From a6f5282973321b452c399b87895dbb29adf50a31 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Thu, 12 May 2022 16:14:18 +0800
Subject: [PATCH 001/134] fix

---
 ppocr/data/imaug/label_ops.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 02a5187dad..8b017b3219 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -438,10 +438,10 @@ class KieLabelEncode(object):
             texts.append(ann['transcription'])
             text_ind = [self.dict[c] for c in text if c in self.dict]
             text_inds.append(text_ind)
-            if 'label' in anno.keys():
+            if 'label' in ann.keys():
                 labels.append(ann['label'])
-            elif 'key_cls' in anno.keys():
-                labels.append(anno['key_cls'])
+            elif 'key_cls' in ann.keys():
+                labels.append(ann['key_cls'])
             else:
                 raise ValueError("Cannot found 'key_cls' in ann.keys(), please check your training annotation.")
             edges.append(ann.get('edge', 0))

From 7036ec1b5f33edf431bbf414190322631c9a2d0b Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 17 May 2022 15:55:55 +0800
Subject: [PATCH 002/134] add doc

---
 doc/doc_ch/PPOCRv3_det_train.md | 252 ++++++++++++++++++++++++++++++++
 1 file changed, 252 insertions(+)
 create mode 100644 doc/doc_ch/PPOCRv3_det_train.md

diff --git a/doc/doc_ch/PPOCRv3_det_train.md b/doc/doc_ch/PPOCRv3_det_train.md
new file mode 100644
index 0000000000..601acddee1
--- /dev/null
+++ b/doc/doc_ch/PPOCRv3_det_train.md
@@ -0,0 +1,252 @@
+
+# PP-OCRv3 文本检测模型训练
+
+- [1. 简介](#1)
+- [2. PPOCRv3检测训练](#2)
+- [3. 基于PPOCRv3检测的finetune训练](#3)
+
+<a name="1"></a>
+## 1. 简介
+
+PP-OCRv3在PP-OCRv2的基础上进一步升级。本节介绍PP-OCRv3检测模型的训练步骤。有关PPOCRv3策略介绍参考[文档](./PP-OCRv3_introduction.md)。
+
+
+<a name="2"></a>
+## 2. 检测训练
+
+PP-OCRv3检测模型是对PP-OCRv2中的[CML](https://arxiv.org/pdf/2109.03144.pdf)（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。PP-OCRv3分别针对检测教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+
+PP-OCRv3检测训练包括两个步骤：
+- 步骤1：采用DML蒸馏方法训练检测教师模型
+- 步骤2：使用步骤1得到的教师模型采用CML方法训练出轻量学生模型
+
+
+### 2.1 准备数据和运行环境
+
+训练数据采用icdar2015数据，准备训练集步骤参考[ocr_dataset](./dataset/ocr_datasets.md).
+
+运行环境准备参考[文档](./installation.md)。
+
+
+### 2.2 训练教师模型
+
+教师模型训练的配置文件是[ch_PP-OCRv3_det_dml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.5/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml)。教师模型模型结构的Backbone、Neck、Head分别为Resnet50, LKPAN, DBHead，采用DML的蒸馏方法训练。有关配置文件的详细介绍参考[文档](./knowledge_distillation)。
+
+
+下载ImageNet预训练模型：
+```
+# 下载ResNet50_vd的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+```
+
+**启动训练**
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Global.save_model_dir=./output/
+```
+
+训练过程中保存的模型在output目录下，包含以下文件：
+```
+best_accuracy.states    
+best_accuracy.pdparams  # 默认保存最优精度的模型参数
+best_accuracy.pdopt     # 默认保存最优精度的优化器相关参数
+latest.states    
+latest.pdparams  # 默认保存的最新模型参数
+latest.pdopt     # 默认保存的最新模型的优化器相关参数
+```
+其中，best_accuracy是保存的精度最高的模型参数，可以直接使用该模型评估。
+
+模型评估命令如下：
+```
+python3 tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+训练的教师模型结构更大，精度更高，用于提升学生模型的精度。
+
+**提取教师模型参数**
+best_accuracy包含两个模型的参数，分别对应配置文件中的Student，Student2。提取Student的参数方法如下：
+
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/dml_teacher.pdparams")
+```
+
+提取出来的模型参数可以用于模型进一步的finetune训练或者蒸馏训练。
+
+### 2.3 训练学生模型
+
+训练学生模型的配置文件是[ch_PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.5/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml)
+上一节训练得到的教师模型作为监督，采用CML方式训练得到轻量的学生模型。
+
+下载学生模型的ImageNet预训练模型：
+```
+# 下载MobileNetV3的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+```
+
+**启动训练**
+
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+       Global.save_model_dir=./output/
+```
+
+训练过程中保存的模型在output目录下，
+模型评估命令如下：
+```
+python3 tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+best_accuracy包含三个模型的参数，分别对应配置文件中的Student，Student2，Teacher。提取Student参数的方法如下：
+
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/cml_student.pdparams")
+```
+
+提取出来的Student的参数可用于模型部署或者做进一步的finetune训练。
+
+
+
+<a name="3"></a>
+## 3. 基于PPOCRv3检测finetune训练
+
+本节介绍如何使用PPOCRv3检测模型在其他场景上的finetune训练。
+
+finetune训练适用于三种场景：
+- 基于CML蒸馏方法的finetune训练，适用于教师模型在使用场景上精度高于PPOCRv3检测模型，且希望得到一个轻量检测模型。
+- 基于PPOCRv3轻量检测模型的finetune训练，无需训练教师模型，希望在PPOCRv3检测模型基础上提升使用场景上的精度。
+- 基于DML蒸馏方法的finetune训练，适用于采用DML方法进一步提升精度的场景。
+
+
+**基于CML蒸馏方法的finetune训练**
+
+下载PPOCRv3训练模型：
+```
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams包含CML配置文件中Student、Student2、Teacher模型的参数。
+
+启动训练：
+
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+    -o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy  \
+       Global.save_model_dir=./output/
+```
+
+**基于PPOCRv3轻量检测模型的finetune训练**
+
+
+下载PPOCRv3训练模型，并提取Student结构的模型参数：
+```
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+
+提取Student参数的方法如下：
+
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./student.pdparams")
+```
+
+使用配置文件[ch_PP-OCRv3_det_student.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.5/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml)训练。
+
+**启动训练**
+
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml \
+    -o Global.pretrained_model=./student \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml \
+    -o Global.pretrained_model=./student \
+       Global.save_model_dir=./output/
+```
+
+
+**基于DML蒸馏方法的finetune训练**
+
+以ch_PP-OCRv3_det_distill_train中的Teacher模型为例，首先提取Teacher结构的参数，方法如下：
+```
+import paddle
+# 加载预训练模型
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Teacher."):]: all_params[key] for key in all_params if "Teacher." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./teacher.pdparams")
+```
+
+**启动训练**
+```
+# 单卡训练
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./teacher \
+       Architecture.Models.Student2.pretrained=./teacher \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./teacher \
+       Architecture.Models.Student2.pretrained=./teacher \
+       Global.save_model_dir=./output/
+```
+
+

From bb0c95965682fb08195b02808ebcce5efd58fc6b Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Thu, 19 May 2022 19:46:40 +0800
Subject: [PATCH 003/134] add benchmark

---
 applications/车牌识别.md | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)

diff --git a/applications/车牌识别.md b/applications/车牌识别.md
index 2abb9c6f7a..afa8d57107 100644
--- a/applications/车牌识别.md
+++ b/applications/车牌识别.md
@@ -357,11 +357,13 @@ python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCR
 
 |方案|hmeans| 模型大小 |预测速度(lite)|
 |---|---|------|---|
-|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%| 2.5M ||
-|PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化|98.91%| 1M   ||
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%| 2.5M | 223ms/image |
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化|98.91%| 1M   | 189ms/image |
 
 可以看到量化后能显著降低模型体积并且精度几乎无损。
 
+预测速度是在android骁龙855上预测275张图像的平均耗时。
+
 #### 4.1.4 模型导出
 
 使用如下命令可以将训练好的模型进行导出
@@ -571,11 +573,13 @@ python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_
 
 |方案| acc    | 模型大小  |预测速度(lite)|
 |---|--------|-------|---|
-|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% | 10.3M |
-|PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化| 93.4%  | 4.8M  |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% | 10.3M | 4.2ms/image |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化| 93.4%  | 4.8M  | 1.8ms/image; |
 
 可以看到量化后能显著降低模型体积，但是由于识别数据过少，量化带来了1%的精度下降。
 
+预测速度是在android骁龙855上预测5006张识别文字图像的平均耗时。
+
 #### 4.2.5 模型导出
 
 使用如下命令可以将训练好的模型进行导出。
@@ -618,19 +622,22 @@ python tools/infer/predict_system.py \
 
 |方案|hmeans| 模型大小 |预测速度(lite)|
 |---|---|------|---|
-|PP-OCRv3中英文超轻量检测预训练模型直接预测|76.12%|2.5M|
-|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%| 2.5M ||
-|PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化|98.91%| 1M   ||
+|PP-OCRv3中英文超轻量检测预训练模型直接预测|76.12%|2.5M| 223ms/image |
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%| 2.5M | 223ms/image |
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化|98.91%| 1M   | 189ms/image |
+
+预测速度是在android骁龙855上预测275张图像的平均耗时。
 
 - 识别
 
 |方案| acc    | 模型大小  |预测速度(lite)|
 |---|--------|-------|---|
-|PP-OCRv3中英文超轻量识别预训练模型直接预测| 0%     |10.3M||
-|PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的`·`| 90.97% |10.3M||
-|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% | 10.3M ||
-|PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化| 94.4%  | 4.8M  ||
+|PP-OCRv3中英文超轻量识别预训练模型直接预测| 0%     |10.3M| 4.2ms/image |
+|PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的`·`| 90.97% |10.3M| 4.2ms/image |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% | 10.3M | 4.2ms/image |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化| 94.4%  | 4.8M  | 1.8ms/image |
 
+预测速度是在android骁龙855上预测5006张识别文字图像的平均耗时。
 
 - 结论
 

From 16bafedeba9cdae18290717278dc9079f32411b4 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Thu, 19 May 2022 19:48:11 +0800
Subject: [PATCH 004/134] add benchmark

---
 applications/车牌识别.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/applications/车牌识别.md b/applications/车牌识别.md
index afa8d57107..7510f3c5a8 100644
--- a/applications/车牌识别.md
+++ b/applications/车牌识别.md
@@ -362,7 +362,7 @@ python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCR
 
 可以看到量化后能显著降低模型体积并且精度几乎无损。
 
-预测速度是在android骁龙855上预测275张图像的平均耗时。
+预测速度是在android骁龙855上预测275张图像的平均耗时。模型在移动端的部署步骤参考[文档](../deploy/lite/readme_ch.md)
 
 #### 4.1.4 模型导出
 
@@ -578,7 +578,7 @@ python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_
 
 可以看到量化后能显著降低模型体积，但是由于识别数据过少，量化带来了1%的精度下降。
 
-预测速度是在android骁龙855上预测5006张识别文字图像的平均耗时。
+预测速度是在android骁龙855上预测5006张识别文字图像的平均耗时。模型在移动端的部署步骤参考[文档](../deploy/lite/readme_ch.md)
 
 #### 4.2.5 模型导出
 

From 9816aebdb078ec14ca3141560f2431981c8948f5 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Fri, 27 May 2022 06:49:52 +0000
Subject: [PATCH 005/134] add rec vitstr algorithm.

---
 configs/rec/rec_mtb_nrtr.yml                  |  10 +-
 configs/rec/rec_svtrnet.yml                   |   4 +-
 configs/rec/rec_vitstr.yml                    | 100 ++++++++++++
 doc/doc_ch/algorithm_overview.md              |   3 +-
 doc/doc_ch/algorithm_rec_vitstr.md            | 154 ++++++++++++++++++
 doc/doc_en/algorithm_overview_en.md           |   3 +-
 doc/doc_en/algorithm_rec_vitstr_en.md         | 134 +++++++++++++++
 ppocr/data/imaug/__init__.py                  |   2 +-
 ppocr/data/imaug/label_ops.py                 |  35 +++-
 ppocr/data/imaug/rec_img_aug.py               |  19 ++-
 ppocr/losses/__init__.py                      |   7 +-
 ...rec_nrtr_loss.py => rec_ce_smooth_loss.py} |  14 +-
 ppocr/modeling/backbones/__init__.py          |   3 +-
 ppocr/modeling/backbones/rec_svtrnet.py       | 102 ++++++------
 ppocr/modeling/backbones/rec_vitstr.py        | 120 ++++++++++++++
 ppocr/postprocess/__init__.py                 |   4 +-
 ppocr/postprocess/rec_postprocess.py          |  34 +++-
 .../configs/rec_mtb_nrtr/rec_mtb_nrtr.yml     |   6 +-
 test_tipc/configs/rec_svtrnet/rec_svtrnet.yml | 119 ++++++++++++++
 .../rec_svtrnet/train_infer_python.txt        |  53 ++++++
 test_tipc/configs/rec_vitstr/rec_vitstr.yml   | 101 ++++++++++++
 .../configs/rec_vitstr/train_infer_python.txt |  53 ++++++
 tools/export_model.py                         |   6 +
 tools/infer/predict_rec.py                    |  19 ++-
 tools/program.py                              |  32 ++--
 25 files changed, 1044 insertions(+), 93 deletions(-)
 create mode 100644 configs/rec/rec_vitstr.yml
 create mode 100644 doc/doc_ch/algorithm_rec_vitstr.md
 create mode 100644 doc/doc_en/algorithm_rec_vitstr_en.md
 rename ppocr/losses/{rec_nrtr_loss.py => rec_ce_smooth_loss.py} (73%)
 create mode 100644 ppocr/modeling/backbones/rec_vitstr.py
 create mode 100644 test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
 create mode 100644 test_tipc/configs/rec_svtrnet/train_infer_python.txt
 create mode 100644 test_tipc/configs/rec_vitstr/rec_vitstr.yml
 create mode 100644 test_tipc/configs/rec_vitstr/train_infer_python.txt

diff --git a/configs/rec/rec_mtb_nrtr.yml b/configs/rec/rec_mtb_nrtr.yml
index 0426750085..dfe2cc9811 100644
--- a/configs/rec/rec_mtb_nrtr.yml
+++ b/configs/rec/rec_mtb_nrtr.yml
@@ -49,7 +49,7 @@ Architecture:
     
 
 Loss:
-  name: NRTRLoss
+  name: CESmoothingLoss
   smoothing: True
 
 PostProcess:
@@ -68,8 +68,8 @@ Train:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
-          image_shape: [100, 32]
+      - GrayRecResizeImg:
+          image_shape: [100, 32] # W H
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
           keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
@@ -88,8 +88,8 @@ Eval:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
-          image_shape: [100, 32]
+      - GrayRecResizeImg:
+          image_shape: [100, 32] # W H
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
           keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
diff --git a/configs/rec/rec_svtrnet.yml b/configs/rec/rec_svtrnet.yml
index 233d5e2765..a700e4bd92 100644
--- a/configs/rec/rec_svtrnet.yml
+++ b/configs/rec/rec_svtrnet.yml
@@ -77,7 +77,7 @@ Metric:
 Train:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/training/
+    data_dir: ./train_data/data_lmdb_release/training
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
@@ -98,7 +98,7 @@ Train:
 Eval:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/validation/
+    data_dir: ./train_data/data_lmdb_release/validation
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
diff --git a/configs/rec/rec_vitstr.yml b/configs/rec/rec_vitstr.yml
new file mode 100644
index 0000000000..005db0184a
--- /dev/null
+++ b/configs/rec/rec_vitstr.yml
@@ -0,0 +1,100 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/vitstr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration#
+  eval_batch_step: [0, 50]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_vitstr.txt
+
+
+Optimizer:
+  name: Adadelta
+  epsilon: 0.00000001
+  rho: 0.95
+  clip_norm: 5.0
+  lr:
+    learning_rate: 1.0
+
+Architecture:
+  model_type: rec
+  algorithm: ViTSTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ViTSTR
+    scale: tiny
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CESmoothingLoss
+  smoothing: False
+  with_all: True
+
+PostProcess:
+  name: ViTSTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 2
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index 6227a21498..934ac08537 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -66,6 +66,7 @@
 - [x]  [SAR](./algorithm_rec_sar.md)
 - [x]  [SEED](./algorithm_rec_seed.md)
 - [x]  [SVTR](./algorithm_rec_svtr.md)
+- [x]  [ViTSTR](./algorithm_rec_vitstr.md)
 
 参考[DTRB](https://arxiv.org/abs/1904.01906)[3]文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
 
@@ -84,7 +85,7 @@
 |SAR|Resnet31| 87.20% | rec_r31_sar | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_ch/algorithm_rec_vitstr.md b/doc/doc_ch/algorithm_rec_vitstr.md
new file mode 100644
index 0000000000..bd83b8d9c2
--- /dev/null
+++ b/doc/doc_ch/algorithm_rec_vitstr.md
@@ -0,0 +1,154 @@
+# 场景文本识别算法-ViTSTR
+
+- [1. 算法简介](#1)
+- [2. 环境配置](#2)
+- [3. 模型训练、评估、预测](#3)
+    - [3.1 训练](#3-1)
+    - [3.2 评估](#3-2)
+    - [3.3 预测](#3-3)
+- [4. 推理部署](#4)
+    - [4.1 Python推理](#4-1)
+    - [4.2 C++推理](#4-2)
+    - [4.3 Serving服务化部署](#4-3)
+    - [4.4 更多推理部署](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. 算法简介
+
+论文信息：
+> [Vision Transformer for Fast and Efficient Scene Text Recognition](https://arxiv.org/abs/2105.08582)
+> Rowel Atienza
+> ICDAR, 2021
+
+
+<a name="model"></a>
+`ViTSTR`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|ViTSTR|ViTSTR|[rec_vitstr.yml](../../configs/rec/rec_vitstr.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
+
+<a name="2"></a>
+## 2. 环境配置
+请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](./clone.md)克隆项目代码。
+
+
+<a name="3"></a>
+## 3. 模型训练、评估、预测
+
+<a name="3-1"></a>
+### 3.1 模型训练
+
+请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](../../configs/rec/rec_ViTSTR.yml)。
+
+#### 启动训练
+
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_vitstr.yml
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr.yml
+```
+
+<a name="3-2"></a>
+### 3.2 评估
+
+可下载已训练完成的[模型文件](#model)，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy
+```
+
+<a name="3-3"></a>
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+
+<a name="4"></a>
+## 4. 推理部署
+
+<a name="4-1"></a>
+### 4.1 Python推理
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) )，可以使用如下命令进行转换：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/
+```
+**注意：**
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应NRTR的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+```
+/inference/rec_vitstr/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```shell
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_vitstr/' --rec_algorithm='ViTSTR' --rec_image_shape='1,224,224' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![](../imgs_words_en/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[1，224，224]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中NRTR的预处理为您的预处理方法。
+
+
+<a name="4-2"></a>
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持NRTR，所以暂未支持
+
+<a name="4-3"></a>
+### 4.3 Serving服务化部署
+
+暂不支持
+
+<a name="4-4"></a>
+### 4.4 更多推理部署
+
+暂不支持
+
+<a name="5"></a>
+## 5. FAQ
+
+1. 在`ViTSTR`论文中，使用在ImageNet1k上的预训练权重进行初始化训练，我们在训练未采用预训练权重，最终精度没有变化甚至有所提高。
+2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果有需要使用small、base版本，可直接使用源开源repo中的预训练权重转为Paddle权重即可使用。
+
+## 引用
+
+```bibtex
+@article{Atienza2021ViTSTR,
+  title     = {Vision Transformer for Fast and Efficient Scene Text Recognition},
+  author    = {Rowel Atienza},
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.08582}
+}
+```
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index 383cbe39bb..213d95807d 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -65,6 +65,7 @@ Supported text recognition algorithms (Click the link to get the tutorial):
 - [x]  [SAR](./algorithm_rec_sar_en.md)
 - [x]  [SEED](./algorithm_rec_seed_en.md)
 - [x]  [SVTR](./algorithm_rec_svtr_en.md)
+- [x]  [ViTSTR](./algorithm_rec_vitstr_en.md)
 
 Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
 
@@ -83,7 +84,7 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 |SAR|Resnet31| 87.20% | rec_r31_sar | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_en/algorithm_rec_vitstr_en.md b/doc/doc_en/algorithm_rec_vitstr_en.md
new file mode 100644
index 0000000000..666798c3e0
--- /dev/null
+++ b/doc/doc_en/algorithm_rec_vitstr_en.md
@@ -0,0 +1,134 @@
+# ViTSTR
+
+- [1. Introduction](#1)
+- [2. Environment](#2)
+- [3. Model Training / Evaluation / Prediction](#3)
+    - [3.1 Training](#3-1)
+    - [3.2 Evaluation](#3-2)
+    - [3.3 Prediction](#3-3)
+- [4. Inference and Deployment](#4)
+    - [4.1 Python Inference](#4-1)
+    - [4.2 C++ Inference](#4-2)
+    - [4.3 Serving](#4-3)
+    - [4.4 More](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. Introduction
+
+Paper:
+> [Vision Transformer for Fast and Efficient Scene Text Recognition](https://arxiv.org/abs/2105.08582)
+> Rowel Atienza
+> ICDAR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|ViTSTR|ViTSTR|[rec_vitstr.yml](../../configs/rec/rec_vitstr.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
+
+<a name="2"></a>
+## 2. Environment
+Please refer to ["Environment Preparation"](./environment_en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone_en.md) to clone the project code.
+
+
+<a name="3"></a>
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](./recognition_en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```
+#Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_vitstr.yml
+
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr.yml
+```
+
+Evaluation:
+
+```
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+Prediction:
+
+```
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_train/best_accuracy
+```
+
+<a name="4"></a>
+## 4. Inference and Deployment
+
+<a name="4-1"></a>
+### 4.1 Python Inference
+First, the model saved during the ViTSTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)) ), you can use the following command to convert:
+
+```
+python3 tools/export_model.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy  Global.save_inference_dir=./inference/rec_vitstr
+```
+
+**Note:**
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to ViTSTR in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+```
+/inference/rec_vitstr/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+
+For ViTSTR text recognition model inference, the following commands can be executed:
+
+```
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_vitstr/' --rec_algorithm='ViTSTR' --rec_image_shape='1,224,224' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+```
+
+![](../imgs_words_en/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+```
+
+<a name="4-2"></a>
+### 4.2 C++ Inference
+
+Not supported
+
+<a name="4-3"></a>
+### 4.3 Serving
+
+Not supported
+
+<a name="4-4"></a>
+### 4.4 More
+
+Not supported
+
+<a name="5"></a>
+## 5. FAQ
+
+1. In the `ViTSTR` paper, using pre-trained weights on ImageNet1k for initial training, we did not use pre-trained weights in training, and the final accuracy did not change or even improved.
+
+## Citation
+
+```bibtex
+@article{Atienza2021ViTSTR,
+  title     = {Vision Transformer for Fast and Efficient Scene Text Recognition},
+  author    = {Rowel Atienza},
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.08582}
+}
+```
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index 548832fb0d..2dbc92a703 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -23,7 +23,7 @@ from .random_crop_data import EastRandomCropData, RandomCropImgMask
 from .make_pse_gt import MakePseGt
 
 from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
-    SRNRecResizeImg, NRTRRecResizeImg, SARRecResizeImg, PRENResizeImg
+    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg
 from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 02a5187dad..0366e3f785 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -443,7 +443,9 @@ class KieLabelEncode(object):
             elif 'key_cls' in anno.keys():
                 labels.append(anno['key_cls'])
             else:
-                raise ValueError("Cannot found 'key_cls' in ann.keys(), please check your training annotation.")
+                raise ValueError(
+                    "Cannot found 'key_cls' in ann.keys(), please check your training annotation."
+                )
             edges.append(ann.get('edge', 0))
         ann_infos = dict(
             image=data['image'],
@@ -838,6 +840,37 @@ class PRENLabelEncode(BaseRecLabelEncode):
         return data
 
 
+class ViTSTRLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 **kwargs):
+
+        super(ViTSTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data['length'] = np.array(len(text))
+        text.insert(0, 0)
+        text.append(1)
+        text = text + [0] * (self.max_text_len + 2 - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['<s>', '</s>'] + dict_character
+        return dict_character
+
+
 class VQATokenLabelEncode(object):
     """
     Label encode for NLP VQA methods
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index 7483dffe5b..0697baf436 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -87,11 +87,19 @@ class ClsResizeImg(object):
         return data
 
 
-class NRTRRecResizeImg(object):
-    def __init__(self, image_shape, resize_type, padding=False, **kwargs):
+class GrayRecResizeImg(object):
+    def __init__(self,
+                 image_shape,
+                 resize_type,
+                 inter_type='Image.ANTIALIAS',
+                 scale=True,
+                 padding=False,
+                 **kwargs):
         self.image_shape = image_shape
         self.resize_type = resize_type
         self.padding = padding
+        self.inter_type = eval(inter_type)
+        self.scale = scale
 
     def __call__(self, data):
         img = data['image']
@@ -117,13 +125,16 @@ class NRTRRecResizeImg(object):
             return data
         if self.resize_type == 'PIL':
             image_pil = Image.fromarray(np.uint8(img))
-            img = image_pil.resize(self.image_shape, Image.ANTIALIAS)
+            img = image_pil.resize(self.image_shape, self.inter_type)
             img = np.array(img)
         if self.resize_type == 'OpenCV':
             img = cv2.resize(img, self.image_shape)
         norm_img = np.expand_dims(img, -1)
         norm_img = norm_img.transpose((2, 0, 1))
-        data['image'] = norm_img.astype(np.float32) / 128. - 1.
+        if self.scale:
+            data['image'] = norm_img.astype(np.float32) / 128. - 1.
+        else:
+            data['image'] = norm_img.astype(np.float32) / 255.
         return data
 
 
diff --git a/ppocr/losses/__init__.py b/ppocr/losses/__init__.py
index de8419b7c1..6c4545eb21 100755
--- a/ppocr/losses/__init__.py
+++ b/ppocr/losses/__init__.py
@@ -30,7 +30,7 @@ from .det_fce_loss import FCELoss
 from .rec_ctc_loss import CTCLoss
 from .rec_att_loss import AttentionLoss
 from .rec_srn_loss import SRNLoss
-from .rec_nrtr_loss import NRTRLoss
+from .rec_ce_smooth_loss import CESmoothingLoss
 from .rec_sar_loss import SARLoss
 from .rec_aster_loss import AsterLoss
 from .rec_pren_loss import PRENLoss
@@ -60,8 +60,9 @@ def build_loss(config):
     support_dict = [
         'DBLoss', 'PSELoss', 'EASTLoss', 'SASTLoss', 'FCELoss', 'CTCLoss',
         'ClsLoss', 'AttentionLoss', 'SRNLoss', 'PGLoss', 'CombinedLoss',
-        'NRTRLoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', 'SDMGRLoss',
-        'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', 'MultiLoss'
+        'CESmoothingLoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss',
+        'SDMGRLoss', 'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss',
+        'MultiLoss'
     ]
     config = copy.deepcopy(config)
     module_name = config.pop('name')
diff --git a/ppocr/losses/rec_nrtr_loss.py b/ppocr/losses/rec_ce_smooth_loss.py
similarity index 73%
rename from ppocr/losses/rec_nrtr_loss.py
rename to ppocr/losses/rec_ce_smooth_loss.py
index 200a6d0486..22243ed41f 100644
--- a/ppocr/losses/rec_nrtr_loss.py
+++ b/ppocr/losses/rec_ce_smooth_loss.py
@@ -3,16 +3,20 @@ from paddle import nn
 import paddle.nn.functional as F
 
 
-class NRTRLoss(nn.Layer):
-    def __init__(self, smoothing=True, **kwargs):
-        super(NRTRLoss, self).__init__()
+class CESmoothingLoss(nn.Layer):
+    def __init__(self, smoothing=True, with_all=False, **kwargs):
+        super(CESmoothingLoss, self).__init__()
         self.loss_func = nn.CrossEntropyLoss(reduction='mean', ignore_index=0)
         self.smoothing = smoothing
+        self.with_all = with_all
 
     def forward(self, pred, batch):
         pred = pred.reshape([-1, pred.shape[2]])
-        max_len = batch[2].max()
-        tgt = batch[1][:, 1:2 + max_len]
+        if self.with_all:
+            tgt = batch[1]
+        else:
+            max_len = batch[2].max()
+            tgt = batch[1][:, 1:2 + max_len]
         tgt = tgt.reshape([-1])
         if self.smoothing:
             eps = 0.1
diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
index 072d6e0f84..a368e74816 100755
--- a/ppocr/modeling/backbones/__init__.py
+++ b/ppocr/modeling/backbones/__init__.py
@@ -32,10 +32,11 @@ def build_backbone(config, model_type):
         from .rec_micronet import MicroNet
         from .rec_efficientb3_pren import EfficientNetb3_PREN
         from .rec_svtrnet import SVTRNet
+        from .rec_vitstr import ViTSTR
         support_dict = [
             'MobileNetV1Enhance', 'MobileNetV3', 'ResNet', 'ResNetFPN', 'MTB',
             "ResNet31", "ResNet_ASTER", 'MicroNet', 'EfficientNetb3_PREN',
-            'SVTRNet'
+            'SVTRNet', 'ViTSTR'
         ]
     elif model_type == "e2e":
         from .e2e_resnet_vd_pg import ResNet
diff --git a/ppocr/modeling/backbones/rec_svtrnet.py b/ppocr/modeling/backbones/rec_svtrnet.py
index c57bf46345..c2c07f4476 100644
--- a/ppocr/modeling/backbones/rec_svtrnet.py
+++ b/ppocr/modeling/backbones/rec_svtrnet.py
@@ -147,7 +147,7 @@ class Attention(nn.Layer):
                  dim,
                  num_heads=8,
                  mixer='Global',
-                 HW=[8, 25],
+                 HW=None,
                  local_k=[7, 11],
                  qkv_bias=False,
                  qk_scale=None,
@@ -210,7 +210,7 @@ class Block(nn.Layer):
                  num_heads,
                  mixer='Global',
                  local_mixer=[7, 11],
-                 HW=[8, 25],
+                 HW=None,
                  mlp_ratio=4.,
                  qkv_bias=False,
                  qk_scale=None,
@@ -274,7 +274,9 @@ class PatchEmbed(nn.Layer):
                  img_size=[32, 100],
                  in_channels=3,
                  embed_dim=768,
-                 sub_num=2):
+                 sub_num=2,
+                 patch_size=[4, 4],
+                 mode='pope'):
         super().__init__()
         num_patches = (img_size[1] // (2 ** sub_num)) * \
                       (img_size[0] // (2 ** sub_num))
@@ -282,50 +284,56 @@ class PatchEmbed(nn.Layer):
         self.num_patches = num_patches
         self.embed_dim = embed_dim
         self.norm = None
-        if sub_num == 2:
-            self.proj = nn.Sequential(
-                ConvBNLayer(
-                    in_channels=in_channels,
-                    out_channels=embed_dim // 2,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None),
-                ConvBNLayer(
-                    in_channels=embed_dim // 2,
-                    out_channels=embed_dim,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None))
-        if sub_num == 3:
-            self.proj = nn.Sequential(
-                ConvBNLayer(
-                    in_channels=in_channels,
-                    out_channels=embed_dim // 4,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None),
-                ConvBNLayer(
-                    in_channels=embed_dim // 4,
-                    out_channels=embed_dim // 2,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None),
-                ConvBNLayer(
-                    in_channels=embed_dim // 2,
-                    out_channels=embed_dim,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None))
+        if mode == 'pope':
+            if sub_num == 2:
+                self.proj = nn.Sequential(
+                    ConvBNLayer(
+                        in_channels=in_channels,
+                        out_channels=embed_dim // 2,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 2,
+                        out_channels=embed_dim,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None))
+            if sub_num == 3:
+                self.proj = nn.Sequential(
+                    ConvBNLayer(
+                        in_channels=in_channels,
+                        out_channels=embed_dim // 4,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 4,
+                        out_channels=embed_dim // 2,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 2,
+                        out_channels=embed_dim,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None))
+        elif mode == 'linear':
+            self.proj = nn.Conv2D(
+                1, embed_dim, kernel_size=patch_size, stride=patch_size)
+            self.num_patches = img_size[0] // patch_size[0] * img_size[
+                1] // patch_size[1]
 
     def forward(self, x):
         B, C, H, W = x.shape
diff --git a/ppocr/modeling/backbones/rec_vitstr.py b/ppocr/modeling/backbones/rec_vitstr.py
new file mode 100644
index 0000000000..d5d7d5148a
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_vitstr.py
@@ -0,0 +1,120 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/roatienza/deep-text-recognition-benchmark/blob/master/modules/vitstr.py
+"""
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+from ppocr.modeling.backbones.rec_svtrnet import Block, PatchEmbed, zeros_, trunc_normal_, ones_
+
+scale_dim_heads = {'tiny': [192, 3], 'small': [384, 6], 'base': [768, 12]}
+
+
+class ViTSTR(nn.Layer):
+    def __init__(self,
+                 img_size=[224, 224],
+                 in_channels=1,
+                 scale='tiny',
+                 seqlen=27,
+                 patch_size=[16, 16],
+                 embed_dim=None,
+                 depth=12,
+                 num_heads=None,
+                 mlp_ratio=4,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 drop_path_rate=0.,
+                 drop_rate=0.,
+                 attn_drop_rate=0.,
+                 norm_layer='nn.LayerNorm',
+                 act_layer='nn.GELU',
+                 epsilon=1e-6,
+                 out_channels=None,
+                 **kwargs):
+        super().__init__()
+        self.seqlen = seqlen
+        embed_dim = embed_dim if embed_dim is not None else scale_dim_heads[
+            scale][0]
+        num_heads = num_heads if num_heads is not None else scale_dim_heads[
+            scale][1]
+        out_channels = out_channels if out_channels is not None else embed_dim
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            in_channels=in_channels,
+            embed_dim=embed_dim,
+            patch_size=patch_size,
+            mode='linear')
+        num_patches = self.patch_embed.num_patches
+
+        self.pos_embed = self.create_parameter(
+            shape=[1, num_patches + 1, embed_dim], default_initializer=zeros_)
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.cls_token = self.create_parameter(
+            shape=[1, 1, embed_dim], default_initializer=zeros_)
+        self.add_parameter("cls_token", self.cls_token)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = np.linspace(0, drop_path_rate, depth)
+        self.blocks = nn.LayerList([
+            Block(
+                dim=embed_dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[i],
+                norm_layer=norm_layer,
+                act_layer=eval(act_layer),
+                epsilon=epsilon,
+                prenorm=False) for i in range(depth)
+        ])
+        self.norm = eval(norm_layer)(embed_dim, epsilon=epsilon)
+
+        self.out_channels = out_channels
+
+        trunc_normal_(self.pos_embed)
+        trunc_normal_(self.cls_token)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward_features(self, x):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+        cls_tokens = paddle.tile(self.cls_token, repeat_times=[B, 1, 1])
+        x = paddle.concat((cls_tokens, x), axis=1)
+        x = x + self.pos_embed
+        x = self.pos_drop(x)
+        for blk in self.blocks:
+            x = blk(x)
+        x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        x = x[:, :self.seqlen]
+        return x.transpose([0, 2, 1]).unsqueeze(2)
diff --git a/ppocr/postprocess/__init__.py b/ppocr/postprocess/__init__.py
index f50b5f1c5f..4f900ee1fc 100644
--- a/ppocr/postprocess/__init__.py
+++ b/ppocr/postprocess/__init__.py
@@ -27,7 +27,7 @@ from .sast_postprocess import SASTPostProcess
 from .fce_postprocess import FCEPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, \
     DistillationCTCLabelDecode, TableLabelDecode, NRTRLabelDecode, SARLabelDecode, \
-    SEEDLabelDecode, PRENLabelDecode
+    SEEDLabelDecode, PRENLabelDecode, ViTSTRLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
 from .vqa_token_ser_layoutlm_postprocess import VQASerTokenLayoutLMPostProcess
@@ -42,7 +42,7 @@ def build_post_process(config, global_config=None):
         'DistillationDBPostProcess', 'NRTRLabelDecode', 'SARLabelDecode',
         'SEEDLabelDecode', 'VQASerTokenLayoutLMPostProcess',
         'VQAReTokenLayoutLMPostProcess', 'PRENLabelDecode',
-        'DistillationSARLabelDecode'
+        'DistillationSARLabelDecode', 'ViTSTRLabelDecode'
     ]
 
     if config['name'] == 'PSEPostProcess':
diff --git a/ppocr/postprocess/rec_postprocess.py b/ppocr/postprocess/rec_postprocess.py
index bf0fd890bf..df6203fada 100644
--- a/ppocr/postprocess/rec_postprocess.py
+++ b/ppocr/postprocess/rec_postprocess.py
@@ -188,13 +188,13 @@ class NRTRLabelDecode(BaseRecLabelDecode):
             char_list = []
             conf_list = []
             for idx in range(len(text_index[batch_idx])):
-                if text_index[batch_idx][idx] == 3:  # end
-                    break
                 try:
-                    char_list.append(self.character[int(text_index[batch_idx][
-                        idx])])
+                    char_idx = self.character[int(text_index[batch_idx][idx])]
                 except:
                     continue
+                if char_idx == '</s>':  # end
+                    break
+                char_list.append(char_idx)
                 if text_prob is not None:
                     conf_list.append(text_prob[batch_idx][idx])
                 else:
@@ -204,6 +204,32 @@ class NRTRLabelDecode(BaseRecLabelDecode):
         return result_list
 
 
+class ViTSTRLabelDecode(NRTRLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=False,
+                 **kwargs):
+        super(ViTSTRLabelDecode, self).__init__(character_dict_path,
+                                                use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds[:, 1:].numpy()
+        else:
+            preds = preds[:, 1:]
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['<s>', '</s>'] + dict_character
+        return dict_character
+
+
 class AttnLabelDecode(BaseRecLabelDecode):
     """ Convert between text-label and text-index """
 
diff --git a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
index 15119bb2a9..3936ab58ad 100644
--- a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
+++ b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
@@ -49,7 +49,7 @@ Architecture:
     
 
 Loss:
-  name: NRTRLoss
+  name: CESmoothingLoss
   smoothing: True
 
 PostProcess:
@@ -69,7 +69,7 @@ Train:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
+      - GrayRecResizeImg:
           image_shape: [100, 32]
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
@@ -90,7 +90,7 @@ Eval:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
+      - GrayRecResizeImg:
           image_shape: [100, 32]
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
diff --git a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
new file mode 100644
index 0000000000..26facca34d
--- /dev/null
+++ b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
@@ -0,0 +1,119 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_tiny.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 0.00000008
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+    name: STN_ON
+    tps_inputsize: [32, 64]
+    tps_outputsize: [32, 100]
+    num_control_points: 20
+    tps_margins: [0.05,0.05]
+    stn_activation: none
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 100]
+    out_char_num: 25
+    out_channels: 192
+    patch_merging: 'Conv'
+    embed_dim: [64, 128, 256]
+    depth: [3, 6, 3]
+    num_heads: [2, 4, 8]
+    mixer: ['Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[7, 11], [7, 11], [7, 11]]
+    last_stage: True
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          character_dict_path:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          character_dict_path:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/test_tipc/configs/rec_svtrnet/train_infer_python.txt b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
new file mode 100644
index 0000000000..72526063e0
--- /dev/null
+++ b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_svtrnet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_svtrnet_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="SVTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,224,224]}]
diff --git a/test_tipc/configs/rec_vitstr/rec_vitstr.yml b/test_tipc/configs/rec_vitstr/rec_vitstr.yml
new file mode 100644
index 0000000000..427bce4b5a
--- /dev/null
+++ b/test_tipc/configs/rec_vitstr/rec_vitstr.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/vitstr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration#
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_vitstr.txt
+
+
+Optimizer:
+  name: Adadelta
+  epsilon: 0.00000001
+  rho: 0.95
+  clip_norm: 5.0
+  lr:
+    learning_rate: 1.0
+
+Architecture:
+  model_type: rec
+  algorithm: ViTSTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ViTSTR
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CESmoothingLoss
+  smoothing: False
+  with_all: True
+
+PostProcess:
+  name: ViTSTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 2
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/test_tipc/configs/rec_vitstr/train_infer_python.txt b/test_tipc/configs/rec_vitstr/train_infer_python.txt
new file mode 100644
index 0000000000..6c7af1fb87
--- /dev/null
+++ b/test_tipc/configs/rec_vitstr/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_vitstr
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_vitstr_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="ViTSTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,224,224]}]
diff --git a/tools/export_model.py b/tools/export_model.py
index c0cbcd361c..6e003f2ffa 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -73,6 +73,12 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
                 shape=[None, 3, 64, 512], dtype="float32"),
         ]
         model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "ViTSTR":
+        other_shape = [
+            paddle.static.InputSpec(
+                shape=[None, 1, 224, 224], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
     else:
         infer_shape = [3, -1, -1]
         if arch_config["model_type"] == "rec":
diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py
index 3664ef2caf..1945667972 100755
--- a/tools/infer/predict_rec.py
+++ b/tools/infer/predict_rec.py
@@ -69,6 +69,12 @@ class TextRecognizer(object):
                 "character_dict_path": args.rec_char_dict_path,
                 "use_space_char": args.use_space_char
             }
+        elif self.rec_algorithm == 'ViTSTR':
+            postprocess_params = {
+                'name': 'ViTSTRLabelDecode',
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char
+            }
         self.postprocess_op = build_post_process(postprocess_params)
         self.predictor, self.input_tensor, self.output_tensors, self.config = \
             utility.create_predictor(args, 'rec', logger)
@@ -96,15 +102,22 @@ class TextRecognizer(object):
 
     def resize_norm_img(self, img, max_wh_ratio):
         imgC, imgH, imgW = self.rec_image_shape
-        if self.rec_algorithm == 'NRTR':
+        if self.rec_algorithm == 'NRTR' or self.rec_algorithm == 'ViTSTR':
             img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
             # return padding_im
             image_pil = Image.fromarray(np.uint8(img))
-            img = image_pil.resize([100, 32], Image.ANTIALIAS)
+            if self.rec_algorithm == 'ViTSTR':
+                img = image_pil.resize([imgW, imgH], Image.BICUBIC)
+            else:
+                img = image_pil.resize([imgW, imgH], Image.ANTIALIAS)
             img = np.array(img)
             norm_img = np.expand_dims(img, -1)
             norm_img = norm_img.transpose((2, 0, 1))
-            return norm_img.astype(np.float32) / 128. - 1.
+            if self.rec_algorithm == 'ViTSTR':
+                norm_img = norm_img.astype(np.float32) / 255.
+            else:
+                norm_img = norm_img.astype(np.float32) / 128. - 1.
+            return norm_img
 
         assert imgC == img.shape[2]
         imgW = int((imgH * max_wh_ratio))
diff --git a/tools/program.py b/tools/program.py
index aa0d2698cf..745c28b872 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -307,7 +307,8 @@ def train(config,
             train_stats.update(stats)
 
             if log_writer is not None and dist.get_rank() == 0:
-                log_writer.log_metrics(metrics=train_stats.get(), prefix="TRAIN", step=global_step)
+                log_writer.log_metrics(
+                    metrics=train_stats.get(), prefix="TRAIN", step=global_step)
 
             if dist.get_rank() == 0 and (
                 (global_step > 0 and global_step % print_batch_step == 0) or
@@ -354,7 +355,8 @@ def train(config,
 
                 # logger metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics=cur_metric, prefix="EVAL", step=global_step)
+                    log_writer.log_metrics(
+                        metrics=cur_metric, prefix="EVAL", step=global_step)
 
                 if cur_metric[main_indicator] >= best_model_dict[
                         main_indicator]:
@@ -377,11 +379,18 @@ def train(config,
                 logger.info(best_str)
                 # logger best metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics={
-                        "best_{}".format(main_indicator): best_model_dict[main_indicator]
-                        }, prefix="EVAL", step=global_step)
-                    
-                    log_writer.log_model(is_best=True, prefix="best_accuracy", metadata=best_model_dict)
+                    log_writer.log_metrics(
+                        metrics={
+                            "best_{}".format(main_indicator):
+                            best_model_dict[main_indicator]
+                        },
+                        prefix="EVAL",
+                        step=global_step)
+
+                    log_writer.log_model(
+                        is_best=True,
+                        prefix="best_accuracy",
+                        metadata=best_model_dict)
 
             reader_start = time.time()
         if dist.get_rank() == 0:
@@ -413,7 +422,8 @@ def train(config,
                 epoch=epoch,
                 global_step=global_step)
             if log_writer is not None:
-                log_writer.log_model(is_best=False, prefix='iter_epoch_{}'.format(epoch))
+                log_writer.log_model(
+                    is_best=False, prefix='iter_epoch_{}'.format(epoch))
 
     best_str = 'best metric, {}'.format(', '.join(
         ['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))
@@ -564,7 +574,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
+        'ViTSTR'
     ]
 
     if use_xpu:
@@ -585,7 +596,8 @@ def preprocess(is_train=False):
         vdl_writer_path = '{}/vdl/'.format(save_model_dir)
         log_writer = VDLLogger(save_model_dir)
         loggers.append(log_writer)
-    if ('use_wandb' in config['Global'] and config['Global']['use_wandb']) or 'wandb' in config:
+    if ('use_wandb' in config['Global'] and
+            config['Global']['use_wandb']) or 'wandb' in config:
         save_dir = config['Global']['save_model_dir']
         wandb_writer_path = "{}/wandb".format(save_dir)
         if "wandb" in config:

From c956f4a79c229c907ba60a7d4bf87024a4e6dc76 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Fri, 27 May 2022 07:50:48 +0000
Subject: [PATCH 006/134] fix cpu_thread and precision

---
 test_tipc/configs/rec_svtrnet/train_infer_python.txt | 4 ++--
 test_tipc/configs/rec_vitstr/train_infer_python.txt  | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/test_tipc/configs/rec_svtrnet/train_infer_python.txt b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
index 72526063e0..df7638faad 100644
--- a/test_tipc/configs/rec_svtrnet/train_infer_python.txt
+++ b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
@@ -40,10 +40,10 @@ infer_quant:False
 inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="SVTR"
 --use_gpu:True|False
 --enable_mkldnn:False
---cpu_threads:1|6
+--cpu_threads:6
 --rec_batch_num:1|6
 --use_tensorrt:False
---precision:fp32|int8
+--precision:fp32
 --rec_model_dir:
 --image_dir:./inference/rec_inference
 --save_log_path:./test/output/
diff --git a/test_tipc/configs/rec_vitstr/train_infer_python.txt b/test_tipc/configs/rec_vitstr/train_infer_python.txt
index 6c7af1fb87..8b58e220f7 100644
--- a/test_tipc/configs/rec_vitstr/train_infer_python.txt
+++ b/test_tipc/configs/rec_vitstr/train_infer_python.txt
@@ -40,10 +40,10 @@ infer_quant:False
 inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="ViTSTR"
 --use_gpu:True|False
 --enable_mkldnn:False
---cpu_threads:1|6
+--cpu_threads:6
 --rec_batch_num:1|6
 --use_tensorrt:False
---precision:fp32|int8
+--precision:fp32
 --rec_model_dir:
 --image_dir:./inference/rec_inference
 --save_log_path:./test/output/

From bc5522b6f38dc1119bd3af06f985ee80bc9bda9d Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Fri, 27 May 2022 08:29:32 +0000
Subject: [PATCH 007/134] fix svtr tipc

---
 test_tipc/configs/rec_svtrnet/train_infer_python.txt | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/test_tipc/configs/rec_svtrnet/train_infer_python.txt b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
index df7638faad..60fd25ad4f 100644
--- a/test_tipc/configs/rec_svtrnet/train_infer_python.txt
+++ b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
@@ -37,7 +37,7 @@ export2:null
 train_model:./inference/rec_svtrnet_train/best_accuracy
 infer_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
 infer_quant:False
-inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="SVTR"
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="3,64,256" --rec_algorithm="SVTR"
 --use_gpu:True|False
 --enable_mkldnn:False
 --cpu_threads:6
@@ -50,4 +50,4 @@ inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbo
 --benchmark:True
 null:null
 ===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[1,224,224]}]
+random_infer_input:[{float32,[1=3,64,256]}]

From c269e1e42f79d33aa87d5e0ba49b488c56f97ae4 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Mon, 30 May 2022 08:36:11 +0000
Subject: [PATCH 008/134] modify vitstr name

---
 configs/rec/rec_mtb_nrtr.yml                  |  2 +-
 ...{rec_vitstr.yml => rec_vitstr_none_ce.yml} |  4 +--
 doc/doc_ch/algorithm_overview.md              |  2 +-
 doc/doc_ch/algorithm_rec_vitstr.md            | 26 +++++++++----------
 doc/doc_en/algorithm_overview_en.md           |  2 +-
 doc/doc_en/algorithm_rec_vitstr_en.md         | 14 +++++-----
 ppocr/losses/__init__.py                      |  7 +++--
 .../{rec_ce_smooth_loss.py => rec_ce_loss.py} |  4 +--
 .../configs/rec_mtb_nrtr/rec_mtb_nrtr.yml     |  2 +-
 .../rec_mtb_nrtr/train_infer_python.txt       |  4 +--
 .../rec_vitstr_none_ce.yml}                   |  4 +--
 .../train_infer_python.txt                    | 10 +++----
 12 files changed, 40 insertions(+), 41 deletions(-)
 rename configs/rec/{rec_vitstr.yml => rec_vitstr_none_ce.yml} (97%)
 rename ppocr/losses/{rec_ce_smooth_loss.py => rec_ce_loss.py} (93%)
 rename test_tipc/configs/{rec_vitstr/rec_vitstr.yml => rec_vitstr_none_ce/rec_vitstr_none_ce.yml} (97%)
 rename test_tipc/configs/{rec_vitstr => rec_vitstr_none_ce}/train_infer_python.txt (82%)

diff --git a/configs/rec/rec_mtb_nrtr.yml b/configs/rec/rec_mtb_nrtr.yml
index dfe2cc9811..405674882c 100644
--- a/configs/rec/rec_mtb_nrtr.yml
+++ b/configs/rec/rec_mtb_nrtr.yml
@@ -49,7 +49,7 @@ Architecture:
     
 
 Loss:
-  name: CESmoothingLoss
+  name: CELoss
   smoothing: True
 
 PostProcess:
diff --git a/configs/rec/rec_vitstr.yml b/configs/rec/rec_vitstr_none_ce.yml
similarity index 97%
rename from configs/rec/rec_vitstr.yml
rename to configs/rec/rec_vitstr_none_ce.yml
index 005db0184a..065cc785f8 100644
--- a/configs/rec/rec_vitstr.yml
+++ b/configs/rec/rec_vitstr_none_ce.yml
@@ -3,7 +3,7 @@ Global:
   epoch_num: 20
   log_smooth_window: 20
   print_batch_step: 10
-  save_model_dir: ./output/rec/vitstr/
+  save_model_dir: ./output/rec/vitstr_none_ce/
   save_epoch_step: 1
   # evaluation is run every 2000 iterations after the 0th iteration#
   eval_batch_step: [0, 50]
@@ -44,7 +44,7 @@ Architecture:
     name: CTCHead
 
 Loss:
-  name: CESmoothingLoss
+  name: CELoss
   smoothing: False
   with_all: True
 
diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index 934ac08537..4196ec05af 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -85,7 +85,7 @@
 |SAR|Resnet31| 87.20% | rec_r31_sar | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_ch/algorithm_rec_vitstr.md b/doc/doc_ch/algorithm_rec_vitstr.md
index bd83b8d9c2..ab12be7202 100644
--- a/doc/doc_ch/algorithm_rec_vitstr.md
+++ b/doc/doc_ch/algorithm_rec_vitstr.md
@@ -27,7 +27,7 @@
 
 |模型|骨干网络|配置文件|Acc|下载链接|
 | --- | --- | --- | --- | --- |
-|ViTSTR|ViTSTR|[rec_vitstr.yml](../../configs/rec/rec_vitstr.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar)|
 
 <a name="2"></a>
 ## 2. 环境配置
@@ -40,7 +40,7 @@
 <a name="3-1"></a>
 ### 3.1 模型训练
 
-请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](../../configs/rec/rec_ViTSTR.yml)。
+请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](../../configs/rec/rec_vitstr_none_ce.yml)。
 
 #### 启动训练
 
@@ -48,10 +48,10 @@
 具体地，在完成数据准备后，便可以启动训练，训练命令如下：
 ```shell
 #单卡训练（训练周期长，不建议）
-python3 tools/train.py -c configs/rec/rec_vitstr.yml
+python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
 
 #多卡训练，通过--gpus参数指定卡号
-python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr.yml
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
 ```
 
 <a name="3-2"></a>
@@ -61,7 +61,7 @@ python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs
 
 ```shell
 # 注意将pretrained_model的路径设置为本地路径。
-python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
 ```
 
 <a name="3-3"></a>
@@ -70,7 +70,7 @@ python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec
 使用如下命令进行单张图片预测：
 ```shell
 # 注意将pretrained_model的路径设置为本地路径。
-python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_train/best_accuracy
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
 # 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
 ```
 
@@ -80,15 +80,15 @@ python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./
 
 <a name="4-1"></a>
 ### 4.1 Python推理
-首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) )，可以使用如下命令进行转换：
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) )，可以使用如下命令进行转换：
 
 ```shell
 # 注意将pretrained_model的路径设置为本地路径。
-python3 tools/export_model.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/
+python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/
 ```
 **注意：**
 - 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
-- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应NRTR的`infer_shape`。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ViTSTR的`infer_shape`。
 
 转换成功后，在目录下有三个文件：
 ```
@@ -110,20 +110,20 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png'
 执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
 结果如下：
 ```shell
-Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617)
 ```
 
 **注意**：
 
 - 训练上述模型采用的图像分辨率是[1，224，224]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
 - 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
-- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中NRTR的预处理为您的预处理方法。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ViTSTR的预处理为您的预处理方法。
 
 
 <a name="4-2"></a>
 ### 4.2 C++推理部署
 
-由于C++预处理后处理还未支持NRTR，所以暂未支持
+由于C++预处理后处理还未支持ViTSTR，所以暂未支持
 
 <a name="4-3"></a>
 ### 4.3 Serving服务化部署
@@ -139,7 +139,7 @@ Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
 ## 5. FAQ
 
 1. 在`ViTSTR`论文中，使用在ImageNet1k上的预训练权重进行初始化训练，我们在训练未采用预训练权重，最终精度没有变化甚至有所提高。
-2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果有需要使用small、base版本，可直接使用源开源repo中的预训练权重转为Paddle权重即可使用。
+2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果需要使用small、base版本，可将[ViTSTR源repo](https://github.com/roatienza/deep-text-recognition-benchmark) 中的预训练权重转为Paddle权重使用。
 
 ## 引用
 
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index 213d95807d..f2284542ed 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -84,7 +84,7 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 |SAR|Resnet31| 87.20% | rec_r31_sar | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_en/algorithm_rec_vitstr_en.md b/doc/doc_en/algorithm_rec_vitstr_en.md
index 666798c3e0..b6f26d39e2 100644
--- a/doc/doc_en/algorithm_rec_vitstr_en.md
+++ b/doc/doc_en/algorithm_rec_vitstr_en.md
@@ -25,7 +25,7 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval
 
 |Model|Backbone|config|Acc|Download link|
 | --- | --- | --- | --- | --- |
-|ViTSTR|ViTSTR|[rec_vitstr.yml](../../configs/rec/rec_vitstr.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
 
 <a name="2"></a>
 ## 2. Environment
@@ -43,24 +43,24 @@ Specifically, after the data preparation is completed, the training can be start
 
 ```
 #Single GPU training (long training period, not recommended)
-python3 tools/train.py -c configs/rec/rec_vitstr.yml
+python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
 
 #Multi GPU training, specify the gpu number through the --gpus parameter
-python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr.yml
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
 ```
 
 Evaluation:
 
 ```
 # GPU evaluation
-python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
 ```
 
 Prediction:
 
 ```
 # The configuration file used for prediction must match the training
-python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_train/best_accuracy
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
 ```
 
 <a name="4"></a>
@@ -71,7 +71,7 @@ python3 tools/infer_rec.py -c configs/rec/rec_vitstr.yml -o Global.infer_img='./
 First, the model saved during the ViTSTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)) ), you can use the following command to convert:
 
 ```
-python3 tools/export_model.py -c configs/rec/rec_vitstr.yml -o Global.pretrained_model=./rec_vitstr_train/best_accuracy  Global.save_inference_dir=./inference/rec_vitstr
+python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy  Global.save_inference_dir=./inference/rec_vitstr
 ```
 
 **Note:**
@@ -98,7 +98,7 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png'
 After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
 The result is as follows:
 ```shell
-Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617)
 ```
 
 <a name="4-2"></a>
diff --git a/ppocr/losses/__init__.py b/ppocr/losses/__init__.py
index 6c4545eb21..7bea87f62f 100755
--- a/ppocr/losses/__init__.py
+++ b/ppocr/losses/__init__.py
@@ -30,7 +30,7 @@ from .det_fce_loss import FCELoss
 from .rec_ctc_loss import CTCLoss
 from .rec_att_loss import AttentionLoss
 from .rec_srn_loss import SRNLoss
-from .rec_ce_smooth_loss import CESmoothingLoss
+from .rec_ce_loss import CELoss
 from .rec_sar_loss import SARLoss
 from .rec_aster_loss import AsterLoss
 from .rec_pren_loss import PRENLoss
@@ -60,9 +60,8 @@ def build_loss(config):
     support_dict = [
         'DBLoss', 'PSELoss', 'EASTLoss', 'SASTLoss', 'FCELoss', 'CTCLoss',
         'ClsLoss', 'AttentionLoss', 'SRNLoss', 'PGLoss', 'CombinedLoss',
-        'CESmoothingLoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss',
-        'SDMGRLoss', 'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss',
-        'MultiLoss'
+        'CELoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', 'SDMGRLoss',
+        'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', 'MultiLoss'
     ]
     config = copy.deepcopy(config)
     module_name = config.pop('name')
diff --git a/ppocr/losses/rec_ce_smooth_loss.py b/ppocr/losses/rec_ce_loss.py
similarity index 93%
rename from ppocr/losses/rec_ce_smooth_loss.py
rename to ppocr/losses/rec_ce_loss.py
index 22243ed41f..b837ac27bb 100644
--- a/ppocr/losses/rec_ce_smooth_loss.py
+++ b/ppocr/losses/rec_ce_loss.py
@@ -3,9 +3,9 @@ from paddle import nn
 import paddle.nn.functional as F
 
 
-class CESmoothingLoss(nn.Layer):
+class CELoss(nn.Layer):
     def __init__(self, smoothing=True, with_all=False, **kwargs):
-        super(CESmoothingLoss, self).__init__()
+        super(CELoss, self).__init__()
         self.loss_func = nn.CrossEntropyLoss(reduction='mean', ignore_index=0)
         self.smoothing = smoothing
         self.with_all = with_all
diff --git a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
index 3936ab58ad..ba6728d836 100644
--- a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
+++ b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
@@ -49,7 +49,7 @@ Architecture:
     
 
 Loss:
-  name: CESmoothingLoss
+  name: CELoss
   smoothing: True
 
 PostProcess:
diff --git a/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt b/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt
index de6de5a0ca..20d984e38b 100644
--- a/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt
+++ b/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt
@@ -40,10 +40,10 @@ infer_quant:False
 inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,32,100" --rec_algorithm="NRTR"
 --use_gpu:True|False
 --enable_mkldnn:True|False
---cpu_threads:1|6
+--cpu_threads:6
 --rec_batch_num:1|6
 --use_tensorrt:True|False
---precision:fp32|int8
+--precision:fp32
 --rec_model_dir:
 --image_dir:./inference/rec_inference
 --save_log_path:./test/output/
diff --git a/test_tipc/configs/rec_vitstr/rec_vitstr.yml b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
similarity index 97%
rename from test_tipc/configs/rec_vitstr/rec_vitstr.yml
rename to test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
index 427bce4b5a..ccd93c9eaf 100644
--- a/test_tipc/configs/rec_vitstr/rec_vitstr.yml
+++ b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
@@ -3,7 +3,7 @@ Global:
   epoch_num: 20
   log_smooth_window: 20
   print_batch_step: 10
-  save_model_dir: ./output/rec/vitstr/
+  save_model_dir: ./output/rec/vitstr_none_ce/
   save_epoch_step: 1
   # evaluation is run every 2000 iterations after the 0th iteration#
   eval_batch_step: [0, 2000]
@@ -43,7 +43,7 @@ Architecture:
     name: CTCHead
 
 Loss:
-  name: CESmoothingLoss
+  name: CELoss
   smoothing: False
   with_all: True
 
diff --git a/test_tipc/configs/rec_vitstr/train_infer_python.txt b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
similarity index 82%
rename from test_tipc/configs/rec_vitstr/train_infer_python.txt
rename to test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
index 8b58e220f7..04c5742ea2 100644
--- a/test_tipc/configs/rec_vitstr/train_infer_python.txt
+++ b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
@@ -13,7 +13,7 @@ train_infer_img_dir:./inference/rec_inference
 null:null
 ##
 trainer:norm_train
-norm_train:tools/train.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+norm_train:tools/train.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,21 +21,21 @@ null:null
 null:null
 ##
 ===========================eval_params===========================
-eval:tools/eval.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+eval:tools/eval.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
 null:null
 ##
 ===========================infer_params===========================
 Global.save_inference_dir:./output/
 Global.checkpoints:
-norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
 quant_export:null
 fpgm_export:null
 distill_export:null
 export1:null
 export2:null
 ##
-train_model:./inference/rec_vitstr_train/best_accuracy
-infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr/rec_vitstr.yml -o
+train_model:./inference/rec_vitstr_none_ce_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
 infer_quant:False
 inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="ViTSTR"
 --use_gpu:True|False

From 275c5a2ba93d0791e7b8009e3a595f4ac7554989 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Mon, 30 May 2022 08:38:07 +0000
Subject: [PATCH 009/134] modify vitstr config batchsize

---
 configs/rec/rec_vitstr_none_ce.yml                          | 2 +-
 test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/configs/rec/rec_vitstr_none_ce.yml b/configs/rec/rec_vitstr_none_ce.yml
index 065cc785f8..c71eba0cc2 100644
--- a/configs/rec/rec_vitstr_none_ce.yml
+++ b/configs/rec/rec_vitstr_none_ce.yml
@@ -75,7 +75,7 @@ Train:
     shuffle: True
     batch_size_per_card: 48
     drop_last: True
-    num_workers: 2
+    num_workers: 8
 
 Eval:
   dataset:
diff --git a/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
index ccd93c9eaf..4e8151f599 100644
--- a/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
+++ b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
@@ -75,7 +75,7 @@ Train:
     shuffle: True
     batch_size_per_card: 48
     drop_last: True
-    num_workers: 2
+    num_workers: 8
 
 Eval:
   dataset:

From 44086577ebd8fd73aa5dc7bc137ce31b9c7399b3 Mon Sep 17 00:00:00 2001
From: an1018 <614803115@qq.com>
Date: Fri, 10 Jun 2022 19:56:45 +0800
Subject: [PATCH 010/134] add pcb

---
 applications/PCB字符识别/PCB字符识别.md       | 648 ++++++++++++++++++
 .../PCB字符识别/gen_data/background/bg.jpg    | Bin 0 -> 2071 bytes
 .../PCB字符识别/gen_data/corpus/text.txt      |  30 +
 .../PCB字符识别/gen_data/det_background/1.png | Bin 0 -> 145 bytes
 .../PCB字符识别/gen_data/det_background/2.png | Bin 0 -> 141 bytes
 applications/PCB字符识别/gen_data/gen.py      | 261 +++++++
 6 files changed, 939 insertions(+)
 create mode 100644 applications/PCB字符识别/PCB字符识别.md
 create mode 100644 applications/PCB字符识别/gen_data/background/bg.jpg
 create mode 100644 applications/PCB字符识别/gen_data/corpus/text.txt
 create mode 100644 applications/PCB字符识别/gen_data/det_background/1.png
 create mode 100644 applications/PCB字符识别/gen_data/det_background/2.png
 create mode 100644 applications/PCB字符识别/gen_data/gen.py

diff --git a/applications/PCB字符识别/PCB字符识别.md b/applications/PCB字符识别/PCB字符识别.md
new file mode 100644
index 0000000000..a5052e2897
--- /dev/null
+++ b/applications/PCB字符识别/PCB字符识别.md
@@ -0,0 +1,648 @@
+# 基于PP-OCRv3的PCB字符识别
+
+- [1. 项目介绍](#1-项目介绍)
+- [2. 安装说明](#2-安装说明)
+- [3. 数据准备](#3-数据准备)
+- [4. 文本检测](#4-文本检测)
+  - [4.1 预训练模型直接评估](#41-预训练模型直接评估)
+  - [4.2 预训练模型+验证集padding直接评估](#42-预训练模型验证集padding直接评估)
+  - [4.3 预训练模型+fine-tune](#43-预训练模型fine-tune)
+- [5. 文本识别](#5-文本识别)
+  - [5.1 预训练模型直接评估](#51-预训练模型直接评估)
+  - [5.2 三种fine-tune方案](#52-三种fine-tune方案)
+- [6. 模型导出](#6-模型导出)
+- [7. 端对端评测](#7-端对端评测)
+- [8. Jetson部署](#8-Jetson部署)
+- [9. 总结](#9-总结)
+- [更多资源](#更多资源)
+
+# 1. 项目介绍
+
+印刷电路板(PCB)是电子产品中的核心器件，对于板件质量的测试与监控是生产中必不可少的环节。在一些场景中，通过PCB中信号灯颜色和文字组合可以定位PCB局部模块质量问题，PCB文字识别中存在如下难点：
+
+- 裁剪出的PCB图片宽高比例较小
+- 文字区域整体面积也较小
+- 包含垂直、水平多种方向文本
+
+针对本场景，PaddleOCR基于全新的PP-OCRv3通过合成数据、微调以及其他场景适配方法完成小字符文本识别任务，满足企业上线要求。PCB检测、识别效果如 **图1** 所示：
+
+<div align=center><img src='https://ai-studio-static-online.cdn.bcebos.com/95d8e95bf1ab476987f2519c0f8f0c60a0cdc2c444804ed6ab08f2f7ab054880', width='500'></div>
+<div align=center>图1 PCB检测识别效果</div>
+
+注：欢迎在AIStudio领取免费算力体验线上实训，项目链接: [基于PP-OCRv3实现PCB字符识别](https://aistudio.baidu.com/aistudio/projectdetail/4008973)
+
+# 2. 安装说明
+
+
+下载PaddleOCR源码，安装依赖环境。
+
+
+```python
+# 如仍需安装or安装更新，可以执行以下步骤
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+#  git clone https://gitee.com/PaddlePaddle/PaddleOCR
+```
+
+
+```python
+# 安装依赖包
+pip install -r /home/aistudio/PaddleOCR/requirements.txt
+```
+
+# 3. 数据准备
+
+我们通过图片合成工具生成 **图2** 所示的PCB图片，整图只有高25、宽150左右、文字区域高9、宽45左右，包含垂直和水平2种方向的文本：
+
+<div align=center><img src="https://ai-studio-static-online.cdn.bcebos.com/bb7a345687814a3d83a29790f2a2b7d081495b3a920b43988c93da6039cad653" width="1000" ></div>
+<div align=center>图2 数据集示例</div>
+
+暂时不开源生成的PCB数据集，但是通过更换背景，通过如下代码生成数据即可：
+
+```
+cd gen_data
+python3 gen.py --num_img=10
+```
+
+生成图片参数解释：
+
+```
+num_img：生成图片数量
+font_min_size、font_max_size：字体最大、最小尺寸
+bg_path：文字区域背景存放路径
+det_bg_path：整图背景存放路径
+fonts_path：字体路径
+corpus_path：语料路径
+output_dir：生成图片存储路径
+```
+
+这里生成 **100张** 相同尺寸和文本的图片，如  **图3** 所示，方便大家跑通实验。通过如下代码解压数据集：
+
+<div align=center><img src="https://ai-studio-static-online.cdn.bcebos.com/3277b750159f4b68b2b58506bfec9005d49aeb5fb1d9411e83f96f9ff7eb66a5" width="1000" ></div>
+<div align=center>图3 案例提供数据集示例</div>
+
+
+```python
+tar xf ./data/data148165/dataset.tar -C ./
+```
+
+在生成数据集的时需要生成检测和识别训练需求的格式：
+
+
+- **文本检测**
+
+标注文件格式如下，中间用'\t'分隔：
+
+```
+" 图像文件名                    json.dumps编码的图像标注信息"
+ch4_test_images/img_61.jpg    [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
+```
+
+json.dumps编码前的图像标注信息是包含多个字典的list，字典中的 `points` 表示文本框的四个点的坐标(x, y)，从左上角的点开始顺时针排列。 `transcription` 表示当前文本框的文字，***当其内容为“###”时，表示该文本框无效，在训练时会跳过。***
+
+- **文本识别**
+
+标注文件的格式如下， txt文件中默认请将图片路径和图片标签用'\t'分割，如用其他方式分割将造成训练报错。
+
+```
+" 图像文件名                 图像标注信息 "
+
+train_data/rec/train/word_001.jpg   简单可依赖
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
+...
+```
+
+
+# 4. 文本检测
+
+选用飞桨OCR开发套件[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)中的PP-OCRv3模型进行文本检测和识别。针对检测模型和识别模型，进行了共计9个方面的升级：
+
+- PP-OCRv3检测模型对PP-OCRv2中的CML协同互学习文本检测蒸馏策略进行了升级，分别针对教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+
+- PP-OCRv3的识别模块是基于文本识别算法SVTR优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。PP-OCRv3通过轻量级文本识别网络SVTR_LCNet、Attention损失指导CTC损失训练策略、挖掘文字上下文信息的数据增广策略TextConAug、TextRotNet自监督预训练模型、UDML联合互学习策略、UIM无标注数据挖掘方案，6个方面进行模型加速和效果提升。
+
+更多细节请参考PP-OCRv3[技术报告](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md)。
+
+
+我们使用 **3种方案** 进行检测模型的训练、评估：
+-  **PP-OCRv3英文超轻量检测预训练模型直接评估**
+-  PP-OCRv3英文超轻量检测预训练模型 + **验证集padding**直接评估
+-  PP-OCRv3英文超轻量检测预训练模型 + **fine-tune**
+
+## **4.1 预训练模型直接评估**
+
+我们首先通过PaddleOCR提供的预训练模型在验证集上进行评估，如果评估指标能满足效果，可以直接使用预训练模型，不再需要训练。
+
+使用预训练模型直接评估步骤如下：
+
+**1）下载预训练模型**
+
+
+PaddleOCR已经提供了PP-OCR系列模型，部分模型展示如下表所示：
+
+| 模型简介                              | 模型名称                | 推荐场景        | 检测模型                                                     | 方向分类器                                                   | 识别模型                                                     |
+| ------------------------------------- | ----------------------- | --------------- | ------------------------------------------------------------ | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| 中英文超轻量PP-OCRv3模型（16.2M）     | ch_PP-OCRv3_xx          | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+| 英文超轻量PP-OCRv3模型（13.4M）       | en_PP-OCRv3_xx          | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+| 中英文超轻量PP-OCRv2模型（13.0M）     | ch_PP-OCRv2_xx          | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar) |
+| 中英文超轻量PP-OCR mobile模型（9.4M） | ch_ppocr_mobile_v2.0_xx | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
+| 中英文通用PP-OCR server模型（143.4M） | ch_ppocr_server_v2.0_xx | 服务器端        | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
+
+更多模型下载（包括多语言），可以参[考PP-OCR系列模型下载](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md)
+
+这里我们使用PP-OCRv3英文超轻量检测模型，下载并解压预训练模型：
+
+
+
+
+```python
+# 如果更换其他模型，更新下载链接和解压指令就可以
+cd /home/aistudio/PaddleOCR
+mkdir pretrain_models
+cd pretrain_models
+# 下载英文预训练模型
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar
+tar xf en_PP-OCRv3_det_distill_train.tar && rm -rf en_PP-OCRv3_det_distill_train.tar
+%cd ..
+```
+
+**模型评估**
+
+
+首先修改配置文件`configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml`中的以下字段：
+```
+Eval.dataset.data_dir：指向验证集图片存放目录,'/home/aistudio/dataset'
+Eval.dataset.label_file_list：指向验证集标注文件,'/home/aistudio/dataset/det_gt_val.txt'
+Eval.dataset.transforms.DetResizeForTest:  尺寸
+        limit_side_len: 48
+        limit_type: 'min'
+```
+
+然后在验证集上进行评估，具体代码如下：
+
+
+
+```python
+cd /home/aistudio/PaddleOCR
+python tools/eval.py \
+    -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml  \
+    -o Global.checkpoints="./pretrain_models/en_PP-OCRv3_det_distill_train/best_accuracy"
+```
+
+## **4.2 预训练模型+验证集padding直接评估**
+
+考虑到PCB图片比较小，宽度只有25左右、高度只有140-170左右，我们在原图的基础上进行padding，再进行检测评估，padding前后效果对比如 **图4** 所示：
+
+<div align=center><img src='https://ai-studio-static-online.cdn.bcebos.com/e61e6ba685534eda992cea30a63a9c461646040ffd0c4d208a5eebb85897dcf7' width='600'></div>
+<div align=center>图4 padding前后对比图</div>
+
+将图片都padding到300*300大小，因为坐标信息发生了变化，我们同时要修改标注文件，在`/home/aistudio/dataset`目录里也提供了padding之后的图片，大家也可以尝试训练和评估：
+
+同上，我们需要修改配置文件`configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml`中的以下字段：
+```
+Eval.dataset.data_dir：指向验证集图片存放目录,'/home/aistudio/dataset'
+Eval.dataset.label_file_list：指向验证集标注文件,/home/aistudio/dataset/det_gt_padding_val.txt
+Eval.dataset.transforms.DetResizeForTest:  尺寸
+        limit_side_len: 1100
+        limit_type: 'min'
+```
+
+然后执行评估代码
+
+
+```python
+cd /home/aistudio/PaddleOCR
+python tools/eval.py \
+    -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml  \
+    -o Global.checkpoints="./pretrain_models/en_PP-OCRv3_det_distill_train/best_accuracy"
+```
+
+## **4.3 预训练模型+fine-tune**
+
+
+基于预训练模型，在生成的1500图片上进行fine-tune训练和评估，其中train数据1200张，val数据300张，修改配置文件`configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml`中的以下字段：
+```
+Global.epoch_num: 这里设置为1，方便快速跑通，实际中根据数据量调整该值
+Global.save_model_dir：模型保存路径
+Global.pretrained_model：指向预训练模型路径，'./pretrain_models/en_PP-OCRv3_det_distill_train/student.pdparams'
+Optimizer.lr.learning_rate：调整学习率，本实验设置为0.0005
+Train.dataset.data_dir：指向训练集图片存放目录,'/home/aistudio/dataset'
+Train.dataset.label_file_list：指向训练集标注文件,'/home/aistudio/dataset/det_gt_train.txt'
+Train.dataset.transforms.EastRandomCropData.size：训练尺寸改为[480,64]
+Eval.dataset.data_dir：指向验证集图片存放目录,'/home/aistudio/dataset/'
+Eval.dataset.label_file_list：指向验证集标注文件,'/home/aistudio/dataset/det_gt_val.txt'
+Eval.dataset.transforms.DetResizeForTest：评估尺寸，添加如下参数
+    limit_side_len: 64
+    limit_type:'min'
+```
+执行下面命令启动训练：
+
+
+```python
+cd /home/aistudio/PaddleOCR/
+python tools/train.py \
+        -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml
+```
+
+**模型评估**
+
+
+使用训练好的模型进行评估，更新模型路径`Global.checkpoints`:
+
+
+```python
+cd /home/aistudio/PaddleOCR/
+python3 tools/eval.py \
+    -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml  \
+    -o Global.checkpoints="./output/ch_PP-OCR_V3_det/latest"
+```
+
+使用训练好的模型进行评估，指标如下所示：
+
+
+| 序号 | 方案 | hmean  |  效果提升  |   实验分析  |
+| -------- | -------- | -------- | -------- | -------- |
+|   1 |  PP-OCRv3英文超轻量检测预训练模型   | 64.64%     |     -     |    提供的预训练模型具有泛化能力       |
+|   2 | PP-OCRv3英文超轻量检测预训练模型 + 验证集padding    |  72.13%  |+7.5% | padding可以提升尺寸较小图片的检测效果|
+|   3 | PP-OCRv3英文超轻量检测预训练模型  + fine-tune   | 100% |  +27.9%     | fine-tune会提升垂类场景效果 |
+
+
+```
+注：上述实验结果均是在1500张图片（1200张训练集，300张测试集）上训练、评估的得到，AIstudio只提供了100张数据，所以指标有所差异属于正常，只要策略有效、规律相同即可。
+```
+
+# 5. 文本识别
+
+我们分别使用如下4种方案进行训练、评估：
+
+- **方案1**：**PP-OCRv3中英文超轻量识别预训练模型直接评估**
+- **方案2**：PP-OCRv3中英文超轻量检测预训练模型 + **fine-tune**
+- **方案3**：PP-OCRv3中英文超轻量检测预训练模型 + fine-tune + **公开通用识别数据集**
+- **方案4**：PP-OCRv3中英文超轻量检测预训练模型 + fine-tune + **增加PCB图像数量**
+
+
+## **5.1 预训练模型直接评估**
+
+同检测模型，我们首先使用PaddleOCR提供的识别预训练模型在PCB验证集上进行评估。
+
+使用预训练模型直接评估步骤如下：
+
+**1）下载预训练模型**
+
+
+我们使用PP-OCRv3中英文超轻量文本识别模型，下载并解压预训练模型：
+
+
+```python
+# 如果更换其他模型，更新下载链接和解压指令就可以
+cd /home/aistudio/PaddleOCR/pretrain_models/
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+tar xf ch_PP-OCRv3_rec_train.tar && rm -rf ch_PP-OCRv3_rec_train.tar
+cd ..
+```
+
+**模型评估**
+
+
+首先修改配置文件`configs/det/ch_PP-OCRv3/ch_PP-OCRv2_rec_distillation.yml`中的以下字段：
+
+```
+Metric.ignore_space: True：忽略空格
+Eval.dataset.data_dir：指向验证集图片存放目录,'/home/aistudio/dataset'
+Eval.dataset.label_file_list：指向验证集标注文件,'/home/aistudio/dataset/rec_gt_val.txt'
+```
+
+我们使用下载的预训练模型进行评估：
+
+
+```python
+cd /home/aistudio/PaddleOCR
+python3 tools/eval.py \
+    -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
+    -o Global.checkpoints=pretrain_models/ch_PP-OCRv3_rec_train/best_accuracy
+
+```
+
+## **5.2 三种fine-tune方案**
+
+方案2、3、4训练和评估方式是相同的，因此在我们了解每个技术方案之后，再具体看修改哪些参数是相同，哪些是不同的。
+
+**方案介绍：**
+
+1） **方案2**：预训练模型 + **fine-tune**
+
+- 在预训练模型的基础上进行fine-tune，使用1500张PCB进行训练和评估，其中训练集1200张，验证集300张。
+
+
+2） **方案3**：预训练模型 + fine-tune + **公开通用识别数据集**
+
+- 当识别数据比较少的情况，可以考虑添加公开通用识别数据集。在方案2的基础上，添加公开通用识别数据集，如lsvt、rctw等。
+
+3）**方案4**：预训练模型 + fine-tune + **增加PCB图像数量**
+
+- 如果能够获取足够多真实场景，我们可以通过增加数据量提升模型效果。在方案2的基础上，增加PCB的数量到2W张左右。
+
+
+**参数修改：**
+
+接着我们看需要修改的参数，以上方案均需要修改配置文件`configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml`的参数，**修改一次即可**：
+
+```
+Global.pretrained_model：指向预训练模型路径,'pretrain_models/ch_PP-OCRv3_rec_train/best_accuracy'
+Optimizer.lr.values：学习率，本实验设置为0.0005
+Train.loader.batch_size_per_card: batch size,默认128，因为数据量小于128，因此我们设置为8，数据量大可以按默认的训练
+Eval.loader.batch_size_per_card: batch size,默认128，设置为4
+Metric.ignore_space: 忽略空格，本实验设置为True
+```
+
+**更换不同的方案**每次需要修改的参数：
+```
+Global.epoch_num: 这里设置为1，方便快速跑通，实际中根据数据量调整该值
+Global.save_model_dir：指向模型保存路径
+Train.dataset.data_dir：指向训练集图片存放目录
+Train.dataset.label_file_list：指向训练集标注文件
+Eval.dataset.data_dir：指向验证集图片存放目录
+Eval.dataset.label_file_list：指向验证集标注文件
+```
+
+同时**方案3**修改以下参数
+```
+Eval.dataset.label_file_list：添加公开通用识别数据标注文件
+Eval.dataset.ratio_list：数据和公开通用识别数据每次采样比例，按实际修改即可
+```
+如 **图5** 所示：
+<div align=center><img src='https://ai-studio-static-online.cdn.bcebos.com/0fa18b25819042d9bbf3397c3af0e21433b23d52f7a84b0a8681b8e6a308d433' wdith=''></div>
+<div align=center>图5 添加公开通用识别数据配置文件示例</div>
+
+
+我们提取Student模型的参数，在PCB数据集上进行fine-tune，可以参考如下代码：
+
+
+```python
+import paddle
+# 加载预训练模型
+all_params = paddle.load("./pretrain_models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 学生模型的权重提取
+s_params = {key[len("student_model."):]: all_params[key] for key in all_params if "student_model." in key}
+# 查看学生模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/ch_PP-OCRv3_rec_train/student.pdparams")
+```
+
+修改参数后，**每个方案**都执行如下命令启动训练：
+
+
+
+```python
+cd /home/aistudio/PaddleOCR/
+python3 tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml
+```
+
+
+使用训练好的模型进行评估，更新模型路径`Global.checkpoints`：
+
+
+```python
+cd /home/aistudio/PaddleOCR/
+python3 tools/eval.py \
+    -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml \
+    -o Global.checkpoints=./output/rec_ppocr_v3/latest
+```
+
+所有方案评估指标如下：
+
+| 序号 | 方案 | acc    |  效果提升  |   实验分析  |
+| -------- | -------- | -------- | -------- | -------- |
+|   1 | PP-OCRv3中英文超轻量识别预训练模型直接评估 | 46.67%     |     -     |    提供的预训练模型具有泛化能力       |
+|   2 | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune   |  42.02%  |-4.6% | 在数据量不足的情况，反而比预训练模型效果低(也可以通过调整超参数再试试)|
+|   3 | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune + 公开通用识别数据集   | 77% |  +30%     | 在数据量不足的情况下，可以考虑补充公开数据训练 |
+|   4 | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune + 增加PCB图像数量   | 99.99% |  +23%     | 如果能获取更多数据量的情况，可以通过增加数据量提升效果 |
+
+```
+注：上述实验结果均是在1500张图片（1200张训练集，300张测试集）、2W张图片、添加公开通用识别数据集上训练、评估的得到，AIstudio只提供了100张数据，所以指标有所差异属于正常，只要策略有效、规律相同即可。
+```
+
+# 6. 模型导出
+
+inference 模型（paddle.jit.save保存的模型） 一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。 训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。 与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+
+```python
+# 导出检测模型
+python3 tools/export_model.py \
+     -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml \
+     -o Global.pretrained_model="./output/ch_PP-OCR_V3_det/latest" \
+     Global.save_inference_dir="./inference_model/ch_PP-OCR_V3_det/"
+```
+
+因为上述模型只训练了1个epoch，因此我们使用训练最优的模型进行预测，存储在`/home/aistudio/best_models/`目录下，解压即可
+
+
+```python
+cd /home/aistudio/best_models/
+wget https://paddleocr.bj.bcebos.com/fanliku/PCB/det_ppocr_v3_en_infer_PCB.tar
+tar xf /home/aistudio/best_models/det_ppocr_v3_en_infer_PCB.tar -C /home/aistudio/PaddleOCR/pretrain_models/
+```
+
+
+```python
+# 检测模型inference模型预测
+cd /home/aistudio/PaddleOCR/
+python3 tools/infer/predict_det.py \
+    --image_dir="/home/aistudio/dataset/imgs/0000.jpg" \
+    --det_algorithm="DB" \
+    --det_model_dir="./pretrain_models/det_ppocr_v3_en_infer_PCB/" \
+    --det_limit_side_len=48 \
+    --det_limit_type='min' \
+    --det_db_unclip_ratio=2.5 \
+    --use_gpu=True
+```
+
+结果存储在`inference_results`目录下，检测如下图所示：
+<div align=center><img src='https://ai-studio-static-online.cdn.bcebos.com/5939ae15a1f0445aaeec15c68107dbd897740a1ddd284bf8b583bb6242099157' width=''></div>
+<div align=center>图6 检测结果</div>
+
+
+同理，导出识别模型并进行推理。
+
+```python
+# 导出识别模型
+python3 tools/export_model.py \
+    -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml \
+    -o Global.pretrained_model="./output/rec_ppocr_v3/latest" \
+    Global.save_inference_dir="./inference_model/rec_ppocr_v3/"
+
+```
+
+同检测模型，识别模型也只训练了1个epoch，因此我们使用训练最优的模型进行预测，存储在`/home/aistudio/best_models/`目录下，解压即可
+
+
+```python
+cd /home/aistudio/best_models/
+wget https://paddleocr.bj.bcebos.com/fanliku/PCB/rec_ppocr_v3_ch_infer_PCB.tar
+tar xf /home/aistudio/best_models/rec_ppocr_v3_ch_infer_PCB.tar -C /home/aistudio/PaddleOCR/pretrain_models/
+```
+
+
+```python
+# 识别模型inference模型预测
+cd /home/aistudio/PaddleOCR/
+python3 tools/infer/predict_rec.py \
+    --image_dir="../test_imgs/0000_rec.jpg" \
+    --rec_model_dir="./pretrain_models/rec_ppocr_v3_ch_infer_PCB" \
+    --rec_image_shape="3, 48, 320" \
+    --use_space_char=False \
+    --use_gpu=True
+```
+
+```python
+# 检测+识别模型inference模型预测
+cd /home/aistudio/PaddleOCR/
+python3 tools/infer/predict_system.py  \
+    --image_dir="../test_imgs/0000.jpg" \
+    --det_model_dir="./pretrain_models/det_ppocr_v3_en_infer_PCB" \
+    --det_limit_side_len=48 \
+    --det_limit_type='min' \
+    --det_db_unclip_ratio=2.5 \
+    --rec_model_dir="./pretrain_models/rec_ppocr_v3_ch_infer_PCB"  \
+    --rec_image_shape="3, 48, 320" \
+    --draw_img_save_dir=./det_rec_infer/ \
+    --use_space_char=False \
+    --use_angle_cls=False \
+    --use_gpu=True
+
+```
+
+端到端预测结果存储在`det_res_infer`文件夹内，结果如下图所示：
+<div align=center><img src='https://ai-studio-static-online.cdn.bcebos.com/c570f343c29846c792da56ebaca16c50708477514dd048cea8bef37ffa85d03f'></div>
+<div align=center>图7 检测+识别结果</div>
+
+# 7. 端对端评测
+
+接下来介绍文本检测+文本识别的端对端指标评估方式。主要分为三步：
+
+1）首先运行`tools/infer/predict_system.py`，将`image_dir`改为需要评估的数据文件家，得到保存的结果:
+
+
+```python
+# 检测+识别模型inference模型预测
+python3 tools/infer/predict_system.py  \
+    --image_dir="../dataset/imgs/" \
+    --det_model_dir="./pretrain_models/det_ppocr_v3_en_infer_PCB" \
+    --det_limit_side_len=48 \
+    --det_limit_type='min' \
+    --det_db_unclip_ratio=2.5 \
+    --rec_model_dir="./pretrain_models/rec_ppocr_v3_ch_infer_PCB"  \
+    --rec_image_shape="3, 48, 320" \
+    --draw_img_save_dir=./det_rec_infer/ \
+    --use_space_char=False \
+    --use_angle_cls=False \
+    --use_gpu=True
+```
+
+得到保存结果，文本检测识别可视化图保存在`det_rec_infer/`目录下，预测结果保存在`det_rec_infer/system_results.txt`中，格式如下：`0018.jpg	[{"transcription": "E295", "points": [[88, 33], [137, 33], [137, 40], [88, 40]]}]`
+
+2）然后将步骤一保存的数据转换为端对端评测需要的数据格式： 修改 `tools/end2end/convert_ppocr_label.py`中的代码，convert_label函数中设置输入标签路径，Mode，保存标签路径等，对预测数据的GTlabel和预测结果的label格式进行转换。
+```
+ppocr_label_gt =  "/home/aistudio/dataset/det_gt_val.txt"
+convert_label(ppocr_label_gt, "gt", "./save_gt_label/")
+
+ppocr_label_gt =  "/home/aistudio/PaddleOCR/PCB_result/det_rec_infer/system_results.txt"
+convert_label(ppocr_label_gt, "pred", "./save_PPOCRV2_infer/")
+```
+
+运行`convert_ppocr_label.py`:
+
+
+```python
+ python3 tools/end2end/convert_ppocr_label.py
+```
+
+得到如下结果：
+```
+├── ./save_gt_label/
+├── ./save_PPOCRV2_infer/
+```
+
+3） 最后，执行端对端评测，运行`tools/end2end/eval_end2end.py`计算端对端指标，运行方式如下：
+
+
+```python
+pip install editdistance
+python3 tools/end2end/eval_end2end.py ./save_gt_label/ ./save_PPOCRV2_infer/
+```
+
+使用`预训练模型+fine-tune'检测模型`、`预训练模型 + 2W张PCB图片funetune`识别模型，在300张PCB图片上评估得到如下结果，fmeasure为主要关注的指标:
+<div align=center><img src='https://ai-studio-static-online.cdn.bcebos.com/37206ea48a244212ae7a821d50d1fd51faf3d7fe97ac47a29f04dfcbb377b019', width='700'></div>
+<div align=center>图8 端到端评估指标</div>
+
+```
+注: 使用上述命令不能跑出该结果，因为数据集不相同，可以更换为自己训练好的模型，按上述流程运行
+```
+
+# 8. Jetson部署
+
+我们只需要以下步骤就可以完成Jetson nano部署模型，简单易操作：
+
+**1、在Jetson nano开发版上环境准备：**
+
+* 安装PaddlePaddle
+
+* 下载PaddleOCR并安装依赖
+
+**2、执行预测**
+
+* 将推理模型下载到jetson
+
+* 执行检测、识别、串联预测即可
+
+详细[参考流程](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/deploy/Jetson/readme_ch.md)。
+
+# 9. 总结
+
+检测实验分别使用PP-OCRv3预训练模型在PCB数据集上进行了直接评估、验证集padding、 fine-tune 3种方案，识别实验分别使用PP-OCRv3预训练模型在PCB数据集上进行了直接评估、 fine-tune、添加公开通用识别数据集、增加PCB图片数量4种方案，指标对比如下：
+
+* 检测
+
+
+| 序号 | 方案                                                     | hmean  | 效果提升 | 实验分析                              |
+| ---- | -------------------------------------------------------- | ------ | -------- | ------------------------------------- |
+| 1    | PP-OCRv3英文超轻量检测预训练模型直接评估                 | 64.64% | -        | 提供的预训练模型具有泛化能力          |
+| 2    | PP-OCRv3英文超轻量检测预训练模型 + 验证集padding直接评估 | 72.13% | +7.5%    | padding可以提升尺寸较小图片的检测效果 |
+| 3    | PP-OCRv3英文超轻量检测预训练模型  + fine-tune            | 100%   | +27.9%   | fine-tune会提升垂类场景效果           |
+
+* 识别
+
+| 序号 | 方案                                                         | acc    | 效果提升 | 实验分析                                                     |
+| ---- | ------------------------------------------------------------ | ------ | -------- | ------------------------------------------------------------ |
+| 1    | PP-OCRv3中英文超轻量识别预训练模型直接评估                   | 46.67% | -        | 提供的预训练模型具有泛化能力                                 |
+| 2    | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune               | 42.02% | -4.6%    | 在数据量不足的情况，反而比预训练模型效果低(也可以通过调整超参数再试试) |
+| 3    | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune + 公开通用识别数据集 | 77%    | +30%     | 在数据量不足的情况下，可以考虑补充公开数据训练               |
+| 4    | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune + 增加PCB图像数量 | 99.99% | +23%     | 如果能获取更多数据量的情况，可以通过增加数据量提升效果       |
+
+* 端到端
+
+| det                                           | rec                                                          | fmeasure |
+| --------------------------------------------- | ------------------------------------------------------------ | -------- |
+| PP-OCRv3英文超轻量检测预训练模型  + fine-tune | PP-OCRv3中英文超轻量识别预训练模型 + fine-tune + 增加PCB图像数量 | 93.3%    |
+
+*结论*
+
+PP-OCRv3的检测模型在未经过fine-tune的情况下，在PCB数据集上也有64.64%的精度，说明具有泛化能力。验证集padding之后，精度提升7.5%，在图片尺寸较小的情况，我们可以通过padding的方式提升检测效果。经过 fine-tune 后能够极大的提升检测效果，精度达到100%。
+
+PP-OCRv3的识别模型方案1和方案2对比可以发现，当数据量不足的情况，预训练模型精度可能比fine-tune效果还要高，所以我们可以先尝试预训练模型直接评估。如果在数据量不足的情况下想进一步提升模型效果，可以通过添加公开通用识别数据集，识别效果提升30%，非常有效。最后如果我们能够采集足够多的真实场景数据集，可以通过增加数据量提升模型效果，精度达到99.99%。
+
+# 更多资源
+
+- 更多深度学习知识、产业案例、面试宝典等，请参考：[awesome-DeepLearning](https://github.com/paddlepaddle/awesome-DeepLearning)
+
+- 更多PaddleOCR使用教程，请参考：[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR/tree/dygraph)
+
+
+- 飞桨框架相关资料，请参考：[飞桨深度学习平台](https://www.paddlepaddle.org.cn/?fr=paddleEdu_aistudio)
+
+# 参考
+
+* 数据生成代码库：https://github.com/zcswdt/Color_OCR_image_generator
diff --git a/applications/PCB字符识别/gen_data/background/bg.jpg b/applications/PCB字符识别/gen_data/background/bg.jpg
new file mode 100644
index 0000000000000000000000000000000000000000..3cb6eab819c3b7d4f68590d2cdc9d36351590197
GIT binary patch
literal 2071
zcmex=<NpH&0WUXCHwH#V1_nkTWcYuZ!I^=Bjg6g+m4ls~os*M;i${c)hnt&6Qb?Fz
zL{>^(PF6}rMnOeST|r4lSw=>~TvNxu(8R<<Uft5x!pKI?*u)5A2qPyaCl5D|1TU|I
zk)n*E5y{~H0R}-1Mjb{SW=16jCP7AKLB{__803NOWMu>c1}I=;VrF4wW9Q)H;sz?%
zD!{<R$jr>d!pzFb!U9xX3zTPI5o8roG<0MW4oqZMDikqloVbuf*=gfJ(V&YTRE(2~
znmD<{#3dx9RMpfqG__1j&CD$<t(;w4-P}Dqy@EqR!@?sXqmoln)6z3Cvx-Yf%gQS%
ztD0L{+uA!iyCzMZGIiSY88c@sTD)ZGvgIpQuG+MD%hqk%ckJAC=<t!F$Bv&kdFs;T
zD_5^wzj5={!$*&wJbm{3#miS8KYjl4_1pI!KYxMz#mK-6@fMJTcnr;7f<S*Uv9K_+
zu!H=?$W#u*%z`YeiiT`Lj)Clng~CckjT|CQ6Blkg$f;}`^g%SK=pvVxipfLOk07ss
zeMX$en#l4Q++zrT-D2QjW&}navmk>#!}nhriyF8_@n{%~rh(BkFq#GiWg4*me-i-S
C@2wdC

literal 0
HcmV?d00001

diff --git a/applications/PCB字符识别/gen_data/corpus/text.txt b/applications/PCB字符识别/gen_data/corpus/text.txt
new file mode 100644
index 0000000000..8b8cb793ef
--- /dev/null
+++ b/applications/PCB字符识别/gen_data/corpus/text.txt
@@ -0,0 +1,30 @@
+5ZQ
+I4UL
+PWL
+SNOG
+ZL02
+1C30
+O3H
+YHRS
+N03S
+1U5Y
+JTK
+EN4F
+YKJ
+DWNH
+R42W
+X0V
+4OF5
+08AM
+Y93S
+GWE2
+0KR
+9U2A
+DBQ
+Y6J
+ROZ
+K06
+KIEY
+NZQJ
+UN1B
+6X4
\ No newline at end of file
diff --git a/applications/PCB字符识别/gen_data/det_background/1.png b/applications/PCB字符识别/gen_data/det_background/1.png
new file mode 100644
index 0000000000000000000000000000000000000000..8a49eaa6862113044e05d17e32941a0a20911426
GIT binary patch
literal 145
zcmeAS@N?(olHy`uVBq!ia0vp^8-Q4fg9%8^2y$!$QW2gmjv*Cu-d<YBctAnmz=2DD
z%j-mxgd29uR#3`Lv$Objn{m%kffhv$XMr{a9!H5L1C~UNLj(nqn5_+%4b$fQhzHui
N;OXk;vd$@?2>=}dC;|Wg

literal 0
HcmV?d00001

diff --git a/applications/PCB字符识别/gen_data/det_background/2.png b/applications/PCB字符识别/gen_data/det_background/2.png
new file mode 100644
index 0000000000000000000000000000000000000000..c3fcc0c92826b97b5f6abd970f1a0580eede0f5d
GIT binary patch
literal 141
zcmeAS@N?(olHy`uVBq!ia0vp^-9RkC!2~2-J7>BAsSr;W$B>FSZ!bFXGAIZfFgURP
zxigdhw+vC8oAYM6-uNkdpxL>U@2X|U^MJX;GGgEBG3zSK5X}Ud&*16m=d#Wzp$P!_
Caxi58

literal 0
HcmV?d00001

diff --git a/applications/PCB字符识别/gen_data/gen.py b/applications/PCB字符识别/gen_data/gen.py
new file mode 100644
index 0000000000..4c768067f9
--- /dev/null
+++ b/applications/PCB字符识别/gen_data/gen.py
@@ -0,0 +1,261 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/zcswdt/Color_OCR_image_generator
+"""
+import os
+import random
+from PIL import Image, ImageDraw, ImageFont
+import json
+import argparse
+
+
+def get_char_lines(txt_root_path):
+    """
+    desc:get corpus line
+    """
+    txt_files = os.listdir(txt_root_path)
+    char_lines = []
+    for txt in txt_files:
+        f = open(os.path.join(txt_root_path, txt), mode='r', encoding='utf-8')
+        lines = f.readlines()
+        f.close()
+        for line in lines:
+            char_lines.append(line.strip())
+        return char_lines
+
+
+def get_horizontal_text_picture(image_file, chars, fonts_list, cf):
+    """
+    desc:gen horizontal text picture
+    """
+    img = Image.open(image_file)
+    if img.mode != 'RGB':
+        img = img.convert('RGB')
+    img_w, img_h = img.size
+
+    # random choice font
+    font_path = random.choice(fonts_list)
+    # random choice font size
+    font_size = random.randint(cf.font_min_size, cf.font_max_size)
+    font = ImageFont.truetype(font_path, font_size)
+
+    ch_w = []
+    ch_h = []
+    for ch in chars:
+        wt, ht = font.getsize(ch)
+        ch_w.append(wt)
+        ch_h.append(ht)
+    f_w = sum(ch_w)
+    f_h = max(ch_h)
+
+    # add space
+    char_space_width = max(ch_w)
+    f_w += (char_space_width * (len(chars) - 1))
+
+    x1 = random.randint(0, img_w - f_w)
+    y1 = random.randint(0, img_h - f_h)
+    x2 = x1 + f_w
+    y2 = y1 + f_h
+
+    crop_y1 = y1
+    crop_x1 = x1
+    crop_y2 = y2
+    crop_x2 = x2
+
+    best_color = (0, 0, 0)
+    draw = ImageDraw.Draw(img)
+    for i, ch in enumerate(chars):
+        draw.text((x1, y1), ch, best_color, font=font)
+        x1 += (ch_w[i] + char_space_width)
+    crop_img = img.crop((crop_x1, crop_y1, crop_x2, crop_y2))
+    return crop_img, chars
+
+
+def get_vertical_text_picture(image_file, chars, fonts_list, cf):
+    """
+    desc:gen vertical text picture
+    """
+    img = Image.open(image_file)
+    if img.mode != 'RGB':
+        img = img.convert('RGB')
+    img_w, img_h = img.size
+    # random choice font
+    font_path = random.choice(fonts_list)
+    # random choice font size
+    font_size = random.randint(cf.font_min_size, cf.font_max_size)
+    font = ImageFont.truetype(font_path, font_size)
+
+    ch_w = []
+    ch_h = []
+    for ch in chars:
+        wt, ht = font.getsize(ch)
+        ch_w.append(wt)
+        ch_h.append(ht)
+    f_w = max(ch_w)
+    f_h = sum(ch_h)
+
+    x1 = random.randint(0, img_w - f_w)
+    y1 = random.randint(0, img_h - f_h)
+    x2 = x1 + f_w
+    y2 = y1 + f_h
+
+    crop_y1 = y1
+    crop_x1 = x1
+    crop_y2 = y2
+    crop_x2 = x2
+
+    best_color = (0, 0, 0)
+    draw = ImageDraw.Draw(img)
+    i = 0
+    for ch in chars:
+        draw.text((x1, y1), ch, best_color, font=font)
+        y1 = y1 + ch_h[i]
+        i = i + 1
+    crop_img = img.crop((crop_x1, crop_y1, crop_x2, crop_y2))
+    crop_img = crop_img.transpose(Image.ROTATE_90)
+    return crop_img, chars
+
+
+def get_fonts(fonts_path):
+    """
+    desc: get all fonts
+    """
+    font_files = os.listdir(fonts_path)
+    fonts_list=[]
+    for font_file in font_files:
+        font_path=os.path.join(fonts_path, font_file)
+        fonts_list.append(font_path)
+    return fonts_list
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--num_img', type=int, default=30, help="Number of images to generate")
+    parser.add_argument('--font_min_size', type=int, default=11)
+    parser.add_argument('--font_max_size', type=int, default=12,
+                        help="Help adjust the size of the generated text and the size of the picture")
+    parser.add_argument('--bg_path', type=str, default='./background',
+                        help='The generated text pictures will be pasted onto the pictures of this folder')
+    parser.add_argument('--det_bg_path', type=str, default='./det_background',
+                        help='The generated text pictures will use the pictures of this folder as the background')
+    parser.add_argument('--fonts_path', type=str, default='../../StyleText/fonts',
+                        help='The font used to generate the picture')
+    parser.add_argument('--corpus_path', type=str, default='./corpus',
+                        help='The corpus used to generate the text picture')
+    parser.add_argument('--output_dir', type=str, default='./output/', help='Images save dir')
+
+
+    cf = parser.parse_args()
+    # save path
+    if not os.path.exists(cf.output_dir):
+        os.mkdir(cf.output_dir)
+
+    # get corpus
+    txt_root_path = cf.corpus_path
+    char_lines = get_char_lines(txt_root_path=txt_root_path)
+
+    # get all fonts
+    fonts_path = cf.fonts_path
+    fonts_list = get_fonts(fonts_path)
+
+    # rec bg
+    img_root_path = cf.bg_path
+    imnames=os.listdir(img_root_path)
+    
+    # det bg
+    det_bg_path = cf.det_bg_path
+    bg_pics = os.listdir(det_bg_path)
+
+    # OCR det files
+    det_val_file = open(cf.output_dir + 'det_gt_val.txt', 'w', encoding='utf-8')
+    det_train_file = open(cf.output_dir + 'det_gt_train.txt', 'w', encoding='utf-8')
+    # det imgs
+    det_save_dir = 'imgs/'
+    if not os.path.exists(cf.output_dir + det_save_dir):
+        os.mkdir(cf.output_dir + det_save_dir)
+    det_val_save_dir = 'imgs_val/'
+    if not os.path.exists(cf.output_dir + det_val_save_dir):
+        os.mkdir(cf.output_dir + det_val_save_dir)
+
+    # OCR rec files
+    rec_val_file = open(cf.output_dir + 'rec_gt_val.txt', 'w', encoding='utf-8')
+    rec_train_file = open(cf.output_dir + 'rec_gt_train.txt', 'w', encoding='utf-8')
+    # rec imgs
+    rec_save_dir = 'rec_imgs/'
+    if not os.path.exists(cf.output_dir + rec_save_dir):
+        os.mkdir(cf.output_dir + rec_save_dir)
+    rec_val_save_dir = 'rec_imgs_val/'
+    if not os.path.exists(cf.output_dir + rec_val_save_dir):
+        os.mkdir(cf.output_dir + rec_val_save_dir)
+
+
+    val_ratio = cf.num_img * 0.2  # val dataset ratio
+
+    print('start generating...')
+    for i in range(0, cf.num_img):
+        imname = random.choice(imnames)
+        img_path = os.path.join(img_root_path, imname)
+
+        rnd = random.random()
+        # gen horizontal text picture
+        if rnd < 0.5:
+            gen_img, chars = get_horizontal_text_picture(img_path, char_lines[i], fonts_list, cf)
+            ori_w, ori_h = gen_img.size
+            gen_img = gen_img.crop((0, 3, ori_w, ori_h))
+        # gen vertical text picture
+        else:
+            gen_img, chars = get_vertical_text_picture(img_path, char_lines[i], fonts_list, cf)
+            ori_w, ori_h = gen_img.size
+            gen_img = gen_img.crop((3, 0, ori_w, ori_h))
+
+        ori_w, ori_h = gen_img.size
+
+        # rec imgs
+        save_img_name = str(i).zfill(4) + '.jpg'
+        if i < val_ratio:
+            save_dir = os.path.join(rec_val_save_dir, save_img_name)
+            line = save_dir + '\t' + char_lines[i] + '\n'
+            rec_val_file.write(line)
+        else:
+            save_dir = os.path.join(rec_save_dir, save_img_name)
+            line = save_dir + '\t' + char_lines[i] + '\n'
+            rec_train_file.write(line)
+        gen_img.save(cf.output_dir + save_dir, quality = 95, subsampling=0)
+
+        # det img
+        # random choice bg
+        bg_pic = random.sample(bg_pics, 1)[0]
+        det_img = Image.open(os.path.join(det_bg_path, bg_pic))
+        # the PCB position is fixed, modify it according to your own scenario
+        if bg_pic == '1.png':
+            x1 = 38
+            y1 = 3
+        else:
+            x1 = 34
+            y1 = 1
+
+        det_img.paste(gen_img, (x1, y1))
+        # text pos
+        chars_pos = [[x1, y1], [x1 + ori_w, y1], [x1 + ori_w, y1 + ori_h], [x1, y1 + ori_h]]
+        label = [{"transcription":char_lines[i], "points":chars_pos}]
+        if i < val_ratio:
+            save_dir = os.path.join(det_val_save_dir, save_img_name)
+            det_val_file.write(save_dir + '\t' + json.dumps(
+                    label, ensure_ascii=False) + '\n')
+        else:
+            save_dir = os.path.join(det_save_dir, save_img_name)
+            det_train_file.write(save_dir + '\t' + json.dumps(
+                    label, ensure_ascii=False) + '\n')
+        det_img.save(cf.output_dir + save_dir, quality = 95, subsampling=0)

From c71f53ad5f3cf4e0cb997a6ee715fd8cceaad89c Mon Sep 17 00:00:00 2001
From: littletomatodonkey <dazhiningsibuqu@163.com>
Date: Tue, 14 Jun 2022 10:57:40 +0800
Subject: [PATCH 011/134] rm infer fleet (#6577)

---
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 28 ++++++------
 .../docs/test_train_fleet_inference_python.md | 44 +------------------
 2 files changed, 15 insertions(+), 57 deletions(-)

diff --git a/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 08e1fe9ba0..698f1e9849 100644
--- a/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -25,29 +25,29 @@ eval:null
 null:null
 ##
 ===========================infer_params===========================
-Global.save_inference_dir:./output/
+Global.save_inference_dir:null
 Global.checkpoints:
-norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv3_rec/ch_PP-OCRv3_rec_distillation.yml -o 
+norm_export:null
 quant_export:
 fpgm_export: 
 distill_export:null
 export1:null
 export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv3_rec_infer
+inference_dir:null
+infer_model:null
 infer_export:null
-infer_quant:False
-inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1|6
---use_tensorrt:False
---precision:fp32
+infer_quant:
+inference:
+--use_gpu:
+--enable_mkldnn:
+--cpu_threads:
+--rec_batch_num:
+--use_tensorrt:
+--precision:
 --rec_model_dir:
---image_dir:./inference/rec_inference
+--image_dir:
 null:null
---benchmark:True
+--benchmark:
 null:null
 ===========================infer_benchmark_params==========================
 random_infer_input:[{float32,[3,48,320]}]
diff --git a/test_tipc/docs/test_train_fleet_inference_python.md b/test_tipc/docs/test_train_fleet_inference_python.md
index 4479a47da8..2da1170554 100644
--- a/test_tipc/docs/test_train_fleet_inference_python.md
+++ b/test_tipc/docs/test_train_fleet_inference_python.md
@@ -11,12 +11,6 @@ Linux GPU/CPU 多机多卡训练推理测试的主程序为`test_train_inference
 |  PP-OCRv3      | ch_PP-OCRv3_rec     | 分布式训练 |
 
 
-- 推理相关：
-
-| 算法名称 | 模型名称 | device_CPU | device_GPU | batchsize |
-|  :----:   |  :----: |   :----:   |  :----:  |   :----:   |
-|  PP-OCRv3   |  ch_PP-OCRv3_rec |  支持 | 支持 | 1 |
-
 
 ## 2. 测试流程
 
@@ -62,46 +56,10 @@ bash test_tipc/test_train_inference_python.sh  test_tipc/configs/ch_PP-OCRv3_rec
 
 ```bash
  Run successfully with command - ch_PP-OCRv3_rec - python3.7 -m paddle.distributed.launch --ips=192.168.0.1,192.168.0.2 --gpus=0,1 tools/train.py -c test_tipc/configs/ch_PP-OCRv3_rec/ch_PP-OCRv3_rec_distillation.yml -o  Global.use_gpu=True Global.save_model_dir=./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/norm_train_gpus_0,1_autocast_fp32_nodes_2   Global.epoch_num=3 Global.auto_cast=fp32 Train.loader.batch_size_per_card=16    !  
- ......
-  Run successfully with command - ch_PP-OCRv3_rec - python3.7 tools/infer/predict_rec.py --rec_image_shape="3,48,320" --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --rec_model_dir=./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/norm_train_gpus_0,1_autocast_fp32_nodes_2/Student --rec_batch_num=1   --image_dir=./inference/rec_inference --benchmark=True --precision=fp32   > ./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/python_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_1.log 2>&1 !  
-```
-
-在开启benchmark参数时，可以得到测试的详细数据，包含运行环境信息（系统版本、CUDA版本、CUDNN版本、驱动版本），Paddle版本信息，参数设置信息（运行设备、线程数、是否开启内存优化等），模型信息（模型名称、精度），数据信息（batchsize、是否为动态shape等），性能信息（CPU,GPU的占用、运行耗时、预处理耗时、推理耗时、后处理耗时），内容如下所示：
-
-```
-[2022/06/02 22:53:35] ppocr INFO:
-
-[2022/06/02 22:53:35] ppocr INFO: ---------------------- Env info ----------------------
-[2022/06/02 22:53:35] ppocr INFO:  OS_version: Ubuntu 16.04
-[2022/06/02 22:53:35] ppocr INFO:  CUDA_version: 10.1.243
-[2022/06/02 22:53:35] ppocr INFO:  CUDNN_version: 7.6.5
-[2022/06/02 22:53:35] ppocr INFO:  drivier_version: 460.32.03
-[2022/06/02 22:53:35] ppocr INFO: ---------------------- Paddle info ----------------------
-[2022/06/02 22:53:35] ppocr INFO:  paddle_version: 2.3.0-rc0
-[2022/06/02 22:53:35] ppocr INFO:  paddle_commit: 5d4980c052583fec022812d9c29460aff7cdc18b
-[2022/06/02 22:53:35] ppocr INFO:  log_api_version: 1.0
-[2022/06/02 22:53:35] ppocr INFO: ----------------------- Conf info -----------------------
-[2022/06/02 22:53:35] ppocr INFO:  runtime_device: cpu
-[2022/06/02 22:53:35] ppocr INFO:  ir_optim: True
-[2022/06/02 22:53:35] ppocr INFO:  enable_memory_optim: True
-[2022/06/02 22:53:35] ppocr INFO:  enable_tensorrt: False
-[2022/06/02 22:53:35] ppocr INFO:  enable_mkldnn: False
-[2022/06/02 22:53:35] ppocr INFO:  cpu_math_library_num_threads: 6
-[2022/06/02 22:53:35] ppocr INFO: ----------------------- Model info ----------------------
-[2022/06/02 22:53:35] ppocr INFO:  model_name: rec
-[2022/06/02 22:53:35] ppocr INFO:  precision: fp32
-[2022/06/02 22:53:35] ppocr INFO: ----------------------- Data info -----------------------
-[2022/06/02 22:53:35] ppocr INFO:  batch_size: 1
-[2022/06/02 22:53:35] ppocr INFO:  input_shape: dynamic
-[2022/06/02 22:53:35] ppocr INFO:  data_num: 6
-[2022/06/02 22:53:35] ppocr INFO: ----------------------- Perf info -----------------------
-[2022/06/02 22:53:35] ppocr INFO:  cpu_rss(MB): 288.957, gpu_rss(MB): None, gpu_util: None%
-[2022/06/02 22:53:35] ppocr INFO:  total time spent(s): 0.4824
-[2022/06/02 22:53:35] ppocr INFO:  preprocess_time(ms): 0.1136, inference_time(ms): 79.5877, postprocess_time(ms): 0.6945
 ```
 
 该信息可以在运行log中查看，以上面的`ch_PP-OCRv3_rec`为例，log位置在`./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/results_python.log`。
 
 如果运行失败，也会在终端中输出运行失败的日志信息以及对应的运行命令。可以基于该命令，分析运行失败的原因。
 
-**注意：** 由于分布式训练时，仅在`trainer_id=0`所在的节点中保存模型，因此其他的节点中在运行模型导出与推理时会报错，为正常现象。
+**注意：** 由于分布式训练时，仅在`trainer_id=0`所在的节点中保存模型，因此如果测试多机的推理过程，其他的节点中在运行模型导出与推理时会报错，为正常现象。

From 6437875ba37c978a6c3f0fe4dcdf448016743dc7 Mon Sep 17 00:00:00 2001
From: littletomatodonkey <dazhiningsibuqu@163.com>
Date: Tue, 14 Jun 2022 12:49:48 +0800
Subject: [PATCH 012/134] Revert "rm infer fleet" (#6579)

* Revert "rm infer fleet (#6577)"

This reverts commit c71f53ad5f3cf4e0cb997a6ee715fd8cceaad89c.

* Update train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt

* Update test_train_fleet_inference_python.md

* Update test_train_fleet_inference_python.md
---
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 28 ++++++------
 .../docs/test_train_fleet_inference_python.md | 44 ++++++++++++++++++-
 2 files changed, 57 insertions(+), 15 deletions(-)

diff --git a/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 698f1e9849..7fcc8b4418 100644
--- a/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_PP-OCRv3_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -25,29 +25,29 @@ eval:null
 null:null
 ##
 ===========================infer_params===========================
-Global.save_inference_dir:null
+Global.save_inference_dir:./output/
 Global.checkpoints:
-norm_export:null
+norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv3_rec/ch_PP-OCRv3_rec_distillation.yml -o 
 quant_export:
 fpgm_export: 
 distill_export:null
 export1:null
 export2:null
-inference_dir:null
-infer_model:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv3_rec_infer
 infer_export:null
-infer_quant:
-inference:
---use_gpu:
---enable_mkldnn:
---cpu_threads:
---rec_batch_num:
---use_tensorrt:
---precision:
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
 --rec_model_dir:
---image_dir:
+--image_dir:./inference/rec_inference
 null:null
---benchmark:
+--benchmark:True
 null:null
 ===========================infer_benchmark_params==========================
 random_infer_input:[{float32,[3,48,320]}]
diff --git a/test_tipc/docs/test_train_fleet_inference_python.md b/test_tipc/docs/test_train_fleet_inference_python.md
index 2da1170554..9fddb5d163 100644
--- a/test_tipc/docs/test_train_fleet_inference_python.md
+++ b/test_tipc/docs/test_train_fleet_inference_python.md
@@ -11,6 +11,12 @@ Linux GPU/CPU 多机多卡训练推理测试的主程序为`test_train_inference
 |  PP-OCRv3      | ch_PP-OCRv3_rec     | 分布式训练 |
 
 
+- 推理相关：
+
+| 算法名称 | 模型名称 | device_CPU | device_GPU | batchsize |
+|  :----:   |  :----: |   :----:   |  :----:  |   :----:   |
+|  PP-OCRv3   |  ch_PP-OCRv3_rec |  支持 | - | 1/6 |
+
 
 ## 2. 测试流程
 
@@ -56,10 +62,46 @@ bash test_tipc/test_train_inference_python.sh  test_tipc/configs/ch_PP-OCRv3_rec
 
 ```bash
  Run successfully with command - ch_PP-OCRv3_rec - python3.7 -m paddle.distributed.launch --ips=192.168.0.1,192.168.0.2 --gpus=0,1 tools/train.py -c test_tipc/configs/ch_PP-OCRv3_rec/ch_PP-OCRv3_rec_distillation.yml -o  Global.use_gpu=True Global.save_model_dir=./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/norm_train_gpus_0,1_autocast_fp32_nodes_2   Global.epoch_num=3 Global.auto_cast=fp32 Train.loader.batch_size_per_card=16    !  
+ ......
+  Run successfully with command - ch_PP-OCRv3_rec - python3.7 tools/infer/predict_rec.py --rec_image_shape="3,48,320" --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --rec_model_dir=./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/norm_train_gpus_0,1_autocast_fp32_nodes_2/Student --rec_batch_num=1   --image_dir=./inference/rec_inference --benchmark=True --precision=fp32   > ./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/python_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_1.log 2>&1 !  
+```
+
+在开启benchmark参数时，可以得到测试的详细数据，包含运行环境信息（系统版本、CUDA版本、CUDNN版本、驱动版本），Paddle版本信息，参数设置信息（运行设备、线程数、是否开启内存优化等），模型信息（模型名称、精度），数据信息（batchsize、是否为动态shape等），性能信息（CPU,GPU的占用、运行耗时、预处理耗时、推理耗时、后处理耗时），内容如下所示：
+
+```
+[2022/06/02 22:53:35] ppocr INFO:
+
+[2022/06/02 22:53:35] ppocr INFO: ---------------------- Env info ----------------------
+[2022/06/02 22:53:35] ppocr INFO:  OS_version: Ubuntu 16.04
+[2022/06/02 22:53:35] ppocr INFO:  CUDA_version: 10.1.243
+[2022/06/02 22:53:35] ppocr INFO:  CUDNN_version: 7.6.5
+[2022/06/02 22:53:35] ppocr INFO:  drivier_version: 460.32.03
+[2022/06/02 22:53:35] ppocr INFO: ---------------------- Paddle info ----------------------
+[2022/06/02 22:53:35] ppocr INFO:  paddle_version: 2.3.0-rc0
+[2022/06/02 22:53:35] ppocr INFO:  paddle_commit: 5d4980c052583fec022812d9c29460aff7cdc18b
+[2022/06/02 22:53:35] ppocr INFO:  log_api_version: 1.0
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Conf info -----------------------
+[2022/06/02 22:53:35] ppocr INFO:  runtime_device: cpu
+[2022/06/02 22:53:35] ppocr INFO:  ir_optim: True
+[2022/06/02 22:53:35] ppocr INFO:  enable_memory_optim: True
+[2022/06/02 22:53:35] ppocr INFO:  enable_tensorrt: False
+[2022/06/02 22:53:35] ppocr INFO:  enable_mkldnn: False
+[2022/06/02 22:53:35] ppocr INFO:  cpu_math_library_num_threads: 6
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Model info ----------------------
+[2022/06/02 22:53:35] ppocr INFO:  model_name: rec
+[2022/06/02 22:53:35] ppocr INFO:  precision: fp32
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Data info -----------------------
+[2022/06/02 22:53:35] ppocr INFO:  batch_size: 1
+[2022/06/02 22:53:35] ppocr INFO:  input_shape: dynamic
+[2022/06/02 22:53:35] ppocr INFO:  data_num: 6
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Perf info -----------------------
+[2022/06/02 22:53:35] ppocr INFO:  cpu_rss(MB): 288.957, gpu_rss(MB): None, gpu_util: None%
+[2022/06/02 22:53:35] ppocr INFO:  total time spent(s): 0.4824
+[2022/06/02 22:53:35] ppocr INFO:  preprocess_time(ms): 0.1136, inference_time(ms): 79.5877, postprocess_time(ms): 0.6945
 ```
 
 该信息可以在运行log中查看，以上面的`ch_PP-OCRv3_rec`为例，log位置在`./test_tipc/output/ch_PP-OCRv3_rec/lite_train_lite_infer/results_python.log`。
 
 如果运行失败，也会在终端中输出运行失败的日志信息以及对应的运行命令。可以基于该命令，分析运行失败的原因。
 
-**注意：** 由于分布式训练时，仅在`trainer_id=0`所在的节点中保存模型，因此如果测试多机的推理过程，其他的节点中在运行模型导出与推理时会报错，为正常现象。
+**注意：** 由于分布式训练时，仅在`trainer_id=0`所在的节点中保存模型，因此其他的节点中在运行模型导出与推理时会报错，为正常现象。

From dee5305b7212744f62edddf86465b7f57f35878b Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Tue, 14 Jun 2022 08:33:43 +0000
Subject: [PATCH 013/134] add tabel tipc

---
 .../train_infer_python.txt                    |   4 +-
 .../configs/en_table_structure/table_mv3.yml  | 117 ++++++++++++++++++
 .../en_table_structure/train_infer_python.txt |  53 ++++++++
 ..._normal_amp_infer_python_linux_gpu_cpu.txt |  53 ++++++++
 ...rmal_normal_infer_python_linux_gpu_cpu.txt |  21 ++++
 .../train_infer_python.txt                    |  53 ++++++++
 ..._normal_amp_infer_python_linux_gpu_cpu.txt |  53 ++++++++
 test_tipc/prepare.sh                          |  29 ++++-
 8 files changed, 376 insertions(+), 7 deletions(-)
 create mode 100755 test_tipc/configs/en_table_structure/table_mv3.yml
 create mode 100644 test_tipc/configs/en_table_structure/train_infer_python.txt
 create mode 100644 test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/en_table_structure_PACT/train_infer_python.txt
 create mode 100644 test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

diff --git a/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt
index 11444a3ac1..de9e336332 100644
--- a/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt
+++ b/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt
@@ -4,7 +4,7 @@ python:python3.7
 gpu_list:0|0,1
 Global.use_gpu:True|True
 Global.auto_cast:null
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=5000
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
 Global.save_model_dir:./output/
 Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
 Global.pretrained_model:null
@@ -45,7 +45,7 @@ inference:tools/infer/predict_det.py
 --use_tensorrt:False
 --precision:fp32
 --det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
+--image_dir:./inference/ch_det_data_50/all-sum-510/00008790.jpg
 null:null
 --benchmark:True
 --det_algorithm:SAST
diff --git a/test_tipc/configs/en_table_structure/table_mv3.yml b/test_tipc/configs/en_table_structure/table_mv3.yml
new file mode 100755
index 0000000000..adf326bd02
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/table_mv3.yml
@@ -0,0 +1,117 @@
+Global:
+  use_gpu: true
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/table_mv3/
+  save_epoch_step: 3
+  # evaluation is run every 400 iterations after the 0th iteration
+  eval_batch_step: [0, 400]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/table/table.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/table_structure_dict.txt
+  character_type: en
+  max_text_length: 100
+  max_elem_length: 800
+  max_cell_num: 500
+  infer_mode: False
+  process_total_num: 0
+  process_cut_num: 0
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: table
+  algorithm: TableAttn
+  Backbone:
+    name: MobileNetV3
+    scale: 1.0
+    model_name: large
+  Head:
+    name: TableAttentionHead
+    hidden_size: 256
+    l2_decay: 0.00001
+    loc_type: 2
+    max_text_length: 100
+    max_elem_length: 800
+    max_cell_num: 500
+
+Loss:
+  name: TableAttentionLoss
+  structure_weight: 100.0
+  loc_weight: 10000.0
+
+PostProcess:
+  name: TableLabelDecode
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ./train_data/pubtabnet/train
+    label_file_path: ./train_data/pubtabnet/train.jsonl
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ResizeTableImage:
+          max_len: 488
+      - TableLabelEncode:
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+  loader:
+    shuffle: True
+    batch_size_per_card: 32
+    drop_last: True
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ./train_data/pubtabnet/test/
+    label_file_path: ./train_data/pubtabnet/test.jsonl
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ResizeTableImage:
+          max_len: 488
+      - TableLabelEncode:
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 1
diff --git a/test_tipc/configs/en_table_structure/train_infer_python.txt b/test_tipc/configs/en_table_structure/train_infer_python.txt
new file mode 100644
index 0000000000..d9f3b30e16
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..31ac1ed53f
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..e8f7bbaa50
--- /dev/null
+++ b/test_tipc/configs/en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,21 @@
+===========================train_params===========================
+model_name:en_table_structure_KL
+python:python3.7
+Global.pretrained_model:
+Global.save_inference_dir:null
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+null:null
diff --git a/test_tipc/configs/en_table_structure_PACT/train_infer_python.txt b/test_tipc/configs/en_table_structure_PACT/train_infer_python.txt
new file mode 100644
index 0000000000..f62e8b68bc
--- /dev/null
+++ b/test_tipc/configs/en_table_structure_PACT/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:pact_train
+norm_train:null 
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null 
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..d7c847fcdc
--- /dev/null
+++ b/test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:pact_train
+norm_train:null 
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null 
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 65ac2bbebb..8c9e12511d 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -55,6 +55,9 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
     if [ ${model_name} == "en_table_structure" ];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
         cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
     fi
     cd ./pretrain_models/ && tar xf det_mv3_db_v2.0_train.tar && cd ../
     rm -rf ./train_data/icdar2015
@@ -147,9 +150,12 @@ elif [ ${MODE} = "whole_train_whole_infer" ];then
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar  --no-check-certificate
         cd ./train_data && tar xf total_text.tar && ln -s total_text_lite total_text  && cd ../
     fi
-    if [ ${model_name} == "en_table_structure" ];then
+    if [[ ${model_name} =~ "en_table_structure" ]];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
         cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
     fi
 elif [ ${MODE} = "lite_train_whole_infer" ];then
     wget -nc -P  ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x0_5_pretrained.pdparams --no-check-certificate
@@ -172,9 +178,12 @@ elif [ ${MODE} = "lite_train_whole_infer" ];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar --no-check-certificate
         cd ./pretrain_models/ && tar xf ch_PP-OCRv3_det_distill_train.tar && cd ../
     fi
-    if [ ${model_name} == "en_table_structure" ];then
+    if [[ ${model_name} =~ "en_table_structure" ]];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
         cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
     fi
 elif [ ${MODE} = "whole_infer" ];then
     wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar --no-check-certificate
@@ -335,9 +344,11 @@ elif [ ${MODE} = "whole_infer" ];then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
         cd ./inference/ && tar xf det_r50_vd_east_v2.0_train.tar & cd ../
     fi
-    if [ ${model_name} == "en_table_structure" ];then
+    if [[ ${model_name} =~ "en_table_structure" ]];then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar --no-check-certificate
-        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
     fi
 fi
 
@@ -382,7 +393,15 @@ if [ ${MODE} = "klquant_whole_infer" ]; then
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
         cd ./train_data/ && tar xf ic15_data.tar && cd ../
         cd ./inference && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf rec_inference.tar &&  cd ../ 
-    fi 
+    fi
+    if [ ${model_name} = "en_table_structure_KL" ];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
+        cd ./train_data/ && tar xf pubtabnet.tar && cd ../
+    fi
 fi
 
 if [ ${MODE} = "cpp_infer" ];then

From b215b786d4a545ee84ee477509469933f1b0dac1 Mon Sep 17 00:00:00 2001
From: MissPenguin <lichenxia1991@163.com>
Date: Wed, 15 Jun 2022 17:50:29 +0800
Subject: [PATCH 014/134] =?UTF-8?q?Update=20=E8=BD=BB=E9=87=8F=E7=BA=A7?=
 =?UTF-8?q?=E8=BD=A6=E7=89=8C=E8=AF=86=E5=88=AB.md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 applications/轻量级车牌识别.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/applications/轻量级车牌识别.md b/applications/轻量级车牌识别.md
index 31b1b427db..7a3d7b238e 100644
--- a/applications/轻量级车牌识别.md
+++ b/applications/轻量级车牌识别.md
@@ -787,12 +787,12 @@ python tools/infer/predict_system.py \
 
 - 端侧部署
 
-端侧部署我们采用基于 PaddleLite 的 cpp 推理。Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。具体可参考 [PaddleOCR lite教程](../dygraph/deploy/lite/readme_ch.md)
+端侧部署我们采用基于 PaddleLite 的 cpp 推理。Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。具体可参考 [PaddleOCR lite教程](../deploy/lite/readme_ch.md)
 
 
 ### 4.5 实验总结
 
-我们分别使用PP-OCRv3中英文超轻量预训练模型在车牌数据集上进行了直接评估和 fine-tune 和 fine-tune +量化3种方案的实验，并基于[PaddleOCR lite教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md)进行了速度测试，指标对比如下：
+我们分别使用PP-OCRv3中英文超轻量预训练模型在车牌数据集上进行了直接评估和 fine-tune 和 fine-tune +量化3种方案的实验，并基于[PaddleOCR lite教程](../deploy/lite/readme_ch.md)进行了速度测试，指标对比如下：
 
 - 检测
 

From 08cdc9e2091354958868e3bb697a92bcee0a8ce9 Mon Sep 17 00:00:00 2001
From: MissPenguin <lichenxia1991@163.com>
Date: Wed, 15 Jun 2022 17:56:39 +0800
Subject: [PATCH 015/134] Update README.md

---
 ppstructure/table/README.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ppstructure/table/README.md b/ppstructure/table/README.md
index d21ef4aa38..b6804c6f09 100644
--- a/ppstructure/table/README.md
+++ b/ppstructure/table/README.md
@@ -18,7 +18,7 @@ The table recognition mainly contains three models
 
 The table recognition flow chart is as follows
 
-![tableocr_pipeline](../../doc/table/tableocr_pipeline_en.jpg)
+![tableocr_pipeline](../docs/table/tableocr_pipeline_en.jpg)
 
 1. The coordinates of single-line text is detected by DB model, and then sends it to the recognition model to get the recognition result.
 2. The table structure and cell coordinates is predicted by RARE model.

From 10a3f2573d8e875c69683a71dc067d9060fb6875 Mon Sep 17 00:00:00 2001
From: MissPenguin <lichenxia1991@163.com>
Date: Wed, 15 Jun 2022 18:13:01 +0800
Subject: [PATCH 016/134] =?UTF-8?q?Update=20=E8=BD=BB=E9=87=8F=E7=BA=A7?=
 =?UTF-8?q?=E8=BD=A6=E7=89=8C=E8=AF=86=E5=88=AB.md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 applications/轻量级车牌识别.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/applications/轻量级车牌识别.md b/applications/轻量级车牌识别.md
index 7a3d7b238e..7012c7f4bb 100644
--- a/applications/轻量级车牌识别.md
+++ b/applications/轻量级车牌识别.md
@@ -249,7 +249,7 @@ tar -xf ch_PP-OCRv3_det_distill_train.tar
 cd /home/aistudio/PaddleOCR
 ```
 
-预训练模型下载完成后，我们使用[ch_PP-OCRv3_det_student.yml](../configs/chepai/ch_PP-OCRv3_det_student.yml) 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：
+预训练模型下载完成后，我们使用[ch_PP-OCRv3_det_student.yml](../configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml) 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：
 
 1. 模型存储和训练相关:
    1. Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址

From 995a6d03e14c1cd9466533d1cec03578f84ccc7a Mon Sep 17 00:00:00 2001
From: tink2123 <y_tink@163.com>
Date: Thu, 16 Jun 2022 11:07:06 +0800
Subject: [PATCH 017/134] add digital doc

---
 applications/corpus/digital.txt        |  43 +++
 applications/fonts/DS-DIGI.TTF         | Bin 0 -> 24448 bytes
 applications/fonts/DS-DIGIB.TTF        | Bin 0 -> 24896 bytes
 applications/光功率计数码管字符识别.md | 454 +++++++++++++++++++++++++
 4 files changed, 497 insertions(+)
 create mode 100644 applications/corpus/digital.txt
 create mode 100644 applications/fonts/DS-DIGI.TTF
 create mode 100644 applications/fonts/DS-DIGIB.TTF
 create mode 100644 applications/光功率计数码管字符识别.md

diff --git a/applications/corpus/digital.txt b/applications/corpus/digital.txt
new file mode 100644
index 0000000000..26b06e784e
--- /dev/null
+++ b/applications/corpus/digital.txt
@@ -0,0 +1,43 @@
+46.39
+40.08
+89.52
+-71.93
+23.19
+-81.02
+-34.09
+05.87
+-67.80
+-51.56
+-34.58
+37.91
+56.98
+29.01
+-90.13
+35.55
+66.07
+-90.35
+-50.93
+42.42
+21.40
+-30.99
+-71.78
+25.60
+-48.69
+-72.28
+-17.55
+-99.93
+-47.35
+-64.89
+-31.28
+-90.01
+05.17
+30.91
+30.56
+-06.90
+79.05
+67.74
+-32.31
+94.22
+28.75
+51.03
+-58.96
diff --git a/applications/fonts/DS-DIGI.TTF b/applications/fonts/DS-DIGI.TTF
new file mode 100644
index 0000000000000000000000000000000000000000..09258773c7219ad9e60b92b918e3c50b58f43c9e
GIT binary patch
literal 24448
zcmeHv3y>Vux!!-KXS!#mXL{a`ou1vD-Br)(xg%+{T1kx95z-3S0V5;~f{TDfTD>el
z7D7n)d4YR{1F>-`RLBD=g;NCy*uh|7D#i(A-%I2hLvoYrbpm#vQc22jZgKC$0m?UM
zHs60v_e}3gfa_MOZq@c~&vc*Z?sNYC{O|woLxPABw3ak_^_o4k^^ZUF`^Sj(u0`u-
zc5T1vx@Ya}r*QsNoNwHJ$K>%-FZb7ouES}dzW>z8iup?FzYyK<-_dZ@!Q+SSc>LyH
zJVg}zOPn7#bnMK*Q+u)_M0${@dHvx7cbu-icinAx-pgn^e|T#0z<+w;wXfjXkFak%
zj0WwraTSr+fb(UC?>Kq7c*$!eqU-Bud*;|(`zI-J_)Vf~Z^rp^cTAo>E^ej2!0}#O
zU%7Mgj;S5L`nG}d(?r@m$M3rPWNe@LA<>>Ne*fV3iK*i^{p>Gr9%2B#O$<Bsul~Uw
zzj|w;{yxTyzA=z<KYFM0Ju|MI*B->VNdb9>yhnR*W)u4O8PUv)cHV90o78qxwj8D1
zWXXmAg{eVN93P;c;5Y6$Cj#nXe7-<8y2paDBIh0lC?<eyxoNaZ?01iYxaNoMaftSc
z3+}N_*}$#svG5bQAO4IA(>)H5E>^h58Z8%>xW_?U^P+nkqSfM0-D91ifzN_EL@<IU
z@i^778mBr|<5b6Loa$JOQyr^us$)6Mop;@NvU1&-<5LGG_fKu9?AWtr$B{!vPEH;h
zs%$?oHF<LCK;`x`l^yroedJ_i&ppQ{@1DGO@?>Q-Unp-~w|VoXp~|*n$11yz96EgR
z?#k||yQfZ^nmRx`=`OmHPErMQdWMeE6dj~V+K=xoXxl-1XbnD(&>`$6@yjt9Li=_)
zfh#7_GKER1(CxUFufB)wme*Bq-#vKRB+e)4UU_AOR=1zn>2)1##@{CS)NOeBF<i4-
z_IDUhzgwP7;d=@_p5op*urD;a&)jw5j>%*689;S{?m)lC<{-t7ye*aMr%v2`<gPm_
z>xR~S%;*=x2De`?N5_yeE-SPS(5$=oICt!+?7!>8ROKKh7Q<LKw012f?#{}@<einZ
z!<Dt0wyfK{W&P#~CT_EYvw}Hcn5OXjgYKNEzPY!x_)|FN{wDx8f3Kxs`MC*m$#~vE
z>+#9sW1Mfk_y|G3jZi441!(oD@yhd>Gd?hRxHxe5wjFC$x6jpSrSkI3tDDl_ZHSGV
z)~v2no^NbQKQA_N<C@jt`JJaKm*dWzlZPvN`9TAlo)0*D!PKT3*EB5KE|pAE2#D6i
zxcIv-i<v(-HGX(U@x_3-lUBW`fkaom7!>>&k}bMyF=R_P!R--w9F^@cc^sElTJkt4
zuSv<zv~17F&#e44CqMJ@*MdAQ%Fiyo2IKJmvwR3pKR$wxavB-5wP71UoAi(^Lcx+2
zF!Yk3fY3EE1?@al8!ggUeXK5O^$T_0DS4re|8;9$2vca)aH=mgh`nF@xBk}c+Jhf`
zv0rO^^eUcwbmnO~7kGt&6rng>-RO_U%xDl###byD(L<URD1{BdW7rpur9{B%?9^CM
z)C%<rg*w%ylhgI&v`D2liy>_2)e3}GWm1D9qxer5nStE!xn`}_ESH;2G2D8qdA`=1
zZnl0}Zv9~skk$iP`p_N4_C~iJG5}oIC`Cd+0<e*AP}3kHW2e3DPmR@!V+9|$+&BKw
zbCi2F21kYq{BgGGG>>BsY<8Rn9>7j9z_~@d7r387&>kr&P!BDml_27cRHw`63L?IJ
zv{x7dgQLR<!JquB(ZZ<w%oW@pxJjFLsl3R&`Aw%<b#8ZRj#E2cy~c5_xse+lbFe+&
z>~fr4xA5U(bFX^0bAh<TadtY+ecXcWr`767$2lb125x9Kh>YXhQ>{KDuW=l!eTG@9
z&wNh2tzFxOwi|RcNIz*D$ZJswimqZQlT907i*S^}y_KG9G8wbtG0U)*_xIVo%LeRf
zK-idwT-Gk6O}k`w7e!a4+>?rxtVp;-rcRYi!s>CB^=XyT;6T6D)2r{R=1&Vt(aNc@
z4y`1AS+DcHPSIx)lmo)O0Mb>v+kC;D&tzTAD?XETY9G@O=nQkc+zPD{M&ICQpPaIS
zBPU41dBMmO_=n7BAvcg3&KZN^ZKw67<A^GdRI^*p)XI!@c^;`&TmRQ__BqvNv+4x?
z0opG-QLUb@bsw&k->OzGJi(U=kPoi^=Z<sBg$gDPmIExHz`nqbeYhQ<6;z`Q5PPKP
zCnJevurN9@XcRy&IU^)n=C_+-aE$*oLqsHTfIjg54EKGGN1Ac2sY)usr)+%*<N5iy
z@<U_V*+bv22A;%)tzS7#)mpaf;cE3^-j{uAp+;f0{^;PZ(|dnyw_Y=01u-0D@Sn#K
z-NbOLq3<{L_2nd)tVyTynV$aMm1|Z-^{}Cb%f%8%r^{HL%*E{?eb54t1PA)<-j&&m
zy}FPW=~b&rYgR1l){3RUs;!k0RzwSzQZZB0K}?liQ<OYW!g5PW68H0hBnnUji!l6;
zO(!pm)u-z&QFO>a9Z4dQB*8Q>RAQQ7rd2eNk~A?WsYKC-`c)k}gQErUobwyU`ORJ%
ze9E-3LQ;bWwLWAT5%>xJGgNiTHE|s1lmr6VRT~H0oRM_03O5!Nt^9~5pjv&iyH-Ra
z+j%VA&PwSv)JQ?FL81mJ2&>C4&x)?_CI%8<ggi0E5==@j$Y{2>9r>gvB`HT!jkPH@
z{W959BxFEr>WN&;&SNT=TT@9hs)b5cJggb8_0uU+81qo>DheQ2%5QY;vYFzryy0R~
z3c7+qW`?mVoF)WQT(MEP_T)DmXJ_CQ@K5V)g@_L%5?XINRo}>&!g%Bd8nsL!mWZda
zX<!5jCuj&eZ^i8d2A+<DS?!q6K9o#QCT)tu;sdWOJn%$lm73xamjDI^K@kHy@TyZQ
zmupUSgEQWI)Nvkdj^8L}@w4UH`D%4TwRz!)(}Xibb+Jp_JBKg6(HQUVEk}dVh%WLO
zBN@-cAuV;o3dBTjPe==v?O<QPE=H1eJY(mLSeLL0>68{L<+2IQ%0rFjP7AT1XiWZc
z9p(db!NPDF`mIgsZRtnQuTse}tuGVMb%Rh<1BKBVJ`#d;k^<O(C>b_J2gXFfn6QS|
zt!TbBY;9;R&oYto_6<H=@mO>D`e7?DG;EbGI~gCTY`@%bHjiKS?Dp-@JbxYE(7og4
z%SzTT+%QnzOXA^x-j))l8}*p3M~w`WM*yCa9n9MqNo`DeT2#bJiAYIL;FdxfQwaoM
zT7$uTne=IILTBp+z|Y_lR6FbnK2(#dpii<nQ^dTAc}iF5XUI*k=lxPlKHAZ@0;CDa
zdA|bNftG|^Wxkwwgw9eEkj>)DfJpf4jD`4;dy|$0RQf-X94<DChs6WHOiv>klq_zC
z&;x5VW+%aur3W!`sg%S)$6<w|GzyOd_8yN!!~G_>S6X#2e55%ipMv{?Y{-oa=jzQS
z+f2)#yr-di*mB#@NJm0iKre-%bqoPhO_--i;+D=1S$`z|K2qPrW06LZ=INUcUh;H^
zX!sQaa{$15TS}ZT?4~s3$)+=nt=UXCR0N#Sh#3{7RL-)?)MXjE9m?nIj9rLp5vYfl
zW+oHN$!QDXE*zq4#uVl}PIj%x1!cVw>pbML=}!D0ggZogsvwD7>l+<*j1;8!+jecj
zn)uDc)>^qK_D<Ant951qa=Qt!{SY|2X*HWHn*{j#Ss0+*t=G{cobDR1c@!~$bAj(j
zY;I|Eu|S$*{EE1bDmn%Yj9*t`OPsb%US;ljw(j|6>&-Gk9Bq8EJsp$Q=beq=Jlr3O
zWzyMHA|4Azqjr%cIItO%HkW{vWk#)d3CP4Vf~JW2aR?fd7C#f)9P^23&&434uoz_&
z4h#y$Sq?~Z*f<|lH2J4MojB{EOl+O7Hz>0RLcMumM|1i-<EkkkDKIMmnhBRWI?x<$
zte4OL7(fzBMdrd{#UU9035)?37|joiawz<uDEulea-EB!)`4TRjq!PKfCg%Fq4<^Z
zJ$z8GT(dyqYVIw{(yfivnPh@BYA`?%E0|8%*@OTY3o@h7rKNaG*Fu8!=^zQ$`SQ!L
z_H5MZe$^y}Kn_ftNwF0;XWaE>O`P=)SUdzmJzv{AZ@?F&q7+k6JQ0hAt+|*tqoqhV
znSeduB0lU#{HLIza0mOM1Ik<~YJme7qS;g=w)&tNmSp6UG9De}sZ3vN)Zuk#lqnXv
z@{*eJu~gLV_E@9TWtR&=5)wcHQRT9wOnNZ}S%|JY^*V=3y618UM2$CJbAe>(QeFs7
z#VwoW0jq&$nA%=}tY?XOd82}ZpdJtB7_ex>(DYKwgc-q*fWBPrfWT>6$!xMIYfaf`
zmd6flk`j`cACu(7?AyB6M@8%t&f!~@>V5QZV`oKTyt~}vVcZ7F`%5NU_fjDB_Sk)8
zKh}$doR%$>x-!~=xc8a$Qjc|nJeXsV3FyDb1S~4B+*z2%@bf3aFTxg~y)%tVB<+Fp
z6f>prF>5aANq)+tlUf3XjioKfRSJBz0%nP{IdZ`jYuyVo)`B2=+%*s0Sq$a}HGohl
z(UXD8L3sv^G`bK^3h}gtf{5;v3>}J^!#yz2RU}EVjG8bg4;e>-`JrLo0}}*+oR46L
zKzHte9mUTLf}(dcdfA@WY!MFGx)(kIG2{93?J9p4t`?2J?~(Qu_wyus<W07+#CKI+
zbvh}o4O}7h^tJ{j3A#}?3|&wu0006aY~&Cu(nA>u1wN+h(gK3@<9#cxVRa2<I69ny
zV^9#+K6m7uk&$<fJQv73`%Y{1+s`s&b$S<e-5Zd&CNP$lrX74=Q@6uf0GJ~q5>|mX
z*NtLPBJDj^x3RXt7n*JMg<SLR4K`OFgKJ-veccJjwtJ8nS{Sx-D5Q%(iGo2GiIEUP
zX6T;`vH=+Z8yFcC*B*U#<ehg$o)x!0H}XEdS%1ttEoD#$F^mMfu`ar=v8k(A$Y(7*
zVx*F3jsQ}?G9xiG7Q==e$)xREQY1_$zX((z?u<}ZF(o4Nl!%|&y|AD{i;EqfMS=*Q
zvgQlyp!W$j<Q$1C4Ix7|=^SU+fs_?ZWyraqnjol;Z93(xJt3{Q+gY2m?rOpwhI|Wb
zl~!v6@MWn)z2wj;1f*df!$p$OfRG9tvBwmWE}}rZ2(~0Q7#ye>EXY7oU~9S7OiZ+%
zhr#|8H#D%TSu3|DWO(4|){ooafoICKs<Ri@J?J<AQLbGB1GDP<4<6Uk&2JBt)6z3;
zt*>&3?%$M$4tFyMGAKAp5-D4|K;6)STgw_H5Nb(VN$81mxDe07*970HNSxFaC8;;E
znUt0+<#X^dp*(c+w9wV$_?3riY`fxUY<fX44C3w4m6p%2o#@kmbEMJMw*H$fiCR<B
z2=sP)3h?#jbj>e}d^*UB2&g&w$Bk>2$-qEgrN12;015U43kVQY%XU@ow!AO_!c9&^
z^x7pYU+OM*X~p?*gQA~27gt7bG&|~ujb=r#ij|p{tHt@@1Oy#nNQ0nKh>Nrd7jZ_!
zkQCv<qTY(R|1rBE@JiSdaS3>VfP>hje3K74)U3R2&jVAuikOcH5ia83P8IXX=29x=
zldx<vm*kjFCaq#VF(f>G$CwY8LB)KSE&3E&h_@$NGj7mF>}{O`rHj^k92lDw`9Zi;
zMZjbX#kDp3^Zd2O`0VIUdxk4zyW*SS-)(L?M#duZsx(056U^zwXY*rcaY0*su8%a&
z-}4!@3)!<uYtAw#q@Te=+YSvWU$2jSy*}ydNgh_RRZioCc&zog37$X2!kZTG_7wXn
zZ_o3SJj*phH#Jtm=1WMq<l3(p3!xv&CzDtZ!HQi#5P24nvShungVnN9^7IXMB#(d8
zuE6n)wlyU^Q~y}~aSM)=vOm#Jk2J>ZBC}2?qW5(7_I39rOr*QwrLID(FPBT(1Fn@D
zjfik5luso!vy{!iw5#-%`nwTAEf-7?_e&J^nv_&Q&<ZD#@?diyAJ&Q8b>s=p4j>$e
zE7>~gk3_4`su&#{Fx)%_Tk+tq{_1s(bDauC%NR7th_q<3oKbxD6f}$*e(pqFtBxX1
z9}RM>$f6AN{oRdC#gv)L<}IZ8py`8|NLsTBdAlnM6N(K&X}cv8aRe?AIo6^&?F&Gq
zxcR?X{%(gL7tkk&Rs!S_hqD(|C1)4VBBIaLARC}NHX~L84I)#9iX3IkhGg!NCCj!3
z=7((-5zYCuK#=ws3lEyw@otPR<_6nG-2nGFjt+B_n2n5&9OuH<9p|H3H*B+x=cIHr
zVg)GV2*)EPJ8@~<&Pc;B2zxqVA^ufLK^{kv2*@E(>vF+Zy%Rp_<Z3%%Y#HFis4`%L
z6+(t;Mn)Z<bJq^isDMH9%;%avvA6lMs{Tlmnsc^+e!o%gN+!*GuAnA8o50Mb5SxQ>
zmn#<RQch$u<eU9?3~{Fg=e<q$Aaq6e)Ad@%(spT|nHuqGC2j5b6rdq5H!y-+DKt!n
z1h$U6JMy6tMhnn@I5P6TM?m_BKWTGjquJ<37Bi@a<GO9}H2CCDbkP}5mb=2f8mXo5
z?h~8hZNI2=Lt&c)Q|2Oe+t(V~DFKtMt5lR$Te%SL60YH<71h9WWw2S&cq>Y~t%Mqd
z?&5;h8%vj8A}Ed>L2{0cW1m*BfF#Vg+4hU#x{Dfd(!`aKf>}n~FPQRW&|onW?sdm`
zx>}u2V=Pw?wSinw^wpIu74rFlD_e?%ST5_Qu1_IbWO)GyO<JpMHxAUL2<&14#z7%K
zZlFV5FjHH9HuBQQ0;HwH%fH|F5*Np{X^W)`R|#-V^pVDn{%*TejFge~K(1Vg9K=cx
zb;}6Ogf7c=TX2`6eHFW>kmi(+#R2eC4EYZ;N&`Kn$odtCQXZIF*MXF&&V`kS=Qc64
z%d#E6sgvqdMw%vb%}5KOYQgiMWPY=g_ij4XS{Z%{s~69Wnw#^6B9Sq2CI|mqE1!|&
z$$gDWx+3rwAy5<|pi5&SlcO}FAmU0C1g%ZT84i%RZet1(+0~-1XVMlYo)R6K9`hr?
zHOE~+BDE)bX`Bgw&p6lq?%3XQt?o9k?L@$@*h{HN*Kv+d;?4OA`gY?6XVGb2u?*?$
z>LAa)W3P<cE4%EK3)8^zK~MpxDM~Ar4`@C@D9FN{g(yMh0_TZUm+n3CXQP5Qkj2VR
zNEV}#DL#fc;O|-^dq>(h1A)M``NMWmgr7ie9Jz617SFJ+!ZcE$CmL7xGL@8VS3|SK
zl`9nsg(4G6A>ZxNOHa35u|+;-W@nL1R}rj(U_+^E7TwfQ0Hn$WTnntNiDbRNG$N$!
zz)!~%n?*={ix5(;km%7=(8i5Huq)@0)9}b&j+9&t<+0BkidsAJS0iqE&u4#iWXabX
z+sg|T;B=~QdSHxl{wJ;cd{o<@$c8gPNWcb>P&K{s8n^C?d8p2Uq-}>nMYxcAWi;(9
zvXq-lI3Svyk1eA4_|ln+*UZe))Etqyu2Bs^PX~1~Yza2HHJyTqXhaJlyo3lidp!^m
z3&g&Yzf?JLsJ^zr1Q?Ps1P-4;MsHLCoV}d)njd<}epYG7>pMTMmG{^~Kau$>&ze*7
z!cp+AH0n7>^JG^vZpAo~NdY6QN5gv1&I`MoHZ3^WF*yyz0!#K%7sxY1`%u_&G5e0?
zdM6#!;b3JRsfj4V?oc8;<QL(beUOPd<%_mZrsN5483jhBP8s@d`Xn~8M{J$cq{ZyB
zTTxLgBYvN6tcT9Bq7+C;A4m_C!jVWg5{(Jn&LfDLM5qs$B3Y^e==>sETCbAanB-sp
zDldDEUX~>0+J1SD(=^z7V*)^?3QA4MtZm=>Bky0*`Y}gH7&ewT822sjk9;5yV*?-;
zzNhT(j#l=p+9LGwHyc;vOJPhFl6G7JBg&b0E?J1$J!D(ZLb+U`yDNgIh=JlGK!nH{
zq9Z9Qim-G-8o_WAAa^>E5)Z<~4K9=@jm<T=T$@W)%s@R!#Uyjosw^WjD^tw{Xev$p
z90!1(^lW;Ws|NI(s0QFqaH9IVj`Iw}&9s-U_RXg!QBE}0XA@zq5KBihky2jI8AdK1
zvVx_QNCpB)xVc&sMOhk2=Lpq72ofUD57om}lm#v}nO7+wCzNXk+LZ%5ol<5L8OV`Q
zWCI2er!RB|Ft3^6Ob$_JQGR^Y6NjH%_4wnfo;-vvacp<%%GPg%ab;_F8Hd7X{T7El
zsmOE5QD~b^HbzlXh+vQzi>8!2$HA>~JZ49`iZ)m<Y(zrH@?jQpNlYImE)g|F$TO+@
zj+N(`uvgx8_FM%ZZ7Il&a>%u#1aq9SnPdTH!ObRfQ9)}T0>H{z!e`A!Es3n&P-Kca
zuxaGS*-TPwh~g3m7vCIt@a=D`%vvE06<fjxhcHozQZkl{$8%O3j#oGmmiZXY{OKW;
zgmH~RW&X~aj#&l(fXTZ;urI?AI|T1AKOw}j2S!+czupvIF0>wn#aG>YnfBC%*8f=^
z`Q|7Dw;a5vhL^uS`t#sr=Qg+_=Q=YlAA5acMJOoq`B6LV21$Zi3B_4l(<ig9iw3G{
z7d$B3VjgA1rQIZ71{s*ac3^;u$|l}=Yhvdf84-{&+bca#0Ur4Mb1Gj%1PZZa1dzaj
z;@F3{_eCpTB$5WTf-p1`jil4nmkx@n%I6{t@vj@E*kWJ!(LqtKre|S+ef<>O&{&zU
zptnM?TtuNEm9#AsD4S8v+Qz|Jm^(^Xrl+}li8LP?{JGn-o4b{Yi(UMJ#wBWn=6Lge
zf$ETEu-<BpOLq)}cD{MxHn(zxGe+WP;(0gAG0wTLkedu6f!ZOI8G>5^v=1+Qz$Q>4
z4LsqAY46)(g($Pa;7B0{U6V7!7r*>M>yhu=Cw}&&qu<@Ajr_}__aOyk&y0(2%{&U;
zb?cl$WNU$>G&z{F+I3ED!s|=m6l6z^D<6U@A$#eE0j^nc8gUBBlpR<Yg^>O!gSY`T
zhk%)kQ4^y-wAOEH1a=Qwp9fcUG=R#Vr=@4MlqG?k(vG1oWl3aZ>{6CQR?aS(0b9zF
zcolI=SrT{!OIZ@n8&tl<QkDb(=A|r2U@1!?-|>^$Vkt}F73nTzNtUuC{AQp(2)>jh
zS;~?i3bd3ZQKfE6SrQI=_+Kzy%91Q)NxZO|FDL0$$Sh?^5dK`sk}PFO{#Im3LdZjq
zI4I-qDtcn0BBoy80#h>)>natHi{uJzt}HEc(gcb80y0_2gg|OFpW-@gq<!SO`9&0#
zx~pZ(&KdZV`E!$4GT_b`{Ibm8qYg+FY$M5v)Q?Jqt5@|8PT)NXi(gJySskxd$9Hj>
z_3RlWvyfT98~dDEP)RMPHQml413a3V8K>{aOi(YfBfJ<3iY)O~K0^N9f&~}=znWY~
z4X38A{DF4wM_+SudtaK1n=TEhuT(BwC>34Y@Nxx=8x6QA;uVuTN)Ssa`3`wWu3mtI
zN>_1_e99-ojZ7oRcmFxNqN;31C5C{f(cw3pq2`&rLuF3dm4}WyI~`}|mz^`sA<m4=
zUdjNa{7&mBUKs$H3=tI1iziT7yt2Vcjk0zyY+D60vX2U0He>8U?ShvvEdWKCkxRw!
z`a*tq{l?LpKF~iT2CD=8dOf?kYei#urM$jexjNn3<Jg&eO$2W(<jSe;p8czWk#xL>
zx?RMlv~M7L;^lnRDjjOIkue&_`{UPQXzt`Q34FR$M;2R`<ab+*M0~i=Y9+tx8YKr|
zNzGf@M9zbMZ4Gc<KwxbRWB~<MX>-?D@N!=$jbWwBsU==E0k1|C&Z6ERkj1(#Jfr?6
zp_o9GKH!m`a+QvGs~O=D;gB9d`GGD+hxeZM=^>=zQEK7ltqVxc)v=tCTT~&(sfHRw
zO^6&Svbdv#4>8{3-k5ngBhCW;e_T6>S1Nb}Z;q`+O=P~I>!d{^T{JdEDyJJ`8sD&S
zczr&X1$kHNU$}nneP6iYhWo@VH+<o~z1V?gQ)~|Oeu}jceha~bYXA0H3AkT7|H<#y
zzJ=0}uea~#m9=;~FlYZYhzaWz26lqSc&)^vaOTe=-I&JyQzGMjqVS_ck=wBU3+%W)
z+JpT;EJ2918O=W<is8QaVIm7_BU<=9F-?@bi70gtYp?9W{tQ-@%3{Z}GVc;)@mp>S
zcC5>o|F1+v9s6rUUD!+L%Lc!fCy2T$*#ACJ&!t4Y{n+uG3i|5%GEx86i3Uz!S;iNL
zoIfF|VyuI=6D`M@i7U|8$};wMh*rHywE8}xHAU=SB^tsrhyEV+|43BJ6RpL|4eM58
z|8t`C*oR*t+JNVb9LLIwJF)*OqEX;%(-qj?AliJMXv;q%y5xIA^~Z^}{%4{~0ms;T
zM2&}tww)ll46tp#jA#eo+VMY$#{VhNPT*_jD@2$76QV1UL{~1uekb;qi6*Mp0p|qp
z*2I{a*Mqv{HuDfxTw7>6ibk+17#<o3xnTouphl3(Fy$&=7OF~;$lIq;pOr;@Nglc4
zB5IUT0EG8Jy0Pw8FX|Kf@Mg#Wbh$%Stc$fA>mIGdn}Vyc`q2<pa$1Xb2G?VGyA3pg
z^}$B5fFLh#cM14lD_u%s)SzuxGi*DS&Kt)VFURU(SK<xtCS65WV{yQ1XcuJiZoGMW
z9eozd_uW8yu~y?v^m}wO?W0@hR=N#K=)FcS(f8?lbPj8Pyg)BXKv74)n^*?;cq_A6
zA<szwpQ;hxe4UFWc^!Y?M=coA4ODZPvAC5;rqY>gE?+2imF#kNPj98Ke_)wY9bCR*
z<*L<dhH7iqtsmYnvT=0N<}H`hw_ZBd*ml|W9pgJMzv9Y?=2cgJ=9*pC?%s3VXRp6u
z?~OP8-p%`Nx%IZmc2`)CQEmGVOdULQ_(<o6PkQ76`bjVETo)9eWoX4P;MeKPm29P}
z(p~AVteu%bd!>@86f5QSdFMWXd;WxeN^j7Az#mZcx0-eiV=>76kY>D`<{_SqKo#_i
zO$6P&myU|{&=Nxw$14%_z(2%}Z~75zm`Qhk5#RsTZS%hWnwIf(0{6@NY?^St<^4Er
zpVMD~!m*1-)N}Y-UVGa8=6)C1ea~O?n#Jz*w@X+sX8HZ+7{?F6<-c9LU+><(vHqVz
z-)Wi>PoZ!0ul+lPPl;#XU(NyJV2DhtP-_N$03Fs0?8M?oq+KiCn0Zt@0li9r9M*@}
zJoCuRI96tSi9U}SpIunNFG%mwBakaidXgTX=jeV&geGK%1=$20A$}paDkl(lCGdAJ
z`QK3A;*V*)ln_i13vSzvoxGbG;Fh&og&@<h+T;&#p(|cQ5EeY$g_Yp~_xKj^T>1WE
z;lK>GG3;pHy%ytkZ-GpwV-YO%6F>nHDw2YDD?b<v;E(^+hF^jZQ0-Iip#9B1`z~*{
zdj}dG5XMvRkZ?4@eQ6rrln;4*g-|=>ZUOEUKe=1G-D|*=+?RJc|92ioM0J3*j$rX%
zgYy!S&kbOewpo1v;*0bJn1PUs^aX&TvM)UvL!GzmONZGA5YSUV*EQYU81^83@=m>j
z_ID7=O@%s_x7)n~9=6GS={gF44el$5r41JD3q$^xzVNo7_vH40*MQ$bI{FG?z%fhq
z6$}~%iqp`d2k>X$RBf0@U^ncYw%ff118@Ni-s$|`xvv=aWf)jqE{vrUf<Y9G8UUBk
zhdbN%sJ<c>c^1}cyjWitiV<h}Hp20w<@OcE5<!eE4S(Ta*xeX*9VMRLsdv!+21yxu
z=kj*Dcfft&RU!Zrwo-`*!-aA`*_Xk6h3EDa_3msx)N2Tb@MpA}+XwC6(N`Gjc7;OW
zNFtr&zCxi$#Ef8ORbLUDs!hTbiF&8)cCR4<l0skJ>HOa@9pNPR6^SG>DXd6}zObqo
z_Z1Ffji`vfFAxTPa<}=<($QP^Vts+KQnIfo))t5olXWy2i-MvUeZfdH6m>VAQzL46
zr|ovHA&R~r=DgGPf!7cPiX&;J?`Slg%b2n+Ga8SA`ofW5BpB`V#dyYAS>CPlpJha`
z&Pcntec;_7`^w0^%uGJZeSxXXxQV_ZvM&>-YLk88CwFVNdkrS~f=KmF=l>44qOvd3
z%ocL+_|R9(v}9jc6ftPd>MI()XkX~<VtoO{(JaGd#&SjU1%YD3Vu=_iHX{3q=`nW$
g{X`=WF7DQD_ZnhhGi*jf-s$|`(U+N%QX`i8e+h-kr~m)}

literal 0
HcmV?d00001

diff --git a/applications/fonts/DS-DIGIB.TTF b/applications/fonts/DS-DIGIB.TTF
new file mode 100644
index 0000000000000000000000000000000000000000..064ad478a510a9c581bb16d295f659256e1f920a
GIT binary patch
literal 24896
zcmeHv3y@sZdFFratKaX(?Vg#w)1%Q$k6ts9=C)=e2Bv4w3kfih1W3pO3__yED<q6$
znTKLKu^otwjDxX(UB||+z*b^RAbzb~EW|SQ#=A~z?+Ox(*LLj0yRqXq1(iyb)U)4z
z&b{3|FCf&`R&CYv%zd4E?|J<1^Z)-nB#0<M>q)0)-?DdT<HL9T?!!a}{t2a5UUli^
z`~JA|(o;m&)o{J((5=(ArJv}2iRkLD5NTTuojO^LKbHLhp2s`1|Hy5(-1_iafB0!4
zGmGn+Z#jDBksGi2(hrHOQKHL#TEF?$JAV9>bqVUb17+W?AD+JX2cLNUH+c3u*I!40
zJ{$N9k#KRnvVQByJ95AN&J@wLub}L+M~@$xro?-{M6|aH*N@*ieaCI$YWg|O_v88U
zvFTe6U;NaCAg)or{_)$6pEwy^7rBjS-&x#$<+j@o-}bHBhrWsa;XeIm;<00MKl1pm
zZb*#1N@QG)658~|KmV0~oNdf*GkWy*;kgKD@(uZp{=V5!Jb!Sufqjox&M&Foj4YX<
zJrtJ(8U?9FVVob7?Q+XGp{a+le~J!x=Yp~#=bhs%vCcc!={j+fcW#g)zT=&nv|sa|
zx5&{>H~Ydhayz^ai->owks(%j=Q^zuRqx!Og!sC5Zqi!uzr1sc656ex4q=2)6Zcb{
ztA46;)lYS<`l-%UKh?SFr#hGY+<yGn$@0ECZ##Ts`q1Gm<*j?yZJoJg=H&Fz!SbcI
zAD%vW_~!CWcb2!l<HXF#^4@pcHhp6Hozo}FYxzO>>W0mmM+eIjM~{~G%-m8xd7`}M
z@QK5>pE`UqZKvaOj80OS_R*bm8y%)2G);$a+=8;Lw3pUlKSQ@*JBd3-X%OX?((QO+
z8YPG6W?bKdclqge&<Xim8SlLVwN2xCn%*g&EYsR%jjdKU&}RIN%33E-`%yfzN48f-
z?I+~bVH^*m#Z%l`YwU~lZu0oin`shbI*KRff}<U5Wqxi;`I^JGpO`s*th`}x!|QZ@
zDPWZS#w56|k=<kvFi^|10mItR>ac8ev0md*zt4NihmPNVxO@Z%NB=equ3ry?9xLyf
zK2}~oTwXuAWy9tz8#kAM(9N=@GH`WVVxtVqDO_@AuaS+fhp<w(;O^`K9{9MPhUI<~
zoo4uNp^eyccNrI(TOD2}!MPX_tn)iuGfhL+Xzi(O<tO#NZN1a=LT~-Xt?Sk{uhnI_
z{FT{fN7H{=6PrfYtu2?Itc|9h6q~qk-CFVF_EY5@cys%7y}X|*>K%Pj>*EIwj~-Z8
zi#txK6pI-KQ8KwrocW5FegCO#^{s`c1qgV})0*7ta&N}C#FG1fED6eeNS1^Rz8gvM
zJ}S@Sa-WcQQ?fiQ&olBoE6;QCnY=8qWl2Gv7iD>e+;_@-7w^%t_J4?fT7jJb$*5C+
zF0OS342LY!5vEbnwSZL;q=|q|F%0L_&`5#C$HvFR(Ab4B-Y9ut4F4O}KNTc%&2XwG
zRl&AG{B%X*CjEWqPp{Bx=buH*GqWe@d)n8@pb*7qXKjTr&}b+Yb;8kzW$LC@iiLv#
zVWJOHk#Iq1eq*P`3u4F~yI{+<hR%spdNU28!MJ9w$)qaP5&WkDncnK~_v#Z9^-85)
z7rl*_>z|sKd32`n@<2m};?ZbGoELpw4=$~hEHi+C1p}p!X%Ge)3L3gjmWbcsH+^b+
ztT1l34V7EQKbj76(}7BL*yi0su3P_*Tb<E1RI4BSAU2AEY&*0YwBIHZ+9pLdm8h3i
z(GXSXBATEb5aoDtq+0}fD<i`R!F#T2#2%Sb&L#Ss5`MC;twgM-x^DH9JLI}UA8<!q
zcl0zDJmXfY?z`pnZa#fx{!?O0>j|+IZ@BJhSN*=@x^Hpa`<rFzTI_Y*{nhHD-b2~u
z6)?7C_D1oleo$gJK{@KA9;(n6YX^(5a6FRluvhf=rXrDWIvY+ybM%+XT~4XXNjU?d
zpkoPg^k~FM#HrHntn7^yY^Srs0A6$|1p}Hcdb&z|-A*Q5$|b{kV9LthAsj^>r^Y3D
zj7e(bog$QRocfz0(BuUVd&x0{MeLJf)I1l1Oi}}mxkmH?<|#mUI1}ipjPx*`N9>A3
zR7IW(SuSdI21WggZsS$g6%m|IRJOQ3a@`;Oe4>JqsM~mj584m*x$gN}UH5A3lfcA<
zXI=NqMCBqp-!nPznCo7^Q!talfZ{XS^V&Zzpcxl}5A?#A&!}$LBh^a4Mz?YSbBS^>
zF7$-ZE13~c$cXEXROS9@wB>#ao4Cx~RIP4u<^E}=1ukv;>%AZPFIDSY|3|OWJ^%t}
z`~o*J8#e5#R`>C?;j>)$D?Ew9t6cXg-ZqF2|3iP{v$6e+GanLK5f;)GFx)Qi+Fa^i
zM}Jy7(38{qV7aVIr}G(SMfd7;Yr<AAU<JDh9Yy>H29hynP!o>PyTa*Soy|CF?Yxm*
zlZc1(U?~-i=+>Ijx>YMndZAS5@6<cMz2)wh=<vk_%O*(#Ll@LBc@7c_^uS_D{tu0v
z8}o>viGGg|ByrFnh=HkLumjZKftP83iDEt_Fj5p%^i@V|@q~NUb<f^inE-?qkOSX(
z(RC{m;w0{B_X4yKPm`+w_D?QG6Is`NtTNGqXW8<D{Gx}X{2HvKAi8v|WDr<nj>IZ>
zqM3Mr7=q~wDwa`FK7mJvP~D`5og}#R#@d=(A`%K(v3N9{NkK59`hb&9Mx8hikTO|5
zf;)4mm@gZfI>3|vP4@Bt$q-aP%zHh69FtU=0l)$}cJ@QeCE%fn%IRwLp_##Ib#O*J
zEm@*5CgCC0SKXNlpT|?^wT)h{k>d{&_xe<A6J&TM3!+FzB56%8g(z$oVJi`J@*p5R
zmvz$dSWpj?l98~^(j=3P34H;fG>0&B?w-N;S^0N9ezyJGgJxc5vEN$(8@A{H!h2-+
zZ{4Aphfh}~S;|gUcDfJG3<1BaeJ0NUy=pkW#yBiM{~7U%dA#wq+WHmUU16iA5Dr;d
zCKijvlNr<K>dX~IJP@_hDLq=sWfOWlPg61b4k6}<KS}S&my<7#Z2+yG8&gUW;!R;e
zAp_K~QlS48Xq{ejWJvVd3Bkjt%E9QBaWNbi=^Yn#U^25~+vd8+?#%9-8SUgT@%9%x
zGP~-Vzc(|wZD&@zx+Bv+^{LE`!Rrq}0!&Z+(+xN5!9YKg*|F;ScTe?Zc1Sw=fH<dZ
zY)XNLYqh9lg#-DhohpSyP}73o^SqOhSaMRyh#oegr9`M?B{Yldv<U@}%j&vj1gA3D
zJ49mfQEy0INM0bQYS;*Tde5MF_NzbW`~J?z*{^;df0481dE@L?*X>xh110Oevi7CO
zS*Ft@n=OR|GdbX}h%Oc*Jn~Pv`=RTemiyCOApWRYMUk%w|2yu2MZux7_t8VN8zY~?
zp&^lJvndwi(54kpbKMUrUSBQFi3b2Vrxw;BE<w*GEKJsE+;hR=$pSTcs_2~Ccn%_c
zTwKq>0zJVD?;)zg?s94o<^{Iiz{JvQleOS|gAKOoaBiqxufskn%%0Os?I-Ach}>E_
zWa^q#3PSG$L_kc5<Q*iHEw%Fn^NO^^*dT(k85j811H$-b<$c$g@B8rcRWZ<b;{Ko5
zr~kIE@%cEYADThD3hHNDGYZQn4_o1rwX5>!Y%~~7#7i2m9S)UD%QC^*Mj&7mim*UQ
zbf)sru+xRXP{4AGf}PW|rH*1oPbU+s8saffj}DK35fkABJl--On<{{*#5fN&U*zDa
z2LP!ntaO0rft=3;QX`cVE8V8qDSlc%TNe*Dj@Hk<$foCncu*Rl+Sg}hE^MuT>P%fE
zDsK^q`kBcq8^5WanV30KZ~SHe@SdS8EaJx{ytmYfkh&oQhA_lB%wT{=n-YmOx@j|m
z6?>y|(hO~OMOnWvecPG=1<i-JP4BH;k~j2BI-5$w0Cf-C@q~nP*eNiKAw2E0!H|x}
zN`P>%1R1&rblK)&q6Ao(I3;xFg4BbhB3r3f3-e*x#|C`GW>q^Eo?>mivI7Pcv}HRy
z6%1K1LNl6l<vg0Q={*>s$s@$@LSacp-tbb<(82+Na(Ik-&>P|6I-)r~))VbS!y{}T
zp5)8k3X{x86FUorcekXIIivep8|u8#8NeT;tw$$u(9BJ>b(v@~kqVgsL!-D6N~fG`
z!ia=f)5W5Y3LqgtLYuNcKi-_px0B~s$U{v#h<ndw4`?a}sgRjV<ErahnzXI@EFM`v
z)lG~nPID#{a5fc7M8m*}<a8i~Sv)=;CCLQybGQ`pF%oR&=r=`k!4@Bj*6`-jT;o|a
z$~kBe8`?&?Q&HVwlB|pVuy#c!73|_`h%+h1O3Y#%MW@S_ESAfbGH*<b&1W^(vU6$C
zyPC^%Kx=6}&$Y?1`9Nq1vN>Q_D9b!p=-@!6QI_gK%<lKp`t+b-g(Fg+#DYB7a44W#
zrD(*|VTX|k%Ji5Fa;<5yvlMAE7Y{*FWZdVcCNC)=@mN#PHg0KS9OC)tJc+<sriY%c
zU024e^EHq&N7t<2es?=gPnSf}o7OljxdGE`QEV-SG~jH-Yv61tiP_RXuY<ceGIVKb
zHbKpcu`HA`Yvp`26?f)QGe}uaz;27{ug$`0s@|?RAdnV?&q8WmoSPQ~iqF#U49x+_
z9Il=N4ffa(eL>qHt(0o52#=V_#9$gw%%y+@CC!=|q^{;c&5>X*z?<}h!LZV?ZJ(K$
zfj{bh!o-5k-wq2=)f=R(wQdvs_J9K`*AW4Wt7adRrU~7I#n_bGbNK^JFgVamh$S_e
zkiLrTGnguIGZ%g$ZAkdL=%hHH?UeRKt%i9rEI_udpvtX4z|ur8fYvNCEeAn7E0(66
zKlQ;;IYp#x7r-PZg_(~n4m|XcAMe`r<BvQf8XtMNvG%10*tbT11$*_E($cTi@*y@^
zVdzJ~P5^K-OardHqG|{I$3H6^$=|GUqD@2AR-P5#oW+FVIce5@8PkuyhvhB(cyq@U
z(1lha*v3Ij8%z_<Bz7F>0qc#8t5-Vl;PMd%9{kv@AOCpQ$Hc$;$gY2V>9Pmdve+{F
zb}83PcxTwJRit;;Mv4VHpS41PR5Gom%JE1jin0Q58Oo%cT+&EHpov4!!+0@7#X>=Z
z7AP<l<-TY*H<pxsO$C(}aMTpKBgDGP_dWFJutMz-DYW(8+^~3+y$w6u7we)1$vyFN
zX)gspbdSi%=g#WPg+GBuv@-GgthyRIK#3uiklJs{*>nga%Tfn*Qy;B?n_YP(D~t)x
zx5O<=m!mEHl*-*<vh%58!$D>Tw6|3zXR`Iiw}GrX-6vf4iQ_YqmBy%agZ<fg*M0DD
zw>mJ<c!hg<8V@~Ct!j~h$<BHfWcy3*)tBr0Cn}Db{NLhMGp>7kWnzD20<&klz;<4{
zjAAX5mR{BSYumdLa2%!MNvIr{3{2wLoD@!WEG>0UXNOZN7}-opPnPmIOroV4DoK?D
zWn{(f5SBl)Y%yD#D-i^%*kiuc2Yj`ePL8E)jLahQgf{bUU;8|vC6XFfe2TH;&lRNq
zTFw=E)c9Lkjb~^roc&2{f4tM_iiD$?RI-?}^En%23BP>_fr(@a!dYnf&`P7n>DQco
z=syQe_e9u$*_iBIQPAyDU%5li$w>z|2;l2>D4U$;>X9O@HH-2I7qe{z8BDP;mEsAN
zC#*ftD2kSPyq`TmP0;iy0%o^&%fsq9TR*KJ)GsOdU>$xl=~3I^Vyd<FZuS_7P}{^x
z0;q8euB0}SS6)^~Ki?KmP*dJ6s3_A2LRi3nM*JZhO*whYbznI=ln68Lns_rNqVu@6
z=_+c|NEl(8w=0S{H9NObJeiT6d(ZFnCD(nSUVp)LvrV@byX3wO259_CyVI+k(z(xU
z^B2LLkJPTr74lH2nM4ZazrwLyDA=&G7{|$wV{|%Y%_-Y?rvQI?xYU)_Q>E^Xte$AY
zZn1zdzczNIjD7={h3qcRL8|^5=*+w>9vk;Bio!+bu<zkHEYBxe6IjoCkb;|F-W2wY
zcnIdqV4Fs0n`Op~&WOgK7ffKKF|8<Cyjp*uE*@(fsK4+J?PNS>0;`x@F}6R$^R4F$
z9j*<fDbX&_lntGU!$;ORPk6zSUPO5Lnx#1qCRlF-7(jM`q-yKDC!ZAeakm>+%xAW_
zG|K+iJar+CcwcSnic)7G2#1p}Ll(^E9_mg+V$n#fqiFZ!qS3U|3q}iv47R7r-6hJW
zl6s_+&A?hj1hvbKiC8<+DY?_`U>Ij*IoE;jDNN<Kml*1eDN#J4{P#*_DgS+Dq|zJk
z^%;-_sbRXOT=$fW+OT^bK6;28G0uWUG3b-9Kgtu8N1&+2T=z?riN-HjC_^)Se4^4u
zpNd~I;MZfdY9STLW%H0PVAV`0Zh$R|StFT<!Izs(!ROG%lom~CP(xnO)0?fg%Ul+?
zENR;W54^ilu6E2}!Io=EsT@z1DzB8%HX`I{n%<d5;gxDdN<B@(zA0)(YZvoP(Ue%c
zxAc0xsa9i1w`s~|6(`EmbZu=ifG7mw4**);5RRRQJJC=uVrA0MOr=~F)^e7nQqh=C
z_TytM*BC=c$!OINkLW$uh7GD$c8sY4Xe_7$8J;x=gzzRRjcqa-A<>_ySAbpw>K=35
z>n8_ZVSJ0HWk6Y4hTaU*m(U&bJGHTVCLe)lRw(btl?t|90OE^<g!0jJ6e$*kBkheL
z9vBCqD^7+_pJ)2;x3XD|VKqm_!76Zc=ORQPHEd=$7wE0Fs6dgx3%kzlT7(RC%2BS~
z{f*s7yJ})b&FZhM4G>H$%Pa{@QDQ;MiL<(F6Gvf?1(c>Ij@t3!<D^ZTs;NCNvP!gj
zR>i9=$=URtGMz4A|H0$Zm4NYA>?ka1{}tJFPCE1<JP##H8G&(W`W2)t*wNWhDiszr
z0~rA?Bcy;0AqI{uU^^$ra!y%2i`avV7>SYBv<9W|EZrD$tiiwMIExawOBjP6bKU<|
ztuCa%3>E0!+Jv2oDC&d#m(S($`CKWB(?T{|P$CGP!#5*;m=N2BJji6o<Y=^!qtcDV
zMF>%H*sz`JZIK~n<BfYCzGpEaR9PEh_x-!uaNZtC{=HfaTBf5ADq@b=U3MTUATZ$8
zfjN*7h##-SOf1|}cHqK>#7M+ha-^b=9FZ{fIx&%5s9|_ErlzK_dzxal#b^?(Ejw>z
zwtY&*arxFHsTJq6Ay?GoOh$#Ji1~ce{DXL5HRI3HnwbF~Znp*-FEdNd$2ZSKJK*KM
zIG!)sFkiFDw3>}Fu8YM&Due)uQ&fh&%$sNvIgmngbWPiohoRt^_AOtGCvH56gM(z<
zxdKO-4@$UlszlEm6JA{IK>_A_Q|fR?L@JzsKTU1-+_mF-6+K?Nx{oow5Z{&l(yEm!
zS0ytd+UKl}Ijf7#Y9IS6O5LG03K*z>22gL9Rt@y(ZDe4}geIkJ0RhA_nxw$tL_}2Q
zN!~eRp@;$^c=a2ki4mj?<$7O-GL9g_V)xy<1QGz=?YcYR`)D|>d+|i&!%Z{fKR{vq
zB~!=g-KQb1Tk~VzThWQMxlhzC&l+9bC6Gr)o}CoPwvc`aGNFo`?rvz<uqM(D&Xw;%
zkX0;{I`i?E^4KCQQtqZ8$JLr%__+XUYALC(T7<I@YM!B{R8|k1cnTK!yjfuz9er>2
zw|D<byB*z5OW)r0t=+FlPp#-zi=IAMo9xQ7g9=`QOg0@5f`VZRW|PTi9C7!6<FLz0
z`3yLPAQdfTkbQyxm<0zQqS7%DZ<im<+#@EM=DgTr8b8};UaApmY$whlq?KW$293-J
zI8-vqCI4!BYEeT<4xk%}MQ%(5#W;n;%O_G&gD>#J1H6X@@0~rb#hSSW2WzXMnAc0&
z$_RvG@tBj0A}g5l?}%e9x|T{3TXxMrdrLNOdcg)y_xZjmMkhv+MhIQ?UZoO@C@<dS
zEbo}{pQ|=N!{T?=yY634R89>{B5COT(wjFQKkTQ!tG2llF1DoIG6VxbD;%^sGO!1`
zlsyP!$i#qBo|6QS6tSd1=*#uRgu1k#IY$D6=$dC7a-NFkn&@Fq*Fp;sK8bo`TOH{u
zzMTkbQORo->W|LAY8jAL?(7BWMf5!oZ?9EBaazg@Ab|xTuxKO_&4w)zMdC_4X6BM1
z-9$hL@jP%Ua(pm4;3S}kZhj<>nNP~7ruVi@Uh}~60Z&z#0Zda(q^ndt#{`R@I+GQ|
za^OGh`s&}v+@#M`CO<mUn5>B6OyjNJ+V!o*udCJbm3I5fw|NW5YJ9YIRVE0Mg4v6~
zbQDW@8y=`aDU57COLWDOcGz*Ci*mUHa<?!ZDiwKnfzA$$52iB0L5XYv6Cw;VpNsj!
z^sL9%Fu^V)JtJ&U*OM~A{Gr2pnzCiGOG5559L;ApY8wdl0+XNP^o|F-VE%-{B70{<
z@63hg5WKG(cirz+tBrTnvmDo-SU88V=|pW)HW6g6G%_|!j-IDdNyIzZ>7L62ujvft
z!+O}x!{aZRj$~#Jf(cMWoNsE6okOy#FFlt!O1%U87I=`=5jOMLmtE3e?t^r5xDsp6
zj6L`VAG`QNXD<2JA3QjAM!fyRi${L`^CK^wsGNA=$j^Ru<i*<;vn)D!3?1<hv_&Ku
zPD$%ZWK&L8vEX#(4bIsO212GcBTE8<z+D19CDWIS3uk0XFwU8ku|eU@Ub;+=Mg>pI
z6zj=ESljcPN=n@Zay{s}aBxc>;4vu!JtQH~>QN+Iu-sGZ2u!iv)S6=&@Zodb*rD3m
zR1CQcy2xaaPgsg45^+pwf}s$i(jpekLi}^?SPJ7xr&Nf?w+E!_VLn1yN=*_SCyB9K
z@5zDIMuv5uV_ald&O?MWeo`0r?42^+w(H5Oue@ROXN~tmFlIPa?7OGKBOh;k<6S%N
zGaHz4a?kxdJwI3G)|zY3qW-QlPjWa}<BfW{q^Zl!AvWRt92gcHN;Rh!jXa9U8D!|u
zEX@o~N|n5bPm3coHZqCyRwj`^pvou(G=$A8T})Z8lSw4OSJ7?coP$5t6synQFQ)px
z8NW{a^?|t9+jrqV-Y@o4ZNSIR2vqaa6ahX7<oQ`~WS1eUB@mQR023J>5<aP<6OV-1
zsEL8?fL%(&-#|W~gb(K3rkr<s@bN>^Z1!_l+SVaagk2w+5n04S5dFvkOha-i$2oxY
z3rAUez}LdjGVulRmtO2;dktd33(1i{iWVS;$ecz~GXFWpLP2s9dp71%p6X1PnWs{<
zb3N6OTtK}4&Tlks{PR1-7w&xU8@qOW<H0-m`=8lunl<}*c&gg-$s*F$XY!R6J?E2|
zQgv&V+qemY`H6cZ4{`2kgI>~p0GRjHV#u`Qc$<Nwu=t#$Fw6xZkC4lcB*`H$@=|8U
zo&(yp9hozjuePjHg{|f2jf%J}M{ndDa5;Lj9KDgD+vVsDoQy3WOq;`0$!WIem!mgo
zw#lJ3XsPAsjq;dELCW^ma`eVsj@~RsZ<eDsw)Z{ca`eV?HF&<h<><|F^rqcw!$#F|
z^kzAFvmCvFCu=!+vmCuyj@~RsZ}@9!Kc8+ndh>6M-k35v@rd;DtNhqby0dnXji9`U
zB%;L*BqJh*hLB)debkJE0~?XGB);S<#q%jFSb(Hhgh-JTTR>9moRkAEDb{xrdBGen
z?&}#g%~(#C%rj_aqpP)MRK8LMrNiQpnIR;baG?Goceq*|M&xXw^2ZNjJsFk!s@5?0
zDMA5_|1Bfk9OFj%Bi%K-jedd=>eO9}$nVRMho$4A?$+uMFs%49b2~Mhy6cv&>F+%M
z`-nks+-5Fh?yF4#D1b<@qhzNGsZ^o0oP!4-9RMmQ0L4oY37V1w69PHK!eVipA*>0&
z@ni2_MZ1<--GQNo&UVS*g}=gsw<Lf{<8NE%VPc@_9;*(!?(k;*uBElygG%K0L519%
z|A8eR_<MX|h|h>WM54l)S~?0?WgWu^I`NK(F-0AI5M}(r_yvS?WE9bsA9r0MhBw;z
z;f<R{a#rt(K~d@NU11Gn*A`dR2FhI<yUJIjyPdwyOdbu~VCTA0CFjr@d=VTg5WXA0
zdNM}O66?u~(>APB@LT8{f2i=L)|FXu<(W6R-ppn02$rc)|IKA;-t;OoS0a;g!9q0(
z7j5gPfEzV|91EROSL6P&8m;7m<~l08at1F9^bmidaQ;3p0AsbSCgeb<Y>j{WiYyD(
zra~3?I7N0Wgj5M+Y|)fuB3&TfOs?gq;20LhQu%6QvY(2C!<^gCUFAFgTlC=rlMnfw
zzK=UDBar{^*23b|DDD&A*VbbttbEO~NDqdxG(Ju$dKqOxcQ*}h%;&P8@&2K^ui1as
z-Pc}wm$?4gyYJeMjl^t2Y}V4^y()f*4;cY!$He*(%-_l{4&Jp43|xLeD1O=B-x%%h
z7w_%E&qAbE{%*#;efv9*!#;px63Nc@BH+FU$q;{p4QpmvI0lXp1^44)xt-YXIbi7f
zSft6qc9bZBwKO9?B8oms6#E324eus4VFftUmBe%SG7ul9V0#<3b3|E`XP>}Qksa7x
zA<92aWMeC!j^YDE9iJxZd=I{VKpiDK>)>7IE~0Lrs|;o?<Jq1s5v}+*Q7`(o@;;)z
zZxi)nP0q^S5DlCrT6H7Q>H(rPmk?Gn#D==pJxnxs6}GPs4PpJvp`Q}1UxDoi(FSZA
zQQt7`yLh*X_p0dI2;Lp}E27a+Z2wHO8U5daXD>QQG<F@)#em@@yNJerkEr(Vi6+p-
z<QAe!9c+I`wDmmEwr7a8qwXEo5bZ=gJJI%KH(`5`XxBdy?MB(|-^GUdFGt&#f11}7
z1!c<L?7i{}Ty@dDXU{gU4|auCK}<P-@6EC54=5{&F9PHEDlQ52lco&55y(N>+Q^D7
zB2%^#U#FA^pQ9invIpO(^wLUv?bwes$_B71+iHCGxE4!E4uVqG;|s`*Sc1^SN^P3}
zicze+wFPUojnTz)35`<?tG7*JiOH>ypWCsH+fKR+OBe2@%jpVy1^*Vh3Nm#M7MR>e
zS7VvOYiU2$*}M*Gg}#-h=z6+=Zp1Q&&(oLb&*+PE4?RUsV?-W(ij^1}|GH3uGhCCf
zc);oh{%78u`@>D$Fs(o^6pln=@kBC}&SZ0WyHM=t>_Sdwxo1W1%D(=}z^c`2)~*{I
zTEAiAuv^_UGP-%oMPnCVGG3dQymaff?K^f}wrlt0S6um)tA1<G-hEeJbM5{E*ByN8
z)b%&qINfURCja!%&4-WNQunXk_|a#?H~xnAq#^P+iv!ylu>P)cwp=Wi$}7t2XJ=7f
zE@yDv)x2(151CGXLH`wh-^Cw8`u}ITxdkiw9D)SnWl(PdeUsMDYd3;IAZ_Hfg=Qep
zhO|d$3*_w<UEp{P==XW(ssnPfH5>O$P&&$*M=V;w&kMZIM{HMmM_Et%HN675$0qJo
z@5>{<^EuV#y!ZU~<*}{)Ips^d(<{r1UcLU2`-!n=U4(DZu{ChcToT4n^?k{@|F5<G
zB-%YlcL_dzpN5Qs3dR!7fWO26^q!WTZD?QDeyBaJ?GU^k6X^%F4YXVQa`yAkW(4~o
zJ9`dR^hvA>@Imo~*&FGZ*%SDd-Vm$BW;|b~N9a-d96bj~^&Dhb9a85h$fo-tBS`$C
z%<R?B!UsSE_p77$CT*0Gf(c{M-^9_AymW23yk0L8bO=`d{2Ctg#EdXdD;@7bZSQC;
zwo0pC{b*2|#cv!N%J-~a-+U#Bw~vPK$t;qWbbNwu;$RpW{*Vjg{cy|Bam)KP%lR7X
z1DL!n2;N!xKHm;;14umzhgDmcFmPKYR=LBQ%{nzS44JrSU9u2M<nTM*ulb%|fb|2g
zj+`0{MUeKb;|D9uYPMzRmiLowSpomDS<VIMg2l`6pleItM_XpZYbzR+ZDF|)WPlm?
z)=I-)t1XP(GL~pdHo&*r9`_q<9vRM2^cYW3EEd4HFbfHUT5Sce8jtz`s(^0LzigIs
z0lI)idt_Tn-)9WOytd+Tzb!0MWS}l&HlqgrWweE5EJOZfv)nJhlNdJMP``!uxh;I;
z!!=VP5kz*HVIrF|2*|6pf=1B$$+m(a|FT)m1sHBH$Xy9Ei<>9S`;5~B(;V7LCb=!s
z3<o2~X*PpFtRRINnw1-%8Fa(@HOu`1JP8mtn^<@sZ3U9DtstdRA=Oq03rqNIh0Kum
z15SfhDC}Q0%eer<1$T14jHU1Muv6StFi7dNY|Dy-kSJ?e$d(N1sKI|(wH5I%o8^8%
z$O>6YwI!bc{RHtjr)330l*xp7Tv)n6wH4Cw7dFG*4>%14!cqUSS<VF*Za9p#(6y!S
zb6Xj1D-@z^HcXIDRuqfUf(P&wa>zgp{>$^)YL@%=@nm~j3-1Hl!7OuEC``FrM79-+
zBqE^8aM-|K#EN)7&=1x}iTRh!axMttFM`_;b4%X`t%h>kRya)gylg9gwxVb&!fior
z`7Z;fpwPH~*(~=9A_26vL|gKiP#_e_Gk1j}WZRgrfmY)9l!V)g82F0@qTUZ}MS{_|
zf7vYOf(ZViXbW9i`aaqU+uT+pLWM$<EXbNfG!+AzM7b^0;J<8|A*{TT@GqO?enB)C
S4KCG|d<J|GF37f`^nU;-1?c1e

literal 0
HcmV?d00001

diff --git a/applications/光功率计数码管字符识别.md b/applications/光功率计数码管字符识别.md
new file mode 100644
index 0000000000..33c515a5e9
--- /dev/null
+++ b/applications/光功率计数码管字符识别.md
@@ -0,0 +1,454 @@
+# 光功率计数码管字符识别
+
+本案例将使用OCR技术自动识别光功率计显示屏文字，通过本章您可以掌握：
+
+- PaddleOCR快速使用
+- 数据合成方法
+- 数据挖掘方法
+- 基于现有数据微调
+
+## 1. 背景介绍
+
+光功率计（optical power meter ）是指用于测量绝对光功率或通过一段光纤的光功率相对损耗的仪器。在光纤系统中，测量光功率是最基本的，非常像电子学中的万用表；在光纤测量中，光功率计是重负荷常用表。
+
+<img src="https://bkimg.cdn.bcebos.com/pic/a08b87d6277f9e2f999f5e3e1c30e924b899f35a?x-bce-process=image/watermark,image_d2F0ZXIvYmFpa2U5Mg==,g_7,xp_5,yp_5/format,f_auto" width="400">
+
+目前光功率计缺少将数据直接输出的功能，需要人工读数。这一项工作单调重复，如果可以使用机器替代人工，将节约大量成本。针对上述问题，希望通过摄像头拍照->智能读数的方式高效地完成此任务。
+
+为实现智能读数，通常会采取文本检测+文本识别的方案：
+
+第一步，使用文本检测模型定位出光功率计中的数字部分；
+
+第二步，使用文本识别模型获得准确的数字和单位信息。
+
+本项目主要介绍如何完成第二步文本识别部分。
+
+## 2. PaddleOCR 快速使用
+
+PaddleOCR 旨在打造一套丰富、领先、且实用的OCR工具库，助力开发者训练出更好的模型，并应用落地。
+
+![](https://github.com/PaddlePaddle/PaddleOCR/raw/release/2.5/doc/imgs_results/ch_ppocr_mobile_v2.0/test_add_91.jpg)
+
+
+官方提供了适用于通用场景的高精轻量模型，首先使用官方提供的 PP-OCRv3 模型预测图片，验证下当前模型在光功率计场景上的效果。
+
+- 准备环境
+
+```
+python3 -m pip install -U pip
+python3 -m pip install paddleocr
+```
+
+
+- 测试效果
+
+测试图：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/8dca91f016884e16ad9216d416da72ea08190f97d87b4be883f15079b7ebab9a)
+
+
+```
+paddleocr --lang=ch --det=Fase --image_dir=data
+```
+
+得到如下测试结果：
+
+```
+('.7000', 0.6885431408882141)
+```
+
+发现数字识别较准，然而对负号和小数点识别不准确。 由于PP-OCRv3的训练数据大多为通用场景数据，在特定的场景上效果可能不够好。因此需要基于场景数据进行微调。
+
+下面就主要介绍如何在光功率计（数码管）场景上微调训练。
+
+
+## 3. 开始训练
+
+### 3.1 数据准备
+
+特定的工业场景往往很难获取开源的真实数据集，光功率计也是如此。在实际工业场景中，可以通过摄像头采集的方法收集大量真实数据，本例中重点介绍数据合成方法和真实数据挖掘方法，如何利用有限的数据优化模型精度。
+
+数据集分为两个部分：合成数据，真实数据， 其中合成数据由 text_renderer 工具批量生成得到， 真实数据通过爬虫等方式在百度图片中搜索并使用 PPOCRLabel 标注得到。
+
+
+- 合成数据
+
+本例中数据合成工具使用的是 [text_renderer](https://github.com/Sanster/text_renderer)， 该工具可以合成用于文本识别训练的文本行数据:
+
+![](https://github.com/oh-my-ocr/text_renderer/raw/master/example_data/effect_layout_image/char_spacing_compact.jpg)
+
+![](https://github.com/oh-my-ocr/text_renderer/raw/master/example_data/effect_layout_image/color_image.jpg)
+
+
+```
+export https_proxy=http://172.19.57.45:3128
+git clone https://github.com/oh-my-ocr/text_renderer
+```
+
+```
+import os
+python3 setup.py develop
+python3 -m pip install -r docker/requirements.txt
+python3 main.py \
+    --config example_data/example.py \
+    --dataset img \
+    --num_processes 2 \
+    --log_period 10
+```
+
+给定字体和语料，就可以合成较为丰富样式的文本行数据。 光功率计识别场景，目标是正确识别数码管文本，因此需要收集部分数码管字体，训练语料，用于合成文本识别数据。
+
+将收集好的语料存放在 example_data 路径下：
+
+```
+ln -s ./fonts/DS* text_renderer/example_data/font/
+ln -s ./corpus/digital.txt text_renderer/example_data/text/
+```
+
+修改 text_renderer/example_data/font_list/font_list.txt ,选择需要的字体开始合成：
+
+```
+python3 main.py \
+    --config example_data/digital_example.py \
+    --dataset img \
+    --num_processes 2 \
+    --log_period 10
+```
+
+合成图片会被存在目录 text_renderer/example_data/digital/chn_data 下
+
+查看合成的数据样例：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/7d5774a273f84efba5b9ce7fd3f86e9ef24b6473e046444db69fa3ca20ac0986)
+
+
+- 真实数据挖掘
+
+模型训练需要使用真实数据作为评价指标，否则很容易过拟合到简单的合成数据中。没有开源数据的情况下，可以利用部分无标注数据+标注工具获得真实数据。
+
+
+1. 数据搜集
+
+使用[爬虫工具](https://github.com/Joeclinton1/google-images-download.git)获得无标注数据
+
+2. [PPOCRLabel](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5/PPOCRLabel) 完成半自动标注
+
+PPOCRLabel是一款适用于OCR领域的半自动化图形标注工具，内置PP-OCR模型对数据自动标注和重新识别。使用Python3和PyQT5编写，支持矩形框标注、表格标注、不规则文本标注、关键信息标注模式，导出格式可直接用于PaddleOCR检测和识别模型的训练。
+
+![](https://github.com/PaddlePaddle/PaddleOCR/raw/release/2.5/PPOCRLabel/data/gif/steps_en.gif)
+
+
+收集完数据后就可以进行分配了，验证集中一般都是真实数据，训练集中包含合成数据+真实数据。本例中标注了155张图片，其中训练集和验证集的数目为100和55。
+
+
+最终 `data` 文件夹应包含以下几部分：
+
+```
+|-data
+  |- synth_train.txt
+  |- real_train.txt
+  |- real_eval.txt
+  |- synthetic_data
+      |- word_001.png
+      |- word_002.jpg
+      |- word_003.jpg
+      | ...
+  |- real_data
+      |- word_001.png
+      |- word_002.jpg
+      |- word_003.jpg
+      | ...
+  ...
+```
+
+### 3.2 模型选择
+
+本案例提供了2种文本识别模型：PP-OCRv3 识别模型 和 SVTR_Tiny：
+
+[PP-OCRv3 识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md)：PP-OCRv3的识别模块是基于文本识别算法SVTR优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。并进行了一系列结构改进加速模型预测。
+
+[SVTR_Tiny](https://arxiv.org/abs/2205.00159):SVTR提出了一种用于场景文本识别的单视觉模型，该模型在patch-wise image tokenization框架内，完全摒弃了序列建模，在精度具有竞争力的前提下，模型参数量更少，速度更快。
+
+以上两个策略在自建中文数据集上的精度和速度对比如下：
+
+| ID | 策略 |  模型大小 | 精度 | 预测耗时（CPU + MKLDNN)|
+|-----|-----|--------|----| --- |
+| 01 | PP-OCRv2 | 8M | 74.8% | 8.54ms |
+| 02 | SVTR_Tiny | 21M | 80.1% | 97ms |
+| 03 | SVTR_LCNet(h32) | 12M | 71.9% | 6.6ms |
+| 04 | SVTR_LCNet(h48) | 12M | 73.98% | 7.6ms |
+| 05 | + GTC | 12M | 75.8% | 7.6ms |
+| 06 | + TextConAug | 12M | 76.3% | 7.6ms |
+| 07 | + TextRotNet | 12M | 76.9% | 7.6ms |
+| 08 | + UDML | 12M | 78.4% | 7.6ms |
+| 09 | + UIM | 12M | 79.4% | 7.6ms |
+
+
+### 3.3 开始训练
+
+首先下载 PaddleOCR 代码库
+
+```
+git clone -b release/2.5 https://github.com/PaddlePaddle/PaddleOCR.git
+```
+
+PaddleOCR提供了训练脚本、评估脚本和预测脚本，本节将以 PP-OCRv3 中文识别模型为例：
+
+**Step1：下载预训练模型**
+
+首先下载 pretrain model，您可以下载训练好的模型在自定义数据上进行finetune
+
+```
+cd PaddleOCR/
+# 下载PP-OCRv3 中文预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+# 解压模型参数
+cd pretrain_models
+tar -xf ch_PP-OCRv3_rec_train.tar && rm -rf ch_PP-OCRv3_rec_train.tar
+```
+
+**Step2：自定义字典文件**
+
+接下来需要提供一个字典（{word_dict_name}.txt），使模型在训练时，可以将所有出现的字符映射为字典的索引。
+
+因此字典需要包含所有希望被正确识别的字符，{word_dict_name}.txt需要写成如下格式，并以 `utf-8` 编码格式保存：
+
+```
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+-
+.
+```
+
+word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，“3.14” 将被映射成 [3, 11, 1, 4]
+
+* 内置字典
+
+PaddleOCR内置了一部分字典，可以按需使用。
+
+`ppocr/utils/ppocr_keys_v1.txt` 是一个包含6623个字符的中文字典
+
+`ppocr/utils/ic15_dict.txt` 是一个包含36个字符的英文字典
+
+* 自定义字典
+
+内置字典面向通用场景，具体的工业场景中，可能需要识别特殊字符，或者只需识别某几个字符，此时自定义字典会更提升模型精度。例如在光功率计场景中，需要识别数字和单位。
+
+遍历真实数据标签中的字符，制作字典`digital_dict.txt`如下所示：
+
+```
+-
+.
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+B
+E
+F
+H
+L
+N
+T
+W
+d
+k
+m
+n
+o
+z
+```
+
+
+
+
+**Step3：修改配置文件**
+
+为了更好的使用预训练模型，训练推荐使用[ch_PP-OCRv3_rec_distillation.yml](../../configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml)配置文件，并参考下列说明修改配置文件：
+
+以 `ch_PP-OCRv3_rec_distillation.yml` 为例：
+```
+Global:
+  ...
+  # 添加自定义字典，如修改字典请将路径指向新字典
+  character_dict_path: ppocr/utils/dict/digital_dict.txt
+  ...
+  # 识别空格
+  use_space_char: True
+
+
+Optimizer:
+  ...
+  # 添加学习率衰减策略
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  ...
+
+...
+
+Train:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./data/
+    # 训练集标签文件
+    label_file_list:
+    - ./train_data/digital_img/digital_train.txt  #11w
+    - ./train_data/digital_img/real_train.txt     #100
+    - ./train_data/digital_img/dbm_img/dbm.txt    #3w
+    ratio_list:
+    - 0.3
+    - 1.0
+    - 1.0
+    transforms:
+      ...
+      - RecResizeImg:
+          # 修改 image_shape 以适应长文本
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    ...
+    # 单卡训练的batch_size
+    batch_size_per_card: 256
+    ...
+
+Eval:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./data
+    # 验证集标签文件
+    label_file_list:
+    - ./train_data/digital_img/real_val.txt
+    transforms:
+      ...
+      - RecResizeImg:
+          # 修改 image_shape 以适应长文本
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    # 单卡验证的batch_size
+    batch_size_per_card: 256
+    ...
+```
+**注意，训练/预测/评估时的配置文件请务必与训练一致。**
+
+**Step4：启动训练**
+
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```
+# GPU训练 支持单卡，多卡训练
+# 训练数码管数据 训练日志会自动保存为 "{save_model_dir}" 下的train.log
+
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model=./pretrain_models/ch_PP-OCRv3_rec_train/best_accuracy
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy
+```
+
+
+PaddleOCR支持训练和评估交替进行, 可以在 `configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml` 中修改 `eval_batch_step` 设置评估频率，默认每500个iter评估一次。评估过程中默认将最佳acc模型，保存为 `output/ch_PP-OCRv3_rec_distill/best_accuracy` 。
+
+如果验证集很大，测试将会比较耗时，建议减少评估次数，或训练完再进行评估。
+
+### SVTR_Tiny 训练
+
+SVTR_Tiny 训练步骤与上面一致，SVTR支持的配置和模型训练权重可以参考[算法介绍文档](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/algorithm_rec_svtr.md)
+
+**Step1：下载预训练模型**
+
+```
+# 下载 SVTR_Tiny 中文识别预训练模型和配置文件
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_ch_train.tar
+# 解压模型参数
+tar -xf rec_svtr_tiny_none_ctc_ch_train.tar && rm -rf rec_svtr_tiny_none_ctc_ch_train.tar
+```
+**Step2：自定义字典文件**
+
+字典依然使用自定义的 digital_dict.txt
+
+**Step3：修改配置文件**
+
+配置文件中对应修改字典路径和数据路径
+
+**Step4：启动训练**
+
+```
+## 单卡训练
+python tools/train.py -c rec_svtr_tiny_none_ctc_ch_train/rec_svtr_tiny_6local_6global_stn_ch.yml \
+           -o Global.pretrained_model=./rec_svtr_tiny_none_ctc_ch_train/best_accuracy
+```
+
+### 3.4 验证效果
+
+**本例中的评估集和模型路径，可通过填写[问卷](todo:add link)获得**
+
+* 指标评估
+
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。评估数据集可以通过 `configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml`  修改Eval中的 `label_file_path` 设置。
+
+```
+# GPU 评估， Global.checkpoints 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+```
+
+* 测试识别效果
+
+使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。
+
+默认预测图片存储在 `infer_img` 里，通过 `-o Global.checkpoints` 加载训练好的参数文件：
+
+根据配置文件中设置的 `save_model_dir` 和 `save_epoch_step` 字段，会有以下几种参数被保存下来：
+
+```
+output/rec/
+├── best_accuracy.pdopt  
+├── best_accuracy.pdparams  
+├── best_accuracy.states  
+├── config.yml  
+├── iter_epoch_3.pdopt  
+├── iter_epoch_3.pdparams  
+├── iter_epoch_3.states  
+├── latest.pdopt  
+├── latest.pdparams  
+├── latest.states  
+└── train.log
+```
+
+其中 best_accuracy.* 是评估集上的最优模型；iter_epoch_x.* 是以 `save_epoch_step` 为间隔保存下来的模型；latest.* 是最后一个epoch的模型。
+
+```
+# 预测英文结果
+python3 tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.infer_img=test_digital.png
+```
+
+预测图片：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/8dca91f016884e16ad9216d416da72ea08190f97d87b4be883f15079b7ebab9a)
+
+
+得到输入图像的预测结果：
+
+```
+infer_img: test_digital.png
+        result: ('-70.00', 0.9998967)
+```

From e1f947b961596623b7340700f8881938300833d7 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Thu, 16 Jun 2022 09:35:09 +0000
Subject: [PATCH 018/134] add case

---
 applications/imgs/det.png      | Bin 0 -> 66713 bytes
 applications/imgs/rec.png      | Bin 0 -> 13765 bytes
 applications/imgs/sys_res.png  | Bin 0 -> 291470 bytes
 applications/液晶屏读数识别.md | 577 +++++++++++++++++++++++++++++++++
 4 files changed, 577 insertions(+)
 create mode 100644 applications/imgs/det.png
 create mode 100644 applications/imgs/rec.png
 create mode 100644 applications/imgs/sys_res.png
 create mode 100644 applications/液晶屏读数识别.md

diff --git a/applications/imgs/det.png b/applications/imgs/det.png
new file mode 100644
index 0000000000000000000000000000000000000000..6f21ee170ed272e6ae9c3ef475f60855a064cbbc
GIT binary patch
literal 66713
zcmYhj2Rzk({4aivWABhXLnv8C_Exe&^fj`HlgQp%va%1d9m+^{$lg0Lv-jRBd*Ao@
z{qBAI{~nb(j^lIQpVxSf*Q+oMwHHMAwD=GRgh)y8nI;4RRRn*(#>EDI-WQQv0iQ6P
z<dk0Hf)8)px1r$Qc=n2VP7nwI8TvPLc-HqCJSgh?T-RC4&eGY<)X@U+!PME_#?INs
z%8c37!qLgf&X!q#M}P;;&HT>U*<Os7_y2nWkDViemzClf76if!QF<o(+C6QjLC?l;
zGG+d39>GR}h>d^tp6z*TN{;4e(O#pl&rl?@?C`tHSgI-p=5Mj7IKm$x9-+&5%)tQ?
zY{b`ZSFrI&Vf<p)WUP^y_qzCF8oQs*)4u*M<+*aUrRLXXu@5{HGaS|fE}z_<krjL%
zt$L;Xv;rc_HX!9yvUqhne@8bqrIeEEIX_o*w(H|HKWAIxH_71sfB&g7?d9+Nzkk0<
zIb~#?|KBGy<~HDT|L@bQO04#w|IptEzGOZ1|9^S|>@AY*P65d_qJ*4LL9W>O%@6qf
z?~6@at27RXb+ak`Lt3)kSR*qOFI9)<6-SQ8&c?$2?+Z)*+E11B<=!+($#t?y8DqaG
z%a%URxH@QTu-l)$`jw}Bo`G^x95%nY3Co@j<yA(mAdtSNYvUwG7xhTC!+}b=qtz6_
zm+<lP`Q?4T5ufYi`7zfzx9gfS*2L*wFKZ0cl#rk9YF<d8ba)a8#&fJvYF;4O-d9a0
zqncFw8iv--@2cm$Xs(2P*0<Nh7*mZCdskf<^k<B3JNLcxLhP<?wl7yuyQgU-UZ=J`
zceUG1N}Wx;Q|IuTd0SNF^sXw(ps}&{$W=elZfo3j{x1BNoxa<xUF93_BjztN_vx6=
z{tgS?)MTSnG|O~VuP!e!G!!-KW+^lAN-N$J9XJ(fYh`KRj@y3tfHlaKpq#wVLhtM6
zi};%Q$C8v;YW>10<;Upze9ipt<>gvIa<YbohRE!_J=>GxWAr1dj(Qztul`tSU3eQ6
zX+NH68hVF*_{39EqHfl<0DPwg=Ud2E@Y#UJ+Q~_)!;GKx0DQ-Zry;V{J*hz9Th_$j
zpnDXI@C^Fhy7JBWEZrGG0-PH7QQ$@TNPHZa?W(3t(=NQF`uG)~2a{7%_|`3V*E2Rk
zqqbH5JxcidzA}DbJrKEm<T3jGX+p0><(nc^KfN}=m;Z56{IwBsd-@ezSod#R?Ed>h
zEnVwL9_tx}(oJ2#Q|p|K(gLbYbdjC<*4?`FxYvz^_m8Hhrwg<Ta{l~znPc_CvYck+
z^V9tC?ZtI)p_b+OCnql8jc=^>CW7OAFG#zX5rJ|T#(A53a_HdYWPf@nOdH9p!V^;H
zWwX(^*QkVi=izZtG~i}EFDxUK;N0#jeZ6KkKVe21fz^LM1k3W3xR}`A@$p2ch>~oU
zZoR^{dzh$wH)rSX-@eI9N=p8C{RsS09(kRh{M^UKCyQj)RUe!+QUIe5uJ@Wsq&Z-{
z>>Ss<#z4FB4N+#W$zcQaAd>CXw;}g!kG{xxj>KM5Sr}wkM_7;;H%F7}7&TDc1Sera
zWgjFzu7^S9&s<Kg_))i$wsV-~j-5pI|5S0_XW(=@J3C{-uuR}x;Ly$Mb&c1;r49(2
ziZ$~;?bn({C#TUlGx7dRVGF|#lG}j$nG+ds@ClG&8F#tX6`7S$cUITb<lo-<{#P(+
z&*U8FCGf@P>eR}|#}Hh3TU#44J2p19kNyLQw^@1KvGMUXdJ{#%elz69dJ(HEyNH(6
zFhrJ0q3F%$<O4S=aBjX2H$jG*Phod9@odh|Z71H-rELvBvKZ*=>p!9l2M6@N^39L&
zZ4jn3Rh3m$I^f{8wzlvIOAL5$PB}iPb1saIj@~a+Bjnj5tHbIdgBclwXkrL@t520h
zfH+K0X6G(07sO0DR1t4ou<4onSi!o{$Bw^dE6!b|%N>k>Va_3NG;kP&iTg9?NlA~B
zSR7EZ)KARUJo?+~FI<jkG#oSYN8dv^V`wHy1|D0l|K2*H4}ADJy48%Eo|%7zctNV^
zFiq6yt=h%>c+b`F`^vwQlQb()9^cfjKUiC<fIH*8w$S&~|LpS8GnfT#w5B*EBzTNR
z0~<{8f)r^N<ZBmT$rO!mgMb6?Q?|=&jV|z^qk2=ypIq9=;NPo`N#rC!BK~_hJ|&Rk
zTr0Q|QW|+Yz_!YMqi&Tlcl~9*-Xi&8wxxyYM_rvhh)Ix#Adp@}(GMLQltwX0XS|YY
z1l0-7Be$cr?kHMRQetglBD1}{y|uA{E2sG=;iJ7t`eyj!ka@V#O44lg^Wc1u>KcL_
zf+*%8dg_@E>*D+)E*24nm1bp}O5ipG6m`tIK$epGZ~3vMs*XpJ$!+cn*XVLvu7x#<
z##tB?7QvYzVYlbrnSz`}C1E8&<%~E|I4^KK5F*E2G~PB4`e(C2uB_WX_Z(OIx?6BB
z4(*ALT3tZxFi@9H=8b6Vr7~aen!~wdsGHcU;U7>!et-Q4i+Q<r;+y}u4G1if{5M6|
z!*RW<9~}ZMhK2o#=dbGcdiEu;lXEvw%9kVi{}`yax#(V)$$}~_Bqa2AbTmgUD-&cl
zgNVHhIXU@Mm7Q^EBPUaHVCB&Sc6byi3_CS7wFL<k0JWrAV1|5g?RIag8l&2}jE>bY
z)4nQY5Zb&D(K;C|O~CG89j*a0;-l3O>r^S^)#fao{NO|XZkF@BoS~k2du&WQi5Exr
zlE*@pu%lUq&sLbbzm?5`kYp#l@#5Tkd1s;9$6(X2h6WYlWFNPEQY$7d9>{%Z@IVVv
zMHiR^<@cKz*Aqo`Z8JopgMA=Fve$br3(_Ps7CRws+GrrP5kg0T_)YwY$8DxbgA-jz
zK=lB3-Nb~sFQ?+qM`2;%lckRE2u9K7F2>(R(<EjbFD0zX5;~thO6+|^f&WHNZ)oV-
zi<6-jOjh>x8lbR)TAZGqZrW<T-Se?bgELD14drmMa>1x({%icqjpA>QU-gxm$)0vN
zb+%H6Nw(n6oH%#LaHhSi{U=7gXa~ZZnzX7`ijq34d~xD_xS>9dNCuXc8&!f-ZEw5#
zMI$^?mqrh6EI#IBZT*-$oSldAjHQ_NKr{k6w%q4mHs4003_6sF)}sUgpJVsZURar6
zqR(Y?Rjcj{rwN}=k>T3;jDFqh*vyQ9yEeVLr9oqV<1d4{%F2Yzyb^1|pQXBIoab~A
zQkI?l-sZV)K+Bk)FB!>G-);{hH<gODhzOzk+wjYck(mKg;%W^dCy}SY`Od}VzpiMe
z@en3o>Q5;h2W$%ElZ1B$-RZLYW&K~%@WeDc>xnuPICDP8Ztm>#Rlnk7`@(0bJv_~f
zB{H_{vU+r(N*}(>uwsDgR=#yCY+1rtd*&dFWqNxvSC{~I_AAEOxTn1oP^7KGz`(Fr
z<L_U$QHUy66;|%QRn9>w=V+cYWSZUT8l&ulm5C!9TD#)dYi3+E+XLptKRWO+_0j#_
z;}#L(E)-rqD#Lhqy|=xQ8E)#L@JHf<tu6L42Nx{6ESw>b>8A!rYY#}G-uO7RjMYYN
zNf4b7>TEWN*~yTT!ujIg>cNh7L4s)l|ADW**E*C8`fvV%DF}lZngvJnV?MXff^om%
zQ37#6qp&5nW<k-<{f^Pi!F7>fxp;kT&9d{qLx|4~V|ge~caR)QsZldc?{2a5fd3E%
zg>5BGI4q!?X5YN;v*hU-<o>|7>?(nLHcW^{Cn24iM#OR`v3Ep}l;>T;+ipqByGFVP
z8C{67*h80f@fp<JW?bN-?G+v_ne~W~uC+x{94g^qYHr~QMwpA9+b_B}$XXq!Cyo1J
zq`lt*wFm_Z%LAKw%X)Fl5A7qmqz#*9+K%|52CDq*Kz)_I*;Jp$cyB!aKR4Nmaouro
zW&KT2F6i-*k&##hw8pHvrF?EL=`#%+4D7Exxf9Co>slDRk!N5G-&A6t(srxTZP)Ft
zo-1-T(C<773+vqb@%Z!ODvl2tF}gj}tS(WE7&04B>^1yOOit*AR9>02x_3`Yd&1}2
ztkA-^c%sfOenqChuCj~Niny}K1aWL8;V404DmlDHwK;X+8EOH<hyL1huDV0l+)bO`
zdYq1nGRoDV>j$bh+MNt_=f|@|iCr3o#e8GMOveMWr@Ec7L+-ZA7tdZ$6Py9el!8Ky
z_r!6)JNb$AcXAqzAe%P0&l7(h(~@#hhSSo};LXd^7-pFVFT}kp6mb1;KDq1(pcN-y
z_<Bbjw^*w>iqmM#35Nnl(7$`XW1uul@a5>_WHQu`gk5CA-EyFpUBYtR31h-5)L)lf
z>46+>KdDa?1EEQ2F?FM@L7Ra`fV<6#!BBm6_x*XOzZAa^2~o2M4a{lXl!Pz={-1A`
zj0Z}hto8aSZR7&7u9q6_lvDxx2v=TX@>ur7Q9<05*|ZP1ecSQhJnA<PA?;Y?af=)h
z9mm4i6v5tLVQ%H<;}`4F)k_tz4NE(8O34*--q6@kv0*!#n)C3VdC8w-Kd`5)>F3vk
zjg9@9C-J|Mad_B$H}zzDoM3R1L3ndfx(X{!z53!>m!5z2)s9brG8F+Ob|@SudR%G3
zW6z^1g-xNY`F-EHYA#_6nt@ujKDzMEbx)rqW2EX`b)1@+iO(Lyxg^G|`C7|jRaMn1
zhzo7+xG}m4eh}ZgN*L!Lw1V9_tBQlj_(Oj+)0%zlhr_0BD@x6;EyeL=^Yeh(lgo|m
z`qY2}_uSr>HtjT20?+0Rq|F#)izT(YFZVj<>Pr~EILZ^4bcIojb+vS#{AqYVB0V&q
zYn&!iT_wj!L=#o;XvPACLXm{&6TYTzz5I3B4`b72D0Nyq8BPGmw56q`R6{k512c)@
z*B>;pf?A|9|JE!RZEb+k-Nwcr4x~S=&cRAb`D1C~wRbuCzEU-|`te(O*6i<JusH`I
zOW>A|jF3;zwD!$F<*2_Xe2X<L$cVSEVqASONV+GIx!<~9EY!*fi-c<kzwRUWy%Z61
zax#(05$?_?@ONk^HkRXZTc0mqF_cKa3A34nP6>&Lg}HM<Jbo@Q>~@B2(8B)_^bt_%
zXqeX!!oT-gABcM^m|+M&x>p=UU7VfY>i#JAOEkZXhz)OJKD_=G!nkzs6acM%e^Nxm
zPx&FAGwd><ryHaYBH5Viva&ML;KO~<>6=_|4%zW8P7)bG8vL=P<P(FSyIKz4(||M+
z<d)gYOdD!a;sh^53{rJG5hK_v48A@0wWok$o!DgVe`MaqFE{gEb-XD0y65Uum!z$^
zhCoTXZo2mkM-uoviz(e&bri)dy6dNJchDnY)tK1sd4SrM%OYJHkIaIRnxC9F`mZ%I
zPT89{F0+5wzqbFiqQrPfX=m&-(5_-_BFrdcAKkIup(p6ao1iQ^+3>bcwoNau-2H9W
ztSOGgzI@+7b-($FEgM&v<FAcBE+f#9$Pazm=`LNWf7e+plOPOYL9u2#{WjCPzoHl=
zX^#in5MCdosr^^{P%;lO!n@bwN9?<wOQA^QcH_wP?UkU^^r5MLa%*dAKgzWeMs&#4
zT+>ou<^-@@z$FRP)9|tVGhJL=m8bF>R7YNeThS8qkp<EmalhGARns**>9IiB^UH2E
z9A4cuFFq`3>St4v*rE5`a$iW7ZV)^6edDmKX4=b3{82HzODIxT2r2AuhSC-vo@WK(
zV-L~r=#P7+8+w@f*oK-1uCkC_V)m|-A}840L@kJu?QmH46JgOI8ZmHnv0As)Lv;>J
zMp{SVNt&;85BA}A;?{37W?kGYH^R6Y!`04jVw_gGPJ(MpeI@?>(;;St;^ytnk-E+(
zKP(=4_s+-X=1F3jS7%s|$ijB>Bd#>pkG(r%)YzRT)x-T6+3A;xRVl~$MkNTfoW+{s
z5b`kk2@v+BX>7j3f}@?C3JSP}vajebvE!h9-gZ1J5JXZ$;7&bUGNp^tT&EMYYce)8
zbU0Qka)mJn!8l_~Qh9H>3I8KcbS=-AX83bNS<~_=zg#%waTcsT5c)e5Is-lu96EC(
zyW{CJGq6%!qOp?QayPMOCgZwnd~B#OGrBxkqmixGt;ewh>W-ar^k>fM11`k!`}~%|
zk3P9li<@ItGr*ScBL!bpR#e3E)6Zkk=P*HJ9?;wJOEyQJ9E3e%2Im@36QVB_AB9TK
z(k>W<fWqLP$RR@NZ&{u^`4GEz@q(E9+L)7~bxc<Mp-0+?o_(6B3l+XFeb%9BE}S+=
zg(tDM{i#FV(lC+P$I3TPsF^d?Bf@-HHh7Y7`CE67`fm8ifKU-<YBE?{&uE;Tg$rT!
z?=?6|D=I1?l<}I6&CxEnzx2gu=FicR`vdG2i-;?7?3Or_Cx*A)Ql|D9-vetd)E<rF
z3ICHS*^8>fS<E2viG*)Nq&vjbA6>DcB4*{a9Hf1U4I4L7y)gK%T>YQtCdsSc^5bmu
zOijHR<l=3ZvIYnQ+4-PjPbl1wn^JzLs*7mWvy+{;_?bTustnQ?Q%p*11lStp7?d9&
ziW>rdhI8v$$Dm7MOX2W3+Io?Pwb_U_7^x)g@QAyS4ktcl5VMC1QOJ1b{Cx2f8_u)%
zV(kKpPti@{@5D;ss6Fhu$I<*@B`%pkhZhoX-_Ca3naozQn@jMLYppHgvdxkZ+MCbT
zI|ePuFLW%|dgbq_VN$~e)~BkU_0iOqx6HOap>`s1;3Ki+mT+Eq+g{@2HcKS+hn$62
zaU;l!NeW^6w?W;YcOq=$r*A~y;UVu!t&_7egd^!Z7V~@$%&DokW>85F3znH}L$B(Y
zYE@&s5^;|u&$|u2zDHL6F5^!E#Y8IrSsEJBv0YeD7K=kkZ1S!;cb;PSM=m9B4(-O&
zsfb<DH*Qr{jas1|tBg`Y(fghf%9;u-v(ygimz$N=#Rq~J+j|!Dy-p7^a;;1(ImyR?
zNP;f7GIM$r6;A)imRFk1+@8|N)2q95xUnwTMElLaU{rUJC(enp=5~9hwFRMlb!qN+
z30(>4$zNG*H8$#^!>;PWcO#;2A!bH+vM95IZ~F$zwL7L9h->4#JBN8m)2*y9OQ)+H
z<yN}B8>4A)k{CwvV`j%fcQ*b~WN{2*)j<z}Eb9z#Z@0Y}Jd5XPJ)tp&(r26$_&>A@
z(9}J=&Fx}7sJX<HMUrg+68zKI#l`BwhbQLjIqB8N!1}u42%JQkfOfukW%ds?Hs9;&
z#Bmb|?hialP|kW+S6&Wp>o(B9V-PVZBW}k^mKVn74CWBlaK)i1ly*`%ih+I{m=Q?q
z$CPyKy1qJ3o}PH=5_6!NI#3fB5vaorJ8oMJAQpnWN{c-ab+DxIQepIaHpln7FNAJM
zWO2Ev+%;pxS$>@m;LqSdx3I-au@2AOJ%RD_8$<Q#s74QI7I?<7{CFJOPyh_)cw^{W
zpX4b%a$iD5d2%PI?un`f5ZsK7=k>$a#|(ZmI{(HS)};J|lJ?68Gk15WgtJpTP$DZT
zEUd3|=Z|snbE`4HyO3`5wX))Z9UGjv;8T|`>r!hF>v-bC%yKsp9NGGr&twD~i%t=-
zUGF=P&Q698blFN@NnBowf18QS5jp5UUFjMGFP%72l7~>}UQKkkYDCILQNNMU?xZ>E
zMQvK|I8u2%X|JWx*_o~GTI}qRC$LR3^ff3pw`)oME{r2ya=0S($;CUZ*z0(7p9&_r
z-d4-<Ke;;0G4FRa;8c@VTUQ6ikk5UZyURcEN^tIdB?dV)H8t*D0M16v&M!F~TkfCb
z_YfbBd!OXqML}x#7Z4MLRN`LVoIC;cRr3k`0ipBeS+10uPE1`%^w;h}XEs0{<!iBB
zK6&cw|If>uRO1~E8wy9~k@4fPuBLSb&b_8+Duw#gA^SuzR}S--FY$Wg?ou;x_RDm=
zwbHD&ljDB^Mpi!~lqSc;MJC!8zs<g#zgj(7RES)^=)X^0Y|JE_^usP;cXyY_!b@Vh
zYi%39j3pU!S)t44kDJ%6?QS}bVwFrk#5>k3DEpY*)i9Z2;N9ZqwWACBDdHaJryLMs
z<sTkdzTQ1`MFzn%#Kpzi7Hc8Ywh{xodwXat?kqdnCe9fO-bWT;Fm)cR+K%IyqWLd*
zCq9?nUbpYzY`A+o#77D}pY^Rj(N)bwe{r&JzVGOt82^fA4UhfcPVaFd4%gMc(plZd
zsG2LR0!Ndu<?nk%5N;zQ@^X%heddj`)5K8G5sMrCkGBofua|#gzLm@v0C*R$M}CR^
zQ-;4cN3#;dyk)YMhK=X8g>>D?bOw#<wY!egIu6xPHv@a@oh;1b6rh@I>9fTCFEc8B
zcPRIO&nT3>>k+BkLnPK|ADdh!9QVD>qx(0DOt)D8S^&T+;G)l_)E&`z!^+dO9^Z3*
z;dPvTPUIl{JDx2l2&h%2c?5-21OzgG18Y#WO_S8s)*fG;Y(sAS<#ROg{ClrSSJt!F
z=H})O>MFj)QzVcNytRco$fToax<Gq&H~Lm~^O+E-fvl*l{gIW$!rtSWub!hh1iluU
z2UKb+z|61C_DS-h+cY38#l<{8XB@;)vg6V!7{x!y1A;UVt^i|wV`M};zielfvIW#y
zT3TBCK39-L6z*{p|7!IJEx?n<L7IWup|74HG1T3ysJH=o9oDb;_my&ntTwnZ3$&+{
zS2Xbj1qHG^Dmh3-3HQuE91?5>$ae|Ko}M*&?zG%#m$3LM=+*|!q?=ov;Q$ab<xenw
zVoZqEz2)w{1GRhNGCb`-f~(Fc7<Kd9@ml=^=U+!dL*Y(c1rw%>XFUtHa$&SS^%hCP
zN5+NNpJhJ`coI!rZHquoJxciOdw1q1%}4j?GvQW&P6>v|<sMb@uh)-)c5V0s$SAL)
zm_6_8@5=r`WNQqFGr76=FfqeaxxyHdp4yBOv_J*?Dq+jZcASlD=n|=!k4c*2o-vTT
zKRWuKi<=wHbLJX(QU*%Q!Pc2qpR=;E4&J;<A~*_g;}a2~PGY)q4`f!wU=1kft~ZmZ
ztE+=Gb0~ey$Y2EA!s^{SMI?qfLhb763U1Q3543tZI=ZaCghEd&&V;L`#r5ZjnWPGf
zi;GQ~B`{Z*TYV;G+o>TwEA%AY+t8c0U%jPDDl4C!oStG&zSU<P*x1~}o3|3AH4UM?
zJYMP`0Q&kThx_<g;Hp2l2|nTF4H;))E;%MLj8uR6Z)LX9di)sScZ#4fbUqEe_0au%
z^^fLhZU5_Jaev<`-Hq`|pyQZ-;cW03EMPl1Ik9-3OKMBWDJ45)C5ZuRlf$i{8UxzS
z(>6^#O#Bq6sDFtr_j?~7N$#kwzt!qR+P!~YH~QX){auFQ`cwWc|MiQqb60<$G*im1
zpV$gQ_E~*42hM#O*|uX+RQ(yk*hgk!LSzC2*LUXTtUHco3^NTN1GY9dd-&+rhRCH$
zUrmT|r}-9^lpucn6Lo5_KENElysTxj_?_pS7v76w-F01?y7PSEwb;4o{r+=_U%%_?
z>NfX8#xVqsUrmKsYc?+J3Haq%-lt(XUSzk}c->$$zP0r}h9ODn5}U4z(^DkCUZuuh
z3_O-=_cqA`yLLU#66%TQPaER(Jq%ZM?_RDh)8AVOS(`uHhBzn={ZBWp5b>2Y$1pKw
zgh5}Mec9w5+K+Us4qD6bwgV?Y9{mqe-Q|{Tqd-IzEW#c!pvPA1Sw7iU-#uI6l#8cW
zt-qcbr2?C%MrO75WmP!Uiv5l-;ZH0bm@ou7yey5xZE()%14rTaM%+Zq@U?&dC`eF3
z$ndyh4SiBL)dY;P@%Fe(yI>QfrEI>rhuN&I3Uz#TbG(FvtTf0Vn0#P|2%ndOw89q$
z2q_Oop)m8v>}>j^;ffw2k*agi>z{+3q(}GV<ylU;&q4o`BB9y2L}wSAnacK=ME^^a
zF4Y)8I?|PplrbE>dj2P~Tk@!FM&#<Wl-5Z?P%wLSb#;M%r}8EUDxid9Mc&3{!;qRv
z(&{6}Ou#2JR=$j8YjAhkoL8!vfBT(H>6SprPfJr}*bkYzDLvg=ZYJycJF=Oa!_3US
z;M$Lu-Ze3N-V_Ho8?&>svL+Kakq?|ml#-7i9@t}3vyUlB7M4Yc@geyYUcr;g2#M75
z*vG`J1N(;r2@sQL;m%@SpJ%%FoR>np*PPU(!$eyPC7zX*mEDu+zCT9JZ&lk6GDhPc
z9U{6uiUJyJiE$G#4}2kd9m03>FR`hw8=0sKq?NF?^R%IsVoxuxz@hKj1uf>#ikUzb
zXxmXJgB{4BY#s0s6Brdm3^Hot%y<86PONK$+`+q3FFk@cu{Y?f?W(1}|F!9gt%-T&
znz#e9!-d>j);%x=%KuteD&rVqVu{vdCA<ckO)?@_KQh)u@5Dt;9f&Oz_4U`<WLwiy
zC2h>9$lG$kmraVu>=(lc;Mguw2X5;GU*6F*i6|lA$lKcNa-;(LjkFR{HP`C0Xy3%y
z#2Nn&!nP&ZM#rCTLYVw>rn0!QnjxD^@Cl|&_D*<^ndyR)lzdD>jlEX{=UnSr2v@rj
zE0qX}ZZEJ`MTiz^={tkpRp$+Pse=*mnIOOE^lw*7`O-O84<<tHTI~*4kBZbSgb;0A
zzUHN5-?jVp$8wGo3R6AQcjsNw{cl~9g=O$e7OrfwlZ0{9#&)+QNLNOM8JM+fZIa+}
z@&hR8aVlf)dj{@=HSnj15C`mO^ImG#{U$HA>cyRhau2Uhm?6ksHtd)(aq{_ziSG7G
zL!%3xBekzw4@D)$w^<F>Zk#zFv51=Wq~29Q5s~jeO%6G;-v6J-dpUxoL}2G@1nlZx
z+4A!T3DPmgUBG>2n^i}yWMoIP-&|Bl_7;VVcr}NbNKG2xh<yqxKdpDh$a;}asylBL
zK*~I!D|xw!A1WhuJ_tcl@fu-<tL*SdJrjC4!d%GNe|vKUA1j%)4Y96P@B6&TA+yTL
z)Nb(qW&sYBxxDc4&5sP_`{*!M@raqKvYA(QxE(2}O7!d7x)Nf~Sqll<ur!p#c^@6Q
zJg`})dz9qln}^|FV^Sh%9c5M)+?beJ=`6HQNX6HTTTPdoJ2~RYhXfu0O=9#RP)$G$
ze`~Z2CGYfyz5~7q%gRCmE;V~+C!Q8@u0QE^c)(kR&;av;nlWft4Yc~`BAsg4^shww
zstLZ+$!@`MuRO!8&%StVv!{}sS85?N>&af(_-P_ppek~uI-7*e=fNfi@|IHx>8CTi
zK4qEQJB2-9JUO1RkqZ59c)DYBxZhRoA)dXx&hVXRrYNgSv!3pfld9Ryn;|A{F_Yxn
zd;nMj$;uZPyXR+S{w6%u0TyiD;h_UCkI}7>CH7~YwpacOFt2d96m)rY1-m*-y0;gP
z$WlrXq08L28{_U%s(kYw0)ZfPMdX)u01oaj$|uu5OdF|yE32^;kSShjhG~`<<?(W4
zms{ct(?4w%`Vf*7KbuOTf<Xw>McmYtqYFqpx^Xr!8rf5kuB{xJ8N9pP?d;WiXq_;y
z4lj{jup3tU`iJ~s_XAUkkxCijm=`2PgA!`}^TtiS3~M;QRIig9JR*h$^j;#_;<Naf
zy6ITx_VDhtfcJGgeG+@vv8pN+62XG~B7)Adbb5~wM-y@F^`S{f=AqK?lTE+4V^pqH
zZ3<hDbo{YfB}3wM9r9LR>2A@jetm%LMpaWGfUN=Ri!Q?8Cy>0}kLyyq)k%b}ehk7m
zKU5=*?`NKUx>&<V73V6j%so|(7&r!Y(Pn?TxM_>`3t0H>A+Oas=gq~1$83|IF_Yi5
zL#W#&-t4eiG7<CoA@$aXw_6~bn3`96EO3@HFpw{^%?-3G<F^K={(d#XT4yx9CHFF~
zXOxGfT~mt+3_FNGsip(SI}}8LzZm?-9Md#LO5Mm2Yul8_9(F@}8NjmQcI#o*LQ~(>
zHS;~Q+XySRcU2CXW-`zULjoyhM&QAk(c{?y(J%S}dJNE6vXhf(e85G^+D`5MDmHE^
zCg)IltJWZk;G5dOW_zZI{kKugRl;bCUwsdU?sVe2mnE_H^${uQxJOoX;(h)uU&{3}
z>=-2^MI}+xok~jX-Bl`7vTEQCN9k@o)flBLW@e1``t2$OU;6QEt|MP_Q;vw7hVqFi
zdSDZsE)`dFsizIOYovxb**oT9A2}q-OgX$XFxATh!w0;L=W<PlbWJQuZ}j!GK+7L3
z(5VEbI*HV|%y@}03FNp1hiT(?!sD!m`!h98Tf4iXW#(PZCtITttG@6qC_O|BL4sRK
zK=pf#hbRoAgl5Hyn7aShCV~I?VhGcFGz_jD-#&?P_iqqWkv&i5pkN~X<Z*SBpZe)u
zoyu-s=d738k>U@<VGfk{R{)|0t6ExQ(?y*gmev4^UtQ{g&7b1zu1DSXt6AB1;H@_+
zD(#3eo4-}ks9EYH3=?hff2Fm_ucDsh!HGM0LdHS(wAd|6W3bWZ3h47E$D3LVqK<6c
z1dN|T$~Kw;GvFr^qn5J<rwkRku+GIE)?$}rC&}Y^!?}iIj#Dzp;(6PuO3)N<6`0P3
zebc#JA8h)?JBz;@X|nTU@*qdWXZxuPM+!^dg=PDtvGvT4yCT=Jv(?WoB_7KEu65V_
z@F^P4;mxiCE6zO~UmTjvU7t42`#XW|<an_YyH=HNJkJlZjOyL9m~J*+K)vyUK1$tP
zI}wqP6lhG$MLJ&mv-s2B{{tAQV5|kqU~KL0j{~yM8MM8zF?f7j5k)Ubwp5Ipsc+vI
zjM0Hd8$XIQ(i;Sc0JjlPF+BO|R9xK#-<vm)3i2gJ4VpZOBfozuCwcL5C%T@l{8T&a
zd9nAkBe`47)T>L$8=K9#&qo4xKLb{o525N@SdNwkV&Cx1)mb1vyS!1s=<zbsoruJK
zIBz5Fwz#k`C|ZZjG1B0AZ`SN?En~Q5{8nHJ%7nBB$K`C$wm*OVK;|@ENy<VV@M$on
z9KO-Tzo|I|5#xDzoG*2?180!-HE?orx;$TwmLLAE!MQP6j;QrID+V?1si-KIRTt~$
z$h;8Lj12TEwVCkWh=Bb|uY|O0dqdTj8JTZw#~sPvIY{$8TwH4tPq+EHQHqytY=)qP
z6$txXi6d{pP*!5at{G@<OieTge*i29%hsRj_rEK0CZVyh;+Q2d;fEIEO4T^n3-RPC
zK<!^-?<ia+Dg7&lGtbwwF1gq5_3}2o2G%aXig3uU#>V1sD&8N!(MI`RT4|&Jc8I<=
zL+>`eMN19;Nfnl_A<ZFl;3iu3&xBA!r0;%?AQ}JS<@m?eK?5@qk!wY3U<0uQ{86Xa
zU}$XYi|@ZUlp-=i+OIexBLh;1^8Oe0zB0x4%*{`DCUxE3kAfgZ;hxmZvDVFTp7b9G
zMrF!h?^>P3KG`YLhhX@G#Y-vTZ+R!h>o^g!#>WYtA0_<O@4?LYk0((c>InGdG--M|
za~?#mOqZ~DYi1@5{e!?IH+WxYq5*nyvrLS&f8bs0tLS}?q;6E?;Tl|cr|=FulgFSs
zhW`gJhoc!KhrxRXnf(~N)2+cQ`C6Z=UsBfxZ6@@M*Z)))#GDmzNEqWQ7{=VM&&w2E
zya4H#vprs74OklZg7@R>Wqu=>KKKbE)4HF!J-_AIrFVa_9b$fFK5WcrswA%C_Kt10
zQ|s=SuJP!M?WSS>JS{@gbC7aOH~+OjMaI6;Z{0`H!aoNy;vFt`BPdk%Z)8j*`&*9C
zbhs^q#M<3mq)mfMzSiR?S3yC+d1H_Tlx}p<2fPqX)A!6k$S*B@1((LorR(PZ+pUsM
z8}*G%Nlu$-klAB-N7OZ+M*8r9Kg6=5a%5x#RJ@TKl|+G%QcWa`)EiXj%YVJRF56>I
zA|oQuN*_ELS0Zx(k3`^x@hC`Cq}+@sn}PkfhBL8uu^FQ{_E(ivabY3mLV$&P+HZNv
z{x6EEBJUdA59I*<B5AIuuKo_jk5o^oEl+wlQ19H`b(>D~n}E2T!}=m>qcf7O)_GHN
zqUt3^7=n)xN5ePR#qPt0peKQ)E@MgiBqSu(uCAv=hB}OOwnC|5_6z>R^dh8d7vL75
z>jDt6CnT{wYMj=iq8P+;!IT`}u`ge~+yrTrlAew((io&)0~sGSw}nksY}b~Y*xs@F
z5trUNfe=w*Q$O|5zmna_K2%EhS%&BDubZ9dA_8wsU%qlMz&SWXWG3x=ZQgx0Q7=WX
zTH||D3y!qS*NCM1TQ!r<@hj39U<3kd48ntZ4aPST*rxSQ_vRW>9b}7*8Xj38=AV8^
zx;3u0XK8!0nTnW8f9fvKzTUi?RGI<a{9~QsjV|e12L;8NsOHARx->~IZJ?aQsY?wF
z4#sqLbp<Xuxa9%I{q!N=6mOLcb>WD}ZW($VODT;|K<rjNJ%x5_sE)yy11!ithLS(d
zAO(~jeU^&u26e74(TAS}jYb(Z`5IcKV0ZRNeB0>zmiV}%mYHiO=hQ~P5#M~pE&|L5
zy|=Vf08?V<;{?F$81yU4iHXgTJeRA}4?r_`Z)PS3E)3K#aG?LGt49k73qJ+uIwvQG
zd7bqBuYO8=EOb(+sHmWe95KTbmu{HL17bhsk-eUp!qT<w#{%6me70v|bdkZ}=4A#2
z1x3i-{5lPUC_W5Ta%$$Lz-Q@SrTnCH+FVGG+EpwjCMKXioAf%qbJ?xBa@SQ{OJ&@~
zmdT*3vN9);*BD5^hWzz*#B2Tip)OM1=jd@6R#sLqm#x>`#jOu>^YWhZ^WXb~9D65N
zQBy-R5m6^>D{9&U!Iq)bw{9lR)H5*n3&OpXKGrnv4kUb)?d<Txgsurc5V62yo0#KD
zC>oy_HT+myT;$`RW+qezLpOLa#y49!w2X|z3!vIOm6FPM#jShaU$KRMnVm;DP3`xW
zNkx2%=bS7k4@Z*c@-&hc=*GLy0t3OwmDsxpj_=KzH^1ABO{2gp=K26i2VhzI2I;Gv
z^5#aq_G?vo{-{b1>71SZhw2YHO_A>_e*k#&Np}xqyvxR5Rw@5i&HQ)o-z&r6@XO;7
zb(hWI9KIKoZ(g45&yV9Ud1nHYvOQL$Ps_wKo(xCG`5DXudio!zC`M6r)x`xH@H1H{
zz6aNR=jo!JVd`n!?e9)cPq!|PV&vzvfD1IdDG;|$*rD<=?fD{)`_2#ztBtZ}on>E<
zVDrL_jP3>T&V!d)>5^VTZ*6NB18^vO1}erGYrPtPVn{&6qd3)5>b0o>WY_^$)5UBQ
zxO(H;e~Fj7^S8UIr*KNe%a*5oPq5pn-jne<uX^2H*JMa|eC<u-eVo{9Wojym7y#x~
zqw8*pT80EoD>Go;>L{sE(DboCL3GF%!^6Wp!S{gf-r3nX_gg{?7orhg3BjetdJ7xd
zSPUyFdg61rnJ3$rW(cY8r#BV&1WCFAxG6UF9%#r`Ab?4jr2c~HpZm+s)m2*wxdCnm
zO|Tazah1~_Kuf+ZY7Z%?lQy0b6;uD+xRx-VlWwTjJ=510fd(5AA5{{0tXy4%;{U6O
z74tqX-J7kWS~Ih;QMI<VUOc{?n=@ip`da6<zx6j?3vH%!(x$h_w5txZ8AX(jPR-q`
ztgcQP@)_4}a*MdvsBk=a$3##6x38~HpZe(KF3yp*yj53@*}a_f2yTp(RUysO51##)
z5Y>kxXLbL$Z?hu$G1;+%pyv=*e`*JkPvZ$B9qgh3);~cyqzc<%hpVE%NFpUAWvg4m
zEtyQb>rAm$o|=`nw?uo7hRBkP6w03)HKd*)Yh!I|t0pfm4<%)k_ALacAD}O3d_Qv|
zJVZof<mOz1mnn%S=={J$`M3l{exkQuW&$n8v8K>zy&r7`k#junNsDx^A##fBWvDZT
z(DP4%0o}^V%05#nTaC1u3nxjd7yq__0aNR=t|s#L)4t#B#X2yUYTXabV#d*T>2wkS
zhCBX(>fL5ba@pY{THH(iv@gBxn>Ur{19G+N%hTQI#iXb8K+teLKUh5A$=%$vnyq$_
z2hLSAlXQRm*DTHajf<mo5a%Q-D}Bju^mzwIK7IBpRwr6qpeZ9+9m%Fevgx^I@rt5i
zKbQcV=4T!zo?eqKqbXpENxA!;g|Vs-C@xU`{BHF1W9r(|x12;6n*;KY^!#Sk>rdo+
z&ok*CzbVo_`M26j`l5}<`O8AGMQ4ZJXFf6N;>Wci<bZP8w~_jT8V;sxl{~LPy*n|E
za+xxUu=r@^Bk6>lug?z$QC??zXuZvpAQka-Q`jLtKc7p|?^c5I;BdLCvbL6X?@v~B
z1ROQ6eSL8>H8rLA;zh{i<t0$yX$6)t{12+L139r<l5y}u?`>{v4Ff4&F}O5p2gmJs
zU=V*3P(Q)_?G8jtf;=fgN9TN_513W0^*s3jg5e<thrAq^skZX;e1mjeYz^wegR}kB
zCHYyx$a-27R9MEbTWnDC)q1=*$VmnaFaY&9ZIZv8{);kux|Giwz>&p#E-Tc)OD81U
zkBo`}E#N`i%ZH7TvhmrVDq&uHsG0_fD=-LxjT$`fXRNYtXM{^=v8gBPr<sS+d^h(u
z|HQ#$u8!lemJ9|caH*!XE+809<#qld58{+E=iM0EX;x=HPW4liY4>B|ZfMQ5NWVXp
z&{9>xFNWVFxE4NIAgunPoMX?(#@u=S;>C-=uOLE|rnum>4$Hy+F6tjXd?<jIC+@cQ
z&e2iet-?Lh&Ih~Iro^57QKT}i+`w-I1GU>bljX8pJ%cc6PWCTe_V#?_ogyw<@gPy;
z9*$XKy_yOL2*{K{R@lu;J>lm5{-y|x^|IKoPZzUr6UpOgf=mr4&dpUD?)a?L#oU9L
zSi>dSm0&V^TcZWCjg|-mfb02eO2-%Lr~&|w0lf5h-1|qu^jRAJh0?A+k%e`)WDiiL
zb3h<qX>0?@-D!1u(nd(09#uof$k_ksduWY#I8HM^BhO<{!$}ly8N$maAPsg_$7P<#
z;PnGZDI9J+Ekg5@To&2RDUpI%VAj)Q$NLmyRcJj?{|SSh>;`4r<4){NKKrXUDfb!@
zv+h1a1_^%Zc@cu@Tyo&W{M{RxoeueBmk&$RXfA2IR0heP#ak`TNxBoV(zmvj3yLtA
z+SS_07hR2L<5vQK1PYW?YnXiZN?PBMP9#$FQ%iusFNn-q>7)&I$5#|mj=dRmcY#6>
z<1*6Izcw~Tffx9>EhJ^66Vv-)d=J3IjqX@hbO+s=Zz=(!_5RKCVSBbuw?k960{n_O
z%=-<>I_p0})iDN-VMqq*Svc4&7zR`u&c6gq5}%V%9hQ3kMMsn7B;{$lxnpH4#7W}^
zijMX5j*gDf)QDEQ8MkpRnP`u*ubv32W1{gHdCqQ3|M=W|bQoSfn`o6Jg7llm+J0hq
zHp_527_*M4tM1&Xg!D47dAD%kWwkF6m|dg7wfnc%^rfL<SvQ|ZAy^Gb-gYBb{`r|t
z)NS_KUd{Kc{DI`igpH8P+1{*4P}a?!<rxsy1hS52;k$f<#STK3fE$2ZHUOvy6#M`W
zWpP`Uj_p!D#BiyJ^lw7ppQG=se0_}o79Wq}P!|LecnAp-w$M2k0VbXP7%N0eB<Xvt
z8$!&W365}VY^>tv&#z$bL7?LC<Ht1qV8~wU+tA>@e?LEWd<8vRe!fAhCGd`7@W^0|
zd`$64ZeVuB(Vp$Z*ZuEik$evtwIO-gszUcYVDyj?(DMMUe0$k@?Z~)>207ksnG$n1
zzzZym>m(AvI=5+QZ+;LCP!Dk6@Ekuo0NUWEt1~ut(k(BaA?rHnPzrq;qXfyHwAos0
z3sD}~O|$?VbQ7e~VmsGca`o<$pxuOiG6u-(tScoQ+$~yiSZ8b``ss|yl`JHi1462N
zyT?v(_!w&9?_jU{JEMlv!b*Z>{Cy=Ye&wv##j`G~(hA^4fd&<XC;{GImQAIr(NcS>
z6KZpFQ&p}jD9{1p5di_g8)M_~zkgqA*hgz!7_j*JIwvsEFKFveVSgMNdIg&8(9n>E
zzv@GGz=huOTm4x(Rm%eV58w-+BwMV#&=kRs#MlJ_Jea%q6c>4&R1HWPsnB85x4*~p
z$#nRiwzpj(`80KzoZ1>DW!ofb4`pZ^*kd{T6{mOv2iB{qb}p`*V$wD9vjO`J9`RUW
z5b$keQAiXi>?(rVv_2pf0=x-&z>f(^4Yi{p4AMqTMMZ_yxWdB1NHabM{8k@avfG0$
z{>gN!U)cj>M9lrbgs-1*`gz`x0F!SNxhOTv?Jiw!xI}kc@^B*oZ+^^|6tc;WQ+3XW
zI@D!oadewbdd6K*5jT4JM&Q#EU>aaWk(iy76M*#<0e;FkOPwO^r{dzNdL|@9g9xND
zXA8(@K&64M#@OllZ>2|7v<pmUPn<XaplCu3nu4XJC5FBF%a;unFHO~sVUU*3S_C+J
zSrMSEj<vTx+X#Udf+GY5E#?`XHpCxnq`b4TQUdx5@@JO^za8M1X}@4}0WIY`g@;G$
z12BIGVWC5PN5A^XX<|XvM5&bjK;+|KJmy19fxQ|K(V%sJy84#PQt*sNq?p*`a8}PY
z6YVyIiYbbwI%{N0M=G%Ly&kjl_BKGnGj!)XT<Tz!k06Kv7@e>)F3@L0aP<1x8=|)M
zc@Q7FqlKKLMP(k%>0vm;Z&z50_1`w~9~U1D6KBQ#EKDpV4Q?g5b@8$60|s(H&mMX%
z7q3(!FvOm@@NGldk0Z%=c1y-adFUBYN4%@6O=e8HfG)Sp6zuf6-v^KmIkG@nIq9+~
z6~UQSbSVOZ`7M<SVfPe2mArnmd49N@+|X;40(#+6ktRNM8luE>Kmztlb|KA+*}M34
z(b%ccOtjRcBG{Rzw}s@ac=;<AX^(&w2}mP+9Nqid1s!%b3Gm1$ATj_Hy)iT#1-w<@
zP(jMgEqouAU0qm@2wE55upzX!IEg3;M6&%E7{Fl-{n%<GnUUPPias5ac5fyTf%X#z
z?h&FNA><O-3&Fu&_C7%Bo0v!furAGX3BtQ7;ZNj}RyYsHFM7=)C<qHwW}w|K+ze7>
z+;4LILW(Z_U%q_#J3UP&1=H}^f$(m3iOkNvh}Pp+NfC7_flk^F!vg32XFLqgo@)3B
zr`#^Tay*^=aT05bc~$#ftN7E9FwgpqS!2vESf_Et!0<XbKQvdGjC)(Z(@w){ekek_
ze6~Z!P6O6n0A&FSRdI1~TYG!Hy)GDiW?z<i@R`C9{*p9Uum=$!nbFycE;PrRBeKjn
z=@WS+A$3<fDITXg$pG?$Z8H}Ds`@5_7_}6_d?YylzH$HPVU|9-+Byh5;EbmGPp47v
z<;lUKJU}a!6I?QTd)GiA2iikYn8#Jvv;I$^cds^^)A*!o40!4VTFF^$-D^tdtZvsm
z0B`bBX#|;r9<|7wzE5OsZXUqdoR;|IOKZcA(I0y=HQzwEhj<-&PU%<KZUV@eyu){d
zJ?JUW(?)u9bR?@FsrQNapSQH#@alcZLqx4w9goLE*Gc}YEs%sbBiwk=!!2BOqcIbh
zam&TO0>8!22CuSH;8G}u&{58}8M2dbnm_mPE3Rmu=JW&|0n~ihd98}PB?I5-ry-!R
z!I<ILIPVNPJR6F^IK&p9=8OKi6Q9pUs#q|5A;0eS`PC86N}~Y6L?0n&NnlehB_(Az
zUEHnpoQA0{)WupT6)hJ5iG(sf<IGZ=(_A7n8O&Y;?VRoG6ao){YAp~EtMoh;&$N1b
z&4HU@WZBjKQ3(%FxEw?*Z9y{w<k<6g<JB796qrgR)82D6f98Fgo0}_O&VG12s$(29
z!QZ^SuvfN3O)crsnj|T;vGABH2A7&)8|(~%!btIm4{`|5(CVrMkhg<G=@}Wd^Is1G
z<qb^QrVK@j@Gcxl2?^y+PEPU-^czVY2kx|{8B*$5?=3Dn$p?hLOiuTDWw36V+JK(p
zxcmw)V-pK)>WEv7Th_*kS@4q?a+7j5Xq%+qS5!(C?p!2O#8<vT_i3+FZf>zj2ZKHk
zVPN+lh(BwE_!cN-=n)jq;@G(;0++uM6A|U-<-Mq~n;*NruIDkT{|>MvW^hl^?1`BK
z?t|6_eaPG_l~o<_$Ur5gAlP667yUzUmaA2v)NZym2XuK$vsScncz%AaZyizDg=H;B
zi(V1{JuH{kw~$^L=Jm~QFINel<rB3GzYwR96bg+E*gkL=CJDPW88>><7(6dQBv1(K
zP(PY^mH0v_IqAEOf((T3O+=e4YZOD**-3RiovBrq{vu6pZ?CDZyDtC<zkh!QVJl#y
zjzHuE2L}sEN)~}F8yt;1<G)^>|M*{Oq2*?fe#w!9rE1c^hC((ANKVprv6;8E&%_<e
z7th8mTm=ROdR{t@JJq@YsP2E<!-Bg<_)#hDrz%P5VaFLEN?%aj(!K1_;e#q=ZUIOw
z-3b#lwT7{=vBsQkvm~~nG?=?<3B(DsZy~E0!y=?Kc$DD0HzSG`BzcYML0!^)m_Il@
zotoU`*}!O)#d!CxkJ(S&fkp=~9<&PtQ~(^X5GIrhMyUWJ;YysA$~#2cTJ@k}1Qaa>
zGe~;A9`3hVFHDs71Je?mbj%Bt;Zi8bV<k?C_JD7r?}IMYr09U|fp(9Rgl@c#H)5xD
zci&?Z;$+?dH4BIw>&fK7x_$zS?Ul=eVIfj6FH%C-A;e61bP|bAapJ<l@Fd+2n7a*t
z@ZZ|{aczBl^qrvd&O}+CgIRfbdGhd25c6u=i|H$*(3QxMUC`>E3JOxpakPn@Z^J>2
z_>!2|w%E@Ta`?vMI^h<2didt+HwVQI^GJUVM!KVon+br9C_clXd+#}v$cW(<KbICB
ze-3;e_cO%Rvj-O8zZSkaHUm$~NNO=W`xGk8ifvi-dG8Ta87VL}dd6Bcy=#9U3ralj
zK7cL=HpvKPYh4~ktb<cuoDhr_7abIk(s_5em4$mN?eQ-gs1u1wjsW>=v<4A~x$UJ%
zoiD(Em!tqxr|}>HDgmB5P*hwNngf{6jTF@gqgPZx)d7>7YuvJz?PoPO*87JABE-Ga
zNlv$}o5Sp2s;rqbUs;B7!cZ&!b?YLzfjR@1lhe)wKZCfdDnPjd?PJ;L=ap~~U4iL)
z;BeY>(^i!xg}p9+{TICsaKnF(?t|ALtOk36n?M`@%n|R*PAFB@$n>_G9vdtzgE4iO
z3Q!h*ul@w`9W166@q&x&a2m~=mdHw=o?F;~zHr-!SaJ&I*lh;ZC2x!YcEPobg{n$W
z3u<#*g4*7+NRd8zI8i>rW8#fM(kdmVQ>yZP#Efy*9!@6W{+5@=@eNag#57A#af<6h
z^KT{b{o|yJjDB3jl-w-bw6d~6uw<B4?09*+SxOKh=OO)cwhaqBbTphN;!8z^^KGY|
z?4wU3CrqK89LpDZ1yUE?JGYok``g(~$-#y;P!2(NBgX$NA~D3y1zP}cpwXKe)BK<H
z?4M^=OYfOR?*G48fOJip5!ToKnnI~B(Yh(<Naa{$G@wJmU@)`@3o^LvDCvcPP6l3_
zTeBA^!R`+i){lAz1bO8U83m(1`ZFX8z!hu(<p3@B0T2PE9g?8LYw2Y*I>A(A<IA<L
zmUh#C%_K_<2K+vz)vk9=x&P;15*@--FBrX@R>-W+D3>Xk8eyI`li0XdiC$n@*dXLk
zLTV&-eAN0*NJ@b}z3=VN8$`FEeXsNx^xgvm8RUDsq%EH8J3(52T!<{5Km@^}o1TBd
z6TkQrMf@wL{rM&)1J%x@d6zR&XJ<kGi;WFHs=wII)#n3H9<+aCIRCz*n6PsZBWQb-
zhZO}Tv$sKjz6Y%dsP}-D3q+9FGnZe2X0nN9ksklvR)>sJ|KoN`kc?kAe0WEdc*4wD
z`pV71u%(Cc2<-qo6%(WF=UI6fQQ8h<((N`9X~Lf|04RabUeH`S@?el~*C1h(7z7h+
z7|isNo`P}72<`})W<Z*OTpSOyraU@fTTIaqJulo4{sZB?7irGGp8y5ZR0qFP82kO6
zdQ*(vmMpu)SZicvT?w#v{_aj!$wwe{XvvLtcXs{)sTHkLRn>n<!fbbtALmyJpi%kl
z?&Zt7@_@Ldz3di@c}ih#QnvcJ>JV69E^t;nzn4r(L7g~Z{WzWLvRp!u1m>*a`p^SZ
z%s+j7pUR0nzEs5JAGza~HG!%4_ars*&CULRW}}HskUTa$ci4Xzm`3`#pvZ!E*&GFt
zprVu+&`T)at{yO07Y_xit1gSJSVhIfXq+I6hXNEo!|QZsvRMYtT%<S7RgL5Un{0QB
z^%#9Oe?%i3X5nVi)m@{qj=v{fg6?Vnj6r7JX4Rk`=V#js^n*Bi%}H?#v@q3&i3cB1
zF)NVWV!_~C4NgqfkKFFTX)7!;a*~Df)?jb7p`ih7Wum<cBX{c7<;{$W`0Sh<_6E<B
zM}1VLB+*1)>8j^JRYyOIm{H<;W202Mh{J;!)UEsjSCz~b-(?Nj$V!*C1u(IoOZYLQ
zFeVG^2m^$MP#mv9j~UJixF0;V6!*va(szH_&<sKA|E(f!bJcf-?$JQ&0ciAc;gE{y
z4*)kK%)<8G$0@O6uXd?J+kthswY@z8I5n`H?@=_oDGG+cOj^MOLubIgFG#2>ZB74m
zxOBgqB>6Ut+gd8W=exF`8t`8$0RM|G7%iTl4V3}lRkS3m-|nfg4Eoa*q8F@xjvtfA
zuO{0>e0rQ-=Y|wum!dD~Y@qqJKXkI|(->8353Chn8;k*`>S1T-R{@3k`-HNf`=6h?
z^9=&7KrcYZxAb|$+q?0hl=fxPfF0_Y-__}^hHP`187^2v0*ZJ(z@0(o)&-S@G_6^!
z#M#9m)Ff|nCTD#8{wyt-+0Q~PJ2wQBQHC!VDw1HG%bgV8TFxbvg8MsF1zju6A-CwA
zEjFUX?5r8wnkISbZCXU_j0-Dvm%~Bb{Zvg9B~HXW!6jP8!QA%Xvx8x`TQ{LKS#Qip
zZSbEtjEuzPM}PZ;d?RMXVY_*#iSyzT&!5%cHi`k~5tSe${f*WH$<fOA*K92<twp%-
z!~Rf-p_T!jZfc^nzzbLPfa($im*G8KdBXc;Qi<Qr2Qk)~$;BJj$IbtC%ctZn`b^k*
zSCiZvV3PeGn!W@Y>ivJ8WZ#!E!k~mAdm%Fj#fYec${Iq-8b)JZvlj~4Z<e%JqU`%J
zLWqRy#u^!0#_s=`@9+H2x%ZxPD;b~9`~6y;*YiBD&+4Q7cjKsps-qWDs|Qw2C~mtK
zk&g#V#o#(#cAeIIXFACx7BCOx?GzKm5${IRGo=KtouC&p8{R;EI{rFxzv9DkiKbn^
zXgIE_Fw5NuyRT9hhiWE#%xKIzVyS=*`$+)MnF4bzVlH65%j>D?#M@-%9Np=2eNw;s
zuPE1{D(!LHp@8--8p9iZcmUkGA%eTGZE3FwT7o~SGWDu@BPuwmmyPwmut@3+#hZ^)
z42+BsYCPf$!Rr+~#|SI(6oeH%2}Vo8zc1Z~ee&PajuV)aJ&hNvyJph7tJl4wS+oD+
zqM5Q~Hl@qY&-Xao--&j^%BVUnS}{u~%b?@VO{rflwOyfPSFOYkhRNKP)u`#^bm;3o
zr|OgR&$aoF@0KR}QXF@q>CUO`OPT?14ol{`yDC{XW<SrX=7Y<LF(2W2yr$@lwDvTk
ztHU?=2Cc<~KTl#lRj@+t#Iahh`ZI+&l|A^y-C4)0oMb!bNYSss?$@K(*iu^)O&s}c
zVH7{-)h}P0n763q>~}O0e|Ysxp6J<wD24j#ef8%SXHxxD6E;~%yA&av!9^8)l7;~x
z_><OA0<=%f?(#QG4@Ff^>=Qq-26|X|Os(34B=J|?Z!%dBZI~SCb&1+B{p~7HJRbeS
z+Cq{~On5m#W%sCyN^Q}O(@oVR86+n2o2{IZ(&vXh$(>Sg>aI9QzAPZ%J}*e5VdfT)
zZhdtP881vv%a1!2wPTfzxh3<Em@g_*Gtrex${%=S8|xbES~*3+o_BrF_E+L0VKQ{(
z)>oglz2?h`%62RZ`9;-o4qjg<|1&9kfnnls*V#Di)oMsi%2NthIjQ1)W?@6XIuyU>
z%`ynFcHN>X>VP)hZ~lAFQEASRai9O;9PvnO3^`mBFTAK9ZzHVNAL?Ai!#O!@&Dy>@
zX;y&JUdeQqIvP{;^2FQhX*U01WAd8y5xOgA&h^u&JeZLsz>;$8ebqnDL7jM~iMn*9
znZ!XFgROpWK7si$Q?vV5ir_qd3Te^?Ldw@wAxg>NQtH%9)~y1rkXX)AUvt1gKHV=@
zZ)$th%*;lrcbkob9wm7=DEvRynCdbXx{~Z>x=DvDMbp7YPq=TpBW<&bYc-z#O@-Me
zLl0!T-$O$NUS3`X_gI7)?`$P2bltrr@@oWKG2jCLHn-bx=-d{F>#3<&E>ZbdDW~PH
zvJra4a_^yJ(rssF=VV@5ua(A8Wy@BYL_32U;Xaj(>&prUn~@BK9%frEj6LoyaWMrw
z$^d2%-P#UFzYk?8WM6IiLp`ChOYA3G;ZV^$>k;P1Zfpo);{I0akMIBGrR+0fEkD(Q
z&nfrf#!i}d)J+G}%a!<WUsbCZ1iqN_L><LbNlm2S!|Vv6;BgoMPNhpoAUr(Na@(it
zv1_N(Ub_+6n8_)=a&$|Grlvp|!$D@UdzCRUF=_wybPWp+Z)UgMl=55}bCE8h(_P+L
zS7M^oxLq_zE}_Z1`bUq>KhCW2KdJ-tM_yJop3<0Be)O>d4bt5@CfOO0PL(Bq60}Y6
zS<b*;^pNIL3+#bPAGp;`err9dN2pCln8(5gyJuoT_mP!)KIH6#V1O$WgLnM>{Vksf
z_-$Lw*%X*M?J1(8yP#N?lPnwjt3+uTJIXvhI%=4E>l;OBrc;Xpdf5jw&K42UC=}ng
z=FsqP0Svv!BnB82Jv-R-$P4>wUbF=0amp&?Mq1DOMLD?y?-Xvwv$kI4N(RrYAAfW=
zy)*4TS+$_sKik4nOoapABmBzW+yh&RnR=j6_X}gTuJ6u+ThkhAzr5Z=z<kF2cwyE4
zuZ$Tf!cz7gucueeJ7M69=IgQ%$RQX-a24=wRPZ=OGv=3?@Y?0oIl&|e689AzXDH`j
z;2Tn#N9OLR9Bvb&UqgZpmrVdz7ntV*OaG#Ka~$4lgtLS$d#Kd=BBvraMVqh@AD%IF
zQ~j=Z@7F-<PUSAR%FoFCcpJbJz$&IRH^_Ld>3%D>#RB}8Ri1zEwdF!#_y$V!mId%!
zfl(CyZlD2oRBQ3xF*Tl3r}`!*Ew7zK)iimfA5?uH()6O)9h7Ojw^yym&^i!cFZ=Cd
zUnBtI5SkBi6AZ5jrfdu&3im(#{YZq;lw5Gi1_{j^^!=YG?pyIkGAjLV?Z3wpWs_mu
z8p4p*dxp}4%kEtuv&jIiXNYzJ9~YpVPtnOI+d~eYnXyp}n1PJML$rtSuyi++Y+qmR
zo>GNwGWrupmICTZy_3Xe5|4c6lW`Z3*CNeuBsYY@`cyWh-OGt3?-velRD3s%9)O*e
zVIMv|gK$|CBYWCm?fb0$S^knD0l$s%?F`kVO?M1{kp<Ni$nXTuke?SK72*Ob?yNC=
zrutOxvGRQ*n2kd`{&S$+N$M%@^_hqpX^QMCol{@wya6I8xSbD@xHEu@o!9Gsp_i7L
z+P}SB5pPpnHV^Ozd8|&Z`oKiU+$vyc##O(e)r^I!tu$`s^p21BFCLnp)El41S65xw
zs&sPygKF@RpPx$00AQMM51!6-rAR9P$&SJK(M<YO2+Qbzhsb1#VdjXUnEv$llma>3
zfb`?sYD}pcFWUb6c`v*O-||s;`9yUo5X+$7hXFkRG`dipJ=<EoYx$&atFt8ojXRk!
zyY>qZn2^l$E4%DYuJ!?+d2Kkpo)Xoo`PhyY+Ev>QNEm|v@d>7?(pp+^$N^|*FJ8Qu
z3)MWCBLQ|YWbzN-UqLU|jz1?Bze!629S5suSVBtz7sFp6@o~{OfBe6$8*1B1o+V~R
zDRM=*uRj0l++_3?dR>>a<wyzsM`rPE4sQx;cXlk2HZ3r_oS37GoTJ_EM@RV7qWy>*
zC?95kHXfEUIC0WHaeVw;&ja@QRYg|L#VbVM?{*7Ck74RXqqO*iqRWOyz0aghBw+HK
zixXmEm@&!$Opb}3vLDmBn%Q@N!~w=Y^v>d%CdG%Sw?#$znZ8?hrP;b&fD)fc&n>Ou
zg)@y9y>+I{k6Co-#gP4%dqOj6+LYXJd&H4)eR%wRz!(A+CCqd=SdP*&Gkc=rJmSAv
z+@Mi&+rIk7$!^hgHeIjv*vdzRZeclrj3;N^$DEv#M9gZRU9{GP(~`Zjvm<CtCc@0j
z<d7RSQ>XbMS=H@V69+D|0bhq7KY~cve*9t(vXn*>H8tDD>q%Vs#H$*YlJD>y?iiUD
zf>q$N2t#p2UDeao4OwjhWD3UF>h^(}wgB^Wb?GiJ_DtG3LR(BJ(C!0w2Tnq;WXZT>
zcmRLN-G)Bv{1y7`&KW+j%3pVLf`zk_6jQ1Ox0Ynvsp1b8QzOK+cY&>cXn~3M5n0Rx
z*@(hfbmHdVxam>;Z^^aF{J*l9N0w5ouHvVkh|A4aS$U&qgw7=tfK^9erS@<|=)}a`
za}?LC*By|Yft8h6&&~xVNmC?ZKAZJ5QRWrPdCIPuCjl?XK(lKD1}w+KvshsWb!E-+
zm6HU5d*cM{JUTE~woT?9MT1-iUr0mR?i-i=!?#*-V1?GS^nm9GdM!Db4!9m+>6Zg2
zi;T77h2`+59y5KL7OFMVxChU+2AgFvwt+n2vJft6G@2g##ck*~{v&($TJ$!sbT?;{
zEZ&2>bzA1z$XJ6(NyL+NWyeM2iO&o8!yguHQctKJgWX1-JhRm?JF}WAYrnrZlA!`9
z3XUnQMk^nHVHz~dIWXpd>yrmFGMMAo+*n!TP>S=w%h=Cc{G~O+LbKwy+$H1rFa4!5
z<HK$r<=OHt7L0=Z^63drU6bmiFuTTHJGk?U&c1{56b}n^)8dzn+N`hp)druPAj#6M
zu=nGP)rYe0dw%?nh<L%}+3b>mjGTKMvwN{2nYcwFIa*t5kg-JI@3ylL4sU<%1A`0m
zS&Zs}*K(M)a(<s6!d!lZ0l$yUMo*BEEkArY*g}21<HoAv0FYHaz#im5I%F9L)O{eo
zYr5X+G+M#_YtZc3?u)#U5f<+sukAvtZC9(V1#q=mMSp*|@rjg(t?A&TCxf~0^%rGj
zIn-LLt5oVMT0$O-4rKtD0VEPa$R!~vwU+Av(*IF><h`d*9E~TuT-kL)KTF6fAai-*
zuYG?g*CKM)f@L1T&p!YiH<_mf@d@~ZwLoM9U?y7?abm@+z_1Mb0x(V3;Zp}|Uvi5+
zCh&UcH?8V!w%1~chOfq<vzIPU^=$rk;Ot1;w>nW`iDCFTIvjUD1ffXyC=oSJTD2=F
zu28LDmSztbOgTgPKELrMIk~r~NnND>x%8QRGFZ{uYd{eWP^pQa?U}w2soL;w7f&%m
zuwA7=8~0WSNg$6JWvL{P%XUsE8YnJ-?3P_wSt(rz#Bc!DFGxy;*^$Qpla0Y-O5B`x
zS7r6(3`HJb905iGBf}4X*a4@~$cs46tQD#B_JwC+v-R^r!|vZZ%F5<x5ztSK)#8kS
zX~%M~NF=h}7fl%fYoY~sP&^8mN*<xJ8D!R+^8}0Np<wuObC~z%Cmh~xM4GVfs%N94
zV>))%y9Gr>M4F~}AP$q~p4y&&C+aQYpF$}E*?6qlOCKidw~dXHpqi4~VAz(FlzjNd
zEh#AphHe(*$vJrd43=0BHE8&Fy!h(Ya&-8PMf7;LuAZi{q#&)k5?>d&2)^P{w}g@Y
zzL5s4ikCjR;RaKzV{%r}_4lh@areP)Pl$2;z#gg#ijc8C(v|r3_UABbZ~7*^EI!w3
z60f-V*62fMUnXR=l$4a!H<ag=&V!TFAE?u=@PJ-KTbjaih2^rlTi9P2sl@&SekXgk
z%D)MB$upD=KN?zExyC)fHquevvQ&|Wzs+`m7MaBU8>UdZ{~XgVd*~=QjrKw~MWUbO
zeEWt0iOyK1yXa5j%f;-3pue)*E#d@^H7`fpXc3U9;n@Ou_zPIX!LsY!DKM%)`)o^7
zu*+OfDCQegb1fsiN;ij_L8&IK33vhsZMzY^`mx@hDKB+;1PE`mPwW^=oNUJE*9oM0
zPk(8Hl7`H8?(4gy%|=FFiZ!B1*&5+Z2v#`PP^B@musi}r3|6ite@y)Psu9XCaNJU!
ze`fo>>DSo{G*|qt&84HOL}$i<uneTB@if0?7R8Q<PkLSdJ$S}yOH$+*Y-D<ee)<Sm
zJ-<wORmq`P+?1!>1WF?SB*F0zhJKKh?*L{E<6FhUKQI8-i6?*Zc(NLhO&<7fOpzTG
ziZ{cS(!G~YzV%3oJ<R$^5m&7Wx<0tEZ&Fhe%6xs$S!>I~axlCM;b0h2uoJBkb(w6S
zBug3kxw%Mvxz{O$zg5B1(-e4k@cg=M1TjwDVj+jvuE-u~u6S7-slYZZE$t_idw?rz
zoSqvhv4&^&3xW%PT2G+?w=_Vd-ta^Oy8;yuJbGMi?E#ufmUXQ9aEOevPW;o{FjL+~
zkFPK0${`qS>gww);Jwn>qPp-v4kSkj5b}0CjP%6qoi)q7g&_+Dm-PFncdoTz?RMd)
zgCiyxAk;m2nO^a%DMs<~%-tt5cr>rVf=@S=?#}W__nb=Fiq!lsWHYe%Kx`UiIsKAa
zfhM1LPzUj3e#0$gv5LLp*3&zO@z1zg7+<2VVJ<xI@+ub&-9@XeU~UbG7>PX=u(*Uq
z^MC%ee?jjy`R)P)0PS8m@JB#a5X{@`!^F&txpj@gz`Q}mQ`r}n9Bi33xteSOJaI_F
zavq65W2pGK;y34}1UFTcLh$0p8fU)dEtu%W#KqCRb%WyUVyYhbtlwT~Qk+hXFTI+f
zmr2QS9#Qf7qUH|Z_GEIp@UZtMhyT;gG+)*kDY-4xa?L1NA+ys|GJmApnRc#ycV=V6
z*$V|R;f1(LGuxTlw>(L-H0kpcu^BejF%i6(D!!LPT&2Yza0o@W_^`0Bz@R+T_$+jE
z3qTY4v%JjI<^@!@|8+&WA5105hH{{R&2B(t_JsFI7E!o1cOqV#JzUYMYORVg{TwSz
z7>jpB-utiKwAtr>hHHJ+5x_tZFg+!b6rojoAfYJ&kc{$a`|g~abjw1#pp_*>I_(hU
z4FsicoVCxbOCk$FE<p3?S~CckglV3Ok_&D@OR~K7OmpKKB{yl!6yL7sX1OxS1{H2P
zB;_ttejy1P=h5RcD@f%G80E@kb@kXZp5wW~gHTnQoQNWP>Jqa4w|#EqB12d5IesPA
z>6ZJVMN|Q%gtJn}+XO*&DWL~?3BvTR*!gHm>gqD+6=vKjU$9xl<fg~V44tM}mK2pZ
zuX5`<WnyLX01^BYmO4oY@ulAr69PBpTz|FsG>g!S7Dwfq7Xc(I%}IMz1WWVrl+yH*
z_>AQosg-zKS6WZ6@7zcXt!&E|r}p0Y+A;uEbJHcbbdXee+|fG97(TA%D^yc9jRzDa
zhp{KDId%kWH{N^#!N%$}O>RqYPV+<!-AJDTvAt!&+R}Y9F{nL_jEu-0fA>C}EF}F_
zjH|2)z^9<wDB5*~5k-vW5A4iQR9!V}TGPsYQl?gMKXe!lNEdVJEB^TO5L^QsOv75d
zQj{qMV_?ShR9a-&wh#<pqqSGAC~-!1-c8G(wDp}<aIE@Z0YZ<i5nGNyvgOF~US4M1
z7tZEMnm{EAO-kCiR6iL{T7isUEjGMlit-92y<i$)NwX$?C)&$m=h+@(Rn305@89lS
zRQ95+qxqY*(PtN*i$1n@azQ)o7`L^)9#`QHMT905VqM0)epUQ?^;yc)E2>%jS)ZM|
zyPS!RotF8~y$=J11%`&7jHqB#Z>{jFG<fqn)k6F+%QM*SahD^P7*jP2at-A$U(6_x
zA19EzWg@&WtsZL7M&#XcRaDd-rh3k<&23<p@=+=>x_3`Gl&_uNa8Wff^nT2bNoBB6
z-A$j^Nw++HSHF!V;dI`-)u}^D>5!+-(Z(MJ1l>mkM7np^n7BAS4bQ)7+Ivr_XBV&H
z)!ks4F=|GwOA)tyw><|>gSMI%l0Y%cwKG?Aa??<25UUWuC#+pQ6QcVm$62>0ok!#B
zP76d+?!?SaWD?xgp8TbyeBtUqcWe@q2$h2ib5zeii7@f<qn<C-wJxIsUL+T78XI2-
zP1x1Pp`rh>N;6p(0*K(-dgnKN*WNdov}ZA`T5F4MlhUzY4La&DGKK6Ot8>*h-w215
zO?a6zc4spL3D1T;igPsaa#IlcZrgpcL*m@8Hm6VfNsoX<+_`IBgkeCUfRLaYlO4y*
zCIRkUG|vkjlu;<LHp-{MN171wlsxBtOl6)@5v#=LR_Pgrw+kpOD=qD67EjZjX$X{Y
zNg5iY_iko;CQGlRC19eQsoF#la(aJX&n6DHihMVu;TjwxSt;cH>y(_6GtzD3I&OYR
zFn3v5lr6HmySt0=HzkkAOB1GCeu?L$h(o)~Lb+N96E`~7h~AdNX3Q($z17ErbXiPF
zKKu>zJ)~5p^Di;+SsLr^?j8M36H<G$(w41B@v*GgaFJIh(sA~2|6at))kADW6Xg&y
z%LcthA{STt*ki6iPNDAYoq3Y2O1i7!_0vXOL++KP@02!A8O+&$p+U2bPuD*4$v@Gf
z^UwFzQMBF))1#R?uCKeiMma5Ox-ErRTj;Hm5*^W=f9ZMWXh_MKi^^_MRuA{rdNygt
z1oZ<N*EPA36g}?g%^%k5*Z8$#JHN_Vwaw5R6mh9t#%5QDWy@XUVpgLx1TeMw4uz0~
zi=AJgni$teG3wg%yhNWCs|0Sj|I_)w25Cf8T=>;oeWyxFC6i!Lw$TgGW1`&8yr?Ky
z97Ic+&s|$76bvy}jx6Rp1IqHHF@UWAMWE<uW<sU*7&v+DVQF0D+M2zTVrYt81vmY+
zy<Z?FlRO_}Axk%hN`pn#o9#AvtqC+^Xb#g2c+}<@Xa80NBHvr0_thTEp=f-6ad=P5
ztwc~>F{-I7u^nFw4>v=k;?OepLu@baJv}4zH$=<i3?}8z3;BbiN?iF+$JdMXOR7n=
zK6mS0N!|I{>6K(wxHN9(UPZ6ZcJA0^_W!P^zuTbrepR0R8<oEBg%?3VL8sW%8Q5b^
z6`Z)I&flb=u6q0m<2eQ?=FJsf&w)a>$_|p_j^BOrt<On=Ggbr6ACLHE-M@Ne8om2-
z?(WyYH|||Bb!%ov6AEqG!TBpUn8y8>IK8KTz7@(3<mD}LtNi}&Zpe_6E!vitaguX(
zJF$9h-=8yMszJGDa;5(@<BM=#v5R&T>M4)Dy#4}%C98EyPko6-?dXd%rM1LD!<uo9
zSzdy?N`_*WM~&h8@G+J?!rfxe!RAwK#K@6PN<p*qlx+)ap|~*fuR*d|k2@T$^2W<<
zt{&oi#W$8CpI1Jea#wM=;-=!}{L}kq_HpCo3=@~&V`FTO!hI2Tj}2F2P1!ALd}}Q8
z5SHP$_xTGAr|AEa3Aai!=4*4=?_o~iJ)a+w({&I%Z2U<kY^eIsp>4Fpq{hIwyW5y$
zX>@LP2CZE_Bb{lj@3iK7Z0uBZkV_~H<+sGDV^wu^ES;k{%P(1FRHm(B27NI9W}n2%
zA{j{pjB91+#=oiz#kJ#i_?+IlZ2uKX7C*5_5T+$eBa4Qg=H=-l!^XM(Yi+0}gv198
zE=5MTs{4mCI<0GuM!B($ahf9dj5$!m?8@+0y+0kAS&3`yB1XStR2aJvcF)a{Z<^fD
zVtT=Qt>{Xq1<KSm*1q{woQB}&ybE2=nrhqk*&kAAC%4pR4-nY=>s|h3-2KSiAFoyC
z+rAGzRQ^fGs0$xbIJZQsR`I4jw|IuJvrrg$Yl?b1=lZ`dJ${p~4DsXJ<7=7)m)O_d
z=q1zqSB(!frXZZQ6r&-zp>*BzghL*DCfHC#<DHvSepA>zy{cI&Ms#5v=oESI(9=1h
z_?r-AezwWIn(F`01qdHS%}26axpp^*QO<9d&Du!e>?c$D3O9GDm)>qysib<Z@x}Dg
zu*HOT)I?nzD8Wy^46)1f&No!utG@T?73yVl_LGIhY{e(oMea?j=-yaQEDQ0ct&qyu
z9}bJ-AH-SlxwZRYb!#7ug;y)>8UT#dZ{yo2<~=7l*NwKzXEh$Z)Z=};xn_KBSAr`i
z@!uq$R!E93Pl~nZXpc>$LdQ&^PHvoRsfm0~qMWPr2fGZjbK^P3`Mb8htr7WTCw;ME
zuXJYLc9lB6nYw$=?y}U5Yf)fN4O{Bko(cc_tbaja4WblRhpba~Ch3^E_ztEkGN`%s
zOGWb#pHgP?_XIp!@2owK;qh~&YI4p`#l;p@j9O2-MgK(e6l7Fru<!Zw<@FC}ctoFw
zj9FWI{IS@ra<_?&WYB13UHKqQg(WWbLF-~<NBRk8TG3;ze+;5TcOO(Ku{K2cgefeW
z3w^tYR~=hubrZm~sxIw5<`UMOG*1iv8X_a6k;q*?=KeHv%0wVqd_bK5GZ1OP<bDks
z)V1GL=j6D>+3ke6GDUBv-5qO&DFK5CA16^K)-Ij0GAj=ES+b?BZkYOe3=<-Duvdm=
z?pP7s**SCiq<-pM;01*&*M!$QNoV<|nY=f0R4gLBs>3C>)fTEfSZL)`(Bz1<R+4&l
zWa?eTzPXlFToclsnd{%%Iv=34rOiOa@@euc%DQdQfU=X>jAoU|ySvs9!_z!Fp)4_N
ze!9$^XHrQ4l<d~YP4i*%JZDIiNpWW=zo!i%{G-hd3)Te9veGK8KZbXSN8doW_g~{e
z{+DGrcFXAeE8VH`aQ{Lzx=>-;8YPhz5&1<nU(?xZ1a2Mp#AAJ;J2`oymD>yt(hGIR
z9?4x|6|#2bCXP4|5#pPXOTTJH&$r=f{1J@mx7${9|LSzlcku9EU5fnfFpo$dWWKCt
zpo@6Giu3KZ@0sFr?$I#%opT}n`P%SL$)B1RTrcDqv-ZAjW<orkr<`ZB=hI+tTD7j6
zROpnC&_*riIhRq#O(sr$bXha^uh;Urflof{c0W|D;nvdb4eGSr(2og9R2jEIe7wLy
z&#p-lq?1Ec3J%Em-|`&@H#Baii^Aa~*<7`k#&h)xC-m(Xv6<Fb<h9FTvjHgUfC6pd
zE1a0Xb3FCp)ROpXXg;m&N<7c+nBef6g*H5nX_KyVAtELC<<;YaJc~pw?|1XPVd1Y<
ztSVgJ5>@O9_r+i7XkWrE#?-Th&ZyI8su6~+ugy?x5OhLap(5+5vfpmBV%Q~`e<3uf
z;KymgUXo;uGBuLUp^UIy-eV`e;2MyJF!(+Ap`U+_l(o#)ku_0Bw-}tnNm;aAQ~7q1
z@HBVFcTluQ(29X{9vmVWnVyG+X#;0UxA6X<qVDqx*Wdi=5Fb)gdoQk?k?WhekSdXG
zShPYarg5DAtC;tvs|A-%dGinUd*W%#myI}fo8+HSy^9Iq7_mo)QcymB#Ja@A;n+`~
zyHC||R5I%|+dkSFP+h(#+T;C@Tc7S!h3uM5syH{vKK-khp1u0JMIP+#T;ynpJvPQm
zxn*)(*uNV8C$dxM_*?_J%%r3@N%5JBil^=D2mO7sRc@jya7Sroq6n8z@^j`Wt>2a&
zsp4wmv(@rStgCg&qG|arL=~R(CfQjNRbsWBaK7sd96=S*x2@?if|9V_-aJj?Kp~Bz
z*vov)?;bX@cW`hJ?L`@^c#Y;)(`@S$(n<Q8uI2Ka+%?ks(SFySR+(==LoGg=DyLk%
zS1~`A9gd4T4F2rdy5#He0ZF)45{}RilB~I5ERfj~k+j%?;IMWXl`zMBujV_i_Ixo;
zJYMnXrt6wH>6mY9MUCRb2>c-9l}9b~u_pADH)p2ro|i`jwz{<^k1Jf!#l)LY&bUqL
z;B8snD5A3{i8d1fLz8>k4~T<S=fhL7m=A21M+I=vDB?b+@|E&Ra>lT2rr}mApNWkD
zYB$Tx(;+V&anAF`R4L4ivuh&34~_Wpk9ihjNcgMC;9FnkIzOc|UNG0-dyPAd=cGoy
z&r<X0v*P-_>F!iv>b=Uss+4B6)|a+bDEy$PLq^Q*mVY>oYMhpIe)<L79I4WCDs0#{
z^kvUoJ~WmleQn*Gm0<RKw4sHtQAw%rw5P4vdM<G$yI*AMGnRH^5=Cf=n~qDOYKv63
zR3~<ob>OBSvg-2RTHnmuM~5PB7iyL|DXfcz77-ig2(!*@QC@r+^o81Eje+5&caJ_U
zB^bX@VLJ>fKx?|CNo)xVDvlWMXttb;x9jQY(HP@38Z};ZQ59m$ei@FC*_uWs;&cz4
zNcKMI65r@o74bFGo|Y#ku1nWtPOOAtJaUQIst0jv=r%4KzX~mKy+ZAWls;ccUu*3M
zViKPa$()oEmzXJjiq&}ROf6IryvCCe)s(Ztsi&F!7T$_)rq<{gH>{>lEM@S7wIm(2
zZV(RKRl>HQ8kl6jzoI9$yj*p~(Je6~98HPG-7+nxb`9j_-*tW76;o;f8w~!AOTT3G
z^;C>t^xhRBUp;HD<Zs{?%rkxc?nfHiIYn*LX|+9Rx+9RFbo*~<GAB$~eF*N=y=8jX
zG(jgqXk5R3w<ZL)pB^8il`M?W?sDnTXZ)MyeNZuwoSl&K)hayv%dJ``sl&8O+ue9w
zt7mWWDAy;I+ad#+dp9|)t`ti%AwAtP`@5TMN%zbGs|Oc{-P)Qb;Lh`f%ypbPg>Sns
zUi9Nc5vdtJyXh_!@8#{^C_g6F62+eLxFBYOk7l%1L?L#Ce=pf|Yn0%H^mLZlde6g~
zS$V~Me1m~wu5an=7&gb@&6=7uY6#ym>%J7*@uW!gYmjcd4G@4chhTYw@R1e#>ku!@
zyR~CcZ0Co@#P0}W=Hi8SxSe-0nBI-|_OiWxau&m}s%`O+m?5Zn?Of%9h|-u_V|(_F
zj<;c6cg;ab<Mcvd3*7*pMc_;9px&<^%|6`MXeX@1ifj`;QxgXV|Dn&;nu=uh*u_SB
z2*q{Eq`l03X{T`Dix420H#9EPJFg23kT#yiB^dO4st@mwY$zk>{AzZ?H%(G17_P=7
zeQ)_&I(w3)Vd=89#7NxUs?BrUuNvN6u;(z4`(yo7BgSsoBe77zKy0<zx&-msqyV=$
zoy`AawQlJCnf3oHkgnd1=uB_i4QKx?5p2ZH`f;;;wDqdWX?GNN^O?erwY4Lr{*S3)
zKZWU=JXM|kO|5u8i2~t8qW|`8{S?_rYbrce<>jV$e1^{MS<I1Hym0w%(!NL*;;a@=
zRJi}jPNOLu$|{+AyswUPWpYx`W_DD$@jdSOw2*@D>bE!zTVweeRim&;iEbvY+k4f4
zr?3LYqKbv=D^35hoi}bPY`3z{Xx+0kb(;0=epP$_;nB#eg#N)=-x@xZc_otRXfuI-
zKUPD<*hb7hKcaqwxtPR%l_lmZEI$DE=mReR@4$K+n;$1*bgb|x>@P5=tCQ`D5zX{A
zSwSseKjNyo6U6i~#f?uEj>14(j^WKA_CX(7XGU!NWoBnO9sVI<C+b9|7IO*hVwuqN
zs$bQyxk%%p)|&0oLe&Dcy&lXx?f$2;cB31e@rO=t@@D1rl7zos-_3XKuMkI1ybJ6M
zO=-M6+VOgYE-vY><yA9w%*E-GQLWMV`N(FkFRo#+2}}}!Zq$>=Q;~8ksyWL>#y&Uf
z^(?H^SSv@BqZs=uCo}}4MU%gyl@?}>&5MU;(OY}Y&RI|UO4pSy9-B@+Q;!~<HZ%9)
z=2Zx19qn9oRoa}?K@eKdGM=4<y2zV`=T88R{g6Yhq?~EYH?5AU64x6$vc?qo+p7>f
zo>r)wvD*>T(ahKRyeooSTY1H?OmQ2M>kG(ATr+Y&iMXn9gYsF`4>u~_{&DZq8P_x8
zl8uidurGf66OZ>BJUkM|#MfxY&y*(!TMZ5iVH`bscEfvAs^F4*0bdnqSo!i)it$+S
zokY1bIgdBLwh|K_g@;rpwH&aB;XF^Zp_G)qSw2HH;XD!7+UGZTYGPif+;FK-XsaE6
zeP1P{IW&NTAGWd>TylJY2E(clTyNd)=1?x`EB9F@uX~8!$aHQJ%cdr^>9RcWIrM3!
z3yF)rz{-2jZp?L`&q;}Y?DppA%^}5Y#Rq9o^ZVtz_s0dSpTtuCXHREgEph5z+iIM%
z3*~`Nide@<YF9r$q%5f$SF(0&{;A)lSi@rA>5V^B1$A->;bn#P>4)YFGrjQ|wvt-W
z7hdzs%P8pVd*&{?FGbW7tNom)Q-m=IIsSM1j@GB%P9$!s^~DK4!R!|K+o!@v*62-M
z{foRssfyLc5F5K46iZU^SFRXkjg}Oto=ER^eFFopO5Qo00CTu&SMcV!hTL@JmUL&L
ziQ^_85o5<LBj<FGA~7d|h`#d~3B1&H=NCG#*3mJZY?qw-V-mxaUG0~3BqT#4&_;Fh
zewRE~Rr;Js+SbJ1BU*IBRwK+7U+#U!j<8(Eer^pNuC>h1$6;hHtj3O-1W#3J&0Ou>
zq}}xf2F59W(HidliC0LUj{U0Bg!JBXu}`;0gb;Cw4YK^=9{I<Emt?;=D<qo&;-mk2
zU1b~PefNd<KmR`DOlH!imctvT>$l$IjX1n9&v&@@YE$(c@|v^usgn6HRi7=nBCIr!
zbzpH%yEJE0&qE5fgrnRzR!grQyzz0DBbc_ybzEe*Uu!?wOv&9lRQ<~7p`jg%f2Q?w
zEMu=Q3%`97(aeNhdaL$nCy_v!4IFn)EX8q~BKvk9A@MB>ap9IetTxjm$VLp4W)V1S
z^aIMH$P+2zT=wk~*e0?2B%hpJ77dlm*v3}szsS#(f&X@9%<3W{wRlvBC?aZ}{v`ga
zHIgtr#-h=heaYIY+xN>$UOCd{SZO`|nb;*U2kw;TBW}d*w<i>Oodhce@uzr=+x{TW
z`&S3m{rxAJEqL^wM9qzOo8G!T<Fme`DdI%JCR`19V5$f?-0M4XPvx$d56(X>W8K4Q
zB^cs*%|m*knVXgRW`@s-kT}uyvkyHNCoh42#kMz3+*6WtYmrfXGK10AvZicm8~H1>
zq-p45ovZUtw4yO$<%i-VaXdyYld;sFI8OOwmEYu~_!WcCbi5fZe4np-NDZH%N6B+b
z3Vg86S!s0_$o8yxaQ2d^6PrC_PokINR(l0(k7&R-bavKb!?{)7x{Q|QPOKtir=LjQ
zsB3m~&yCQn^qRjpp!Z~IO6dB^@MS}n+hrW$a;IfTuf6k|O$xOS{NoU4pZuehhV2@>
z_1@jV9*-IIc;P4Uhu)Z@I^TT@5zP7-OhJxq-Sz&>JMqH)`hAX=6BvIby?)}I*&9qv
zdLb$F>JBmvmWjG=#I@*VdaAiK?Hj*&nM)F~+#i3*IW_(b6YTWQl)`keLU7OM{^nq_
z-Z;@pWKZx}CTgbo9M>$qmW^Ax#;}J1_wBi;DXJwt(P~q1X38yi7V~^4c<Vp&o?AK|
zvkL^zxW?LUljJdkVqI#_yv9>DDRmbslIxD5X$l^TaG!t4FeM;#i5PJrnD?4MwZ=u$
zSpGefqUy7+JDZ#GgjzUI@ox}f0#R{|doA0Yv!6XZJ)ipd6)r6~l0|f<*}~P+z8<pJ
zxA9f>6MXT4YllN}|Mo@+|HP)gT5a{@0&xJH@RQef%XvR~?<g;)eurw}rIib98d^A{
zq(tWiA70)d<Q96=gZg@P9b>E<#>c40(D>wq3m4)aMQt9SPH-?o%|kI?Smu0rE}ovW
zdVZ@-F;2-STIQjMEi*F-7R_~eOG#!SAN6@55)1}pE>3(DGU_gy4{M@65f*y5N!i6|
zrMc~V&nqUp{h;gw@9DXhTRgmkY25FMii-B$evO;k-;G0iGly{5M}?-Q*;SOD`VmTA
zXO7jP-?)yiJ?}ViI6Ir`>E(41h586Wz6ND1xD-}fJvr*@iw6?Bs@r+fz}vYc|F>I>
ztY?A|HJL;f8?AM*_ZyzZ%%`iHG_D$c!c>ItE>Qduos+OL^f=-UIUk!=*#djmZPW`M
zeKBw7*t+9NA4T9*;ko6zUQ=;G#7Zg6(1My`%6lG2QErv@BFGE?8ZmR2dBe!oK$vVR
zq9h{ELWOE^oBdi|o-|CwnZ4Q@bvUi}gQc|>8@i8e2kWCZ^eOy7F2wdJCJrN4^qI2F
zT{LSt;JTs9h1zD671FDG>YU<FxQsq(j*7#*xU8h~1DtSS+78nLvgSM?ffF|FrKj_g
zjjo`o07sgtnwp<5tu~qO!TEmNq@7&~9oZborXX#hbYpi0b|&*)$*<fz&4@BCFnsFj
znhjrjBI+s(Y^jkz5xSLa(9U<F<m=ZwI74LZ>Ow<0eKh9w?F5kh!b9e3tFNg^YZT<c
z1Fs@)5~&i6pK@;|tvcw&VMkrf3PnBkiK-q9eYMmgVn(J4m4~rWEk*;PrX9cTR`vxe
znZDyMZ{CBB{MpfA?MZI#V7IRM4f4LUsc8@~0^5_U|GaqV()&-J^gy;uw%f5eWIGiS
z$<~$)+@8?M$~K{NL+iBR;o)c!WvLFsv6U{)@HkiBbJlipY@`+$*}9C~>FLupS!<RC
zlF{EvJ8pT_Uz>Yn`=W5cw4wrLTMt=s!~IsMm7HQ4tGtLAF!L%aFV6?3KG^LMx(0?W
zuzT<>j8H(!18zw0UvSc}I+vSNtmvPXo=#>qkQH$-5+C|GoXsVrbcH>O36(;@KZRL9
zS`V&eM8g|^(dHKrZ_0xp4wzNoV_|3iBJbj3)y$p`CKB&KqzE(s`2Bc8x$W*fZ1Md3
zJg6z0+M2;6$`EL$3yd3s<Ku;(babgxbel`vueuqKj}`r4v@9XL1r$xtv*~UxmrXf)
z1v$C<uH)-2I|&Xn-^G8N{w*!S@X%dCgatijmh>lRaA*kedVcCG+cV{ol5X%=0<i?_
zy?aJlF~j1)9&ezxH|y1_S6N)_w=i&cejXmyK>(4LmS#)-UW$i-a_Q*O?k<@$3gp@j
zv1e@|Zl`O!^oJG~r_A0w>~v0h2?w7Zzq3;bKTX+}FDGKlDI(Ph6fRu~lk<DP_s&%E
z0|;7$E#dfk5XWoPI)j6$fJC9Eum2%C+mopognCugj=5!!dM~3;KY=wf*MgCvTze6}
zvxWmJ>>!S-CO1)_M!OPN@4IJ2AQ0gE)9cXB66qyXPbG(No?ye{@mIcS_knQ_umOZI
zz+HmlZwOLm8Y<H4d&{o<zf)(4DrhwGNTiH8hh7;75jfkTe++?v4G<!eMFsx@GDELj
z4KX%0HqONm8&cJ3TKKh*v4foo=i!kHem7*Pt4#BtB>vVVb+umzjpz-I&;8s+jFtOY
z2p;#E5Yi=b1By;`?_4O9)qEQFqn3m9W&=CgalQwSxasJ)q?JnD`%u!jQ+MU>-$k&P
zr3?lU-2muVAU+NB^{xMS{f~ti7&{QL!-~`P_)cZ1jYK^~?}YS)I~e?dJ#f-ouP0oE
zj9sD!n{U^s{DwU=QWkY!g%kj91)SxBn!KpLSEo&V7}#}4>^*?@O6H;b`}gIBzP{Yl
z0?2cL=?ThiLKn)@!z1*uG`&n>BqQra%rWxk0GhJ2JgmBd!5zHBTlep$8W|gtFS>da
z_#g;c!Yvd!4>~)R15*(6|7hs$j)jr_CR|f9Q_~RJbVL$3EP(nF)Jo?uk7!%}5{YCI
z83<`R#F#Uqu)GS1)Trq&I%j&N3V1U(T=jUJ?+5s+1484#v1on$7VJP%6B~GR2PUV$
zkcEG;`IDj$q9`<)%)!1UX^;%B6MosMks5ja$8n@m4Mn=s`__N1*4i~bjMPLPKPldI
z7yi5TfB$_G*M3eYyg2-rb}>bmlq`0V#MYo;g5TzA7JR!nm0|a)*6%UUe}M~@!toN0
zW_nth{gWqp;8g{%RXmr(hMqWmgFD#X!*-@@AY;L?qPf<vu|5D8`BiAZP66iy8!y*W
ziz7jrMam}Bbe8t@w#p!Py|7rsIP=^NxNJnf@?1r;JV95U#CfYRghWXYsHkwOlUMSR
zhd3qSM|rll^$83RGHzxRCZ>4Jt^E0OUij?762w2?8F9US_4)o+!`lX83|vt~sjqXu
zz-;m&lAqrJL~UdNH`s2%dm?}LkTM98@$#Ub<KXQguC0|oUJD9uQM?&seuX0Q@(tdN
zQenu{p)yQhz-MQt1qN$@H^##TFP>xwZLovi0ftULNPFo@!G)1baRNs=ELSTnYbHFE
zhYv+U+S_Tdl}TcOzRe0OuodFM<Rq_61E&>>R>iTWaU-u*o)qaLgQHU@rnDL?u{7V^
z>ftXCIX>kj^>gXxgVonTkMz&04Iwhm?CZt<Ub&>}bp3i1kficq><=U@*fp>SI=pOw
zGJWj~;V;5h;a#__%SwiDj!Np?W@ct~@b_1Ji^^h+e0}MGOWF~-#q;q+z1y|^N9J%;
zJLWeY!Qln+HcgaUO%kadN0LnF3Yx=C3zuT0LPxL?kmt!u+=wwnV3Lrfi)M_uV)Gb&
zF=eBihw~u(p~OG3DigsfglZZX8~fDOw#!aU-oI5>QGrka(SI)xj^F|PJ3PSELhK<k
zbb;FSZ|*xUkg;??mkEIszQIjk-1qc!chk&1|KWmFdmo;qiDrqxJ|)ZLU0NVVlJ|<e
zQ7i)+y8>|2fQ;%0<m(3dkDT>YRb<XDBteg(Lto(NIR2j**&};<1IRpC7f=dt3lj-Y
zRv0|b&~owf81`6x6(jdxXXA05^$XDP_3-ohDdd&63aWZjFDkxGyrAMK5omqWbq({v
zGB;<r5G}?>duWBKxuP%FqVuy>P{POB`YqXHY&;4C4%rjH<>JY0G8wP;FA~PUa@7K;
z8ZcnA{|wxiS7D<sut0Mm7=Sq#*?$5!$xFZiYb0#^cmI(RtizNWe}Dg8lQ;>;Px`GS
z<)@5Y6JnikY5MzbqrKOKgPCu|s?U)DY<Jz`JkH0d-_AUBOG`duo}ypWx_O-0H#ltN
z)s?usJQ3fG?+PiqM#f)}Rda5Y#XQFe)lkB~l2zdPfyXp=;-B&fzYjo63;6Q2tjuOI
zwX#xacEg+O5~;862rF_yxb}N^_#1Yj5bJZtIPIgGRD>UC>(_fCkXj(fg3}?Q%?nB$
zFjl^(paABZCSd*YY;`ga)<w+C&22(ZZkT{m1((Ze=)nv3FBpmwQLVb_)OV@#7C;fa
zF^SRFr;`ouV(<7dJK#-W-;TcWAK&iO%e?SEt8H)%R9eEM>0Zj!aSoq&LzG}H#q*}h
zzDz{5IF;+aPTIn{(XkD$)xx=lO=4keOViWp5bG!<{7Eb6;OG+c8HGZDlH{hldntUN
z*tobm5+^SL8LSPMp5hoX?;XH8gD|iRAb-GxB3>r@8Q#l^e-WA7SCy%99WI>Fo3J#&
zq%d>jyb1qWQ}Zz_jOI&uIoX#FwDuupaOxps?7pMm-8bm?g^bs>N_*Vr0?z~70}W2j
z{!2n|<Wmw?FQt@1aRK^RdpMzlKujN4?$QjZ3*Nm137uRHf>;(DBd?O3u7uX7rmX0c
zz(oOk%7V3wVDZ!f*{Xr;HVS_P{5OF7Nw4$$RYlVC${?*vquT7P0STAo1DVKWmY1v(
zQwtF5peR6-FLZc#IGOr~wdjrO!K?!&l%A3C2O6BoJ~2531JKz)^iSD+FmvgWy;B<M
zSR`I};J-EqHmj>93Kflq*+N^8fpZ^UJ5eOEB|ZIEDkIDK)JkZRuG7w_tzAs9T)uvM
zPFIVUhX?i93NSnT`}YYb>+Qko6)xwMKzQKfR|3XOKztm8EBSk5L_>glblTcv#T{Y)
z83f6PDQ7#Lz)^eq_U)CMc*Q7O-O534x87Y%uDZ>&z<)Q+;0wtLUl0aBPI>p`%N=+G
zDe8|BKWp$gJ(V`M?Bp6D++puKCR6diBaSkAwf&eaZN@yN7#{b+^0EZh8>&6{Dd91&
z9FX1tYFq2QA%fL|8qD%BEN6kD@-pN@I3Z*Upp_Ki2>6iXS{Y<5mrH*<hR{m(k$v{;
zy@K;N=<~G#i}-b1Mq|9ERj^<(cn732U|>!E?JU<Mu=p-Qg;zWPg^fXqfy7CV`EGtP
zp%*ee1iagC@=n1PA|0?TC2#mJyxA*&())cE7?<Q!aRe&1#l-@6WU%=Vv=BcO<|Bj)
z__sZLeE8}%R!64<Z3bebHeFO)(398Z_6mE{uRQxTaNQ}5?oeS|DCqC_D;$GUtsp{&
z>VfjL=?eNMu`w~Wkp;Y`PRYy22o?Bsr0XL`ipj*;=7kUgxRNQqyP)i&$f#bY!av)X
zi-83SP1XD_Xozo#w>+tX<wTJ*2M^z}K00^S;o-x%Gq&ML;uuKpAmz`-RzU12y~y9e
z@wNqlJ`L5!PjK>qvJwj1tdbHL@~RY>7V`%vuh=M^4n%;-Io#|1661JQm3gu2mchap
zc_k&DCJTJ4BC<KD^?{3zoL$j1wU?lI|6q8YTEFvpndwRUW``#B3A2{}&jqNtp{omJ
zuOqlFL5R_bzo+*QS|_rg+r=e|K(L0qM*Wu2qN}_6(c{Mw!bR^gwBt45<bkr85mOJ<
zCR`wUFw=x`1oGP*Ev<0qZRIM+w@rGzd&6~smzOy_SbD|DCdJXdYw_4&y0X>^-{_^T
zLzw+r+q-&3)mvc`Memy7wQBegk-BF%)v76aO?t4lR7@kIqaR^;tA~e2L3z0X+xS{M
zR!H^N?-QS>R?MpeG4L%@X1JQJ?7_!@vvLzEm4-;R86Z}vogvAi3Z{r&e5+u?VTFQz
z<sKNY!kMA^CGK0}t>x(W0ZyOnc&^-AUe3(>th|AUEY)103=}ka@reZ|HfOWV|4i*q
z)leP&8}pi2)KcBOX`lj47Uk<~qFAe`B=>3+oc~EoqX0R${3~94)ib6r;lnxp-T6?&
zy{5F3vj#QSAw<U&R8I9&?KE2zh0;s4>{C`CX({caJA;;!dkbfUVpdTAPNpMRRYR@T
z$Q~fFD))mAy8uQ>kgTx+o1}NG{Ha3`uGFVQpDt93-|#<ap7c3vGdyHC^Hq@WardYd
zt23ePVqYrW4ftCRoICmY&%A@$V|=H<yF$~H09bWEt^}ZP${vqW!P*BD3uFF+R7Cb2
zq3h8@N?ZRPA2$Mt=Czk^M}bZY-7GMN%c`q+W&h9}!w>d-4X^3_`RQ^8``Pbkh&&M3
zf5KfNi`bI?lamv_y80rSllu{A2BARgp5J%|Dljngpdo39$vnWSi0SFKu=6N}Etn@0
zoNV@_q@;prWx>LCc=#@u4rsp*_uv1!!D;>E)|UP4jF&lZX>=o)k(X+&9K7F%z%4dt
zt5HU`p11;m<l&;lvbJ>6K5;+h0kxbnlsd4t79a<o1JE11FolL3h^S<@_T>7`&g+n3
zY2R7FVI>^MNI4bzMvvRoxuYpq<&2RsSeeRBd%o_yG@procg3{z*|x8hnNCK)&(_!n
zTCM`a^S1B*rW_BUYjlfaZk?)L*Y02&Iw6<x843;Z7AcU5fQq0qWWY5A1C4<&7?kS4
z2j^*>;UU1c9s?RQEiEniF0-Y34*vS}6#jJa!z$>ZKYX|jUKPe*R+}qanB&XjvqIew
zE2DFfT60cT?=~bqvaAy<#IpHU71Vt2E0&r~Tc(9|9in>ulWRc6rBH478T|n)1XiQ#
z$dR@Som5ol!a9DUw&?Es4s-@&yK(U4{H`W5=^|`~5HOL=_=J1oR4ns3@iA2Yl3h#S
zP4)BV&&1(uNbI)(6QC0h=aX{RXfqjI3c7^-)%(&DQV6tQcRu}M*h-h|y0iQV<ZTYn
zegM`I=(WlM`_EJ(ABr84jby=A9SRjWn&bGeIw4LO^Od38&dcnl8G49#`%3kHUe#{<
zLW6eMg1)EGOrCKJQq-rt7j$`1gsIousZCfQ1u$N)Enm_aG8hG4El?HJ%kWL#9p77>
zYF@ojqpp7D_eUc~-Z6-M{|h^uCQ%3bJMLx{gO^Sszz+Fem|sJZ-P*yr!I`fuv9+#0
zXfi@@LuFUJU*Zh9hgP1ng+<lZZlm-Z?HUCo%4V-KS4wpk;*<~@Kl}PP{@fY+(4zfc
zB8Zk=pPQOS*CeDr_Cc|HYpEasO7gA{3zk#ZqM|g{xBAopiDtJ~P{@7`r7YIE^~fA9
zbAe6+U*x3zIJg``1)3803Yssl{6d`6*ba(;fC;aXwx*wP^kpbw#YL)*u@&RAbD}dB
z_pAL=e|YDK;t8+RrG%p3#Lx4?yPB=Z)W`m`Ozj~2H(#`VQa$b&MW{&6F}M)|OOZRk
ztN|PA)s(Dq4$FU-BF@p%;J?<&4S|6yGXyn(JjiN*!l>Q=`|P25xA*cg{-0CaX$|h<
zgK@Sl`Z8_vY%D7T_Zo(bKqqsx?}#+>l@v8mk}w~wbr;ul1^J(^b?dx0t^qtluAIS(
zm-|wui&U_;3+qJG`^HwRf>n($TqnDb6qiiD4W2JAVlOCPn%lq({{qBtvKAyP(O10g
zWw%e;w>CTYhaZI{-MDdMbL;PYIHA|z;##S^poGzPDsS_L58_XpooQ!BfpH6;d2Xtj
zSjSj4{(ly<Id9f4D=Hd*7Y`W?J~}*LvYCB#Hg>+6b|IEwHF@Jkmp<J6aNYI#qu8kr
zBc_?5w2S6oXm&4MeFlNI4jSXYt$hmaIb`|$bapyID1?_<&qSAQF<-DsRA@j$tr$TH
zuw+Vp?%GHPmVZp+IdR6f=u~~rP}niFaU?Zj1!`6M@Vu*i`iVjPZA(LZ79n#h-+tU!
zE_t@Fu<ZoxQb#8`ze2Nx7<WA<>jNPX;GJxF>a*0kgc>$$ybY<Su(0WeKv#Hpi&zMN
zQ)&rNBH7&N)#TYIub-%OBe|j9Or;|+Xy09V><*wxAce)gyPDfo0NpVBZ>QlcMd4%m
zR7qLw+$T>85Bza0FE5u9{{WCF`B`K|TjzHBZ1-X2X$82ggD$xu{1e={lzF@7OCEUS
z?)eVd&0cwSrv`U=;L$;^yWO(<r%b2z<E`7-qHa0MG_@;N{~JPSinu>F7>9}p`f(l<
zzV=kXiY?is@{VL|l8j0Xx9nd)q5-<>gN#9Eaf+8$PMjB|G~;Whlz!ik^{Dtyw|t?Y
zFH_y)=sVIY^p)Y~23Ywu3IeQt=<mMCb7<JaDoBU*YrhwmJehQ{mEp&Tl>{<OsRn%3
zVBTG;TgEP4lI#Vgdr;pN$2kQ-?5B8j86}^VWYZx~Y{_1?N58SGS^2l4ybbV<9vQ3z
zL`crc^UJ*fu<06r69dtgtS*B?zq?XR{NDkCXa_+4$KmneVfyt~odcgqPz^#SN+%?Z
z`~!fqFaVz_sGcEV-=|`yHrn_EaZYK|oo`<&z9e+8QO!<O_wR(057M{L;#QNO`}dQ-
z(YHI(Vh?;yW-^g%Or;4-5=1vCSm*{j2{BNNLhkFrCL!lAU%JZW!VN8)`K^VazY&t(
zh`D$3d?5nbPI7ZVPOZ_@``<L=@v{${WzA1QRyk4Q;%p5+85w8Y-E>&^fcu6F0;3?Z
z$PXI96pr(AHt^d}PmrD730H*NS&0wFN98=?WV?-IuW`3v@8*hsR}AFdpmF>JF&LVy
zW^CFeF_y1#R3D=7zeY#-njX368L`8t+Uh-Fq!6*X3X>J6?&AZc`pUj?a#RaJ_enN`
z6z(3SOYF3T?UiJ4Z7`9%G7xO4x+tGm05-N(LMb(CiljNC!4l!!GW9ZBix!bJv=JPO
zWUq2(6W8FHXT=QpoJ0y!{gX4{Q?#4pN>Ue&<#eIEQo(kan+-qX^OouyZvWxql(Yg6
zv?dBi0Pxl{Q^FNV-<lxwvzdxwL^!%x!&WI<+*bYmahfUcPcz2haI1!v_fYJtwa&*n
z14V0S+-+^oQr%G_!K46iR@;TZpzNZ<!vAU9mZUv*y}Z2(pks$(PKwifC$a#vbmYBQ
zXQ#h>Dx92R=59(w(NCE1w+@0h<<iJ=@1BwmOH22j%mF+C*tu**E^iV@g8%>|tBg#O
z8RNz}yFf@Pr~DLmr50yiw}rHwRy2c`mF#s|d{^c2<@W~%zF;a!Uj4`E4ZR)|0`{)1
zGKt}{ZePAj!IzRvnaRR&dx~8?K0avE#B7RhhG<G6&$S@FY!vp7@C&T`tDK&8RyeLA
z;K%B|tx?n7(-f>Z$@Y*my3Lz+)x?y3Jq4#Ac9Qln|HwuRtATK#!2{c+Uun}*@}G#s
zdSkB>(JhR5*-|ylj#UrDBRHVKAmh&~D}^9L+kYuKX)Z))NOgaVuchoTFn9sT19;i8
zUi6{ZfvHY*z!;~_q;9Oh)Q&#XMts+AHQXiFCJ=iBDX;>)5_*n%2B5-%D@OM226R*G
zZn#F)hr!>iPhAjgLL9{w#cFtTwgS-y5D4-z6>`T6Fz;<eebt=b>vez=0OVpDnVYT(
zz#O<D=RzETnkrPK?G8e6p`?-YPW{%yGX${1Exyh*l4)jP;RrJefTSR=oc7->uC4t9
z-8%Fsw1-pM5m46)O8tB*f_aju9WGsSy0O_YlQrpnV!g&~goS_Dp0x&jty2^7<=BXM
zW@yl6BaUH$nvV1)!ZkatS!R)-XLjHgrm@(Pa2;3jdE+?cF|8B^WoD(5&DD*3J8K$o
zfHPeXYll|(#rIH3(-&!et|@Wh4x<_;FW%^FNE1uk*l_1ZBHtGkNtF^$6L#lPotu=i
zZv)zqZD0WO7c#OV0K4TH`=Qo_E@)3|S$%=*1l{p<NS8$3*h6q3j^pQt?~l)#3sy78
z_)V}Soxt4s$}&jeO;1U=Sz_`N_N0JL9Xy&7r!Jmm_rn!1`qn|SH$I+0R<uD&POf2d
zr++Z}<`72-Lb4>ViX>x?Jhq1bS@`bECh>>{F^jLhoAMgJ`r~L}!T#D7K<i`-7(P8r
zqiBla`C|Y=X_^yv*j=NKR}ttQiFL)_AFDi__*pcniZHp}MUpY!(fB-6ek&Ifzwa3@
ztPC=DyZ=YicgIuRzyFuLsR&sKA)B&S6dlQ`jI3luviBZwQszM@Gb^{1kiGW~8D(#>
z_vUxKKi|jiulpvC!#VHQxUT2*9Lakib3XwN%E@NsNh6qdu=>E*9cWqx@O{-kLbb=!
z=PvN`|K3ilt@A5UG&i_)tUnR2!4c4l+>$U0B0oEM6fCFLn^2CC@2jA0pMUdM{k1@t
z|INy*$CZVFyXNWQn%F{aq+i$t2+qbD=NR(U2JVtJFOiw<?(YZf6+;07D+2-8Z$$TO
zdV!u0-WLo?pwyxj_5t951ov>2vKV5bq8FhR0Y~)4q~I=HfFHF{@b<6#_%6V>W)#Qu
znA|PihyV5AyYr&#KrXh)?H;>g0GZ&y!ojBm@QqObsSkN4zWRC?^QTTz1R;g?@iVv>
zJV4n!_w#9`A0V5ov{F`9MzUiNg*2dy-?{@u#=g16k%bZV{;)vuI90v3ufev#rAdWO
znA`Xr(^2UCEpG`@y`HKmdV_<bkMl&LzBS_Mv-a{#diyhXK5kT6=O|r-rPuN-yQ~(~
z-d?^?9@rHDcF!B8^NSqJc~$x{xr~oOzm3u{q>b)AI$pte-`BXayuym|=6f>Gn7sE&
z>nhp02SeiQGlCp0S#m>$Lzhbvc09K~25+fdo)PCz?Z>Zo@G}w2aE~mT{ZfUlx%rNF
zceh?%G55T^Js6tf2GN&TTJL#Rtx3NBGoV<zmqLJKzrYG+%hn`%%6%QtLvunzK=}Dm
zxLoq)KW&4B7J?hkbDlHM(|?0brn*`m&TMyxA%cpb;qz<qco;YW`j-d!8#XmZ3LuMb
zf2qTWB{cl2tJBc)bp3e)OO5U!)XGpC3R1kI<NTZP(hJ6tU<_YTR)+L)AZ(VetdYqi
zUPZ=Y+0S5Sq{9&p^Cjq;xEc!>uQVT)?nrws9sD4N>6fV}eV7;u^8n`NnDm;O8j_?k
zI6+vmrwhY$w1A&QPL7t_guz;u==2*~OOlT}OoC_utPZ;k%+ZdULwQ|Y#jOyF&Uqhs
zkz>cv6-^?+Cj&^LnCtEq-{a#Vrt@<qwlkJN$~pKdu~;ES2KZPi7}fx$^)vA2B9|sP
zJYYM+q)(nDx#0UZO><9BH>@?WSM6m4OLj8rGR|C;bN#0Lo?sTW_r!GEA@axbbIAbv
zChTdEX$DiCU2L{Q$W~sw-bzDCc&(+2Y*a(PFMvusKzz2@b?%8#_^=B^+aUEblupdS
zQQmc$5E`Bj70CDDDe%@N_(OjIYT+<IG55f};osiMFo^lp0B`lZ30tKhsL{}C?>WVF
zzUEtc4)Tls<h`3;3f)oRu>o#0J@g28^*6Y7Q7U(s5$xe$1OsS3lB)wZC<p3aWf)7u
z?jGE%prsjtaUH<fh(`oojlT^;gB_zw<MIT@+0N1>mcrUvrv2Ae*4D^N09drQJ}v?1
z9Ig&F8Tb<Yf>&_z($}pRfddIDV%vPu_I^n2`R)*to<a|+CE!}!7`9G*ZFH8~OqaY_
zwgKNt>JRq(SA0UkL3RX87hYdh7VB@G4BQ<*-aPO?oiUv=Vv5ZU@_Y_gv%JpxVp!V?
zP{$?YKH#f}rq-sU>F7<Kmg7A7`}*A%=)0K(wNu&6$Nq^a%o&OKFUCDD*VBZH?{+#c
zqn)0ri>;{(p)n4DvOZfa1oOrp{}F-m=;5PEk?Ba*JDaDF^015cIYb*#JA|~sS~GiG
z0(O9q6%+V<LY|@J@ht+j_7T7|VfdR;P<t*;BbL3M*Q(GpP01T_?JT%aX(2gHEQQ$M
zyT{i}th-qnmp`ZKMX>@n70W2?iHsVcK|~YbE>fWoe=NYuphkiNZWnBb-nnhPqnDi9
z*u0_FI+4hpLOd@I@Ed~h*4O*NxEGn6^<-^5=Usn}L{!<>@VUN&dJsBf61t0-Zllv`
zla9qYW<owUW4$ELy`}DI6Y}!vWJXTcXB3S*cCPCN%Wv~#B3Yk6(GE^bxmY{qr3bng
z@5dPEVKqv}c$+vL6r6_3KhUz~|7>o{;Ij22+7&~QL{nOoy1_L4N<*qDtkH&s{_@s|
z6#XYkN$KAI{0koK-FT2iWfaI|sQR+mYtb8Xg_Zli!DAgZ*8&1T#|Np8iF96TSxdV>
zO@hqAb+>E5*9fYs)bwDnkVf77FOvGnF7Mi^XSC~5@DDs*H61kwWu6;EUzdZ@J{!7*
z%`s%4gvnJ`#k{aL0q^~&vjgF2T#-vv!$<uLXi<y%(GAPtR*f)ep(MelQy3sChVT+1
z_UjPv`~uPvz$Jxg+81Vt%<2_Ceq6P5(l^+;9jP^G?z?aMivH?HD#O($?v#XiWn)*i
zoNbOTPV$o>EfR=kahc0QXbShm#zi`e!fMSldGgzDjF_Dxnd6#RQL-l7G!MM+2r9gt
zis~<vprn%H9i%GYlB2dISY^X!oJrkOvxLwkfCn!KPHayFTN)Gk&KL^i%$4nyR=;1y
zYibl(52Y1sAtD^4-HA2yMs$ocwU9FdQFtUt%5XSd+7-CX$Xi}21>pDorz)7fA<;lE
zY`GI7EiM*WQzQ`?tslF>__?=~iz2Ji<I~m;W!KP28v56Zw&+HtU%DFF%S}n2Y92?+
z-&tp9(3MW+P(mBdP+(5zvvxfWD+>#MZX?bU%u(ZD*be9FO>*EmB<Mp7GlR4J<oG!0
zhhK}7FB~X?IwLyi*&ITeW6#BXM43DuEEcDniC$@vJUVqJ*kC><-@4_)h<oIL@qXTK
z<9T>i`*+cY4TY)Ad4X9t47hqW$#%Bbj<zaI_qpF+FuCwho3ILpU^)I#9GwD-#f(w9
zE*TByZTIcVX-!;qopY(**s)%zc1gv<U0@b{g=ieNla@GRfiXw-@di1q!X(w-7#{83
zCznJIW}Dqbn_TjLEz96f?&y$sHNELFUc$2ZM5?22_hb1#>P88nqOMTy#@(%XQSRxx
z(^8JsR#v~ih}d9-LeCBmRwh)gP%Q-&<pDa`_-DaDZLJ0Ga|M4zpY42`1MEHKY1=5j
z-H868I?s(O@4`QBf&S+ErmqQUI~^tE1n<s_I{p=TC_d`=mbIW9_&?u$#O?%utM6Z`
z<gXhf*3Y*z75<KMIZ8O%noHk#9iqNXh9*hyF~+^{8qT8x%ct6$&j-V?U@BqxXK6BG
z_W-H^7q+}d$HOvX=j3c>xTqbB66So3JW(1hBAKY&3-7WGn{=kD!lkZ0huu@-{T@a_
zBOi@F_Ef3Njf~ES(0%Y;I#6W9I+HDIl4x?}@P<M^qFVl!=eA@Y1u@~<R)Z*GMHr>V
zSJPj4L^IjgDq4<1YV;ubtFE=7eCLQhXGKGnjll+9_ZH>&qsU9!GcM%j#x@O6R|HI^
zcc?rHq<uUmVZlIp=`G`DXEuk}8|3a4t+RN4i`EP4&5~d=A~?eC<@k_*&y}W|W-KR6
zUc<Y!%w7+(aP~pKE|!>tY9mcr69c8lZ$JoOwgq!h;8X`mpHwCcCm8ApGw6nqn&}lR
zbu#J~R-hS6=@h-!&n6ct-d;c9JC80nA0s`;#GG%dS}c~bo$cqjJ2%b`1rR<sl$!mg
z{LkP9&i2*I=CbP#gT>@4Y~5T}D-+97^b8HrQncfvq3ltw^r-D^@D<wU?eARwcp1Nn
zbM@es)A7y0z#)0o)=|ljFw)0*-#^|%_qc`_*seCL`7bCj&65l*cl1}gEnrooJWA-R
zCX4Z|++u0kq4-<94E-mZQpy4q9zh|aR6GW;;XM19iDtjpq?3NfM4;Qd=#2fsLIP@z
z@>LDuo=AjTJKx-zEIumcvI6eC5`bPnF^KhhQ`Vb<`?_~|{cY;zEBu>5RC*H@y`g!t
zE$iw#i7Nr3AC7BQinKrSEVC$>24V%6J%4Ae#TM9L&Pn&51ol><2YX>@b9%W;or4C$
zY-Boos0g|MkODJ{G%QD$vEkefDhGHiI@4n0%|`P_Y6L_?_yOQb$iv5n-(_cG15gKN
z2)05a*JX*lbi%U?K7Q5?UkZ97r&B8W{^}B>4d&&A=qn`%M*=nkOtAZ4%p;oHA<wXY
z-a&z}PRaN06dk@VPhrS_x)+3CGnL-}0YNgUrq38vDyWrpK^bb$A5dq+eZ)yg<Yx|l
z67aAxfYZ0OwuUe$ip26~wKsna8~QM~4#ZsAX)<Hh=q8>imX>qdVA)yf9T2`au`#I+
zzwKNQSp!N@+nK59=~M)Mad!Ts)8A_2tii5q9>#$E@IHc*r+r5T|BS|~4{teGd^W?l
zw#Cu8Ow5b&R;>m#UvrcsuI3j!_aS+Hmb+4<V@Y7dOn=WGPzR&899=Ubs)gFmU@R&+
z`MI-E90&K+k7=9_Bz&FmlT^{Oi2wsyzIj7@vYpC}M4LfF`CYQJGiZ~MBMvAT1SyBB
zt5!4~(Czj0MMB!86A)lp-nksZtn>Rpvd1m&)J=Ylv6ZAt2h%+Ez)a#mS?Xv{M6vG8
za*EdCu$`@yPMX=4BbzL04BGv8HjA$NHZ}z?T^odh3!o=kR~iF~Gp9hb;+&G4hJu~D
z{_EVPkIi2pI9b$|cw-ZYO5pTMumfCp14xo~y>eUyOTyxFkjXT-Rt0PW@srT}rO&~*
z3u1#u@hn_+iI3t69?yOi?X}`vffh#d&zVLaC5c-2czF@a&XtwFWKwW!_+?V&$!i^B
ze=@ieBd9=`(Q=B6Jzxe05E)4oPGUmutME~l+pfb>swRYu^{(TcPD^d0yWiRpzW{Vw
z<-hx#6NRDR74Wkg`S+6dodXx4@v*Rx0PnQ&`&;|Z>9l3Y3IEX9xw(nuW=R3f0NOIO
zhC$0U?im0T5HbV|%fPA=@C_{psj@2s=>0>1x(O`F&++knfQZ2RAJHzUu9o6n4zTm&
z@}l{$iKAdj%^apEJEEpJwvodLHShmen|FT;yNrYF7xd4M%&Ne#$m+8lQ0RbQ26K-<
z?0L7+g2%iuO7&4<Fv*5W4G#6+$lciViE>C`(Z1YQDDBeb-SI}Wi{4H8FT1_)L4i{j
zWmrL700v>0!QgTkaP9#}@?b;tCP@(|N#G9T?!&sTKh#icvzP37gg8#Y`)Z>P2o!g`
zYV5yGWuH8JqXi||ssbe*+)bI#wIZCT#$q5#fnDfVu*mv{pbYj{OrWq;`@J6b5|#zj
z;Ek1PV-KD{HwqOvgd)m3CJ6W!4}kBA^4si2y{-I3-jzBD#+gdLy>DtAGru=MpQUPW
zU+z*V<4YA+tD3OuU!$sbaEOxScSnx5a${XzmeZ%GVpa(qHJ%)b=3s|YY+xwB9QL)R
zryi_602hdc$^bj$erqhnhwe4h*AHCF>c5L*b@uf3KF`wxv?r$i>6Y-O^{d`rtq#}1
zT1Q?<=0f5bv@A%fM_d4OIBIh|gRwYpEyK%7?k<kn)I#^^m0?qRrfKN86^LO4^04~|
zGk9k+y=qx#>1u_f41{z>a)T(WDN-Ig%<qVIT7!XWf0Nz_b_o*m2bnZ5%S0#<lYq@(
zKWI>S0DMr4WvkvU5F_s`y6!*+0tGM8ADgQdP@%#63096e+~WK<;EmC4aB8m8NJdA@
z+PbQ)-z>ui4uu5&Dhx+3mYqn;34U*1k=bpwvWm*Txw!|x_7Inm83y}1SFmg;0_tMG
zsk*@BTH$;5aOtZ7<A+ktFCVz92x3;;BdVWI%XbPvN{yT!>aOKKh&M+wzsjnM$>UOT
zJ^6_+Gc&?_>-K}UH?Pm6RJ|&APblZBMw0IeB0rUwYQp*;Nt|0v_~edLvEFxKVp&?M
z1#F9xlM_e>k9o$I1dbSefOLW{SimhT#jke*H{I;*0V+rEx_3+PU-(hKc%YmM$<5<-
zkxk$~jDG(Axd0w^f<%p%-@J<urno98`u7d+)sXW&3a9rno&uqZqa#@BTf(aWNCQNR
z(=|qLBOy79rnDqeccs*+6q)z;S1bD^Agq;c;NH*+pgqUf{#+EFNV7IFI|Y&f3}p!b
zwz1IeMi^{xNdRoP4CLoJ3dn;(l38G?CHL)o7VtHekl$N|i&BzVFqu$hU~XYr@;f>7
z@$k?EmIAFC7SVk_Y9d@9Dx+t0!J3IKOjR2QAAtLVB<(%Odo%sjf%Aayv~w)=CSXCz
zQ_jt9Q+zoJk5YQ(eTeH=M_*|?cO?ToGR9}xnPUAV{b-IVy>euoZAPq|Ur1o;1^+Go
zbswSo)~2rbt)WUzQw$rpxPCXqhVJ`p(GVJjhlOG7q)|ffF5NL!5V+@ZqG0VIhZR8{
z?FD#%!a<R^tec@Baq%}wAfH2jNmA18Baf+SPt?B^OU@fIfZjr^PJ@J4)Vn9_33k9z
zf#&|@FZfJgtN|YjZmPxWZ#;f!jD4zdyP@IG<_S$z#~!<?OB5f3ybkelv#Sn=ta!2F
zfZ^@!`Nv3f6s>7;=X`?#Jxq~}c%j*(Ib7M?R72nxlP@q*3fP8Vok9o>0Xpj!osl}d
zl({1&w(hS%(B;!z3);Ntsfy_@907SB0~1pYy!6w{Gbm&%Z~O07AGn#l_LYDCqUsp}
zgVQ^8$@y#N^FPZ^r|<Z%82d9-Q7?b4h2$aesz1xr*#qiJ7i0U!@PS6_*DNyQ%wILI
z_8+EU=c2E3o}T(pDBj9-g)BR%$~HWT5N`d!=H@^MlyS7RrTAdAm47GZ@1H@KIA0dp
z*)fImRWdjS!EO)$HYmuwo-xj$h?W|pYJSqqa}2)O=ZLSD*kQ#E=DCVPKzq0KDeJQe
zp9^n(9LAO+!pogKfYENs!c)pY@Ok0!&V!~aSZW-U{;dy24Zerkvb^SCTZlMl<TH4I
z!;{LQbW5eF+&*Kn3yFAE8*&w>31{zx=RpciU{LnJoOJpT)=etTO7^c@3SSvsujbfP
zIa7%F@ugl_GgbpQ3_)PwYzUbycxL%!|27H^3LbE0SZa@Ahl0uIW6`Hi*I^F20v^lo
zHFKAlF)@nyXtrhPHs7Sr$r$q_@4>p5I4^Ox<c4q3l*hVL*5wR0m>2Wy*qz@0&*r?t
zhJeP0+PqD#e7k&OhuY&sK4rC6v~h^ToP9~mO&}Ny!=Ves28@H#wF(R@5WBr<7y%eH
zw*kxs{0_K4vtS2Ju|f6?1gx_}Uz+Z2zGucEr$s?S%Bd^z{?$u*js=*|P7#?~I^Xd3
z%gW5WD<i}7?%&~|Hr(;`bYSd_s0HAd6Y6@5ClcG44b%qMr14!VfR>Ex;o;gkIZ;jF
zQQkw5lgFx^76*QH!fhVoH>Y_@z5+9;tC9w&czgE0Q<SnYU)TX>gw!1*ivz*02M71m
z1>B_n7gJof47(UOR#y&xKTcap)_PWURbF6re420^hm*emOlpT=2AwhwYX@{5jfM2A
z8Nf%_+g*MNgd`;29swxRKes}-%2OmZ5rztvli5=7$P2M0Di3944m7=_%lP>>zj0}}
zzCS4+wsROJoiNK4k?gO0>&oM$z?eR3=^1peG|h^$$1|m$igIecN=)xhXZ~R8$5`hB
zLZj-qGT?+?s)PB9Hbh-312p;%9&5aK9&i%j*o2}Q5E@mWc3rOIq9g+Bqh0%DX*9=s
z$eb+s`O^~k&2TZ$zOIjN|M;Pr{8uPb3uWVV*lx^@bU*_xt^{Zlj?j^2^ml6u3(n|>
zNB5d{pm?-h|1dCkfQQrhAhJ@{u1|5YpErm!g~}+)TzXvh@xWg*lA3GeXajX)Q6Oag
zRZJ11>8?Ho5H3VccG}Ru0FFcsR>dmDp%Vs0KD~2>vsWQBSE{roRIwc32L45873qy|
z838n@UqKWWaNi;XG-+u@Tjy}e?IX{+s(Ml=eeYSDH?&3&Z-yz}+f05Y(nI!yVo#Jo
zl30b#`aw#J1iwh&pVDe+Vq8LO*TcC~KjSxck4TNnvheG#&`Um|Oyqji`%HpjwTpQ$
zR()E^KwlpjZ`9kv4$p?c3%_*Wq=`RpHR0ecsi|Q+_Jo-qSi!v{m|6R&Z7c!vHjvhQ
ze{i*VhB1q)4npS3K6eg&XlXy(slITf;T)9p?s!*c^(XfxL>O+q1<dF-?(udvcu*pY
zP$&&q4coBNR#Qr%Nwc)`+sF9%!lc9}>$r+%KQKM)p*Juvumsi%98L8cs~a2WEr@u8
zVI&HP%Z090P(k}Y276~$*KhwLA=4htfDW7E*%Bvb<F(yBdD%r5ayoSq#)tgEEpw06
z{?u5#e0j628J?xKZgOS`&R@6yMQaf!9Ug=d;b{Zn1E|^%Lfu_}^17mn%GLcBtMFnz
z;LNo&TD4H27siMaXK=cns3-13bP79Gg2fZZBXSu}H|ma7myT&40@=eQIlM>?`qqwm
zPlETl{Jk`SSEfE#H>>4CD+9$1;Wee?c*hSGUJYlf<dVN(-vc)qt_dzz8~(q*F<wI4
zxuFvioOKNPI=P$ZEkvgJ+Lzg_I!>M%=&zPGHoegr%=A!p!(Z8{?ebktHNjPrDb?;H
zdAHQ*gYRD4G?xyvQ?8sj|C)%pW=G%*dL2~9)E`8)@Ivi%)5)u@ein*j#md60vckgD
z_V#xE+dn_e`uyEnLVKS`R%!+IH8d(%02L93Am}P<^o=CuEY21s<oGH^SIyar0Y!BQ
z&Iz#5BwHMnrMADcS-tVR&b*D@(=MuNCvhNb>*ORQH1?v*p)uiP#M*(a6qoDbq5_x<
z1ArnyVTiceGuusT4i&kr4@ofJjC*Ey(BXEIa!08oBe?ePV=Ya?+Xk$-dt@-HO?}&1
z_wr8Ft+-Zs-K2vIzn)9lwR-smR^4&*j$zfrf1Or;$67TXlAX=F9o$KboW-GV6qS%5
zE(wt)S7Ca2*?g51)ld@{CVxcZsV6W#CVS*is3yT6xtf*UR878FJ~Aa9reG*})o!U-
zN_v5jd!*5aq`FtaEA?U`$+&UiCtdo;$fG?f9-2$ZHaE%0y_TcoDP^%4EnZqxU-cJP
z3Ll9>AKKU02VTIn4_h_pi1Yf#8w?C*9JlPxL#q}3F8brwoFBh|fATKKJe`ZWtKnRF
z6~myS?5(#n-d`4bdH<1r-Nrlv)V-<IBmYDd{IAY$+YXI;k+11lRAAYK1Ye9h%y^5C
zSUVY!HMrPdGq*gJM>UfyIASO*PxLHKT+ig{cd7;$H@>i98>jkkSlbr5_DETMY+X^f
z%%aFtbsi-xu8(w~RQ*zxKC#5b;0L0o$DA4+NBnxlvp(S$?@ow1<PFDlpY2|*v48mf
ztGzc&v%B8S7gEnVI?i^F9ahObcFFfesqJsx_%J?Sf7DO<S@E4f=BhpOU~j3NY5PG{
zBUKumJ)a1Iil{~CQK)sevsyPNmcP7X*tXG8ANX4e=hK>4>Ji=LDWTRiy`!?tvRiwE
zb0@A;zQ(y#Q4zRo7}oaC9Qh^eS=`wuOAj$j+%|uYz@NA);EZ=WclPLu<QqOYm+*;6
zzinWAK3TX#j&>gV?$-7Y6^gxYFu8DWyo_FtFPF0?@0O7-Z)HsPF#VO>B_p|HAfirQ
zMHEIf?sk2ejdhR2OE15{Bk|`Z(cOh}qzE~Ynz2>9R8#-njdIQ&H+(x$W@VZu^0WLU
z4)d0W7qd!~Egg$nXN){@$)fIdt*tC<$-UqV#)aFC({ycVh-#+UNkh-tV*I}*EFM#l
zy$5RQS6y%K`u)6eY27LJL+~yJFWW%(k*>EkwU^Ss-Bs}bj9uUq=pH;JS$n7K%NJ*r
zda5lO_;B#~;_SrrXDG&&@tzP4meFfD5i}WLk?+yu#7?`~_k@OXeG)BpnBFR^@D&Q@
zjXO_35AwQtGbMWnUpHMoZAmy<WK^iQ{4QwSsiKvPys4c4&Eh)LO3I^mRFUd9zi{Eu
zELlgN^1y;UY=7*^6m7!giRrVQbN6TSZw}5RTnr9RRWtAjjXZ}ft4IlazaP#&2;1U9
zF>xx(yjAN+y3+mbSb?f)QcB7E=D`ow0yN6*+ce|-1U)%{rlIKGS9k~CrOjR+UjcuD
z%6tOJ0j0`yRAot)ln74o!%A0zn%Q#fRqx3sBY9kWDVL5Hl*E%<jvsFKNO4NOIq2ud
zN{V@yI*r?c_r-bE_8LuHugXj*`cR-}#dE`d-8Rs}GXHRjMZ}$kCcId!N^({~T9da1
zJ7(wm4$kRSnGILI(>0%y84}^{%_G|z<r=1B<7tj;OD(%VC(7{O4Gz!ytgF%MT)F*~
z$DBk9kL*z;)^-}H9)%uf8vV!(<&Geuu-Hvnxp(&O?QTWWE#9lVY)ejxOC+^Ky%>D0
zVS>`q9RpqVyKPcFf#62=&8aU7HvujQ^YCxrJ2;HfYTZl;eb)Q6mH9>PgV&3<<sa!L
zZS4`W(enx3G@6gg@(AgXCrIU@+tj+l?)f~LxL6CjEOpj~PcHU{#Dr@%SFwEG)vc*O
zA(0;WUj*^*Q<)dU7=0wo4a|uyZlRN&JJ&chJ`UD}!X#5AelZf&`)=opTN<AV`BKpI
z*;A@l9?pV{Wx6lT?~t>_y>tD-LM^MM!k;5lGb>N<6>WQ7{md}*ny$SMb@H{V<rrI|
z*%PRDAs*$?e>MxgE}e&NGO0PwCYi@m@v=V7ezrcUHfdrem{{x}Onp#ReRJ3@|8U%z
zUsAq<h}7IHtALA$JVwg%L*ug+94|fE>BMRMiYu>p?#|lwUNzXX{Uq)2HSm4o5k~Dx
z&DW4bG=pVXyU?wwedlU=QfxP`ME!GW^wyTA?BiC;qp<z)U#>EV<obUqsn@;2<*i-O
z>8kW99(2|uUds8Gs!3JF|Idt0sOn-PM0J1VcE;RDjE(p|gY?}(k(Prn-F{xt)XLSK
zxGHKWIJZ-4^Q~yZ(I)5K2jU#`ip!OKwh4SKy(!plD-Jr7KB)}T^!wKHEA6r;(NWf*
zJyiI76ahwD@D*sySmL^u!>CdI%f3A8!<0CAlM!U_NzG`xdLB50zZiS!&`VokJVt;0
zeP~?ivO<g0{Z`>EO1CVoCumDOfo8m`?q!1e6r-kXt+RI>X36H}jAzej3U<}Tk3ywI
znTm)bI%O_AG=^HI9Hu_>u|zUmHS@Vn)E|7?lvvLA@@BB}%v4`}u?QzA4x1*09nBq6
zmnT~h*Y2l0IrzI2WlmjwYEz2g?F#YBea&ray_HHwqFS3^aQ>yF(sL3h0)Jfn=Azkg
zzA?>9F)CGl-8yI*NeMU%yWiY9^`s2TNdCnFW|7~ZurGygtNmk`$M{@c3u6f%nQXjA
z+_Ha{XewQCzGhDRQ_|^s6P>R41qA`OGDuYo%1WtwomM$3M3&U98owjAh}v?)TFX2m
zmbBVuP3Ce>b~wIN;2@Se(_8UX{%4g!oU<?Un1i|G7+q)Efj+|jA>8`)qQiP%ews{q
zPE#E3!|X5W(ur}eY+<{El4-Bai{&g27L22<noL=rygd%CDv5#$=C8eQZe^?lr0_^f
zme$rL$D8zs`L#m@<R5c8;DfE$KDDmrL^5kIOxpqk?3JH%Bu(Ix0^|iM#-uBFOSeo*
z?>MOTtOzp<H2tlN7|?pxHBHF7X{dEGU$4H@lYSTf;gUAbPEeu4(xr;jwv+;oANjwn
ziP+pkjGa=Qo4vif7@XN$aMzr|*RNb_h#H#_`!0p%pZWFqVmvBv!CMu>YNU7w!Ze6t
z5}shob+(CN3a2dNcP`%W@U?H=A+ixoTINc2Z6Q5-qTZldS)+9y3?fER7^pn96TS8g
z*-+bHx<m~;oikR{y62&%Bnq<w43h2X#oAJHZ{Rv~g7H!m>yd8z+vjCJa&EqS)yfe6
zNcDkjWQwnyu@&~$<8=9F)G7o3JLG`hJOGduK+A#f6Vi_-v>3EZjaY#yx7R*)D)e4}
z`KLg;-lZ}?@Fe>gkz-0+qO^nmtv4|MU}CS0Y(p2p{j291fkt%_C`Xk5LF<^eq#cGN
z0Njeo;eSct7iS=6>$=#w6Am1KkeNtIrO4OB&5Nn_Ri6XRGgg-|@3utT7<wMpFb6kM
z{~G()RXrhioqGb@0HixsEAPwx>AMXMJyRRC?Lu=d;08eRUOTvfkK;C3Y1F?7CN@Jm
z)_Kv?@NfZs=hl1>h6yT!dKVU4QUWN_5-8Ik8%!-I{JJAhhGv@0=UqA1v7%VcDaGfT
z+^=}Tf9}G1_7+uZoP1W3Zm2p?3{`^<^YSp;*&z8OPrAi%8r|QcC;!W6(Jd(|@<H`q
zIDb*8V4oQpre*!vw*IFsgqMVZkZr-Ykp0H%k3uE@OCl+*G?nnFx;iWOARXSAZ`U@U
zhYS4tD40kwD^CK1he($?w-}ev(far6f4&BY4|L;3x?a@jpy&FYbMJa0&nAKiO`wTO
zSOK8`N*&s1`B!pdQyjJA^Q^U&76pD26y{rPt;}yVSj8aY4>~W@RnO2E&G3Y>I-t5j
ztsgxi<Wz81*fD8o)8tz7AG?_k<};6)t97+#b|W3!W;(A~7l&Rp<*7!Oo-+6_dL-wM
zh|ZC|$6|Zbw*2dT>2$n4Nf6(|2iIN6?*@_q&TO>IgokAS^C2Vyx)>Na5TyrTh=qyC
z0}x9CLj`F5<g=U3Qp=R=%VV9PZ`{XcodY)_<2|ZN_j+LwXIdPGN3qg|OP|QBPRc~=
z7M$e@V_0dUe}i2Xljq?DJPWRh4FA(j<)8>g=hh+8X4b(*5gdOFhRG`}xLv&#u(5mk
zpy>eR*<e%CJ7o9nH5J+V-l*?ce6wih8w}0~Z%j=e&N>?qw_YDBcTfbDke>j6rHI-;
z#RT}F4H7U4E&<Km-6xIKvlzr`0m9P0)2U%t;o<y`1>Bh{r8@8opyK#$nZtarg9p0u
z^j(=&o$>Erpa2mEYM@_4)PxgP92~s7@|KsE>q2sC0QOfJ`1_Ac*!<;z<?*sl+AhQF
zoW#N1lzWo`1|boAYde1*Chd{>Pcye}FQGlyc3VqVX6#?IMuaKSdbUW=FjYN}p*7Df
z9tj<ndb5DG71y-iW!M06j|^>q@#^jl>@6TU80Az#m5CGJ;PhPL?@!RHz&<1GYun#<
z#@lgb!{0bbXq~R&dfeJzOg}Jc%W~XWe$MK=a|9o4@BYOD4C~}0@yOj|SlK?P8<=A2
zgIwU;v;YOrFlrYlYCW$#%mojK;vX1}Gn;<2DEvXj+nO2@Ep_K^axWK{MF3qI<KL=~
zce6{b3?iAqoSo^I<kNHBSWpa~b=7!<_z8&H_B64hzQWKW28hUOYn&kZeDz8W(;yu{
z|2TpbHNs#9PYBUtLf!;N2&ig6T&Gs9x6ZOcE{_OVR_WTxXV&xTHu)hLg(z!5jD?Sz
z+jEZjP_VJNx%p-B2xr1MGr#qvqsjZulo;dS34Ce}YT;hxEQ=de+*<sba`c$9;t0ML
z#%fhYqRCy_SoG%!CJ~CTX*)jZj-->*!z%}#v#e3GaUu_`tPY7M?RsH=2cG&Ms-y`E
zkS#A{Gk}CT_|>{6S3b2yYLM79GfLgvbXQy7!b$I~9q1heUB3fo8`yOjgGtZD06P39
zYiOoLMKWtUL^pv<FX{Z3GX&J}plaC$op;KY8Q(GBMzfa8XkoaOf1)j|GsZtxY^n=K
zK>)q8b0**(U45kVX||r8xj%pYBryZ=#nqtzrIM0feu3mhz1>tb4wDWl(bdk61_cEL
zuy$|Iq!kt-1C7g`$XNo*Gr*<l`}YWx-b<R1z#D;_oKp{6->&YUwWj4D*{uXdXe5`2
zYEo1012K|TlFWGC;o+@p5MhGN0E|xH%z6q_9Z&`#!b^B$rb9DdDXXLhVv0N>cpZVN
zyzx<b8YSm}qUCFkVc0NQAN#U9%dO{@_AX8fRZmft72rQ3#1tjZyIJ0-WLM&gE}#0$
zOD@f{`H3rXq`dB%n@M1ZzJWnd8%Q|Kg!usOBJ{X1_41|zaE7L<3;V$Gy5&Fp;Fjks
z>wCvSl8ZYB^7;R>7Mz>J>_UQ&!!jDGsABHy)9H<ZC{~?_UA6w8bR!SX!$$Mc3cZzC
zyE=Lf5}Mx|p|~v@;e62?8Pxpuy5^$r9X=_sApt2Ic+^d-*`|JQMhT6&_9Y`{taI^;
zJUsqt*g;30YtI@X9R*N3WZr}1%ZQ8j@9+Oa@KjJ-a!cqwW<&9$i2_OubRYbA5g^ok
z{74KS>^o9JqahG=BFFQ<q9n?1b9T=F{@+-YVIqQ37|jEq@QIfpXaC5^Hy|=9H?Tc<
z3R6RvgB!xV1ln4JxTdDYFYe#7Xy<lK;g|fg<2cqo47F}8%RN<19lv72*>xmWO(oY;
z#V_i?)HYw?Jwcw0f5=iAd-$uyn}fuEjwYzSa_RnML19ZJGQAG?`;3og#-OG7JO(ae
zM8o`h?#CpjL+cg^D0k_d7id<VXtS4J{^#$-tv3;*7*R~yczsKvX5-ixS<&98ejnE5
z<i`WWCdrdAj$py&gkg=sWU}#OTDV8Bwj(BwA(p7mK2Q67JVp=Pu<Cgvl*6^IRQ{`Z
zOcphpB0;ju%v0iJfX{A9oH9`SDfcx{2mxz1^Qne$Az)^0%08l#6`=OihOeI^gJnRl
zm><Oca|%tmRPJ+81B3p6`91q+xLaa6=cmfUAg2ZM5sV6u1TcmS<Dh^V7*7G>#vv{v
z#v3G&z>AQ5t+oFKF*_7A#a0w9RKv&&`CdX&^!9^LVuday>%%4Xs%hW^fMpS<tIWD7
z2p<zLf}RVhZxSeMInbY<kX%t3eEdz%O=3gL@&1?1EE}czllrUU5@B`~G-k&Z6j5B(
zNr8+n+Yw3X*A4J*u;tD_ILiDr5tA0)$3-*7vWtsreK&z`_UOaWDX2Fmb8bTQrH}H_
z)G)-bu3*k9EOl1QP_-A+^FHvQ8e(>Pm5%997(LW~O2>~aK5GyhGJX4W#pXGWn8ONA
z=<b%Z6Jt<wYN(J@*P_8qgGlsa>k`SkF!6#RafgY-!Tx@O&Y-IYKV_w$R?c(SvIwe5
z;2uJyem&q7OBBHbHp?!gR)(0%%Lf2)!JndzUV8%;Acdw~!ksl3tKs8K?}{w}iIWQl
zvN&O^g($nQGT+%B;NbY<h5`#g=D$0#mz6p=3V`<e^QSPlsRWJZAyTN+mvCmp@ASm$
zhJ>+;R{gt479jP`xqioltRa^5!qCI~x~|rz#6=&&o&b*E-_JgOxMRsX_?wsJY>I|3
z!gVP&Y$u}F(27-oghe<4r%j%YjDp3N=~WY9M3e0=1-hRBDH3=IsZ0r?&J=ZS@5k5I
zpFP)lbMs-ahPtsjYqML~L6W52`H8pr@{b44OLf0`ipWMt9*jBc8+hdToV{7gcH<@6
zVCQ1pnW`)g#>>JMn14wtgxf^8c!eW($B`GD$Qlj=-Z?+&;*Y7TODG~vBp}vdOC(Dq
z+)IIBck)<!L^4^&kK2t79%aOC^qy~@i(pAGT%K(AwrZ}E$zC;~JYm<>JKq*QIIl^c
zoLnffc~nuH2AT?C^NfS*MqQ(EJ>(=Lb31ZY%GVR9gf-kJ9;-@=iN%u}7`}Lc`7Qdi
zS;+^|hLAkThw3K#3-4-I{+ZkTF@h%`{daJUe-@%?*lgz?aU2lp=~zS@#2MdxQ*Apf
z=RFk3qF4--_{7A7=jsw{sU$j!I_L9m93MP`1fP<^TtVFH*s;gO-Q*0SBi#asxiVGx
zSqTo!K?lSnHci<8w?qlBT(26x>={y??D0(%cZg;zaW*_=%xBz|h$}Ky3Cm5VYBcih
zJFx$!LxNApi-%euoc&f^yHZbvDmQFfe{q>Z#}r4tqwPRfSU_fFSxYi=$jy{@FGn*6
zW^y$U`KR8|+>(&(GjKjeHXP-<;$5_^$ekkZ=<#PV@=VBShx*1VpW2U9tqu8umSW0w
z=glNp8rdc{XPdMGf__WlaDM`K5*QQkp<MyHW(!p1Tn0>7HTT=MXE)M*L!IO0U&77H
zyO1qtmJhkvy*mwKMr7)_toUT~ze4Fm93hXK#4PD><>0S&*N5)c3^NGr01`COgNO-}
z_ZgDNofCZ{ssqVMo@om1Zy}WC&ehC8$`hzZbh1gqpwB^gQ@gvxFtEG}*;}I)x7G7B
zkrYS>Zi7N`tS1_SWE+BpW4z2x9tgq}aAF}|Z@Q({7BK(D798NBcbvV%u;Yk#<3PQO
z?%MaS0bp*z6WDI@U9Y4uA@jGS!)TUPi{DqxqkDoX0mV*DvMU>I>)G}7KkUAoGGa07
z2-bIHldHs7;?4-@-^|95>oJDT^*=b)p#B5$N84U0@6gH~Mv`0PfJww4%;NIsok_Z^
zNF5n%^8n6KEB>&7xFk(yT-m=L#TRdLqSNi-8Kr^>qU_#&7<~Qzxc~~PZF=dX-5pgw
ze|`#QFX^Mxqj`m_O5ki9-4zWY?jjYi`S<kPsw-h)7L22kdlQ#AA4U>B?odJnOct4k
zgODYH9IHqVP8Ub08X!LF0lZ6bhXrL}-XLyGaaEfcb2Iqd1KiK{hLcCyM}225QwU~2
zFz4KPrPInMcott)tRsZW+`ps4>dq*$g88Gvr|$2Y2bmfd8E0@CAF|(zr}m3nb28MG
zA&vf&*l3$&@`~dT-W$dBD>mVn9d0@@r6Zx*SaHSMbPkCoFMm>Ji)FNV8U2Y;*A45E
zfH{uY?e=H>3`C|gKeeSlU0K5@hg(qjJLfp!$XjRPwVk`XaBhBKm0figr4if={y(26
zp)6o2BF;|m&5|VyJI~J=Pye0ZFE4M-*pmnxdhP8)Myh?6^aV?DXEMLn&`@AL@9dp=
zVmoK_k7*>8-{tzCh}&(3xL$0}wfEl)99@#kj)AuBMUUoJuyX_1LTucRoy3~!qzMX-
zh8RgxK&D+#z5jv}tqDs^SXlE_U6)ve;i(nc_grSdeU|e2frvxeCp!Gfw*RU=B?r+a
zawY!oX)}1$rZ{fWgAshfIj@78P(JNKtK7Xna`>Bkg<`V%?UHr@P|4Xfu7ASTVQ`9b
ziUq?aBmFYl>ENUK^=qZfe^W+h)6}^=qDWU9yU;BJzEf?SPYP|8ekH?_31FQMgHU1G
z^dY=JQGe9XR7vNqf`HYwplA`v!dubjZ5=OtHy87{qDCgH>NalALKoCdyvragQQGbz
zp6)VQr7)UG`sW{V<KMFWI4O?{kxMW(@1SeEQ8s%i2p?@#Cahw8K6O!k;<h++tS6I0
zRrIqf?wtRn1i3l3b27h=6-!Z%ZE;lCyPkgE>3y*--Ve8gq$Wq9L?7TaOSJ1iybPtU
z#s-k$*soBqJ8IpD(Wuon%l}fKI~k|ASB-;O$n6$1_5u@~yYR?fnR$!U_u|wG&qs&r
z+^n|+Zo_aaC)IHLl@F$Gyqp=h4NHT+#Fu#$?+_&XmZ$!qBwmsgqklPK_j=dE&yk|s
zrX%fDfu!FwbLPW#@`W^cRYUe={qRLbn9Vf1igYZ}7Z~5uKXhb&o)bDB#Hem|*f%_^
z^~x2Ggv^RU*)u$z37?uKNN9E8?rlK}F5dXqn8~%2E&jWLl?#%}K{Klo!>w-*s+6&)
zcWi`3!x|*6zC7>1n{l_LP}Eo!sts^5o5%rcJ~#{AzIPreD01OhV;3`oo9gBX4@FIa
zJaX~awJko4bVW5!^C_5B;pu0IlX!P=3Eg|}(zK%fEqm(A8i3*WB<^)WcXpyvnkQpc
zo?nFjiDu!(97&Vj^vt<)HFs&-dYsJS)C@%WB92Qy6VH|Z`BWvlNcr2OGtN&RalXv!
z2LIi^At<qMpZl(dg1hGg*FBMp_fxaxQKT^Q-RIzv#|jt(YaUKT<Cm{Qmpn*s;D`^0
z`)<7H6Fn1?7Yq9>*#0(i`KR{sH*Ie`d2@dg`gc=Pk@o}Frkaeon4VtA4-l!_ELy6g
zulQ_tl3*qCu_rvF&&&LJgJj9Dt5C?ISAM~T$hBK1FF77BPMtf-MCzAM<ZpmeB_ieM
zk=U5q04eoCF;S!E#fJ2O2dU!__V98?441gex*n&!z|AA5+RieZH5B6%9YI+2$n=xK
z0)vBZp1lvrWOo?KO!>tUWiH)$(n+ZL=j111{35+IVY&Add{ntAv2OC<xZhEE>8uFc
zvKWQ(F~Y8gO5If%q|xP%uwrDuM9q(kK|CE|2?0M7b0HxeTR}?=E`r?+!ho}A+60rk
z_`DmZR%831qJ-FHrT{9fWyp7}x_rq+ug3G}M^Ca$@=EUkrv3Uu<0xXfY2fR}Xc+77
z!hoKQGhAtu#zGB9U2yPcT;`)AcQ6DT#;v0VEcAkcf*eow-|APn78Z|f!EemCOe-O_
znB4ZN-E7<%?V*89QDux_7>eQ2a-43!MfF3@9HZk3pf@75M^CDc7QTS1PTR{~DIS^Q
zB^D%G5*i=3?;oG4bH6Fjf-vj0sn==ua=;?$(1O8!PA)NHC*t#fSMD1;azz0HyI0jE
z20EGE3X(XT%_{8&VqXhqjVB`9Y!>_yRhVagp1<I|va6?sp~g;0!(QR`i?18D{QglL
z70ovKBkBj&rG9_A{Juo<q1jDLIZ+T+`l!JDw-24-<_YtYdbM-zg}dnkak9|3Rf$NS
zGEn+>DUHMg<fMO|0>MO(j{LqK_oo)%lOZ>>Uvvm_Ce0XLQnQumLP~H<Dx9iFHROF-
zw!b!-m7mY6UvAG<uuv#>oQQ-G12uZ*0<`2vWFr(CPE&P${Z(Q@%7PUW69)Q~&QFkx
zONhYTe#S`WH#!c)*l?M1ZCF7-!0V<n)q@wK`;QgEi8Wzzj*w7*ee^{!03>rrO_56+
z$zYo@2RshM%MKnU5OQLZ-5~pM3?FXMVc>*0!pZ>hi20?T0)o9%PSJg#C!gpf;<li+
z{?uc;@ZXIZw8EMTKfC44ROzVdsA-H}WjM)5&FyT*4ps)1r}t{C>x3IkYx>@n=-U~g
zCl#?89WhQH)yY;k?y)EANEBcBOEhVbIAZaK>JqBaVZy)ov#;#N6_<_F&8p{dF^?=&
z679U(L-IAbY;bhJa24Y2fkr;hZv=`xu$3MLB4m*waEIHCy5Sf%Fjp3S-E2g##(5R4
zcZ7jkS0|f9)I@q2u(WNWFJC#7x(0gpz2NhS=Y*Jc&k6&jP$>rEM+gBBjGbcT=ok+y
z=7l-_sF(AeC+!3RQ$3m$l-cGt1*6g~Lpr9UEkCee-K+E84CkBPq@|%j!o`7Imj?(E
zVj29ijekI*{jfC!!2I*7pb>>~0Plcti|~rrTFn=S1_sjn{QSB^$9;xG!gl-LuenOu
zPB&yf9MWBs&e(Rh%;br?m0Eb>&>ICKb4{@J=E9J;PEa&esC-&E2i?FhxbYy;Ey!73
zSOkD&Q7o^ZYWoB1FQvWjHA9x6CCgBI4#If@wNA7~KZtt~M*rJX0K=eS;BgpamGXAR
z5y_-43no^5V{&6*V2&w6A1lI|a)-X_C~{x-wXoX4@F@v#a=Ao6IzA2w0UYVTZyp97
zr}cOR8%m74RRsvEutTGPcExsYqyfBcvh&R-$<kKG$E5{DvAY59C3{Y*65&_gIBRwp
ze(j&i-mb2>wIK~9sBGQ6nX30-f0N}Q#Lu?Rq~L3?k`^wlVQ6S*Ak_0LAh{OY{vOBj
zTG-jCBQ-0)`c7+q@4@H0K%r(y12}mpxc=BPeHPQa)6K%agu?_n<(?iD0D4u1^YuHG
zBd}$F14df~!nkeUVEM(qzCMx(JsuOUEnazy<K|TTs@gdI?<d)wqMzqG_ysdjlC^z5
zULE_&b`tojN3678qmgQJUFf0(5CXK%D@bIjo?h3^`}Q_^9UZ>B9hG|R1)%MIPD&br
zlW(xEZ=^if_xg$y(AiiN!@dDvU?X=61@9Z-EP=_Qiq50??$>cdU%S4U^OBU(e%+C%
ziVZW%GS#}oDz6^YNSR5SnJwWEc^;A~P(LKM|KBAUA7EJ(yxp&!NUh``@;cpWW#fkh
znhdH2L?8)WE?j;6&jO*rx|-4#H%5k51cXg&7l5lCx7#j`N}M*gjcmgopOM%ir-zOX
zXP)g#gW(`(4}|I@Ld>i?f3Upndo3jh?4|hl_p#Y-$6<efx@|L|>;eoRG&z%|I%S33
ztb((wiSp`>JemwLpp@X-JtXth2NMGl8&y#&g1|k7$pN71%XNO(8Ik*7`~&2~IsHL+
zAQAEDrkm#{8?btiq-0G)!(`Y4S)~*1Ih5VMw2i$owm+Mz8W*NTM(->+ZmwTj*b=Ri
z^-|%qFg4AHR@#o{8<m9xVTOj~%p<@lK|wa><l?ddzX(3|p1$j&RBuCHu1&%@tK)-2
z5&I#YwEzD77se!W{y=oq#g-ymm>%ETb@Rqfy?{$+f=L&{M*1p|uZgyj-;KB#)~&b!
zBk7(xibm$!t>d&O`)i#YH}B!pnD;V1zY{trG$GNKx&$7IK>I7UnUt<R{+E-%DF$f1
z4kry>WuusYmgSH8DB5DS`R~6!)MZmaEj*!l4xiKVzrh^$`4}yw0BkE6k@PaR9kU+Q
z34rxf#ewkFqkOZb)4~@{1jOYP0q;Eab1h3gBbp89dEbC<C=)sJJ(vI7;M`SN0Jwxf
z+EbK|_MYqYhmi+~^dOT3)Op(hc(n04Ng}p;=P3a44a0?s%nVk4!HR!YFzmFu)OTuy
zebN0zWba>2!Gbdwfye@1(oKz9T&pwcX@g{?|I=U^=0l(Cs+tj5)9MuC*rc(WqWNt|
zTb|_LzB*D!-(&wLlQymXw`@D*ySI%_((o^4mGCr?y$@%tbVp+^BRzfBp2_;}9|$^b
zByIXy>;<m?$?Z$We{hcCY)sAt#pBI6p)AQ*_>W*iw6V#$OVySrAkczbk#7dP)!AI_
z`(%<g@){)<J6fhmA5>SzRez1zhU8|Osx1XhnMdJWgg-O}X2lE@DCT-$cW?iAFg$65
z4&8+#4M;|VAjPz3id)8CkXzZSSf8v-4sFL)2T=%{4I_T8Xu~pUxNkHnRNP9+c1qR&
z#e7c~r`&zd(tykYzYJE@Do%l3Ibu1;AnwRfbYI2+Y)|0WP|7Z3GE+jat0p}w`8$%>
zK1_0VC=00IhzCQ#-UdX@Ls&hK-7_2(dt$aH;ZnG}_ih!VPQ*~2NU--GXeLJ|&ws))
z%!lg}>4QNn?e=@p7nN9;kjS-2U<Xbz@L+aw@fO+J?aPgVPJoy0_%4xf=NzCaxAHvv
z*P@{x#AG_7f6nW9;oMQVC?wT2NT^x-iLrAqsTjQ2((lE>RKChil_)VNhz&C}ogEi5
zdCfSu@C=_TgDQquRjGSkfBhYldCL$DjFfhGe<LHWK<+p=otb<e9T{mutBWa1z{>l-
z^k(KnF^5bX+pd6q?JlzM>x-OBGTsRe@@9c|L*lA_l6V?)0q-(>VVh?|Z~Z9~j|g3n
zPGj&jc2*Qpd}KV1RZ>B>C>fWgmah*N81|-qI@yML_P=2|c3$z}m2OY4)v=!(z5K8d
zF?R>GIN~1&CyAN4X(MpEdjlk)WQ2EqJQYCh2*>>3iWA(pE66DW)!?(7!QE*z(|@4U
z$Sa-D0C9=*>83pL1KabRpfzukFpDbD4(mf*qhD^yE-+JqE$Wt?d;gES%FTO+JMqQj
z3T$|qwY?-UZW=|}Sw@olibKs$Zh!dt3-*!YLRWllZGO~&nkgKV*6e@M-*R20$54^w
z3jVhvblON6BQsmXaS-=n6b1Fka!>kK{sXpb9KnRlml<2tcvxJdJ|rACzo8sBKR+k&
zL`<7m1_7LL`_~6+WV;<YlJ)g<$qFkzUereIg?EzlgI^&^vl+p#MiI+$i|^SQ7Rv6@
zj!SBq%0i~YaAU~By88`YVB6LYL0dBfRPCL1dw5TO2)azbP)*;S(ToqCEYCc4?fG88
z;&VRU+zX&fLSbG2h^dwD<nd6cl1dxlnU;auLk(l4@D<zB!!3<6Gh0xZbWC|eaqZ-E
z&qd(Q2RX1uFmq-F5D7VhCR}Gpp*mNG+d|yuylTSx1S!_l;b$@@+!h&N^YM1307sLl
z;J&e0kBWv_tD~}AesW@5V!`Be-`f=q2Gxvynl-1-enku<M9Ptr4~=DQ;L1=8qi<Ia
zNbk2-`r$2PH$3X}bKR8+1VMvfJ0)$vod4(N8Le9NiI+O1j=5O!a}Behp+cH=fs3~h
zChvZAEI-(s71F<zBAvJTfnv1H=HXX!)Ax$FEL5ce{cTH;>D$xF31;72HVi<8Oh1$r
zZO_}8xUwunmRI$$bTQg3D=Jze$v{X@5VNvIw)1Ej&I_}b(38Mz{tQIl6E?jo_HRl)
z!=Y)vBnqYPj>&|qyhgB}n-j{%rxqC8PDl~3vez%Z4$l~*x(n`ZWTr|eG#Nw3wZU@B
zcOCi#!#g-{u|H-Ru8voh;<N#TRIV!3vD*F@x$~+g;;x!B0=sn>%7mR=Dmsug7S!w{
zZY-4y8+5Y#Iam43f|c-;r{sl$T&YXbf<_CFxHVXff29_ESB~4+o{ETPrlnlGEy#}g
z^lvtxy(Vp40}bO`nDgj-kOS0UJF^#d4m>c=d=dOT4NLlKf>Kh6_@h-yNcgqAy`aS5
z(=3&keOlWKxI7o<Q;`kIGP&n2Mb2?GoksWzYgr;5s-v+vn<?|rY0{$*Q3kHRV*?%e
zmmXaY3g{_`UvNlD+As<W58}wi$=4KJj0T6@^|A7_zC7a><6>Yf)+MG`+sQbbvUvI_
z^zc`zJgh5NtuR|329>2lQs-WDv`Y^yEWNj4@K;*QO*jech(h>z{z+Tgd(e7-ZQPgM
zFgvo%wn&!$hF73Pu`^SWzUu7mE>8CZmI;_QJkAaQMzZ8{aINC8aYjt}s{U(v`+ci+
ztA!2-<iA>(d=eM}<XyAtMXQvfa=WK@;CkwDa^qbB^u^+CtwDh~f9|t86}Hp<h@lPe
zuE87vbOK!Hec_uz_YP`LM)45Sq((IPqUR#0rV+I8kFXB&Q{t}b3NPnJ-ssWS`_J-(
zg+Gr&jRAcgXb_UasdzuMY~6r+<T*T47`{DrP`23gHiN&J&ZbJe{|q2l<P|w})}IEj
z=w!35{V7t6<<-BFVfAmXf>*!d8PPDI{cT3QGsIs6?*7;W|KIP)xJ7~*xobogERVjB
zz>z3dndx9>xOSJk{LLjncj_X={KWZwj)i@iudQwx=gGwrL=Mo$t&WRJ&`CJzwoMdf
zXk^c27z@+OH0Cx9pBx`U3$Rj9n-u@?cf#*=D$$4Pa~@&>dX^C1r%Ug35H_8+eHS{z
zBT=nRzKj;4Yxz%Q+;?B22&L|Hb3VrNmKokolLYdtuoc-=n&P}Eyzpp@^vR#zg=J2>
zyfnpdAf^iyJ-IzCh2OvJGX9nZIZ`Z0?K3p`Kxz2p1U--d&O*CTf|!*9c^=Z()!;m^
zp#5Y|;m`7Wot!$$h~}~mBiK@MF??9?;AUEhAa^jXZTZ$g7M3>QY*38bp=kL-i|pH>
z4*jW}{;j1lX|ketX`_)&g3w`<%Bn1Fs?G+XttVcib4P62ob{KB$^SKyy;G-2)xM#g
zPT)kyCY3F-`1H@m$K8Qz8b#Pu$1MyL%Vm=r%?G<J^xUZIa<@A>eCuj(3&*_o<;gCt
zXz1OO5++*|Ka<R4YP0Gpo2`;4dPw(aw0pB<qa^ZCcU4q<sFMQ&NdmWL&`JCDR3>$o
z)V5HE$nA^q-mz~DoOs!iPrCNLz@~>l2130LU)wrW7<!uTV;|gyxiw<?B^qH`UcLGz
zXdQ%w&WxAlZr^~i>JRhkWNqzTjz#Gi*<@2?+kKa|PaJRlNhiPkZalK`lb`&tu7gim
zUT)M$04I)5^4?dnYlq)1P9>xpf1=QjE@de@zc6t1Wfe;I!`#!^;-FD+i|9}cXh}4`
zb;&j&l`7)TVI0e}p72m8OOLQ{c7pj=cJp&qaesnSO#+qN%*9fBhrbb~vNui39yqH^
z?VMH-sCAVKc!pX{P8NpnTpJp7pzdB;)z${X{Zp%;k9a~hdNVE$FO^syIVyGb3&Ni!
z?Ja{(@ajwar2QnVY~p_EtgVmAlPB|}B4fjF<N{UysXC=Q=zAZF%PuY+j}Y>Ef<p=6
zi>_hyR(<@@@OfEtQ+A5P)Zb(GIsG$N78Z{-4)sd;#C7(+JnriJOJ;l=U_rRjt<srK
z)129d>LA>RW1luz`POzWO>1y>BJLzbg!jYO;=Y{8#m)8==Oan1H2h0?_%dkJ8^cBE
zuxFa3tvPNaAsRvjuY^%k!KuNqPd`6tQuo0M4t-&%tX;}{HJzJX$gDF$i2Pf+v+H@6
z*MgYWxy3!z!Lvu^3vbp{iv&=A;#=k<pYp7~A=^GW>PdUyZ7qR+DgIwdfU%FpZR^-r
z(VD#>T^*)4iCz0{(2AXmRWwIG?HV|;$q)T$ZEKL0uEDe=vSC!K;r&zK#|s7FpS0Jp
z_$GT3EKd!Tbs}_*eFSy>5!2QtmkgYy6=VA<bDn7`EgLB<%j=!YNB1X2_j~BdBsZGZ
z#%WX}{lB)pI;_d)Zyy8!l@93=kdy|IMoI<}Ly?vkrF3@+C>^6DMhpQ3q@+`%W0W8g
z6OfV^-SInozwhsQ|9N-qpKZ^zJ=@N6&gb0s{kd;wV=DAa$N0Lk?Cq2;<n^oz(M!_@
ze~D|!$k>fy4J!-ocp~R;N&QTq{=BuZk$LYQD_4*UM}8L!Fa3!DNn}3kse~Vb7e}(p
zLx=|NI^=TvQmXwM5^_T8_6d(rL|`f;jOIf2m5eraA2DAMpeLl0P!%6>ZVNe4T5Sb7
z2lRckm51YJ#4vt|9SeX&T8tzkfs-y+g`4+!Bvn?v9~p@gUh?6rT3O!hB}D1wT?~Ti
zM4G!56gXgs<jAdJ(q18BUtA<OF0)cnC1reopgxYQ;Y&x<Xk?J@HPN!3RQK-C7)-lV
z1O+hf2BTfzq-@kxZw8hPX#Q$G-9)DzVAwAm>#x2>=bxlCCqNsy*6$^mrn$cuV=jUZ
zI|f_;7CQoIrxuh0Ju}pyOhX*GJ{}2Ou`RL-q<j~?I=s?bhlw@)fq5@wE%No*(y|DF
zqaMFzy|rdngytXDT^ZtKT7IF4_fOcr?pxdTVj#;qC=6$M7isTSe1BJM0yZ74gLc+T
zw~jAwWl<YV{hVoC3z&GbTa13D2=AiB-O&R(ab^qa4`Fez_3h5Y#6)r|k-_Ejb=&l?
zrHJmF$Txp>&%AMt0CHk#<n+hdbVhP@-)<0ZY4$s?PX%TA8o(S*163Js$m_&fSAx`K
zun^v88v#OigM%1v0|<LvoBtnsyr0K4A5cVsTFa^BP%q99OxLEhc8t-%UCz^6QOJk?
zB+eiI<10wBx!qeN+?&sJr8Krlj5zjDd;KWNqJ=9eGpStttfLIEp{z5@6=k6~_Ns1s
zPKR|Pij^@WLFI?E@xi}AT~dj)#7q9_r@jMvX3AZExJ(HXi5SStPNlcs_En&meqNtv
zSi=MF*VPwc?>FIv^r~<>MFijpYJIW09gO>Apu@rJI$dXct<TyxQtz~?7zgS`-edsz
zG7*A6*}lV~MpL`_mK_L8evr4)>m&;l5yG*vvDq&s5h)^(yU+%h4sZD81j(wpj9d~X
z83~+Tn{iSOn*q@go0GBniK*V(sP!H-{~Exd8-4aaf%O$wv4MWn5+Gy*G@*p5To#$R
zNu|sV16v*Nco27jeQ(WcJwJE<n=0ZFBCd+3mQmq^pkgS*@Z+D}<1QqDvf-lJJhsx2
z@y@7&$!}Zbb$r}~i09k^Y=s9zVazR&iX0_&aJ{2+)hH&keMqkSQtZN~lo`xQb+LUQ
zwG4FzlOm<CP?b?wtRgqgo59_Rp%+-fa#JZ336~lr4bh@$C^GgZRgQ{+-t>Zx@P)mI
zn+01I&Su{C0^!5j)}?gBJHmw$deOW5dHbOA9D%S>9Y}~Xr#lamw+eonVqvC_Irgn;
zvr=jQ#y8#O@<Q^82Fvp51dN%&09BJ~Fx;TF7+)Ynv$e2zuxf8<^eB0HdYWO6S;AQ!
zz(ftL0CK)TQ7%EYY&k$Kx8|B27a^dK8<ZJvFY5Ph8++A_kH6>pk}gshOHM)4@nbCk
zRYZhA`Wpsc5UE)v&8NC+jeQO1U#;O&*6}alit>ui_fE@WoPs;fz3P7{G`RcB_JHN!
zBW{&pS*ll(3o9FYg%U=cjkV&eDKL#Ttj!5S=f+d^)OXmyl9-<-2BG|E@oD_#QKjic
zmaYyVOQ0-mX7m(svd@!9$U}&-NA`s&iKNfLZ*=v(@Q6syT8BVrN%qH%;ZwQ(oGKV`
z#&Gy%|K0(5cC$lXEwVQuf!K%#p<tPvr|1==ujhV?fhKy-3O(Kj-Ing2+}cI@;=ob;
z8DNQCn|ykQm?cpR65B=Lcf37x{k638@4>r6E6_XdUMo{TD#;tlrf98V49|Em934o$
z&uP6`<IfY!7qzIML-{SA>t9l!ZY-%`()KgS5Tke}lbP)^-Fs4q&EPzul%CK!3;l+A
z))J{Owq`v|Uh3FzM71VUBCIFzgUf`cx?`a<Z3=bOqi_&8H!)lIK6`8Zxt=L6{(hn(
z7@?^}u@QZ~p&<%c!w;)qnqvR{z1gN@YN^BS!W0eq`ugt4D_<~8b4+s`%0oXqY2^1$
z!|oCG5PkYk>yqUQQp9S`t^&`YpI34yef8lT!9Pv$zk(*8*JVm#9D38q^Dd;lEl!&l
z0q!*p_L%ApwK*eXXW1IzC)Hr|WW7$MGrt0SfO=dJ)NZqB{n*iKjP=fK1L?!a%jfbO
zANE$Wc>eBD53}e(zWzZY68@5H0}r&PkP^;9cgPs6+UvA~)udr0T(c0~_>-ph^tVVe
z=FUHj$>lh)Uq60GhZ|8U5Brb}p4f*{5{;u`PalO!9J|EaarTth=g9hGe%t0vJ7j_C
zVJphv4?iCs$BO>dOS=EB1t6duHp)3@$HT5m#&Ss5m6Cb3;(6lZk67-+j*uYgx$L6F
zA>1Ie*ruME((E`sJ$IDq6q*|Hk&@%wCjNT_Y(3?KMmnDb=T$1&AJLv?xIP?#Qa>^Z
zT?cLt*{|@0z9~#P7&M?wY(&I}cMKdEI*!jgj?xZ?8ae>}y2`jajlb>1PhpjP(;<vT
ze+8SuH^8R!riOk-@;V3vj0{_jm2bK0*XIUr6$Oq|QaInK(uwCnY4J#9tn32?=Bu#=
z_?SCn-G=@04P<==%r@+oVrtjkU&zpj2YyNX>SxSJfq~I>dK{ObNIAH>@)*WpuBm|B
zKsQVN_vq-Z(%2Mh%)YHlHC9ksX0jx#lI1e<5B)y<AhQc?YNx-*7aEJ&e-n><R>}Hr
z*6p@25B8<XIwTpziE$9Gm@Za%^>Tj**DdyGCR$2`mv74BglwIa$)~9>oL{O-LAA>I
zT2<8zpW$CH<BioBx9HKQ9~z){9@1l*5CS~*Rsu+(^Lr!R5(^WD^aHcd{Z{WB@n@`6
zx@S!b$D|?O8|127wAL4%xwt<iM9nb|#;XriJNeH^E_<oRI{#{F_QNj#a}j8x1Pc~-
zq5sLpuTQ|9MKrFW9-z&sKLL^{225^8%a+9aN9F+Ma3k6H+e-%ix3s-c_asN0amvjw
z?^wd*@$todt^wPgnUo^UeeS5m|0ta}XRkqWZDH3T5NzX1eSXQMqt7H?p(BLL%KNuj
zhOW3y)RG3U0ePck*;PblIhqSL1`BH*b-z=IhNJkz6vl%nNbjhW7IZk-*_DXC>}9!0
z@&hUm*z^zQDYVze4)e&4)KgA@j2QXVMj8f(1Jj4Al5ROg8{e%0u7*Bf@E#!wp_IyY
z;|83kkBYn{b2<loTXO1h*(8)a|9CZ!*b9PkKuYA9u+1%rI^87}YQ;d!Eti?5#>T2y
zX|?(GcK6`yw)4Suo3LX_X5tmncHm%JZrqTxApt;tu$TcY+^WcX|1so-`DNX<-j`UD
zJm5b9XKZ!*{<Zb&bNTDGa)3(8R4P8%>OJ;*jRNhaD|V`0+p~RNB5t;*we$s&ZDB`>
z(~a2^$||N4HXETc_2YZr1;pI*(yZTW6cpv_0rVT6(1+;rU`k>~;^KIMEooxyIRi~x
zt>i`3En!vL+fl4`{>Oj*{GrIH0_>&V(WVZNg|2gr!J&55cS@Rh6GLLR9_W4>&wKiD
zx;_qYZ9Y)xPRl8&2qWVCrQ8fE@BapL(3TSYf;n6bb%+FQOS;L3XNN1A4!pZuftI}3
zJ8~JmXIMM<B-5md2~U9TZHp<2n>6g50=#PPn82AoHPd6D@jK~!&$pxy1G#202p}%3
zdjRgx=)MlMt}d^p{p=Me+<wD0|IjQmx={&<NIqrm83p_(;7u4BfWjoNCIXafw(jmV
zqV;{E6uSD*uwCHLVxwJiDIB2K^x$N)wm7XOF&l1zlb|PGEgTWxSVQzqeVvctR+2(b
zwv(sx$<9J)4U898h&Pn;G)YWftxUh1!p$3(m^em%eTZuj*G@RKHwN@Ys5foZGdVVK
zI_lmBXxnshvO;}MYtP-)0KSGoDuu{?MwH$|Gs6a82cJcx51k*4U@aMc=>Mt^+UMs$
zuJPoq(U+f9Zht38R-Vqi2F-9cV7n2Dc5qXgkD`40y^rS6<MUyJOGQ_AA1xNtr-YYL
zm=-$0DoIwdo=*7pYG~r!RIy0-Uu8T(%Z4|R)i`UVr;Zz>AKqoB&uGb1#*xlu;KUwQ
zwu;<(5GF>fpX!fy=ULYDZVlkQlfG9zK7R5DHy>Dl46gAB^5)BAOh_=N|FI~j2SB>)
zOn4q$C!So|odEhzonEWHG*^iRRh=2lHmyR!dQ&|tQ{eXo_&VT(;KTI~2_#d@JbMLy
z628lQ!?9;JvbX1o7a@`Gx7MJ~$B`HTOlaKR^NL}2Zj^5^7~4ctFbrDU4Hv;Dl#C(H
z>JdHPYiVUlzDSiij2(x~YE=#eM#Q#(;$rO1Re2q+W$O?zl1jUK<Tr6vfK|eNV3u*A
zWrKd|Fm<EM;IvCZuRLazcBdIS;IY5|NOi*mC>YhK%L-Z%OSrIp_9sl<!t1{rAM>sl
zElERblg)<Qe}auB)-yi>oP<s--aN(Ik&kf`0c<qe2a_UPx;S0xi5EGc`?Kj%D!u*U
z8jVIHq~^O9<1DrZx~;<VUG)u!-sth6F~qSAkg%kT(2VL0v(9+7in8>21r&sbo5YvY
zyUcsOA!{KyC0XUKos{*L!On&&Pr*RqR~sWcgOo%}Oo`tpoE?8RcCdL8y}_#mbw3cb
zrsAdg&$<5b7-#^suOf<&M+Yy4hoQD@e^|<gF3#g4-w67UtZmzC#VGqetEDW^w=t1g
z!a;ig$b={kI6q2>!xJy2EhY-Y72b%7CpFj14s)s{k!0Za@sJa=`XkM={~@(WSI50Z
z)-_ae?I=OPNGpBAi!ptLWgks;Bt1TdDyiS+BjhcenDvTlAdH{z2}H!^+)DlUv~KG9
zWGVi}8U?7I!I`d%I{EVXHGv0eLsrqUtQFvd3r&e>2Y=UpiB93sbgw;Tr*N0LGzkkv
zuO3OHsaRRm+l`2G>GR_*0x+F?<C`mpf=LtC*zqDHkd-O|VmFfZgi#%zfcTJUhlA=Z
zI1Ravq$Z3JTv4Y=ijm2KLoUj9J0Up1X;_AOcrqKe@eeNeM3kjke!VVx*r?t=97=%+
zh|eJP%BT~`vt-Ntt@fjD)bSh$1cvW4UTsrHTxq<l7}l84mz}iPU_r6*M0FQ8?=NMY
zV3<|BV~~Rh4r!oo2mm!;hp|GjP>ZI%x3PYmQE=c@cOLkqYJ&#(VqP@ON5#)4GF~IX
zNT`$m_Wr;cKwIO3gBpD}4b?KATR_ViG`%nG4_5{Dcf1wfa~1)5kc<`&G?}35Asycw
z*x>+TdvAJTp41xz1|UL!`DNM7gcS-L2?3S#xNe$lBJs9<YHDN$I(SGi!y<igHuavW
zm1X(+2I1{iVB2kM@e1mu{ekWGppaVsXs}kiXuaViUKkwesXch=0wKkZm0%nU*lu*Q
zD?A_1<VZHJqJep{P|r|yD&o|JOrpK&`?Ex1rT!IPuduKdFa=p3Ikr?@AXi*k0;zSS
zhfzI&=ERw-br^?~>bP|Zv=;rcsCbp>G3T<a-@^jY+_OM__URi8u20jo=ZAj;_Gr}b
z1zWftrLwgjq12E#W48!~GbO~#+?>_Vo7i12=*NeeR?RS1;&H4dK)v3u$F(QMRFM{~
z*MR9grR#*@;3-<iW1SJ;RgtVX8V3pMuTak6Ta$>s5GEP$22_Sz+HfAgx?Nl<-BDmq
zoZ1qNSr!{WsaSGa<5ZU#5#=ZN2wk|k7^O#4`rjePeYygBBp1h*X)<k~(rz}4+FN!~
zQQr*=!}a(*<((aDnQb1VbXt)!Dl|r1^0*o8vD%(+{$S<SWi!p5>o;nQ#|+VqzbTy8
zBN<1+MbL)RY6=TUl|badsm-cZ^2p4Wg){!q58!W-MDFtWeJ3=MGI7CKmY#^aN;6`D
z;iu;u?f3lvJzqzrH{sPnBB$HVnK*=j{3%_f9mjsFZ|IP2fy4JQF1ru90fL>%u}Wn4
zVUBt%Ik~feFXmKCO71w)3dA0MjMagvsW>k6_eX&^FIV7zJIu0xO+by4f$powWaSPL
zd8xQ+(_q9u!Cf~ubs12_T00lCKRw!;VR_tA_PvjQZ&P(1btdw%LolUQ?ee!{U0{`K
zTV}hapx3S!fv<H_vx))KG>|{!(g4pE38W#oio6u~S~Ru#d5pwDJxiLvG<jz7d}otz
zY`54Rxm#JCZ)4;AoC8^8MJy&8f5`4<`F>H{_c{MG5yFN8DNO$Nf|xEDvG8FmG0|W>
zEj1M~E?~jcCC#&Op%xP&Z1%Y@W%Ja8P9k<PyjxSy)tvqpr%oI{13T-l2F5t~(pUZh
zZcdetU<ngX&{{=`OEznov=woqE(VT?Ayso^6@0w^xE*)`<B3C$D=*6q<nI3qQ|A?y
zKWFo-@wNU^#fxWW=yTUM3{VK<8!Tqs+*o>eoYQJejQK*i6^Wx#O(P-(LcJ_BaldHR
z@{F)?70Dy$9ivqp`h@3pVXzm^=)yL7U$WxTiwN{b<2_Y{O=m2hO1?@<a3PmIX|@$D
zrQWyHV0JcT3>odhQPH4yHe{z@-Ah=KA$8PpwS|zdo538cHE#5@T9wBgetp4H$nhT6
zJaB5Ea-;uPwa9EKJ~158DI0?ua!KZqWFFw2kXTm5{41Vt$%pw+Kz1CVo3jmss?q2U
z1(u>&!=FK%R<8t~g=XKT_ByCkwJFvuf!B=}Rr-C#&QKgosTXN5m0F4YWlqRU_&p0%
z-`Z*_Ins+^&txtB234U?oh}Nh9;bbJIz4+ldYlmOY`MW2b}Q9JM|AvdWoy~5Y(!s2
zyk1OVT*&2?Z2aKhmQA|#lQRqrU3HF-u$k<rtNG;x<7D(8;#4RIK^U=BmDl6f_uEp!
zNrt1EZLI>m__>UxDv9hY1FjPjV<B@fFHtpk#i<*O&?&!aWo+F_Po!e3JI_&yuq4y6
ztQCs-rRK9h%T5$K7_CdtJA2V~@V8@$P?5<0g_-c(h?k?CwZg7ErUwUCOq~0TtvVg{
z6UJQ~>%7DCWAdl`yy+V&=tSBP8vodK)%Y!ABV12mHeMh9I3&4LH21BmqiTYaL@Sz2
z?$r}7+_^1%TC~&Akaortx5%)+Gqd*V)1|k#gtE!^XlIWri8dAg8N*ak5gQYbuVvS%
z4zPh%gk*c`U-V}6FNFL2g(<J9%7}%jxNzSuPZe0H-yqKZDZEd!{~k#K6u4WLtbd^W
z?7Wr}ymY;Bct$ZWE%8r17b%qV!KLxGoh;D=t1gTqkRSW!;o<=Wjz0rmeyA=h8;wg%
zmd3bnlm?6Ni>lBQSsw9aIz_>@84uV@#-6iKd-ql0{`(JJZ3BpDmOIw4hovMP-EEBD
z&qTs<7@0kyGKbFs%2hR@Nu^78aS4LZi4flVO8BIz+e;)En@n!4Nfe5G%@A!eIJvkc
zULbso!&r-^E#N<{oax2vGWyFgGI+H6_9Y&qYvwZK$H<j5?_jBEZB_J}CYTeY<ghsf
zMoeUVrINWf4TWVvPwi>w^Fw-)gYMWL<odT6<Y9lzeNzzaR<qbFNZ;aCJQ|&(+~coF
zcxm{Ym9^qWMVdo&LC6cUz{qY|BB@a1J*#%l<aP43wR!?l&z#bG9raxO=hkPk18lc6
z-Ni-iH!S=Sha*9{rkc$9wZ?ZLt1Cy1Odu+QUqMuIn4otvbIqA1LkLCPxz-CW!Jojc
za{P_6ylm4HS1DjQ(QR{X0z*t-Uomf-*X(`AfgR8hk|7_ixe#%2CaNzV)@&NDq)JB)
z^*a{AynBiFfuGTBg((PR<lFb2itj245z8B7Sj~hZZ&M7IZxK*MA$l0nk>n>zVN2Un
ziDEiJB@-}XJ+n^_jY%|vs&pFp2{-P{2Z}F0&vo{rwIIjhd#$NhMQE>$OEMJ_@O0(x
zci%IU*0z+ZXG7msQ$^fnal6wy#e!ALqQogY1`#@{OV^bqn{2Xajh(bKNruAal18Zn
zW1F1+?qOn&lh?Zy#(F8e`WZ|Bl>^TiY*TU?16uOcy(g;FnE33soy)y2LJGnxTswMX
z(AsWqoS_h(r!j6tFEFhEf%aqfVHcU^<qDIB4pjfne1y;{inZ9@KW{UV6TxG6Lcd_*
zryj6bz}s!k&dXB$s$|M3Xv@(Qu}L^Iuqj6`UKT-bir~3pIOh!g47I_7?0amafsyh`
z83}XA&y$II&b0N&n!Vf5RLcww!>)_Ms%`O!bXJO=<~|Fo6D&GiF=juSsu?#Y=d+Zw
zxKXyjV3LE?YRiJ64|6j7qI1{V=FD*|cuU=cvP?Ve_=Gl7KQ4rBFklN+-MMCb)$`*;
z@T*lpN>1GcHv%}k0I_PLZEl-;g>D}_>&{A|Lhl`{8W#<SQ{+lGCAF{1ioIR&2vv3`
zb$2IRc#wPMBx#J+>J%r~X@K8Gbonz7CGS<n>8sAIx(kjbu<aw0J!=#_8P20nY%}>H
zF%6Qf@$Z7Ch&;DdH?n4IPDjb=T+=7+NFM8%bz>@+V6z(EbWCG-*g~euH|c`iRe`CR
zMC8a|hT3x{hEC2<)h|zAFCIil9K0a3v;4KNGq)fhH5#5d+MUpzz2WzO;=vbl0;LFr
zP7d|M<Mc1cN+kh|Uzvueq}07h1mVYE5JRBzaZCwsr4<l?mFhz~X}RjDH2bk-Si#+p
zp0%fctTQc6p5?Mh#h}=eKM`v)X;4g`HWC|CQ69dF>FsJzkZd9N+&b)cI!lm~DsujT
zxKsGO#*kz?rdlh2WEmA`gU}`^sJ8tnKE<}uv##;L40>OSS3p`m=>t#z{WmSBvgXTs
zjOD7!bAB1})#uY|TrYwi(}~~SLiI8Q%<=jUa+z?(4GK}Zs=oa4le~RTL`|zgk1l4z
zOaSAh<hdeA4qGRrsoib4k|Tbq&u{Gw^9m$ke2yzFDIEF&79t93OCAsiAfbIuIBqFB
z<q<V-X>nEgQ$*jQc)*{*ElK0KzLU|=0vk)R4uyYTvjGWp+Y~yMUc6enhagLclcwM<
zIj*MP1GZQnty^>(y<9w+v2J-X6<ev)oI9#AZ^gp<xIY)AP&2U*>Ei{1U}X@71egD?
z+I`|AOyp}y9Y#}R>YVw=i`c8&)BNhge2RU4_aAZXj6Ui0^<HmpC@TWf#^`&-6c<DB
zw^Uk3VTtnIt%Pv`=>8whYTKiGGt9JaMT<}PtK5Rn;kkd#$S+PzZ(%du@5nej#Pm+m
zL^NR11)_!XMI5_E*s5(b`3vEXADCwrQ@CVSul>5vsLjqD?dNvJN)Yd8>?R+_E007|
ztFBDfu`}A^XS>9}YH?Qd$e<cdq|JSX1#v!^P1$hpRSG%Xcy3s5hvntry0pSEXR&*+
zS$V-|GOBX0@expqX6d_#xJQ)Zu_|Rz-~L_tj#Ne%5v*GLPf=yH#8xEapQPL08O&e)
zzn+rrSb@|jkLc703nYA*I=>T00--9%cHqi>ORfmUz5SzzYHO1#sfZv^i-Bk?b>}Fv
zrP?1C?5Y-)IU`N9UA`x7jsf=-!t&U&iLJ;z;KH1w6wiJIRXX6A)Fz`Hy7o%k+<(VN
zb2aI!{!5T^y6romd}IYHxY`7AtlxG%Jo2C5fxLwWf`tdH<(EbOVJvmI3y*8d<3GXm
zcBk8VyaHLo;-XprFZcL^-u|7gTgS!<m=ItOm`(JH9^2VUg8^oKD@Rn$Q*=Kj=d6>x
zdgCv8P`9guCy}bMx^T}H_GOmo@#CHORNTin_$gs{&joi@4;oA95msS4#B-@a_$AHg
z2l&IYSG}Ky$}j5h?@On-smnq3rAtn@Q_@zzgcNOGNW}(wwz`oOwRq<Dk1QmcMKj=o
z{X`YZzS}R=1k2B1z6)&`<h#cf{R_##c2cYAjCpaV0Xl63feD4mwW_pAEB%!B5wv|u
z8uyeYx4kOSB~tDTn?+npfI*NE@;+6w;w0PyX7Drul*+qaQTg6YTv6a*EAl!fp7YW`
zhf*$bqm6YPqb^x}sO!7p$-$GS;Nt12chl0a>hx9N+oBr>mO_}E@#<SWrLX8Wk#E&F
z6jySRvs@KeqE-H&FJ@4J1p3p(>@M#{nzW2G`Ec*jVRg(Px{SG1^`Go;lu%k%UnQE5
z?y$`qHqG9j%6n-`Z;%nx^oa1@P|cnGP5l1s<Ff`AZO%A%-zm0;YMoa`Ei+xE^Oz4f
zuW-k-Zt)*Ezzq2lmm6Cc+MVT|j6KMFi1SJM0S#FetwTN01h%l=WTr5u^PXvc@)4cE
z+X=G%L%<b)xFWQW1Vjq!zZ|3e7gVkOA3)TmX!)u@)~VDKm31!cMvLb#kh4swGIp7&
zl&?)qM$ADMmGgcWnSg^))l)Mb5t@5(tCXr=QN#t<=O6b~5>52635(}b456m?*HCpZ
z{XIit1M@DRs~!wvg!vffbz17Y{bnexAH+n(eM=p8u2`PX++NgOl)G@u{PpO_1H~OF
znKby+gU*Mxjco)^T`t|eh`T04wvRy&Co=tESmpsctJ3~47&mh+%(V$uwFGrCCU;n{
zjC7f@qt6WGx$pEzeD-xYwdx4BQ7x;adq3*z8NM3*PK*1nJ+&fcD>&&o`EP}6jR^Yn
zm!F`+qv(bZ0Gm;GJX-D4(QE&dH;IW-St#Uf5fpN*H5Q5tGNkOfPIT886Wz|3Z=9?`
z!UOfjP>|(j<HE6DRS4CD$qkRWMi2GJj<Yuzlz`{~8SZH)Q_ySByW09AgfIdN_a2Wt
zVP9Wu>83PgPM$wuh6C0?Y1#LOaaqX<4XvWhy@GGP8FH`w!&I0G4c@Ivdoj%0$dskk
zA6O~W#0XnF^lcHH<D`=!(RH1%m2gzV6~iGrk2?F_XoM%qj#k&}_UE@6y5e+Bc$Az+
z;mo)@()WyKxj!IVUQxz&+1pU4NKfa3`WIPYbsMQW44Pp(r-6E~3|w|%QWZT&s71z)
z1L9@+5QO|1M)K9x^<}4fETepkWR6~|QSgOY%Jh<q8Ykci;m|W|wQI4$gR^Sc8m=hd
zn~=l~yauQe#b{`ToC#88Mc<c3_%Nbu4p_Bs+BLAx<XLmVhq#{%c6iezdaZh1784M6
z6!fS{sps~RhVw<V4<EK)Au#W4Je0yYwV@3r&^L~B42pR)k}3lBBT*2w2BlAL1ee)g
zhc+W9GT<+m%sDc>@+g)>xmq*r64p)#YV1Wi=C>Aqcvc~|7<fB^S?E1N&giq1c}Joy
zw~xAZ(ui3?1G1`=w3nP@Q6BG~oa+Zkk_APg3T5=Ev%UZy1%c=B7I`tJ{B2>Hy9WJ5
zc)f6F5weYA9pAr<E6PJ{!zd_7q)?Q5y%Z^x6t}d|Q&+xKT(cJ;%HXN=!WHe*VepGd
zIi~bS5Iw_?bbtlRlr`whe)fi-<)+hnu$GbZB!J1l#`cFGTd1FCRC=AIW6}a|OCfq-
zglEM$-FnB%;>5Dzp**=pZ)pAWt&Z>LX-zvHX-D4fo-6SIq2i=(r2d!N>0!%mckCzV
zsRONhlsn_FD5qPZ<=lS1bCX>E`@pS0Q+Z6@Mwzp>_LSC9#A;#nME@1_WcV5FbdFls
zd8)b8##XDLiOG;#xI=Q;q;p*o^9-3=pO8|~b5eRw!&Z8{(<u&<M^O(Vhq(*GlB}ci
ziwohVx|V_~Dms+a+=b!rF&mD{3?w4yz{%?@zn6dUMNXMl8u`TmSE9l*q7XN>h_%?l
zhXGpL%blMt8dAO%6!_RmNp)du?98uYi|8|taEPH?^MuB7iJ(AIMC%QLav_t0sJrq8
zKeUTd#_g8*2wh!9ewecvIB$yFDn#(6Q0&2aJ!Q*=FF&Yor{E{fn%0G*U;&nabpVdM
zT9N{Empr3Y$FcImzeinM?bl5Rf2U26{=U_FLOd0YKU!<zbeK$4na6qD1>Nx`ILYt!
z4|eVCU0!Xi4HbEwX-oC9+~P^2PI|s)GQoX66rm1l;CFwf^SX{#>vk`E>=pavOEr8b
zWIX8#O!}`=4{pmmk=XA{y7Gb8=|WA$P*lnb*f~MD-S52-$m`a_>Vl{8`S{WEF2K(m
zRhIOE?!b20T%u$)hklEnn>9qA=PO*$`P4+(ABfE1SP33+J-0yBKL};q_VV-uPCtT?
zKU!B_U{Pf~;aK@<-<!kpV_gx~ETZ5F(XXmEVAEuh_~e{cG^&@tfcX>2MG$P#@wK!_
zWL_TPt+xAXCDQ!;_X{*3b6uB1&%zs2S5g9oT*al_jEsl`mFBA54y$Ril>%aL9!+up
z6@Iv8u<v*7H*w_B+UwQjC1@f4RX-@C@VO<;t{RA-hsd~G&RxeC@`?0k5PvKk-8u{a
z>M$CF-k4_MflmtX@U`&vu4k9`eM9X#Uc4iU=SQ~Q_s)UP#v;<Fw=6vsF$JQj480gJ
z(JU05ICQn0&QqNj>(-$uHs_B%^+Pf;GLVvHzJ$^wJk=vr62@%2G*(YwR08yV?gHoU
zo_v*f->D6Ejpqw?IJOZ|I-Gs}F9I<jif^C?h$1`d5R41!ds_A({wKF%(5viz{h6EM
zRh|(+F{!|N$*qk{$wFxf<F*nuP2LfAwk~<nWURk%ta~J6Y=zdTomiJxjvFsB6msg9
z{4M1XIUFlK+!`-w8IXlQfI#g49>_LbF!I6a<4$AVa5t`lBp5%8wk#$Bvih`Q`5R5n
zh;#3Q2lX@2%2ta~iv*7TBpA!QMI}!q<+vyE^yL#UOY5Im)OI7a-!oEvub)U=Fk*KH
z^%(QThP&pC7w9h7a?jXKgXTNy%UnL&XSm#2RMi@W;dbu&qAzY??RySa8<Ynk@R`%^
zKiF`|u0ILkj_w_luo!qha7o_L6P~+$&Am~)FT|5qt|S<P?2?>#V@SppLg3x}_$XwW
z-=?7EZj&X?Rv|VRz4^$us3s<KkE{IH;C}g!<I0+u^Sw0c%Av3Km0fo19&bhH5`Cpo
z-rc@=@e2~(I_iiSFuJPx_UiYX!#q{y%Nq42aF>oB!|V05O9OX@en+fc0U69)e}sBd
zi-6bZz@(`ib=_$=Gx9&Ld-|AKJr?|*gIA%yv8MUA#`w3Yp8J?%3l|<(au>cWsl<t*
z+up~*!bw$AQh0GnjqnO+eZ?W>X?E@TX0!e(w(M;uccC1F`?>w)e4B?%VfXe1kjF72
z;j;VwZ`-Pq3^7ly2OWU2?qKqg0%&>ynH;9ora!f`WkDO0KQ?Uy5hE@a_58jT!UHL%
zyuycz;F?JIl^0~)6|yeHT^LmFak;!<S?=ns6UgrI>?8c(Bhpa~quzMZym2V=W(JPL
zS@``Ga<({$ehohmsJ*(%qI)};j06TM)vG4gjJ3Ow904k0k%-#ZqzI0=UuP{r8<<N}
zlLKmG9!wBghO>Sy>Igsf?B8SiadXK1B-6x$jIy@lXLRT{Q|ePa0z6?C3Tji)`Sm)K
zT@gs|Nf&PA<F8?-%}Dh0*}2!mti$DgQed4eTrQ<XK+Mx7C{KOVm-@`cVJRMRR*=sH
zCgt?x1J&OPLX2ok)pNf(UPRi#&wQ>)@Rr=`jdW+NZ{1tDqWplW=j{eYS?$%tFRb&z
zjU{R-mUy)@g6p6~;=fI$lu1`MkoC-xO1DOKM*gV9`APIK2jl$Nxm8kUb?vT|?YV$`
zIJ|AySy#ZoPZv^<G_PYx-<eh(|N2F+f5rd_7&f$Oa{TM7x)8ZRl4zC9T_~#MY+yu1
zjJSk8`Jk3gYU}#<e%bJGD-w?PQ=$gLOr>1ejxtS^4b;Kji$hYJTxWQ)1SP8LP<qt{
z%E1p_PY{$bNx09ST{>?ZrKu64es`P99p!U@+DnHwDxZJ-@y|7zbq+fxNQpX;^F_R}
z&dH1M|DMl<o0<(20?M<wH}=El-vG>jX#}4Nh4@|6A(;~o^5`7)3Ffub&z!&l{(MZ!
zA@&xi*eu@(HkC=grN7o5i$}tboWhIbsyrNC!4r6|?ueZkys0?`I<g%27`K?`+;yX8
zA{a<XNCEXmC}mgMcU6Jc%yJl=d@UO8^DMdo%jBfMb)TtKzll#%6ExZWugAP*`PJBx
zuF{Zq)Pnfzyf=K44mOO0Pim1~2rqiiU&CLE$@CmQC%!t%&>B}!3(|^JhyjAf?S027
zk=kQ9`){i_Vpxk*v)AfITP+XQdt3rutu-T=p=|zF^8eA0`q7@(w@g%+{4O;4ZI!u~
zAMxwvLL1{pwbWo&{m--id)Y$hS!Te2g(kJ!4#D;}SF-2yr~m7~{%d?>0(_>0JWmJ<
O+|-mcmCB!*h5kPSJgL$E

literal 0
HcmV?d00001

diff --git a/applications/imgs/rec.png b/applications/imgs/rec.png
new file mode 100644
index 0000000000000000000000000000000000000000..c9cd069e42198947cf65aeb14923765be296ab5e
GIT binary patch
literal 13765
zcmb8WbzD<#_y%mFC@6w-horQ`C=nRxNF^LC0@B?eAPqxsFltQLMyP;vgGjd{q;tUN
z8X^6j@9+2fyzk%dK4<5goj&)T=YH<{x~@A$2ck+wN>6(2+BGtDHD$eP*RJ~j>#y$K
z0p_M}I75NOEsvM#26ur^;N5r8!0&smYQ`SduHAq9-}m~^Oz=6dQQA|*$W!0ty{C_*
zyUjHxOHWrv7f(lfD^_nCcMp3PXI4=`Q9&^QRy$8mSCEj<|960(i@U852bJ>eYu8w>
zsVgfQ_-1V(-~M>F<~4R@KBJA)oXbn6n-G(tNd3$bt@vg!S~l$~EBDJKu2`$j^DW8d
zFGwFtw7q;8#_h}Y@skdiJ&c6)dZ@}f6<YkKv9^0|k7YW#IlZY8`wtFk>IbF0*3DQR
zzP>664nUgD<O_LB4mTn^J->NxU6}r@7QCu#>V-(#J|EFhB71(|sJme=`86k!{zE|y
zw|0V?O98O%f2UAno9w?M23enXH2$}Dl@8NUBl+Jsa^WMi$NzJ@?PuoRntZr?hTf@(
z#@Zo#Xi&Bg{D0T=g)@6012hVstsQe+L}OeuPIEDd%UK~i-Oo$@_im~=or>FYLTiUv
zM3&_&+vH%yp+trB_y2dk)QDfa?HtFN_U=Vya`1fcLq<l%rH!!?&1Clh4PIU>rZbvZ
z!*~x=aw4y8QuHQe$7}GPN5dOKL(>5Ec3Y_T%ufB}%Gtlo@JWNtsc{syqSqGd)#46K
zVLgmn+P3Hof2Bzg_vGc-Xh^gF`cg!*X+?q;zEfU7jofDsF}K5g>=oqqsGT5?SxQ#)
z=KGQKu;5Ii8p?_;$=1tD)(+~kcWAj>w4EMzU}I?P(y-UDl|}52vTw!H9^cp1Sffmq
zXHy?5HXY8FKjf{Pa_T2|VBsTM-jz_nN+SVIb>hkSv_-{~lNX2y*#4ms?(ydPsNVO{
zp%0aOq&sK4tPMg~%$g^kJ-=sg@bDYp3YAk^?t|{N0w}}pDU(}O?gJ(Q)IHMQ$2ROM
zO;V1~bJ!x?kBDBILWf%D2YHlc^5mAcy^BWS2+p}$({J;8%3su)z{T7}+{+CR^v;?3
zY<m7>|AlAcrg{Ew;qJ(m#Nagf@t0*+E$X<4H{W0L4tmufmpTPWsF}h}GWGa^wmd-@
zgM{TeKfK7k54|9<&$_ct)?1n&nzv24Q#nn@dY;+;AtAj)m$_dErXA=EJOwbhxRe|e
zjef8fr+*^1Xr%KLD&L#;pu*%CV2)?h_Ycp4)Q3DluQk+>5LH`S+q$4L@51J>jkU>2
zhjs6ILPywsYw5)E)IG@~q)oc*%I(8e>tFJ3A>D6#k%uGRAFH@QWp^6_6>L6A3`(>G
zC3<%p*6g6pm$<Kt9;zo=PtScmAZ=>)@hEyJO&H^Be5zj3boxPpa(cJ>ot`eU&rt02
z#vv&O&r)XD{qSdS1DIV_?u)%EtI`4<Zo1x6yLdaOeYBSDiK#Zy67zFEn4az^6xELa
zy?aDPiNfx8)N!CLGSB+tA0dI()K4IK+*qGm{8ul)aQC=&fQMKcO~tW;{*rhRGq(C*
z;{A>2aYInZQLhFAStZIcNLK>A=QATP?BrN!5@9Re5!ar3Bl+K41NMi9>;#QE){NNl
zu9Z+K76SEYo|~Hp>0vq{T$VG5V{k2c?|D_d_Nkq2?!+y3R3-T=-JGItx#QQqSM<fD
z$H3DWcSr7l9L()17jv8sZ&GzYU+O1co?bA2b(_EPt}oj2{ZU>nHdF6Y;yD>DPxxVI
z9`w3(;iuIm&|(w^ym<Jq28uco^Hgn9)h-UDMm$)|#YK=;{mlN%f6v<f>z^clioxqU
zdPvvlT9165$f<ghqLGb5Pft&StXt?5vl7Eeo23s6tB3yBKtA@{sHD3QeDq>OkzB>x
zXQbW(wIj_&0vcyvbkYj37KDV=%@iLUAK!o2Ps1KS&v2A8w?os~QT3=+wL)2?Sd6x`
zxzS9!d}4Fj3-(sI8~zp+Vlck?bvzPyg;)8*VbiaG-MmE{eWM?+!ObO*eEs739bB~5
zQ%ZHc>Z#Y&Q={)hb*rbI1#NhOu&-9`otY4>p|t`W^)#?EscwluK&J!tmMPz2=*<_w
zvQ-6kx}hn{wH`70K?O*Vko>P!)spl9H`Qb`df-zF@_p~ilkGoaS}b(d(7le;Q`pA%
ziXVf{<-P6!0gc&!0T-~SW>0*!hZzQb3BwGkf6zQV$s!sG?n1d<FGA8{hSw^>^v=(G
zWd*|+lmzG{^r;xd`oh?XZo7-7Q{DcIL8H;41ys&_BZA5-a<wEAH>?djyuI_XJXG$T
zUNixQ;<sl(C@$E3vRaN`yU!;aAsyNFlwk{`)Fxl%vQ$q7(>?3%Ru>UGJ3FH%k0N~5
zpD_}7JzvUAKV?(1p(DHTok8iZuH?Wu687nCl59RM1{YD`b)Yq+m9Njl!a@^1KS!p~
zHy7M&9v{Gz99KvMqhNZYT)%0kT<NJN;2@6zyvw)YH7ahviN!wH-r9OXSLT**svJD5
zsDg~8I_VLyxEkl)X!1f@t6s!F<nlb$`p}G9?(wN9cV8NT;hiAkVtl*GYN`9SG{HRe
zp!cC$NJJ^|1d7I~4X+sl1O$AjG~qJjaZ+k_u70cN&HOGvS~NCX`=J_UVFNhqOSdvG
zZ#3J<qIsGTt>rV7JM{N3-EYRMBp+vI@aa!M2kcjLTZ08;c6BVsdqUZO!b_FI_7=qx
zzB~-()PTlMCbOSU)@O<_g6vLs7-UiQ`jd#eydaySh>MuJ@pBip@}*`+?SYjGaDxSf
z-sd|qNQg>tUx3VdyH$=^iH~6Wl7x}j=F-kgy`eqI4}Xcu%357rRf~#>BJ9t(J(#HG
zNu=QXDW}}i#GOYMgj)6#rllfy1_u5B`rU$S0m3}k_#(JDS2a}^AygQa%rP!JH|<;%
ze&@^SudX=ojP4&h@u97$YJS+bT`kZNriQk>pQ8pVtE&w^#!am?iNz-Js+%Y33U@~H
znh)q3U%oFlE^KX9AZ^4`eoIPoHx@*2$$CXSeE5(boWirPJyUOUjr`%m(Jk*mk$le|
z!#GEYVMhsmik1(R-!P^;R9Xyaw3ly<ogS_0h>~HSgp?ZDfsk(=v{?x2d>`c_w-Kg=
zA0)SKc-NZ_J*M^qNmC|dP(+idC}RUVwx%Nj@fYVaLG+28SO`KAu&WJGh@OL9W<QA%
z>UFnp;Af}Et`EKa6*$U~xxLg@YRg?}ZzoGMK?+H|Tv}|p<z3$;bhaQLP#lBrY0@LN
zW9`P9dO$YKk?8p2I{7i{i7G)O61ff-$nLIOVUK%S73bV~yj*7s_^&KsXGsK!oYLg(
z2?hn>k3`V(M;WN&o9^e@FDN@-PTo+4xPM_e9`26&gJtjuFyZf>m`I0cYXheA^7SoX
zkaRX^z!IS@%@<YM)3s<k-kL0`HLNQB8av0DoGLoWd8V#%d$&W^h&5i`V}Sm_9hFB$
zPg64Pb{(G3v4iJ;MsZl=5~Nu^p+!PM;y)NLi{>{hHqiM}A(7_Nbn@P~-<o?xffOC_
zhdlJ9B}SV={eX_*^VzSn+}uuAAJO#Hg98KnUd>nIzY6JEq+!^{s#J@jOzW8hW^}Z_
z_1k{5Y1hp0o-~-p`)I{Df0);)0_vLm;@gV4u{^QCi$XOEy>{pil9J*9x8h*L&DI5-
zE$_Fgs;V|-;T@uRY+d|d?`<I_23<ii0bRMI@{<<_&f#*5<S8D4h2Rj|6Ol!<mRr{P
z>2DD(dLRGPhW9;vPTqcu0ds?cj`zLB19P^ORiQM}r2-&Mb8EVldSuOYJ9thMR#S5O
zpejpv`O`fQQ}FjUck{GX?Xg=QjUOE!A8Te84X?rL@Q;=J)vEag6U#pf2UdH;e3A1O
z1fMqf$TB-fWTwPWKCHQ&!p6lx!^fDy&lkHBr&+(aW6%&?jOP^YO<e8TDtF8Ks^?1g
z#P4_iU_LxnKjB2ZI?B(5{z~n~Sl76d_|=btZ4#y&zojdKF2Dz$s#{YcHMB02BdQhM
z)}IQ-l|wdnb{vFh?LtDD*=g=D&p1~vHlT@%<Mk-$o9@^4!I5iHVO;y|hqdPuj&I(m
z_P-Z<^^|gJYYVB=C{dxMP}Yb$`6t;^=JmOGOhxR{c^+uiJs5GD?yX#X3yoM!Z>EGc
zaf$>NWO-~^_gVkVo3F#bAu!EWmPo;w8Wa*~FUw4`00M!AQuqz|zF^<ylAV629J;)0
z!Z;21)<V~pmN1{Hph>tjPjUJLapuI)50yR2*b3^$`RaNabpgl49e6k6nBkq{K|eS9
z?++yI1ZS5YQTyg+oNtbPH{>exwSAb^`}6wF@!6A#h1KwlIeXcHkAkR<r^S$%yNZ)5
zr)yd3Iw?%437^QyxY4F8hWoCMr@Z#)cM`UI-r7Z@!yY|J?&DTf^v?54dy@xPaFWxB
zLniVf8F&`1LQKyz%I+6>IL67g-d;8^v77h=G7K2!wVtY0T$?=Zm&oF-?{e0CJ5h_D
zq2A5of(RwN5`MFwZQr5sXYr&T#g9J)XLW~T6eS#iRbP@aI%8OWJGrNeylUebz1$>h
zJve6|)GtbbsUV$Zh&Q4Nb`7<YPeHqDp5Izql@9$Iha!PS_Iz8~78r>(08_C_2wlZd
z-lcx@@YGex0O$w>^P8JLK81z1%xCRR>C&=Et7_l{i;xLplVavt;^2{uxfuwAimouR
z6aQF5stpE-!+q>Mu&CcL(;MG8eop(Kr|qGxE0YL)OhK}><!uiJ+7-%ICuZs?uOub*
z1-`&O@iGW*S&4T9w{?Y&tv7Ybw+DljX^DmLkAhfI{x;VN<XYw)NvS@XdY3zFAuaL}
z6Y5IUQ^Cmbktb6i^Oz{Fz``WuN6j%gP9k%&_qy(Gv4rc7ZW{U%9+{?rfa%#fJCRxZ
zzNrwJrYb71XxJzGCyx-ko+lGiSIRZMDCSw57(0p8l-TCQKJK8nF7`|WHORkEUGZEH
zsxQZ!=;iLN_h-I8Q<A^Orp*QWt`3NlP;U_%pqdfQZ}k4^PZm<X@$`jxySwM&>@Yw$
z6dr!FargKAl!I4DpB+yS;4lhC_6|!z8b8$<mY6*|=>xiez2xnwwx7c`L%c@P>x?E^
z&uZkni;Jho#{zfV%ZG=DNr%cUWES1i4Fn=Pyg#O<+i}*^bsUdeNcO-We0g_h7izAu
z3G<)q>t_TTY%$a^cM-UllaANa;&2IOEjD^G)e71K?8X5!RNq{ymsWP~B$_ek8OZ2L
zn%(=IzpJaxb*AF6D<(PV1ZAc}lwp+AH!n^KeRuFL&kBkcX@la6;Voa2kG?Xg^QIiV
zy(F`+TQ!zMYmPB=Du5}V5z9xgPP}NxQuA_R(ggYsHs6M}!!0Z!KAw#s0YL)$-Tj*L
z!Ct|;;?^6Mo_%mts9)=6e*gF6!dF1Df`klWInE#8Ft_E{-$3Te99O0%@}uFt3Qsa1
zFrFZq9xDZGQSRfzju;*O=coiVF236w7|^92d9=VxWOUu)zjSCb8JS<(=gB(~`(etn
zqX{;Q8z4?Q=FM2h>aVo2ca6?sUSW4+*>}gYqgQ7TbzQ$;UAki2fzmX}N@_aRwXlw{
zI~n+H15>T3(jw&=9GhE7w0^ZTp^O^EKP;u|WVQ4{iNTKn&D+Cv_EP>2tlgc*dK2F|
zgr)cK7Q`Oj!fOgO8h|tsJG<XN<Uu6Mp+_G<$GuFMRFO4VeS?qtajwEbs%^_<uLQ%B
zJ`VFU3y?p>32;zCMvQYv!b6{aQFkI2rx!zLb96O)fR<VjOlaikmumyr?Lj7xEWbW0
zjZTB(!0)eTs$YEDC^@n4T(7Iip0UZyhbKiln+k@44Thoz)>s8vMcNlivAP!9>01Zp
zK)~HRC*)M#f}#AHlLIXeQr=w3ISea&0K1=1+Z#{bjmsP^SP@}*drLg^6}=4_Kl<S{
zdrM7hl)n)-S}TcysosTxRrHBex)vXftTUAyHgUTrty5U==2KzTYx_i8NBdEIWBJ4n
z#|L6A%n^ce&p4cLqzFe5ZTbYE1gC>l<!8{KpZxHJWEBQE{()j>!NxaAz2LlsZ%XfQ
zUq46B&XDAn@ca`A`t6VYovcb@uWE2mw};-x{n4qF>waaXTR_+Shpm0ui_t?tNx9?9
zTKw_yM_8=23P!tWBxf>uMEcUv+Ebzw8k~O<NV;u;qt_IxP+K`#9o~PX4c4h_BAxPc
zmcB8r+f6@ls*T2xqC`Y0yQ90QIup<LLuOmcvYgS@tnO8%@$QT#ZD(k*#Tqc{{GeAM
zJuGQTBWkek21n-Vw4eq;KBs3U2&^!4@@$9q^;q0lxjr*_rNF0wDYtxk#?N*?u|x-R
z-f6Iobcz16cv3qx-f?g;MS2p+`7?T2(n4H9LW5sR^sfdhR5Wx#;ffmh(GP#tuLOFj
zpotZg7Lr9mG{?8eAk6n=EcrN)c}RMFNt{~Evw<Xy(R;Hz4oQ;7cqM^w4JIV~Wch48
z%ga+IThm=;weMMV{L2}^#F*ac(f43O02Gm$S=HHiLz^`Bm)qPrGyYv;?5AHZCk5p^
z<x+X)Xj*ial#u0d9b*@0w<Q;6lI2w07);r(P*{Yn@ZqqHSXlJ6&|V``@N}1o3>Esc
zZX~le!Tn75z(7JorDcUvlQ9q?s^a2&t3{LpcYn$#(_S9VOIUy09<6hN{Jm7AIr@sl
z0DK(BX5mPQm`fs&7Z%7Biwr1r@}yQi@&ZpdxfxSz4UM137mJH%ptPPoAt4k^QTyMQ
zx+*6}ztA2U&6!WDi%GgM#emck7dY-+sOytr`Q^UodnTzT4)N3xA&ozrDjl95{#lAF
zbtyG3b<z@Q6SwcnBtYdrt*Yc{3SM1OM@UfXJ@rHxn@_yTEWCYnwDv?g5NQwMFiX}l
zAP8C3?PjB}h$-(amv?=hIyM@xj})aM?f`jZnUVsSw~3&;Iy5YLkbAKGpeT*QD=K03
z>>^nGHLaZ5>k`8#(p8@58!<oZsQPrq7Kox!_c%v6l&=UcCKTIk3$m2$1<(*-w7cfp
zHdXnbre2ZAEs>rQJmrQZy>%@#bvaL-rd3n1=-lANU&88lXZc^}u5ymwJA2kx5_Jul
zK;q_z-!YU!%0J;G3NE(4{Zfr+miek8w!}+nGi(6Z=C=ndV98qC05^XE(*IhDq*%+y
z7r%rt2VEsFNXA+&J*`-kVzeJ_nXot+(U7rYFeE-_cc8c0U^h%lN9mI1HgtH5`fbbM
z<T+fcQ%={t_ti-4<#abp4;*O>Rcc{yJ&|qZhetc6Up#E@S_=*6B;JX;*%nBhk`}gj
z@4jdl@4Z|rc3E>tqCFKaZPuZWxgBOJ@W+Dz1XyVxxR!Z^P&EJM>0e{jO>=QaFC3VO
z4@S<%?x4@~E%jSYlOF|vWYw{@p-AvXjAd~zTJ~Qsza?@n&Y%;;^Hw|+_hR40&=mI(
zoKK&WuFoQ?7Pp!NDS$*;4`f|%2wMbLb6UaI_h*<lh6Clt*47I1hV5-y*=R^>gi3cr
zO8P1WftU_R;K4FI_sHzw{db)Ujs-H;MDjjiolf)Sb`W=KVfkK=5t{)5vFBEEh1p`-
zkbledv!0t2{rBxN%N(kwQs5juOA4f>{w5u4agEq&6S3p26jAXus$Bp*@lKiC7*0E>
zm0BQ#R(^I{nF?#$aErK4%vSa`Yo|1%(&35Dj=$sBXD3w?HyjnOPDo7fV)1+9r5q^2
z&dL%fM=#DNS5i*~U3iI}x1}9zYsEsTb<6$Mf*-ZiL+6y<(O3V}-wO&F^GIdjH&yNu
zS^hOjPT{BWiP0updeam~>BDwv9L{poF4|g~U*~<MFf3dwG~iMT$&>3!1Y)(}VSOYL
z`6Ru;d&|Tgh2~*5+3}k{dptb|8gcaV{4gRD))<&{EaG3LxI`*6=W)Dk+W)>htRTqy
z;UFad>eYE~Wx!yt*e@{_y+<EW^)tmlmik}vK9rnV4rJLtC^*yLrw@1+<#M0x>B3b@
z+2)Sk=Pt7ih489Q`OepJG}!!P4@nPv+b5Fo*hjkj6ev&mYHsA9{fQ65wSd>=)ojIa
zcMCWlG#hb8#2l+0C?443A5HDJkLN{y?GHD-AcU|F89glNl;2CFQZfbQME{`GIGe4R
z{74>y*T8q|*?x<5g4u2=%Li<t@%fHz2FMcK7OR7nGInzglexpplQy1V**H6RU>*f*
zKta%E$@kP>xeLcu#^kICHw~xX`Ii@5qZP)+U|(#Xe_nH**8INSo;xinasQ2>V5k)|
z{>9}Mdk@8lwa!ad3M2*@bHXqjy8V3^Ucs@n(fUVUT1tRjF!wnm?U7rDshw6Ktq9Qw
zSNEtl8ue0R0pAe^ptBQPE1@3a%1Zc@TUpBS9TNo~>nSIiK&>l;X+vTwB?O|bjHl&Y
zAwr$S=W#zfUB@F6Pi*gVB<g!zF~!t6&GP>WZ)F#Hm2I9Q@W(=@rJz8I;t{JWbX+Jg
zajM_0r=wSvaO#_Tl*Sj`F@3BOEH-^yE|Ow%ZOQKV;GhC-^T+~OrXL&hyxDupKc0ND
zjGo}y@Fkhz;h2B90`oDD3&W2lYHH<K`WFm7Uj>lVd9_<xoVZB?ZZGS_I%sAVcms0t
z^1u@*x9U#*I;gz1v5k!Bds#eoY$K(*S~{WU7`*#{e`mEyDc8(+z-i=yoo1@Rr~T76
zgV#?<If*g#`@z&h(M{3+toas#)kX(J4kT5u2{>hO2`HBmS9J?IjFPu>vt~V?Tuuz+
zT5k>4qdDz#m!kG+PS0gqKCDBSs=aM(;=f@*r!1_@FEVCIr~xdsijz)K&`SFJ@EoFI
zz4X+{yhkyglF4DId8%VI-pTzeXl9<B9?IaE7b(yT7fMx;*Eho(>E#8QD=&)BR|xFW
zfo}n=b_v;Nkbb%%sa5KTARtzKwd!Y_xt?s)0Zv4kK|{~Yw>Z-Rhwn1r<aVuX5x7h$
z5scF`szpyk&dB%`u>e5QlST`^1lfZe`xTt4lq|P4O%9NF2kr?GtM$C>^=Cf7qr))7
zlI!L3l!uwoi|VCX^cO}CpMWR9G_sBT-=!8|wIHld5ZLy?Xl{5;PR`!svQ}=6ErXrZ
z`U6|#lt4lH1==Rc7}fCK3LFX$R(GxSf3ImwYXqyR2XGz1XEL2*UL6e%Gi6n#PVZ_x
z7taUJ*7d+%PVX-oP?1yQABn=y*YBg+-C~0Fq(tTB3Ax#6J|j=6`a5P`2!i>b!H#bq
z`RsXvknXS~T;;}5kAS_*^)-o6WaFR#Ibv4SD%KsZupz#6pWBMbcsI?hxpQH6_h&<A
zDE^6!lRRo;t9q{ZV`QCl^8X+=E`^eH#3SoS<~IX1@js@vQe!yzDk8eRluGPoe-gnN
znON}t5Ey*<D~3UyWmpBwG%dRr>^4e*{semGz=o(IF`w`o7s*hiu8G#b^HCyp{v?sU
zxJjK*d$uq)ZkShQ_p|FQ!$Sxx;P*rr1NZMPnnY<bO1~Wm6>*Iy{0`$(8o1Cy4_xdV
z_b`*VWNiAOe~a_#<WvsC2<8_#q(QXe-2D5VA-2B^KUujq95y@1R)koSUh*-9GxWwI
zjpMBx`=$+>K2G5`Mo2j{6Oku`>O>Vn5>aAsVC>Lu=N40j-z#B!iY$DOHbqx(^BjoD
zaUz12oR5|@dnU*gJa%T~07v`!T_r@4Cm@Sv-$TB-^SFs0cAoOhA-YhDqHz1=7r}6?
z{FZWewb64g({Vc7{RCs>c1IWGG78Jli%V8cb+)gi)uBKT6Bv;a&dXj{cRjp8tfY1`
ze!zLmxAdaHS-w;Lp9}>=w(}{dzV3AGuC&?ZKxrZ=fq-^zZxU^6scH{zT|YIKd@nA}
zNU(b+1(tXO`?rW`i0P8+v6f|oea#abisi4g3P1`;BnGkbA_%2}a7|)3d8V4&`irH;
zVMgbu&<uiQfwlO4>9iCi)u$T88h&>|_MLzHnv#NpRr+N=Zwi3g53C>zVhpU?TI~?l
zHT<B!M{-Z_f==9|_@zVxq@CC5(lSKPi>BcA*lf1LY<hX3AEc`D#G2!F2toIkvyfq9
zO%ZNY^uj~I67g-ro3~)I`F4cev{SqTvD^B%{iUPXBO4JC4Js}t!zXEpfeZ;7XbLxD
z`N)qA?`5>W4sOYV_?>%=Y9fu`<)4qiT-<YOtNp6{m=LsEVn;7H#p|C>E?Ax53l>%~
zb&VY=Yg_H|0^fwBTXCxYv}&pmbdf4+qT3?2QvlfX+{u8ChJTCD2!O@bxi4$n=8@Vz
z%5WFrt&kr&`6owC@kd4YuX9OanFU`TBy^3kH>UHmE`MmGpG!K&zxiAFP7j>!C!E-c
z^TEd5HHAO?TApc_bea7z&ZbK3Ow$?hwe_ekJ=%0_i3F#gu9Lt|&55M^mqt9eFmqaH
z!TNBC#@P*NS#_ygPpSfQtq5EDE(a@Dbjt#ymXr^OQ7#0NmiqIn1|OyNe+1<BJ2R=Y
zjcJ3<2;|G@cpF=zu{qE?joXVgKesMfn8?RYrXG!{h`*8L`-qC8)WhM*1lXU0?weUR
zOI3MY#0I^~v>uR7Ri}$dVj5-_=~r3OZgLa4*88=^Hq(gN=P!5yBeK9p!|*iMXiS>~
z-!x(2@g&7?-Zv1tMQ#kf_kHvT%7s`v+;j`14Kf*cHtZ(5*f{Nhzsx9!{{f{A&G1wH
zYEMv`^?&^qZet&}X!w1rE=b0(s!%Y_jiT-VPZA%NhB&D5iV~rc5VE090H@I0_ITM=
z`(lO9p0%Kh(e-{b1?TV+5t|nL-2y4<8v|W3h`#ibx{+{{rJV9xGxC+sGX(=qmJ>s7
zaM8}W$#1(?AJILhNNF#K_{~HnlhUs`)^S*C5Vk6A|7e41%=#DTWUg&P9AhB3Qg2Nz
zeS}&?x+}G2dlWPBb$T6p=6R71Tele8LMtR**XNgD6_;`o&G?eMNyG7c-O6(^z5<#U
znv|$!VV+eL6F}ZD&3i8<_u925)c<J#e$MS|0LYg6(#Nw@JwAJy>fWSpCzdyRBS<d?
zc=*>Hn0DRHXvuc{u9%U`1d&`>oxi?MANhUAeA0>XP9p#H>ri)XD3N8CRWAlZoLb$W
z3cV9MJ9gvo*yo|u^b>Jf11kG7GmYHY=v~g&zh!f2Nk3hAyz((+mMjkNxq@nI-TT9%
zYH4B8e$FM7o`|JnPc#itrYmkHzXwLsk=<i@dPb9E^J8#uPeEy9RUG}e{EA{tmE<KF
zZ2PLIt?MHUJ0M-E(9mJm+I9rqcOl<FQ&j%=<JUnrL6!YRB9@*dN#p(r26>@s&=Tv_
zNebKywNc^zb(%-)VX*JM{}?(Gk0J*|u};b_rD0_@s`K)y!5Io7hqNvCZG>S;6?BPp
z8XcZ&_-wt?NDQUArg~f!F-lKkjOzlRH`Re{s6OdAzZcS67z{$(KQ@vTyTxBZPEO8O
z`&=wJ$MFF!ros39O{IJ_Dr$BqK9-QU$UirFQkcL`voZUqjl_}*cX2TQ7^EMImXCQS
z&8rrXn^f(#8$@x>FA!>o6v@NKhdEuZBua$XzmTmmaNZBaD>j`5cy3e3m2Z{GhU9y^
zqkR>%^+a%mjZD-}y_Gp7t?Stl*ZzoC#%lDmVg53dT=Aw&#sC|bY(<W&0CB%g4wfrj
z96TT!52-JPm^ZuOBQb}`?EvHUrUwX=V;N0$pvORVL9~E$)&y<l6SknU?zJPnvc-Lf
zr1AG?yBi9s+*t=0wAkQgSDFU9E2~{ugMP=f_j<<Frt-Zs<_P9H#4BtTk;G$wWis%J
zPTjmqy4ie)Mq-WyYI(OcS7%5`>`?L6BPCsS717`Exr&x)F0qj>C;7NiQk{rmSpseh
zjRIoy6-JU%Ov4}nA^M+oOY6asd{(s-Tl%~9jA;G1Hrs9vv>58K`T;$K^OB6No>p9%
zLY8X<aUSC;E>Rl%x8GkXhi3VgQw@uXylyu>tyFf_TH01a7@QVQYxkLh)#@!YGW6rH
z&B*N#iEHWo?`@W0S5tpne^^oV1-q4}a5y3~l~yV@puAKzuk_p;M_TTRpQnpzXRD_B
z1PP9bQXf7C{gf(JKJ%InbMJ@_U2b<rC-_tOXI1niYMyqj5StMP&0H7ZXL2t}-akE+
zN!;|}3zN1HOPn>;qhKWD^3EGDJ_H%MjZdj6>CtTKyRNB4k7T{y`#?d@AVYH`w%>kN
z6eQZrId$j7Hwd&t7za*W{SyME8Q}Q{DId6-!EWB?<4ns|A_r3rVPrU#Hj~tzC>e)n
zM;bHRxt<5yx+lQ{o|EuQWuhnPWlBl8BiM600fKa^UL^W}^C@h~9*i*|A5usp{3xky
z%tn>Pn;@h@CDXNfK%S~Ssazn3p1X>aQ&uzVq6uqSJv1ouqQCc9vg+T{bFjZfZ21+M
z_No_!!@1Xs4u_nW$)iAWaz%^^x&-($O@KcISl<jtJ}{*aS2wODpRFAYx6~t{_Cy{}
zU3@mz#UNZS^}yqQ^blztmq#ndSKMpuaa8Y?MfI)i9Kpiz*vHaTq8y8*X}m@HWM(gE
z4W9`2iK^Us!S?LI$IsI{bmn-Xb3os5DM~qDQ_SO++v8{K+*bqrU<0)k#t8^WCGY0?
z=O3QIre(S;SL^P7-2||wlCCO*-+NrK?5G_YrnbR`u{#6%YM+e8BJ1|JGq9ntz*xe`
zZX3%EcJ#9OCqqWIfh{jj+NC>NLZ4Z4?2tiV^3n@^m%}{mZb{!hvVyg7wQms@GQ&lG
z-DGI;Ug~LZJ!>sT%I1wctPZ!=92*QM@R1dOk;CVE=kV;e>0N?h?vyrZ?xx(W50z-1
zZ!5q(f;q4U`J6%W*n^N0U<`A4a3!t58}xFBn{#b_^gBS{@G%(|{ngCQiXwiXW3YLc
zedXb&Q$O=}VZl-a0?pr4%<O-^oc-m?m!9qCVndcW1~X*cO3_mD%$)2NV2b1(3G0&d
z!;{BHt5k3D^3EH^Xjm8-0Zz)icUccWv~#cF?$$IbYmkRuKp^}H8-+^W1vNn-X>=!+
zp~M_@gGhjwA=k%aT>E5HZ}r_<SnY$(0iqPf`<lwk`t;($Z&+1L4LDVqraczsa$Ac$
zI1Sc~T>gB~kdX>r(Vc!g*d<gI+7Z`Doe%Fx4NS}dNCZJ!dVTa3khudoA|@uLLO?Vm
zOly#O!+kWB%uf6R;~Z%B<u%*vxB2#){Tz93(VLmvsX}Y#VI<Xqx*2}mWMKeZxtX!-
zz+T(+`-SLAk5eZg-vGEO6_jZ;L5tlEUd78z^pwc72O+6DsSS~)aKU%e9s@3pmvew!
z!9&NI{$khz_koisUMzxZW^!-SA=&H;#?Z`<YiDQ2@b_}{T=3h@&Q72yj{<C&ng1Hr
zgK2;ML~Ci;y^zK{urzD`@5W(I=$S;F#1i#oczF0Cw)QBMw0p)Mn)4D~tvL}hAE=il
z8QE~Ml4&A{P?1^`6}o?jl<@(Gt={)B0OL!=-RjeNio_<9P6CWINJi%ST_iHh&kEgs
zWRis8E0Z)g%0BB8^Xz(<JL4;t7npWGZx7bf>mRdjq%bvjD)*k^6;9W8fAUbdRa=#r
z%#X?CoQC+%F_G+3^VGG|&paUL7~R=&m0skGv&<WFV(8mDwDGA{RPX#Me&ib<8<r!>
zNiIY{dt#d<ha;%|w<RC(vCkn5$vl!jGGR&kRx<2P$G)GVypy%azHl>_^SC&NFUR2O
zmeU8uEIK5jekGTw`2`vIe|6L5-O~K9oTV*+x{Aewu&*IFFP^Dic@nRh%`q<-B=0*P
z934UXyz(jr=$Khp5LsWV4>kA(Shs2MzlIzrMHuL4M}<=qcQkK5?ZG5f2#bPtH-iQ?
zx&;^l0|J`N1PJRfu4YmFDT>ZFlXl;U;s6tR9&gY57UExWl@b_e<G&ui^h|4otk<Qo
zil4;D5WsH^LS^^<`3<zdI>82$%x)U0-O$F94=2$%{GCyxR7x9m|BNI$VNS&HS{CI&
z@u6F{1f%x1&amqfN_n)m{U5w9PlrGtBhJ-chwGy~<3Ht-&usi;WMqcByJG+%9t%i2
z5N7`BTSrF?$-Wp7T&WH6Sn4zuk{a}M4qYsE%Hb-iP~#G=izX|=^nh9CY<1?RW%V7H
z6`Y%rt9VUk38Fl2Wls^m5<AAqTI6K3LPd65viF50k8xG;`}6Ycey}KP9P2R+SI~QX
zMb)8@KIyAO_z(@(1pz+P$#qeRK{cpFc*Q=d4(|Lj8ab(1nEP{rR(*8l9k@hUk@`nX
z@~48iV^J3^#e~&_8!SoTa>0EG<&6SL-<n=Uh4e+$?@$#)*O<~U>GXArddB0G6b~-q
zec4fqwg^jG9^~B5_I<7TA$qqy-hb!X@Im+srVI%M=^SuyTXZXxKG_dcS5|>!kG$VI
zdEF|WUXj-;gAzu6pRhdNuy(!39JYns>W|{#-@BzAe!nV4dPORvy^Po8srIyg=tn<l
z!aS`huH_^c6r5kqSnA83%4-r_wJoDPD9xVN8-E~O%q#>#Nga9Kz2%HM(M_bob)3zK
zEM<FKa;VQ4dR#cEhn(p0*;tpD*1s7j>XP0#+L}r^&mkHBiUYunVPpYm2|sBkH0^e8
zm^c#Rb#}C+M;;+f{;TM1HEFERIA`HfGe_*tFb`E$rQeWzG>w^W4lOLJ9B>>IH#ndr
zQndht$UEi`_{{e~$^?nCb51DYn}csqUey^CfyT5f*Rz@CcUntiuKoLGq<4Y4;VqS{
z!93QpaF9fh36#HgrUutgf~f4eD9?)5d0Jd1bcG-|zI=<4qIkD!ocF~_+TcxaUV|V)
zGXaby{WIUjR(*DM;yVjFb1_(Lqyomul&w=`Oll)7R{&!T6ri8{@KosD97=H`_qP6q
z`U<%WgpK|HW~(<Zy$W7c!r^(AE5`n~dT+%;>)0)ZJCsh_``v^V*JC3i?jOB>g%a5g
zqt%e-nBn&ooAUCcrl%94+iBdez|To!d;h`3(co`$I}R=`+9I~vT-aR7*MMYj|FZ3$
zHys2c9kHlbs^~dlVPIh?!N;y4t}#zoa5cK;;WF0*=qF!4{SG)>K=tW7U6TXoh!$LZ
z1x<AbCqex@5#MbDpwfiLE46c%g9%c&KXk9$9<YCtm6Ln+8)8|(o}s|a&7IQKIkxq+
zF$HN&mlU$;!=y|8Fa2)Jg+~An<tBQ^CN^q7h+e!3Js|uTDHG%s7zmA*;a;zqmLxzY
z;H@H_Car2Sk!&D3%1+Btqg8;J7ftl=^)-@?9Xk|G&@fAyJD=(JBQztdm*+-&3wI>R
zdXJ?{G1#um5!-c>lg$pNNE{BoB^B?wT0kJy!V(h!<t^XZYwZt@Ienp~#ZJmuKM!N&
z5K6A^33xZx%+<SVA6X)@dmN79*)iX8GfE#Q_CL^m@Hj>D8{@2Ys0aBlzvP|pmRcJF
zYW7G+5sZ61fSgP`)38?MTJf1vRb1Nm$4-iNKl4$bY1Kt+c+WKk7y<(?@PA`>VDg*s
zj-UEHj&;9f!)<4Uv#N7VN*jK6-_1IvtW1BH?O_3%-m>IWS+=fUguREMdTj>Mqz%-c
zVA`jjCt6~E{~fi#w~R94YT9kp3``i70X$sBXB+<7z}bAj81R$)){Gpt{w`OY7=Xc9
z+hU>QezKYU2Y=tX@DhFixacYnU2w&Ex7o=FCqhd)>jiB$LFZwP+7`()Xk5Xy;y2&1
z-wjLn2wv~DMl}PkegPuQ`>cx_>s@cF$L6%PwR__l{oX$rbaWdb0@qZZ96%UX2iBp@
zN$f5&34nlh`ExlQkJlLS+nH^AJY!bx{R+Si1pJj>NYEz$Dm19si*0LIQ7<4;%43em
zj&;ko`Xd|=RFv?!;`t?X$aW!zMqDIb6fH+9Lg?8qRDRGBv|}MG6_sJtD9(BgLECFX
z+8j9W4GIZhXrp3l>$wa?Bw3wbl>>}5^7j%-K|hBhqg-(@QY;be*WL7|tLADmJm$n?
zKXJ^uo<k-3g#P-?DZE>i+hluGnpL#BMs0M`EfhL@{u2*EwoIPL1%9SGFm9+1w@KBd
zxu+waEBLRe&JJc%UL^4?HK!mX^>)64qd?5jmaHSgrlTXL^ga6-pOe1YFD}z<D{4M~
z!svkNjP{J#$p|TzyS*Ru&GT_XhrXtC)y}G37q_~uj&I}C*81t;MmueO#Y|&oYfv9F
z&j3RdBJ5hQFfl#BGY(E!6c68YWV8BYG3aDj{X*v-M{b$B$s;+JM@aZXuGo%0a}$E3
z1;c^>P74IuTsF({{_1p0l($7!2<W7*k3@ej+>g7+>V;Q9emNA$nTd3E4-@<D(~r?+
zXR-v(p8w3P4PLRlU_5KtkM48QiGPDm$~)KNU2{CC2jZ-kvjL;H!;LXbX`s0R0?u|t
znOF{b?~oaZ{6~!}zped&K6eD$-F!+fP{kFz;sYkXMdR6kQ}>Mr2$hkYW6hox(KfYT
zFG6_%6)DHo&JIFOSi;@&6SL$_qKMH~@&J3@7G@5|jc8Qdi-E!RyB*f+)vbO_z?fev
z>dfs0{Z<@SW*uwKd$U5dzcxSm)~vb)FH-Gv68rk!%qt<ax{C^iJ<}ItEPqFKEj7u6
z{wm9#ul*-__xo3tbT+nsG+#={(9>wf?dC3o7cR7>(wqHQb355x<!<?(vVrFZrkek6
zdBIv)!kGgCI_D$yyunDnHR}2Ez;V6SxaV@zEP`GyEC07Zpf~5~v*G_KQg|?gr8|ZX
zsp@N(Yw|W`>(%|?Kbb9p>z$s!hl=1s2)8YNjWr@KR$=_SxvDl>Pvh!CD`HNa_aY*H
zY@b_`7i$PS7*M6q2p<C06dG$UW%sQ<>zx863ob6cdx?@`&D_`nhfMh^D~-bH+S$3l
zkRamOS@!!glsFbX$&39jj`b#9_${|e0$0k^JJr)zlav_Cjt+*#0x?-7*M)<+0^i!r
zWcgIR(_c0t7x2|N^;5f=stsg0u>B&lSzGN?5;4-z`!xd7jPSVph<T`aI+f`b!t5uK
za?mPG*!gmxsLt|=H^^v6;mbj}G(mx54Tec{*}kg5$YgFP)YtA9=b};%UI&Jlov@mo
zUyMA=XPK5bhg_(iDypMekF_;V4}cq2Tyi#Q4<Yt0fDn5&FxYtG`Bn0PN3#4IpSkj*
z{KNDsh5ZKN%ORbitE7XQCN6qwj>R-`Ne9yE3N<Iw6Vr%n?<thd)m*)ZiGsjddGZjk
zE9w=hiES;ybt`7I(*|oZ`@c0AnGS=LTmFPikCx0=QWpfSi;Qf2Ge<oN^P$rs8x-Oo
zDtUV)D=6~L*<0VzlZ$K&17d2?C;k$6d!cUI&(nX1-6ZyB!RvEc8-~dm+G+}d7mn$X
z+gAe<wn=&#B*~ZGG0mAC^5ldjS;>YjX(6s4VXWutAzRLiiTr7dW<Gq#JjIiraGuGF
zO_bd_3Z(hN(TqA(^fd0eAX6`yKMRl?%jWhME1G-$-5`jtDb}9}*E4izy<~A8xN`D1
zsa=ZTB?KA{m3%-3*JS94c3Fko7m{c;PdYf_k~~CyZRS}k!NP&prm2ARpKHp61b#LM
zQmFh(-rRsP@t3`*5x`%nsPD}6MIUos!2j-81Z8*aWTkxD$-3ZD5AhJAytGt5ew5t!
zowtbXv^nKqG>fRvsXfRc<m~*YUQZ*yWKb^fwFAc$W#$ySP#+rg`GDW=i(QbU{=mtw
zG~sw9Yc9W%>0e3~aqf>t2Z{Yf@H%`bN5KJD3T>#bb#g%%&(kn-W1zu*4^c<T9Wv8n
zSDb58$txy=cKKqj?A`mxmoJ8{l0AbwT!#F)a4Z_9rs^#IlUpHSf&bWwJYt(U)-*Gl
zSppz_dj!slzmYV67&}p7tQoL~roZV>NYcn0l*O^Bx#{F00Oidxdp$c)dB_%JDDQZB
zK9$*AGA=03dtv8YAn|_%L7O^G|G(;n=<&1v+p+bmrx}3CqsVN<5Agv&J<`Y@F|{j}
Z>2m3xtLkl?K*`lLbrpzm@k@)a{|Bz*xt;(3

literal 0
HcmV?d00001

diff --git a/applications/imgs/sys_res.png b/applications/imgs/sys_res.png
new file mode 100644
index 0000000000000000000000000000000000000000..6a9305c62d529f87a9ab063f9933b686c8c63228
GIT binary patch
literal 291470
zcmZsC1yEd3lQj@zupoip?(PJFySsaEcZc8-Ah^2^kl+y9A-D_{T!II8xBszQTYuH=
zN7dBuX5PHp*Sh<hJ~vWXQ3@3a9|;Ny3ROm0Tonomz5xmf1`h-ed;%){r3t*jx{AuE
zgMh!UAhQVI8qrBw#}x_+8T0K0J-zDx1boQrCZX-7=4j#OY3yPSWpC`}WasE+XKh02
zVeaB;?dU+t&dkot!ANT5=H~R7h2{U>!0hN^$wGl84hIEA3MC^hqVAP_l;hcKD3Nm~
z#Nj&fxRxjhf|Zx1B2%3utpmX#gQme$Q4vwWtp}~GCd(2M;+*`c>nE+P2lViwh{2ru
zU*08<bG2CCmL$STJB%=$KIXbw_y>$-T;}?^1~I{!WGm(J-}*s(N5-;UOTLz~4*Wg$
z;wO$9-1=W{M`J_ttx*5r3*c_Z=thd}fBY&phq!w7;y*v|$_t$Of0}u_)v|Puq2WK-
z;Kk1;VEAL_e_8h5PhZ?$B$H(#`&}y<(JvZdE*imd6k+=x8}@qfKp#lgx(OAf{etMk
zSp)Bdn{FDda~EYe^uIg(U%$Wn)E9d;`x#c-B6^>5fHSZ>D9N+i{Xg3mKyd^D&#iSI
zOy6?dbN}B_e0$(tfW(=ufi~z|<!{mbd`4e5su)Nl)tnYQ#2%xXXyNs{#~4%MC7^7b
zzYQM7*U{+UB`WQ1d~1=dna1U8hUS7K+H2o$9DUG-KNqKzAL-5$7{n(o4yO*&SA5fj
zR(A#t%g*6rjnBQefy8H!puOrG=)*i5J$Y~b++**ATXVjQ-h5QUU0|MHbiJ>pq>NYx
zj$T|Ij{(DVzv%Bj9$whLV;y+Hy<-bR?<MqDD0tfa<QZ>{ckLr8Cl~he(#m8;3XMuW
z8u#%FOhr+~04Ro)uXlg!GSft~J5ajz)~L_pR;R9`ufSg;YVUp-T=B~#iqTj<ZdP3G
z(O|cW!L%zY$Hs%Vd6pP5A#(4B_cQ^T`Fcy}kXNv#1gR4D9-Bv*KxGK^O7I)K6NFqY
z#h1`Q{jo{u>=fMOk)-Q)9wZt(Zd#w9h~xkK52SGEjVg=zlXIQ=c%|9y-ENwA8+E@Q
ziOsQ^WY^d!F@#|-KI$j?eVk&{)||LiCijTH^o^HcCd%Ult+d2tic0bt-K`P#yeQrc
zwpz7^w*W9buMy}bI{}tmL<O@xxastv=@A+h<M3;bE(q7Rw=-6=L+oPcWnZUCe`uC-
z;-hw>s9KqM5G=iHcQkkiL=$9O;#K3>@Gw_zB9wi&=<-+DTZB$06U65mCalRCo_7I9
zmiUs-<m^t38T#YI?KKF8ppV~L@u3Y}_|j&sBaFd6&3>8h695%5V!Zow@+n0>aaWA3
ztnYeB5n?OepbPHm%Hvc(3=Whm3Q~=FdGQY>wL<bC3mC|eruRyiU>Y?FG4X(=tQJ+^
zucmLcd5sDTike7MTRG)SQh+eaZ-`&8i@2fO{}c<<9N$qnqB<dl$Zf8rt;`aHVQ>Gu
zXro5n{|7b5jJRe#x^;e3*S;ysWGYV|dguH1y<NZ~hv^sxi|5Kht*J~REEr1#E1nWh
z1YNoOkFxC4FR{Q~@zu(wZN=UDzur{n_?eAVBlcbxBdUSWm_bCEf{=A)lNwec_z}&~
zESI46v4FIDd9^#KnJsVYZQ;Y-V@W8stGzmyu`Y`?az&jKA-w*TFKvM}H*pDKPuqDY
z5L>ZIO$rc9G@H1JIPjkg=0Gfwo|iLMn3AG*0X$;lw@frXb4{2~z8d)4`DjsDI%}VU
zo`6x?QDja5z6dgsax&54mS%MGp7y@1d2fCeg`z8+l$UoBuenp4kDguFhWdaDtN&gn
z=3;|q>hR9tF^{m`xbN~;W8=HblM{z<RDy6ecsdb~>33c2rJwwi1`fVovj1dSYKac5
zKzzFWNDc(kQnWdRkUQ%8y)Tc*Bgi^&YyDtR5r@7OTCWK93TCc(%mJPKWyk#cB7&g*
zk*nD@&0><fdUnaY77|K}ht-~rA;&EFK|$%YH(~{Ffz=qQvZ3f8vVSKh7TZG!P<9J7
zWa|a(x9I1v$fuS1W*TD&R3187*Tg^D$i+N#Slo*`N>leVOh9_7drX_Oy)|vC7tgbQ
zw*-hG@vuAUmpj6eh4+HIN!8HRX1rwy0T94-GY`F2v3XX)5XF?h=5}0@x3x`aY~+Tj
zV-%yG)J2PCr-4c!7dH``#*6fpVIhVXQAl1bx9d`qH+4X8E+-C@Q#;WHcK%*K!uC=u
zX_c66wb0gBAjruhB|GX+R<Bk*{46qsnZ_Ju`+m*^@G-dQGgZ-ah?@LuAM3Czrz5`l
z-xY={co-SiJ!N5^*$90)YnKF(4Y3o%6joNoP|Ex}KBkHyr%yqegmh=ifTgiTO$BOY
z;f>7EvM+Rp>ut#%gPJph6kt)V<}SPpDMlV<rgg1qOyp9yX-ii}YT{}=nXiHE?xHV3
zqYSD<NUI^i7@+TJbRD}MVFV@{xn7O7A({<@#l*w}8zgF*b+^1DfsQ#jzNQT|6rS@p
zYv_gA@9ECqy-L&VT<Caz8*N6(UwO6c0Qp7y+NZ3+OC&YS60}A|M&BeXrTjp~!0?(b
zsSGr@zvu0QGG7F0g87?LS;xt`y(L>coy)pi*Hk>`#lOgU*Z&0zTqF20ES&cG+ISm^
zt>8<^MkXhwrhW`XpS9u2L}I#)ZoOyGMy!GF6$$sx3L0-IEP^ZRvT;9e*EQ<~<yTVH
z5*~%^%!#h?jb75L>ZWFiWR3b4_s1dG^dfCBK#ODgJX0K=w^PDI)Yy`z*|_r2RHU5^
z(a$OT*BY+{x=pHOhd32Y6DK|bHFR_%Z~BYt@P|5Qng7A!rMfynj2%Hiu()1-YU0nD
zUQ}^m3S0CtD0@zJrDDbe%FlE+8ojw~3EUCE5$dJ1KZe{iK|;yOR-C%F$?723jNzzb
zOr5We#oXuh=!NK=q2!bleaR(X2XK>>&04T)+vbFW6kOZH6U`7+WkoTdC-bf_!Z6^~
zlA(o>T2>n`T7J!ivQu)=iGr}AqEU<`Bt6!j^4cs&gG~ZVQFKx<;HCmSzMF+gnzCA%
zE+g9wVb8yu;Ac!It1kLoUYMETnode$!js8#1SM*q!<)LjR3uw=g}~`j9csi`A}HUW
z<3+P_2c)viNyvp`To+3kNI$TN(<&$G>QcNn2`qedI*rM{7Zknpb3TnkhoQ0~N&W{*
z1{(yLRQ&RtRxsrFldApWoc6#Kw@+k|*O#1&&fYK5^~u`(#l_>c1Y)Exm^rs_u$0tf
zm>3Nel4YV;QQ@^JUG<E>!`az?&GoavsLV19W0GShWMWZaVakYoiZBy5NfiA@gD$bC
z?<*l*+>%mfn-!S9u=^lDCO&1^#y#OCNqKy(ld7VE{g5aK_0t$TvRLR4fmEFg8rofr
zW#9vyIh%zQCv;DsFL;U;>xT?t7?+TOd@$zEe>et4Wy)7H&%ceI3U4Q_BUKj~oo?Aj
z>k7vrWkBM}{6bYIsH>#$J%P?rq*0Bs?y`(qg-=xitNtf1ZMN|2xHWmUo;k0s=G?eg
zjh#Pg2a$}c50}OTx(ZOfJ1?zL1d&PvN=GJBjtJ8rGNsVRaQ6Hi_)X2iMepT7<hC{#
zDkA^$c*JldkWwK*^<%h7kg8e6_#c~Jew__-p$X)dXc2n>kpv;&WL#5FAFGz62s&)B
zweI%VwC;y+50tV$HtO^#CLLQ3j;1KiggCbF^450DzF)AXX%E3WsfeOC@8p7*NxNBL
zdNeZyOms%L7N9Dy2AtabMwdtMoN?V^)8yv1tL{x6TaL%qQRy}%(#k?^J>9)poP0xW
z9aAg41uT)OWe1yE(}w$s&=yPhqB<_SG=z;gLDto_c|UOUuG)2XE}tq*W^b0^x6d&G
zC)R}!{Y2{#Wmv+If;AEKE)6*9OirxO-2H;-8i${}6=j;T)LNfcTW!w#-;Lxz(pI^M
zc1%YfQ|4_>Ja3o(Uc6sa4B6cXsph(C4>>eq4tX@Ra1t1KL}_^adqZ&0E=KgJIS0PU
zuZ2*oL8Har;h6XB67u%~wX-w?VqiGaIV*5)+xwXle+is#AR2EjjRj2=rYWdz&MBpr
z!qCO;LEcjhq=W6jN0fegbQRf?+)}BvNK8XRb<wO9`Pp)9WmC0BBAX)?2iNe&Ouh+b
zS;)||-HrCk%{!@jJonp4KvEM%KG<LD$cZ0nY$Zl0tHa3L$0}}XVCM;wSXVcUD0AAf
zM_fotE`!<2T0CWi4hyYo8%#=yBKBV3Y7%;|^^y)Oomjjegd%ou{h9uD&~wLakawk#
zQYK2m&DZxaP@?Yp{Asdc6|qbrx`c{~mS-p9XjE8Bo2CxN&Ows7!p1M#kb<h6gPjBA
z#t4r+mL}7psV|3fl}9amxxm`^6=#fgvt){|!UJE#rCSR6k)UZB=v{lAHCP`q#VH^)
zf|W}d$PG`2eZ6K_7HaozQ9f>LC^7jX(=a`>%zDY@v!|?|4yI}|#5(iibIKvdNb1^2
zz3NXdv{m2jHBT_swNDOh)KObjau}IOpmEUQw-vsVwM)RnI-VdRZ4tvn?V@0wMwk}H
z(aGm=_U7AFgV8_ZIa+iY`qk*NdoaB{^G?rE46X%NkjFe1p@W0628|ji74Z&38c=Yd
z!ir^Jv15{k?#_A)+q;Jw+9=imtY`g}<5l+&clT?zV?0<9wwGofS-Rne`ohYQJ(oc_
zI++=_Kvpi$5qd6H<Hy^@7)ys&p4jh)$pM~_Z?=xNdKr>MB3KgdPE=Z$Mj*N9Cfx#z
zaU3UWVa7aviCg!6Bl;Qe>g$AZ52=5k3Xe3sDN#k;ufATk4I|5#vf@yt)RRP+sNY$n
zipNw(pIq&ToSX9qRlb@(s;8%#J*2Cw#PIV&-K71Uj<PY0wvX}yhW(fZJ@=Kj3gT)i
z0%TzdGf#`QH!Tx~BiHt^fF)p6<K;&CUrkLj>+43*de=UFd?Ag{Sogee)b*!9;J7xD
z%x^Y2GOFd6Y;X&ozo0n#RTEsUG@l$Kn|oAh{7?sS4qVUC=en%wW;QVY$F#wG7l9F^
z&bo7x`;{sbha(2H+%#HnW7dae?_k?QNeV5WJ=c<pr6Ica1_aHuOO-JruCTXl1oMH=
zJ64?d{k)_0zlzZ7!YI%L3Ftfq&5$gY)6C()*@@K^-SobtHt2@6JBw`F&?*RJ@8QQr
zQ9P#1po(M?o*Z_HIG0R+qKcX8g;bdA)th@udkZKVLZs<_%pZm&2z#S$ta;MpP3e9Z
zbat*d`~VN}{LUr3R@(eBZLr+2Wu77FJ-+*(PNPN#=^veU+M^q_A35xH5lW^U4!p);
z{qk}=v2$=bay((c9FnBgVw+~JfATOv+82Xx;-748zT>|5*MGS|gZFn8n}u-R@e-sV
z=P+dSkZ4}BLAMfDFUM{oMJU$v;QAp@a4CkU>iy33Rh2MFW2Gs@U0a6*t16v|uo<50
zP@4W3&t6rI_}L+VnX75}RMt}&ez`=W;})S)i-AH^<99INIcj5?!FjA}(#mX{_;{*Z
z-MFF2PX}6En;Dk5(0`X*Ux%XcICzeK&A3QJVsn41%|UPY^=lzaCZ3@~J#Y@um~e7+
zH6}UHxG$Vbj11;-4^moR@5)D&K;SkS-1ZpI6eyEK4ed05XPMs{Q86V<5n0~i#DR3b
zG~d;^8hp~<-mp?w#pw02vL`4dCx<M%`+*x)6e-v1iOr$&-XfOogy%iq5vv}vjB`#H
zGR{XBn@__T?B=qa!XpqctfX|X^u-tlq{ewmGM4{B+AVQ`lr|`bw*OChb-?o>m1A2_
zQkRS0O6aIU?27n5zf)flwNgbY8O-ToRBSV8Xqilwd{-sxXmLB55jSF8_B<V5H9JUb
z&(0Mdl8ggF1r|~yNX!kl*4|dAgDI*EGro9IEI8UeJsnmollnK4v&<CA=UZlw9tFNm
z_(ynCeCeH!Fz|F)$e$<kVRcf;VS>={*qx%jq5SzcQ_7|K)5Q5hKeVo8KTPOks+4Wi
zv#2=^p9&!vJ)6M2D6-khN$><tw!2H*qO?kN>F)wj28KIX%p4Z4Oqd2r8BjTyp9~CA
zu&_zAu&hKtCb_BJe6D0p>di+e{QE*wAZ2VLg~1D-y_BHh8NH7teIKt{2iBx;sGd}-
z{y;0jS|RJ$p@_TmQY@rk{i9KNzE#;P>=^q%w}^{Vyt6DZ@w0=mwf*0U{84)AViZ@2
zSjLew^Ki!U0&<DMioqYpEF8hK`K5+5(PE)cm6e-@prm<~5Eo=l-&M&zpClQ&1QoSr
z>a8dL2KB``t9<gXIchC4M0LBzC&|>EJ<kDTmq60FGN{T63GSuXekL-yFY?UYmB0JR
z^+9b$=pc6E%qZy@;r=N99CY=NLV;g%;!M~N^<%N6DP`K~G%}*Juqo0Qq0oqUF`a@-
z^cu<PS91?4#;!<ejY!1VdlU7x2^Q|^`DKm20c;q+1O`(Ts(j-Nk|Obr5<|;Dlq`s$
zR8FL<-B<`_rIb@M)W<$JZ|6~@ODWlh3er<jLIW`D4|Tv_TwPyd!@COz2tc6_a7Qza
z0`LP2;1JzP$y9Uo@;~IIb7vcU2)rCEgmqZUaPVZ-J~lfgi;jGMvA!7njp??4r&(^@
z%Wd;o=8IjdQq-l9-cNQkM=KUF$F1qAM_peVwRtohIl*Vy`=cS~QG>HMxw9yku<&JP
zyxT*&-U8cnAet=dr`Qq4g4*}vktsVC1-hZD_j^@Be$X;R(phU?!%aH41CKLx(&i)q
zYD13ha6Nv3D@g>GfbX-ivIZGK^irO2_dkIwhMOK@-U(z-_)V~^j7nMBm9C45Rrp1x
z3$c(ngf@hXs{czw1iTk{ia<TQFJ4O>ng+G!doAXm{1RJrGjm}B{q2yRH9awJ{uU5&
zRW2gp2Kw+Yyi@@0h^d_u@9Is#Qt8+irLt(0NW$rN4rc<+c?j!=J;(gB?g&P&{q586
zzN^n%MM0HT!~q@{cu`nlWvH@0kFzV_z$y%Jq;Di!8-1H3%jth6^1zEN{0;hF4R8wi
zX+SWI4#!1clgg_iY4U#V2ZBRuYwMumAmw02g_etv`wHC#Bs^`FB3ZeiMDu`+taYDg
zfEP*xUYKNXBDm!0l8N?`VRhRcYu~BGJ~;Kalhk=%TEu=nptTs!l*la6_gFNSh-h!d
zFK~ZX%pCSpb~u{8Di5z`dMECk)4sXV8JNW{vf0L9L=811H6q|Xjb)krhXpR$Van{4
z75z6GOnULegjD^huG<i~yG@)ft5LEn=9Q*Ze<EYXmgkAXef@JR*WIxm+`gY&ws-fC
z0Ksm!)s|p6y8DZ?E|kE;a48bMWIDAHD5n{w%l*T{t-ug!wy$?*D{=*r`WjD{Ld%Ec
zPG1B@3>|h@<8^JL<ICb#zEc>lX$Wn~89gLkW1-`d4GFYV=yep}RgESS)rQ{i{^IAu
z76sY(J*)Sok3S2P*+j>pxZ8XE>p{?N_b|sGr%+If6@0Lgj3w0ZGJzLS_R1`l2>}QW
zhmPwx^6V81DZ(DX$0M=C5?B@KBCPjHlH*PeGO5#l{s7}Omd+M}67rkWg(dB`)uRi1
zlzjJ3jq<x=$AIr_;ij>?skNqqg)t)i7=FxO3$Tipd?*E8pDv~8@OWJ)b)yQ$AXUDp
zk0))_XXL(*%j*-lz9*4m9paCdH5e>*UxV6CTFXKETNUdD3o)c1p6z%sTp@pcKh#5e
z<G!XPZru(aE}2AD>uE|=Q(;THm0t;br|tD1Z>&MKQ=j>^GR^Ye&hlcGxK!5vqA92B
zj@EwXc&_PWFNZB42M1cul+?byJT42$&>wgYk>UKL;J_!d&(o6o@Hv%ncS?!`tXrvC
z{)gvchywbpRd!`*ss!07qjIuP+IBxJfwoixVlSvlE(9Vsbf5<;vT)vAD@RHpr1>h<
zAEZUJ8$RY*X9lZY&Dk*OH^<ZDeXRqnx}5bO)Sq@flJ`L+l~~b9s}g8Zab;T03yyZ&
zuIYucXY;yJ1Mvv%Xkj{UK{-+~P(5ml2f}8HurOZ(3B{O0)`gAwZS``s_f1*$+%w?T
z6W|*|v8-U9Ws0mU3n=4+p0Sh!@3#YMx}bbdJ3dLZUpv9j{X4GSo3eEoof->&9ErXi
zc-EtgCU10Dmi1i*{el(|c0{W%xFsAU^kXF8=7$jb9!;Rm8}$*abEfDDP$1dAR|TU8
zbm&DE&CNG|%<YoY7s{ZN*)XOG%?BKzPlsZqrWjp+1O`PEy5jNiao};Oudgqj;8(BO
zyyV2SHN#Dwl8%lXn+Bw-cVy#)nqdA~(QRJ5(TzB&UiBaW)+1yA4e@MWvK>W92!Bgz
zm(Zz@_Ut*dFEJc7T!9PniP__5%T{&sCbc5pF{SSEk>lmdA@$mMtY^*b-7ep3JzQ&B
zQ?1=1J9Eo6rbD|kw0$3Lyvs;2sciYg`pN|<Fh!)kM^hsu3aH%nOTTx=vlf3@f5_st
zH#u2pVbrb(98Qv=2!8S!hC-;JB7)SO`kO*|Iph%Od8`Z6%?}ebudr@Ztv1p9U9A_I
zSoLJF`m7rnw9~dTAMiKE@vEsyBd|@N481ns9v&XFn;eMJ3_S2Yt-5_kDm}Sb_w{;-
z6Hmb^gT^ng>xlk!3-Oujn{52~N_5HtaQ<<@1vBS6>l{a}2f<~bxbzwMQ3XT_wz?P5
zWBT&e&^quxud-Z~xCR8zbe9#!kdTi$B|>rk@UIQ=KNZvt(wal6xrVxlvu=-1@P!_F
zx?B%_luir!^S@hXP<ddGWv=f%Ni)yCaHW!|f0D53u%O{NZ4=s_RZ}vDoNryG-vkL5
zaWq;PzK}I?>2<1CugH!58FWXHjQ<umRIb`8qeDE}Xw}r?7h#LziMGP4;IBAy6dIZz
zaeXwhcf*k8dFFURxMOQO)1FG37a;sCEv?*1ujEcy-xH^9U5U$Bg4~9iJPJB1PbWm?
zgE;OkS(NL_H~xo;v%bK=two8&$YfnHQ`jMX`JC@WD_Vze2^f-i*AEtjRw!MqFZ2A*
zS2GgwYn%!h-1hj-&(9TlP47)n0-m-=_D@cB@3!Jd;)fRdwGapfe&|+y92B})FkNwl
z2&k89q%s@mhK^)?cBgj?cw9#K9pTv`HDU}6Y&nOIEQTlx{bvBQ#pSY5$MC2&tVSDJ
z+@WO%7#ShMS;i%es)!%xk5)`9R{j{I5#e{&N3y#o^vdf@I5RcX(~n~`_B--}PKS?s
zf*P1`bb+aNw57qJb)Py}RX&3q3~E`Jo73Ee1BS-wuD|CRFI#Q2zsGX}D~8Q(r8!}$
zNGi$P9R&|J1PfXew~i4_j264(tkwaZspd%Z11>}{208xScpz8*Oc_^-4K}+ew$?Kq
zV$I)7Td$i8FeN*~+8(EZ&Yza-{h3nNhyBtIJF;{MEG7=k8}P&dPv_9*!}AsYOs0!v
z#xl9S1C6OglK*O~-8_8uQ>VukrPBFj#x_5er-U1>-WM48frk87!4KM|)0c<q!+XG0
zk6!GIBE#K6{GTWcUmxDR!F~}Cy@>QuO>a9;Q4lY=Kks@a5Ct)}pNSTuulO8RLb-W(
z0D*J2(C7(5;=yV<fTGBME}pEaUKo*^OYHwp^hz7*%WS5m(?b^e<$m)$bMs#qAif|5
zCI*wDEYujk(|S58m8RnbVgVk91<K{_xas~hSOUclXX$GF^GD}N5d`OpVVY&Hqk@>0
zwV9c|xYy@(`0YEFTeN)FVG|Ro3^r4EsFdy-HC4;d-jlsvbA8wGKW^^s*ru}Ru=@2r
z*iav+<^KZ99SjDC{Zc8Vs`Hj9o?1C8Kh*hUrINs5t20PWt(ZNuXqgwoj$dFOF-(gh
zL*Q;g#rAo$PZ8QWcnYs+#p~Vok4*H*K@GYw*rTcJ-WO(Tk88Id_f%Em!HA#Mo!$`<
z5lI*_RB|z-u`nd-TIRTaOu&i<{}vIZ(4x1v_rd@$HAV!Z#D+^w4+)7^;QFd3iFGEs
zjW*v*i9&spI@Pc~Ee7=AoWLmRyn!CBLvyyNb!MH(qN+Hq(=1s4f=(1=zAoxy%<ou&
zKS-izN5?s$8hd%;=-#LN0nf$OvU++10JbhtB?@9gmvmZbQ#5{Vb*PiCZE4HfiVZ+x
zXv-C~mC>zBwy`CWOf+M0hKsYbie{U&X(0bJ^tPZ2CBF0QWxHD0+M2#R-Dmrrv=F)s
zkpcA$Anw6$U*igX<>~0?=s!2c{Ww6lH5g~v^?cKJV{_6qGZc?pMyaY@#GE7b@bPED
zs4^-HLF&Y$Iussi-s`j_@6Oc+go+1o*w3Ro3FK*X@E%!7Ge6N|h3>T{zMk4m2)%v<
z){DJ-_UByFwdeOY>jAIir$D8Fq3yVlz~0_oeVJlOGIfcO<$J_R+S5b9?ktC*$-~wD
zQRA}1TOz*E_l?8n$|^jLP;mvH)x2x!hVZkHKYG);AA^=bTV(auo%$7zJzzhLr3?DO
zFlbiny*!@oNeMk*#PRzWk^No1y*9NXGOV>%iHgu>HkvK;f@IKexF4tkYjn6wr3eae
z5y{Vj7Ffp=dg=Vne^jiXi*;gL-r)<&=6thdMs$2m-&-7ZvKUo`vJQ@*;0*taTGf0Z
z8(zIS(8`pB3#{lpzs(N{;4t)+mqR1hb-!C+8{8Z%LbY!&Ia;42F?fFzyv9NkxI!o4
ze|;*sGNiNOZ2o$ppR8)P+!#|7CzvT_Ic`JmUsv&C62{jqwbo}Ge;}H~u?bA)`Qm8R
zL-;>hfZb0k?(+tr#zS<a*}Y^^gers=tE0?YN&ycu9=tev2A&4>9ASxx!(R^Ex%qDi
zeHGJ{^|vfv@AIH44BApVZkE2I1#DA<sUCWIu8-e#y*`^%y?0tL{sw)9hGVEr>wVIB
zt$6}C&rtMi_c@*RjlUt^*WEfM(EQKd0W}zz#L2a_i58M)D;3L7TDtGDNaCgPFgAKk
zF5_JDlfbSWNnzjx7T7|-?$~DYnnG5(P=FvnsQoF|;_$t(^X(dm>TB<OhZU))vzKhA
zI(`281OL-mU;C>V2|Z21_U-N3=$lQV4wT^S2#VOhAM)<yFIS9tl<ylDzStQYOvfjF
zGaEBvE}e9d-7Z)VWCa6M=MnyqxNKz)qWv(!bad_-@sh4bQT5>aO{U{MX*GogDvMQD
zg66}UN1aAH9gHmh>W&*xy8g%YSLj4a<KaXqec$5-vy5Gq0Po<~SgNCXh~QWTN9kE&
zglqW#P<}eB&xJYty2o}@JE-l@vh(qzWlwM*h9HgL4z~Sql?E$UR<U3wv}Wh4R%wn0
ze43Fz@0X4W$v9#)Mi+Z|pNK-)89v*1%+U$D>|o?*3`!YId1<I#)&X2GES+*c&tpyn
z+)oZ9FZ+4S+pULHMfLR`=%~uIs=;0ByB`QXIS?k%C;=7kI`h$gM@M4<Q%9?y;0V@e
z*C2}G9~0P~WY#}b=*N;|4+{Y@wa~@30B01<0^^>41RYi-K}vcAitZg>mx!@hpyKc-
zWmLA<<_@t@!FR;K!An{4vB_Q+8u3pss+GeB?mvy?(cy6Enq6Kr+?WgooX8N1$>l2M
zEx8&`FMCV#Hr6mGyqA~PG->gna=j0%DT5hr6%#>;J|R>@-}iObhG_)P5jR=Me|(DA
zN|(h-;>OJNyU`)`yThPQP$+wx3<)qf*rsbCAHako(`&Tb=<<K47iC*Z)XPT@=`OU)
zEdi_UG;hV$dpBLUe53z6;;luFQ7j`%D2Z$8b#YDuwVXEKET5_^7D$Vtb2TQK%52!H
z)p>!PC@Jb|OVAsr$ibd8y|!fWn8!&m@o165P@-GT1gpxmOwznp)%M-j`L*M=AlXF1
z5+_KV)21=5{Etm>26nr^<?IZss&g+Qjq+*J^O=3qN)%=yr!}>p2eFE-ty|{Q-1JiW
z`VwiFaOwE=#FrE=gHb+YGNUpWg6b~g$9<sdEl+%|_ivyN6@*#N(+)KDX5Ml&lCQmS
zzU!jPu8}u30RG5^u_I$7-{qKx*2XyIfj#zBBTazJbP$-<c<u7R;<fw+o5QZJ`|MSn
zHiDMf&hR3JYeuV1MSfMql<Lqj9V#H5G$tJ+WkR(|&C0~8<YqltnQ3WwO0zkg0RZUs
zj#Odi+Y990jc&<jwX*1>ew1Z^D&o3bac^R(u3d+V=T0Q=aZdyf2~SO9CL|_Ohesg^
z6Vy?95BLAXh(`pq2Rf$p34|Q?NluoDCH7DmJu@2^KvXLm3zE$a=M1AUzV6oJoUe{{
zg(KOmt5nZVex{1u`fqFy%@(%N@92L?(D{7q_^!tP*+;Wdw;w2!H$NPd|M+8bb#;YB
ztp6pYRws38dO8?A7|?L^vQkEox3;z60T>GaFO+k^0;kKNu&SiAIt6l}R-fMml8rY)
z>jlP;E@2p>Wz<SCYw3H#Ir5mHt=%-@J)P04Z<y`{RAiu0LuRy3WOnj7HQSD%g`Rgl
z@$45T@$3r(SX}rs5&BMxj6F}R9A9MfUo23hpXn%b$7x$slr~NjF_tA_cfFr;*|bKu
zoNMY#VbW^^D9rIoNVUoR!F!7?cW#6+LIry64?|D|7p{n5J^3+`i2bE6=O`9kp4<U<
zmz+0zR<Ka@kc@sH%s{q>liqTgHw%>oR@_DvvfX-RtT<JcCEaS=hx8fXLx8r%5E0H|
zJ=;Tow%)i(s9fKwq`lV0^Ul7t!RSxXzsu}zxI2(?l|7Eeu#54FIfaAgFsb@_%Zjr5
zOYgP3zcE4HQ@m@GJGOui7&mCxjQAX$$^P~3JBj7V;7erRpdt+->Yo3ZKKH9{uh7>6
z5>{OQ*Ka~+*RP(uw?Go2Yx`d4$qDLh(ajF44a*N8ud<vS8og$o0*ol&@4oqGU~XSJ
z272arksG~Sr14&yQLy1Ee!i!CF&9U6%)PhJs}kOQ55DnBkoLcuA-Q}}G7!cW{;5X%
z^-84kCZuHrfS+bM0~I>;cl|8>J)LSdaIlw-U6Gd%p*-iU0Wz@J)<DeD%WW48Vk6{u
z+eQ5OD(#ZznSvxAW8oNxi4<SYzZ;D~5cjfu-wOzR(s$~C8xp=as|)-3uRf0v1rD5i
z)cJS<RSaz=c?tOLTo%YPb=TwSDNxqxLlep_ukOs0`9PhJ5v?{Mm^8}V7L7MknXrZy
za3$6WtlYcay)LMpzfWNbah#u0r77!$Zj5&)az;bXKP=SQBO{P}Wo=sXVNffP7a`}}
zn<3e%8)Y`_y+gEHYt3kC-(gz$dO0r9ffQ_!%N)#)kT$FwWYTv@(0o2XaKCojIjCcq
zBO$=qf0=TSjKG(>U4)|xm&$>^2p|+0m&zDW-}Li}JE*3!)tV}wl}Jgo?W1bl*|dv5
z02}r80KuBto6hFvk(n&h{PO24;_|;PviJ#T3fJKl&-JS-uGcI81|ib%gB39g3wq1R
zc1BVtWqN^#4|Mg>?q=ChsPiJF+UdAqzRqkIb9y#>>tSiWbKq)!hWB!e<GV!Z2R9z2
zstgXxi>n!>nZ?CnAo-6hjF2qq0mzYI>(U&_B4aU*Yhb5FmB@lcfszi2*hfZfVc=5}
z5Lwiu?nI)LtPUuVMm}A6_Rq0UI^!siKo*sjmMUY>f&U#G(DQe$7-`M%#PU8K)yvEz
ze`o1jK&sT!1?W^lufIWrPTkM|dYGP?QkiZ0fh&hxjo5TIr;CHsELo&NIpSexH&d#_
z!vB8cn)VwxED{UKo&p9ev|JjqIe;7PDJ7Qs!QB7G=(~&SHZa`rwZqoGQw!c0|KR6)
zZz;wkD+-gOgs<Sh6NaU(r<-q_w7cdq%EWkjAa1fAcAi87qp1vdEc*p@x|Y|tF3UN-
z^;i_qT_ay6sD?2g=~$9aRcBW#?cN+fVsDk>GD_5CJdw={`zPJ!Wn#^1E5^;u4P64G
zK{AFFi7lwun4Mu_x@9mLLqA|9hSEwL3oCV46vS<}&<zZ!_Z>@|inZ;|Ssz+cE~k!U
z2}f$>KhdmcJcRb#R>hXxOk2`le&1As>1Ik7f8Q?D2pAX`6zg7&mRD~ANe8uzn&5Lw
zz&23sAVujb3^+PI?#Jzm!e$Vrvp8xZVSkU2Kvef&%Vk?L*+5tX`+-J@JgKI}lng!u
z2}`Y10Sr(sv{4In=E%r6%;GR;0yiL<Y#(;r(lXXBGV?B&oN6V@a9Tih1@Ob`>uB8o
z+co}gs)$ToA`hnwiXs&C2YG>J`FVFAT0(w;^HA}MGARfKpg5qkTU@BDQ=NAjv@V0t
zzIG$=1g*dBLvv=*C6%%Q33;;WTUv?Dpjl&Qj$@mr2T5l${ikG~zRFsq48gXv_DAFU
z{ISkWf`(4hn$!DYyEoDG{$dA;abGu9Gq|-8GMMx|m#ks{V6~_fbEOFf!!RO-wBqDi
zQ|m7lx#Y8XrUXgoN&np3V+@cM1xlupC8P3rkGszU1W;Y?jFJR&1EUB7fQ7aTIH0-h
z4@zvu=~aIjpzK8<-73mwamPGcbVdTm_YY>7O>(@Mj3_A;Y?Qj?JqQ2UGvk^FgjuF&
zO86%LVC{da@HxYFHE-JJ>Pbzmr>RyB2<P~$I`y+uZMLNa#C*k<Yybxb9=8t*<i5V2
zc&AC%4|Fu0$C=!AzG`RV8woWC21t*A-O0jueR%CtTNPotygqm6o~$|s5hx>wfq!!^
zMN)AV0DWr55<YrSqdv~`8GOf_a>(}~zl(KHQ*^4fIONTZWjS_c2rsrgH|_anK>|)L
z703#e&8WqYc!i;mHM_Giq6G$zti>KoQhY|!SoS8`PRyPkt^vX%_WE#sW_fw^jlKP4
zJri{)OZNRzJ1C0|HbVrx2Jj3(+adsf1okLeWo0D*AwDSa_xL{^OOb{^%QDs=QQqnl
z-A~2&hpI+tlvik7?U&|FXt6xd!}Qc2<~5imD&DW2`b*)46W&&BU7(+i_lGbf^^n%u
zCZ#ZF)d3+*G8(sM&G!_wyT-T=NCTyay+faG*IzYMEO-;c^>DtgAhXm@F&h2?g(T||
zL*8$L-AW|3TNx#^yZjGidd0LLmfK<LQ1+}KMS*9|H}6a%bZreATy1oLD)=(og<w1*
zo_)Q>3#I=wE_b)66cCOTh0zwX*?uWFlIadf{D%S-MTsx)-W;a!^62wf^SQ$?Tkw2Y
zK?{hG>vX4^H}Vw77zT;QpE}BABuq8fP;JpDl1fmRch7e^vynV6z<~-tF#Y?MIm4e=
z9zP1boX4q%Iu9Jxd!0T)8pE4$nLCbDJMU6h_BPh+%ZnvHfFdWV>wmfO?)m={3oNes
zP}S}uxZZxG*8)3smdl}N3Gj_yk4*Erq4azjov)@&>6d2dhL0@KMbBg8x`s<%&pAK^
z=+I-lkAC`B*VX5jdC%EQ0?Tivmnh{W$jx(}3f;aVbp8!wU#@u#1Mk2uMtjbB?=VP1
z=?6lkUM_HR7_?aZpJ11kzqu}8d~+Rjbl957d*Z|k7ZP}~oU8Cn?HG(J@>_6vpXW~b
zY1Icg@8zoW`wg@uX!5HGJEtlwJ4%r!NuQLUi=gd%bsq>?G%>zFSsHIr?{KfIXiYF5
zb%hto*3OzU2;@=OAe?C&yi(KC;atGwHJt5A?t#Rbzow|zBZq<m=*cd%#uN)@F~c+p
zKcMJ2GVvXatPt=jk@WJ#3MBa5tVTrV_L*VG{$jJ9W_=@6-r`g$<31o6(X2Y!oR(wR
zNXmPn1jyvETtDgo+^XhHBp{8HK$iw?`kN@nY;mzY3^kT3u)O<#-m)6YJ(l76F-_k^
zrYcY-qA<`h*E{M{m$yYhyEF6Tzyhnv1is(>1sK)=iane(o>E5O7DgIFw04lV?Ghjj
zYkjUqHWz54A_~iRs*y3u8cOP27Xw*Cq58g|Nr4E2*TM@}i>*UQm5YC*>`K$M{3{Hw
z0JsuHgH|juSS6un%6X#zB2iF@LRRXxJGmithO=6G<A|+z$s*0{(R36#Y!$OZXjX@n
zwKvBY8yglTJpZQ;k#T1B<S=w#qIbO|&Nc^Q(Ae28L?agtniN*YhILxI+`pD#aox)U
zZlv&C!Z%yi3<aAFQ3M$IWltl;&yQn(ls)wCL3y<WK$+!J-o9?!5_)eZ@VA`d0Z`KH
zsCl04@#q$tw<#1IAjP;C4cpVh<UZDeo4&E4L~f<q?m{fjH~Z)C+f|x)o9f1P+f}*8
z$}Sgx^qkSC1t!FJWP7-MZ;T^KxzySQTLTEZdsUC8J7-Pfg+jHwU}E13XH{c#Rd$b+
zLsDhN;{(PA@DH_8my!KaROL=1@0}Dzqt0vzxwLp(h42FV-%g4?kRQu*T`Zk<CID5M
z%b%eJ;mKhfEdzUZVM(p{3?5TbYyE9f_r7ggAD>x&_$Y#Y7M{WJc4;s0`I5D?fzud>
z5fQ9^s~%5whA85;PaaQ;s#R!E#3WA~`Ur{quBO7IR2F?3b9bVAsD?wwNvQe)bu(*T
z!uE8Ue>dL#9!}yhojB^X>e0cN$odpFF`=01;jRi%MF}4y2_Veb%f#n9aev!cR+TNc
zYrZ<Q7F^eiJ3uR2Kp<wu@*mZzB9+0UmQNS{>ryHsB4#-xeBO%q=(MC``LTBz4*BDP
zWsj9Z+vWRMfg9~mBvR$s>Cg;E!St;-q0WEbtJ^PRKlAY&yVAj-J>8St;0wi;VJ6}A
znJV$evH=~nRE6{n;+v*>0t2VlI2MkHl6<W`dy_E=3!?jOiSYNP?Ua7RE_qwPGFv6I
zK>p&%DP~0&P(8z7pu;BW8lf>AA75SHVqXVL-A+}ndB|+?P|pP4+-Tw>w!a=J)fzv;
zZ(8vHp~e^@aI@L6H-FhX;7*C!*A~l{<9lKVC2V9AoCx2Lqth~Yc>yR*b@lZ2kB@f(
z-@X3=@NGc;UV@u87)JsDLI@O6e=0}(q%McaZ_qSnLxGM|41IEWkz`C^bv28Y4VCa>
z5^LNS6(lKIY$~cW(P3HtXbpe78A&Cj0RVzzJ>3m{avAP_9e&g6S5PQY%6V*E=H-oo
z)BOKP1`a`@Ypvx&t&lnsRc%r?j_nzBn}Y$Sy5?U-4FB`C>run}A<y>TjA$ahRDqj?
zkT+D){&HO{#ojJ`xj9|k%*k`Ik~8X~{aUfRYa}FIG?m}g)rHGp+kh`c=nShfXF;J-
z>&MA>+f-?u9wVB1wL8ugPT=SjXwxu4uzn{wX{u}j%anrH?oj>7Sf1cB5o*b5$*POB
zfxb?V27?WM8}&|V)(d}k#H8W6a`im#L0M7MMF&1C`|l_Q1w+k{p|BId?fkulVpcg~
z`d|s`dJv2>DpDCN*-RX7>Yi3r(6BG21gVeBk)bsOcW~wE>eI$eTl4h++;7daQ5BAL
zm~PLm7X|v36<SO(D>1HyQw#O>*+j1I0P8$v_*=rxW_--g@+`9LWE7Io#RulsxZE4x
z3Mqw@*c{!C6MVY(wtTvodOwe`6K>|ncc<L>8sj*1u?59GF66g46Tj%Sa5z^XaWl8*
zyM5B)*LYk{fwxUmv3<Z$z3oTSF*w(QOJ3v!nHoG7owpf?!G9+Tm>T_~aoHA0giqw$
zTeyk|ifrNP>5|`R5P-BhT#cy8g#MHAou235-;%8CWoKXa=sy!bA|rI3#{`=7WnRlK
z`7u^{6Qp(oHKAGBbP3%Le#32eJn{CUy2jz=aK6fIjA^jg5Q}5CTVNGm^Y<g*X1}(t
zu0Ph9sOBP_*YtmJX%?=u{-f!EMZne_(L(jXn<(pXX)80w*-Z9+zylkHs<I5mxDfF>
zPbWMD9AY@B=O0W95M-F2aYK>(?%In#S1UQE*pY)&l@+?Tne+^k6_kV=40m;<XT0y%
z_56V(yAy4R67Z<xmN_@h;((+CvqXL|t=j+kFf5Qqfur;F`E@W1#}Kr@PvN@T9><D;
z06P85rv-He)cukGhxZn=CiqVI2jWQ3te9K>*@h{r+NOVY#|P@S49VJxcRCZ+=1jY#
zA5$+#FV4cqtZgbI@nhio!ejE=eC1B|DW%n7=1^@KBsZ{k+B1)OWfRTeHyE)SEM`i^
zSDXh82wlh6q<>tnFEsx8F4P1{1o&6N&g)9Y1WLq#2y*p&CJzUVjBk>&ONJ~8mtksb
z#_^zni&j0bA@0Ea>9{F=j9}<L7T7-%@5biDwvFNI^}Q{tefzO5d2Ux+hcWpUk+H0b
ztczu;NRy@E2DVe@1_TCL45BI)?ORRq7~=M@C2ef6Q-(`?g|?Lfaz~|1I()=we`G~}
zi1ouUY0~^@UrFfFO)!b7aKum2V0?>Fv_CsKr8|SE>Al_a?brl0UYFhY)K6KL90OL8
z4E>1L$5V5c?9|^E(m@C~Y-xM#mi8p5o@ALvlW61XgFmwz3cg^}48%~C^l(3;7pL{p
zgp7uC-{($__>3AqX#p|B;ltY;k@-Z$P?WpTJdzDmN3VC&pd|*Tp|Km^<3AsBzaFvB
z&Y<*hxNlO5k|FnV!c)hre*>yEMWNfBZ6~`xm5Z;W05No2CZ&YV7~Sd5Q~6PChQ-ak
zrEn%jYOyg)5IZ{3GE1YzPxTNa9KU^|sG1=55BZh7?fB{ojw5cG%z8l+oY>eOa3v)A
z#mU9Zg2!3x#+H|y_KrP)^GHA>;RgjJHm<@oqNTzo$G4oJzuYaaV<5HNslq=By$X!>
zp(*CEr&NJQ(Yt5jgnC2}h{Syx;1aAw>G0N;Bxxx#Fi^@Hac+MwIY?Lp#|K`Q&pZEN
zsiKs@6uj#}V<(I27qXx3%ziD1!yIIzkc$14x=RNuiGgdLM?lBBVK7*9*&b^2d?BGA
z7EMX<-UVWahq0>`!YSi@NcK%dg-sMW*XMxF-*v`5P~aA5JG9F3xARyO*Z`K6S}{E$
zdD>?OOh*PS`zZ|V<UYb&oUxK7WE#@|((KX0B~6#XW&6=|FcyqSE1pOm({OTB{>RGJ
zp6a+-1(kx`eiHN7sLnBY#y@@oS)|BM%IhuYhzerVI`~~Z@hQm)nQNVI@=u(u`OEEv
z$fLo3={_d*c@7;rXuqeV@muqF_Xr{V>vxQM*_XFZK0Qc3Q&gVzgA_l1TxK;+D*0uq
zPVzY>k|Bz67>Obi;3HI(psks7B?JR7@x@@s#gJpr;WIQnzUp}+0rKOvHiEze3MpNR
zZ~8sk{CU~u$=}oS8<-2RIq^LNR{_S6tt6rjm@c4jv9yGwEUpQpeM31rz#<h<7Uxx#
zt##TECHDDOK)wVctfhqmoWvuD|J114urZFqsI83aA#nFM1c*S5w3mQPLL`yWdrf+A
z?Apv$Nk`jqaQ5QVA(!BbGIw(71kdZ!?Ip10Qk=Iuwn$#jVcrx*z>~?=?kD=t3-dT2
z{4se?+cXli?p7l&Cta`pP#>9ee|zucXoW5U+-X2xG#-g*o}b5FecrQn?V9toc2{Ro
zlA@2LGcn!j5-m%c_3kDAL7AA9J$ANJFJd$<8w?q&Od{4TxTz`4eDq_zNk0-0%fC;W
zegP<%!ltHqA34N6zuQ$KS~PG4o5PC3(_z;OgQHhpQ=nk<)Nki-^k;6jqMtan6m$}u
zg4T+=7UMn~DN(zj9~5yX1*@KPZGpo;W7A+FpvXD}RDA%HYlvHqb^B9&{gtCI2vrbV
z`7{^012P^Tf@Z#IJ=BwU(Q0)s7`z-IcLA#{${Fh1(u4+T>jbs-kiv)rbGfy>NX^%N
zLOQs)W2kCPbQQb;U}>={Z_1qJ&trSiW4@-x1p(-Vnsc~DyH!>3U}2E$ue7drOAD+U
z_7vflD<v@=S##1<IxQZk1Wk4&b7eS^P1`Ll(8S(PEMwlUixR5<-%>ceBzt&fJf}Q`
z>p{{+b7<t1dosQnO~_lv(=?Yk+Vv_Rc$op=TvyOI;46+%ygV}^HoaLqWoQig(i8Iu
zYkpSO^*Z>_j@RsDKWwA*5XWhd$n&fhnZ@Z}!M6)Qyp{X3U7=l{nUBd~m#kowZjR@B
z%8_fCx7=NFq7aSG73Jvv$R6vn$Jczf`JMN$qQ3cxZ+%zKwafHW&Yx)hTg#189ZlLd
zzuELvdk4#$+P_P-5xPIKT^_7~b1Yx(9E$Adqc}Gn4y&ak$!I$EX4@Sx`@J!i4>Pkr
zWwCK1IGWh~Cib)w$=7yzhH+>+<_1`F@5`+9F`Uk`uuV(4Rl?ZKQ)ZF1ltEqP9$s_g
zQ0RU748Cq1*QM;0;ZL~#2>0HEcPM<3s)t||Su(g^4T~rDH_O$VN^#E!^o>c`La+Cb
zU$M$>6<3=ES>UW;LqmhIq6}RIpC_}U&rBB2N%IyUvI0bs&P&#F-~R%ZvLx#Pv}plQ
zxB_Dd-484AoOQqB-APuHo=IM;w{%@JPfIYz14@GLby884zwpAPJ)Xq2TusYCM<E7X
z4vf@AA69qe?u@44csCa(hZ<{#in*;5#7Ne%x(t5KE?o#g6O^{Jq`}d5f`$Uhx?k4X
zy`|~ydvQ9Ix1V7Md<N-8RvJbN(2QbT{`Kfq4vc?V*AIP(pY{oEKIApeYfD`&{Tg6!
z*F!0z-C%=_W8j8C^7$5%Iwr|(xizLSRaWqRmJ@$Ja7!Rpi=ecbUTvZtT*G^;8w|>^
zkS2=<t$A!SvKNsJmu(kB_Ogl-0HiNB`p)x!@i*d;cWC^=WtmZjb2iv1Sv2KOIdSTN
zimIEW;xu$j&TvJjc`uJjn6S{!`a1&zj>U8HQiOw=*#%rGbW=5$@R1X)42Xr@-ipD=
z!E-jz!<fOWX1{B=3x!py=e>7-(10kVfAD-6IMiwPVxxwH!A)w~i_n}HG>jzKU{SbL
z70+w(!oug6OM^jRQN-c8emP)P+E3i8#mGRR#VU?BL4b(mUo{!3ne~YCU^_AxD{_vT
z-vE8vmbA@g+InQV|2sy@{#`<glil@5O8Sbe!uH4XG{4K~Li^m26eU2r#=e9L+t|t6
zZkSm@ONR*VEyfE{{iUG?xY98fVsHYIi!;TqreB5V@>%TI)}-i({m3{NvJ#m#njY@h
zyU?eCH`?+Rd1=aY@~H)(r0PgHx1|NDM;s2>$4NQDHKpH{9ol00zkkGlhj#QH)l+Vn
zPhtx%&h<sHytx;*t?D{yH5r3VXY>1<M5D;a|89#_eA};Km|d41+sD^e=D=N1@J)O2
zMlS2S|IxVNMGo#=Q<HsM5x>7wIX$J3ODr@@3)tfW#DGRdBn+SVGB~X<Ejvy-eyfB^
zSFlo|h&f3TtD3mE5;Ath^=o!{Ee|jH6ME({I1^4UEk&_~12Q&xXXF%FxOmP<Qd8A0
z)0DO?_Ivt?XaRS9#dKc*UDDo!UVvPx1wh0DRO-9#11>aGTM#SV^E2H5&*@#D{c<D5
zVC^_`a?o*;OHhO|ZqnZGNtV>`<@zGve70@XSrE`fvMQjwn4X`PYS3j8HE)|-)@yQj
zBXWVD7Z4znlar(GwGn1j3BWU8A7X>CVkARkM`ue1{znS{?#)`ayDWdZOG5hGrokr6
z(_3iSigT#0!b@_+9qKIv%~k42Nl769^fsV7xO@XsfN9v5C#xRIs5M*DfS5T8qARo5
zI7sZ<Zl|Zc73H$<XLm)Hl#TA|O)tb{&3n)JZxDh>bOE6KyT9CfQ$__m&j&cq=~z1V
zz~Q_}2b}MB(l~7ADhfMg=>SbFTsMG&-=f7t!@S`rK$8M8bn6-FR3^OvKwahicEZ%t
z^Y+#9bo^Zf{lwL)QqB%@$IN(PS;Z=8L||Z4p`?m<KT{78`yyw?@b4*}0WSO^|Cj-*
zC*DZv52l>-NH**?h0Um$Wo4wno6~Ne6zavqPFFp7<@{dD)c)*oD-NuWea(?Rc~n~O
zl*{?-KJnw{t>4<Mx0R#gr_NF1Z$tw_2^2vu9WN6(eBOZMXn4oPVho_V-;US<yhEt5
zH2t4ymWsia!ouLXI(B+``gA~)w(o{d1`DXOH#WbsSR5_X>bne6wZA+ZW_#~uKvihh
z2D^PG(&#XLD0s#g_bI5Bzzqk&X06o}IX|SRtt~4y_B~kaVNMTjc7I7}it|K6L!)@c
zdSrAI5WHm;u*9l0+X~lMNjzrN(?(R$U6YIA+3AfNA=+W5LlF3%pGW|GB@b}00^!SG
zERni~Muw#p<Uapq9w;He&rV+;JaBXf<pBbzn3$N^`FRfiCodl#pSy8Kf77psr~`o3
zxR2xosvFNT+qjhlT-?y0p+PJW7>(XLCbnodH%6r@yR@`qWo7l{?rg*Ny#(hxKY$tD
zdIrybpmdZ{atd%lU|=|aV%|EXdWBI}E`WzF%8j0p0%w$ja`IlE_Mr5?p7#Ue(ii<L
z_EUcnS8;N-=bHE12=QM{ihbSVTi*aSig|A+qQx<6^Zjm?POS+Xg#R*{mxai3+Vq$Q
zs84R7SzB5CedA!#x32a+t$9%Z{dp59QLFSg6S`&-IJPGMk>YYt=>O4lmT^_D&DW<J
zL?opVkQV8Z6zOh|Qo6fa4oWu&sDv~MNQ1O=cdC@M2uMjkv+w8ge{o)%qrg7w{kvvn
z&3CQqX1wc7%y)TZ#cx#{_napVOM7~rF{@;Df!?e}i1yBXL>NKIAkgkYTGyHUmL87O
z&ZK&sC-rw2^Q;}rYJbtB8qOf)l-bC#8;+kcTs|K<jjB^+c)*nlHLH5`&EEIGQA>GR
zf%1JON)KWrYE)j8FgmmDD3a!*5mCi^-LB+(HiV!+V#BtDFWP|+4N(y{-+QA}dhHqR
z?Ct`X&Tdd`g+K3*<{>gHbLUCd`DS*7hYZ6)A~zmBiIzYXQ-(5d#_`1OqP?7eAZ~1=
zXjc!^(bY{-Eh;H5FKlk6TCBq-Yr_}|SbCT}3R^n`knWLw{P^MXL5?SbF;O0|L?#eg
zMv^HZkDy1W)s3eM_QMTZu_$H&_ZC*7CmpWj3I%Z*pT~VRs!x(<JP?a7EoD<wR4gtj
z$*rlGQfIM~<8D14w5qPDt%d$h)!MoU+h=F^N0aBS&|h_yM~~z%68bGzI91pTOPif-
zsH1gjMi($p496<ISmx?e5v9Evy406mmJdxIUcttFo`rTjnnmCT?Bm_hBXU7iThS7v
za-Qq&Nd8-6wn$IP6|@ZuD3E87a#3o)Rk4mPWww!bPgN)s-ey#KwEgBI0zh4KkrFpy
zRSX->jdP44SBZ<Q2ONVhriwKpc;t6qAv4Q}lZ8IiROHEs$yMDea?~No@^kY0xANN6
zc$Sf7i^Ko30hS#XwCr~(9kZZ5j}1{|is)jQ$Ca_y^YNP)8&jL)D@@PL*-#bWQwqEd
zvX}hnO7%1NHW2zK1di{}qk_p`Jj+jzt|~A#qTk$ExBN5GQUV1QQHd89e)sR+x3RT-
zOj3&k8W>OL=yR^#v0Rz+%eGtpa6&w4k*Vi0tOdeF0pe;<b-b4X>nSE5zwCKyU(9~T
zL&A7g7OUfA)>MrxwX=$2XJ;qmMl$qfxWRQz2sSbh=K$IlTP1xe0~Si5>U+CC(+vb4
z#p5gu&?8B9XDbBHe!r#-viU>ZQdka;pKLVk%Z~Jso_-nq(ggzP1oJo9gY?fg@A<B+
znWY)$TVAm92D;{*80nF`;l}Y|O&84?UTq*PUrTb)nyXV00NXmsyPE$T8=DWV?0+xR
zX0>}OFl!3M-P_)}`$LW^qgAZ`Y36C2DpJtpverBA4%>W9(hp?IO$vV-@qYZEP*hS<
zA&ILDpLZWnr{lm4S0cBaAUU!VH)019JyLo(2%x}TTa=a@)9_-Uz(+17Ew4Pu?{wZe
zcI0<EBi~>ZR4g(iox5KBOzCeMD=RAEwb?8;%<!AB=*ge>@jYrjKnV;CL|_R|8Kh)n
zP+|Q21=7neSp18}t&tNxOq74?Yr21S;-_w4pg@QQCt_gf;X&X7Q`-d$EbNAxrffI@
zHkXA6mj3k!_E}X&^mxK9SqyM9gg~yXr+0L{ADoRiPY&05-2nhPSbr4Un6t12y9i5Q
zQ(=5P>cFJswj3bFz`x%mOJ>}aC`Z!|(0g%d=_&#(q`g_8Rj%~T!KdH}2YVYVtYsqb
zkf(UX;Y-%k(4-`x=)#r_PQ+KQY5d;USG3H8gkEo!hkb-&@gW@LoA9FwzJ6T{D8I)t
zvF#yMsc-7t(u2h8VyL6j9g0cwW=#l>M(nLJ^EYsZ;IM=vFDx$3*V>Z-|G5S)?biT%
z1>fdJWc8L!Lg}~HBe&W49eE^s?$)T}>=;3XIK@w6qSg2AzS>HT-7nur)iYTEr^VyZ
zvY$phXWtJS3qE}sYM$}G6{=hpBAA|@<}j!VcMe#ijI!*k5`wwK2aw~~XAMEErhz9q
zNYM2YAh~~6LwSbrC;Q|Jt<N$t(41Yjl_I=&GA03svVWR(xsI}g5<Xs6l9~8~nqIAv
zNd_|<JqkV_46Eb%`kUwGJo}bHJtvLk!PXgl&1n8+D+q68h~`cCy+|sEDes(vv>uBg
zAP~NnV1IFe4@>rw5uN&y-Zf~SyQ(-qabsv|YWg>z*ZLx<8W{mXJ1lMh!Xg4iFe>!n
z-$W#!N`rytY?QOymGoj@EihO*_Pv6~nlIZZts8;ATBaNK6r`KjU3}_wbk6ceZLg<`
zN2yyq#G>R2Q)V>AJl0dcAeQ=^NAlUmzbKGPl9L{-KfgUXy!DbytYC8GQoQbE7Sfk;
zrwJ)>>UUGfO%^;5*#M<oMlFx*$@_V?AD**^S4X3WQQWPO3Q@oigm5pFJ6=WGjR>WH
z!*ZID)8p%LWW&9&uJ@MLey}5ij|y8VL-weakrAR<Qbsxjvg01~2Px_4l<G>4hWx4#
zEh1Rd;3ywaCCRsSbhMoLWH2=~Ra_cxa|EvnQ8^p83$x&e&vI&t+%5ka=OAE|;7eXK
z{f{q%)GX{{og+fSe-?YcyQJttR7#SbX<%w*=6%#EreuZW_a7HWt*Sk|y?NpH?)7^P
zu_R8j)X>%K0sfuh@e4FkPc(B$@pPv5idveQ(p<B0<kBL>Vf$qN{fs_sdntYD%qaUi
zkz#8ITj$Ty5j_w2jPNepfPBkWZ^9=FekFFtuRQ4ZVsMb!_OYlwJz%baK2@2S<cImw
zSwbuw+0?_oih67lOgInjzo$FsAo6K_UBfB1y=HlN5_?5Qm#D2OP3RLNVS`w-TN>@E
z(imL5#pe4A4Oh>J3LCctdsLicdxQ$*7b3gzjK@bVDCXnoqZGqjNO!kY49f3~$05bV
zK9rGWG+6p3$1;GUv`y<9I@%kHkEHz8gHV-C!I`TJWuf);S}D^m?7KA`|Dz|Few8AS
zyq*w^#G21{J&Z0MO=}ufLqgr{i%W(ymOd!Epif`B*8A9Q3hN}ulr|?!Z@=xfbs*vU
zT<6StQM`m-3#38h%lar1cvxt^z9B7s>3@wZBw$WR=P-C+!XofAbge-jJ3XLSe81mH
z`}Q7}xMoSwR^#xOFUf3#Wuw+(2_1Gpa;tLEBjsxh4$*XYI--G^Cl3v0ce3u6INLp4
zh*}uMMM1~<kKVwPi4tq}@k`^SDJ&MutMT4+7sR2r>CTv}(4+Fs?2qGMUEgbDA_^n9
zsm6tVw3fzCP1eS*Ti;!@Y?g^%77x^eZ-Ce5zzk+8m7K;eUlIxXd+~Sm8+WMVvc&I&
z{<mg`CD9c((<pf=W2i|i9xZ#x&VQahmP1HkthIPIvDJ|9hY>+qB@26P9|nCRRlwSA
zu$u2YUDMaUQ5f@Mqw~Lc9U5?An3N6z(IZd-&Z^jj_pf-d@}6cW3*^`Esn~Wt8wk;6
zlhV@@EGSr)&rK>J4lFI$`XVFyP0Ql&!a-&KGjb~+UVovULtBLC(@jfI?L_*O_iy<n
z3rf)Wh0>In@r`p;d-UrZC;+`}%!u}wkm1T~%Snhe+Zf7V{<knajUY9U$i;jr$5{g|
z_v(>Ta|%$yOSGe)<dm8X4h+N;P6b=#;<*~XU@EaIKkakhjla2rOZvS(d>e;~61YBo
zets*l@~?kH5Q+=N9o_p$xTG9r9VoTTJ=bXsFi1Hfi9s*-9Z(9`7X$a`dwYG>czbh+
zcqS3D1JrM8aB3hJji(IerL`V}N6`9lU~!vxjLQOH^p^1}6ouE|u6PXZ1{*s&0AID|
zj_+<1moUqB|90I|s|+}IBM|>*c5iGt^x*GbH+THiP0Y89qSprG5wI8ck8j-`)hmCF
z-QMhbR1EE7Y3TstOxO12T6;!G#L{3`sVw_~ByW;Vz~2d$)qN#}H+0H#zY;$g-qH_#
z@QKO(9{4)<8sD-H`GMS6lRp`~CKJ9P>7w@^B)SY$4dLRb1o__YUg}RhOKU_@=*^FB
zTzf~YPO*E-aP%0#FY#hE|5ZwG#9tK?pF{MxY<KA?gWM|;rE6+oKjOtY=Wn<HREtt5
zqK4i{YzkUW8ZZd&e$D#zg(<obOmx@`EvBLEcc%Q-<_Lm=uK88PY@Kc&=e0={6Okag
zzcjH!A@83jT4>VmlT+*h(Dm;g$*{<rikbxrIQ_I3F`s%nUgKGld_-JZE?N1}^I^fd
zC0=RC#(SMItVHDukyOHd1M6&+QpVx+bkmWNqHL>?JxVf@K0&T#LWxrOdQPLq51yt!
z!g>{UHX^d=fB+9#PfsecT#+Iz7v-0~AC0d49elLOZhBpsw-MiSZEt$(ck6uY(_M}&
z&7w=pif!?RiNoDt^7U5cAKm~{uRnJD+OfGQ`dE{%Te_bjc@a1AJ5*m+*#(cE7P_(+
zjz{U%gq}7c)gt}%l~IauWH-6o4?6qWeD=q1;?GS)P$g7l3y&_|UodP8XYvC<K-%jH
za~U_ClE{0m;_$JNoydnD`ru7&fWZnOYTSFMY5^ZDFeLp4b5c?Jm+6oUujaM&eZ+8*
z!}EA;LG$9RIe8U;x!<}g%H9tQ3@DT6iDwJ$S@KZ8(5HO58AmNT7kqmS6J~>J!dh<|
zj4s<-TR9~qk<ju!lNI5(fBY&a=U_6kob)ho$wGAf$QOt=Nq2W%05+iWN|a}0;pF6}
z(4BW6ZQmmQ{j(+O^wg`)Vg5zw9aZ6V;^`tu9oqb4g@8wrI3(ffhQ>`e1u*IU9#+RU
z{BPej8HH*V-Bd^4uv|u9H}QlR2_41s;@4l{pxetOEmPBYOM{^&wp_vzat!Y!4Ws;N
z+SA36kz0?yg#YI<_3Wo=!#|#<bO)VCw%B2Ge9Lu?Mn1U%5guAxop-}gDmi)1pHD3&
zoNvAxmfPLw|K<M4-Y&oOO9xu=?{NF7>%XD#k7P`YZX!fJq{Q48#aQ%o)lzV4wYFht
z=5EI&LCR-O&u1bgd_b3|dKU*P0S;awv8pYN`~sBcMen@W=19|AD)WTJ*xuz-7g4Uv
zy-|1heOR3;??j%!rIkwBbTOPxf9H{3_{N46xG6j)6vFS;2URf+>&%*b92wU|c4Se0
zx;1Py2mVCg(U;rNkBXN~`3mPq&-)Bi(()72)Ysxg)L1{xinvdB9$3EFU++Js$qh+M
z!{__t3?oF(OdMjImS-)ls3?NN+fyZ0?B?sk3_<4)hP?^lC|7iOhb~iB?~k4&YkW9*
zGQh^(a5cd017Z!lz1%t-kD9ND<Y)qkdGZ2k94RB|5|s~I%E~R~t1Lm4x=$@{ZWN$@
z`<POoWXCkH<tZ^Amb6YR2}bCuOFSDE=d8x;h-crk&0WTO7Jr4cLVc!dXWx3NSImo$
zpKpDPN8n%%j3@Z`RQ#40zG6I~?IY(6UMY;`xH=RH=CZnXzhBtbBS+V>d=Ey>pH#SN
zf0#Z9y#?tunu3_xkM#80BhTX#3_^@>dMrwd<L`^GVM&|J>StBCL#3#wogk6DXVDSn
zT)j1(oy^7ltuCjmIyNeb=dW+Uu(pQAGxG4n9(F9&3*PYp$703k(|LpE6}qpK3}qCu
zFv^Z;o?Gu)R~w-E*kWMLdiGmYqfcvP_rOSroa3>oE06_RTWV92NWfB2S=nHSZRQVT
zv+Tgr$D7MdPEzP0g-Ex29tyhovv05&%L02cJ?;Pf@orPf*s3_4SWb0ZYzArLlSxY^
zG-Cb#o;A6t=sm)p3W<8S<LT{PZ9mHuuw)IBNDR%B%)&zEH91!N7YOS#Jl-1%_7(hl
zUF<|Psi~=lpw}dT=V@|grUz1Q2BB+H1hW(ZY*0~Af%zq?wdazo;wBJWDGflDLLA>~
zVO{te@U;;u(prV~lR7h2$}dy#p^?+j=s?(a5Z$qehzKmIj*w+l>io}~5J>oaN=nch
z1RT4zXL4^paE2MmE-nts=!6h|`fb=I$HvC@x!`jLgT(%q2Xh^-+Rpj!T>pA_ce-L3
z+H(;3Rdc<D^ds;o1~R?%c)=40=hmM8J}UR++*)NG_Sl5Rjr!`G*VAz)^F_PzdfDLB
zawatA=2H<<u{Gs48%cCnK8NRI|N7N}UNq6D<&i&eBYq$k^Q&dr9$Z`^nu<S7k7#-W
zSM=9$pG{gid*9jrq~AZ*ArsS+m_juBr{Zbg&e?5`vG<RHCd!JSDhHkG^DNAUlNlqO
z)qFhPoAXU!@HO_YyW`P{zuWxa9|t-Nv6237ryt#YsL$T!mQxxHF7Mn&{19HNA*AJN
z{(aXW4Y>!e5>z-TDd4pL@szyzp=-A7g8%s<>)Tz&#hx#}!-DLdy-8LR&&a<1Z5=}^
zK@*anl~YtknMXeNz0n=P6P0}a++F85(N8?ziic9b|4A?XA32$g^tdS#*~R$xM;E=U
z$^G2WYMc0NGXX{x!Gz}c)r+^GsptE>T<xah!-0yH&dy>a_80%TN;1`BrOb%2nd<J}
zSrwhuU(rRBki0!lL=kaV>=}KH7DQ}8g0sxZ%4!Zn0a_!QOEt3*FN_nqx>m(89ytyG
z^&s4}J&i5!IZ4n=8eGSzLgrUV9y!Pobf!5uISGIO(5dO^#~xeiK?kpB59VIRe=+}t
z!IOap#ii>ku+Hg+qO<cOgu4=?xt|~HltgP>sml<nqGs6R2*rH^hb^Axe;X;Y21QMu
zeWodjRhB0Tl)$-b1r9i?BlAT8nC9UL)`ZEH{xy{}Bzl4Gswa1nE*axS{PXtVL`Ygn
z(jD=C?J>^o?$yYbeEct(e?PZ0^Jmu5jUNV`7J;R9P3xJeKss2mtwoN2qNdq9XU{*~
z?V95^-a|8~hV<cITHN0FNxXUU+1;09Y(PopvE{>*WI`Y0Xy8ym?ODDQFR@5MIPJb!
zQ<buaNJpj*Q{Wf2AAPwyhh;%{KcRPY{R<=GJ?y_39B7Ibg2xvfj$wpKVOUHVA56lE
zSD7XC?AR6xg{$K?VT6>9AQH@TEB*YL2~P$Uhl!djEa3z{c1~K|NU!L)oW~$6Amjl@
zSb(-Aad`gsaBL!jG{+13MS&;ppN!jMvZnq~8N9G^peOfj$5Pa?{n)>5NrE){V-dGK
zEjBB`Tz+rv3nR7cIh#6hJPM0KA)>pjXuT-KF^ouRbXWUzK=GPECh{1Sar|78?{$fj
zk)~x7hLohGfpAvdW0kqM@qvk>GA3f#XN5(2+vc*f3KyYne+r%oI+SOp2ua&UJ^8!%
zT3&?9)6)~-@k7Eg=XD&=rV09Ijv(1z?^mq<dl}D*EF!0GX2cfFYq#h*{hVDolG0az
zsh_xLRFq`t3}2p+BWu>=`W|Aot+Tv`p?Jo2JY6elI}(W|)t<zpgalm+ow;c9O_78<
ztO*GH#GPVVCp_QXxX1QjBj;`=?+$`G!z!42L9Cb*XaDDpoPHl&7irj-)-Ju2{DiA3
zXV%zJN1cl!<*ciGLi-n{$eLFqu^-E)CZ@9Q4^7;@%G+jHFikc<y|Kd$9YYholH8+y
zuBTCSdg3=%7ya_vA65P1gSM4*ZO<2MI?o6%eCACuvjiU(tb*_6k8e^xksC$y#01-g
zp9oSy=f-~p6_Zuwlpmknd6?C&rC|`ef3m`yit&Q*!pfgM4=ePI<~S!YelCL>>w>kV
zE9w9Ajg-_>4O3I+)Bcv07Dgg$%@X3T-us0$H7WEm5x^QKl1BhCI#nsWg~pCAB6a)W
z6#Q+`e?(`0=W3Ot<W5Yt{#QQYMg<iJ;UKY=Oa}kV?AXRaon!GKX!k?HT_*<lDa@`a
zs8_-L(iEg(Pe=!fG9Ld?msX@3(DQ{VE(|jV{_qMcNc#|BxaG!Qeb!U;-a=Cd83<-l
zi-$o{5|aO*syqgpE5;p~q59Sn&Y+9QY=(mTs;VjpWahkMW6-G8o}6e)jq#Xp7F1X&
zC}1P(dGJLAO--45i}lrzm4f(M@HfD*_T!_PE3Gu9c)i2Cw2BHon78t`-9ev2_-91D
z|L{$vzy^!I|23)stB0wlX54NQh>Ou><xNo_Hoz%?f#@1RCtHCRY_-}Ii5@g3sVHL{
zDnQ7>0Xn?E<U+~a2=sAN_eG^Ay`^e00Ryi8`cDK8msPwaO&BWYT3$X>yZv+a&TXP!
zzjO5~+3me}@nde@!qN6R2b#bwb0n_ok?yEXV_Hc?Dzy2Zt>nYGet$+JE~!*Swi#Rh
zo>WZMRj2AWS1doQ|7q1)zq3a|<+^45B~FQIQDaXja<}1Xt^Vc%W^kay<eiKN!m#dm
zj@7E*rl++lm}UkJHVzE7qW|XKDToo~@=eR^fA&lN7%x;6?0h#ICp1BE_e@Uk#}HU*
zC_)v&%1^z5>VLVaif>ya)x9E7B>2Gn_~QNY`ljh}bM=UBEINH>RRetQjSkbL(U@(V
zDL=PEBCg9KB^QU+kA_5|f?GCC@;7A|kj>?_u0^b9kN(3V^pNcQ%C0*xHf9zQFjzO-
zsGn=Gp5;s*Hl`iha&kfW?x^4#3ylHG;LDXt{^m=$*1obwx1yT?wyky#B5FRFoD(3n
zDP#)1bUH7RDjiRZB<O6*8jVt`WwF(q!cx|Ff*d~EJ;L&}-9EbdfsU)Zn5M3CVl_M$
z-6@NFothZoq+Z-r9wM)#G&Y(bEkS4{b9bH%CYn0Eg-w(x%@LC@3&zLyu+T5&q>JhD
zEbn%q#AYU;T?j|%fgupe0lt|hc!b}68W!4g5-WT3UNw2x%GS>A1t2$k{QR!a$hg5j
zL=p4|Z4_iO*!Un^<6(5mg;`FWp2j0j&Ypkd%(G<=CsfWBryf1RTfT>e43Kdia%oLk
zY~kr-lfZ>LJ)E`nlYzn4C$qw!_Ttt!C-vv|v!MTjrc{|phCCu7N<&W|z;ILP>;VDK
zlanbeq-TT7fAa6-uk>vm6{6sCrzD;Sw6@cy^w~5kVp))D3>Vvf9Ib|d<<qcs*cE1q
zwPuLgn>ReU_xde}kG-)2vYDc3EVYg>I9L^gxn3)OfPFAkWEh(^(!=$Jfpes!^62d0
zM7JqkhA=`s{bfx3T*kh2g#`kFPIW%CabD_;Cr=4i^#ycUT>q1Od#3&D0h>`3V@8*h
z%j)knR!&Y?#eM`h#*k<>o+qoHJK#qt>XtGgAx(%uuSJj<PNE2(F3t`qx(9N-I7|~q
z3i^pZNl>=WdnhKd7d6IyFMaUQ6b^579+90>)$ad@60UT6AU7$Ho6nb_`H~K;?nk!d
z!FS%^WeB5@L8CYS`EtayT6lHrTjkw<Qj&_V41&yRvhznSVv3)_2#-e3gEEu9S#}gx
zrBXn<JcmQhrocG8DJ|SUaNWVNCwc7QkyNQ|soDBSMrqQIA0v0PreEuVJvL9c!E4G!
zfZv(!IN;{`KQiI97+zu~&cuHK3{qT85@=z@i{4P35j&NmvdmG=h&$BJ;2Sv5_nld~
z{I`>}-eAF#!IH(JC?-kxtX-?K>hLd(t)Nt=W+o4cRA<$2EjEKOaxGU$&X-p?jJZTS
z@CzmUh!Iz%ySTXMO7%J_3a6H4{Rx%Q53+xuP0ob>S_}?Qs)8`|8p7}1QOen--o~ss
zE+xloh&r9M3Z}3m*1i4D{Wn@%GF{eTztxfU9k=c6&{xdV=y@rU8ztp5&p;lHLuIl|
zedl_SE_5B^d-bQ^y3a~XvtP>N>^NgBEgy3{ecS4fLeUqmJ<)tEY#s{QS=Xs|6zJ`a
z=9DdusEHFb@P@0RG(txgRxkTN;HUQf-T_u%Jj?Hpvo{B3w8eU7Om2@KABu{DjSLMN
zc1p{z!rK2~w(cJlNnH0^q`#VAbu~`^X+OW0`+qFJPm_u58il?IvzSd^3myvCEc-$}
z46Furs9Um~Kr)6%0)7W<CMKqki8fE;S(~+<v&u|A^~t3hrb*<<H`F-nx>y#uD3Rvm
z6JR3JkfXMY+xMkex|peFf!-2`O)+>_(Ea$cQ-;d}!@8r%+`CBtj)fg_G7#}D@a;1c
zF+V$cfy8#yj(Ue!ySuwqV^v29>VgG?Um%qI{OamtGSh%W7>IxR?Q#NJK>cxv{5JM9
zneR=F)$o0Um658~9NZR2P&-qsvjfv0@K4jkWCXBu7vpr=n}gTTM&55@`ssB=E=7pQ
zB7kkBeuE1=L`XCPa)spZqbU_(M{-+Ngwx{+#n96o{{?o8zsi4-<58#CbCUP@Xi3aw
z_j@<r5_8d(`=oCMyV8EVIngNfXF9Kufu#*%qZ|9&>wF!t(0=Ye%>T<x*or!k59#5D
zfGRs}#X9Gu5qKe(G%&Zp)-LdDF`jos=oM_9Y3b4AsNM0}F6)C)2Unq6T{qqT8lUvn
z&7wFYhY7q^dtcZ%107fka=(4Oiempa;@$cUw$yUW$>uZTa_*h53kdr4Ej)@;l=dH^
zCTl01x|ur1hreSF;>laMkXD4E*o+m^+RmBVE>0?Ze~rrxKc5G6GrjG~a&oljgKvcB
zep56mflvs-VQjf5pmkCZ?|#UX?g9vw>mI~2%+B_%fHw?#d_kAjgJssln|i73#*aYk
zK(cC{10Ns%>guOIZhJ@Y`T3~t$+VuZL|y^El>ye`#up}~e<#jC-@xY#*=xB7e4tSw
zGFiFm-(hd#{>^0ur0>Ay&%?vhb@y}q{TZAmg%V#kX%NJR?}d;iL*C=XDqmh;2-XFG
z<|2;(g%0s1Q*pmQlkZ<s6Ysx0dvx0gD9U`J`$HqxZ~ghYFQ5><W%sk?v*L`dp<(Ov
zqxa~?_S4-5U-NLaYhcgi+%Ha+Lt31l)OV*_@%X54F%37=jS7+Ylq2;@o;f*U@d~j<
z8BF6u<)#~l@mK{tg`lR}R;ebru$q7Nv(f96Bv}1bLy#2WWdf;U|MX4mx3Z2LpYBOe
z#A1+Y%Sk8WK8$e3>gn!5CIq70)6tO?c`wjeaDNNYd&R!3Dy02h=a_{Yih<&BUwdHl
z2(j&hCnIZZ$Bu7r%&zkH<d)s`lC6sprn;1+P#mRT9Ei?gDBuu@DiKHZY=K3uqPMpo
zz&W6rEn1A&(W$*MjC>l_j@wQY5yYL`zak`S{l(^1lvy@cXjk%{CNguJ<%+&QOYZAf
zx}jhHek7|^m6$)7NQ!}P4KwMy&6e#CEP6C}-grZaM~`4zJug%3ygBk7qd5DCzn2tq
z)6DMJiynA%W@q&zhtYNnnqhSZ%W^GM;s3DjF`F&c*?I#307D&O7sai5t4oW!ob!rc
z)w}JIy6!VXk=Sq_y$?F|{_~vz{X`lx_ItQD^6q~$6;*|#To*xJ^RW^$8&Nou-K8*0
z>jpYxcxSo{`F~I#?D&TQ@;o*8=z=Fz1XqLzVCps-dAFvrm|2eCRfmdrtw;FWOUqZM
zZ)t(ag|^^sL~d@^gnP5ss-W{vhnWo6w1R)jt~v-({|3}O`?h)v+k#p$;vs=dg<Bi*
z=va7bpUbcYD@5btGX&bpOCya1Q0(w-qx<ISrMTcy0B_tHci@Ew*bD91($<LeFRlng
zw&?Cr#hC`Hd?vm#vZL%B7~fqDv%VEX4kICAP^A?W9343k=fYC7{mPss_ZKfJRV*ge
zBs9`rIy`UIuX>-K2XD`FESM%Ll!f>aUa-3zx8!VuFiyXsYHE@Wa@}vM^7qDAR7ApF
zSokVmFxcnc^6N>qa(R6H&!E>1BM&mz9%S_=sovo^vteFa5n6Xxs*5bExRAKonz46v
zz*HhcUfh{aznkzVTsQI!3s!;p{gCW!NfJ&~YCYJ;0c(IrqXRZ6x1gY#F&<WWkWV59
zBfF)grIk3Xu-h-1%O&8CCx?<zPtrgE##8^8JaC+;Z(z_r8g$AF;UhPZPZ>H87?1o>
zEiK3H-X8x^3q~m%3m-O1c77`NXye)1HFM+HIy>KLuicIK&n!(`H-!MSS%kDA?7rbL
z$XQQpvx!6hme-o#HMN(1#}R3Eg{No6Ou+@Dcj0@UPnBq4e@lSnaBiF3eAU0{ilFYZ
z*0U#ec6Q2bs|s$1{P&hCzY(nc@LnpYt4o7%0sce?bc>Rb67#VfDQ#WdXPXOZ!3DF0
zh*b#!KZ4L#mkEW3$lMlvgh&t;#|M+0@Xx?AV?d8!`;P=K?K_odc>$H$#^mlOg8PCo
zr5)HW3LXnAgb6l>ueP+bbPmdlGWzp?-yWvXX1e1y%$(1naqjYa9E;psVYYX@dicfG
z`vVRmNpAbS>uCuq0WXI?t!FeMBBCx-ud60_jv6P{YKwnbvzT6P$Oi0Xk(IoBcA=DL
z-BHhVBX{$svjSns4XX%54^`h=B(;~&A8b&#haG<5^gY~>UW0=$7u|cw|IT~k_d;<p
zwhAQ^|9MiM%+`?yA4qvvUdR_sM43F!+cCN9Fhn8{vA{8J9xlEm61pljIQ~^txPKw)
zhlflcx+6~@VMqG;t04uSje5#2W}My!D}R%}?js(mt&ClF+;0gF+FbB><AtNZ3kUQD
z$;$-Gv-o<#tG|M6g@BLKr8R%N{BPH#ll7tNYO~hdd{Z*K0}g}F7_jr)GyQkOHa>{N
zQ;T}58ym-#C1VY9=JlNaG`)?2FptT$e}<2Lhg|>LktO>(9nHld6vQ>bFS`2nqwv+1
z0W7!FH8gTdhhCh;K4qZFQX7>xKIb;lB9KefNA9XR$GIWaRw4Zi7}mm1r)*qxC1GB2
ze5SqT|IFoiRu0BwqHP>BvLbvc1)M~-wuJ~l`~2?m4WBCL-wMDXz1rGR!RSd|@^FGj
z>^4E2K57A8#K>sz;e_IrN>+If6v|-7ck8&1n#Tgl7spjAEe!w#PlEq9p!%S#p-)wO
z?DbneGb;-!IcUBScW71NxmjOd_wZm5p`P^wF<c2b+y9CRD69<S%ufUGmu4zVo!bGh
zG#4AgA2_7!!==5WzVoTSAwm!MqF0!AfZclMAriYLG&M6b0jyD+4s7<p5-ndei6|l_
zCIpg7N=hC4+9;Kza{}~90phQBnLoBWyEbwy+{U3lR#&>^Pi45vpPP_SNrf`{%?({)
za3xZXZ2bH87|QUNQc9(fGoOuli6q4WIJhl%`ayO~WZ^8keny4V4tk=E{!&+7cYLC1
zk;+(!<xAZ2fDcgVcx_zpgNT2h1=7;XpfCm<>+L>4mk-Mc{yDJ0^X$$}hV^aV>l51Q
zrJ&1|CN2mbLR*wJ467c|3C)eCoLTuV%4<7;V)E`B2FHDVD?CcW8Pu`4W(0=x5lFSc
zDhMayTTsP}p;Bn7nL=7!tlP_Ry~enb<neN#e)al~3>{Oa7dE3%B=LN6e6sp_lPZHO
zbJjX;FLivfeBvN0=Q}yFM^;dGum&|rlv+n=v;9Yg%njKSi<|W31bmk^hwGOLJ<psx
zsXq!&odwPa)-MZjVN$suz1;1QOYzod>m*$CIta+V`>}`H&w(05O}=`E7vnFZS|6Fj
zE=)~5eIpAexCx!G&j9_^;$(1pmGKT`rjJAaJ)pLTS(!k!u!MLLJ5_ucRgraKomdVu
zqdF%hkZGO13ssZ)NlCuRe3X@$<?A!fAZt%CZtpl(iOfp!=DvL7<|UeDq;nfk^uO)w
zc_otyv!-4?_8TJK>511fC~h4!V%J4-0YdG;f9iW(L{uE?ILuhN@8)biInFX=jo!17
zRHhE;L4@*syOU-AtBDJCCi-dJ4H_>IhC@0sx)2A`h$wCBptpjP!Bl2(qLqnMnqOY*
z^45i>;`-|b$r+59Uj@oOVfoa#TxH3cAc)6gpq8L&1cox|h7tAV<}S}<I?k#L@c}l)
zt?h-84C}x|+S`AmH;0SqhpQ9m3h8*vS=-SXA9SE2kcl)JKy5SV0vB9gIQF-`O-7~y
zuJh*@PnBPzdhTN%1s9j{N4IZV+ZkfwUX-jtmFWe14csO#X3LF{5Kk6#!AnEdhh6OW
zE}^AYA|Su}ZWp8<1ZK++2%~ManbvN6QdJ#OcnLiYqy{1m?8ypBr-qJCQksgeo4`38
zYV!M(m)FTi22f(6Vga5~$LZdpLeb=g*bQ%W&08~zYs2T)f`zJR;~q=pyO~GPsCJT4
z!N|3zsAo&T5=9^U*nlDw*aWE+5x77?LedOW<i<6~z6<IY1Ze&jz!O*_$kgN0Q$(Jv
zzrQG8HSl+%;mR9Ox(A^5Ygmb(6at7)r7FG-kV%Ag1l!C`*dUuArTu_(kpKv;-FW*l
zZUq>uU!%b{yn=8l(X{`3*fO((h(Oui-p++n6}}^?aygJHKfZb+bTnv08WC$-F<&?2
zoPx!LvYO{0=C_%lE6{w2;>3Ni5M0$Zpe_GYM2b`a=R|h*k+!Rzj_%&!2aI-y!JDqL
z>D=iKr3t2#d?l=cOgTPzH~<=!!*Fq0DFv@TI8|5Wxy6sJ22YlFNN|%3a~jH>0<3!3
zC9u|eoqvWw<J`AQ?;jxGBE~kre){h-%|j)9ywdx}k&Z|WabJ6VdV96oXhbZwXab{y
zo#)2Pv*%iGoH|t!ij|3*1rJ0~<4Zc&uT4&J?8N)UZLdyoj_6kuH7BBS3@Fu!(|c|s
zKS+q`*4C!SJ^SB+g5vUfh@lkH!crBr&X42Eq7SQUcV<?s**o^`Sa0W8eAgInS7lBH
z3^fJDUnA$1ZfFu4SEJZ*km!bA(V!tiM%(??eO~9EVsNK~I)b}^UIbPvkKifgrHj4!
z?DeniS*YCmRBSdSb<x)Qgo@0`aNJfo{SYxUG_-MYDy*tX8W>Q7Y4X~Sgl^{fNzT@F
z%{?qO`{}blK*Vgrk`^+i4W9$G{+OU_<exXyYq)vGy<8+g*yVTaOVZh+36#l0$qqp8
zAe9>*>W;%w>w8cVl3BG)O-+~8>uYKNjs>z1KBpuhwuM9;@#R_6I{NuqAu51s0nZX{
zt}&Shtvl)L_BrDKFFTq9-d*qx7f&FP(PsDs=@}lc!CC@|IAj3k`WYi<SJ151d_S!4
zwZTOkFv9Q&2xLYT0x(iFVm+K`iG)rKop55w>M>K@IuQoBB&D{CGYlMsJjIQT^V-zg
zQL{jhW-b9`GV-R=numg8h5VuskbQUHlAk>wf>P8Vhi*vT4w#cuNN@<E70brpCbfXM
zrEt|@Z=p8e>af@D(|=~LRQ~;SzcuRT<3uxHArb$1quvnnQ%Onxb9PoHMCu4Y5x`&~
za+neHEcozpbCI{QX;1h8(t-Qp%$fTrisYtWZV3^9Io1*mcKtM28JTYncZhN5^()On
zGi2f+e;OeI0C<l890JE2l~_y4%03nrVz4p|4-bo7RZb8iNJK=)3KFa;!0;9Z!a;rt
z;O9RXIkxPUx$zK65}pg6c_$h$*NK{lt#j4<Cj-nA!xdi|_{sdc(~zdpkBC{*OzO9I
z`^3LaI8HpGHR9su@7-NqsTJsYE6oSSr?Dwp_p>~?Xt40UC9HguOE^5zFCO}7(aC>S
z{|1TQ-4W9QQz^HK;-0@z>Bi>NMQFK@_wI)i)uueQRK=Bb)YLAbutzInGVysWEy?Zo
zpORyBs&Bg)=BlXaVe0Vn54{&n4?NTH@L<;~FAUUsSmJ@6OC-XW4l5As0Ewqq{hhe2
zomtMD-Cxk>9VA6knX$wy9D;K<@^8dme(_e|3*b_=e~uin08D1~1{|^0aNJ_FBSta*
zRSyvrut$|>4-6@Qw(~4_S9Ix|0IVZGu)43s3I>%10oX@5Fx9|JW)@OWUe5K=XxI0H
zg5>0`8(0(|(>3tYtCO4WqV22K9uLx7@N9kx7!#`HX8V0J|5g+h79ug;?>oq;t+N{a
zFboEQ4u}xNlX<m&sr;5Z6rLFb3%vTw_`Tw98xV70-iD0jc6HjG4&LhvKxWNh6h)BH
z+TMuR2I%57jg3_d4CZMfy%xMFM7=obmx4rLEUU)8-*)wPaN~Qt@CJD&_b+%JECG(h
zU@-!I3IaX~w`HheqAp{ip-%wC7-3L@qZ#!+4Wu~0I-J*Lj1C5yp|iCM13!HB9UD*x
zy74{{mrX@=?iJ>BNt=~)T0eidu*UEOtD_ok@Z0TT5S#jNPZ*uWl?{4lnaUc{P1o)A
z%dU8>ca%}5reBa03G<JVh#XG78$`*r%iX7TqMP`#8*B{P1F)gh2sU)|s8XKd#-}$8
zk>Cj*XL1U*g|fujZ_rofe)<$+!DAcpvEjidDH2xDeh$%KntYHoT=lY-0QM;kaYgUb
zm6=~k=0af%hY#}2zct^(Zn8tMQK6^>g=kbU07jdz4YL92k!<TpwKM=t@8}xL2l3#X
z8Xl9dZC9L(*<LQ*ZGtH|W7ai_B?P3hOl4i_<r%skFzmgw(Y3R;M}+1g&Qb)IjnD`p
zA~4q8+CH4FbF><HXIt9g;qUkS*q$B@qATx1WHJ{Lu(XKJORwr=cDHR=jZ8%ck9sP`
z!w)`rjV4iSoL!Lvv&is^J`h{d{sX^>>{@hEo9%RUR0Co=M;JT<MlqGln22hu{jjK%
zkUN8HwqiUgRbC-$Dti6-ZoB1##MN@F`N4OC0$Of-^;}yKImirx_|+lzV&qM3yu+4+
zR}WRpUk+N)A*d#1_a9gjujx(C2lRZPg#vR3_faMW=*0KCug;I~J(E75jtH&cO-Iz=
z&1o4L4*W;P*7Qd2wq{7%dM6~s(qL+OM#I?H@ff!tg(~2n12Ut~M~{TIa$XHJ7umij
zNU^9k;BCXESJn&_--p47chtt*=gI9>qoUTP?fVYI#Ed45smW$1xeW>kPfG0}j{cKJ
zySKYLJO3uqTf1OmCnqEGDtC8sx*^8~)AKfPX%m*`1w}=#5xeZ?YCrBQ)*)i=!MqrX
zodf}EW(H}w7BR2vKufTI4LDeJvL27S`@-A+HdCs|r=20W2w)gK<b`n9e!`FY=uri!
zO33X{hhVXngw1GGZBX3<r(uSDS+d|UITVl~ZF04@+#Szf`6Ts%$z=3p{F5FKMRAZM
zdtHT=Lr@{umiYBhCbk!f1ZBblx^1EG&@K`JlN7qJnDNZ6(udz}!x?I9Zk-*>3Yu>u
zROL!I$P-k`P^2z=C(d-4C*?Xh&95&=L0NmsmfBuV<z{dP>!6vV6Gy(_xrmqZ+Qn|U
zW&f%|bG;Sk-q(f7?~L=xqFA-P@{EdyEl9Wubb(-+&~FXkm!lRYF~_~QsM6Q9;ltFq
zMfJ#6)o&vD+<<(}p;CdIGouT)$^erhGYhKy;6klEf`QEN=kHD`w4cUC_oMj=VC7kH
z<+)OtY7wHrDld-%E&1+X6`}_MPs_>@1V$qxoJG(sMo_PSJBcmCV1Mv`7OVSkKwA7&
z14{04zb%_u)axofb{-)eS6N`w1|Hvq9d1ae3y7GO$6Mpz<)J`Q->1f47I^12!1^4P
zN$-4CBN5Uv1PczSr|({0B6g{;nt}$<qA!6SaRLFkDd6<I@897XZW7<$^XA;GCvaoV
z7LasBMZMbfpABmf!?%pm&!0czQj!qX1JI}_E-dKLpkv*rZ@d1ifZ+$s3UZ87tav(t
zh==|KVLpJ7?k(8yf&B@ex3)3uui}7S2%1)qv?hN1P=R?B7S96qdr-2KpcKFpjBpjg
zPq8$(TE>(x0~i%*H8GMUD*H6_n=Ruwb_FlAir?5wh-tU?AEG2k_hW_=QAQK9AeWE6
zP?!!$O1Tz!Dk-&b988AvFFx+Bgz8-jmWAN%-w9%gG>m3lCY<)8J2k6STK^Mb`#^}*
zWk{5i@00t|L5IZ<%UsE+BQ6ed6pFQn*^IH+JvFS3-}BWfx`E!__5C!n9YcWqKq(vc
z7nu`uT;N0WLq|b`Xy^vl7v!*_?8(jKHV@mbCm^u<@o4ER_N}K_wbt2#5vk9psUj}L
z?>XM1FD)+GkEmhhYn2d#A(QqXkapHH18!1)-z0N}auwkBYc{G-pTHFsP6&+y!!t`1
zh+&md?9b%4S3W^C3;QeLQ(885ENnMi^^`#)r{x~4=wKBzU4>6(Jyj~F2G?C2)4A0}
z_(}TR!YhH7vlz<>ij4u`xuSbj%9A#d>|(B5DK2D9O*ryLg(P%*Or0I1*E$Io=Y-OT
zaZiiaf#u13Ouw6IYO33xFkI+y%n+`oc|_<b)Wip5#ln4c4pogDW{Jv-2|S}6lZ6xG
ziz=|_>-?p_mc^b*S8=lE@VdMeZ}WL&2%Ouzk3t3}Xut+YF)Am-g)P`2dwbx>G032!
zTV6WO(YIc0=j3K=cdC&>Z2iGg0Jif$p)Cw00>krSAcfT;pdfX<DJsl!CwFpt<TP<~
zGsB~gIDj#TTq6sIzex3ohsgWDCn1><aKFbhIwn;Sg`Pnz!9z*az7a(GW<?Nd-J?`;
ze2%y?#a0JBw18!2M9g0t-mYh_l`xljVHAYj)zJ6F&GkHikNnaY;afYrcLY~>%>Od>
zsAcv+F0=|t7y-e^+s85*7!mYd6R}~R1d86d-cK%~?PRZLezD=m%JiDb4GRGs-I1Z1
zrN@$CR+9)HX{7;V?(eb8L7pqVH0@xW&}G;W94Nvt<fo*hz~ytv%qlxY&Q%;dt)&h}
zN66OigSi067pDz42&I@s5bo|9%AOLk$1x^bYnVwdH>uR*oCOhjwJq1gmp(t5I{SNu
zVUiK@y{pA*t&)0$>Dp}F#~H8hVjZByCnSWIwNF8Pw(dA@zCK+n#F?*ijKHJlgfubS
zkbNw-(0F^JaIC*PbjK*)!;R%_(Q&>f9STM@$pv#3?%USNEoyy}Gv@ZJT_tu;b77Ed
zY7=>g_4ik^SEssqQaK%#m|G)_LodBfW`5wA*H0fuvb443LL|3*O@(F_jfdH1@(s3B
zlwyw<GivVgNTMk8N;7I=A-9q_3O%4vV#Rp66D3tL`W<73o)a#%xFd0u04EIO@i@b8
zMdEr}4S~N6Uz04|_AVk;A|^)rS`x9_zZyjUG<+Mg&jx=d2vV>m2f?6-0<xwx6~(U(
zWDt_<qDV;|c$2mQ*WHA{9RTO8IrN;_hFZH4y!2>@>=UrJ2^PI`Q|z#Iy?&S@Y3)vD
znM2q)cDdJfd-EOs5SmDLRacerhVB!IT2jEf;O3lQa~LN=@?KJj`EWx@KNwt&KGI&l
z0iGSvyIq3Cw*Aw#Yj9r`$HX=t^vd~-jTs<T^h%nad`+3KamS{A3&<KYx6gXDvrc^>
ziJJz*xY6Z~R|v}_43j%Mb_ick2#VbaTt=4ht4oNyqU_i0#6Mza+L9IWN(t@Tv))P*
zS;paqqBOasF`^!$FvaDc@fJ@_{AWQ=gc~j9OT?H^cOnsi95C@X<o0llc+Yukf^<)k
znf@R&;pX)q^XWfqV`|$Q^b`+idIRhKkK<7za6GsIv2}J_wRj^X-a1$&w_Nufj&xi4
z=q7G)TGhM*viir*^b;pw*etOB$j(~cdSM5XDtwTAk*AX|GQs5tvm7)eh#MD@L(7GC
z*K*=#)FSKeR@6o5xos^0u}T#AG*sq38nUR@Z!?g@{kP1(2GeO#intE~<||Xv>|C^F
z*cDfDM1$gR1h%P4P#sXow{b98nIK|y6DU%iHS$rv6R~)eEP;%u<c5e)0?Dpgfg16!
z!sCfRg0Ob&Vqlr1wR=1a46p+O4U!bHVCdz^csFnitj7)6&`r2aW18>eeTn(!Y58T6
zrjhx1pJJ*<{xK*BM)s|DxB^a#v5YZvgr*KOum21&x0u}&c~5+{ndE2g<GY<bb<<ac
z8!;)!V<+x{@Lcj*Zz(cAc{E?AVr$C^W?7I|K~=IIQP9)s&YlJUmJDe>BsVSk)z!8~
zCKx&s78VS(qzaA?E;e5V-he^?--D|6arX92cL_27+}$FBDR+Y6fjqc0Ba2@evvf5u
z7FFd!6hX;f{W~mlG@dTFZ7R3#Q5)5AZHe(obprbY+-C$g15qR>##ethI=Q_#0Ut+i
z7&`7y?Rf8ahwReXa6mco!}m2vZ`mFVRtZ5|J!1Pee7~rcGP&PEiX^K%2aytkxROo^
zz5K&1|MkZY6|d-)YbI6Mv3z#y3-`$eHB<c&p19$QvJx!Mh~TRx2$@a(Mj#p!AnyDD
zJ#d79HlfBioObmEDqHBcJE*@NqG1n55tar=`56Ysk{Z9eZTBDn+i%ww>c76Vv!Xt3
z)V;J4S+Kc?3m-CPBKjBPKfbjW`hP6IQ{HsC1OJc&i%zEWUD*f%p5sIYMnZzRf+TAf
z4wNOC@W7OO>Nt@LmpCyu5x6R8yZ^<!xm0%)Mm1Vf(nIac$nLQz5cx(GvsBR<0jI@1
z&5D<}5rRw6gyLSF4jR_8r4P~<@J%cD;ZLOel9ZC3eP{rib6}74(=3bkK`YN1<^k>I
zTNf~X(5zbh!KhV*Byjjlys&XENI{qeMmIrQk2d%Dx1WiTv>K9TvX6nn;oA0KgaG-P
z#OT7exVV@_o}6kfEy$S;#IhC;hlae@%*NW6?o5qM%rX(E?kqvr>^Gad?zWHd!ee))
zrbiDSkEa@f4d|R^lT__(jD37JdpBm7!J$A@i0=5Q*Wkyn^*VBOI-;<b=08)yv5dx)
z-k7dVMBEXQ*Bd*^IMzPnTBu&l(fQQL@@3<Zt^c)N{P=CQ$Vl)lx$LD;t!ltu1xnh;
zry)#Z-;t#^V`Y8yM6`2HY6GHj{zv7uOhPA=mN%HBpSHjvRCf=ZCRHnO*y<ddrfMSZ
z3t=Qgn12eXdRcSI#t8c1(ySqW=sTGS%U-p^_{b76UXIN4?l~;wAO*Gwvt?<uI~wp6
z_#&Pti10Q|<1l#fPEqAG7C9R+zR2{<45`iN)^grV6OlCe{do8w1(hulNml~-fXMBE
zt=y3R9kGKDY{DQ&h<S?D0exfx%y_W2f}%+IdKYqjmSY8f(q)fk#q=t1HFlvP(F7iA
z+-pA;8%3kzr(RfeZW~5yf&E=K)v%-j16$66;fz98E%Umnu^I(Ca-V@|QxZ4c_V#Y>
zCq`I}gNrdix$w(gDE~DqbgZqtJa+N(1(T&68D!`Z6<6emyPu5|ai~wnuo@-c<;?z-
z7gew@A&Ll{xRht^KU@EZq^d62&s^}I<JVu8e;GM(=^4}3pSdUOs2d7}Q!F9z+KC@3
zEOPOX-AC%%Kav-<Gdbc-j59-u)OW$SIC<p2>Kcf?p+J#9B)~(NKy=?l(@~L?RjGJr
zhntKj!pwlcTtHGWG^SMevP`{We?;?x^6DiKrj&R#Z#xz;TB!4e`>591O4Ck#Zc*Lu
zy3b*{PINB+&D2)jH3{DEpsG1M3PPLECzR(<BD$ZWIo!tN6h=p;ZdcYOwVW8UrJhTB
zrMa9-ht>IJChE=hcjno2Uia2x6h((x(*Mp0sW&G=CD1dGeIqUB(UGM{T&n5O7QOl{
zc=9z_|Ne-IE7R8M@Y9f-z?C<$3rh1<$Vg^BmhM!ivwTB1?sK}Kupp3Dlh<B`Btdug
z$6Q=;&*;UD2&Lh(@jDIY=6tR9IdCgENak>>cMMA|zjf@6WCo$2oo@*mK;ad&y?W=2
zlBz&oiISylLtjm!TB4>|y6mxqITxowXk4R)QW4G};vCE=P^oC-|0A3DHzy}6dd}CP
zEDIE(Na98ZNpc(-pXFJ&cs^1>ye$RlDzC#<khZqArjCXpR0$>~)-cH=`G(_@b;t_1
zF+Cl5;#6MRe700O^V8qz=b4H;!wS(;dTcWEn!G+8!U8i5wl|P@R8;q*wpQ2Z($-Op
z)u);IQSBV==jl2c_TnOzFq<;cyToV%vXXp^;o(Rq%#157!$UuJ-EQ>#_vWj4ojaL2
zm7XsJ{2X!|1yZ%e+UoO((dd3hQ`pIwquJnw0+LlB|DBuTyDnN(9%IN<hZCmK3UA-N
zvs`g{sb6N_c))nVCSrbl27?89^uPoDokZD=s2O3s$aRGz?Z&E|(@!v_r@Zj;O4BWq
zZCkRovx{O>D1E18-qG{Y95B38cInKE{nh1X+>akiwJp)!_ntapV=G`AlsZ6hB6kN5
zASl&kEQRmKO-@GjCO1TXlVRYVN>gVgK0E1VKe-7c$JK6g#ZN3VdB)<+(pi`GiY?`d
z9sLXAE1VJH6ZzkCdDCRW|A{}X{$0t!G{womU>lWjQe3IgOT43FDs&HPv&HG|67fDm
zqH!d5sL8Ftr=SP1WOZ(M(j{&vCW$t^3&Uj%Urj5sbkr0q0<SMIFjjruKV+AVU%=$9
z$!778`+%ije)pcEVWVuWMoL>t12tDoRhL1vJY&`F-x$K${FFYA15d~ImN2h;GVmtZ
z<6DPCja2luqwbv<OnGkd=5{le!rH>cJ%aN-GFc7T#}sQ1)0SYHkN=tQQ^nt8%_^7@
z?3N7DfPY(HD;hv2d#>(?>jLeXZrC(j6cWkit*7OKi=tON8xL1}<))tHihEwB)Y{Lw
zL5BU`&ET8OeA@w-&P)5g1fI@ge&)}<{JvRw_747qQ9%>i6_Eo4L&@}^anSiX;vO)W
zUb`4`5ZE9y)MPsU4Dkckk;^vo@4iS_D-yTCJO8@#zLn5-eD{Q$l5*`8?G>H#?Pqu4
z9j1zc*}VrTpL9E{3@DA1e$hmVr54}y4aoV;mlp46Vix=0aixU=UjEDdL#BMJ1yU<l
zmXfh&$t5wJ7XdmiaeG<b#iu0Z#SpNFe)^A#@MTz<-S}k7jz6OgE0MwqPJY7l>5M^&
zx+RZQq&atot}GW*1V?|uzq9ooBg=3e^sv^wWG_t)FZHyeWF^)c-#tD}`KBjrtHw-G
zx(hGHG+P4>GzbV@OXnW4Z&-f4B5=Cg%z6Iav<_c#f$MYsJDN-noF#su$Zp;JU3-ZR
zmVscReh2aB0pb!x*L4Yd&Pe6bI*h*PcQt##sPef5hOe4%3{p#eEc+L;;26FqUGQ;z
z7K2-*DaJQ~r&3xF>Ad8^;ko<Jp3^||TdF`z9+Ui2P2&f9vS0oEd+JJes22F%d7EvW
zJ-DJV;FvkzcN{03KD6ffW)^`Vw_x*>P2N(`>F<uH5mu<Q&gJioAh)_{Tb;ex;jbQ5
z1cgKI8AZZ6iEFgW>>j-e^=hQ0XX&-hpUBSIZXVR1VX&8xD>UQKrTp6*Qr8|a`oZQ`
zO|+Qf)JKMIYZ_;pJXd#)#9zLAN!(~|Wfj5LzvZF%R~(XaV<axVEm>JHKh&W3O#16d
z0!|WI)c5qqUutA4OzFm(m&ZuL*@l%xiQryqLr&WFsj>A24Mth>^R`yj*7J~y$Bt|q
zK7>wBG;8>YmxbhENw(tEs=%!%xA>p`M)O$Se)v#MnrvyHyB>+7@lE=lSS^TljH5?(
ziMmOQ%4%wpPW7+R-dxm*MwwK|6-}00j7g1Q-|e-{OxpCVJ>a}TmTAnAHmR&^(TsAc
zA3Q!DiPkHQ7My+&_P6n1(SWRBqiUe+k%+zO)Qo$mzyq({OPtH%*%DiK{YQfDMh^^)
zJ#fSOj{M4pU5qO1uj~1-98PSu{KRV;pVl|=J`-y7)ArO#D2o^Tm_sr@ZTomDc>C{z
z6x59HiHG>9EN&dC_RL8!dVd1B(Xk2C|H(S4k*M5V-n{#SU(x1`hGF7EnW&GFW|cAY
zwyA-doL4XUF7nHB>MBRWk_fGb84~--2Y98Ys%FAItAxt;BJ$*)P?lTl?RYLzY4$rZ
zYlY-?JR^+Tc@bTk1_>_aeNo(>@*y}6Vl>8_>J4C#V3r!pDWM%0LZVB6@wK!PrN$tg
z4F~xxa2Ez6#nVydW7Manv@Fw@X~SofFMXgr`JGk%DO)oND71pAE*y61rQ<@uL%1v?
znDUFl8!4_wn^bT~lU{9;4ps*O_F$&o5pM5fTiCGc-WX8GsfxgkhubK&Py86|kZsv>
zm-Yz^9zA1F!|e3U_WAn$yyB^%X1*HUcU}&8qEE?ho+lf~Dix>`2MLg(@*pQl>s^NZ
z#}X)t^f{O6BA57ArfZ_K6McVWbwM8|)0|=F%iJrxidwoso{oKX{m*RZ3!#(xpEGRg
z2Sx1LT2f*tB&V9bb%rIhKG!g-ea4K<81<jsi^?AkD`)nceKkY(9bINFNU{p!<%zg0
z;iC7ST511}tg{TOy4fB!APph{(nw263Q_{n-QC@dAdPf)Nq0y~N|(~zDP01Rn~wkN
z=bYzU*L%GmP(HEuFJ@-VnzioRbO;_5jp8&W;0u*$DskgIO6|61U9cM39D8W80YwVG
zaD8?ZBKl_ObHZU6-)dA|%es(61zIxL;(K?jAYcA_EP^9a>Fp&3SwDF81>vNIFB%~S
zH^(C6@FoP6tKxNw<Lm~gt8e^2deW0_i|W6irmJp_kx^d7WnVn5GE32ZgL1a|%~C5l
zY5j~ij&3y?V|+HHY|uk`-~3Sa?=c#y0&e`Uy)l}8$GDn1hLpij=vr9a^AAzZk6t_#
zsS-=`boSWlndekpE_VS#w2YrH_t)m@5qFoGGwT(#$O?r7c!cwC*v>mDgrve37xi}2
zikM9&6!+5b(BDL~J><l=ExS4uuU^fG&q@v6izcfvn7S5`$tvM8D#(o4m>PM#Ec()F
zKlDgCEc2P*;FA{<EbNh-RIc;w1KkrhVvSk|BZPF-M{oQul}3639=9fgp^Xj7l(KT&
zqH%zf<e=9wTtE-c%Zs^wWf5;hmC$afsYKV74<<vIrUWZ?jf38%6@^p#DiMQbW?q#Y
zYn+(4f!n5|^(3M;fWg+h!X$*ulZm0MtPIj&R(uo3S7rQGp>@w6Hhend!61iEA1AdB
z*XIqm-QdD+0H8{Oo_EfwL6?NDY^RqS2nuoZ%Fwi2tQQ5(y&<e0tIde(b!pk_H1YG_
z9jN=tTLgHJ9GK4aMa!}}s4UC~rS+aejFDX;L|k1&@be@}Q$Iz;%%Z;3Zuo>^K|??h
zvH}9s?J(j<_bgN$Tm2bt?!BwOk;m&h{043UihI#~FUxZhxT+Mk2?K=+CSRl)T2yHC
za~!Vj>PNu($4Qu|Q$Lo9QGo7jkT>{$|J(UsB(6pzFG(%3A;*2A>UBhveI@^Or4xl1
z_nTRlq{Ca4re&JX2X1R|40KzBb}gGcv*lXP%nV2-BSWojt3r~Td-3H!*U+0ch;ayp
zrn(xMsYRCcgC~GcUwLLW<9%{ahF}=W$8|sB{W}J*y9fVJ+4zDjjJgXxDJhNYlojc^
zH)OMBHm|-?EqQcwl#Q2Hx^bzH78Q8jt7%;LyZ=m%i;Z#vj?kdh6HxmxFfczf7fNtN
zzmqM~K_I`HD?#R|;1E?DKXOe?O*I7Z8PuUXlK!5ZgX5Pp#tmlQYiGF3PyVTuMJrm!
zOpq>eRTaH{1Na`oii^3Z>l-BP5Y=yx{^sV9ZQ6*WH@vsRe=QJ>l~1`}t6trlJbeGK
zV|7d-dds1X_&%Em;=z9W)5|_xVr!r}KLbW!?7(;W&#hTzn)`FH@A-3-^3R<YUh!v6
z=L-mB`mbn)E>>yN&ZQ3Z9^qv+wPTCbE1zGvntNsQCA0mxBTSOEsyusr%uNbeabC16
zJ_XhD*t<oYvgq;CiE;iuNbG{Jt}hFvS@Y^s%j_QB%6U!Lb=Lajf5}WpXJGTau3Bmz
zm1Mrp*1!;*f52?06HP)Se;1p)X*#%fd?FXs5xT)M``_o?&;E@3=g4SN%ArMzyO{uf
zAaX>E{*`|q={1T-wfkO-@%e{-Efi*t-#)F^H+12j()}#l?5}~kq)w~Cs-_H!{A{sP
z1(_-~Xqj2dt&d@s>#=0_`GWLtl3<f&2+uDW%cD4DO)`GZm~Ajrq_*OfC?a8|zIx&1
z5tdPnzFgMm#f3w&PB?g=;f~v&deGEQc_-V&t>?a+tEz{4iz3Rso-fTxGsKovyXyzY
zdAASzQYeAgs&}?iuZ3a(S2=q2t8rtr@L$$yMAlw|^9MzG>i4+hI6@MU-+vcM6?NWM
zLrV0KSiDA1$mw|tZ!aO*^WV2eqTd+rM0{Z*eRG;-%UWvt+FI%QjmuEhE%$Fl5{ji_
z?yfl;zx<WG`^{729{J=Z{;3h~=|*cN`qT(F+an#<I+%PWM2=qrHy3T1W8YQuRT8OC
zMw<q+o{KR%eQ~p$AxPr2tX!3MyR8gUyeqppO~YLrxjoKsW<q?Am2-ky@rAS@Y1KEE
z$hg;GCE&+xgXehdi-f)Ud6Ox*5^)p$^=W514WU6e%$%st$d|@DC+bg+^^||b4H^DB
zLTe6K<V7l+F!zOAg!-1%DC87D1qAl+=b8@-_Qxst@%}>A@N`xp{u4zOvAZ`4fhvnu
zth^sa=C8RLlBXn*{m}W(O$#P<7(@>e*cXr7SQrXLP=h!{U#%zpjibQSFpopFpn6kz
zoS)@smF_g;MZ3t){at*}#3JYQqO5h4Qn*itK;q}}*C;eW-8=!+JrEl?q2SP3o+TM4
zY7EI<o2r<2+X9L4Q8saUj0S?buwpZzZ?6&XN0&+lrf<Is{0@sHk;NQCVzq1^&rr#<
zj<@ohPKX@iFeKo2h8(2{ect-6I7WgTFxFEN_Ot(f?c0y`-EaqzSLBKv`cJQazdT)B
zY&t9%iO9ph{cD4&eZK8k+oS=U5x;3N4!&;IbL1@Gp?fblwAV5cc@Q_)inToPm87g3
zeh%@7<e3xbXg1x(=w$vobo#J_8w=;-m*Vhv3(}YB#OKAG$!Bd5UzsD?DR>hqI=On~
za()Vq*KGCQ=2*mjX>P5*cD&38DE@b%+#(p@ds)vIN{q>`gw{&7?$xVE6Bu((r#~44
zgL^}KWy&PC@>SU)nl&?sF{pbtTbw%fGr6p5{@x2G*WgpD@!~)3!i5(%{5_=T#2+)x
zx&QWf&tLku7jP*iGk4-()OF&ADCI5ojXQ>>o%e@+uXQ$vlK0}D<BSRMP{s|Sl!WgV
zq#q=%CfM_eMN*YwL6liwvi#%R9^4#si_FT!d4O0IB(Usohiir{30GkQ%M~=cb|@%F
zm}|@#z448tl|!`Rc1yyoZFDmgIrRlY`!1Giy3p9v2<`WaFgA6OkLSy?$73my3+99q
zHK(b+qDN;&uZiE$wC^s8vpacI_J^NnW!{{uG&6k9U;T31ouRlf>t&djY4gz7_PSo*
zcZ}ALimjPc#lFg?L;o!%q05KE$kpK6jj-FBl{ks^{65}Q{@TdF(r~Gg+gdsQ*plg4
zM6xfsTTfJ9V6T4WO;DcW(cpR%hdWpW{c7e{g17E!YWIAgnE3bZo2VOSg=&4Fm3mn=
zOZORN9w*`>eFXN0r@Ex;Whpq-a6w5<Eg96!p*$9iw-aLd3dl4Q4LMVi1I`i1;i(tT
zRBVZ`3E@!LwGxrxA{r<W>tr2AxHDX+!hUK=Xv0~PgkjMlXubFeSe~I|Fwfx=JWQs?
z`9+#mHZ8A!4TRw!Q%&3;k`#l$IIc2MqMJNPgpT=s;Ae#P?Coth*=X3h<7aOyLAo$!
z142dF09RQ@&W~^C+zls0Kent0N%cDiWH=8$PNQl;!pd$%B)<w;E~;|oD;W5Me&jW(
zl=#JC*nPIY^?m*ee{xuHQIttrtKyus_fo9++p~F#@WAzR%p;^-X5N~UWrAHW4hbw!
z)gk@!Bw>GL*R9QG*ug0i$xhzQgHGcm(W|b9Pl_Gw;-WCKRZUH}<sxxD@AAt%4<;sD
z46V%@OxwymI?9{43JDx_tJK+*$_C_kcax?Bg`7)v7?CYmhv9Ne&}_7#VFgQwb<B6C
zDcS5X5eKIy@wLg4gu~%6HHemZ#M;;+D|Y)nTmKtD)YN``pO=h{OGA6~PN+}&3$}@-
zKPR_4>$7VwDM^Y>Dk>C>6lMHuR`<$*ASz`5u8nMTs1&r8Ez75siI(|Eh9!jx9}rf_
z()1^z=Z9fg+s>DzACpXuo@IlXJ(>~9C{2G8ObKu|k6uQRORA{qsYW~dWNZwUeB&Pc
zyzTl|k_pGw6h(9$X=qV0lAZROLh4WafF3k^HMbsL+j`PNUcVm0^4z4tq$r~Y4WStm
z+YNo>2pf?mWO9pvmHu;LV|XPhsxSd*t&k)COtFMwc`~Qb-DlR}vFr6bF60LTi!b<-
z63ih7KGR#%Q?d{2OZRAe$zfARqIy9Cb49X-%EOU-oN3rAvg=o2JAx7S_im*>LyBls
z@vyreidGIT1FLszTN_-NAUev<ich2H6I!yobH(Wnx&+1z3`Vx{5}C|>Y^3-zX0oi`
zUca2fX>jfB&K`eHMD1^1wW_!Jbou8u-zbfoUn5DPV42y>+7tX*hESN=jeoCXNHv`h
z&lA7mVW5Y|4tDZ%i@SlsAoFkb-lg`KR*!>}ye}3X@c$d-EzE$2gv*4GlEhud=p&c{
z=buU-JL^>Nt#M02H6pM2;+#psFP1Q=$Qe~j2pJyE{9O}MsRdTj8;wl`CAF~uHX{aP
z;RfPY8s?JP3pvfBH^`K{U$^EbP`8^`Fwn7z7z#_n9F)g~X82z3d*A+fDPMg=re(95
zVjp5rQN|X<s%ms`l2WuE>yV#V`sSuEk+yKWf%7}H*WL3lQn)$})x`AJ#~Y7nwYXh;
zWBdaXWCSg=@NR?1BKFztuQ%)#_Hv=qZ*9{UJkz`PjMu-*;5@ckhQH7M8#1;xC8>*<
zDm}e8Vb0OoF>1`M4jH)Q(tMmJFr`$_c8j#-e%iabTk>w3v9i5_^a*9{3df|+iCwQO
z$IVz9Br1^PTe>pEn8J$bUf74v$85LUe9L(q$B}|ulW&~)B5vqB$U$#6+|If|Rrh3B
zGfzzyo;2yIuIXt9uEFDjm}vYF-%EC{<0u}p2nM`+axdQ@o)8?20@C;Uc|lUU@=shf
z&aq(#=x_qB9V8M}R0fLu{P1~h7sI;M{yTh3)Qx<GU&H)x<(4>C)P_S7dPa~b2PZVi
zKr<mi-&y!+5iUuvN<cO*-U(tfl2)SDI-90uU7lfc`E2U9O$?*;W~6YEEP0xtTvsn<
z$Uq3qD7=qL1|}Ri!Vc@!8SA3_PXgoq8rJZ<6C+$QQ>`83rP(Z+#P?R1Ohk9?HOimh
zB=Zmxe0%W&F>!+4$nt7zz9s8%3{|o?kF$WQcrIlo{LJT|6=uHNc*dup?z3Y6AFU8@
zI3d;6dKV;q;-OnaE?<f3Kc8NF5gdk5s6wsOO@?KV_~Xp&ji3+~q{)r`jTvcK$@!-#
z9KkG!%(MwTmMg2nA~)_;in%P5>8097&G2D;*I<i!%E@;tVxgtJF}Ok5KS+{7%$cme
z?>yX`WI`fDxM=d>18zhxRl2Vfloe*Q#x$?+4V&NQ`-mSjLclU{i|orgH2TR?$TX+v
z>Gmdb$j4z<uj4Jfp;7uD%|I9t|F<rJbnLW27{pX?rL8(~x}p5NuV`~&!fH!faR&w7
zJwu96G$v#3JO6iA)X$#aPGN<TI4q&uk0za5q=8zXyyfoqiEfsM=LiAD!d6n*kLm7M
zlLt{w$@zc?)d>p@Yg-hSZV&|(t-_|<!`r0drvvvshL$uZq*~G0w9q;W6Qu!0G&nYZ
zN5v?n#P*5HJ<s`*DrORK;yN+aB&l2cH?0LYMk+QE4>0H_byUj4s5%K%$F#^&8eJ3g
zSgb$a->}b;i)bWU_}vD^mnY>9YI7Hl4C^EiK$cieU%n}hWW_fw9hhy-;RwGq8S%>a
zaVNSqP<NxVYcp*!%TCU$MRg`q{J4<LtoR{LH>Jr(APZITJ)42~L;LIsRu(cZ7CouF
zFXhFHbS8{>6HISq=22LRuB6LAA?ko#$&$k#0lc40=7b6ga}dT`qdD~I;4F0Vlh$4l
zEBtQ{sT**^N=XI3xP_0V<o`XGZ;ZgYxK69k!ssBOAH$fW*UDzy0Y`$Bm|zS0O-fmv
z^@LaeTSI+BK=3%Z2%Qag;0Z4266K+>CqqoxnKi_DJ!~<G2$ID;zoU`%XOY+@UEjnY
zj%(0kWaiJ?nY~N2F*|t{!o!Z2H?w>g_H|r&JNe~BCiNB0a4c(7O3$^2-_fv(RUB4T
ztv!FK)b+?Ls~k_Ykn1A*@1k{9P{K0{p|Ns}%HvGL##PZVsFmIzZ15LO4p{dnv1NZ(
z5Wbo8@207;?;f*00~(lU6vS(wbyHt>aD!$TllAT(WY*9I#V}f1c<NaXH2ku2@{YSE
z&>V$p1W}F8|E+E={*t=^H*ShifY0tty4~!^2Sl_U+q5dhRgck%qM4OTijbWeZuF&@
zm|{!3sb!?7T``iO64%OO;Tg5{7@UPt?~-2Y(o25!vw447V62(ds++Z_7&vdeozkiW
zdJh<nw%0JA^x7Phb5%J-A)xjC8!>H3l9yQG@yB3=rn#x^Ko)zY<k&E|@*vwxAw<+_
zUDbtaC0<$$UdoY^;s4D89BK`}P(+hR6qXv)k&Jh3838#pC^m_;h)fb`Lios)$olz_
zYcWnEPK612Ar8oC7VRt8<;UWcMmu8|%TU|BOGu8v(wTgxab(L7i$WfJk943Y0HNB;
zr|AA;T0l>bDxYKNIEyp%Cnl<PiEl3QHMWq-(bO;j3m#<in>c&aAU5@a+HjimT4|%|
zL&jFKK(PZSt@z9${kKJ0;xg7>3M9fS0Z|4WtK!zO9jnE22NOrGWlMRqd#)`O6{%@y
zaBJZ35=!OzNLL?qjrVVUgU?Xp!+R|B3}jS6G6_UDaY&X!v7IFTLg82Eubfi-9(+~e
z|GvayGCOf4Nl9U=&xJH<7=`J6ZEaCuWAmG<y?O``%IR!4HPkJw%Sb@Zjb;1z%|v2a
zD!zPC9rJy`JK;R+K&o+P>R5~bf7^tNXI$SaG$^;Luis%#1WUi(asUi65!GbI7-=;j
z6<TX1Md|Q~m*%;Z6>8QJEY{&#BgQ4WDX5TRh{w!B97PVv+r=oJoh~0np%Oocev4|m
znOJ+(%%5ra&H}9aK!+Jtj-W;dbS>-hdcNDed>m0llw1fWU3eZ=#~EMM<&mhWHcCFL
z{J$#<=+9oGti*Zt_SGZazqJ4#^Xym~Gjp{aj-pp4bzi{h?Ukdbj2QmW*<CqWgZ%Fu
zBIs{@SznG|O)XXUrS~fwIq~M^W@J<p<u;iV-OwU3+vIe_YauQkV*<%|bWy9BUukAS
zH_e@jcwfXm;92Ypl`fGW!WLQ%;ay_Cp1aa#e!m<{ESLM&Wx}$OPj}?(5BGO5&1Y5>
zjW5&wEXCBIl#^CDM}vB0J4*Z!7ca;|qz7Lk?Ym_=q3dN$Lr-SfZ?+AjiRQ2zY-#iv
zo31Uoq9!9COT^CJb8+!*YHI4@?qM*GWow#{l(gy(&LNd?-T&@@<R%Ojv!IHRh?3`b
zavS86JoTIQKTZ9WqFjCJ_hYXr-iMsLstWsexHI*Ft9C#cMU7X1L2;j85Dj6CrxKRx
zXVcd$RMJ|&SCWcXp$H@&(xQ?SN;WO9l!`|ew?a@fN?+kTSDc}U*}*bP<MhcF$!X3}
z#fwZNd|UJG2q8T$p~I2OywTm>o412h*|XiF!RvXv$ZUgcxIW}+dWnX(IEIbA$YR=n
z>y~jjTf5rH*^U=;F<wh0y~6gGa=wPqw$4-+o(TO`Y#Rw}3FIXsS?iW;L0<C6wFOHs
z2=h~m?iVQ=c>0K+WL&)&TK}$PG_re~_SlZEf^A%*Tcr?1gvUb<g(U9hi#U4;nlb!s
zP4~|fNnfjW@7@z!{%;(}gJPZ**?62?LA)HSkVJnfNePwS>v_1qcXhnaz8jWLb#V)q
zy>-;+<wXlQ+-%J$m{>$3aan3t2=0K~k#3b5jqar@!pS~xOIM4-({<-KacMd$ix5^u
z2o4imY!Ws)x9OGjA%7p4vlr20*2LSlszPv-N7lD;s~{H{5c`?pMJx>$pkRJ<@*PDB
zov<KFdYH73wRcG=95eS-!&?Lv!#Iki9gTEFOC>3-3S7%HwT2WWsT#(z29ya`%d}+I
zWRzL0@Vt`B%>TW~VN*yd1evBo9R;|K*oVkN?Zs#t!DOh~Eo|!<i;K3H&FBFVm7lI{
zlX6s9VuBz3y?gx$X1_1&TqY<L`I(buaZ?MnJc3UVPz&E>mIu)D{@K?Ev5|}jX^By&
zLv<?LmtZz>&lVCmJ4nx<8i5HDF^nSGlAh#Nx-EH=gj`P0a7N2fO}`pHOK2|5(A+&I
zkL<_UfYSv};g`3UK)+4tE~2yz`>gY3SsrFqvjAOD0I?7rS$u;r=q3Mqjg4xd&5SP1
zu$(ACk5e%&#E;sN1V*X`x-a$@eTJ={*AQbe6t;HKV3#9&!?K~9z20LDm`$Tn<kB=4
zw5am?_1ct35-TYO#?}dI8%f38PEy;LD?KlYI-&M!_v9#BT|K@b3)kdXyt3MQ<S%+Z
z8Lzqo`t(71VL!1ZPOKnyvhT>Kf$zxR7yVpNDn{Yk@{qd8hF~)te(SL$5vbv4&<ILq
zxJuvvpVoUrZQ)X^x&|&fg^0yZ196h!a#~agHIXnj&&KUyW+=M6wlZA`$wCowU&>L~
zqGGgr41AV|XjmOxq*r3x+C()fON<9+XDvrp3A^q|!f!*%*Imfr-LupujNP*miY#BT
zCQ}V4nB*yekT%YxLM=kCN?nvq>);oV`LOg(!y9TR5mdgkwyoD?Cna?AV<%Q#>8-IF
zqAL`!j!MKTR8ckG)lqeIBi4IyG=6*~IrS<1L5*~&xzonhg)ujSOlG#?ov`;o<GOcl
z`^RaU-?NmrUrI*~8VUl4J@k-|po!48;dz2wep0WrcDJvT{5edule1T0QA2*B3gU1r
zY}d_Y1`Ze{8vGMzw4hfG(=hmZk%y%8r5eUrzZqDNib;R{Bg~O~bY3Ps`)$^8k-S>*
zZ#=Xs!*O4mX0bqiRqr@;vQ{<K4#N?W*CCI}dso5U!H=fiFE#hXgYI&f)sOpv9u4<~
zg7`JhZ^n@@yca!4nG;{#Hj!snDp^^O8g}IVHM~>!ZNSJ;UGu_wWSk$;etB8G)x){|
zwdV%8Dk!6~LJh5gL+fDB#V7<hpa#x#h`$tuM(9FaFpevzpX4p|enduU!CGe<#-_sM
zg>hzSx1Pe=m}W0o|AhJjyhK>k>y`KIcz#vZX+^5qF<K{@_7OW@DeI&k+H!|ayey7S
zxW-$3dT!7fJIB8k+I8tyNtftz%?Y~jK|FTc^)ky8E;qS%1tA&E(TqbZy(+sa+~Qz0
zpRsx#rTE9Vq&oYSV$q@P-4w0H!FgK9L3vg$IJxso$vFEVOf)m5>nwxM|Cx7Cz}6Il
zEAprBO&l2svU%CBvPv9RG)Bok<2*FjZxdf{c8@($GyeX|^q-ae(0`an(Ylhhgh~=q
z2*`@i-nc%?sW-@=m@3gROrn-WnEfKd=A`?@k4jond3C`>%^;{THa#=5;>_-qsGkg0
zq>Cxip!T_C#ROu3gX-?P>0cHh3fqEcnyxwCyWjOMEUAxb&T?y!-^i$-jTgl*o(mFe
z#!AJO;`ZX~b(3L-VPYi}SSTBOr%lR<kL<p>Qcxljk$p#o&O?G2go%~8TcC+qP35j<
zfKVWGeH#3wH9NtCpqG2{4_DWIpo&M=&$VhTN>ibD+Q7F+Y0ZLKo!4^q$@DBiq!!=M
zZF6V-<ebB$zQO4wQ}uTa@UxGv|9S+8^=ErX5;_36EehG#QNy~JKqsz5oc0ka<W^d?
zrDN4hFWrV(q1ZPl?v;|0=$nj_^nogC#1G3knYlBT<P|xq2UCMs=n7l9TB$!F#EX`_
zeSH&)SiOQ3YGpNs7Xe0SMqGqdlFm+K=wdvXa@3sL?sCObxpN1qFOA7Nv_lbzh=_!(
zb)Qdn*Jn@-#JVZt{8GSZVkvpitywmcw|cA|aL+hYdKKs=?%j!v7`EPYYa-}vfXz>7
z=<o_r9MWa(hVFARXH`!RTg8u$US+N+RyEh7rCl$-B2<(0p1CeXsdUl2#IyP<`V#kU
zFk;k5m(_#DpH<j({2Ipmr6aXI(nUJtXb4Iq8F6q;w<Gr^sCaN)YW&!Vwn0%O5c*HV
z%NChvKms3v5cqy6IlkwTI}kyJrRN=-UssNlq|Kz96F?SC9bXO9MDCJ!u}Jeuf7dYK
z*3{`BeI4>!;>XCqK!3x+W$u2+ytNTCjLx=>A~vN1vIr!thdd4kr=VcwAin`GH1{Fs
zYWhG2)alyg4-!GNESN*e2JvTz(5g6=m<|PqwY~Kl;%Of)A<oKG!VmNB(h7)6W2~~l
zAaGq0Hjdp^9B^95FViICp~4NJAQz^n{<MFZpIuSZmQaMwbiZ*Yra=jZct+s+8S=*y
zR8FFS`4Eu*z6N=QuB#Y!NlvUiSM)zHnyO)ME#7JU0U__nYqQ#HfFoTtvp#+fCOJqB
zS6^pfVd~2qpMe?Y)F2N|6v%UzlH*#<-V!GlO}mQcis+3=3{O-`4f1Um>$3{n8g?mW
zf_pl!dCnwvv|=+Anq$?vT(8kOyf3r)F1`P=l>ni2N}igx5x1;zFE2Kg5$3Pl`d(1*
z-zy!#go>4_fJ0b_6{vD?<}YH3Og>-dP6>*1dQ$ztmIIU&LHjZlBzZJGn%)=G;K^cT
zr$|YZb8;PIvoCra4c&NH8n>;%+W45D5pgb$`P}h`thj1_>7gj3rL2*X=GPR{e3>V)
z6>BkJ<+69>6i@YwC#Z*F8h=r%s6o#8>EQ)=ChfW2^QzKkEE*XM8ck19Z*U{mGsu}y
z*>W-Isb9iHbZSH-qA0UZ{YY(ZbeEyl+{ic7a%%k}UMgaMzSXOM<DKz6GA3qX5Q`0A
zcDvx)PeO4+G%7fAj<7-6x-M8zz1V0MmRUn5<_)S+(GHe^3Qod+SRaez0)j;=Cu@gF
zGDVbbKRVRFME(t_YaWzbiCSFk2EM`k1a|OOVb{t<r2<AiTCiZ0vq<b>j^55V4d%t8
zQzHolVe?lrjbMqqk)UDoP%grX{*o&}?wc9T{ha)n;VT<Vff`0iNh}cm!zWaS+s~A$
zE`yfz*6Ux)JF&tTAL&#p-VL9(V!6Zi_Vxy!bp-7d^Np2BR4eR~mGc2LpU&qM7vj&$
zj)$O5KWS~V>S2A4kB<7mhtyc6o{c3`H1|U{YoKvPYU+A8L*@LNKi`-XcHKv<_3yM6
zk2oKEI)*dn-i)1>Ms+_1VP1H&{c)>dB+@b^3almDv8>p?x?|L;3k8pNco-d&?95z=
zag8?Do6s^7>o+b{sL=(h=6vLWGIa)_L}Nu9a3_o-39K5;erkT0WJHdE0h{dtP)P8t
z^rvJ6hvmi*b3I*N$*X@hcqYi^N3lQ11Mn$TIObXcK~#ITeD>M*>izPkDOtX&rm=0{
z2`HM{XPG%T-Dmbm08pG5K>SSv<f0XsNLdwA#70C_?K;y8I?~4AKM4mJ0|S!3Z+cg^
zmR4_Fsq5g=jO5N)>ohtHKY%V)N@+IYXdyhWs}}Y1A)1FXPN8TQTxzZ5`W>dC;$rvH
z2t6GXEyMKNN8vX03Oyi(Vvv{-M91D1h}dpEsO-V-L1E@KY&jCJ2r>LvLFW^6^k-wL
zX*YAIOb<dI{o!`Lv=YK|6~SNe2`*%Flyr4~d`!SZq5#d5(EJAgGQ}z-n?BD@Xxdv>
zD9qWu(#EU*lNLomCt87v;xsw&n=-v@zE8Y2cAS-W2I?9mZ2}$RZkM&$&pV;D>w2C?
zzw%!Q3<8R_Gqe?>bq#@eZROafkrXL1>D6;#wxI^01Ou{JlWO-(HJIPmMD|<S>W{mE
zEGQ7V7i0TQ6?wEnj6*Bce|@rQJF4_co_uvqWqr0eifx=j<x>9UV$4>wrGTQQ^A{iR
z%RbH?Fn+vxJMcq*wCu0WNLL842CfkGhRQa<kIO~UoWzY1V?C`IcP7@VJpyQN!{geE
z-}|fbws)W*WZ2yO;Xr?^TNHaK+<4u`O>g$k<h(j03zG%^XD=E4Ux`nsc}863Q63nI
zc!%(@kll&M|8De(<8P)AV)Jlb1$(ksrcE9|=53AMYrS2yRF@R8d)~{19{f)FWq|~U
zv{9ccCwTwWyTj(R`H=zuK$wA|T-vgXpb+{m=m|aNKIsg|#Z<+hB~OZBQi7v<B~d*}
z246BQK8Ak)M!8#1#NF`|U5fXSQ=n{9G^s<U-w>25=`~t4tbGTI)2uIMhl7LgyGFrn
zqFdALIyakKHN^abHEh!?@<h?x5zq`heNin_^GirN3i#=<gw*OSq9L>ce?RBz9Rhth
z)mt~V`FW(IGV1Y>lxWSG5%gO&us9zNUxrUAu}Yex%1h<Q6l1|KBHD(342)G_^qw`+
zdq~qSno=u}^k)K%LIAkE==o#4SDR}KA;e+tJbO&TJi{k00T1_xF~%+vj*OL^Yq#IU
zPdUErQYV}aUCA#A^&3vNmfA{v@ZSq`W}k_53*}g<@K%<B1+FRGkq+NeX&Vm{X%`<@
zu;UDqkcf};Q8YMJ<W|$BGOgq&9?*(U(0Alx=G(g2deC%;R*|&HwH$&^t*=h~Mse2-
z*|Vj;hwL8_Huzd1t<%g0z$@;7UhZw6&wxYZd{w4iC5GGuB)iabQyCi+Kn`j?SU_$o
zQ!a#WEaVXJpj!&l#{Yw1=wmbB`Du3eeCWmh+kR&URh>ae96*|@(nb?%@=-#1>XXFY
zLbc&DuoO~DC!^Pg)F*>7_CRpERdHXtNcUh;<HA>WJRWqIf$<3?9fPXU!$X@_D+qe%
zX&D(?_D`4g6I$vJv?^77qI!u65BOB2aL|Z(1mJ^pHA?7;j$pr;95t(D*M4KxewI`P
zZR~U2Ni_$$rr^BcNY?w45P>bAaROp5D1jGD0#KD#NPCr#idyJaDOyCX+`2OaFo-hm
z{<g)K3+yfc!s*`(ud@M(Z${A#|JkzUzr04!!{{(VbC`_F5hTLG!BXH^Z?>-71g%lD
zzW05;!=9@U^8l1hL4vN+HtWju3iGz4tJd8Qt%uuv0SKyEh3a>TaH4Eq-6};*Kbdsv
zd!JQld}S25fR{|(s1v1X&no>cB#togadt2@Xx01R*a;Nk)@*bpQL$~`t={sp6Par?
zf1(53V=T~2Cje~NEY^iZ5xsW;_y8xskvX-;^Y&TV#0t2Rax3e7k$q<v&37gNMeXg3
zq`Pm$iG$K=6IWMUa(M7`cfi$tsN%XzDr{C(YbVF2&4&t`x&aZ|?S?@f9x3~#SpHnQ
z<9tbYN$-9k;s(;#YRUp@Pdw)N_x$>-!#j0T-|3$0o3jNPk}>@^k<^%{;}~af+iP78
z)Szt*;3J_-ruuPf;KzUpPdBeA8=1Rj2og+7H^|q;)vlsyxz>-Ef)L6BMe;rgMO+&H
z9=l|r2>H32)qa~+R%U(i;_2clBo7oMtbeY&#jwAIuwN$#++at4*qL9q5LAsYDRaq<
z;y|X06>{nl@IwDGD+J1FMT=KX(d2`E4s^B47%9}K7*OFfgQ~S=*ZB7x+cr>8Ery0W
zOfO+&)4CL=?1z5-yTfOlJud`?p?GHC;JjHF^XY|lU;rEqbOLGGJI5next`}1X*yAL
zF`;AZz9D_o7vA*(YtKIrOS?w2PSC8%1qc^hXQm-9>JJn=C7*s+7vJ<FkgoZ*pTF-H
z;jC7zsy`6$eI#|%99{C+w8cN@vk}5r?D)+IW0F&0pFi+Tk{>8}Yd-BO{l08y%vC6<
zGlOU}92<w_VNopfnkif(p)<{6Zd=4r%s<jtOr&eWMEn!j)kw4Zio6Y`DeT#5Rti_4
zv+q;8{HBDUndbKj|A;IBzoJX>wbzNE-H%f<;J7yDzfw)RrWS4JMDcS$)Wv`8Zi1$+
z>%y_xvcm(Ip4`P5UVLlk?$nx6jZ3CRj6`R3%yq|yriHdyN+@$)PoB#jL$;=MFUUx&
zXP2S4K<-VK;_qYpo`$la<CarLu9o#${qW@^^BFGYo`3`z-14_QhBAgP{uNR=l_YlM
z{f@6As4m;g5JAb7N3|XdQrHe`VtB$v&7hLhUOvqyHFnIn2*;kl#Y)XH5o%o84X=8l
zN#~R|G7WYhGZpL5&#5mQ_jYK;vNgf7S)ig&{B-W0ZRZg<RCz7$bCmIZZJdjyKkDPR
zz$NvH9rN_Yv)ZyMLno&+<7htAq{8800LPZ@=-!%@gu&X(H)$4?TeLfs4r%+csqacG
zggjf>k-n15*o&KH?|onef0E&?UelK)z4zJN(xvxn+CbBWz+=RS_M5g>X#NWOdU|W+
zy?-<@QFEz^bTVlJyHPNOjH0$nkM)Mmb`Rx?XS%tgK4F!M4g@mF8tTOEruq2BCX~95
z*d3&>$Y9EYIw}|_<Pa*$0s}<&_kqjPCvtcoAgZ!CovqXzf}&>Ae_s^QH-7_VkS`5f
z(pwB1@!)}67TTW%t=*07xB-tqv(b(Obog*VTRf>T5ftgCzr7~EY#Q^9)~kY61jBYh
z1Gr?<`(OI^qOV494q*d+$tvO?lb2f5#nu{tsMKpum~;aa=U23@JyVi;+_8TwiZrl!
zzCby<RQ{-r*I1Bv|FHV>0l;bWW|&-7ugL0jfei`zcu!9q&CJZM56uOJxOX$y6B`XY
zJkq3@+v<2yR_szho4}dh3$HKv(LCY$H%a<*9lRH@5W{&S78QHzx8O?q2>Naz2Vi0N
zkj`nlG3RqM7t(rA!PMZXEBa54U}5G&O8%`ht=%50DVMBe&!J}{6`SZrC(q+i9<#%Y
zs=DL;wmzjK((yY&6``aAby~NDus-K4*1j_UY)uGx|DpR}j3<EYh>;!^%dfWZ@@2S^
z%2-8<`7{0Its!EUX+@b4SIxw?f(2OG7@GdB%1MfH4%*K?@@n{WN<D70o^#(MDqn-f
z{C=X<Pq-niYeDxt0^VurV1a=`Zy(fOcQUtKQ$v3TFvLutT+OZ><tf8r6(XFp%I<Qi
zFe#cdKQ=!el8)gIHve#-$*A)q7C=W6%zW}|S5M4*cCUW3hK_$!aVcMvzP;pn7-n~e
zl<D^*t+u6Ibv$|bTqLOKc<gTj8b%Ok1P*}yKxqW|(I29pVd(rlh(}IlTnTtkD-SF%
z^tC`*jIY&`Pe0l{Y&4<<lan!MfKkn?9cHHByL)cG^WyO&Mi9Mm`?5sDpBo;Ax?pab
zz9o8cT^4nCuaVDZ@Y%!e(_IzR^XCPzPPyY#xMm=^pCJ`fCzm=3#FpPGF^*Z+J@&wR
zVZYeF-{HZNczinHiK-i6k|5-<<i7S7a&=ND{Ydf!P%V=6;-;)V-?s`bu<?k{NQtgq
zBbZ6is>B;Pl=H0#j(&LJ1MD^v{<FB4li;}yKVkiAIH>F0p&La{V;QtcYB;XJLhltQ
zNDcUKwVcp#Kh&`Wj21;DGb7?BHw`Bfj*t3P{hB_VYZ)L{04?{IRi=us9p#Vs+<vF4
zA-MGMkvHtZNZw@vt+Hj^!)LS6C!u^BL6RJ?_^E-{?%7)L3xS`nJ&(7SK4LJ^9c-s+
zspMBRbgzBYsE&O)O}F1HNRZo@-_-9VALA{E&#s%H{Z=h#7)~$b#H7+BIH&qO<8hOr
zwK;s^w;^>a7RB4bFWsRQ0{_%?^fS}4Qjz_Cg}l)Bne90BeVi5~Lrdum{;snQwPg3t
zd9H6NTc%t7&iu!bnEnMb-KZfF%An)ebNRA|mljnzk4<IPXnSxQ>;at>!^mW@&03~t
z;jNGL`c^(y{o?LgtUtKE>uACLNAA1LaGAgI8YlwWqQ0Ic2MEzJ`uixJob<{<?Ycgu
zIq^9-;v}a^ftY}S_Cl~Tg}EsauVYJ;*!SKMW_Wm%g|h)B|9rB0)H)n&GLz^pwHol=
z|8^o;Z9XFcSl4>K>l4cREt=1hRxWOTov+rv8pm8rXnk=Y7r-N9oJDa!$@%dotqj_@
z_``aJFd9&!dKcd^#_oDgmjBwi0JSTaY@$W+b{)Hv>n!~uH>&pAJan6ka<}LC1tKmu
zYDAoND>7C6l1b0jW5c?+2ORr4GP3zEjy0?SDMY*JuChh5`AV-`@3TCC<jz=agHMG<
zay=dT{mCV%BM5Ec--M?>-l?HS2W?_+EymIR`q};K!ZVo4I`?g4R17FL-aepv)oazw
z(=8%JN{~?aTTCu7cumM4lIiFbe6OaN`JIfM+@!}W7ADp%ywn>iCeto|PFE%MWOu<N
zUxqCS55~Zw8pchiQyKuca70;gbnAfX4VO_vVS5}kPAP`(3>k_&o7Dnxm!6dN3u-z3
z8xve+JsGf_=0}2*Fw#*=#ybG-1DpfEhp8)yNXAzKhNG%l$b(NdR0atey-#=EcQoDK
z5B#~Un^kKF>&1tbSzEoL?*t8|V4l}ijRC}LCvde!6%|fqL0&p*l*w=2y!nSu4Z!db
zvuR)@_zH%NG0{?+!2@19Y2ED1_h5@^kH<ELbW?&hy($pEr@k8PqfXK3miIYJpy#Qa
zJ8%Ri)gY~Q(}~32>Ast(E~=~R(1kGL1K2nQ;dustD6U}*<>4`;_f;#^(uUL1sE2x?
z&B=1T2e=`MgJaFHX3x2}U2kZMKNJ-k?Yn^Tc%3#-O!?25&HV-zMHcN{biD&&Smz}g
zPWZ9cxm0EA?b7{_`4fVlU`nzPRAmt&a8I6)kN`DwL-jwH7Wpsf?l+5>x8B-&qj?^;
z(DV^|qp$VBC`n{E;ahueiH<0hv!pdX?o^%c`##`6A(OuM#lAm7_%Fzl|F+i+wisJ2
zEv1R3Ng&2?3!$hH{}9M-xvqugU@Sl&eO1e~uUfU%=6hTGjG=DvC^);){MX035JLp2
zT=~CuS3(#R(l8lp%dWc>I-fVM+<|)soM-3JqC$gP@Lhq#r=^U3Oi@dLW^hh_##w<|
zgyOo9`Ci}W>B^Us#&FC#8K^iOGoid@=f}QuZb0~f#)um3<4)}!H>MU9+~+Rld0KD`
z^)a;ZRtCAnUMBJ1`AhX-)U51Uw>@qJPBs%k;^T}D+SUl=N!4Ff8m>NEo&LohFm}<e
zMJ8ux=^>YvGW^a(%-<}=6%y~CL7ehYZj>x+23wp5FyD2$>Q~#MBltZXA+4;zD$mbc
zANqXnk@RmTQ+}Luqm>bNZ!W-R-?sQ}ytQ`+x41phIGVet+|zPXZ)bHh@$-zs=+f9z
zFSq^X`IE;>7$7SlJNDRZx&BGR-v#ywrR`1h%o@AIwj4j~BgrN29R=*S&u=DopGe=Y
ze8LV7q(;m7H9(Wqo&B(ot=b2pGKvoK!p<#p`>v7<yP`DOaOq=~k$(HT@bs#VAkwWl
zc5dzgKz~HNc7z`B?`He*U#&O`Fl5zYiLzCrkWBbMTdz}mZ>*t7F|Vt^GQb_;fSPbM
zTisaQv(DQ)u2*4nZn9U`p=#j$ymT(W<%B`BRmx<4|LzghJyA9OB%>{RXj}i0ko0c8
z?|QnJUv*n2xm!Fo_;PvTD>r^NK6yws1Le=eL6*YpQWZFCqe;fd#y5zoA^$mh^vV4y
zo|jyvS>;FDq9jEJLbyx>l0{>FFTDo#BHe6?_a!L5WPFpCe6EFkKJ~?XpPjI(*I@Zh
z<azjAJPD81+WoFB2o0A!a1EsY)`OVusnx2qdVq#xzb)gm$?hQE6y@Jq03V3DZxZNy
z!~2NIqErVFX3l@xxZ=n=_p~O4?juv}c8UevjQi|Y_rP@dA(3`Mb-B~|3HMhcjUbAy
zLIimvuEhKpKDYy{con0Hj}l=M+?w%FH>KWFF#Gi%(-f!ux%O`VFzUKKEpSS$=Xq<1
z&#S$SfGszDIR79BpF+{^Hc}?P+l}^wk&av-Az139fJPC7fYwJB*~`_(Ka`wXGrc>B
z1*f`0&EADL>24hW8gD)~$WI)zH@To8y{?W|qwjomabDTj&A=Fe7kSFt@xb}_hcjTD
z{Qd0y$s>H<<pVg+myG7Vm4L}=o6ABbzy8ayy4w{(YGkvXN3TA8A(Jw&z~tT`o<^6!
zN%f8FC2C+%%9Vr?e)=8@H+w`vG5|;i(^qI<QR5S(cjC@Qkk7_j2aKY5uz&b8A-?3F
zno0-M#QHU^;S5O5N}wIV0!@?iV?5PTsMaIlbN_+psw0E4tD7<3teG<PI}D$e5inl4
z&xhIlK4ukDNzxB>kR^k859Bt0*Uj?>7=cX(l`U?^o~zPnMM<LO)h*Amdk!M>?<v5R
zh7<+_ydWHgiV^B&p;T+GT@CPl>R)%`wQg@N*0(-A9Gl;qZMQ-y>`CaI&CH618;iIA
zcj_M@s;(;mjJlpXH4@1u%(bH{sEpAScpB(ms43#8rUz+Pa6q|l*<kqv>_oTIjX$-k
z2N0NSyb5q^(-8P^tu$iwBTo5?M1R3of#ucu<RVGR)(^K3C`U~*Jlir2^zZ252{{00
zRV|<nfbE5LW;W=?ZN$0-EtWUeeFAc6MLM8y4~FkzTo~HDo785!(#sm^<OgPi;LRA`
ztcVUbR**G-JE2&ec)T0@HKF;RvMs`BwiIi{opSmK3m(moZ@+O!p9Dy^!|i=um7i9A
zRyqSZut0Es&*M{C;mP5jjwAHf8kX!fyV01pfXJI0X4Z-y@&fNHZLT#KI+kHT{X`vx
zx3WFrZp&hO%vs0dVOz%KA7$<>cO-UJ)~*nNYuLw=mE#Rc(i>!`M>Bh^0a0-@I^D@n
zRWR9`$J+bp?|gpDMoG$Rzj*(!?-=U(5R$}`hcYmEeE};hbg6R&I*ae_Q(~9DonJ(b
zK6JN~>g*~THhp}OM)sp~(tBX2G6E0k5PVG^R{e#JjI=Z%E?*nC2$%vKE^`GTmUqT_
zm)xKAYrgtE?2dh5HwoJqvFA%ymj6({{Bu20%z*nWG{pg6vz&JbiZ>&BakwQHl5@qa
zrH@6Z;o<C4ZNoFz^$z;-rIQ)I-S!9}ua0q~F)AV-jtdPO8J3>Yg#0XL3)6tm1YK}P
zvvLt@;e5?B5&W?oR{?fmdi`Kl_nX1^*4kEh#tUz7Fe`qO3Al_`0uBPc59)X<vjpee
z7ELPxzPI|&m+J>0HSR~&e}5Ai;ASnbsf&T&-*J8J%zZHY-r>Y=?3sa$O(Ys%HT*t}
z{(SO1#yisV?ogM`znO(<VC@zM8B0|eorD-Nna_ROpWoG$7Zh>x$T4?jc;^cPm7f4;
z1<XI$D4@&sKK3Myu~{B&zdJmrg(n+tB9~NkY0dSrCRXDe<=ceJaLd;!>D2nz6)YVY
z)>BH?{P`e!q|Sd8{kb)u<!^HY>BF9QT`dN92k}5WzZDmiuGwnAs{b%(O6;4VAKJL;
ze7PI*`{DKP^Q^0r@?R}bb9X`f*Y3Q*RyQH!jNCyEvwEV^tB**;GRbxG)sOtiRr4KI
zyvMq9gD!<jjelOiS+9sHz<0>(3Tmjxo+yo~-tCS4{BKRpeIA>=NCl7HG*-V8$GE>E
zGgF|XdB`s+vZ=)V-{_z3xB7IR0*Dl)9LGN*Ih_sA)Ef49L-FR&{M(GR4>}9ZOwE1M
zo#Sm)d$OtL@h?MWH-0?*L-hz0_!Qp<qWj;+Z6t{zCGf^?g=p<M9$f8O_d8NE_ty&y
zPB*B*$qH05?9k)88Y;CmF8spOri*Jg!x^JYtzrl0m7pjlm}Q5}isg&d#<NMvn6+Qb
zK8mRU^W^b??XwbeASY-a&Ho3#vgz_pMt;P|MyjR$OIe5K+ddLscv03(-eWZt6xE<=
zu54OV$-&PZpX;+r#yn+YWO8Jy!WsX7gP?g^*sG@t*h}|I=XuZ7500a8rUXLQNnt|L
zlhaj%?1d$T^<vht_6<Xz*EJK}8kH(7>^m;cF=Zr~^=H~Me<&SC7=tI&-uG^4x!&Ws
zSRvzr_X}t`2}T-^*#vxjD5Un&9e0JAmgpSlA<Gz)sD&ecE5scyGI=Z)XBIC<lRuvP
z{ghM7I;*}<3EDTz@G#!j+iOv1n}NO8ilmfRA-s#={=QF3RsbrQz9LT=ODMz&qeE3d
zE%=2JP)O_i)@7iEpfiF9vMf8xlU?VD1XX)-SkHpW0w@L-AbyN(Y-}tr2^a5o((QM^
zF&QBYnAVx3=zxivxcz3t-l^yLdxOaJcxk3}S$mHug})3Z-3ws*$zFF1j>#VHj>kZ@
z<)x6rgeczW(}L$YZCSYU!;^CT6~FKcblIs-?DA2=w8*sK&*Us{euk1^G^vM4zT1?I
zWc5J!$wlob6)6_L3Mzu%v5G}ueZ+)E!i@dCZ)Zl3Rz%>zWBx+5L%`~IrI_RWfbKzy
z&ro_s-%Qzy;;TyCDQ>lybN(@2>sdxb4B~!(kiUVl@}2lBM*wg;mG?6fnCLgrz#h8&
zxUZ=tucRA5V?bZ*vrdrATxb}HZ2)#ox6S9km>kYkoV-u@%Kj(iD=3Wv*B6(hFASwf
zWe{^t>f2WvLf|CeJ)={1G~IbBelI12eED~UsH`xrU+4=IaEOlI31E+#j<9=kRiwZ%
z2~8bANzUGPm)8Vs9uT9ktPdB53)pnj*0bg7pzJdTa9`VtfSnFHzP}syIe|O(%Qkuu
zo30^`Q{+!0GSGDcL~s`PzWJX{zHR-1rl?<ofBHJ!Co1(53rXioY^<5p+}{@qY=W#5
zw0cx6?`exh0LrR`o=t{JW&tL-AHaL2Km+6o?Z(L}eF4D5Ridi|!W<~R6b8_6Nf?-&
z!mukinA<L}kY2pr0w+qF4Ei^yP>lvRaGkh;&xIc5{Qw&{Zt%?)>1Oi1HB`hd64I5`
zqz0h4M*5^Ypjp;y^Kb;NCM8A1kBxRKv`kEhYr6tZ?lNZEmCq>6{dG8e6NfHFd5?d(
z==zgc>^3VXD0Bj!0N1BIOkntU(>F!4h}Hya%QI?A^srLH2okd&JuVLiF7~H3&GB$>
z#sMDkdR0-|IiQCBCJRter>9wbAH4v9RUnC%GgFQoh(9r2^GUe7-w23WY!8zOlkLf`
zVsF<>?!{6mn8n;Yv5W)l^y}TC^mhM5F~@*eWP)h@>jTf#J0|Qm8DPzA{j#aCXYqlg
zK_E#EAUxD-9oE@_m3%4jSqYh-mKyQJ%_Qklzu&XFAIFcLH<Qob!>|Bu*J#$eFc5nf
zLcziy(=`GxZm`vvu}Tw!fuTb~-tD02#{GEZO5zO(AyEt~8I_aV2N%9J&nqZNDbFE`
zN#le{D+HCh;yVcg)BEHhOJO-xb#=$xOy{L2Zp;{aAQ_#nCZC=mDF8(5A1k&u^O2<|
z0NZDJcX<Tl5m4K@!g8+gvVf=S@eTl|Zi1iprrH43?&=7csRg#irJK_jUm16I?&K9Q
ztN_9O*u;jTSP5lhx=kLhUEZf(FMQodV?I2(wVdKU9Zp9Mn~P|har8JL?z9RsgeF=_
zk(o%iT1v8=rt*yrf^;>2hQ;tb`gWR|y}~5nw>F(-rietBDxP|SP=l$F`|zwmRyr18
zZ9KQ9_36>7ZWcVg5VTPNyp#RaDw=~i?=Pr2#Fn{*A`E^?gwi~~tB4jTMjKNzDW~H+
z!tsS1!D|(yQq>#>>573%4FJpUPvw~%?=`StL`6j-KbZ0XGey^xw_2snK(H*4X7eW$
zRaI55$C`b3DRzq!1H|XA2}y6ObeiI4cf@{sAA74+Y4!{livvXue!{xCv<-vbE?V#l
zftQ6;L%)4eI_E54Wupi%=16~G@zeS#{T!qsf2ItN`i_{_m1-GBl5GK3DffBy0RmMf
zj<@HP>Db?AIoKtgVq~Yfk`((nN%qsa>-5vfVs8^ABu!=uMepkB5OH3)V$6Jk^{cA3
z4JkhkN~c2bTVV9DJ3~<Ublf%h>Nu{N7)%{#ou%6xUns`SHut&YB;4TV8)TKJXl1mC
z8!3ihnW@zI?#p97O`;RB1U}rw__+!K0h`$~@}MB<DHC?s{lI`SxLV5^l0@opHot>g
zV(wWPQFrcd{uDZm2qbo;bvPD<6O)+a*(yk)=?3SE7!`D)%e#KXv2e8*FOGFvTy14G
zK<>1#U~1n@nIy@2sW)pKXaO*an?``$vZTEHXM-)&Q;3R+ssXl-U?u|3@-C8u7aWcO
zo}mfIogUd-(=#x@0<IQf2nI=BLj&&W)k3{Bp7ZYLHWc0c>(?(GZ^(=UohGoF0S{c?
zmAC8Rd>2r@<FgpR)S8Xl;yAto0wb86vFug5+BSgi29%wdzc~~XJrwccQj&nTW`pw8
zbN(en925T2(PENt2*xjy3`~C&dQrAio6odZ@E^<T!rR4RfEjq|Z#89#AeInRr1NVl
zcQy@rfSQ*_0hp`{ZsinMm0+GF;dO<vMdFpB1HNIGl^^Kj@KC%a6d#`&K~8=zDa24J
zZ1*HV_pP*aV=F;kEU2`uWVGQ3lU9hA*m*BDMr!!KFTXFxeSoq7-6Bc|6=y89A<fM}
zp6IYA1S9t4OG>x%5$aYDR`Ol1^#H7A;Acy)fbu=wb?gG5SqnP80zsv%O_A&!UkY?n
z1I{W^7Aq9<Ab>(vxeg3yub5j;;Dj`mUFKGxm}{`74~>rpOo?l1fP>?PO6<M@JRgST
z)GQ~tjKTC~jKC-$NE_GCCs+c!X$(}F3KpDQfb(VJ7$x@x?(gbqe3nYQWa>>wzAX*E
zOIso*kW+!9noR*frn3-t+tyRp*B=Xh-uu!JK+!IMDpj{FJw4qW)RJv-WX-05B?q46
zaywrv^Cz?KjdYQGwjZXNh5(Cd!f6bq+HzIXgJJpJ@a%nQStZa-*yR9Pz2mqxQ=$@p
zLf6`@h=V^GfGy7#=ksv;Yx}XHli8M!4)X}P;_FM+a6&2yuQ4pRFfsZ#b_U9t^<%OP
zD~gtZ&{{12U-oa;Jy$#Gj~0oY?Dr=NVG7XD@OixO$2v|tjWr3PtNJa+E(p{efR%&o
zE|?iEH&b8Yr$q%g_<4)wbAtnPniPa1D_6gtb%Hk(8jVvI>R$g+M*GTXOW3&Vga!z^
zb_@a(!V@+tHK|bV9{^X6=L#d`Ck~luUQKw6@}0}vUK|X8&!{`FuHXpVOqHNU*_0D5
zq|P<%AU}8N^O~Jee^_A+$#B%~0J05N7LN-H@Fp5(5vl4T>v`PB%VfIJJvpxUI)V9&
z&1RlOXLWN&yIoq-{e6)R&np(#qz|5#CVS(#d>8#aP<SIy9I!iX{B-l$h-<8ve|+bm
zR@Z24V`tRGoV~%j+D7mN_8H>+1NT`YW*K5WUH+m^;5%m=L8)<AgH-Oar`5R1{{;w7
zZ@tVU-kOZ&X@|V@lKL{Kff6<gn}x?TPK_Uf9HHBCH9479I4<dYscCK~jSNTn>W54w
zTdJ_ja}Bw865gvC%OTEaVs0_a>kQs~o5Zu1Jc4Ckz$L=50LPG8m7c7s>MN+BL`ym?
zUHhZ7l#~=cfP;g1S&WBM>;IVg3aBc#u3b`68U*QX5RmTf7NiAf>6Gs74(S%9W78oh
z-6GxHsdU`6zrQYHcpT>p1<vN(Yt8v&)~~gnk+}PlMMw@0|JaWnZ%qB3ws1lA^by|*
zE7dC%zJes>EM923A7`evh+bD>-{$DMcdomj&k9;5Z`LC&HwYT^tnJTl9Y<w~@m!DI
zDe0!O-H3;U<VZ1|J-#{3F!kUNhm%S=**2dhy&;?(RWr~rw&f0?$YG*7d|}4*Wryz7
zs>4Qir`z6y3I%dd0t_C7+e0qaPoxJ>$!CFoTDc-NYrniV*?~7ZoS;DGq=ERh$wHn+
zLyq#i16`%ZVk=_(XvsUFg7{~ehxgS{j}ysI+jZ&JugLuz*fwb?#mS5h1elDt0hIU1
zG(_ytwB7Q>XFh>hS|p-VF-i#2<a)$<MFASGSKs3mN2?hn54M#2AV})wB5+h0>bXYH
z(*LZM;Sr8f7Dv6#chjjjrLw`>b`waq&Bqt_d(3^pwsV0>$!@Xd8efl*MFK5_V&P{G
zPz!+8VrfQcb8}h{w%5s<sTAg>-CQ*cW)`euaHOH4p)HcqWP2TygM{jEyG6)37-2d5
z9y}UmKqDXVB9GBU@Qat9UeSKiB>?pbKF{~7yU<=iD9s451G#vxodAJJmZNbrOlPiY
zhO6Wr2T}mrw)ddv96)e`-GT6+z-%1~=YYW5+QLsD_fzP4$%W(mt#LODxCl&p=oc!a
zx0Iw%wLv$6mKGvmN%raeF`lIaGWg!gtE=M&t0|B{k6%Gc3%^Lh&DIa8>2%PZ1Kv0_
zpu3PxC3}tyRmUcF*Avu*6gKubkqC{O{s=-mhaCZ(yF>!64B(n45paz#<CN#gijDkg
z^*|*})SW;nIoDu~3l5!F%=d+IYX(wgsk93h^(|G=DPcdCfT*Td79D<>fcEHKWXd__
zQNcB1qI2%QMwp9%%@0_G%QliYA#rY2C}d<{_<b9dJW$DOK>YZys|t|tCJhY@oRjL9
z#+6)Qa`+?^T&0x{^$?CBD(pK%i+Zxb;$YO{aV0ekRuIa)Id7Tw3xaF)61+g|w>vnt
zf<I3EuJ0$**D$mFpKe(U?|8tUDlnCm)VuAB%6qTt2?z4@DSN|W*Ve<yEfyNep?^z7
z(2nZgzn8X)+Pwhgm(bM2g^uLlWCF+O@3}cSo-Fg#PI%W-BeL{Y;H6C69!}r%Bm+Go
zZ{NO6kShl2$=t$%g23Kz%qC3(>6t&Fp`osO6M-r@0+SPtDf9J~rn*!{<F5z^H1yko
z53909p~ZQ{DI}sRQsUyUdT6$S;@2m_y>}VDg?97_Qy9V6q?=Uh%(jbB!*<rz>)=@c
z!xxCcx3aVQ<&*^QFlr}}%7_Zopa9SU0SA*T9~Bh^CjZjT&Omm*^SiUxq(~xnojrBM
zloS#@LpT|pIB3NZJ*-3Z`25&WJX9y-NtMsPL!^IQLm-fe>SwWs>^nyD9aBZcur@#A
zh_ZAeuR#X1#AYKb?O!PcF%hWri9PbgR@S3nDNalSa`NK0Ep@wNS>wOEyk5{JiZL?m
zPay`-I<QJo-!5OWuX&RB-)!bV>sNHHu2Ahe{%J==M@M?F4nv4sd$7iGUD-@k$ofGQ
z$#{dcrtaU_Uoph~O#Q@J^9`QN@lPXPlMfCKzE)S4ftU2|&I^NFBxdFdJQAk5`sC=A
zyLMReV5<nI1GVGbfh2GJP>ta~``qvQ94OHFu+5$+ph(o;*74N$;TC8c2deXQ<<y;I
zAH=v$$J%awEMJ4i`W2Y3V0t{Q`ifdx({4X6kAiu=Wv?*0`Mi%9h-(s$`yxF)4Y+S!
ztbQKZj--Ee)=Mr7(+w!(sF|PU&%KOAl(Phy)QhUrW^qS~VIkIh^i<z_|M%fu#mFPo
zh)#GD;3!F8_jY8xSqyUPsZS6Y5<A4?emoy-&JV&bCkkp{N(>NC>0^Qp^t6L&a9sE1
zLMs<X#C<`KF|}-*6e3YE9`eIk4G2FbABVQG+1o=4&SHpMa7xS#2|?O+)nOl+B8-VX
z@}xzXx`u`V=uyv9O3M>{qJhpVV6#JDs#C^8fBkMSANPeCI8Y{gU#7V4gBHVzb;n2S
znabr6=~p@8*JRr?N%C~LB5`0*Eo@x-8@cJoHCp@|BR5N&IWU1=w=h&D8m5vnhFmnR
zk-Vw^jDw&{gk(5MiwdYjOqXS|oNQg>s#9sq=?BaVk|G<_U2pAujM+loAdsz5pVx^%
zT&@0r!Qd~`W?$ULhh`226ZNKNiYy-Of=1&h{<36+IgfdC1)7AOzFNDeD0CtX9UYK?
z>M9U9I~_F<1~V~~KvA9&Yl>7%YD&vc5;`HTVV%|2X94+hVs`vRnI>5t4pDbJHj>~?
zvCNuRzHhQ4zuI0xzO!L^%wd+53hCM8UF^|vN3_ehXnN2F70w`n9$&GBG@lW1=Bt1G
z^w<b=pOp0o;9{p1Zq5Zum5F(5BY2Bu4P2D7*%fV+6&0h2#2KMI<6yZ5V|&I80}G37
zS*P=VQgHieG~C>{L2j{u=XNTKfmITPY>0Bhk5f=qnJt)~pC1qikJR_+8&pXQ+Ep_F
z{vK{NwdHXCuqzsW*xLSgO7Uo^g=>tpq;V&bKG%5qt2(qa*y$ehHR11O0iZ~`LME#4
zy3jypvMW^sWb!5pp0?wHj#W@s{0b5M4BQJr&}3!%;9XDXEyhMi_4G8;Ealq{I2GvF
z4@!hd<4vUKL_okaGUeW46FX4oZ0t=ISyd8oe3Z%%BL!2$*KgmVhQarY3zG1uQWXkP
zqFloNqjZ4szf)6F?w%xt?H}iDc{;U`#X;8yXk?Djas!`;@(iEc?}n0E216F{KocSw
zh&pop(e`EXf8Y}Ej|b)r|3Yjy(UH&a0X;v~!PH#(q07$%Bm0aU6%<S}u(NOOO>P3p
z(c_E)yq(R@H-51gwex9v!XO0d+Wb9rsOZ5Y;S)Pu>4>-~JYM_^HCkK(#l;0sgw}$^
z1r*fGcX;zwR#qZYLTL?fF=FNje%?uaZ?vAGsj~j6>jn`Ffn;C_U5jF78@G;J1C?f*
zD?nlaWSXqt_A&>`PM;H(yiEWexb9Dd_Vh@AS+W;QQnzQln3*EJ0>WOdn2Tkb?hwXC
z8bn0vFJHdQb@~c|yT<i!CKe>mx4IuIl&R$Gfn!CmPC)^>hk*S945g7AjCdusv3Hs_
zHwW(>%yvd}+W}b(UUyL70Iq*og~(l^#S4lzY;by6|Gw+uLMpe99FGe9SeZCsBN`hU
zoqWL!q;!3<45n~2Qi=O!1Lw7Wv#-=aPt72GJVw8YY?bHnf&C(<-%C3meEuL&qQ>*&
z!y!36ytqH2q1oXm&ZvJTrl)^451IHUZ)J%3i>yBE@NPb-DTp^4jgyvr70{3Xnfzcx
zDD(cD{C>U;y@1<tj3n3Z&Kf|SeJ9IppsIiv{Hzfou1TQEOz^Um6c^(<w4c8)?T_0A
zP{<}=luY_!h?Md~NEwws7`aY{l8E?Z0Antv-NHNQX8}NEFGykEU+M5}e@_IOWT1_D
z02G7H-2k@wS^V*>jJk-0so_n0KgrWma})7mf)pqapvqLzP)h*vlp3=^oNwR1L)!%a
z?$TSwV&wnm*1dI^1NI4cBn)HlABBJ%@@TnDFjtX@X1$o^$$(G;Qu&|vapeX0A4Kc!
zi_Ibu@!135RA@5oyX(Ph&rC(zM(pnv7)|p^pN{;!ymtNz;gcnM*1J(k)N8TXH^7&%
zv0*MZYgKLx=(~o$@@vn7M>@cY`;`y-h>{XctK4Ha&Ygw4C|}48&_?II3Cffp`{h>H
zf+&ekDcUyz?}X<;abEm3(Dm;0V!n5ok}m>sU34<e2%wd?_<=;eHrMww;<l{QVzS7i
zYwo-yXTmZJZ8EWy?L!H`jr#>ErwIk!(W%2<{<?%?*?(g@;P~`+>N~(~!ry=JO(hfd
zavYpy^lAT$soR7<>jo7%Q_G<Dy9o!0CAi|8TMS(LzJ*hqvqH(@wU5$bmzgbq5%+zh
z)Q}&{5A4`~xA0Ptj**3p?o3&w5+sCAGeAPO8<b-ykR5%=$I+~;Z7I(Tmu3j+=4Ego
zlD!390&WLVO;zSB3vC%;y|VM2zFhUyZ9Fs~Niw5UuC{+Hbd#hwvA^rE^!L>vcG@h!
z^Q4(?z7T<+Yl$E<G?)#TeUTSP8WQ(qU}BS|BZ(3q*!1j6(ol#DmFC$@cJ)cxhf5)_
zNZv;z5%tUN17Fr2tsi67<z@}!v^v+=ixx4-6c~`}%Yot=40l0pwUg7m{%d#~ltmrO
zj}aK<N~A#jpad|hijPq#!eB;!$MlD-?l0)NXDok*I)3>vs0=LKq@G2M7GUplA_ZC)
z$<<EZTDvSFE(=G%lNL4Z6oivSX&Fz331|dJheT`85&Z)I>oCC9oeW>~JAp%gD{J+E
zZq~@#&LF<Dlny$&r)bzNH7AZ{Dd)C=sPi0rW^KX^U1Vq}CCD$v$L(sl@=E-qs4xiH
zzlp#9)+kT8jVaRrEkO7b?@sv<0@c-JMAgaaBCKK#c3auNcWuj~=;Fc^j!C`)4rn<>
zMkXcz??4k$DXWrC@1dnUjW+MWxd}D`%0#&a>Pj0M8_(PAbkjdXgmCPMasaohI@Eag
z4sU!PfDz}QO3G=`uIbltT&HHY$9h{;^yxi-pr7t$yZZ8~rKqqM)L(DwRN;*H7%i}{
zD<fEY-xvJv5c2^cPbmM-@oLk;_H+;nS$|gCWUb^PQ9k5<v;ashC&WhVhzzS+r=B+F
zS<rGL@P5pI$u*JO|NWgFVvZ-E$9VaABG8F@py4gC9+g_H*4l4XRtiHkvFi(a9XPf&
zS0;eZ!zroEfhMAh685ndu-oQ_^L5tLh@qjS1|@2K%_8Wt&k)*7w{4{{zQ?yA+HXlk
z!|>G9S7N)`?@)m{6M89U^PjuAAJ2^mi#A8g`d@K1XdErn20uOCXOWCJEh|^$4U75{
zb?%x7VnBh~IKdNRP`CsdL=wPY3ASyV?~SjCiJ=G@z-P9K)v0SGs`(N%jh0peHzxCh
zB*1k3;rVX%c3;(hDD2-MmL9eTn|yy?Qqa4eIV-+2sX}Z$>PduL;_J^sOA?OrC7PQe
z^KF)xP0Q}Yl40oUV5KJEbEMEdv9i0;0C*mBP{<bQOCU~*QN3AIEvO;2N}y^ulYEM8
z{;H;4GB}^WjIdCQaPvV%V_)uHswU?6C4(Hja+eH$k8Agx8}VG11ADrG4>s1q6+b~5
z?SO?I0~_bJ!X+U~aV)iW8KsRSD1tO9JhpS)VBi=`Ruxh)F`>lYHkT{N6xiPIh?5DK
zfl7GO?K6}?f2OxIFT{Rc%E`%zBp_W;0u^@F$OS)nOavDnp#kzX(UFB|XxL-@x+@w5
z=;vpg4j4DTu0T@nPus$QqKmK}vH>G?kBcyF{=2mAeogqfH$&F$ajQB0r93NTPbZiz
zj1@07$(P5|Mv3{avt8zzov5j0(M^Az6=d^$LP$4oMgee8+@)_ksAfE{)AwBK7u|wV
zV6-`cVLUMfZDF><?3VgFji$pMqSpEZE+@@7PzTN;&eFy5DU2g{xND7n1i9aL`@e{S
z>|0tCqnE4CKYf|_qk%jju`OTu-%s-m?=Q=!=_q55U<gfD3(}o2L1rM<(PC4)El;$0
z+1W$Ie29FJLXsXu4ubs*eYfRpl(`&QLJ+Gy<jsBsR}CxDS#2dAla)RVCgGa^pp@Ob
z7Xf{Ee1ixOJ5b;;vd&3&H^$IIB8gBSSB54>6O;NS>tAh2i#B6D|6kv{&#Q)m&B>|~
za^gK4Em_4O<>@|ghHY18Q+dzbo$;8LL&MQTT=p!xd7rTXwYt&EO<Z2S5=+gRa<SR@
z=FtS>K);Sia@sm?EJuOwvVu!u_4X|*>!>T^vD9oGj*l4HVBB$=7({K3Qre^-R)cL2
zPA#{uZh=RiH4{CJHK8-y84vc2hX-wWNHIl@7?Q5fY$!}s4tMbO@M?j@`<EaJeXRjy
zGm%rCE*Um4pTchXos7wLARZq-O674#a0-uv?i`9dXJ$8qHAGD(O<mk=FQZhltn=?H
zjKH*`xem0&p(Ld&M3*Udm`H65@rfw&tR=&=GDm5#)`1_mK^424W_x0-x$c~y`*@Kl
zl2~T)^915_(RPiAs4WWlR`tI{_FcZxDG=U^E9x;|0;wMPcTeZXN$TN_uQ7lK#h7(7
zb!MOnppt%1l^B69YJ4GmywZ^cel$N(OP8d6tN*lm*Y$P80EhKjS4`>CP8vS8EkMAb
zg_%Ico(#71<k8n~PM`_W07SimhN&<>zSe*^oSm^8Vt*BvRBf=RbCl!i3K)UG`}mzs
zBGNs;Ho&q)hYSa}4tapF@6FXx0}g~E6F-$SQE%Rs0CWKQJ@EC%k(zG&j07jD<lEgX
z8ML4P0B2H!<b(Za`p+L&ndN`rmUjfSye3$^fC@W4g%d*Gz!D9nBWEDP_}%u|0qpaH
zToy`7dfYrbA9V?2<46jco99O0hocr>H41u>mN!!c`K)yZ+Ey<^YmA{tP{>f=ai3Z_
z45hSS*Yp3K*2w<5{--;Rvn2n%$-z*+<-rL3B~7yA)BTk+L?YDFqL_7(fi2S=C4ES7
zvySX>z6hO&BLq4|)Q<}gvuL{K!+ohAg-J<E->`|V)Y8{&`9zEPxJ8+w&4BHAagz*-
z+|w9Ie&e%W%6`7M`jjSy&<EQaWM+cHs??HZ9(q8khG61YpgWs(%>tUXSIkCkIBpiW
zPSFS*dq831gGT9iwn{%6X*r2TB(a&3ZTP9rso+Q!Z#=l#VEcJ&Z0)QyTTyFp2m-Gd
zPj$z>VQO=1ej|d;*I)>RHuM0!D#`5bly$oEB?w@-&g#U9X#Ozb{hEbgUoU9a`>kb`
z(-cPT6><p57v>|K2!berCPWi7gYClNo_^jodt$hUjF9Lde2shw*`&G?^9rM;FU%H(
z;_+m{dn+A+U^_4#+@LKz1(kHeY@2C|_Tp&xyHy%`p;A)Bfon)25c8sj9tmo-{Fmy+
z(w@0J*&{95fgJpFcyOsjdAE633#Q>A#)2BmKF53*{c$`27!neS*f(rVKG%Xm&kvVh
zyDL2H7U4+qgJAkGF6mJMQ}W?3t%-WRpWmprrlsY>#Y1t!l{RL8BZv2dGxx=k1~;%Y
zXq?y0k3yU`-*^6W@Vg`yOH`nRk#=&T+@oli`(1$8Ou3w*AvatIj_?(uD*{uQJh`<t
z9f&ehs<GH}Cyc>E3tA?>seqxPA|YQsc^WOUPx_)mTPk!=!KRi4jXxw`Ep*lG8NM>G
zn;wUMZ*jJ`=RneF{~6P!ZL!qlf5=)|Ey<sXl@BLMBaQ!|W1rb)cX724JltT|&mr!*
zpjBvd-=3Q+QsQ$H61|&zY{yS)v6*F4Xmc%fDZ|KzN1c*yxWLw+npo!#CDJE7KHTub
zc{tcFBlMkOxchhf^6MN%Lxs&NZT~YeU>QUEqxU@W02?XMNNuR~aBEX-IGSU8%h2^d
zC-~=ImPyl5Sc!Tl{T#RAkC5V~oqT?ZRJNa~GA|r-2^eDW=uD~?4}uzMS;9>uWor%|
zuM^XP3C`|C+Xqsj=>W!HJa6KxAy<-Q#>pDc15u7E+1QW>qA;G)wRFA<`CVftI8tKP
zj}Ua&Y6Yt?=b>Ptk8*XostOQyw{Q#?$GtU?*7twAFNKaAd)MUxPeFgY_1*R1kJ-+v
zmroBvhv*_E$cBo{SQ?s>MS=j@s6EKJ?zys26VGP<qd<e=6PqRi2BgU{Z*~CZL?<Wg
zQP|+g#sydVCIod~^1I$(fdR{;5t6{hRJ^Vu46s{NRJj+7DOSTMDgO)_aNcq^CX9M9
ztQMW?z)Ei$p4E-YnG{00Z9k4@c%T-k@@MwCdGmQ45t0g&$k8O3`NWccoAIesB4g|S
z0f8^35&4@A3GEhVlzgG1k@Qa(2NJpWHi|$xpnsiO89H$*K~l{E)#(t$#RL-jAS5AY
z13Zmucc`ng2=)P|_ivjZ00fXwR7A_U``3r@o4$xUaEea@w^_GArAhOXxLGXGqqfzu
z#U*7ZjaX(jBFaYzfS3#9yg>hfo2ZA)p=oFsYqy9_CX@j{J)II<=|mZKuiV|lL9=E6
zfOrCP9+0n`t%-vUw+TH^+1>u;vlx}1Q5J19>9uO{7Kksi7<~=wf+2Q`Kzi^KwMq^f
zf0;UDt)hMm>JR#lHPm>n#v~g!HK1Pw)MB1bYpVVlk8%Da&j`vq8$j2#91Q~l14Z(n
z<K-xCvO<Z7G$9-)O9K%3?H+Y9Ok2aWjyC}TL0^cD*?p$sr<N*3Q&Vc$c(M|Q)y}_%
zv+Q(m)E8q;+07q5pDYnEPlsT&gEmp<2m#zgeJw3eSqEn(`nb>~x`qNcsrvy;6O0uE
zMAo!bZFY;b7c6)75uNnTm4wt1-$6t5B4mHHiyc_p@W+L(g+Ona8RsBaHTGAVR8^j{
zMzhd`j~Zv5zyE#lf7BAGk3=?p-ui3j!qF<-a8ziQxOUeQ+1^>i`;2z3)y{2;Qk9$`
zYg`pc*gzhB;V$P*<%f&eu8lZP7tg~AF6>GFsgJw4%P_7-OGJUjXA=eJWCD2aZ|A0T
zl4i;ImD0xyv8H;oL{@KhcyE?c6bW_skpqg9`x#kS-rh>WQjR*X)3A{Yi_1mBY-X6q
zN3TAwOG9hbjh+vw0Q8*k#sp4<N`73V<J>jC)wYHv?YL%RId%GEN&MPKxYjm&W}=dE
zpJp-uHI)Q8vVlg}>$>o(DU#piq9rKp-YilkZdoftI~R(Fo=;(>vilRY$t}Dzk#KlP
z1{Z@m2iC^(QQPRG93-)8YP}{)-MgFc_n@N<6ZT8RDR%yRizRO^D%MotpV2SpILC~I
zR!TiF9Xb$6@U{1KrZO#a4dr}&g%5e{*V@t8!Bg-C03C58qB3<9q|EwKa~6zBc`{}>
z4Yodb8gd9f$_BstBi}}@K9l^U|4o8oD%=IM9*}MpN)Mp6hcax@1gufBhk7<EK<ir^
z0u-{Lpz;xn4TJ)=Iu#c1k^`<sbJ*6p=&#>K{E!UGhbvM-_sIj@<KTr_SiXe%c?Xl}
zxvUa0%SM%z@d)r6FVBB7z1s~fPE48FQij;&`agM|G;d<8fPQn^#hr@!@jjHwUr6YL
zSWT=8tChgw25RiQuHLrHRCJkGI%J!Z0KJa<?pTg)^F+)oa7V*r_=qnylY@Hfl~vd<
zn)>66_TPHHE8E-c{`TZ|1{X-U08q(5z>SGoCI%)SJmbgCul3XZlq!wPKg)cRrJ%dT
zAmi(fX73yrX756m2-zxVc>;1BGN^9Y189@u?$1xD(W#_E|9-vQ@|8HUT#&-pCmB8T
zsrUd@N<pDN{S7X~Pup~-u^7{myzK87%;Pa)wP5s^vA%R;%F$^$B^eeg2SBUh(Mh>+
zfVu%x@^~w=`upsgKW!e^NMyZnSzDf(!XbA=qpY!OBSlG2p*PKQRkW&`M;|o9(GN;`
zT_Tj^j#C21G{mISWI_55=@9lS310>peO`E<q(bb4FP!<b5t9{n_=yTEeJq)bmK*D@
z9L*H16yxZC+D%!tg~kxkt_LOwE0NEG(&s(r<Lu9fv~M*_4dhn+gq>0%r>9JQ=Gy7g
zwb*h7X1$ziu_3R1G{>HDcl4DUA{=*?w4iCg*CTih*XMjRC!(T&ma+k?x@#NJV?}j!
zuZG5r1;mp7>51#pBs1P!9mi{s>3?t9L==If#<AkE4omzZ`xhj|WWw`(eYK&i)()wJ
zR$@+vGZ1ghLSHQ)SxMvCNf=P>I0|x-QP!fJA3hO}Un+~DMK;)mq*;k(zt^BHF8s8~
z(_V^RmLq_%Z8<3*ra)YdSZ;>1?)zn8NQ)wJpiqiXLvlZ%0U4br%f|ZEvgsm(lE+Rk
z29HKtyQPYl8AA0(i;aV-BxZQH5ie$F+@c;3R+NxA_>VSid#}eYH<na16acu_3)SSM
ze}*Cn+ZOfneD8JNy#f<37;Ipn{Qe@u&h%{U2awn$>pOI$0$E2O;1M=F$)K{4OCbP4
z2X6BHJ@^y-y{eUou9R9aN)83Td>*ijK+IC1*N6(_D_#EnLNC-sWmYlI`6fztruBdT
zOoC1|cLbCnFEUOGjc<ZX3@9skwAv*amocK{^78dth7n84Aw|06^Gho4Zi6JIFP~ea
zjO-r^J6{F`fVbmsRZ&vCRE1Kd%pi>@t}(vg<3_<VA*dLy&Tf$hyhjK|g<QH4BfdCa
z@;ClRS_S1LG4mSihXZM2q_gpveSPqv>Y|)KeYWk7h#q-u_WT4d<navu_h=@Vi#{8B
zmAsN(QP4fI5>n)*3{Y|NoVG7EZdp^Sfb15-HsZa8@B`)eYCh;71?{_Wf5^=Tx_^x1
zdD8>P67Wu6K;*zrg?I>e1&WxBK6xPPXzMqxOCGP(>tfaG73j7C{X#&!k(h#FYlm&-
zlf)UA_pf<(tj+USwR);V5=GAjH5(G_SAvH!zY_;T0skYUwk)YG30zn4k8Rhrdw+Pb
zBqk;%0xBz;S3}-(v6@`bCXTu`xN9q%A1GEsjjAS;<qOTJ^0?y`zguIKTF!7@A7Eq&
zm`)Z|2zD@gQQ{R>&;WL@^V2RLR;+w+jfMjFHdKgtVTq8#Fi3|^77W{OUzzGJPfUmf
zCeu-7yBB9oZ)hs8mY}1&F1J%kvC_9MoZC2d809s2=cgbS<c@nmj%6U=qht_wo2o>2
z+gv3wmPi|7d=Z*wx^$y?F2ug#z)-TMr>&BouV_sKHZwc|H?cz=q;$~3+?tX`S^Sa^
z?meT@*|!-T_n_023paZZTry{mvo?257Zy+W%TTmA9EdxhTo*6`VHG|a#P0-J$Ei=j
z+OB9wOJC=@vNA(8I>oZ%D|G3kJM#+SiZ*FIg@Djv`#h4_pZCEtHZDgY;7Z$Phs)%b
zcDU}W<pU$G+Ct3j6=@rc$73wSNtNce+%R7fMR#2##J3!)AZTcdUBC7tViu3X_Rgta
zynHg96EHNm%*^kefo9wqns=Xotk85G#q^Y(g;ST^Js=P=UT#jm5r{X1;Ntg4yt-i-
z|G2dDU0hC(Zc#8j#rYuQH|Tu)U<QZ?fB()97rHNbaPmUIjp(%B@-m7Rws7N*xi3Hd
zFSyb4(=u+F6X^#P9cGHgp5uC6jgnhO_GseieL~HK<nCU8P^MBQi9%<58bM%KP|#cC
z!A#4XUqIQw??)0x=1{B|jv@0HqHD~{&26F!(0~Ynd~8Nz9sw#yXteWgK0d^J-GYlm
zc;;s``vc1A`;ZM9w)Qoupl<1b34OyLQ{86!u#nAwu<oIWxp%UrF2Xi5RmL=u<zM$+
zNKAbHOyqm#);kJ$<EVydFiTSU`EU|txNdK`Iy-=%Sf%#!IGGT;;&y;v0UQ&`oX*s`
z+Oj$VBiy(+MU7ODQoXt~O68=+dPt&K?k8y-9A4y!41<pfHzp}N*moZII*>u%&@{Pz
z8RSTfpqRs{C65$P5aoU%Bz7tioKVLKkWvqZk~6m~CM8sBXG}z<n2Kbp_52hmlvukm
zH9g#$BvAWZ^bXEi5J3ROZV1CCfhussX$?cj^!J{Qd$Z@67&D2YhJJmxW?yV?Z7gl!
zG>*QW85bGjf_K0!4t^LyWD3_$J5Y?A&x(8*d|U8Gs;W+o@fV={K%0EzFd3&=1m#xG
zEo;cy`>s2DD5L13yroQ*_y)TGw7E*jNi*ds(c$7YIz39T=+=b*tfD_0uW8lKyUo2+
zqrCYk(U~!MP|eWrWd_3_V6`}%Hl*m3zkn$i$^!z?5<s)T@yt*H?KFniUFJRC<~ezu
zc3hK*V>tp9=g%#ts^v{!9@-QcK2F~I@-=q@SMc0Gf1nGH<bcc-x$=u&N!M@&womF3
zGcXT8c^M#WX|Vncl@-B23z|DmFK;&E`}3X+76&MhW9r^4-AIR%?Yp*;?vvRW8P5I!
ze2m|%IpE;{eDrtkPcY!mLa;z7^dvaaA-3~D2|)liRM*udf(+W<$2_<|{{G)Y%ke-e
zf@Y=9yvyc`uv6m?O2!|aT3%&{96E;(xU!#)HNH3MjqH`loa;CvW2C8mzGbiQ@-9L7
z<c#0LgY+9M@9E`xBaz=J#4>sFQ*o!SBm%0fZTZIDTli-JW@)m;#q!0!?+B{D%e);#
zR$9{c)4vEvnak05*hR%g&|qbd6nsysHv1cqJmQABv%(XOHl8yA0$3bt0J8pFPFLQ~
z<N9%l+fqcB$t!?-hVdX?jM%0tzp_L2fOxLm4|lcs{17UO0Vk7)kHbsLJU;>;1_7nj
zch{$?5JpMP9M0ouJ99c)NsD!OoZ>X_tpiA9j}jR?3?5$qGvsm|#_Gy;AeBC)LFT(x
zmY?5Vk&Sff53)dCz$wcA&I@qw<*ufUH7uH+nRKm4tR#k<#S(XoB~r_N{r=q)dR_}{
zyXxudg8{wxeYH{ZsICU$esgZ(0*18D06CHyVm|Bq4l;w%t8JzZs)3sQo!3N3de7d3
z`aiRsOUy1p)6bSpjkboE%Zj)GD@F6<J$8|cIcpDN+ZQVuJC+l@ST#veytS(;^lO(r
zFERl~G(VPM$^?td@yG+MzUmGRfc?dQK!tgS@mj$pTJf69cTT$_c)09CWWp2{e-A^s
z^Dr`84NKru&BtF%ec3*!3R#EtZN6-;jNXPet#~n$z#vi5zRnVpKk3{Je^@)Y-`6(6
zz6w`o`Lv*cgkX{tPikYaS;}O8cfh;J@!>gf@Pp4m2-iZax8iqMadFNkjEO|UJb>c*
zM?g6ous#BXq)NX^$-_;cED*vJmXxG69`bNno)|j$UxD#sCMLciiNVf9(E2yIfW~5I
z-fuNeONd4gu!poQdHm-6y0^lWD^%)h7qb7p>xvb;(|*!wvQcL~+-JRVLC3|ZKBmWZ
zs%pCWoPqLfbeH45<=7WD*FP=8rg50Mr-sy!1NnbKq)<hzXj9q5ba8m=s&G-6t#J{^
z;u7Y1#KGFbl5&NeR{DqQfRfQ}lZ(~10ksmr08{akPu+m4n9N(#d6SjzwlHZD*5uZK
zQ)DRpH+JFnmvZ_zR_q1=x^lJ+Y5Dko6E+#%`7~3xNVp?;v1jm+U{j(MSttoII7#hv
z<`X&%2S?5*BxKn0A)Vz#0pqd8kF4kw^ND-~6*579eJJx}Ni!4A{==qHg0SF8=<^Lp
zQ1j1ZZ~vKn9_oJb-CJ$qg`^;tt1lqp>6u}<r;!PdQnt=~33A0@r2XfH#IH6Rf~uw3
z%ltx3TU&uHnG*S}(TpYFsHaKiz}ZjT15q22%&%eH6t9Apra<GHS{gwj_Y?b9b&Md6
zx_F^ZS(^Zo!0KWSffn0GwE!8}Ol(FtPA$l`6sx2g2X^WL3&Eq)CIS(!N+d{$=Xco$
zf3JTh#K{EvDT9-L)chmjuSqDgMyKk%F<aeOsSjH`a-(-GzE&<M*l405umX7c3>hzQ
z2BAknkU(ReGSF<Ne1;5SX#6f%f(lDcanqAk2lCcMLNUcMZ6|HQ!pKHEVU$8E00S;5
z?N@c4UI6-T!zSOf?Q&42y4_GN>7T`$b>YvKAo*+-3sR-Xnb<9poB+3q*?k@3TX;$d
zN4{9=<jZdKTt>RMWr`|KQ1BL-uMXmcWVG;3qL`4hztq_huU^&^OEEbj>~-_zDX&sn
z%R?N#bA~eiO8611Tl<$)o7F!H?7YigN#Mx?TV^$3Q5doG0#7MZHyj@+jr_OA=OCju
z97VBiGR0zW#aMiBDW~f?1j62~WnjJr#|=&-0$>fIP=v-WGx(gSI!}KuEDQlx%N!uI
zB&<nb)9mwJDVM|Oqh1Q(zuAw8Y_GRv<oUu;BvY6ICC?(d`=<(0W!ADp-0Ph()80H5
zNS36Lek*U-O>#u-Vdn|d4hvBd?8|jrKESM*w5m?s*JS4iCrOK)fVIr!NVL=Ep$fFq
zZ_+R{zw^AQNi^ePrP099L%0bXTEzTsQoX;6(4d=ut|^9+)D{OuPu%!u3(ow4lE_L3
zp%UCa%Dhg?w_JSJF1wTEnzA7Qd7}T&Iv<Z6>epsfxqNOMfh_^Z#L55>YU%$hs3K_d
z(TC4(O^npv>li<c&G2GBYP*WRIKNmQ1fcw@<}?YaHQ@>WyEC2zu`{2jq1<6<9ks?F
z-Fc3z@xESpWBCcm$mf=Jtm6#MB|dDOKGt~eE4FQArHn2djTu}#?QlAW-NIXy*Svis
zrTvGOXZ&#y&!ocNO|4Oy#<O}snvA5A69>@on_p}V0@+~U)pZo4F2wf3i*Ehlw8>gi
z*&vF1hi#aeQEtN_McvyU-r3&kh_<Ws(f4FG*uZiYeMpy#A?N58@EtzE#Glrzt@V?M
z->Y=jt*+DTVO`4{X`ucxY;|1^LSr+YEaYAILqDXCk07AE_w)FM*hUIl3gTRPc*Kyd
zT^trk#2!Jij7lOsg+v@V;pd^Ob2=eYJ4(mq^8t>5<=xjpj%_3f$4%w>Odyp=X3v%a
zgavlNkOBqYiU(c*xZTr5AI|v`L{x=tHKD<JGI4~~7sc|?t&(o-54*3R2H^aJw0&e{
z{rs${D7KPcK8Ia?19Qp!I3f3}fUL2LSbs@8`^H1gUo(uxZ`F23DGr#fL`#9{X)H(Z
zaw;!(cj<CWI>d8CChqk}&3fvzJ?6E8yRozNA^Pxp77K9S#|StfOMK;P%0y<BGVRB-
zURWu2fpyvSebzL^sIi(XoKu-Xj~|RB2=Q*?>#xxUa^>~RRhInErz^+^AT&X^c-vGr
zN~}?1rV=;Z_i2NV?;NO^&6Z_C{!W-pnvD|4RcL6QA6nT9|2>n;cDq6t<J-h@9Uzl6
zB^0-oR#y+QsmcL@z>(NXR@vFD9j|1GLdos&2tr-Ee%tQ^eR^E8Nm^@RSHT#tw^si@
z5*;=@<<lc-)TG<NSb`PhxE`Y3j}Rp_C8Hq!J&W=TQBM(phiRggRThgdvRxaefVKEI
zzcCt_Mnh*!pF+BV0%zf)>?jX{e9A6PGpd~;C{GbDMf}sqU18W3)*5`T4og$<Hjw?y
zj=YjaPhZYD*~ze&w%p-O^xs!TcU-3ZF{r+c<`w95A(AsTziN|@mg~tERVM2#KM%ED
zf%584GZjYt%}$$V2Fg8*`HkGscC_jXH&bo%FtRji3KYeoK>jDCFnCkDnb?E%I`A-w
zjX|OS+2=R=uTWY%f|LX~LYuB2v!K!<B0YWL2&$5#NC{L_fdnOj4Rc=xEFiTWSTXd*
zp38^3^AY+VEx>3;TElsrE>2_(dFtr>X_dw7im>a+bn0EJ#W$W=wS`fsA#U!*45vRY
zNy;6Hoz*_DWJ)PWuUYG531O*JNZ}|@*ky&h>@16YG2ArKDc!?JmB~<=-;IS5XP2zP
z*TCat7A$E+DfbFbeXfAoOt01_JtU7tS4Yalg}O3Vwmyq8RUxxdYlix;^`>z2i4)lc
zCgYagnQG>(z&mMq$}=BKO1vouvfgOplzgI_p%CrB-Rpk&!cb3ER@}$=gF6}=JA#C`
z%{3%gg}G)}W7@MunK}6-vCDMn=s^!uMmw*oeN`BbK&nKReA>f{w$;a6pq~EuFZKw-
zaF!P24)VMbw>r04(5*M+y*%xV|8GxW=+EzB3egla7BoRkD|%7nGQZ@9WfYNENrg)0
z+q&3BRxK`^%!NYj=+9FksjeD-y5k7l-6Y@WIX}96_^D9rB+L%cMi}rSvj-z#)D!#*
z;Mx<DmyaAAH`e`2`X&N!Mz93%FJ2ZGT9N}%8pDOqlr-rNg&Na-j1au@dQ(9HG)^Nq
zsd<^m$W)jDkDd9|Ng!heIE5dKgbq5f0D2|2ev`elHEDdLxvtf|*k8lo&|Sv5&b|w`
zif>$|`?xUUPbY{%=5iGxk6xN{wFn_Jo3GoFFQQJ+Nrb=>0x<!p*&pgL0w{ZBP0BRP
z;9m7ZVtiFsKK-v#tt27j=K0i}fiKD=yVLYypxqp!xl&A1DT;ioBYK|KU>sQeEe+T-
zFhqU-(l<0fYTS?Yd{!;*Vup!;Q^|cVpnxNMq`s;%DV)snEYF2E?X`*YJ+J9_>6^pB
zg9W>#cb?RKauP5I>+5b<FJ8-2GBO^I02Ti2MS9oBVf0f^`>l)QuSmx0K}6LI5qWwg
z($xOQB3TZ%g_N7u%HG5Wd;EHP&7}RW<DG^=&ViGwdu&;d&+9M4pT@=J?U~bB|7Sm7
zzK{XRZ$K_QfWR$^M7ey|j3WNBLCaj<7f{*|M5F<#nb$9^8t$aynd}MVIZgn1@fGP!
z2G_Vq@T(6#i9o28(P*_6Jk2M+xI|wvxJ^7`7#b2I;j*=dO_%E_@KP4D-sKttP$P3K
z-blQw&QJkRDbPL7%xFVVmUzFvMIOf_>;A7rO}3ww6Oim^GSyA3(M?`!NZwcpDX#bb
z{Ud7R6V;^|?phacdwH^>_@O-!kjx)xW4x3Y6EcL&#!q~2CLb~T#!yk{1I0cn%G=na
z>2g)CpXUijPK416t$*IU(ve7VMRnwYWx|?EV_Yo8Ot5;%z3BdC?Rzj34)nz}vwSkt
zn~DWd6LC;gXjf-#NZV=aIA3zexaWhVYTx~MNX`rM^|mj`GC3pL>4w;?!2!4@*Fj>Y
zy6#F+GUEbXTMU*e**_=4OpjJ>Ie^Iw`WsBs(!*mH|A~4Ql@8_YbWtI%1hKv(FRog~
z35;GLy!>hIBO%9;6vhc}Iri)LM#mNYT`r$J2EdzRE%f2I>sMKg+}v~yet5dsRGk2S
zTa&%ztq=Q1FoA%6p9{s&Z}wG3T;RVbY_7ot$}LuKaZ4-InD>0GNI37{;P~RrfVY6A
zFv)XCb;qpn3FWaV#o&L}(=K=su3~KjR%P_~pqqfzxhHJv-?bq$#2&8fz)&HSRF0?5
z>tH&Uu+DH?<Q5Q-G<t{H&!gcr8{QqdU?(U)XWf8db^AQ$eQ7oP?QZ3oE30v4Wj9;k
zk0z<_DuM9V2#T)^NfgQj0W^I}S%+L-gie)}E|fmL@CrYVyAP`#jNVzkHZrP;LYNNm
z2)C-~Err1sM=1Cf5XAJQ@NOr&{#wt5=QN&wcELMO0FmZ=JMB$ZjPq&7m7dF}J^5_*
z>gx4<)&WPYEs?pyivShNp!-vAteB4<eJGPR(P~%J_F~F2lc3=fAg5E~5A6ygXL)U3
zQAptxSv)N>YQ;akSI(5?(r2FjUY8b7L7S2P`XS$D!=Jz3IOm8^RntCmxc#EW6b`Ch
zZ?8A%joJ?9YMecrYt6enc32u6N@d<xoyGG^_Zy#xL`b?IYOK|vo@;Z!R2AtAq{h(y
zwGoO|Qz{`JI*q7hp1^53zk!eEMJ;GFH`;#QvE`{kKR&<fbu_OV^hbx7=kfYzLAJEc
zxJz62+fa?oy^+8?A@J|hC-b{$6{VO}WE8KS7QO46VO{h%I9K+d9WeXBZ<*5*j43pd
zOB%R_t4GytJPXO;`_Aai-NP7DJ92cq8CzO)_wJFb5X0pMQG5y$O+;!`Dl>-VdC=}$
zlTFxuD0k^mILC;~RVG>6=fCAKl6*gM%dG{xuT44%s%%QHPD}uimNMh`*F-ka4OKyt
zx`X=ijR--Xtp^4VGMc3*PS5_wh);5KN2Z;@#*8;VTUk9=Gnu@<s;G!AOg|2e`V-*L
z!J`qd<S!mWJ!*9(y)P6j6BL|B&ih@UZvQhvV%vM6TsKvbxY_mlFZpy+%RxxwUJ>lR
z`-;@^{;}nQ7cmR`QdODd!wJNcJK^wwSV>0je)yiQG%j5ylV6sB@F+r4M;;wb1<J4A
zqS|Ktx@tFflXyQ2L1Va~VYFCh{!jm&N7ZFXLsN}&#0iK}Ir#9YP-ca1`P>Y{1uUaV
znA5`{7EEl_Mc<faEBGGN0GQL$dr0=urQvtM;xQ)=PajAYkzN_B9XsFKJlxMjHSL)@
z0YaI-$!RPfYN`HucB!t|R_H{t^*EGqhUo+4fgcSkx%WrOJzf+e#gv(cquRW10~gOY
zj<6wpsjAu)P0#3f|MZx%uLha|y*j0D#wcBsw^=-!)<9x|<#KX7b;DMKY23~ou|xiq
zi4j^3q9r*a|Km};r|Y9YF`iK8xjf9%V$O%nx~2~NOzTe$J<YTw+7i%@9%i&yHv?1r
zIg)sV*GR~+-<2mc5~%kJ2OP8AcQV_rg}Z*Z%~q{*<=vu#d_s8RcEP^G51+-Gr5hX_
zrq2PtvDtCz%l^DB`JBt8o%dAtjh}-@rHV@l{YxFsmgq82gJX*R<-#y*K^5LgU$SlH
zec?fscHZ^K{>%)FMzH-P06s9?q~;G#V>80JK>l?vhXDTFcKHGhnlHIumS+Jb^M@IS
zN9_g{%xfYx_-_6FR}O7Ck{F<90F9u=_y?SJrOx2}>a{4aY}H{c(OU|?lg`>-cdn@Q
zS-)(8dND8hqS<6*mIu3@$whyL`=4I(o5>GIyKh%NFp1utKQ6UhkwF9C$R8)koqj9x
z*18R&c^pcRd#no3zJ+xLXrnx4)5--RSg3|TsJ>2}6-V?dwvEpsXW~G61J2kRzpKSq
z>8B%-dg@(t?&@_IJb0WT%$%Z4A1=OZ$GNMh^!u}UEaTV|d&gt2#{6=4=yPyfcAGt)
z^?#fN5q5<s#!nAxw$2-o^hxGLQaO#enVCc@w_~3#3NU>mz@yaiXs<)d7sE0vF|nQy
zRms8(v0oknOgdE898T&^D1k0VO!<04e$8M2{_4H|zxU8MAQIodldb~2TkmQxRsr6(
z$5kmAXKAVfjL%uTqP;+NI}H5=oYfwy_fMou3M{EOc0?~b&q9A>up6~!NJkUEp8de&
z*<4vx72Y%0JmFAdu*mFbaWv9RJ0{7<%oIod0U~&p^jZXW=OA`SfU76*%<UJGi=O!;
z)$+A*YbQ%|3>sO7_^O=^V#1K8|Bjqz_U$HR$rf-P@%f*>^8W!9z)#AOk0QP@=+=5~
z$gjo_hozz1(@0tP>1wrS9*c2^of7)7{wFFpZ2E#@?FISjeIQQXOc>Ry)eF`4IO9?H
z!#bU|jN>OueE_u2%-o+O9d;U{4&4ERD8{(Ry^vyh#TrP4lo^b%Y-;bu^gDyk=5iI_
z2PAdRmnbS6RU3atOmkh$S?=AuYp<a?DNB>}>ykzL-P_7VzN@*SmaB{J=Y-BBMG{Ku
z4JZdOa-GvN;Xc})9&eu4g|9)p`SoRvgI~0!Dj($*k28iQQAoN~>iJFBONN)glKwN}
z3lN#+y2U@m^v^tk*#FE7Vumg0I=TdEH^X{|5Ao3Yb)oyE-+UjOcZeK*-J(*m%;?s!
zlrg<DVu%yI(m8or>41udE7=j?x@Y28G?9Z04J0W}=B|WvMdT!y4Rl!C1^Maoe4ORp
zRN}9F{f1iUa+Z|$cvMEH@3h1r>pYXWTevL*5$Z~{Tg-#ce}7^MY$Qs@+-oMcQ?yn@
zeOK+5<KmGVxhr*nyLpRp)}#tZHj=xf3W$q+sm$7I9hQe_cHnK*@sF?kTa-aKiAL(z
z^RZgfMU`SNJ{cLk6d3(~3CLq2_qIOAiF}+T`wR?arHpiRveQ9DghzZK4+`rU+9s6i
zByumIVOa5xN7m42Yt#6r>~BX+o*4(NJH@!FFjg+whCY+TI@fbWCQ3@9CF$)PAziR~
z@+nDB`$_M7!t8!4_iUlo0PB1G-c{dQ#Whm;JXah*r&5Zk<^Kw`SI4vSE4P4MrIC8k
zd<<EHx-Ye~YX60@LJJdwev1vL?pC1C;nu3WYCma&X+N$^KaYRDiT60g^0-hkjrk!+
zyc3Q|Krh;UY6QgcG4)|KRV^=Qi)g9fwMaswx9*>o9nKTueL@Nuuslkyj!TZN=Nes*
zG{g*3#UfU`Pv$LQ77Vntbb;<>^(kxh=a(<`W%;Y8O~;QzqDzx9ES9=qNx`)b-AH~}
z&|DRO`kN?|@)-5?!<jHk^UVroH_C~(B~qQ~XllNvg;On+lyF0MIIjG#26XqI`k$23
zjo_-NW)Vx87>FVU{pL4sHsf{x_3-jjC82hk!zW(2XcvMKVWX0SpCSMGcu#`BYPyZ0
zn#F{J{&Jn-ac`e2yp~&6X#Y69E2k)qFX@B#wKBA;;$fls@wIMl)9gUAiLFg9YtvE>
zG#mwb3xd?`hxAGhk1b2g%f{-J!J8XCz|Bf341%CP5c$}?(|*~qi)7>3EB(r(NV<MX
zKnRSXEm+|!x<k<O`Tn$vJbO27+&32*f&1-SMsBXXT}icHzm|?oCa~LJ3g2iFuo(>W
zgrY&S0QJF}6A%m^!==pql40bK36#cmsr~<o)&kWWsDuEP3PJPgaY@Xll;yV1^_Jsg
z&=U~UU_eiua1v)^XgFX7FTZ@20rG@`Y3w@r(-X!HyAI6CWuC~bywP(jTJy$5Q`5{R
z2+zYl`7=s~5oY$Ts>nq@G&~m?2?gCG;IkloR})-c>n~aKY7#{6f<?_hJ`OOUB|~}B
ze~Ny0-N8^NGl)rn(&X-^E6{F79l&q;KkfR*O|-r|$o9qjD6;osUHcR#MFE|K^^fCn
zh?G~P*QAmIQqP-$Kks@AQ`YPpyt%^3TC=%rsy0A60kl1W*E73Fm!L;KtI@J+br+iH
zU!V8L0u7Jj2l<m$1DK{tS(45FzEdY~6Y~fSoj#oy{`gZei)N=81>`pz4xrpv<lYrT
z`DZ~BG^}3E%y1xadYtNU+I}qTP4Ds{25v+|68}qH|6ZVQ54_GTFK2}2eTY@hZrVAt
zp`QwapXrkSvoAEm7kaOHoPPvnCwUM?KLGIy#6z|D7K*le9`1kRO>JtF_8SI7NyKCF
zo#cUF$&Xzx<vyK?4<!WOOv?I)X$008G@W3F)KnPW8;U%vW5T~ogd{9FJj3PPO_w$A
zmu4+IIUF#SfM$><p03-huCZ@g-@`iV;fbXwWg>oA7&L?tJNVE8n{t~>#9V3hV*572
zu-4>_5ytQdm4Tv7=w%2N1-H>O<HUbo4q8#c;nPteKWnaA6DZRY-MD+q&2+!olM0~@
zG(a_QCd0g)agAx!{)dO{$9CxjT#@J8%ov7)kC-CeGslabN0DwRj{*yo{H~?V2Mf%D
zsPbA`Z%>>tg&(+>^f`E{d&|G8%E(2h*Vrwh9nG~RM-Do%dEamE#r5gOJ8?^|pkS62
z);ib7r6!}u=9XUto<7WG4+mpDMA#>q+J;v7`mCWrY*dA6v2%LANxoOV$4SL!(Qiqi
zjuXuG>yov;n;u*H_B#Am`I(yT?U$i+q3a$lkA1$2fnMDPAK1o2>UP@SUxxmQGlyIt
z?`NKzN)DhGqk2}Q6xH!aX&s1TFw!G1#9tEZvf@Mz;Xiu4C@`oYK7SG4tEieLpX;;1
z$RqUR!V<<c&JSw)Fn?smq<&Ak=z-)5>H6L=LPHOQ9G`n19uoW25#LoInDG6L(3->f
z+fJ?HJ>JwyqvgbSDx{*99Yl5dM)Rb|{Y9~i3^Bf;87oSO`C08Xk4`?J!&l{4T=7ZT
z7!=I;sUoc~R%WI{!g0=Te&nNb5g$A&r0!L)$hP1)%$#^e%<<vz>RC2NG*!0;-46FG
zh(7zAM}t)I7npeGCS($i4H`bzS*yk0-6B%z)7rYuU#>i!d=6iFuc0wIat5vq4sARV
z_c@a-3FF};0~@~@cUB_fj@_H0xQz+IoWtzsw8+UhN|{}rYx^Wh`1ae4=v&}(iq(i(
zS?R!2Ok0cj!7p2i2X&u>ro3}X4iaEdVB<R#I=MS|55I29Dj<{7TK5xHx?#iHVpAr&
z(KTfwMpH#x`cRS)LCY|iH;dG5bVb-ZV|Ks6u8Bv7!s~ROK>A*nQCc^Rp(LTUQkoz|
z#~b2cxzg^L(OOCz&%Zm@W{Nb#UyMcKd$TTdyOj(AChIL;D!yV_%X$A{lwRZ<rQ#YI
zYR*ui)8*%zfi~m3l4;4w%^iA`+e$h7arq#VsJO9N;6eR1T<~P=T$a2f0jFYnpcljR
z4`uk8UF>z3L0B>R3}Ig#oa(6ZY*)0teM=}kuMkG}oh;cp<lDwKo7KSj+;*(q-*5gi
zNl~uj!%)8{FARHj=v1HLBk_{^g*D9`M3#m_I1B?zpX-D3q3e+RcCqVG-|twsYiw?g
zFr3Vb0kjPP6Jfe#uICHN|5#H#bTMzGIiI)eZLbNRKE71pTSoxDLa^Qt@OEXrk3dao
zcE4OfvN^dqnirb2uH--d`=>%=(-kY|v(Yn~ilOgosHDx=xvtUOCf7VXxbBN{DAuvd
zv*V*f+oQY4<Gw01I1<QIp!7XR=>+uu<rC%ukGY}Kz=e6X^Fld@qyt@Q{}m(KgrHb7
z7bF((Rf)@g9WL8t$>(ykef^~a(3`Qx?S{Q7$Cp6v`-ecb7Z-oN-SsTU*0P0TbbW^w
z<f7+eVSCr(Jr2CE($F~bHzFb{OhX|2XS`54`lkETCn3`>J~g~*BjcoYBu_LCoQ!IC
z(gVu0KLg&F`In?c6KufOeVW;d>VLhh{hL6XW?4eHfLl5Hki%6j^H;zd8&U5_Eam*-
zkN*z1Zu)pTe?f0p@0ClVol!J}N+5CqKt7d`$2ua|AVCLeJBYN~scP%w=e#m`LLbmO
zHI!4w(tH7ON@79c_c@5i^C}j#xK@^|VBl7}<u*rjrZWYfS*PO;z5N>cr?NeF=D|Wy
z2f`~_mjm1Sk5;>nOh<iQ{#%poe+VXjFa$-%Nm}1AHs?&+T;jdIN02-aNWGKkJoP>8
zzn|Cvn#Pd*eX{okPKcVCKMHmVqgRmSR-T_Q9S10f`8Oh{&x{&8qa{6ac{|Ui68+vQ
zP;MLB%_r!LI+PHlpS7V^_-K}?xf9Gf_7X(7u-aM-Hp)F{{b$XT=z&3cS^g*gTqE50
zS4j$-O)#sZPfcP#C37|xiQ8Z7?B_2oEg*(mr^^XmR@jLKjI^9Hp4z0IeD4iEzx>et
zs7SbZu3+45zOp4$)pEtu*C+4}B4PF8`osuLtc$I7s&*&G=4*LPDXKeMXb;M`l9t3@
znfI^HaHv@bqAyI=HC>$#S|uo)#!T&S*V8ri!<}d)6hgF9O6kVYz71Uzkw1q_X=S_5
zd}?=mS1PrOnBH;B3DkLg$*Oi6ld`i#oUz!D*w%NMKfQXv3dnzQW6E#$4~4^vQR+!c
zEA~}x``#OFuHF-2VYcQQH(O6X!JHGm<mAiq8Izdx{a9j8lm0PxENzn37<}fmf|UGk
zaIkMX9HO+stBi+dc6HqcKB^(OW#b+ljykTY`YEw8#R;BUrLpNNn0v{=E`ILl9_mel
zKkHLmh3OtF%NdZw6zz%kb7tRd-!Y4Eny!jZjyfOoM5YywNLU@x35iET$T@$Q*=<?t
znLR(cTvQXVw6!HcUiChNY7)mqno@<p*7Sg-P3F#kp;$dzw3=7?CDUTR@YaLq&;O(A
zEui99)@Wfukip$uLU4CyaCdk2;KAKp0t9yt?(XguG-!|z+zFPybMHC#fA^mEdJV&3
zF;m@D^>y_pyY{x0A!CcwfVYNT@(xi-^gt9!sTIyNKiV0%6OCy>7=h+~n^FZNyVyJ)
zN>Lj8Uc876@3{>hzTy0#he$MtqkqnJior5%1sK#`E~mLm5N*iM1>AjpA%|did~Z7n
z%dBlG9z}>@Nn?Fp(4lxCMH4N79D5IHYcj(|QnfQpAzPmccy%RVU{4>Xwv>^x?(a7U
z_1k!Okf?0R^5~Duw@c4YDcL%Axt4UW0JcT|Wg%=`?GPN-`aJaOxq|C7ig;rH>g#h6
zOoE8O_ug?yc{NI*RFo)^v>0pMa3!vaerbi75nQOQl#<S>#e#3z50<e?R(ingwi_1?
zvG+ITlf`C)8!;hf0EoESjs&VqPXF+RDSq%kkKDUaRV=Nbt*0N*$bak5zC+cxWbb+k
zH<Hu)r}UX>zUU_JKr7*2z`VrVWC_o}uE5=W*Hm*kDjFXm{yqsH$Vdg0Aka(^Wo(j{
zZY(NVJ;|Ae6D0S)%X57j^?&gfLg@XjXty_pq<D32_{zuFv&&lV!~bjKJ>{RHvws#<
zXK-1faaIdu{9wJ+(++bn;QSzP5nb@06`rT8+WQc{y6X%-%jJ)U3gmL@01pnEQEO#$
zp|u@pyM7bu-LmZVxcjrs)26YhLRG$Ie?#qU^Tz0BDKp!a17)3ECl(yVGWS&5Y3^%R
zz08^l3F)sn6j1g=_LHmomsLJf-~aGWp5ZU)xid_UY*%aB6dv&NB|m);z&_u(6m2Vm
zG3TQvc^%j)T|YwpT3DxmaQdM}l9_XJ|Loj1oSex~Id)zLdRMr&2~b%g(lm_Jz~@T#
zw!3?>@LBUQK}=b=Gi$JjlfmaHFtxe1LBh}yp~FCh<gg#Dc=ru8<@O47;73NB?Zim<
z#}(d{MvCpQV~0OKc_-nAH-DU2U|#_cOQiQ0kuNU7Si<0QA#y0S{PK{qj$~`J-L|di
z=0uIa7Bdz3v3oB%g-89g>y7jChrQC_PG5$SAsJNLG|y@BS7U`oB|#|XdTT?|A7pU*
zQh0Hm?@?`-Eo`gx2fm(b1<zNpNaB7W^DF)J?Umm0njeV<52#y3(u4vzjVF%>#%lSR
z)@RK`$k%~zKL^KPkC&!pz4dN~M==kUdBe?EzeD}T_c1S&;btY%L289Ay%{1x9JfPU
zv?B5y92{R_S%yDa5f6=-inJzMa{nKhkED=rsvXI~iN+<+gcPR1_jpqE3p+priEYfa
zuchZ&9-u<)@O(U)>U^pCyi;tsPxj>@d+U`5$nA)f;tt>r08#{nyhFR$=70=B<<qZr
zrRG~b_o*iOG$9l~e>+$|^9vGb0yPjlcz3?`FEg_(IM@$_C{gDcXBSrwP_<05yJ4aC
zL;BpGIpp8qf_c5V5|*i%c)8!NUOY(Tczz(LNJ#DyOdqc^Frpm*(lCiSSCb2w>6;z#
zG(Dfgi9I6f5&y98DN<4y^*x93|DM!9=DShyu`h??8LFr#V6>%b$KZc%xV77Cj2!zF
zJqKBl4tS<}+7gLhP{$7YajnWnlZRK}RX+!6B_uIq=89_tqsIZ2q(Xq6dbPz8g^9lh
z%Vq)#h;7>$=V-q({f<{u)lB_T)5aQ8h4Rru7MtPVLR7QP`damQV9P%gppqiYdN1m{
z<ty|iGc-$FBEQ(i{-$EBe!uO|)8xAZ5H|+7!>Hb`dw!w`O(iZCP~i<})Y6@&d%CF*
z^mq%O<LSqyzG1pwv7hyEjmGIr{OjzXj~N4?a7tg{vj*a3BGUn^iZ?p7fGE5+0KO=Z
zCC+`r<Lw_@7a-92y?iRO^;FhyPt$K-bIrMFDUVXqz!yuB&=*r(9pIS<0v2(aApm8?
z0Z@Ju`21#@Dle^(5bL@2Ba%w<Wn1u|5tb(QJW;kViL0ZZ7`^Un^0azXqQM#^umuDn
zL1Z3)3L2`8GT%gVHXnC6H#CO7dv4GEwU$q-!zWxV_#8hAw(G;0!)X8eeWUVnNMqaB
zogTke7iz5WiF(9jv#&ehm#eyhGXOUV64aMIdb%I#<^*`Ux)b+HuK}it+}%w!YBI{p
z!Lgr#EC!Iv1JV#ELL2!~Iv{_^GyjV>IU85}KxL1Gsv4C~T*=>)Uxsel`_%T~&ix!v
zkF0g~iUJtdZEQvP#D2`It<JW9YfbMN>4_bXROp(|;SC-Ygf~xXEp14e8i3T8;m&)Z
zUvQo|s;Te(s8*b}@4#e%{0U;3jGfXflR34b@Kuc=bi=7f&pOBM9N@qgKTMA=@d4Lv
z84=y)U&|)2aFW;Nc3cER;+_n6lT$!I2Imt~C#%&;VT1FE@s;!G)3Rog#G?yOuZ3S0
z_qwk(TiR+2kSl1R_8W_$u0cGc@jAAx66u)du%aOfg)8wQ2KK(<pjMlooF}}b1#Ic5
zHhCX2ire2fa_*JG=j#=2p<wd(73F5|Ys(9H97?8GDz6i=R{o6zfGa8nySLjD5OiD(
zE9h}F+g*0M>8qM~7HV5h*cuREDVE$s6mOPwZRQyij_3BJv74@cGes8B@cwpwpRcUb
z_C{nRg2LkI<<$dD!#DyH4Sn&12!ta3o{&BYn%<z^sHk6sgHrmzaw-q;{JS?vw{3Jx
z$XJ%DPE3;2)cGXJeki)LWA5rLQFqCyKM4^N!6uKvdXb65;3LvYNu(Rehnb6~#3!WN
ztzHf1*_@s8aOv8Ttqf;LP`xqxZB)&`t<^Ygn|tLl^QEBP2R7`fCT2+j%Z=-Ri-ygo
zolpDGhM5WfIxeDy`owKG6glSJcCB_!t0|P2nh`(<?j>jBK+-r)J06UcvQ)eLA#kp@
zevuIDdSxR7^cJ#Q=cTo5d486e)^gIvWjWccUJ)t^y|4>CUit=llns13MLKWKDVF`b
z9V=p4uO7*et~Xe4>3OoDdHFzbtbN}gIjhYMSst7;+)D2yhERzRje7WQb11Y_354mU
zzLQO^%A>wjki56)?Jrf*L9fp(a>nZSu<_)v9SB>SxaShlnl|r&BJ|paQcfPJ`UCP@
zfXU7)r=!`QNcF30eOASp%Umc<{o{>l`Y2VI+*JH|TP{hpe)J;Vdd~Vd5GZAvAQlgh
zI~`xod>V85NxOIKWqjRGR#8&%#C-@hU_8H^t~Fnv@?&lakKTHKnHG2=jB)z*)a9M`
zQO|!0Uu=uD#N|WRKBt*I&;H7H9+O|R5Ctyeo3hhF-|2JwA8B&DO)$24j``@dok3f_
zHwA7OT=`7X>f@d9(v{Nuq)!nEa(yzq&tC;qZ0}CL5j;x31W8p0@7Zm26VX65)2=>(
z#e29y-Qwnx)zm|4=va5s*V=-Sj)`~1kd8f{%pLMadChc8DFV?s-%2nD1f3eCRZJDI
zKctVGuM4j+lZ2q^y-ag_10+!GR%!#omj%Wtj!C`)M80n!%TvtpvK1@`U7o+V`)2>~
zJ)FKvREEss{n12V?77k`>l{LeKif-on-X_u4<QkFz@T*i(k@8+ZNIEZ>5IYzW;SL^
zz@p9^2tH(D5s)cKIpvNrP`NHNs~3x(^7?P74)mTzb6;pYb6PY423@k<^PmWF;G6x(
zmlroLuaS<|?)-Lze!nNBpA@>Qiw||#EqD;}|LI>sKcOMdVY7_(^d6z|&Q#G3L_jik
zS72?rP?GDQHPX`aTVL>TE*^5Wg+oO;B;n>b!?NW=pxhiA7T?`jNgA%{;pPS|8Ylc~
z?#wF^P}|`2ZU#bvsoXz_V;*1&+FEy|(_+bPCxSt=QydAv#_pP^%rkLWnl;E*(?k4T
ztOJfRG*460j91cj$dyAh=ZM=K3N(|+<c$Hr_CkD|>p{pmR2VNHar)GG?(X2V>TigF
z{;$Iij=5f1V1?{s!%XCio=hJ%UPZlGYn~XMiYzr*{uL>-&Yls>opVxP0n~9YU%#w9
zBMO;j_BQ;i-Aadt@18{C=)ip<wH4Pr+`VX|R4>HbR~<OKO@yFhVM;2CFNPV{y5Rgs
z@8r4iUJ2U<FV{kSv`5kM7*5}KqDOv5S%Pm|I*VNU5yJ@1WwHGjS9xpsk3%kw31ekc
zc<s&QUS$K@4cMNorMN#uanlLV#~dH?OS)a!B{^ks!j!D9t&z&DOtRJZ&M86R5($}E
zK{0L2@1c3Uk%wMy`U6>4$eOgjBxU*vZi_3mar|V@f4gzeC27&uGvpb*$a+P%^DIHM
z{ES$+l`99mc1{%RhGTEj;Ha1$>qo3{>&Y@M+@cX*$4`^rOjVYqW3o2VtFu*s_r{E~
ztaK8l`C^JAUsp`Ght&Cs;uNj|jCQuwNsh{!ym6yFIze~G<Xt9rOi!yy;rs2bE6a3o
zD6g`FU&!_MLFSGh(V6@E{#3OuSeQtj_<roj;%w_h5S@|{y`=c+CBi*_CDv(>@l_^t
z39zlE0ou3t>oA4VV2MT)72z){NyN!ecq(>Ilgg5XTjZ?(7NmBGl}+GJ!iage)@6hf
zd(seUv(Dk1xY@xwaHHcRDOy^^-s5o%EP!SFWrm;Sl+V~7_Cv^_?dLe%uP46l@#*Xi
z8u-vAEE#W9vPUJaf*QA*`ex)Kaofl+chC{f{YisXs}cGkl4&-gs&^dEPf*TkXXtTt
zTaur8z4vmQQIq9QTpkeE(uzU1=2(x5CkSy<mldg`B9Rd_SVl@tp8_uPX_J?gA+ew0
z(oCno38Wd62%^z$^P$o14zO_hQO;5VHmWNDdxg6Pt*J@T%*rU;dT%qJb3fOr6M7D!
z5w_dBn^p%AN)r3wrjNHhS2+x~r7`mgrtv!P*nM@L7C5GF--!8zZ~%0g9a52zV{=i@
zXHI;Nw~+S9OgpZY!|m&cZZytzg`XNS$6N+J4I!N-A*9Yu?7&jk_@c;^sjQlw`NZ)3
zx8Pnf{<p`1?{#dumvnQfRM=5<mvrjoOl{GyRYiZb@J~UGi03Q;=z#OxKb-o1=tFrX
z!lbTZ#PX06xdwh0VFHhrIe_??54FEVIRnUzGwJpL<^8_lSb_m?1>+kD{%L~ODo-SJ
zH?8A>7EcU&o7Nu%2<6ADNunHrkG7qcu{bR~>u$%yvDZw~pVPkB&cYCk32}+NhIU&W
zq@i}dEN7`MzvHB8M}TL*@Yglo-$)2&T#EpSc>7WJQxx~}kethV3l&2xWt3@M6MF9L
zut%5dPoTUZM8KJL7_t98{C<}ol$s#JVfxeK`SO-xd>#-bL~sO1QZb65hSdmNCRx6D
ze7_aBJmMEm+{qg9An}{wuDnfzLV>R>X~w|l`?&u{4Qf|6YGg?$3a0^kAMW2r4QK@o
z(!#$HO34RVY0il8x^L4#hzGe58HZv7*>zgz@UoyyV%lpTBG+=+`p$PqwsYD~XjvfR
z%5b_XTnHsss6&5X1Z6SaC7Z2d^7Fz7tP;*tWZUE6P_-?0ELqd+h0|7osX<s%EKV7Z
zg{kTgj{_fU-F}yS6;7t`*IxCtHW^$s`AJ^OhwI973xu>a?-7cI(1eDX%xhxNT*J!l
zh9NNsmB^Rr+=e*OTc1_8tnVu)_2FMVTy7Wias?WO_r)Kdggxpwrv(fG2l1y?mMXD`
zkb{dLdbej2r*4TWX;llV{=PE@FC?!0nWV^`_Z_-pet$g7_OQGJZ$kqLq9L(QxUvUg
z-?%^LbUX8RIPj^z!Gft_&sI?A*(#-#1;V5-R5D2*O$^kt`NfFw6!n)+zBOf6R3xZ4
z528sbrIH!jq?&`y>zXB`!jp}!zeN)k4<Frw6-tg?5jVgrf}6$k6GoDU4Z@DUa8q%0
z!mXxOf*(yx3WD9RB9ypnp&(=LA<2+D4U{xf4g50{S`egSP!d0bhhzPjU^9=LnZqEl
zM%kg=v=a}>ER)>aY`;^P?k?}V@7?h<(%(#_QTNEZ+1%|CMyp0RB&dW7fe%Q=H>KZ%
zOGX|v`-D)j#dLJAk-F$yUH98j3hxT#s~Z-k$C64amZu=o5|>sCx24WkHTC;Ma4;dN
zaI<zTV!78^wNu^Zl&%T8mOV?e$<_@!1_p*)Ol-c5<OnUcyc{d%g85;by`20h%yt<C
zilvVa|4ry5^9NLODlt;m<k{_ktd|a^t;n35BI+nkO;!D6pmg-Yf4iuPYbx=Ra2zkH
zoj^q>8z7h^^n2LB6c+&?13vbP8tPbAJp#a|{zyC9jTxE-CDk^a`=GH$m2!0*$^Q|C
z?lPbm0z&yY9Y9%(r{Eb#%=h3|vS0JBpiJ8gIO=A(h`c)#0L1qqBjEs8c0cnuIo{Hu
z=|C6~h-__eQ+^dlRG&U@9_rY#3>(=$_)bqw);eqnKE4R**a95U8%wLW3$z#bCbf0$
zG!xDdy&-@Mh5McXfHN{!p8@KjIMX~OB_$9ZPwQLL4-X!E#6ZAi3h1M<&G0ucD1umO
zhcC&(oL%1?n1CD~I+hXBm+%O3*9HMtfj}j5mfzBt&qMrH{FH&0ar(bKVMj+}ZVzdh
zu|L9Y08OoQk7x5pqKK9usv0BqtSIff02R%XyqUL|BnQs})8<KtpU@{M2_fWjnnkR%
zzfAOtM`p){1d*bGguXqXz9)2p5C2ss)rn}h>Zlo8L6vNQBqfjdi0E>+_3Z-o*|5d;
z%CNi`t_FmN)+~Qkzj_z_`>q7#((v_Bb@tb}(qGT8kl%)+A+1~9{?oGApLccs?Q#5H
z-jTfsG4FyY6!7M*dYfQ_JL(-8F9JUA5HgraZGUmGNH(qzlGk63k&mhCY(Qs#`H6LG
zf3c5^oPvO??FaJAVPj(xtvQ)Hg5gMM!j6~xnP#R1Md5MCz{d9U1M`#X4epRwpa$_F
zgFfD(nv7MD=uMdy7*U(kgC30ENjE1$7BXO3P@Im4?dcnj3co>f0bj}D!43+yRX5^Y
zW0#1(DmwweoRCSVx|vCl=r5B+);0VvF%ln1c&ArLg#q3+XAgf50OZcmjC@Vkh)+nO
zKg89kPx?Z6miA^O=t|kjYu{gPesYI@bn{Q>T!l7|okk!OVW1#fYz7}^fw8p=6Ww4r
zLuI5p-PMLk&1&rj8hYS@l4GUyYZHnt_wPO4u*QX7<<-;21smC!1Z}FJoc4cn_RGt%
zNs*M#MG;^2I)}Zd@K{=38@Y7E)P_|}Pe7RpO2q>NGz4qBr=V}V1W2i?CHyc%Q$ObS
z=#8#hATB&v{P@_BKVxqdU&XLtf@q2;N#gnEM^1OxasC8Hw3bAzighBLD$pa0BlTT$
zs>FU#j0Klb?7sbH{mnJ@>FWDeQ8p>Zs-}`^+hmY*-E|rPV-~11%as-!p&{_KY|XKt
zi2@Bwstlq@)Y(a~?Bo+^rngns0YGiGCUd+81}+u}-1&H!b<-YSo>gR=rnL{7BL@jm
zHnc1WO%Xr$+@4avkF-~p%Zlb4avyTv{rFBD1!*eR7}=<GNF9X?O!!`?ghym<cGW5$
z6#B7Vqh3gRJ)zs+T%}$cdOAXp>w_R6>5#}@^#CTajMAnhqH1{-3g!OTJ@udjud#eb
z@nJRg^NO@znw3udObD34zxL5iAFUIrs4GqMePGE9?Rs?QOQD7hc9B_Ej3dhC`(ZPI
zoG%ihT2Ji(W#>HeI<iZ%=uvIyj#4+T5SGjAe_(E3kYS<Aq5IMu{GoQiGO-tpW{Diw
z?*-PRE9*Qgk8lH3plL}4Qn5!bHy2V&SG<J1rP84Zx+^4>Qdu-Z^f(jsdOeaC_hd}%
z;1XwR4Oq10l)fULztgcyYe!K?JR}<Ti%pFUd^q_)v=RU!t4+x-<Xh}QKNq;JM|Rs+
z2KESm5_jmeG$|_I=8}muqBUX#?~f#CcQDVTqIR-LRq5>ZB5be<HCxgY#f3y8?gf*$
ziM!8zU?#LdZ<0X3Bw0JiLWsu3uzSaw&GOiVzEv|;R8q(bqH|<SGEd3-?yN^NK7Ji3
z%y}cKt)O*)g8`UV6j*sVPR=U}foCX0X{Eg`c7a<`mO2!UQkVSQWU<k4L6O)a*%LO?
zfY-z?qeH9ytXXa`xa(**G5aw@C1?VkVh$DL;l?%xQ{v?nJ&U&fr@)QemOi0wOt!MP
zf;en<Bx`n4uf6M_ull&l$^@gv0D<l*y1vy_T!EotA?w#oU3Yc?zEBLO$rn<g6h`(X
zby#}uR{e#2c5zD5YXyZ3RK0Ps<bJ+zj37U(pX)~-%XYo9k8zr~F~LSYwZBhXd2j3^
z(O}-QjT*P-SdVoL$JJHa@?sFtE)3Ah<{Uol;}cL3g`#p3c6fW^5fBV02iJQXIr$kW
z;#C4-4!Iu{2%wg9U8^tqNjXUAORAj;B;2r`9y{iA7FwE{mZ6kPcepecMoiFTX0P3F
zF^bW2wzCW4$#NH4EhfWb41-?s2rforU93-kbl`8O<Kj%5cmHst`gcm{Imy@zn#I;d
zlYKEb&-^_ldaU>@QoMs7;+E>lS&o19IMuU?i05twb!AE=$d-Ixga-o08H+B^g=!CP
z2|hCZ?mDD(xNrSUClrg;TP*_Tl`tcPHwL7Q-&?AG2m2Q<{rURh>3X7DB6En(5?)lT
z#mDKw<iQoZ##z^^GeWfLznwc6LiVE6;J0QONbm_egi2sxNo#CHMu0aV1(^qfGww|%
zW)&}(CX}(?!n<cnlKg0o$mhT`jsy~pR5XrU^+{vQSUwaruYMcpth0YiqV4--J2EX=
z{zYi-kyiw1QgD{Ssi@xtwY%V=e7tQ;;Z)!Eu8~x%HsTT4z)}gDK=-J2<d76b6GM+F
z(&}ka{9xbQ?_x@2S_IQ><YiJ(1-)NI<QF9P$ilB&qzQ?OpQW<jnautcy?2`?O+g5W
zynguh>4?8kAKF~T!7B`Sp2x7LH7go~Fw<6X)F-fvsN^cV-;@5?MhupA&1We=3^7#;
zERXBwxmao%r#K%1)j=EQFA=&s^UnGUnps?%NTnR~t>R~!Cr{6Qp=<ZVq2WiL;D*Ky
zR9THM8IKl*;1N=?GfyAZ2}Tu}Qx!K~QeS?h?GI$hYv3L+UFL;d3TwAq)#E9HIS&OP
zh(V-48#<A}?DG;#Ytkmuv8{bADeu308&Mke0X?kw%)36WH~Eqmx;gCOx<8}ush;HE
z!x9>q&Ie=#+r4hu#w~m$7Pk}{k>iW(_yw&|$GBvr<v9M&dA(nQuPj<d+Gb%lgQu-2
zer5RCO}mTu;Eglv>NS&9G@il|7-S>4S)DPcbAO<n75_m;GbHt^ekUt41lmT)K;N@o
zFT{%j>pqy@fjQYj7tBM%TZD^}(!fy$8RLSBxSN|YC-7I9KW~sZ@hiRWpb|VDiTvAL
zHe*Ro_9$6op|{li4U8Oca1A?-gCkmmNyAT~=$ueu{3CC$WyAWvvxvB0QSHc9zZ2^q
zuG6}JVF(U7?s=Xn-7b!$KsSg@6R5#|w{&r65sR0EfV^Z&lP>YE<-Gq3HY`dx7Kv1+
zBgIEwg#b(^$!_Jq6{GmPP`lH8>~hbvH6yuNYYY>_z}{q-t$XUKbfs$u<^>pA{AiKD
z>^jWsvNRURNU^*NjZh?8gp!YC^awRNV(Y?@E!&lOpWM={3qm?=Qbfqct%L=kA*m!m
zSRgJ*%56xIqzM+Y6w_EqsYplUbIcNKD~7tRDAYPiRAuN;t244bgwDl>?otXDT7qOw
zqM9J~Etq1}r;{H8Hf_pwaAai@Ne4$NqY@(ImkQhZM)3Iaa<qYJ8%C-TVz_}E#FVaX
zeXY}pzG!J^>UrX6&QBlZDN+hXvGj*)pXjIiJ1e8Ph?U_k@W!cQ5VmMipb@x^$JBnP
z;`#o7^yKm&zq^E6KIG}2@GxkcybiR(<nEkF1i>e&f+Ekoozuy>#BPaFkU3priOjzs
zC*?DBp^!x{@rGa~($kF96FLYGi;S^ra}6L!+uI}{O0|uut-f7(%NdgGQmt;GKAh(q
ziiuu>SnN`!mv8pt%dA9v?D=(Fr#ej+ZVZc6d7XA_Qyw4wci&pU6ANxgo(#8!GUd0$
zc3G=VZ}d*i!?}D8^Bi3wz_SjR_MwFi3_4bNAgrM^tyHzq$=%|D3OcoF9XD*m6U6W*
z_;aVZvl$-AoWsSE4>MTc?M_LYB)_kPqX?t?OWs8ZT9#<(!tjxiiNTP8XI3@@UxQ4e
z*6?p0wiU~+<Lc>^sT~t6J<A$E%@IpFQJ0X%g<bhKeO|L%9U90XBk0LGGPF^@Vi|0J
z80l-ujT}h`QqP;rPR@UOqz1~k<aJrN`)!N8!fzm~`FVEQ2uugy1U)g{PX=0?v4`|A
z<IOVbiP8%57Z&x&Cc^qTf+P0PS#~EvK`hXMvLUwpuc&Dc?h3mS->m43E~w3J^iZC}
zIN}7Y0!7Jalp;4|r2ZTpUw5|2nyj2fIdb=A-HCU^Mx&uXJIWrY_&Y@4BO~8K0i8o!
z+c}-kDeHGmOS?X`?+}_j(Y2+DK$L+>QRAWEcIU!0M58!lqMaG;q<yV<kS+mDFo%et
zjI%PT8k~DAmq=}x9U84@zarzB10dz~c{*{d^tXM>*wk=sFR7C;lii3SI8uz!BBTa6
z%+yX2tk{Vj$0AX+4~zSC!|S@=DVyaO+sED@!J$6!CW#UU&f7>qe@87)wuJtw^(99Y
zzi8;K%;BP2tb1MqVss&EUygmrmZa2=13YuANARKc=$07+5bthj6{$Q((MsVhN;S1u
zRR!y3Igw(ePos7x?u6)Y@}wFGXF+gNp!#ut<LJDGFd%cnH3e{Ii(~|W$s)xV=91#d
z0Q}|TCz(0!q^Dfja1LrRD&`z(O~v>Uz7jkd2#1JxHkQL)8!2pJagI?@CK4WzdI^uU
z_;P%NU#Vp3w^=sDf^yprJ<`ry0pt+!36`bVF$5b`)1Mg%bA3kr4x>C*Y0F=lykb0u
zlGVuW*P~ZZyw2xjM)a(5_$MHZv86R=pjf9D=xJ~CD63-2=j!aIzT6XA-rr{?u+Gj8
z13)V^_0h4fy||(!Y&NPRPb*6m3&Vo&*L*kni8#sX(n1wPZ+^B%{TlhX<jgYTGR(Wd
z*^&``#lwI#&6k|4WjiZF(FEFxU*dnRicHN_w@+YYycCN=MK?v74_+c7RG+tg_b(Tq
zGl{u~kz|ZV4l__v{gJdpa$&sLITU&^TIlRuX9ml0-yE0h=*J)y(iVw=`HW+f=d7(9
zoK$7ppUiyOBA!G`8a)A2l$}?z)F4v}i~g;qu)wbosVDGnsszv8YTID0mSI`^<t2ZC
z0bn!xLjbHeBgKednIA;Q)L$@-=SUoxHw^)$!5(RkWDtVno+xc|h^n)l)r%XQ>_-hV
zisZbE8boN{=<Tve3SFjW66xu~3__Zh_1c8;?gpDW1yeWeh=bq08D1u(+|*Vx$O{>6
z<wZr~?rh{{O2p4flG7DUG^ZwG84f_)oUgCrc#Dc1f+QBN#0{-N`?y5T$D2T=vt8A<
z247XEQvQxAgPA{mh{CQ`6VWlT5ltj;iCr4j8;-1R1)YIX5Cae|i|LaIU&2%H($h56
zN19iXL@)T^dI@G;I1W@<i7R4_<6pk<;@D7t)qPBml|@_R;{9+jv3>f%j>sjLgC;hU
zccfDrOgb7YW1Dtsv(2w!QwjTpQb0FppXhNZzo>L!tQUdTQ=$|xDlLp1e{5%uP)0hd
zrAocT1%`%W(^Jkc=`ad>s6}9B8iFvOdDlNJn(79F;mEA7<MB2<iujh-&W0Pd8O|>z
zlcdDeieW5_ThTK3YmN$2^-m|5kmbeUd^eopJVqigrW6g>wH8{Ovn&%tE5QU6N<6aG
zW{=zzkDy2Js3%iKCz&e!b3We<)&UNRBEg(&v$_sr5)=1(uuzY+!^TqK?R~s*G4iMz
zX<W1i3f5QO^vR&u+*vZKuOiseYLya}<SWQtv*fU_p>dwye7J>{d8WgT@n1KPp>yHU
zUgJR8MisLbWCiuJXvqbvIwr{RjThU&Ds-P9=cKgMYfUl1s#fyan$#SmKmQ8n4c?@g
z3si+6M?MHC1mU%l^75%t9KAK=RfdIo5|wxofruUyLjb=MrbFEfNKiK4H-$OGjiK33
zaC?b1qTWi4io$<qKG9K8L@t+x;A8+obMAy54P2i(5wa$SJy)*p)d|ETUq$`5Tigln
zq?DionZ`skzsUW*45Azp9Dn`!_md&3gEJZi#wR&Bn5SOA%^M!+a2;V)o7?~-mEsGi
zf4Dj~60^%9#A+zvg_nu1WW^hHq|J=;gnvN)Ie7Tt7y#S0a|g+MS;yk$O6~}!PD%>Y
zV*dJr>Uw+C-;sX?mf&_%(i<j`OdNZ>fKO%EtlM=qRqw3}8b;wr#?#N$<FNdhfx!Vg
zbQ2+6TcP`>F!c$Gui7}lAeD0wlqN2RM?s17xG7yINM0|;=@k#O0;Ow(AD?($-$-#;
z-A6Z-mXDkeszaE=z4#4vZ#)4Yb_)wHKJWW8bAZ+Z;MBjItWMJJXt|zLC4hLib+6gE
z;?ftkmvPpn&{x{sem5=)qGI;IOAc4?OollAZo?S0YeLK5N*p=#dEAjMg9k67vV=Zj
z=2snmd7tsRkIGb&N>7Y8`{B4&nAMfsuf*Ljd#4QVActeF!wgMCgN!>_qQ~MY7gWk&
z&ecBXx>GO__x+~h2EG=TSA5CKz-^<Sl3<9p)`*oq%B`t)H$fMp>!6iyhD#_ueq4UD
zn831L7Z%8mUy6@fDfB>FGRwuqAmgu!?L-u3y~BkP%hgJQ;f0~rm$HXyw&S&skT)>b
zV73t|;l_wuBE2;o-{?73i<ya?7Oy*f6~?L%oM7(WCm5IM+*&FAfMEZ>S1kx2*xnDE
z_ppK0HX!v~Y3(|>qVsg!__U~y=Q=71ZItzP2etou*U+#JzsmSsAD`A|B+i)d6pn!<
z>*+S5>k@-k$l5x=O#K{eCVdH)jlgjaY29!`h_Lf@qlXt}8t$hInBnJzPWiC)jZV(f
z9Q(HAgDQOd24lqdt!Q5qDDxSm^e2X!F2YIR^55^B)nk5=UFq=V%U0lied*{K;_7}y
z$?|zlq^4I{(cS|S3ZZR9---VF;{W%(zvoY9p!3?lM|2Lvb+PSbvk-ed!1;CWSjYS@
z3hn=ecnrz2o%NPRNd}rF1rZ<EeAr1u9u)?P#imHYfFZVE61`<6%}M)IT6QXXdwk;d
zHx?iqJ{Fa`%H>A*k^H46YtgG59a#*9W29n!#`E>$`c)jY?i;6j@2hiNGf{)Xxn+~_
z(mjP4kwt{mde5}|LPiZg2onq!g00q-;rSiGv6-Zs^PRu<#Hr8oj2}s8K?Us<{dlwl
zNScfKV*6M6zn@!wK_dS1>v?`RZ`FQFfFt(J@cWS;nxyqv<8VdO!Z!TSRd;&OF#Yg>
zk)AQ&wc@<aB$a|9u+-|~^pyByZvcc7?89qr$mje;d@_eHpAmv*6IUH3<#ED$q8C36
zzQNbQCD~6pg>Grb&&KaIE<8`u2A|bnX({%&G6|;ImNZ-p*2`YQ#>P@1ZVTSwH*ws2
zi6SS+`eY9MA{gVIN8RpmECI)`V-<5~E)$vcZwvjO_X~tR5m6gAUY?|c2kr@@twrJ@
z*G0F;0oWYzuy1N$>W~B$cclaans=haB};y@gU_0b1`RzgM%?GQ5*Aa<a|Csx-gtPf
z{$zWrvRnIxu5PjqF@#xe;h*hBDtS9cj|M{0tzI!84}+Tw*GEniZHQWD=e#_GWb8c8
zA$YtT(44E`=6q1bxraxFnQvVm=qWqg&kEK9ySy#axLB}l_6d4CVGYO~Jy^mwjQ`i9
z{<=$q5P)ZUaUDokxhxCb<CLe=d2MW@_QQ)r3>)|P#UcbZ)8#Zp5DL+}Zf9)KU}cte
zgkz`j3;S=5i<kokFBa<JyB0Px;bd*y9y2^J@$;_~gJk+BkMfToG6M}BmyH^{=#%1a
zgoYBDXEkT|QzKtv{HYa4L!n4Um24S2HMXbNX{WpDg{It=4Pu*^{Wo#l%xAqKTCbc_
zv&~g({vQTI?l_yE6Pj$M!phbaUw7%bhxg5C{SeqdFJx}#0~s%2AyN(qk6YsLexk;5
zw%viOD_Z<Lve_{*@loBadyC~z?@Pt>m^7F)@FGD9R*4i2DuP{V78Pp|WhD>kVGibl
z;kw52?0nSnS?U>#y-#Dy)7{GoNtahm1f=o>G+Mi-jfWJ43?hp3D4!Vf8*q}#^rpIP
z70=;|>2#Zo`f1n9=KE3wh3rq4Q-@}nRXR;J)lCN(@s|Xqa?va$zcrystB(Ut7q(B;
z7xB-8ihP!5iEiCL!_h^Y@T4ZYvq&&^Xh!Qp)2((Pb`T8^i!`kr3n0j<cSa=b8<rC^
zti~L<jt>vR5D*Z^m8xcx>{RoenYg$F#}3>>Tm+r_3gYk|)k%evoR-;2o-9x!pJ(HN
zbp4i{X`>3L|Lf&!wgg^e?ai;*La-m{9`pO5vsUu`gzesGl`S{!3y~mrj#-QpJ_G^{
z1xl_GJB}vl{x~#r?oudGKpdIq*Yy(x-_A#Kt=sd|E$R`X%?|Kf(8qSF0al{WgVu*G
zB0;97LGBv5=wsV`98UWW7Sd@MG4xY*G6EIW7WQk3X{96~29xM20_vFJ%K!D+U(rTT
zrh|N<Dh*zw(bZ*-k)*f?Y?ry@1z0LjNnx1ZL`a3;zAOcAl1ykaJRi_5De8!~x0^DI
zyr_mfSvHmHPZuwKG4{n1OK+MJL0r^k19ToLLQr+|%p>L;dSJA9)H;8E$^W0RD9DPc
zj4XTS;Y3=pg}(YhzJ!t!MAj9@VvD|miH@v6c1vjMG0U55EP%?J5O74NrKiP4CsQ2{
zSx$`Ktk<Jj`j62ahcjgmV+fcA3EIAS_fC(5n?jsi8xtX^85Ke_`RV=HVMHbQa#p4A
z8=fW6N^ram4%s2@acHFMagq4g=l955bCMxLeaBBU*z*59I^xR2pvu^?{n+<ulqy5G
zXwr!~=ptgc@K99v3E~@juUgSIni^zMWB5fRdf}{DN|VSn0D^$#M0-Ueq&j;6;~!Q~
z&$T4B7zG6v@V?mnIv`*$d0L(xT~!QQ7*`BEy^|q>L(2XA#@W+1riTr8p0i3(yxtJg
z%3ZY7&?-;i95Oh~Bvq4twCUWbg`tdKCs&XvtV1U|x}*7~BfQbif=DdXEdpB=yRAmq
z(&<U6Aws)-#)7dMy&38lDRoQbzpMZ-ihW$}beBPz|FYSCO&bd85|zpXbZuV%l*uF-
zBhTRes}y}})PAiRg`Q{}W_gn+y9PPpYtMe3><G=dEN}l+24P7PaXBv5|C*Yag}oR9
z1`<`0SOB#)Mnb*W9OU|uqHS%#5+iw}vkAA#W8aQt2b`XFSIhNHh^_Mbf3$#U479+6
zI*1Ax0SZC|3!70RRsQ-3D9-R4j~EmeXa8I(NPD*)Y38*4`Pe}BA73vrS-!wP`(=;T
zrYa`LxAHMHqSH{Udd8n5xg|Q{d5hdRIKoFqj8QNrRP%p_2e1U(lXs<S=MXcxYj+L3
zu%<Eb(1K}OZ}~|)>|XDg0t^l1S^oEg;>yXetQ}E0EW;>o-5Kot$Sw|^AE$cnk(z&h
zkb7p@mP6xlR}cNiJqesE?c|+`Egvzy>9e!n3rer(^|2&`A>oO;_h!j|FDw6=qI*2|
zO4tY%3A9MbCjqoyI7OJD|D4AtXR$Neq9vktRYtGgmYc8ly@fDO0;~?};FwD;(-_5Q
z7Qs`&RFVI^0Wifq0~`&-tK8o(&D0noce>QR<Vx@2^$G;D|6fm%yTb937oqpYhu{Jg
zt&QjbrI+cA15Tnp@;7+Ed3*8>IEVQF!5N@QEW0DNr`28k&+cTApU~P2;Tkp%NX~3n
zY=Js|Y9SQ=^JCPaXSR<b|6^l+{%qS)QkMN^H?+Og7XM-O;>t_JSWW+&?Bpg%afW}k
zZGNi!|7u!?Y)NOWS^oh66-*iu5RtWw{@0B<J{El6O(Lsnd%1~WdNQ)Q=?xLX5&_a*
zzRexM1$JSN?$Z5dvgn4he?Vs0B*QhcNHfuYa)db4`>pd=SCT1Imv4iZ&8zfL+oIUg
z?}-ly3+EzmO_}_T9(4ci?*1ZJyb5c+nL$`6GK|tJ<&+tx<L}<_umV(xW`N3<FnVJQ
zi?1touN|i;JN}#-5m|^IA|pBX;l-wHGt}R`PF4k}YyJJ*zuN~ax=E6vNka-0IA^uk
zLtH`}vd2<!r1;n8%pCw}r@R@j+Pb>FKdi6E?eUYW!?SKg!)bm<4CexF`V{28I^e%g
zfw{jZ3^W$O5{3%7aMY+6y_<D>{o=0g9wb}#*Mz@E0WNxwC)_3O<>~${(9(39;Lx@m
zC&lnekNfoP_Yd+fRe#%tvrT?OCzeX5a4bvAh^bgei8at6LI1zb#sjc3!jT6>0Br!O
z;$22fZEcYewE(42WmA`FdQ&PY<fixkVG0T?>^LLvXlT`~yYPO$@BMS9rsQK-Vk8O;
zCT%h#LjN|7RSIAtzaIj7o#m+hNNBCv4Fg-Qj0}E_YhG1a3Pb}S0t~c#dtKlpY46Af
zaTWJy{=VD-YC17svYhu*G_T^^JA>(JI_vLI8OoL5dCN8aJ)p^5$_0%9p-tAZQGI|P
z;=zBAKI236X$<OIfJFif(y~=ltNYKQll(m-N{4Q5+3$4up%+%L_ugmbewH0cU45>Z
z0r-j(jf!i&&0mADN?<;vBQ<UOX$&AO>dw6MkW*LxBgEx#?s?y_wSfRES$>mM$majC
z%u?)%+RB7_BaU8RYmCD1cNY1D8_(bO=UYA7#ub0{XnY!RxtL>kyX%~2#Neg?CwqJT
z|9_s`bJ5yVp`Hss+;4g_0jI4#cARFxkA?7Z`uko@-e4KNv4)z^>af=Zgzo|U$(J?l
z|6IWM+qYPNskk7Y3jjM3sqEM9FwTP%+W+eP{tsLWUO=_eP!<r_onB+z{A&P#VN|eJ
zThw782gjh<+ULey0P<rH0Liet(BBp&DS}oFoDSgh*=bx*0N7nPIXTNJenNd#_>a-I
z2`MYvhBn$ySg{uX1dm)?<;NKn6^}Q;d2>2{U0}WD_(QuRpiJb)FOaRkq4Ia9F!$et
z0kblQJK)MYggaoy+J`#;2*%>ZkGvf6UjApAfsu${fdoif`~37yv%XKV%`DP`{?m69
zW_CNP&}8w-Y?rP)w>9fhGV%zJ{d&&-t}(fOk4LzsWycxK_SP9NXgqV`YX7-`&m`GZ
zY(p`bRL%v+pa8T@3u?OCd=YUnvPrgq<eNxRBvdk;2+vc<NMYEtMj~M#U$Awk6@|7i
zlPk59k0{*UjeebZ%AfT4SYzdicUKq^UG&wKwEY#TaPRXj+<bAN#OHG!qa}**+8)G_
z-nz{NH|l|k1f#>5dMFdaBO)<T)PR8e%v_J(ZQ@$-Uip)eD~s3*WMX!;sI)=C%(H9Q
zR&`KW?rBCe18*Eiqa*|>Q@<%D0Z60kJDuHUoy>0<verF?N9%Br@BN;Qf<w6-rz5fY
zlKl$h4^y5HL<a|v9ic+3$dxmixKT(yyVcR}fb8oddoS9wK3#)4_`eLOlpL(v9rY-r
z9(Ki)Mz?--)9*>u!1JdYAJ?wd@svwG5L~xsn)#G|{%K@!yy8MwynUGHnxKyTo<1|Q
zlZ1Zr^UeE>sf6K%#BcF<3H(&#5X59gb=GIDNsXD~%DVz-@J!_;Afm*y^G@j$@m|>q
z<y36RfP~}kUWiD^WOZ+yT#HJ2hmkKxQ(+`2rvmJ<L=Z`l2Mf5axs@}f?cDL3i*$!8
zS1FZ;4~x61&*PoX`i+gCT1{YRu0vRbP<ul<r*1ct6Xv%dvfse!`#F5?bhw&Fw}wz*
zK!P-eYF!N7|0tFI{c3jd&?%5?{R@Oiwd$|-;?G3b2lw<g1b+1e1*!FKi=*o#HzmOz
zu%gUu-716U{E5aPPDI4VsvdlK?;H(s!VoMnrJrd-vfPY`rp<wWYPX8dv&jh;AjTXe
z#kM8H<O_C1MoYZer!;cR4ZCNLZx3WQh(lj0`H;g3dKsLAAe&x4t|}PVU2<#3$-Yhd
zaabH379kIA#WbyclW2zvLVXJ(8xb(rgw|vi!a98<Kg-8mM>+RlE})HZYEwbe<<JY`
ziYG(=P8HJ}p^H07-gY-ttR_1IIr~FR>JIK=-L+c}%HFHjgRI|<zJ>fFtNGAgh1t{k
zF)B@%Vrs}dvpZO)f@^#%K-$##>eB{aR<WnDkox>nyG}iGjPx~3bg8Cw>XQr&nDWNg
z0cx_2Z-#^pOSZU4zqx$>l4FdFJX19xZ7_MyIYY2UZ|#h^OX@2hn+e+8+cuQ(4TJPH
zgzd%;jOYQNC0x{8B6LWq30-<UL*~|@kNP~;bIQ&2DJHd>k=f*7l=4?9j?0m^Mw#9%
zxwPig1J(T>{TTS*boV>NV?e98>Ew{RAN$Qo5;1@Rh870hCa;)ul_;J@WU|O^!U+TA
zK8M-Bhe*eSO{JPjoOdhL-!hae%#{<;s9?zy4|-84H)cr3NVH+czECMTf-@S6u-&C8
zSEWgb#cZ>!Izw(&4qpXiGH8paLw>dPim^F*jJd=ZVCvwAZZm^FbenfHcZqQ`NZ1gK
z6c9m6uH#G{t7v)EP>(l{UafRU5qk173}90-p6Yh`$2*{rI7X}<WPDg?Ayj|TIFNA~
zIlrr>c6!QHRd`e1)Jj<)ZRg)m{F@$;i`rJIE#;bz2Z#6|P6tv8wPB_6hp#@vi@4&8
zHT`aF>mDj>FR7~dk-O<<p?JH<ae+6Bt~~Xn{J#@1->#71s#F?QOOwUI(g%bPz99Je
zqt^@NLFb0nwu^-&Raj6IA0L%PqpqyqVHyc?f?(`Mn`}Q^26*0Lq!#>4{)CU!lx)9J
z;uIqblkL~@s~ObB!V^}OZ)&6;!!}Y%Xnb^jWqAO~9dNwh>~wH4I<eTc7a|!wbPerF
zaN5-#x)sn+wjmm?WgTxDorN&P?p%;_O2J<Wd`_n#(G#Ua7DFGizrJAm)U_#PEuARX
zB^pVkq9H|J4(slBhD48h)Q`UjO;Ay3{&`%DO08NkdnMO4myuE)KL~<6jpL{kXW@>)
z;MkZw9g#RExrEJst8xs#ysgsW^Q$M1{7x;3Zf3WXvI0CzoCyMb{K98(Ci7x?xEtL=
zp^0VRqplNnO@{9qzZ|4qf6tE5opEoeC|C?OdC(Jwqt!xzyO%YzymBV8#wwg#vsaID
z#F$yFrd)JvW}Yuwpq}yLMieb#7^>Xbs#9O19N0LyJV_EcWC<{Frheqiu8Y~_4Be_T
zg&EW~l;j4WJZ3;pZC_RLdqPc%Goaz()ngc=HCDzkrCP})@x3egvggvnAls}$gG|30
zIchX*AD31rr)>zh)Xl~;DMrbGn%}k??DpbbELU}fKStj08dS;(xf70~wTzAQtkj${
zQr-=>h4$$mG06UGW6mACrNQoh)YCQJ7fX-Ms#l9|{-T^u5@3DS=Voqv)NsWWgjpAz
z>`Gx2fe@sD-n6W&#M8EVC%tyC;nUM<Uf{foWJM*M9Mxz*fNCH0lK|gqx3e4Rdi%a|
zjF++s&)||7^$JPi;M|nOtbTE#I`LEZAh}L@Re3O*C{B^q;BcrSmiTJUX-<5f7u}0e
zSy3My!>|ug_$Tv?_D|5-t?&7GZW&sfERGsZ=OG+o+Y|S@*wyD~<R`<9c0l=Z-eC>I
zqLCRZj5bQz?@c~-=n%%0yw}}z%#Xz+YV_b!j48Y;my-=2D&qKgwH8W*3Uf@cM$hoM
zn-dPR#2<O&eZ|GwgSo8eASpD_Ku=mQ22Ak9GZzGBk^cbbB4LQum>3#cbWL0?%8G!K
z&nD`KW=Vza`mR0b*U-4K^VkO0DNq3GTx`T3fi5EyptYgVUFAd^qJPo%>)PnPXDVKN
z?`5YA(=(Dev<PzCy9w6a8eyPOKN!)yP+&siv7NWB78TnBk}K+iBCZBZetR<qFh`!e
zscIaH!Y-_&JX2QaMBp0nk}}dydOf(+tC>H+m_+!M$@b23jC=&s>&eDz7)QAx-UWP|
zS8L?EPu-%UXy5_Y0jHdgv$8o~0A|S(I0bn<&|77}TPINme>Q)EG)0-j+i1C11nZ<b
zDk)jSQZ<qi>XF=cfo8Jp2Yr?UmbF|(Wy-F_b~x1GtHag~rHw7ucLp%NAG9^PD?h7c
z;T(wW7-QCKE~uiHmjwjKCSoUUUVbX3{&2)bO8=t1l<FzAbVq``e}?|a`n$%6$$mG+
z(0j7E%x(tv>Om5uEW1MKFc6DYy5!d?go*3cU36oL@)MY~&Z$Td`gaSYsg#`_sv7WY
zzT;qBLismirmzHu%DU$Ij0jr1YOdL;ruD5XWi)-5C?SGBStH|ymdJd)z>yQ*LHlMR
z44UH_UdMN=7Oy<D%*I*@@*g3!!<&B$rWjks7i14Ln~T_PH?(dC7Gz~isw&_0JD!gT
z-2@;Z5Wl`-?S-i#v7G>~tvKi6ymu)s=VF20mR&631qzCWADCz{l|6Ac8ew!a6qy+|
zp-j%`?QUdELyMb9C7os2rL?~U+)Aw<K7Ye-klmXXZj7g@uS&*R9{5yKghihS)wS(0
z*u9>0;iibSzh|;K<aJ@kfKnY!{82UzJRZeE#nhkOZS~`&E1Zjwal>VQkrG$T`DPk$
zJrS}IoO}n$x`ippfA05$rBxN?Q#2C2v)4n{7{@kOZ{1144pswcd?P&;#WnLE?+8m6
zi~y~0FdN^e)6TCm{NgI;8?D8IbLGJ<dz<y<2kJNQ=iQd{w;#bwQGJ~fHq%++degL{
zLEKIBYUX3ShMk+9Zw{ggo#*Rd7=r3QN)6m4YKx8we^^CTd&XQes@K6j7e&u$me83p
zJg2l1-$jP`8oG$D8`t^lJx(M2jZu7}V51%PTMNA5xR<g@R7uaZH$X>g#?7q|q!C#G
zCy{%bGOxG^=5$0B`xHtPPHrX>PHMY&CNXdJg()#$3{6pl@J%ohjlLnOsI7}kXXljI
zk4Y~?K6&fWN9t1k=)^RKapoX5-?KheD&%9?(IexjUWk?z@Okq^R{`~UUGR`7M0?;4
z!z{vm($W{T=mjTX!k0C2leE%c!fBx=<~I9H{Etx``t0ZjEN35smhkTp##xl+=}-iI
znWquc-^@Ji7IqnY=P;+cerlpDLnP;YvLg85q99TDy;vP1DwjZZnRknT$(mtvzMTmK
zD7uv1X)7I+!z|FGoM1}#x<J%HTRWOWaEDYXjUg6<D@_q8%Lk-+lhdrGAO;2u9RA?!
zi#81dut@>>*flT(6noU|JA(OQ4kay{6(w%<1yzmzo0+AKs=uoYpQx<2@BqsZOOs3Q
zrk+4s#`Hbyh2mannX-_x%Zus*p!jWHorXT0XqK>5I|=>eLEK9}KfOd#mnLQ)+SmWx
zQc>}`vgs`rf=j_NOyu9-)8a!#9+JRMU)l6>OkrGgIx~SQ8k2|$N#%W)yd{L4rFErf
z<gE8>hwYTor$dIg%&?kRwGU!7WTOiSV8bFi*n-GtE+$rY?H-6Yt~tEOs%(m&EKNe~
zPqpRm#NPC-7~k<#NK8IgZoAue?_fj)=T#80l$>kZE_8c$X6<j{nLchEwH&hQ*j@;t
znhvhDIfQ{+v(IOkLIRq?s;`ec=tkizVrJd^@(ofSm)(fm^ih7A&|3cL%=!dsw6Exy
z)&{373z*KcbSW;g4oA3D`wM1Gyo&iYtxk{R>k*7pO3!G~1cM%GKhRB5M<!`fNZKZD
zO2COp(~O(MXnnU@$1AKT$(*iDkySYSTIZb^1*CtBtA=#eTN(yt7F^6-d|h8&^*^ip
zF7mRoTIXy*DfhKsPaQLx7g{JLp+1%4Q7FGxSHB$rLgvh=2o_f2_$N!d(vMy%D9L^$
zog}H&oSDV|&r!?loyU2fRwZuG`9*$VwIpi$qT)U#J}^FBa>slTK`ut9Ys~}dVcUwG
zC`p3=jcHZ}jYJG?ME_*48u&s9PK*KsqllnfW*2&{(xJJQg}dO!WwCkp*-rtgGH`n4
zsc;b|`Ocx+kayb1H@Yh;I@p4ZH3zgAd_N}CSP`;1sj7}$FobKhzW5rB>cVFOqLV4)
zCUi;eu3aqeGLj+5U`oO={z+h?xsW1Qc*6I*VXu-v8&!{FZHy4WE;vo&V=8wPaPY+}
z?bU7acT)UEBXxxM5!CqxLA#4)+vm7hQkZj=rRn=si>+7p)Oa{jFaSt;v{jAs?wL&>
zx14uV?9o5u#5}l%%G1euy|Ja7>gmaH6XJ0h8LF+~FN$Ib%F0mYq4C{(+aY=p;#uD=
z_!j9``(T#eYAJiLxjmP5)-mmIC}OT@Y89DvTyl=Bvqs=^DB+MlyW%4==}mV@7~Pia
z%hcJ(C0Xp}@Tc%j`<kAnZ;O3r<|KZl$;M2Fb}rtTUWtuV$x|;j(gm}3KvDv&P8d@%
za~Yrb+|Oo&BruhE)GKMps;JW6e99*X46zm1bbU4=S*E9RJ=QWoCm2vpt;r@6onIIa
zg+dY2n!>pJ|Iu_7Ky@_D8pb8Kgy8OefZ*=V2LuTOcXxMpcXtmOf@^R+5ZoPtyW3s<
zs+%eb$rk6ZGt>R{JG1??>gX(<U5gHjg%wrr`2NQQJVJVy8^;7m5xPk+U##0kAz_av
zgfL4iX^*1#Q;g!bqWXs<BPAa{prW2|D_8N<MpL&GCzp;vOK3VSuSD#U3lgHSZ~z>1
zGGR3Z^GuurLba62ADF_ix_G$whRfeMQ9mtK9vx_RUH6-#8`TY><3(!SbfXp5%^*x_
zEprm+JJC#Y5)InCha-<oe|_x8p$awjk8)TqmyCVmPqV-;uX?nyGT?W0N6R8WDY|sT
z>d?_@p+FnUb_!-L15xV~d48J?Wma(Z?(tF*s6_B(Mwbb$QIMRhR9vXa_}w|2e&833
zUY=8A<%9FHau!cY#>Y3DLt!CpKuhYzf!R1@RpeB~E)q1AK^geSsMzYwQvI8B>UZuW
zjad7Hs2V6xj130|G!l;P(;rC|>OLOGD9mny8)G#bs+GWD(qLGprf>nN-QP1$V`RD4
z(4ukV(f>UQIZX$2SIUN?Jc%oI`om5bC$YX`Y{8Rxe*#R0O7GqlN>t=2z5pxEhzS#s
zkWd}|Hk-AWhtG(lO(g&oUg1A10DrzG0<Q?Yp%QbPtgVRrQ(xKGh_MvcmmFPSaPe*F
zvMUHTf=m>-Vhs-mcK`p+0w5A8EU@B(%NT5uvdM=hq2|B;tVmbjs+%wM9iZA`xCsoW
zgciT~Ho4UfXNE!slet7K8C+TOZ~41j@Zw<v^C^6BKr?zB+%o5H2OD+?aooM8+GG66
z%F#&_ou@V0{;91Wu*8Nd#cv~O!6MJICR^2xsha{I_iXX)pU`gGW%ihkrpQUl(U}l7
za{u%@{KmJK8@e%zuhVM=!tj1$(brEobE}Gd*(Z+%0h6WKWDvK~5c?B`C~@)A$>Qb@
zSy2)J<Yh|;2iuzP(-4THPcP#;KNVU}2R6%yiYm#~sDtdt)lyV6aVgM4N+C6BH$8D^
zLIPA#C@-CtrVWQ?(j|U6!E6M44aS|pW!%r2L-UmVl$34YU_kxTrlpWDqr+0?OuYyk
zrw&m}9{hp_SA29zrIR&yTxB0j54SMoPp8nAm@;xu<7LpAkdB-^Sh^tjquY3PsvOHG
z?~vB!uHuCmEMPd<mfEyOLVK)%Ak%Za*vhU$wYPjwI=S2Mz?|d78^?KW-sI~>P~lfq
z_c_6Uk8TP!E<L=g1T{%5Z|>57<=&`9!*<El`ZzVm8`t5`W8#4D+KC86njjp7A$8{X
zs}DEaq5ogpXscu^$#)e?(YJ28!eCs{6IO|xdkpK4dwml$Mfvp^0x%pZ1k%d=AN}_b
zmo?FJFLxB8niv_n1=0RUN+s4Tg6F^apNB%WK%skAOJ3IPqXdrh9aoPHeQN8b68rQ+
zwlMb8o5dP-J$yZP$IeRfQvDPX{fCrb0!F4|GRu;&BfyDFT!4&%)Pg6V)K;l<rYIzW
zJd7UV)OqVG?&8^Rbi*+hJ90btLo6&IHV}@O0{Ml4LVs4Wu6=wd?A}-tepDv;Ig1~@
zw>1D)N=QR{k!D3xRoAoIeVjt4PDStf{^#!?-gYSCOd?Yne+jtbLzzpyLWD|427>}c
zJ}6c#h40DFpMfHrDVP*#T@uvd8<@oF8&2-wNXbnvKdyUtYQL)gx1T=#4v`i4gj?ON
zuC4N4=4@;xe|-R@DTFiCgpYjgmI3P^E45E5TuJIw2`oL;{m?0P<!F7HFgAmXv3co#
zYTvQ1s69wDMbtOpP6-J+;8C-4`>VaqBeH!%r+%|$%mx?8He+FG4kHP@&<39KMprzI
z*w2(eDu<OhrrOS{%zaJ>2TR%=Gm?3F%O^-sIno&GHDiB<O~;hMMsSa2hvDzaL1F>s
zxbo8_rUJHOf8JLrPopV{Q|a*yFx>VshFJb_=36rqh7ro2jv6<e<&8GX42(n3v`!us
zk7y9C6%KszJ;qKjx)_OxVNRddhSU=kNa=}9gC~(nqFqhj@Qyp;RRB~<zHepw>sSsU
zfK!SYM@jUUvA%ZYdc^04B8Pl_C8N!;m-32Q;4e-A)fjN@=qPU@@*<BV?fz{%jNuK?
zf8e^u!r$!rf{c1+2BDOi7YBvs9)pB@zC^mb^cGhYqwwamYnOIoR<eIqw+*BieCRwc
zMK4Jhh6V=w=6+gq<U#J}&p(}<zYo{hcAygx7+vw)k<A@9A<f>VO9NXhr527E`zE@v
zlyZKDaRj3J>}eBfSb*@Vq;}SkIv7zlz*?^9!<7pU#E?S3p!lql<;xf9A5P|sF_M(M
z?PFKH>0{iq#99fXt$}cgTIILe9lk50SfdJZMu1d}OLx)FVYPwAN+QG<tdVWq%0dlP
zz{;cd`Wwcq@!!a?qsckE-h5F)NJP^_UdZ>b5ppxf;W+U;fitmmr9oFRS9W8LQQ|;B
zrP&^JxZ;VVmT;E@bP62;)=9&dUT)VDxwug5?{co;5f6ffa^;mC#b*h1MT7<GDdy5a
z>C!hN5yr#{e9q3wc8(HD44tIno}5K_QMALt-!7+`{xj+2bGJ`ys#U>ejfYB!Y`)h=
zp6@=%&nii?P3~4b(d{eMzj$Q>AE)IXsW{1gZyPN==Y=#4B4M1SPk64ipbI^O8)b_r
zfaYS1+jVJb3MOKt^K4(unkwdz(^?ld&mec-q{R3gTY6^Gd=4!=1tg_3GtOs^i3Mdj
zxTDzE(Ru8=>KeJ*sei%$iuJ$$jiqD8qJYE5!*|gJMaft#HK*}(V>t2#J~k1)&JmQ~
z1t#qB{3l+_%+=qw!ze%Z`lAkWq(Cdk8DY$XZio}g<}Q(W9iMDGZswh0;W=O!gnhbv
z(mLH-=4HVe`HO-vvryo5*#$%jLUJno8H1hRI%qNH8ICGZL*EebIX}P^#w`J+DiynI
z4Rp>aw|a-bE!_rXdZ8r$w4TpLJ(m31^h?|wmvn1tz2hX3)<inT%K?hfrOLhj+h-qW
zb6KgZV`QG$5t&Li(86aVFUQ!V-g?>2!P@?WCK}IZhu}pK>1#K5XF>EInRcm_U#>nk
zV?J4$VQ_pq-W8W%GC}%Nex7V7UpTJqdoOFqde|uw>UJ3@EeoD`gfoKKC>j1qN(5~F
zRAkA@mKxCa7*V;7v}|xF<Yl7f*Bo2d@n4-(hQm|kd;<le&LE!AQO}17(pxw={Zg`+
zlcM~oFG*^edsXY+*nFB?iKF`#c`j9W!W0tEh~G4>pRd;OrxTupq6Wv*K(aAm!ad!m
z=+(Jiq2f<NnrRP~%(Q4&*~iX}%nKeKKapWlT>sCkyBs8DJbRxL1j+;|+2I_QyPQvu
zASk#t{s%<Ny21T|1KWi$QV^z5ev}Mh6c5-CTc)C3UNJ8XcIltt*>2*1sJQt`YFlN}
zP-)0yGcL%>QF=)ZuYEC*P1C(}DcveZ0Yh8UTRpcBCX<^YQ<`*B0duEq3PS<&pb*LV
zFS$p`I3_v?)tr-=PUO@T6J%UVhaR%bLkc0=hL>W=IhT%@i4||2ild7{|DLnKpyrEK
z`br8J^`t7dINtb>5q=M>QQhQ}t7^JJMz*UzIw1j6PKdH*K(AmahcgjQoXKwf>3%GV
zTFPMKycYQg?$=`#>QKC|3Ew3BD{C_L!KF65LK<qTd?>TO<?x--z52?Wivl&pW?&>K
zqPSbk@L2~%ph4=<tA<6b{CaXjffF$`LTB#$2B|1sF|&=tm6b<C6Q&sQYSWSki_Rq3
zQ?T)sPyzZh7R<EuN?2|6c;)nBDEc8sjjv1E+$Gu^4B^PUro{t8{wb?(L(Do#0cd_l
z__@v_64CTxVKTa7iEtUV$vFa$c?YF}<1<C+9lDOuPS7O~Dia5tR1zVxBLPk0zYioM
z8U+SLu=_9&Y?E>Fakx2h9QFAN3i_6kC&A>61tm3BZ8d!{M37=%9Q7K~NS0ueh_th;
zgTJN>>|R?!bi+euG#+MG;rF-iCSw-Z5LhRdAS3|i-KgNdxG-uY=YS;Yq2j|qsbkEq
zW^^VNzXR%tx~e05<#iV1&B7|GFuH!FXM==|U5oP;v8YmPM*}ecsx330ILZN$ZM;cg
z@wsg#wZ!K*No!kMHvE_;k9bh25<EY@Z$@{@4&TAkIkI0_F*@^g{f+{1pmc~dCEJA2
z>MM-h<7VFbej)m_DzX_>hs`m!8E&;2p}jn++BHTXWs%iTw0UWjQ2kDiE}=BE!OH7s
zlbSIhkH<fI)8Jl=e^-P<`tiw%8FYcYQN8`iOH?LK=PE}a7yR#5vI@ByR7qCd@!3V*
zmI(b*2!v1_OIhKV*AGpaTKw1YW>O&;X=+E)O%;$~!<&3!WX0J=wby7|P{^@s+F`L%
z-e};e4Hx_9*iO9Eumxiou`|E*L_&~{plu@^ua%z(>9~H}{cg*)cJwcf#~N;^*-*R-
zx$bKr?tj7cHuhZw5uqp*Dub~WeS+<IHJ@?08f}J#c@5L7g>+6>)vm^9zpH!G1e>cM
zXmITaXJ7esyhvmV^43|g-5_*AJCa(d1$j;_%y@Dr6*C5T8XP=-zw&9auI6iJ<;R&b
zKeJPbx$}-*Lzk=&l;J&~FZJAsCJRQjGh}wF3?K@nh12*ngZKHjqRRhDDuiBe=d6+c
zCR6ZiK!1mBgtTs1F&jEESx<Q`YZf(?AKP+XKb3F?73zGamTuZ%7dn+_<2N2A)l2F1
z)eBtq!9`0E1TCCd7<=`_)pR!EbGZXqdEK;<=Cz<5k#NY3l93H#Z=CgK-$Fd2@q(eD
zL2J+O8AZOBm>85JPw^XKtHGNsPK;vQD$k&~x(hiS34lUzxxN$KyU0<BlvtR7k6Pj~
zvq&mS*(ROUJ{9lRw^cqUI)u$Aj?833O#Nw_@;qy_euNs~TNC-zbU()u^E$IGzak-A
zQZS1O%?zuY;Q~fD<<?Lnm2CeI$5K;kn&sh@16g|fna3MAyf<D&{gdbculHRw2)BQK
zYv9vQ{{TW|37k-C&rUq9r#4Hcb3J8HjH;xmI7a{;J7KK8s3PMR1nQ|KrX9=Zk%bNY
zIzvQGA1QR+Bo$`$B@J5XPZrUga)$IemqrdXhB#)?RX#kkx=-k5|8^=<BmYyEA77W2
z#uc~1L*^uSho(qNE;*ZmBjgV1qf}~mt-}3|pAJ>+U!An@-TpeY9NI7i<pG2uS&Cup
z?`h4Qes6`JJ}Yz5vOohGhk`d+K1b(KNgF`ShBPIvtS=z*u^lSSTMRj|T0<s8rGwi1
zdwqIm(nmy>Bc2A|p*BiA+#3(~ppywMM$8)@$P~6sG{W4a&k9D?xQ+N)ht-GO(k}A~
z4xmq|$dFLXI_V=+u<YEC-lOL<qQCW3#V89f&D)#Wm47=p=!OwqT?9_t%$Nt#JcYJZ
znw!BnJY%#r>+RJ2BvZrfvRk2~MOo06t4KKggCEuBGB>)M<AtZ~%=2lH3uR}NY#M7N
z)XwK>2O*K4R>5WoCc5g`A0}ER>GbmtxxA1KwgCF1XJ^6x&JBWf#k*uxMiVQ8UQs)^
zqwg8r(#ABXrL{JMm6J~|i55F1i-pyfp*3`FQ*TaC50Dzt*ABo!HwysNd*R~9?-s0~
zlIr)~rnmjM)*UwPQ3@?fJ2_tog?m+<P&)PYe)CK7>tmQ;Z?oPV)1$EU;k&GC_gg!3
zl(8;(FlK$z=&eAk^bc=3I76f~y}L+3VTKr9$(3BJ;$&rQPBhMkO;Ae+X^wDy0U4Xz
zb-HRi{8O72Q5H9~COtV*0IFEbu6xul98Moh(ATm92O~KICQR$AZRL5arMGI*hjjAa
z@yW|&zv^Tnn_8{V!kFRS+4e~|+r7Rnexa!$$0i(U5=8#XR4C3<Rxk|XWMuc_$=l-}
z*UQgjuO--gapDv(bkyK(XZG=bPfUesJR8&cdBT}Vw!0hr5d1KGzLxX7;5c-rMy1xj
zC<xBqI5@9reg8dXqw=Ra6Gd6cO}xA%(+Oh`<9VA(Uz?|wQFA<=W4rHUy+y~gm0nF?
zM15}{LHU)DykN5em~s{?jq>s<P5rodF_N!Zvjbax=2A=<A)_7EEE!ax$17<(YgKxW
zf0%C1G-(&6_qvS&*aKvwMjQJXh4>y{WH<xa1{2nl#J;P31e|zkNh*}sULJD-useq=
z-A~7^#bn2#L2)rkyzoo~u$0hMn)oIE^AfnUQe`(*e<r_x!EmV_>}`m^LuJqA{ynR$
z1>ln9Hg~_Cdhc-gQn0-CrlHJN(hk-T7GSCzHTxvZWGI%=t*LRQq7b~V)kz}c4273u
zN&KwFZRM?PRl#aPCWy^_-C<8B&IMK8?fCRO3#%<SpAO#X;kp|Rt!ftw1r><9(j}%F
zV^}d2T!;U{rD=2+jYg??dVtS(qMIKPH6|F3iPyp5m6Q6ea+G3a9$jDo^s40<WpqjI
z^NB*xq@8vecy6q02yYwnp)(=$b^G_-GP-w{_NPj2!2=IRC61DMx?9g+x=7oQ5$U@+
z=uaw}8}y|cr<_>6N^F6K4V{&TUWaAaLP55jhu$5-0to;cm+g9h{Z6XzAoc|>%ZLAf
zL?|uO=bKs%Tf9e{!_j0zZt)Wy<v7LfP__9Wr_jZ5^d0+WsnxGRLhq|gt2=i<H754<
zphIgYetBlQ;##198;Lz;GaRVu4<Z!w#yubea2Q6sIi=W{OV_Jfa%yVhrq{Riqvvxj
zYeEMm5Vp}WJ}NcuD;_B<d?JlDD<sX?3-)<z`ASyHGVdoS@NX_T^W%<_n^`yxsWnL7
zTx1Ozd8{giC=iG*bG3Ly*-iXxJf2j+2LLCFf}y7j(L9Gp#_(o%m<{?Y_A9(p83kvA
z20S#3S@`x$X2=d=md=2A#RNSD#QgHxQ*TEnJ;nwbzBP%-L6|p*93OATUAw`RyNDzi
zV#-Wk^nG5hSNDaQA$t+jU`oMSI$LGZzt6mJBIps$`-@><lS1xjL-;#>k>J@fr2k~{
zub3$K(|$a-(;L;zqx0qIaAS&yS{^@VWO=4Aj3#KFkSEUQYH6ffpRQ2CMJ?h}>Yqe6
ztUf$$IKv4&RW4i}`MruU(2cck+~Z98w(O|YdVZFa$JCW~ZK1b%#Gy&ewjkecjT^bH
z*|km>LLj>cs9Nt;&AY@!CFG7TeEU_7VVG}$-vw~lot{sdoyM4&?LR(Gd|kLu6}nZ0
z;D5DHQW}bYtZ&$a^zl9<jaBVWwTS?c&IR|FVt%&B$kIz&X({0jB~tmNos=YMGG@y`
zSM{g+c_6O@1b=BNK&YVJTSubf3L&5lKbsU2Gc9s-2Qe8JZW_wPPAOmKi_Au`jB=!T
zspt3feyrwCD=4*AmA0XkIK&5v3q?jQTvV7`WLw;z7P~)we<r#7>V|qk_P1%=Z8W=V
zTrKqg$%|*w{;jdKb)?JZ8E60WRjbD%cB}JTBHR!OJwY&$lW9ZPjt_TPchq6DKJ<L7
z#nW60Zy7(bZ;5qgT6b_OMX9MmcocI-fmFKq6J%QVQ6QP2MRZX;BuOcueXh@hf$qJL
zP!Gku^Vc}`K@=?NZKJTyti`v3Qj{Xc!d~cxIfY|}&<>dba+S$qY1s{;qr5DweF-A7
z@y;VoMKS0!@{&fr+bJ7kL_Kdhk*zw#8+B>x7M!fuieVetILvWWqetBxd$0GGSu0NI
zvA-r@IQcS1v5tGdT{n;Xe*zDYxx+vnh$$y<8@m|LCleX}eOJ1fr_;XZ%5ck8H!u6+
zt}-)es>wK5OegmRO6@DTq|yn$;pVTrH%5}jsH%%Wr$Ib3`^$kZ_fX0p4k&Gy=%NB-
zaZ8O`I58`m4GTS=+d1H@T@2UyFiF=d(Kj^-H8meIV$n!*I>|XUgp2j-aw&mGyXPi5
zXb=oZJVi0diK<R-AP6-P7dS@uus9?*sL;IXoj2(ueG>rmREs))nLO{ujrx^}%S+EC
za~#>Kl9=;}pegRV-a|ro`nJUqM*mJ7hXEE>U_}Nw`S=TUQ^O+zsQR+8_RmlNSw|r0
zA8XEIl(C{$3RJEv?&EM%!Sce;OZV~K39qNx1(ykH&Z0_&pTDgY%Qy@{M!%}0agB^|
zxInJVH^|4CAUK9lJa*iezn<JWL*^Vu@Wu05nTzPw_&<N!j2ZiBG0+SS<J{jr`x{KV
z^PO^{V;sGHbt7sEtE?<nk_%l3nh<z}&{+L}jEy0ma<uNuYr>XsbBp)a$<;mHv{alM
z2GJ>w=<hOTJ@^ROqdY>q{KrHiPxjxazlI=9Lb81q>si`1UvYoo<oK9H4`~9`>R(Hx
zOw)le9QVZ3xeopoU$^czp2^y3*L=n$S{}BeJXUvz(~xyQm}}+*Xyr8UPl(?ZT%v)`
zyhQSv=Bn>~evjaOT#r72b^qF9-)FDdCj>CSHu%fVk$dCg&mMd4_wf3|b;%_CM01ON
zA?i5q{m((hgej7Z*df6vOue`3<6hdmYvcra9mUwzkI4cE)m*uA0mB2V{ye`Oa1e&?
z*})sroH+?5v_rCYe-VZ^h~tc}r?;jr!hj=C>=e$$<Wh36*JbOi3_<zP=AXChoH<iX
zrMXYgeZow(7(YjWfpKrR**t9aNfBmJVR5`K$ki-yMvgt_mfXlOF#{;6aGpFONXhu1
zq$?>DW>|^ifF#)I)ME=na?v$_7xV>PWBSp5sQC78q1h31(W@EjX`3-?je?!M=IF5-
z35yx#Y2`FOtfR%sv4en8kT;AeCdLK`6dC~xKN==XN5}*xEB!P<kVRF$@Q?6OHSE1%
z<Ut+~&kKk(GLmjPY*!~q$R#>~hJQ=ur(yO@FPaZJ#+cGef0YwK+80aks>^~7(6exk
z|8YcV|0FDX4M`*LkX<<#-4V{8gV?!=Tyks3_eAXj;j-#+kmG$Qr&nl^xnAcv|J}^t
zto6Y0Zpq9E__)s5c;x5gjC^;;oIm*_{#I0kt9hLnXxOl>iBdR~y3$l+^^2ljPZS2a
zI?SA+*VA{)Y`q6^Qee<YljJEH_Y+E31rL#!!X_(ImbfGW1-YtL*QDP|ieFs8AeVj>
zWvV!oDXf(ZeHG1_aT7~Lo>Osq0^`1?lG#?>x0i=F!7Q4SOSe#E`7xni3z^DV72VxV
zSO<h?0#(a~Yu|AQfYw?M+V}gWPdAqxgs{JEUia9TqexI#vwCR+9lnA)Z|Jl~1)|TE
zPk&T(0O&+U=4g<R?f}21wWKWU<J0?rfH&T}*lZ*mNkDeuS|qGpIL^;3R*LGq8#6l}
z0=@2Q^tA4WlI-VyVN_yS{Tz7bl2e>7y(_+H?dPw&8_$2fT!h=XJKq(uIt^I&0c^;Z
zGVeVry**f#H^@oPn<5@<m78T<MwLBJy-WD>?X&jteOMNENQi?Rul+xQa4oM4VGpxq
zf&(RKx}$hW9z+<Mc0zgC+8?Z2r!6cJr*DWFJDYY9{m&sJPpGMZ<8cuRn<72jJ!!}t
zXGG_l=yX;2*Z$vv`EBb25y|FJgJrOuXqoqRML*21`GETo?tebTl~&8Hp}*C%s;FEw
z>4*=BFrjN?75<tDI(?nQi;nQV17!PbRG$1EGRS%SOEAqr_?kqvrYDOWOJ)Fk7bsNd
zrKlPDB80^o5`+Ot0P8)shDd+byb&ajYzs#Kh`JQb>#po9eve~GDDDJ7U`HPBJSX3-
zGg$%0m&q)@EEeC(C@(ho)6j%SfL?%j4!v>jvk`Vrq})(;|L1pR84>>#zV-@iE_}ac
z;JtHwpSq}+RB|J;8jZ|}H~4vdMha8M_i?YK-e}qShFD}k?B56Qs`l-u)|0{n<OqLw
zWz5U8YhJx)?-YM*+Y(z<+r^tB!HFugg{ZVy2mmr;83h|*FkYB_V!|)B(EQR@?7BO<
zo0Z)EV6|NQL3#$;NLyrOa_^njJo5<x0{CuSQcS5-r5OOrHH<L>=&Z>eUb$wdpwg&{
zGgzo1OYs8&$s&w<+tmp#p$lE$%a4c!5zaid?eLpou*dXQhN=y0+oHp9b@SC_oFAd|
zG!j*1b#T1UZH3YvD9!`9l=ES6yQpmR7{=M*({#yisBQ0b>xS~rzqay=41!H?Oc4Yn
zXeB@Yo=@uF%ozvTllv5rf=VzNCVQQeo_Z{Bq5WRW&bM~*+5;TCPHvysxz+|5^d4o2
zFT<t&o0BnbdE3u=?57cgoRYG|2@Cqt`A;I4F;g7t<7Xn#2_JZ$3L=B*z4y*-BK|!(
zrL9bWrT{dN9m4HX^LuUXICH6m?om*goyl5%Qm^TL^Ax;^=-vWGpX{S4JAuxSP4oKT
z`Rn@mZN*U}0vsbZRnIEl<E!7JMh5`Qfk^iP0iX^e<h-|=vWj1e>dTTw$Rw}-qz)yY
zBP1VAo8Li7trz6cJ)QUKax<>0F5pgiv+s*mGaEGC*IbvoG(PMUTq5P+^#(TTX9z@^
zheSx!zXLMLB5@8mE2=<-`Ie=N2zgJ7QZ2Ws1QiO!m$68BPmH-Y3UP9wbKH9EfZ>+;
zy-&bxeFUIE!j3;5$gy-LhY7uGRp>fT1WK7NxO|{}yswh-PR@rO5Awh>z#ClyQ<V<0
z8cHpc{#$&%?0O&3=zdabOD88Jx!HujemQ^C6;;EEnR11!Pc>(}Gsk&bKXeT3(R-WI
z`_Pj5(ObYY+YUrxCdHLv5MAxd%nh?J=Pb?UAtqBV|Cv>5rH|qhxg8y0kpm^692-cU
ze1A4+o6A5y>k&hvLc)-g9EtX@oWm67{lJB#f!=#*_cp<DvApmmCHOefz4JkG`<U^i
zS+?74$2B{#d}sQ=V>kRj+v$BgusfX4HT%2g+d%KNuF7p_JTebipp)4J%ErY+$HM^2
zJ`mftfRBP26adfz-+Iz8R(`zQk)U#c(msk1{_WM;Fx~UgoqSrqzx3-t)g$jzuZBmD
zD(2UWfZsnGmjVnuiFeQRJs25=Ujh?Ukd-1ls#khicwQfJHecnl1HW0Hr@liNF_s;8
zFA2L)?^xR2*o?!-^>AP@CJ9Ee11<Uf;&YEaCFI`xg|g-f(MyNYl!JHYGKAwVUdt-l
zI3alpSg~!70jMCUcCYhxdm?cj<36;OFc_2TTswcu1Y*#<H=#e$(sqY+pEQb_t@(AA
zJ6OT*xmnZ7<0dI@)$c8=`zi|JWo)4f2bE~Mrq(Cs1BkimFhulnRQpDM{xlV)LhQ2G
zb;&$*j`%Wv{`Tqq$E`Vp@6GSpoG`X~pIzcye~9!|nbg}3BVYGoZfS7-iahju9ok7y
zdNdL0*j7<T@em401${;t#l0}ACX$WzGz8fD?D}EP@9|UawNfq^t>ftHOPQXh0NDF#
z=b{HW>OSGXcgXEBGtU<V-1XA__(t-$;jvTuVwUUKn}kcS`CsL4@N=x+Xk6$1Z|^a*
zj_`x`MTQ5v?iVNi=dA7kXkX~vpKD`nts~dD&p&jpGZ!wH()^I#SLBj>4rA=RDK?x(
zs9tSn_|A8Ugx;?^PE1&K03_8vtcm|W3-C|wp%%d6=`)e|fW00@@?K;<cvZQ#Kg}Ha
z-rK^29^Cx;KJpE~@#_V@{9!rK4K?tguWg>KH!9BT-TJhQfyUzmLuf3EQ{5m^{Bz6n
z*IycpQMTeT#g|)Xw2!^n{{M`|NrkMW*X`)Koj*m~lA)kS$H?eM;837eSmKvz{tYBG
zM4qsg%u$`6P|cNj_f1C~)oMq#X!x(D<bIx~jnLz2kHs$o7<ioYFq5J57wkS2DZU%*
za8QKnKntGHZfTYs^=>)NFCELul6ymBojXdmNer<~3X(?zy_fAT&%;5OWbDy641dH7
zG|$NAC&Z@AaINrSjtJS9opqwB`Rtc?W+g@CtsJ6{8__w6XI(IRRjEMaoG>8qAA)Jg
zp3Jy1Tbnv^L{{d=Dhn`Xi|O7(Tm%ktY8sanO1_mW*J8vs=*(^Z$f<W^?22)M8P;84
zxz?Oj?HXZPw%Q1|-(RjDYtEmT|HmyoNYcm(Yu3ATwQV~rg++VDrQq7IXXpCi2Cd}1
zjn(?EFe*`*XS`NJW{ykV;C8;bBL&EV4kGDohiwF{ckFrao2_6U=7Zep|Dr*wt}8Y#
zBP4GlqCVI%7}$N-QFKJu9lKO4GUt8u1g}(qSnxQa1m!ledmejvx+Av>dP)l}{4WJ)
z@S8vYl0IxR<Rzg&#X=F;JZtZs1j4j7<P)|nm70H&b4*S5ohQKL^wzyfOrz{;uL#!>
zgl8WN?Vz<XFPEgqq-Kgm|Iwe7`f4CfWLY@@N{TST;q(57yM*BjJE&vw+nEQu>W+t)
z6`!B;VD%!nrZ<M#x+!3!Je1XFJbk7L|Bo*{56o2AS&&Ejx8>2Tz$+?~<2oPsMWM=y
z81!}=D`dOJ9IjgvMwdqZq6NRVbMEtwYu3SwN3dVld-b$=FrtM`epZ(~QS+L!QRTrP
z`Z6lRACit+rk4w}cUBM|4*zvxnbR1`0+!(}`0Yl?eYO>?k7sDb_kLniAgPl4ftjee
zn1qP`?;z#*^<SYidmBqWVu0zn&k1F!&<AwR9r*|`OpR-MtD&!%?ho5vZ1?AnZOIpt
z&-9H6>q8nm?P!RND4KPjEr4~}4QANJyYt#8BPNuj3PE37T-*en=US-;|KOo6acnTD
zPsQz#)PjtbFw9X24Ia+gcv83RI6c1vk@ixMIi6Pg^_NWw<=MU8*&SfSxr^sPag)AG
zr}zz;2|X{pl|`E0E)m%7*2noUeyF3RW;;Dup2I}@>-pyHppVHxHn4YK*A)U@b5~_S
z|4h~p9wSDECAtyT@yKMiwV$Q>&E*`zX@+Ul>zs`EDn85}sDmh`cghZ*+<ebZ*BkV;
zo-&&|Pi=hksK~tms@CJ`oW^xS+QbaC9?+r~;s;=TIO_~y<voZ4Es#YGqd#C*yKUfC
zA~dhvq6eCdcSF<q0U0eO2g3YO<@+2E3}I;eDMHY&O$Of|;7n&qaz*<%iHg?#Bv=Q9
zoM{_*7vvI7G)O&teEz<rPW)C5Ps7zv(r9a<3*m&}C+#o_^>C~A?$Wh$bzl!()7x@e
zzww>qJLf09U-avD6MA?2_-2xOCpyFkIj(ch+fW_@k&3R>ng6kn=AyAox20|)t=pMl
z<9++f#iUyod)eu_)}XWC_0IBVW6&a2qPjll^9TYxB(8zH?MmRtt6G}M_hfqpQkIvj
zc`TT}l+DJ>w1wit_Xy!I2o_|I%d_eK%rE{c8z&`se;9haBzeDdE-ty*)X?>yI``Um
zwx7IAOzQ|X<KMx1r3*%P|9FPc@9l3|SF#*)tMQ5z|4*bnpNy=IW1aJDmbYydz;Fr@
zdI<XA2!!Rg?w7Id@gbqNOW$lbI;_EuuFV@S6B`g0GeWO39dEa_2RvYp$v;{QAjuDM
z#1%pJB|)~=S?dq`Uf2&)5MxWqcUKDMuMBi2)~4>!ChyK+VxLB^xg))6-qC;hNqvN-
zMvXFK)SPEw>3=n09lw4P>i$Y8WeR}=)ig53x+^_^J*x|y-~&f}Dyg+&L2BJ~s9?8!
zR8_nGQF>JAz0#Ax*ml3V5(9twbw0d6xRl^k!p9qUxwX;T;Uo5QcmlLU`^kSgnsp-m
z2&lxaTSBG2yE~6|?+13QZ%Ox;QQZ$whhB4rR8h=V>1O<C8Xeo8|0g8pYJFYsg?t5;
zu9#nIzBo+s9M}Wa%*9lgU~*p=n$u)b7wlb5=ad~|v)%`6*L|No4{`jt+V9!#6{ptc
z)M^GTj%3x(%*iPUcsh2H_gJKTJw{@}o9@+W<0a%huk#-DS0(StyHI;JFZ&I3=Q#2J
zTwy+ka(xuSC+YgVuj+;z)R9d2;ph?58n19xv!tZ>%0?*Uw1K2e;mL<FsDDdy8^+G#
zeCBzN3BwDo-mo3dj<*91fcp0Q_k!=d=ctd+V#FV2@PR$(E9*(9-Xz0*)D}7^9_A<i
zvy|7z8z271euiDQvp*peQUL~z8#&#3%VR?-b^A!(s1JbW`)Wcx<Z1)n`+_pmWMc|+
z0_T#)1$V!{+Bxf(?eCyLO?7!OFcUlBx7A#CC)sU1$*~+9*LDwHG;^2B<y;;NnJ?|Q
zPZIlmj@_{zd;JB0>$8U9{d&7_)#~@l<Vz=*J1#e9GJoI-y!a&~O*eBanNHOma4`>8
zn(an)_kjT}wU{a0924vhdfq`N%hG54Bof9Q|Jg!}6Oo0+JYc^tNnZ%x{}7lT5ERpy
zR6%`c8O7>w@WE|85qT0|OdX{<9c1?2%`54ORPqtZMo-{a1pPLa2p5KCWo<1@e7vtJ
z*B?619{6cK85q?Kfb)yzJ;xC<nZq~p+e$-pi#621eI&JCl8iWROlaCXIWrT#aGG|y
zoW>_PD~q9Os5(gh`v{DvC^71S$x4$=!xw2=TjqlQ5Efhz!}?UTSO`&n0YGF351mqr
z#FUf)!0DtEqa}{p;l^ogeZ1j2G=9t>M?@Ie#_z>0udb%zThZ1(IyxHuWaeYHN)KFz
z2E*zt2quGJOynIOOJgycDOI2gIK4eyaN3Tb88wcyS(%-mSNZXSkxPQ!s=T;3`0rm8
z+xAmVx`?u4-jb3Mx@#tMgyBc^A3x+}WDvLP({0z<s4zWcn9P5F-)nO@HjY;vrvR~x
zj*ddmGBfX|X8Fn`Q7age3eOd}>ag&X-|eeBTfN5nS_#3BFI7-~>U!KK=rL;LiKX!B
z7Ad4ju&6T}_~g-D34I9bgvA)H84bMt2fcbYf#UMfyWeo#RD)gKSw{|!y%s(`uT7IE
zz+XwQ3>sF4E$V$MOqf)-Khg-7dehw}+rKvmUUj9CW+mzkh4S|A&L#Kv@84%<XX`9x
zs2Rr!eV*?c&8J1X{Qi(d4ej(Bf|y%2p&T6@ce5i044seX6}rNp3>CkpN+>I1#Gg1(
zd74CsS1+IcNBJV*sHJcMToIFH*OdV<TM-my(gnOYJ3Vg`T|142Vxvif1SRqWb*_DT
z17WdoaJIQ)n+<4M_Yq1Let`uU7WfPppaLN3HRX>sqIO)6Yt7P4y?jWGa0J)#XwiF(
z;M$qjuCF`0Li|5R3~wV(+cDxI|DBrhFSRf>Gb@8B@I?$CF#P#`GzKHLlw>*QLWJ@~
z>N_;GQf~I2Go5Bzok*u@9Jl1;WT0amn&XO1vj<{HzbC~!3W%g1Hzg%SSw%(E)s+)?
z7Fns@&3_e*J64PoT_3>JIqF9G+7|YhzCPW;KtTfUFbEzEPWsU1LI*wM%XXV_9h7E+
zG2-XXpBI}Zrlm-jY@C5RW7lyRicB2og}tM!r8S!Q)ymlD06=(8HJViQ=h9s=C(!zG
zC~(8bUm~+T*o7XG%RNgtjvZWkh}aT1cPp+&?x=dBFWfds4h#%5TFjVs>r%!qgOyM9
z$Vul*Dl6%Ay-vTbdtWQiB~_1ZKv}P~r4+^T9{C@rcWa;Qi1&un9Dmc$0C?KJ^y+TY
zjbaVXR$J+OH2Qfnh@%Cj-Y%s5HHHsIjYDD;Iq*$r_&ly7Xxwk6rSOD&JNR{VqEZ>Z
z{{fK5noZVtsU{%%P2pU>HxJLdlL(s?M(w8P@4rmzjYlvcLQ#oEfsO)Val`^D^Xv|~
z9txumEV>=(-M~gk2k>D1zqZMWvfNQ8uz(bB34}Y~Xgu83`YDtt=Gg0`<?y*R08`po
z=b#@;2_|1IjhRlOY@3fm!3cP6o&V*cFWd8|A{ziW0KW!6;c^3y((k%domfb-Ef0VX
zd!LcnxU(I$LcaX1*5=}MJ&^|X!lqqv9CijlZfC2><icA72tpF1I8pVkr&_=h=29jm
zC97UII5>`nC24GqyTi%1Jq?U+F9$+S@2~e`ZZa`q<*ttrG`XAmS$3(Mf`q8h7bm~K
zi9qm)^SNJhQ{Co(vJ^$iFDwiM`0q2oi&j=vE&@<^KDRSU+?yL@mi802!*lT><u=#T
zQQ+?Fp2xZW%YDf&VD`D>#)afi*vT-<-OAvzOFU@<x>2a8sSS)O32r4P!RK25J`#()
zU*KM`1%6q!<0lCT32z{++?;RZ@i^^6u$qiw_he~9!=n~AHYV40zh$m>dimA*xag#P
z(RB|101_IL1;DR_(zQOh1`Czxb^Er|$mI$8e4vZ-{N8v9{tF)+m9_19H2oOJr@K>z
zTGh>7coO5mXuL2>arWU&DT<`oQGKD7ny9L}Z`%%wtxhz0-fgMCjMdNNt=Y*p&|d<1
z3-(#5*OYK_-9Nu}0OtI^w%7~=$w=eye@VbjDiHC=#6r<g!w^G6N4kVIX1?@sD0Dq+
zBhA;|7jtL#OSY+&DgJ9yBt?(!rE^gws~i*e|Hjy5mg9oct3Q?GBFAxXLxqU|5B;5|
z?09x=ZshzIV-yC3<nU&xRt6pCI~xB5M9ndq%h5L=EfmBLdCS7d*>FtsOrbE2A(49_
zmv|BmB?&j?(7x$I=*)~vP68ASIRF1d6OW%a@1HwMXIlO>g_`iRh@9geOy<ZvF4JfH
zhY_19eN*1M#kKA(7Hd<h+lc_fe^_^PWGO2rS8tQGpy!uUZ!`pT0YIEbQ8dB10(+W~
z=X7**K#uz5d24xfbyfF+rUbDPft4?oVh`!7LXmc}EuEq#Fx6hx0Gn_EP;VmKg!{m*
zH&YZ26FGj&sBl4*6&v?a8}sAi<A-D$ZF%2-`7|*2&6-ynQ**Tzp)s<DIpSW5aHG@9
zEtAWRIWG>Yd9m4!1eg;*d|Cc&)e019yVr2$b%y^k?>ms$m5Va<1;T@iH&FEB%^|FU
zOwDmiBe34jx}5M{*vJw)>(fK@;N`K#LIBi|)63<l`CmQ~X_Q8u*ggPmP0PYEbZ|Wl
zWMDXCT+)MbX<Nv@U7x5crSXyA5%jnIgaBc>&;lO-s_o4J#PlR3!2~A(LR5CkS=#-{
zoUu?80#@Xpv$b}1*Aq=|L6X|8C;9~GV1UaASv-t5xZ{A&WjhXtWGOnL7TYym`I+R-
zMwpos{i#Yd1|D1T0_HQZ1sC8p;&Ry%BH=LV10goOR0g=wcwDyGKMv&Aq&RSldjF)+
z(XRS$unwPQ%K5s1_fM-=hWrCwFBATKEkAyc$La&siQxE$i$czDR8&<ZH3@Ui2ywmm
z3IVZ`gM%X<{*Uv-X((14#?51dmcz)V576QSm)#V()Pl{eW{6|XC1-W@2Lv9%`Y$uI
z<9QFZKee5A_&~ZE;7|ZC+$;yOK=9gKrl&--S=s;J4N&~qPA;84dCJoLqB6M#P<1(#
zrMHK@B8GOpsjL6(_VW`hKu1AAf&aR0nXhDGMi5FC<td&Z2h57bEvxyTknvt${G}<$
zYxOAmoIz5UGJ2PI0v?b;uK}j<y?#7@oxav2twKk=DTq!k9kY~-1TO$1cQ}1Li~SCr
zXeHy*xd2i&(L(fM_U(GTTX|gBc$2iTVfvt%U++(RQ&Usc)l~#C5TNU57#Kcq>;r(o
zMwL6;9pYE=ME3jl?+5R$WFLNEU`Cox7Y_K2Fn{H#WkW*k7vx?oNRTF`wjCK@c(7D-
zs%bytZL-@?q{Nh=gZOO3=d>?x+i@w`x`8j?$pP5)dX(+YN`6mR!15{bo(ClQK|vVt
z_6SWLKWQJGfX6>>R5XDbbU@`?Ve@dr0JTprAPT~lewi8J!cj;@>;DNxymGqV@Dl=R
z$}~5Yn)Z#2Lu)VPbh1cforOK1bINwAw(S-`1dkU{{2pDo+TF5#BkTc(M`(+F-duCc
zVKpBONEtaVnspN8#h=SU%IPJQTrg0%M={_7qme`^`E*uD{||34J3AYj&+<FY+}zwp
zsx*q_3H=D1RW)t0%Q{<b5k7{jxVs5b#E^-W@SF8x`1ttcKYto8)aW3S2xg-K9qv1x
z4ZG>P^kz5!9R%R9hVXcuX#qMAfD20IJMT!Euhxdv^*;0i8uMhZnT#5YjPheI#${(;
zL{wE(6|`vLpyeN@hiW*t0zpHO$AomH$w$t@f^LY=Nq!q097_zqwLap#^U-VsAUL6?
zy8Ygh4bMJs&2fTH;etMQ=XR;Fz#3H6)EuTO%T83`Mwlv+LQ}8#A#=XodHMVbq(}f9
z#VF(jOzrLKM__i#e-hI?p#pTr;QI^s{pH`~7-^7Lh@K=iWsn}>Y)ti8m_!tgvTg^}
z7A<5=lu^*(i~}Lk&>RF&9*e<0XA%NV>v*JEI6B0f8;SOUf&jn;=lGp@>BZSY;`aiI
z`ofYr<aWMJbU{yn(wARePFrwa<8xrD<$Z9F=Qmj_lVrYB2%A}ofecrVM*gp2J&gXG
zG}>2P9Vh;0jq17bw|@xpy+?CZRM#>KE*bVtI`(V3l6_4!Kiz|cHHR85-kN;=Wom#d
zhnPNe=hyj#`7JaPp93EtxkB~kF9eA#uqp&40!;JUISsl>TJn>z1fQ~gpHHsA`WXW|
z4)pZ&5W`@w*=)Ho9YM{<NCS$A(|R%f&tFHvFk%5uR?~4@4*^4YUC%=*Sy@?HMn+=*
z<lSiNm2AuaNIYj98qA0~lQA6N|1KWAKZa$Dxn17&;m3jrpj;u6Q&Uy6v}AR3h=9Zj
zz=oZG!vkg3m<*iiAiH(`?MrSzARGXDKXL}2iRlrfSk@B01;K4||3;evA}%h!J8R7j
zx?&xM&O!+gWsq@Mh2bPPhMre}jVD<V5lG+eyuGosk3$K={n;DmG%Bh?Lhp}Ch4paq
z-6RXQmCjy78)JM*Kl-S8InWoXt<2l*ptbGHF68`cj%R1*5>ir7)Ct=QAek7#mY20e
zWAKgolOS0zkd>pLtyfYVi?(zAOA?oXtI(KZTnw}<Y;ro7#QyRHD@Ky%`Q(?yN)r<i
zud_ZN41hUh)4UdHv-06f@L%*mn`csEZC$!&Z}o-!PcrGMA41|?u5SSPZaf`CAyL@a
z(UA?*kRxuJ!gG<8*@hhWu<?wH4kYB7RIxDwhIA2rU#GXZpc6v?O0a_wF|^|1VR_Pn
z^=eY5%f<Y73(-9`CT0K&$d|l;YSG|bBjS!E@-u__e?UdWid$^d{A2J3FAnsVJX>@(
zYv8U+0c92t;T@4B2F!JJbG-)xgg>I{!t|d%qLEQikapG*O22`~E*VG61Z@^-zzU4b
zCnO{~28JR)!vVy#(nUTDb1X#ider-qr6lDd$r9F+U~zpQbbiEAp!rcl`1W%?uiv4_
z*4yi|s<N_(g++zDO2ji}-scf6e_$g;aWW5hF(B`yniwppC$^qYYbJ%gRD6VSu0)w^
z=z~b6l|LNv$ZI_gOoa*D1PomDmUHx(U%{ny`N=T^Tn#SA>YTQ#(u_DoO-<@Jj<2V7
zenrX@wNgoyybkA`_l^rXZqr>pUaqIh0{~fkWMrfOX#9*!#GBmF!LQY1O&9mYuk4xM
zl1<Kp>bPyPdBax@2<p2&k+;8UYGO+jvPR#rIMrX7Z_Vz7p`FtE1{5$T>`TnORAl9E
zU_aXvh5@zS2c%usWAF20JD?!14uszER+^UD-B>=>xNk(M#mB1wg8vdo|07PXG#fy#
zqPvr&K0tKRAVEU{s-XuOjlX9`NeHE<x_O=6uNkN!tyfA-#!~5Il=@hWRa9|ON<pGf
z0m7pn<x@sH?5+ST6l4Wv;nP4yusU2A``;4ixouS!mh@awbrT`7LxMHOF(&g&Ne2gz
zSST`N2YDdSVk?wNE_GO7$1Klh5ZHAP0mS~{Yz3Cha-)pFUgUK}hz&6MM0~DO4JLAr
z&#6q>A8B~G(LyYX)5iU3gvMg6&E?aIT&L}CTCbab(=PzjO?1H}*Yk)jV;HX1+-R;+
zjn46=I~l4SK$Fof!S?|E0R&e9M0CKZzs27|2_zs7qSlN7fxc56%VuR1eAuS=;SjK}
zq**FLOk{pd0uOO00MTl(!DLKiX&#W<@-sjjq;^z{Y@KejP;IhaBKKKvMjTbEQlBnU
zB6;Ki4b5%C_99P`2IXPBOQg~SkwDM6M{GtP^QMQAWhj?gU^Zt56B>*owY9b(4{N4U
zRYkyx`^zxqTDlCQwQ|-Yg2%>^5J`Ixt{=S@Ba21b`byTbj@zzs`CYYsz@eet@%D={
zb5dBME6i}<t$yxiPXS@vvta5U8tulE49#AhN%hSk6Vk+j;j7erCL5p2HaBe=YudJD
zfdzx&{_Wmg9cTU?*@hLD5;JO>-8s8_2-#}75El5?0-08$(=L0L*w!0l+@0vQ?ja@E
zr(UfklR>jZ>vF!H!=&A@`!7w;mo9^_)MAvF!El^Z%}8s8L-6gv2+%TNpJOnt;KL<4
zANQ}}HCw%5;qYA<WA|HHQe%|9JB#^(>1K57zs3N{AP<Nw8iIT&u=PNt$a4pjjH!@A
zK{;6PLFbh}u+Kx5-v<t^F9Bon>aZ-&aD`L5$$INA;*M(ZOhZG13Ma87x&JaiY5(-;
zQ@A*Juqc_gVBXIcKZo_GZz-u>U==5Oh`tNeZ{PYCj=N3A)8#s>92j3RLrcAOEOu^T
zoY*a@`%r;^XU`oiDXPMPd-HT{iC>Q!j5VXeZ7pi6fN}>PBm!t!wb9DrENS1UdkM;b
z*~c0<89@Kt-rm9liHuZc2>S54tvbL52n#n*%)vYX5v}+Y1Ccf1zc0vJ?PqOB3NtOI
zO)C$8<NFa_hWNi}>TJL(%oGb?vK3oGfSYUp$0(i0i8?`AoLKL_|9Wl?XZCo0tQ<q*
zxfr%^<Hn}x*Osp$xC{JV_qbOmUG?CKHQBPnC+jAmz@ggjX(C#A0Blt%oBau{jd5VV
zT8W?deSoqN0w`XMzvew?bMc#Iov{PU<14UI$bqy61SsGxWvlu;-N-nUN)+@1mWBc#
zRsc&bFc1n){+CTwED_&maRt2><$MmX%pI0Y(nfwuEIJyT3jyPsVIC<?&bCFh-<aOs
zWCM6|O(fX)Ge9ZGS7}_j!wuXNYQ^kfAbvkS?8E^@7GPsNqCUR*wmBdEX>r)`Hx68K
z(b*r*5K~i2kvEr>0gcJ6dY-f4m8t{6kNjUmE>gWajGw3$0>l?xztgX8^%rWKcFT$!
z#C@zB_<I~hE@PY@TB^}{E`5wzSK|Hc)ncKB2x!Ed*yV32*Dt6@*_XzoJ-AqJWYVh(
z2xbq!_duf+`fWOaul`fB7uXPEiJCygzJR$yY6M&pK)q%TJ4NKD(r!wpG0Ue_V!h+s
zol1*1#+{9tYV)q+YHB(X&aDq`p1J`^2PivJBV_%7F!9UjK(Vr_s-m)z933G(DT!j}
z7f=<QEdP#rdh!%4nDT{nKI}RH7K4dFt}t-x7dyS0las%_9(|Z6tDeWz&$n$2jJ2$P
z90=FlcTj-SJVGC5POp#k-g!+>5t9MGN=Jt&Xyy6<T-MN#2qdFepl)3UY}KA)Xqr0@
zJY#!GVJ304rD%9V?m#oEW<bXSFJsIo^&snChEV(<gGHmbn<&Rz00ei4XhNPunmljJ
zi<dY+Gh}kwWd3d10SefX;$rc;;}xJV`mjWIhZC;aq+$tWhGL2KJb*=2Y$?Zktp1}J
zzj+zB1TaF1fh@{@HB6za0#^$wVPHT8h%Yi=1?38Ou>sY?A^R)WrjQ3TusHN_VW9l0
zMMUQ$?=#lVIbu4!?xBW7A-1>>=PH4(M_aK!h3sC}Ge-UV`7`M56|idSej3-e<NINr
z3i>B6mT7K$xVIxCJfw6YDI*iu%*Gc^JHtypruNbW)M<q9z~%vAmw8*NUH2=F!ZpXB
z%S6BHb}sZn3*GbG9D0-lEdxuh3&+*<H2`}r>WFMuYcC0+Bqhwx&L#{+#uH`4NlN60
zrx4=S?|}e9iTV6bJirO;%VX{TIK%?zgNrA9N~l`V8kyBB{WZ<h0bor``HKX>K^DVQ
zqn2L%2eUidAFfqTPWkj_%U;%{lir6GhfOVvTcG72VS&SXrMHRcMTCUYfdi#fDG%7b
z>e%7Y2^Hx%IQT2al5%jhOWB=efE^)@029RRN~kv#hvgPAE5ZG^n>4K@j_W$Q1dvCO
z8UhIuXFcolBVwXUEmWTMth)^B)01s66N4-JcJH6}*f9fafI{jI(<Id-4V2%uS;BDi
z(2nEfm%N}^aXfb!{;zm+Q00JSQBqh4r7$x*HfGf04@t+^PNEvl^8aT6#D2S%X-G!+
zMG2*vlziiV118`$knK}O0$*NUu&ML)|CWC?1g#Jx6&8|bd_#xSDhMDP68Y6R5tS=_
z5?Rm4%xntWxf`(1_od>f2^Qn4?_o4RYGlTz!oK00Y_r&GA@7B8)nO|@Ery})r&ec8
z5@)Hn5hU5p`L;nY{R@@9mC67br7zwGzDz#j58#DfJih`0-0(Y2y(aR43mq+Oz8N?d
zu&D(fHlYDC5jbxZFM~p+LvwNrT%6A9LJ!pWwyWG{o%g!PeVhM^;`RZvZ5526cKR9!
zi$qILFPrcM<KB3MQ!r2Ij-LZL2w0$p$mOy0l*I?~LerYa?B;>{hTQpwZ6_5a1Z_jU
zoC@1bF^AU}ah^UYFTAUq9u3Gih>yGpvzU;7^`K_gGc5<oEHW2s<N$#nAgNDFxgctw
z2OA|TysAx$ooE%TDJLPs*gN|bS&W)NtRVhJP{>XRb;q{~%1ePlEh1?|+$b;1<2cVa
zc<7Y1{Dvm-1O&mnI8K{oR={Hc;wH^$SIsdpG5?3%R_v|*<G>_MXRyQL8nANG4(e?P
z{*SJ&46AaBx}`x$K^p0l?vM}xK>-m2>FzG+RFE!dNdZ9#N$Hf722nx~a8nXWcYSl8
z@80|Go*(BNeK?-I-*>IK#vEhJv5Xl}xXb5Maiy&jWU*vu;+U4yPXYofiDj?k^OxYO
zzt_r~TwXbPuP~U}Fx?RMWzUjxx^+!ouhywE;uaILpm6pNU>d`2Msn^Ptt|webHf6$
zgs-BhsreKrazNcHD=S;G_3;3a+`t0=9;1-He~)w_Y2Mi$m>Yq`8}kwrk@C}@ZApNF
zSnDA-eqT}&g+Q3Koo#R;5f?`>bGce&ay;{?tg7-eMi;4~)bh>rbT^rIaTO<XfGljF
zaf`Yn)!Gvl7T%h!3IQNn6q7PNxc#dcuFerB>h^5}kjW9tbz})3`k^w0vcdz^DGxx&
z3O+vKig!G|2P#B1l@I#q)Y_2(ubi?kOH<=8BY^w?Xxe~+`JH=x!S>sQVz#bvf8^dN
zh+h`l>f`qIJyE1tkXp*|%@OLS*n<G=j+(xEo?9kx)ymed<}^5~tVXi>flY?Go?le-
zkd+wOVapLp$pd+)fp`i80|+U6L0A4AHF?kqOS7=BAhpryI#pNKD*yWH%k7~3-{Ew7
z=L^@;ulXk$T+O4RqP)-6KOm1GA5iNk`OPpgimH6}tx#qN30szxplCyqEF&+kbGx6G
zqqm2LFf<0VaBvmCvIlxgX&n-=#fFeN@kjLnfiji*7z#8k6eM-3%!pTnYV{PO&qeLN
z;~1npjHU7!;1CwB#;#Rj7IyrpNhTF6y4Lx-^79m8EkqWY^Nb>OV`H`F&r=U(er&4a
zs%x=2!||s<VV=O{J<R-B;Dm+YKbF}0<wuJyZNp=m3hOM_hlt~H<i-G=82=4LlhfV@
znc*F@wMdhNO4BYtT_+@{9&hk&{RpzFjXY^h4m``juPm}X`1dOl3I?)`qkr=m2P;I#
z?)z8e4E6;!G>5xVl2oz81h4;0d{BEFGJ5d1d%Kz7mb_Ns(f7NL`&=U)7C6ke=whv?
zOe_QGfsvRQ0hroUmkfwUI0B0sWnTOtdXUFOiGqiL_I>hiC!O2)3*Ly0<MS$#%OcHB
zR{-bk9Q%IuCx7*Yo}V*rbSzS|n-7m}I8)gA&sd(A>Sly!_wj_vb53c6F$u#&Jf(FO
zMznaw-)tkl_fIWyGw##Vj~23K%bPvCW38TYSDJVAjW0f90*6UPkxF^-OV0@*mP0yZ
zp-F9jR@h5`saY2ud{#642c_lwU$g(!>6)7p0q{yBPsCm}n15g<XuCSIH<)4}i}sl+
zUB696reS2nGy1E-ScdXFW`)diU>|M1y}!XD;(?Ffy_D0PQXpPt+@2HYbIDIMI)mGh
zbgNOq0lt2o3C^G=pAbG*bn%4sj-^M!otf;fu68dioXbr3l;HKkK2GU3`|?#iGbGwe
ztuowTY6YBqk4#%5maAR91FH^<R^keryGUCraxMCRE2JNq4AitVBi~4{B*cSyrlda)
zA1x~s?}G^){7%sFyx+FM1Pr3Vvp*$>U^Z-e@ipjGrSQ0ora-i&Z4c~cq~fyo)4rk#
zC}57460Qh&dWMNbn|HPs0ar9p%2Acv;k1oX;%jsJsMA`lnInc0-ctLWrii+1i&62f
z14_TeM1+LCKESE6x*~qZx}R8PiJSY6W$5)Ma{FK<%}6x#UCGc`a!-w~nE>(g$8Dl6
zqN&)4_Ej#p6BEtmi!VQi%CSx|b1d~;x%J<j-7z`FLimQZl3tVDP<Ooqq85&m0rBkI
z+)<z|xhifYZf1FtV-eD-Km!^$Vcsqt7xG)sWeKF5r3Ra@sVs>Q(Cl`{4&QHOVNFLO
z(1j`yCRq3t+L&$fIfKe!Eqeo;F%<l-%76Xbzt|r|A~wD1*a_WhJIuODKg2aswsUG+
zA(Qa>rxF+_4LwZ})Tk`apSGW!7jnqB^;}?sC#I#P5nDV~`wSYEYWZxPs$sUaWR+<b
z)<b|_0Rd8Qq%z&U`N)`=mp2|z$1XTjUqX5NoA!>#^Fm44K&(7uRU)w%JBQm-sr=cx
z0zrP|gA>8UPp{#%knMcL<NdL4+mE@HO8Smn6#&UQ3CXXBG4DU9m1Um4I;wByfAMGR
z_GET2yF|<BYNj|yZ+`$VKYo*Fsf$M`5c-Bm@i85EKcU*DT2?`21zrUzBATRU$D8Ek
zp~ZG`WXrS2;!p$7X}(1q+Vni1{ngJ-zcR|oxWK4`2js;gsB9WHQ?-tgrm+E<#sZt3
z=-hT122wC}Fp{Euk}a`2=RS%gC}{n6WZ@hHs>*V@(o`i!@+}M#a=hkdCB6l1rnr`t
z7EpuGqVvpTB7coSD09IVOC|g`5*UlRPiN<VDBVEAn*g5_O`j<c0C0ZlK!6aafskKa
zSoz|*G?6AFY$dpuEvUD83_o5Nx6*ddD$Z^J4UdFe4Xf@PL(NJ`R8+hA6ci0S`slxK
zDu4l@oMsTHMc=~>mgw4g$JvL0fmgDzUiD7D#)M*l0`NWf^TsSm)M<`AKugK;-%Q=`
zJJ}E<9Z_6f&U){j{u#&mw-ic1u1NK5@gAsX<x7;hxeR2s6>H&}-kSJdN=|pW;#*Xk
z!nQswi#>_jFT%>zEcp1i{SMc&L01~&i8>#6?#NNgTCR(AlKj~J$J4=$weGN>T|`fo
zHzG32wU?436O*7*hwZIq99~N?!Fv<>U%=h$oeQffyE^o&MMp(3-;@1R1l*t6>0f7O
z2DWDYFw@5yRRzLC(Y}$TI$U8+R5zZJRln-0<~lft>O#NI6$5)FG-PJcd&GzRc7AX1
zmt35pV%(1#5z&iY%*1bAK7aG|aoXC`kmL<S$$)=<4nt)3`++m!M4P66hj&OKIewSn
zvZklwZg}5hQ&rlD3@$RIK+^K?!0@{0VL>Amm7Ye;(rZG}Z=$K|K`75FjJVgf6vz$b
z!MNOdF8V^3+ctys_3PKSNl95k>X=0*n#u+JPh2S9WlMOJh)#-~z_hm6{!eP>v9U$*
zB}Scp`=^yvyT38Lxuy<%mC~&qA~yI!Z8N&>Q1|T;FF%o!B2>+NQh%%6s(|+)(-dy<
zM>pG&iTq6o5Z$V;<Here=KQ#o32(Z*^+?ZlNxQ-b|Nf&#ky#VGZ6}P-M<u_^MOJgs
z1^;J1HHA$D;rxeVear*WI2!I(Yhvgt8Z3M-e#W9~Kak9UT)lAa)mV?xN}q2D!gu?m
zv_aeOAyw}dTd%4T=S^#v-5m@}AUAw*3-}CEz#a~B4eCF|raWBlkSkW@hM6MmdL*$v
zN8BHDnS_SFtQ`Y;TV<HWl}pG*#_&-N*`856EKt|2($?4Kq|9!|z-h+&+8sN}SIMnU
zMjOFTw7KvS&Aao-jj(zN>~Beih=tDiXau^}s&<)dV7u|pfGRI_P89_WRYMt+#+*wc
zxWYtQSd2YA)^0noe%_a}Fue$lTqCtnbiSePlgRhBzuN5L?&a}PNTiZJnhqE{ydJw?
zLsrQzW5j+O<k(%Bx=;HTN~PuG<j|oZ6CoVQc*O3o^F1kOX>X9XZ~K_8l1i_PW4(+{
zy;}(0WAN=})P96`6(N>}LxEVh*pn1VY@K?iY&^HaKjR@FGCr;@O5-&iX}>z;EpI!s
zgTFK^{%Hn1jvM!4HX~338M}Oap2;I&0Heq@pL2lx)k`*M^9)v~L#6;xzsUf~2RQ|f
za}dBV!SMl5N{S<J5&HHUSRvqkK?L7;@M9aq0v^MrN)S4d-H|xdMJ#Kt3{b~Pn(-=<
z74ebY5QhIS6WHN>q^D%wXYKEY{Jb-JJCr#-yqtg|UFnM*IWYCIB0nSu5GdXX_swL|
zZ}?leh?xV&!6Ea3^O^t2@9?Y_-yZFGt-Bkw+(QI3tmXK@+POi(`n5aByB^pr>GH-^
zDj-#1;rT=L;4*H@f-<MuHdPYOa`6gzWdgXRR3?y$6_}I+Ui^7VZ@c}Y))G)6vKzlx
zR|`sPSGL}DnyG#ba+b-g2=J#%1@Y2|0nUnKsa7|4_rjlHVtqlg%}zzUUQKbhC$CVc
zjyL=ERtCjwJ9`$JzJC{as;Aiv4u%^U50f71BCQF){XCj90PgzTyI|ny(vbRc=M6TL
zof+%gage8U@tOiJeL*7je7f}OJ<Jsytap`YmHzorN*>G=0gV^pU_x!d`I5|jD)a^k
zR4X>`t*X9;&GKfAW&568DC!&z+pvhl6CT-FR?@YgWc@r5V8~l~uXXw6Ma~N;t#6%5
z%qY?ZP2cr3g<~7?H_JI<ICcetEoa}^E@X}o+t}Ek5P#_nm1CR6v{Sq5YK*XU{blxS
zTa2Le`6n~zbip<>xIj0g^)b_P>M-MGDQ@!s-ZS*Jy8#BDzL8|I+=|%Gnq%HnYV{mG
zuvnz^+v0a%iz@Y55M1wl)Zd+Lx~Ov$M0TZm-@jsE8+*1?)WdZ}tmKRE4@zR{ySF}*
z2n5=x3BM7GdRQP3%>F9UZt-cBHuLB1>5t!oZkdoj$@-(qNTa-)ZD!7K_vm%kb`wt=
z7XR!6LHzvZ21myFEiZ1k_C56a9lGRKAkGt(_gjL83h_2D=+%x116sXZ$8!$~1>xj|
zYtY^SCg)mMek|%Y3G}@`eQx5$WAtKNfhOo;_xurtSY6kinsi=SS<GIPhQkyl@o3+p
zYN|xij7Tp-<xkT?i#sp1%N7^+-%gL?j4OpQSZ6B~TAE@=)rCtw@#$s!Qc!IokGNkO
zQesF-Y6kubC~a`qqgD*)_}7664lq1&@faeZ?I}QBunfKZ%v<O9OhbdQ;sfj)jw7wt
zkVa7D0h{9%L3rcU*`^ObU>5V|a|L>hJX>L=6c(12$YG}<=1BzsgI43(vs2`_3J|(h
zP{>r?3b0Uq{!HQ)(u7X6ys}a#hFzZexhBjscoQLJR6UPH>$808E+;2=-~a+3aTUGp
zSi(sDWn;n|3G=yK$^HXFPOT!kE#6HwhZM2gC9&#oYUbQ<_ME3Al^ZRZoWFZ&NWX>I
z3AZ+hKF$5q<oV&4=~VjT{<8p4%~OImABm3>17jMrU8N94z)-$27Of<1CroF}D7@uL
zuZYb_ER*j$XOZw_6jdwmDRce%>@&jijXt`dEacdbix`fMN|xB04>`V33m2=R*aw8)
zQHt$2CrGGEqz=l5^}QbCJ$+2f(wd;=pJo^TMcFQrJ`I?h4Xs*g(5rm*ERZY~5143@
zYeg=N;*vwjXzm9-btomx<cYGw=-ywSZ@tH@UuOYmkj8JTiX!X_w4-8&-=B>O#Ud9*
zi;w$cxgqEA!RAzj_1HUEj?L`g9TLqW7bp}+%6GwkA0K#L*$E+J5AsCn@O!egvA+Hw
z6qiSXsoc+wxHt3w=;%P}*aL@K7w|sAwNpmXMJ(s|bT%Wl84p3P6to_(=nBK<P$*4E
zZoP2Eb+dL^UcB!~b(Ho^(-HB~-?cVapZnco?Pj9wH6X8j_J>7LD++soFw-FZfo-Ob
z)^dGyj^s0-i-isrx<10iL)Ez|b8Miz+me2}oCNpW*bo<Wo>zf=3!)X!0eeVRftlBH
zdwv)Qb)8mUVYciS-vT!WUzQ(6$UO+tc`)^Upx4n$Wp&2Sc`3DuspEj(dW5Doj!vy~
z^3$%+_3xuuXKN5Sk<+ax7tobkgBbz%BCI}FWC21+=QqdFHgJ6{X@%&Hgy@9lXIP^$
z*OEYSF4xUE%~m{xsR?iv4uQ|B0q{O*`aLh6#i~>JQi{XI8-CJYnG3u=Ul4TsDTD-v
zl+B1~FmjE-|3F29GbHy8rL$-V9i}tG(XDz(@gkccGmxK+1Uln>OBgNMz>}v{W_{C5
z^zi9dvCG@(Mg1Kw(Oqd^2K%l`;bUVZz1N+l9EVILE8T~)ktMbaqgE7H&aH=J1&Z;~
z?_bh0^az7V3LH#@7{*{=9M4w>R{TvC7k3x;Tb>-YSrr9{YK*rB7^xze-ZO9yc3Qt0
zCCrwnH3zfLJHICpq|wTX%5(LHB>RL90v|EIL772U;L6IMVP*~qXT&Le4A-m+C^;=-
zpra+GJh3B0n^LVtGn`6EjP86%{?h6Tf_NnTqnV~?8c)u_dzYVFgJzqPpP+3d!om16
z*@!FUdyw!_`xaM<>}ak?GKYDxkt^pvS-zVnGxkofi1ws0W*-jm&en;qrknqnYplUb
z+$a}Kg-`rulnvrO<NJUCRYTGFe1JZeUXLme!0P$R7Jm0eD7xF0>ae_^keMsCw!7rp
z0VSHQG~I^;qTY3~Otkelr<b7ts@})hK?3VR#-=Nt(Y@aXm*n4Zf<XovFb!SZNK0L$
za}}r^^hdh*yYo?AzR+oK4E?yc@PGR;7K%73?akMJz7<CMYrkHa0{dT`)33WLj^FKm
zBTZ_JH@<c^HqgHLm?30K3{oV(X4?z{j~`lGI+ZA(g1#cq$dtmRcIzS4{(4MDtPvbJ
z&#-YAu}zZOzM5f=;=NJ)VY<#S^VrG$Msrc4f$zFb+{iYnXSL4axu^LTlLT;6s?9q?
z;|C0rM$%;(nk%=Rv3y_D(gZ%2s4gC=QYP<e#DEHM|HIwuyCiy&UJPCAudEs0utW)V
zyWf6IkYHgF8Ty=_s6$AV!<M~Gt+o%L_v3IioUg9Ux$L|Ar}XfRj2rz|DI2~(9DjLC
z7TR4#F%$OEI9z9lb?exA85=x;Y!9ZsRFec)#J-nGP~!V6q}t_q2c+O0WL3~MzM$S=
z-WriQHb722l|vJpT(q;U!o3@`ew|Cpu0=vpzKL{dLVuggh+0S%QXEMvEQ^%w1RSRM
zPANTFNid&<neM<D;<ozMx*CuJ43N{VOcG`-N)W#j*p^?^kUK#sSE0CJ37K|S^>c2x
zYSO370Hb@Wxr_o&Ep?dz;kBRt+WwDD1CtKFY7!<cnG4jL!O?a&cLWloY{uvtKislU
zFS1q15Q+qx_2l{fXs%4IzH_Vaia4}`kK1(&WqvSg!5-wt3kjzSOiy|u5s7p>d{<X2
z9OB>owIO+zi|akWUyfs5iwVOixTL{jd2LOZG!6`@G;fR?3RxW{{3@^5-j|+69J|2k
z4^E>&7ByJJzZ=(bmVoSE-VVCHhrBn6gjFS>iX%lmWu#mWJjC&mdl3hpkPom!`IY3+
zv^feKf3M+7?Ckc@)x22$mTY#Dv`Qxi^lgL~L9^F3tex__wzeD-KK2Fc_JvU}3HZP_
z7vJqBd6pQwZm;WGwxsm3{O>}HtPP3b3tzgUKu_8!Fa(W#(&s4^xxGJ^wbo+XWTasa
z+Rp9UxA{yvF+dXmNho&9xMEX+Nih!NWW>CK(`lj3k-Gkur~c!YFDs|g4BT!2vx&T-
zYk75b?}xTA>inEi&{5}Pw79hWB-azkU3ui&LqgX$nJe#(%{2&NNb9G~L^ivOB(Xz_
zin17ycwO({vu|bf0<}yOE8?={sz?&A2hvtQv#b!Aa<L5w@P#+!EA8OuGOA`tO;$<L
za*E1(en)b*9Uz9G;V@D^BkIrngIy$09C$O7$t;5vUI%*Oqf7`Z!@ob^tx%v2-r%qB
zBELdFK_3u2SvWsg3NN?@S!oDN#X&;=v&;sghmcd^>l7t743gG-46uLgU=o(+EeE~A
zA-Z|WeE-Oa_(}ZTee&3cD}ov`?)gn^d3mV0qimv{B>*WN;-}(r@Kxn$o>$=LvU#`{
zAK8;k@mwv8C(7RV{S{DHU+~)3QdHd9{Eep)TN|g}LxUTpyd_7FR`Ci`__!yRfk-0~
za>TOn?W5{bgP--zqhLf%FWd7-`j!er_Hw!@1@e@ff1Tb$^8u_Y#3LiDWHlVz+$sJ1
zz=+;+^$%XCAekV(oQN4Bb`xVHi_SZ}I0=x>_o$^P56Srzfr^ebR`BTF1%AIYKH_w?
zZWuTKMWAmoipHBf?6L5vk_FnoGbiaVA^G4>U<*k}38`q=y9ovx@BzW<i!ZW8x;MlG
zRFm+jMV>x?&N>wfCKPCM;Y0>I@?c}xW0(7TzxGc3ft{pPok04WU{VEaBUM1wzx_j&
z>1AsCqolKs)-MLeR-}81g01hNy$s$OJcb?N0`L$#Nn+YC)B&9S5wWQiit1FA(7m?~
zRUT+JnrS=|WAe~tnS(>oIeo1n-jDFIq<=Ij__#yn{<VkmA<?FLKtL<rPRXd@1x=*9
z#kHWT^p{T3Q3H>SP`pnjH>1tFn|MYTCT~@-f<H|BSpKOtUs3lZPAUa+_P*Rp9WL-Z
zNLje=!akakA+@Vk<NFB*B`Z~t3`%*3)2vPvJ5}WUx;xiBL(#Cy4NPahw#NPK=2v>o
z&&JGId52m~bX+(a3Gr02B~qTUV=%LfLzil2HJUTL9eCV)nYgsBK#Fij#An1Pso8<`
z7Uqll01X6xZ!<vnn>iBKX-xkC*B25nG!m3N#!EQJ1QP*o>jAOXLV(wRkyJLf3$xZr
zG@8hi$9S^c1!z>B)k*a^-mk!X8G_Voh3&Dt_ZETwKwI=`KfeRSz1hb1Vo)PozCB`d
z1%+X-s{L?Mzwvh%b;EYs`N!hoJb>p&apHcW?K~KOQ2_sGQr_0q9Z6zSTSZQa!zFZi
z;Qa%gNvT~Ui;ae0P*_2dT6N+CXie0U<M!*}EOGZgQx$YzSp%aL097-jzZ}rGZoShT
zq!$>#<(uOo+nmrVOmc$`V2}v`G^wQV7{UlnA+Vk(EoU302ChA%G+=?0N&lTCi`CBj
zF)(&xA4S7s^*kX*WghB|rKROijudt9!u8eh^71mD?KftSVczB&@R@bv<oK*oIyL;p
z>X;QR>1jfozm`Ekx{Hxt1_jAak3r{YXK#;m3zctHgDMZ|c2n*G^JH4Q%vCz<3rKe2
z*86O7ak7h?4FMumwrx<*3Z|1~a*Z!Hv^kD5MleW^dHeR1X3``GC~RexOg19(pDzBJ
zb!PbNObCXoC9iA=KZS88m7~Ub$IN^G8bX?yBsMlTZD*=k?niB7_;2&Wd4jZf8Aa(?
zF!l!SE%zgbiL|v9;dprYT_-(w0EevY5I`rzJx{JoQRkwjrVOO-^!6VM|5&_0Ckw_z
z$_sYymvl05o5I}zP+Kif+(9Kf!IJ@H4wyM3|6PS%wRC!NQ+gKfLvw~iPR59oa?COS
zA!0vD6cpbe*{HL#vt%O{Y<&;+%8#X`OcZznH_3%j3q)3KKj{NT3YrG6@`Pt(WXR|8
zl282@6DCpJzx!`YitPtq;$=<#_{Z;)<i}%UWB1u<de0(V#eTdcep-qJ3Ny$ErqF}f
zWb}dsjWbg*x{8;JE17hSEz1Qhqi7exA<&=d>h{AOiYwj`7O`WOm|~X)6PApM21zde
zp!mr9_%z}<l%RS}c@$S1PqRtq5)OJG$9tC_aJ|Z6tc3dFMR^c<ndG1my~=i<Ilp^V
z=SL;^zIMzZ(9!^nL0tqp!pTk-p;X_GF-t+K!Rn6{6=|d3y;#)i50tgjIEK84q^aLx
z1X6o@EjbUL&~&XWZ3pFU2Gr*y>%gU>_sl|U7pdk?sG1`GPM<FrV71$TnBef@=wE4(
zPl1EWog<NM5~=w6pNdIGA}dWhrJrkS!vup8&wab<Ng-8u+bj1fM)u*8f}aSZ1GKOA
z1va`c+K}+0RJ+D)<+0D5^~g-HGrA0LeE7Qk!cg?J2Wg}{*h~F)zF@-<fSe|&w!}rs
zCZ9B|p9oo#Q~MiE&D7eHLxZ&pI}STRMA_21v3w9ZF*zucuzf6H`M5a}N0N5&RW1=U
znG(4#l`T}pz56Zag~v5ZUEjQ|`n;VEhw+g%EHyEjRFV0${r=rcmUnXD1e&DI38V7U
zgu||`yCoVu(}#a&wSTe=R_oMQ;ltPmMuHo7B=^@pypyGh{=Z&;+{e^lZFP2-U4=NT
z4n91vCW);iKeG@R@Gt}pZdvl){*d5`9^*UK9-w_50(&~pRDp%kvtP#*4TFBlp3`cU
z8FG7XqcC!nB1-IYf^{ao+Xc+lec4^rHGGfCrl;sR9iJ^`Bp8GDj57o;kP#0(l%g?@
zy1|v9pFpqkFW(DiUJ2`e(a`92M=p&U$RhT1K9C!@pd$HxXxA`@Gey;^W*>aIpOW}I
zBnAz;q-076iRsP38V0Pq+G88Ah2v3+sQfhFy#|jnl5O7F+PX|POW>MEu^9gjoLGj&
zJNBd0b*)oHH2EE^387>%Jzyxtqn;ah{aV*3;x+75E_1!!zP^+}P8Gp8OJIhHpM6H}
z)=Uk|Vbc}a79I3{_sE7<{hm)iGc5b;T1|2n191DFCN{aZfd;_m78e)4S>T<NP&d2=
z>{NI=m`m@o(nu0GlfMRJWEf~ol|#3<MNUNChx<Af`6#+5OIlTDg5FqM)8T=D0rQVY
zUa!oc5$W2|AUlH-1xU#ecq>Xzp2W?C17<)L6EPQhKydHDc*LQgXI9bjVqJrS1@}F3
zYxh0fXU~{ObdmBtIJ8pQ)p%3ALGiQ9!|)b}fC(d{XD$5m9e9J1$8+v`G>mP9i2Dvp
zOju^0o(DZY-(d}08yqRJ^o(O8B`|UjV31*tx(i~@duFf%z5aCFW?o%WW62Pk@j@%w
z0tNnkMGxiu$-@o*TSrE*pPI^*n?3*9OqFL}=w2FNQC?(yWv&%CQBTnNCZ?&qU07Cz
zH}x8HLtm_KuVL9dgLww95+M&mXmL}qo+_bV-i}x-z_EyQFYQMiKUSU)Qn6ghqczsn
z4&Q77hndO*Gg)*$>)-JFjBXD@T~{6*WwYPpkz`ei$D32}@J9*;oV@_P^mK3K25ZJk
zOL2;(Xj5;uS8-!UQ(rnt?Fhp=#mbYaiLlU6^WWfv$($r@%P|*sGk$d1>L_4bTU&ef
z{5g{PKvDv#nIb)|H2=ryMHTA?cyDlYb!9~zY|CIJ=55(QTVz@18cdXW@xt%}Y+6vk
zs0KawE~UXQ1}zHkAjDn{KRveYH^<#!=jJZ_oS)S`8o*yzUE5MZ9IPYBddG)9J~}$O
z_~S<ySVP+6kl1i<=3qnVoCt@gM8kId<;`0!v_+W;9>$shr-6c87lI>V=-Ly=!J)V%
zqMNTR_%!+0FT>t4vN4sn*VTzM{vK~>JO=XJ{OrZib8YxO91GNM=qnvFP3Ye&Pk0D{
z9vy47j7=yf&v?uK3m;5~qKLi0(fz&E{g6V1TdPwk1&6esEhX|T?X<?*%m?|JmVZ&I
z)7~igM+T=W86j&Lz(oMEY6=RN(D?!L@UzZw5X`YUE#6hhxmxIMi>>-#p9jV7VOt-p
zGk}?Zh=*Cy3_8!Z+eZgdI9(0X;BHzNXE>Q^zrwr^PCleV-RE!ZluolJH<Ix{0Wi50
zn39od`6ew*5jgFcD)VrQQ;IMKOTyQ41x2=hGU9A0DE^2YeYxlRfQO%xQ@J{VrjG#z
zbus37QDKhLYOM$kj+-Apel!8IpG{n>+VPeZxE;$k2aN1rjX#I?jC_f(^iV)1P@M9B
zuK%%?7U}<b+Tc=B*v%Yw@8*M}{B(cq1M|DkEMtI@M@bLCd`tpA1sGXKwE(E{7k_`I
zfLR<y3Un$>URw0rs57A_LRy!=V6DljK!_Qw-=b}oAV2MNNI_q?gO+&No6370F)*yp
zURt&oW^9akQf0`edS@!iGHVz>C<c)qzueS&{dYHN5|sjxuN-Q|4$Ll^=;p}RoSlK#
zs4Duad$~W!lHuK5@^l_b=5#ePy0D9;OY{U3FqYo|-gpg=J$9ns5LAf%usy^yv?U<F
z4*soQ$YK0YSiTRlxMXks&bMa5iBK7E?l~|x2u%lO%Yz!|ueeuy<8I>=rG}!XS`}6b
ze2ebwZtrP0^_K>OwGlA)=!pAyNF0E9=|$SRGz8myP8zs>HOX9Il^$i1Z_M3t;8v%|
zJG)~R|2}Z)C9wx$8l6SI$8}w$75$epa>(Z?iOz<R>cu~kCEfgRyBe!uou(JXDmExF
zRKiiFU16>e(({N;nVTLtJO=b1F$_~6)nNc5;xNezj&7v93q|ci8Ry2;tpsiy6y(cu
za&n6LMYg>I#oPAV`^e9#2rr~b5!N9qKxw962Xp!nh6^Gsj<x%q2&K*5Yy<&?QLet~
zM1s-VrC`(dhc@3U9mnvs<1+2Mv^+(2vim-6%AQG-|1{Kk@Vs{0E-&})y^pqfQ4f-u
zf`}eX9K6?s0%!1bi%;MyJxF5z-@l#oFg-KN{SyDGe6zvErl}WH9bE4e`n4?gIPDo{
z-56S_3~0;y+gL#pe51Lh#JrcZf+Ju1N>3JDNmnL!6gxkkK~+S}U;0uKc03+dp(G;j
zBP$5asagbfuQUH>GnfDJFocOQeV9*uY-B_W#~&Qw$Px0B-83@_X%`R1H|$?bP+~*3
z($KXsMRuF@o%5UO!a%YA&lMg+!LZFsgm-0eD$6j+PC?NTm?8!X9M(%F=x`Mc<%57B
zk%o8&28MlMK<48A%zI-@LFR%<Y5;gNgu8bTbAUV!b85@g!O47ax)1)}ND{Wh+}wD_
z*({K@lGD=6b;*EQ;3&JH{s}qBB<WpQ<`1Y^*naZ8Qu{|d(l4w^#ZM$Lw9XI)CH@5;
z18<R5Z&ZbuCV+hdIgO~A1lxem34t*)bdM|=>5-k`FIqqJ5=(Gcw*8RZdv>^qTG3fj
zSvgkPIWE<ISDnS7dHxa`4b?x4&>>#exmhG^mH*g5l*!qnfq?=~gFKc94<qvXBoUtX
z7QrYMg#O(Z_@Egn%g_7cX61o2jos_YC9!WgCEK+o)Ws=A2!9yBNW6^JT<?U|0)#dI
z{#~v#MPOZoyoIAuNwRIof53&(2JQy+=g)y91n(`e>I8(B2%Z49IVPlfRI2mY<`#JV
zAT1OZy!+q?1EM~*MY?@~_go^<YjXQ|5NibKuz($m-QtCu;6W~WA?JA-Il)sH0X{7<
z;U-O45&AB=IEAj#;((BqssVcroDEEUaLx83WY;Ip_=dUWgd9b60uJIk$i?uV)qISG
zdf0V5pkLSo6i@Jn^~TeUf^^NV`mDMPntcn>($Tf9?PEc3RH8IGA;*Hcmt6a)i0ag`
zj^8IeNlc7IQ1G)MTS4HF;CKUVbkNALvn{GtH^vbhAMa_V@~G^01CatE{CT&Zo5Jkk
z%*+f2X<a^f@yE_AAkqt8EBv`a`L*Ykq}52(Z7w1$^-TP&ELJ#L7>^DYL0po_fY^<?
z)S@c4ACJL2`vsoo4XjtePb60n`}Sab$w{E3lJ*x1x>%y~f%)(TL=@-p=?^<wtyI0g
za*0MaK3(N*HX*s$i~Zc|WyrBIgRlC{6X>t4KIWgGJAl<4i38c7yGG96c}qS4-P*&$
z<HTPYOekP8>xm|Z;SLfc#yJcG>K?U%ZU^S_9`I2}GIG{B_C4%VPz=Y2Hk=Tm(I}mm
zSfT&X7Ey9gug6jGPq%xmdvsmu<M&G~U*pf{RRKLYe`SX0YBh{g(uu2k5j5*l9|icn
zU%vZU%9*X;!aNd<ch#J5#Z$7EJ8pCEZq^p8gfPF{2!`*Ynq6Q&i-4?w`K^7>Qr&?>
zf$=YvQ7NGDkh~j!g$MA=m;HXgGz?4&_QRI*0gY8->lIZdoR!#+TLC!%NN*(4N`f5L
zMG8p3v9C{6<hJA>_w}4h7d}S`H8iDG(6(j;9L-!n;u6+>adP14vV>uMj_o!H=@P?x
z2#U<e9KA3;*Hrwlup85HAs@zHvMK**vI%Blj4qEhO^A8M`tHu6_4dxSe~b?L_oa!o
z_XCU1;z@F~^8{KoNr`Sz!z>H7#Vbn?)FP2e2_0siYbAo9g2(?gD2*z{!Jx-K66dV4
z-TqdH3)~c74B$e00W_d7udD>#{B9AlB$n989aE;l$P%X|dR#Ij2bK{eQ8v;s*XSN8
z&-mGXDo$`zVw25~iCOeM<I8+jLHTEjzr|n~JIq?mHMpu|il{#3a0ICwxavInqGUH0
zVe;ICa7!QRXwd~<{E2ODp<SDpXr3mV9pZ?bq{|7|=3t%U!2PYRcn6cYPjf;Do?<T`
z`{8ee%76|Y>k<j8GJ$)DH92%>9+@Nz#a$T;p%*+#JfFcZ2QY`gsFErTXoE)6XnDrH
z=4Ox0UU1_=GZvdHaob&fs2+3(U^3a;nTT+ZLr&}6me$tFk^FD|+<uay^5@R09KDR9
z{+xSu6?wo9YUt?T+;0!OTs}GTm!1(Ff~`F`2<Zrr*6@tA5WH<%qD)QYB=J3k7ud|y
zVf79(8Y1pIT%58~t$yyi|LXmYjX+I(d{>#BgCno5j*OgK^dADHskk6|WJWO~t|JIl
zK0dF(l4NKAt?%Jf=wJB}@*hu1F8u-K?xk<|ljOd%Fi~Xg2@N3Wi|o*qq+NsftT&jq
zpremVrp^+uya^V!9F>D8`?nyM#7iAflyC0Aki8rDTF@Uol4BqwkM7~ale;P#U@N+Z
ziJ+B7dvg2X#~b(FoIKCQ;!o`yQ3JymNY3%FH^HU1)E5ttIfzsRX5w9oCyo`))7{Fv
z1<F%S^Mb#dr@A;RS`rDRPwo#;tv8JT4XnsVaUuKU@LCvNIM@InlwhWp+ku^n-X!A>
zPq)K1?QE?*+uAcOGN|1Dnl^cm<8g2OHIjgNRenTPRgN7Zi!^o22OJyP@@l;uL=Im7
zTl4gP&9hn5W=4}!Q-jQw(4@mq^oMuv=ot#LlB+*EW<t>iYu!`vBZQ<3shM=NaB*3g
zN_iTDg<ydg!K0cXgqoN*3_KNNR)BK5`S+8*G_(d^ulwek6=ywDFmu|SM;ebL;{Eoq
zFx2;0B~=qd8M=w^32m!FnK5pL)_vY2OfkI|_TvYOPMlr%kFN19H=Bt9{bDp>o!df=
zwaLjM^uq<p`mW#|RC>T@vQNYre=Y|3MXMKQ@#1TG?tk5%(9XGU|2np~T3p>~oq`>{
zCc1VU_@mUQ$0vLW;9@u0-=SW_W$lOOg>DZ{rXSyqXsxXd;*|Kj=oNTge|m7ePbqzn
zN62rl?-m|gbgM<X?2fW7wfV_R?xpVhUTXbLbNhKSw9+0ZlE{En;D**7srh~*K;M8Y
z)L(e&kUQAkAAj*Re&<cRzXcA>*8rzM$E%+emsH#SYxi77b|>rGQeYf62E0|lY|^`t
zlK))eGY=|<n#PUJ6Z`tg7HF86(uBzvh^R?li9Qn5z5AlZzx?R#gmz9}b@!>;%!leB
z`k1Mv@3D{jv~9nREa-c>t(>Y|WgeH#efV}O?YK2-;;+>l@7lAK+R;Auu>p<ew9Q-#
zODM<?Qm^D2zc!m!d<SRy1EP2!@MHdLvL1;s_mgB!le+f3fSDr*=-WTqb3a5T!$Tlf
zIx?U#_w;QprLR}(UP=W<{@h8A@r8njs3_7`c><GPBf?V#U8}3CnVFeMmuFknGJ|j^
zNlbf5U!RWd|MP0kBtnCbul6snbVV=zL0T+r;MH1C4q?zq*TsR_wGc8T5IfU=(k924
zPE6bcPa3=azFmj7L>jA?LTbVCSx+J-IpJQ-g?=2akDuSCBujUKKL$LOc+c3hXRN(_
z{0DrU8n4_E6nN5F^ldvIue_+DdY&*q*^e!ABJ0$ADH3z5ZL_&88d7~I?J?<^DBjR_
zX3ih5&L6iO@AiP21&N}IA#@k?C;*_BZg5J^GzjBNS$w|Bwx5{K^q7e>=9Vi122cBi
z8&cUqZm53^O9aSYx54wMtn);~(MR-toC{~-<R1~QtE=-~e2E=q-v3s@r=tn1$ftX0
z?J5h${#)jBOR>%u;TX9E%}ehn@hkYF{LIZmmh}6=dwcg!X(HaN<{g_WZjvlQG^98m
zLEIFi<S^glcmIM(zEdRH8p-Sb8}|a9eUFO@ulE<+1_)MUrKFkEL%J-$Ig*;ku$y~*
znoAh<O)&RVrMzjQgkQrDt|{>#MTQipPV5Xp<Z=EBsZHHcq;YB)ehLqS*=@d$#%_;<
zXl1s2MyZbkskJH~H-XlB5mhq<tOQPCK{E`=_#5J9)wt*6#kp))^L|z}l*rgVb*-Ou
zE6{^uQwYfATr9J>FU5-PfmbU7rf^R`-i4w`3z?B5hEuLEU`}}Hu_sb3=d4TFx{|7o
za?+XYwKElTnrr?rlOzWsEr3NAgfzrrnakVY1NaJ(rS1z{J+Fw~4o1hWr}#zIj3eS?
z3>?;9<Z!B9I+|wP<-n&tZ5z0m!5;=Ko2y+4R`wQ`Ue4_`BbSl^spC08d5YDwL-=%x
z@(n=JpVr8n3&!wYTrONo!4R&<e;;naUl8no_FH*g$G<?cF6=t0udhGtz?SB-_>p|e
z!Kc$$qSU?anNN(axDKu8o+~momnKVDS7do8KdXv`5u!sc9&8DEPf4*?;^1byT~c-S
zCn1+U?Nsz1j)SnMm|H&A?hn@jtq8N(C-vlCleg`>Zdu$&l{syf`7+gZM!JJq_8a2x
z2*97d?y}G^J$8mnjDS~ySyX%G45|RcRbKp-&KS0GaCUY-*)g5p2@~Ccw9L5nv(k2|
z$#%eo4I3q;SmCQjM@M7RXKm;I%tbPHTh^qU8^Xz>F(GQt{bYXo(&*|W{yN<`a6j|D
zKka`Rn1?0W_^7CsSct&9l9s=Y&P(->DjVJr2S<CR@%5qcS`=qF3dM3$ZR%*y)qWR6
z*10=40%t;{%O{ea9Jl>7E!>;#`T1w)m4y;>946r$;q{*O6{dFJwnEEhP$2o@pOC%`
z0W@E0)=|Ri9m1uoB?u#0g-i@I3Zt~RV1aLe^)`Q1Y@aqk8q9x}3?RcPcOl%6Bp9Yj
z?GtPF`_%#yfjBwL0#Y9Z?(cHz!WZ3|Hu4NiUfOf&Ep?+KOq8>K)g46E%=E0Zg|*P!
zPF%TU$l&L&OJkOgwo^e?#q!~xLqrSjX}}B{r!!|fr6pax%IUpRzvjQ%ZyRg3lm`2R
z4{|w6o@$0zuh2i0u&1Bjukh~S3>#o0?5W1%&E+s*G|oNx{9zYrnBQ1PYP+^{^QlkJ
z?CgA4#Ljv9^?Ar;$;H3DwzLJ_<CQ<-1uZKOP6@F@%K_Vdbkx+;7m$v~7*25K=Pe?L
z=S~Y;PdoWxe5nN+^RCfBKDjHTKm504Aeh?KC5KLj@?Dhh7FA-yl6Ncvdot#2dCzj=
zoYIxe+9`>ZR&^{V9<Y!24N9dCE{1%2TTAmjJBX>>D8*<b<)-F`=a!=KI2kOC9aK0!
z(I5P|cQs1bp1agXyQu6HL~3u=A89SUR7~&Q;>=fl!%CV0bHQbh8#$&q7kK_RWC8&W
zBALz2O-M4Vc;~3!R8EaoID>QMf^Jf*^qQ#BQd1Tuf<N?Qz`v97aiLpEB*&%!+sMhI
zKGZ7)&o6kC-34f<=7`77b&L)W<mP3CoYj=8(U>)LmiH(RnJN2eCEYMG87CG_ibRo4
zIXBpRds%CeqIU`kH|uQm!jA<g1g5O)9h`es_a@oRBq!tUE%<$DB<NYRe5g6NpMw>e
z5Z6as2}89ce`Y!lUJie%5t|>e9i3m)x?U9S@+99fRWhs@T5nvYOmpuS<jLi_*^wo!
zH^+5iMon(y+P~~<l}VsSa$VHryY4ad{%^|Tz^L-64tL`N$(-~2wJH_ke>J`~p|3q#
z&T3`Eq<Bc`r#8{3c0xb5|7v@|bbx~8E8!-pu7)l6TVcUVIAMh3{X+ct<*zj3ZZG3N
zbv>CUVY~r9Em0iK#f6A#q?zxO_P0+E61V#Y^i1$9M{)A7M4+mU=dq9NC$6;<QFBEr
zggZ;>)AV4Q;)j#U))o*jmG;d&pzgT)4!`xjb;OLvx1Q!+y{v77AdzIb?RpKRLH+|F
zF|V`qS|MYcuT;@I!DZuq;XP*0Cdq@{xdX<t8nz|;gNt;Rs25IppO)7->L;hZ3!B#`
z>hm9;5VFU3o#_;a+p10L?Gi`0i+Z}cAocZ>+@EcBKS!A-0BhkIz2bNL)t`9pQNL^M
zZSTH&VN1?d;f<GZepPk9_dZB7J^5xCJCEI6Mxy!WjDU{n)y;Bo`KNMI`eoU-UW5_u
zrtdqi?K7eoZlAVGpK&CouK928Kh;Yh>Qjx3>Ud3Vc&jF`(*G)l*F;55ZjgmsI-A_c
zU6|a!pP|o@Fq#`h#9`136}M_kN62B~gjP6*Cws8^t+VaH(39<?;m*Tp6Pv`24bnJW
zZ=&fsTwi-JEAB}Ns_pc{PGR0g0%wHOjINDe7fm|)`<M<)Jlga=W14KS3e_M%{Zj=p
zQH4{4a@(4Z!JgDJ-FCw^Tk99L>E6E&E6_R$yjA10nM}+@7K9hB?`ooBMVjY_ipg}C
z+=#&1c{%?tes0kH6?R=+!|AZ|Dl&r{d~*el_TOIn&pLMbiVqlNg!$kYTz&uW&B@8-
z(qoEvI)5Bfmi(YY{k7Z|fv&|fS{##J@z@_1qKky@KILi5@P>+qU=X9~D)N^(cLrVW
z%M7mmru}pO&IpdHC;5Gve4^LSMIE-s@=`@xetY;`=%q2%qh1gkT8i?^oT}Vj3(Y-=
ztlu@iW=|E}I?s9MUMAld@VA!t5hwOE3m=CwZ`8Yn1$C9bbGdGpEhEKwvJ=kOfBC<?
zlTwoC{dn_zWYhhuAFJlI^scsLq9#}#N?{1}fCKYlbYWL%?#f{iS>H*UZyj3l^ZR4l
z?w(V7Zq%gF{cD4p$!|@4E`Lq_`g^!ncXSrpzkTxS*j`-DvEK2gWcqn)kabEG0^vw^
zp9(#e^599AM5aSEzwrmt02j}*_s=#P>?~jJylBXBeH2o^ON`r>KF%Pd^DBn1UGEh}
z99o|q<ytAq<i8gtLio<tjk(lQK4PA(o?CTWKX2RcEm+Q!S4<e!wB;yFyvY7_6yT%Z
zyYtGv+?3$6b#_nuRks>m^y*3*&z&xr!1LMDKIfqGWkIz%_pLzx1j|ofaD;B=7g*Xc
zV5w=~e|vX=;*xtWv!MA-pY8qFN_xIQx2l@^UAJ((a$9@Xf-6w1XTsvdEV3{izPV^F
zZgw=<>I*mshT(qD6H8kBEI$6yD)#rIj-KFIiLeIS%0VLyU)~vJKQ7MjgMOi<TTWiu
zZoWSsnW<u|D&EKY+be+^N1|jUB1KSSDJvlRI4AAxcp|&hk8%$yjS4+W?J%3*cM+)e
zdTocb=Wr<R5;h*s&k~W~7`aTk0NvxZk}~=z$#Bqre!QHrW!tq}melCDiTXcOBG0hk
zs(9hI{FUG6YA<bnnN5ZgW!o=<RWUo4CeQljOuJ4u`&yChjaGz1423ZlnfjX<a~r;-
zT{(VMiG2Cr$nbEl1rI_K7?6ubUb-P@VQsl&j`On4apq@KRJBr!P4o2?_G$J)qeldW
zHfhe$XX~A{>uuw`sUQz*<T&U8^gToM^|#Q=b-VmdK3pH<jvw}^PZUr!tQTcQmErb?
zcWKVdeX`1uMW?WKWp`htpvi|DdwQqJvO<(J|6um;&u@>9bKJxi6vBz4M9mZCCJ5gq
z*HICUD5qzoUPvmC2*k=~IJG0Sry;w;nto?|@h_^bBu+<}EfLd4_O{%bb>FP`LK*9(
zo{XVg!wus<8mD=c9d888?@hx^uh;*)%!+?LDn{4(*}MB|;W8VU%4=$FZcf28f9!lk
zjy!m!j~hp;OqaTA?wxUGxMWRIw&d<tlN<FTzAm_qN*#}dE_Kx{T(=!BUGP&)xD3X(
zWB@Ji+P7v_H-#N|T;G5UlimrJ?pCQM3cbDzxRy9_{Kjo%)r;*MAd2+#U9bMSrskeC
zN9jR)N$+W_RUk1V`odY^$3H?xEf9$8EL|Z6p|5O&=F!@{tXrLRYui_=w9?19c*f^@
zh7OK&D_T>1d(XHeM1D!kcOIF=Rle&nudAQI2wc$)v}B)Zs;%vTC=AIM7S)NYM3qku
zF~)<>yrs&7Xxzp)B#ly5r$a)Eda$~NaCziegmJjI4xZaSXUdE5sgJM}QNUs3Tgcx>
zbtUU2rY$PlXxgUreXSTAVZxGnu9GX?X_iSB@Jj}+mb&BuBg8$h?uc&6VFbjF+{?O~
zX6Q-zA84L)LC?Ocx-@nwB8bMlTjk8%9fA|mOJyO%Xwk@RH$7ZwSga`^G8~U!vLt;~
zKSU#@i~is{c{LNu6ytz^ylI4iM1S%`)mArgrh^i8vF7KPKD%!G%i*&THErLM;pI<T
zIz{a7gRz+Fl=(#l9QYhYX(g-7KbA4$#_`*abn-F5?Wf$yWO)iRMN6oi4!EOaC#rD$
z<0qU5XS=9_n~2Yk_IUB0kYS7IjZucOM(rd&GG}(7fS~j%+=XoFtN)h^fciOblcCPY
z=erVizZDK^Y_60|V*UMS<q&ObUp$<wO`gT9CF3R653W;ep&dgyLyM=8AzKn*Y_w7X
z*Smf}UoqXMw<O)kb*4MQa%;vh>;A@^TW>lX=?61J5JkCuWu_kDdl)%(np5P8u`s@P
z`j)QsMsr&uo$KU>tS_dG2WRp#Xd_;kI*ZrnO|^Z3Y!X#dXEsxgj%Tg(_auFzN?+cv
zX8%#FH4}N4fw>^1_G5Va<!r@HfBYrO@mv5eMYQh<G+M0TBBLwFOC~3=9z?uxq_*eI
zmKi8>JqX|$T`qIkSFKn>;yM&0)!s|-@AR9Ye@ut6(H=1}^!%*6>nk3Ud1E}u7s(;k
zj9y~@J`&m>RIfQ&rs~Y1FA}IXC<46A)o^kaN%0vTCQ5qKNyGGRZB@JW6IJ6U`%beW
zZAVpGi`s4e+vS(GG4cf~Bsn`6<*olLg;eDzHjX;?bAz-Tp0<uBEZvK{6TK@{^S08X
zh3WG9e8fAW`~YQ|upDuBE353kry`&9gQnfp#rG4YS#S$qAXqW+aayabL~vx^j+?lQ
z6CxbT2Hgi3-8tiLm!d{8)OGi}c{*UK&{J7z!X=BuCaVbZ1)MLzwuperU$KQ;j^?}T
zL9*zEe~8E}b;+Bm=3d6+l`wX;WSW=#j`j+=6;D?X4wrK!M*9rg8wm<!6Ee8`aIZiZ
zZu^oW+hpUjOkUN2)_^7bg^s250IulIt<$TPQ>`(ZTOL($eu-1{0lOj@wP6tJM=?Pm
zeXuPPYH$8_ly`0y^XjzyWKEsm!{tTi9d&7HdmQ41ztQ!(aY0vcI##GG%;8ttm(2@H
z$opoJaxa)Y$+A7RuNTkxdy&Z~5HYjGU&Q}RgZYTa|95oGXl~%D+UcI5u_gcZDZ<fs
zIgWQetafKcEr2T6+UV$o5Cz=TTqt?mGpeY`*NZTaqd!9PhxGPA=$yl=?>9OiCMPGy
zu>V>R`^F0s^uL|EM4KHIQ(XKx$os${iV-90mFm@PULuv`+ob~Qnu|JgI__2xc9%oO
z^H@SJO4tc_)p;ITk7*9O`XbzT(~>0Z9b?C0SIUfzwm!HI*W~m9b#<fd>_-JbZZu9f
zA>C?iMSvq-!mfL_(1Ctm73-K1y~CZJBk@5-yyFJ)vE}IpUu@p?JT2Bd>O3acL}2`k
zQb2#llZx&vD|i2~;-syA(ZEKiCCasI*~PM48M}$2#5LOA45m;6L~#qn%EhE){w8zP
z{9?k`q>c_fh4ZQP_1+!1picu|5Z->Fx?U7(`rn6kt>QvG%8b$9795EQP7vwvW%g{*
zGSIUtegANm^I@Ro@W)*Ff;#G~%beTsi>T@`SR-~Kh^)g~)35Hvze<+6cQ;zxewr|j
zp}TvZC*Z;W9skZ+UFF?Lc7A-akY#qUiI<*P!xGm^)PXpt1n7dV8~H47Rz@=K@Xr;B
z`I@j>VDwnu)RE*SyZw*k<n{3t!$O8I&Bka<L`pTAAwkP^f6xi~`5Coh#1FYpohtL@
z+}PUUEj_%A;bIpKqW-(W_s>x(VUlV9ybzhXNq2bfJRUsS#Xd!38+Sw~$TuD!l)Z;b
zjfVA_BLyGO`Ie#M=O}p5@S~cWz2HyJt2^Q}_L2MPG-~Me3zOaW>>KjlvuO202dDBr
z4N8Sy|43wyr{!q{)m=t*5ajI%zyxd}uRI%C(-&KNdT9GD;E&^H!9IWKKJjq=2MfHF
z&-oteaMk$?Jj?0y54>f`q-erEilZ{Xoq9OLKvi>A*&*Jyy2=pe{0shpRf5~n?M82A
zwp|WnmX5LU+DR3+h$tcYm{(=p<h!2?a#hlTDvX*a!UsmKlX>!rVVSi!soQ?LhZM^(
zoDAwd?=3#0@_E?xum>s1RaCN4M^JS#VyaO_ANiJ`C2+>0m72HL&`T-#z7m!=lOyMg
z)2Y@nH`mAfeysC%rY`gT3n#}y=?C+EbwgN9uC)6-XziW^*Ooj(R+%KahDW?xO>u{Z
zau|!cye+A~xjXA@Bd-k_^_{HY>#a<o!i|nSY?<Qq76{t^R6rq0EnrZ!gV+wjpA|VG
zm1xsh{OMuwRRC9Rb#v}c{<fdddGB?`Yj=mY__PQ3HMPU0{xHewM2;1%Jx{*npy+NC
zJi2i1G0A&EoZGtOY%}Ze{?}r~_Nkuan*CbkPR<xMPwUproXah1cZj;|-Dl6)ZaVQx
zFgiI7+MyGx4#=AQcY;QYR->78v2T2K(~$h+xpl+F&oo&Z@oKTKbh)_-7lNweM@~b{
zfCJ9XBxYx4<<{quVyASdE!Z77$>ghy#4R)eScNc?n1xmIxhT)DLVw_x|2xkau1)PR
z`;w<S_<YL!QcJCigvFRAJJBuM^LArE1H~J~R}6QJ?kq}`v3z)kV)XmWy45G)YwN4Q
zx5*v(A+=a>BYBQXXR#f|0slM+n#IEOwO-+5(CA5vH@@K)9ZY`nF6FT$_GIwJbhr{m
z)+RsCVY$7=%e!xH<(7RsrYB%6Y*Uja)X&)o9y@fWn)-H*IZkAEBQNi`kMH!epa-9l
z0Lg|$j&jLB6$#y$&L_rx6?HzdQ@X2W<0q8&evAbWt<Q8;#TCk~;~Y>};oy%I(Da${
z^*7MeU~x0NNFP&B^Rg`5;Ss+fWc7m}AQq3EAtnBH8FnXH%u<8koIUSibca{N@I>t<
zcA}g7(38w>Xm4Z)g9|Ntt?=a0+xku8vNL}9>)e+6gPZ-8ib$a|3)kjN-XG1;jMvZ2
z7!2e(Y`cc2`Cc}dO{OW0Jb(1~fuvmY6@{SnJNHT(apzCJ*XA!Cw_%oL?=f8;vCpM<
zEf@!F({24}40zCPN^MXdJKoV*?piLArK;Y*`d;c5dc%0X;e0Z-%gBMwJBix6_inuG
zVYIRvwh|ON%=O){Of!vP`ID6vv@mEkp~M@3x=XuV@C>&qLZ@kj$;>uGN8pjL%h61g
z?-rco8q6ubswN=}yz2M8`S4}UpV$g6wj|B{99ns{dPbKUQ#;r}JJ{2oCt>QXxwN{b
zCDx`iT0NMdgYarGs7F!yf-Zg73VGw1`Mg)lK_PFZP|(r5THH8=AO)xULUL9W=>rsM
z^Zs)lD=*By>W4TlPid^U<&GZUyc`}LHk}RFt{S|Q@pPTkwJ$&&?XyuUeL#*jS-#{k
z(v-u|WwoLXrNWar1%-mKNFzvK+jbalfiqV1NpwPvpSI%F4;CY3?wEVx-ZMfqI{c4h
zcQEKIyEAyEx9b0>zKEb($f#+XM+6LI2CNG&^nH-Iv^{HEvz%(nI*i4C5FucdpmiI+
zzY2qtEIN+bXm=xz{Pz2zuQ<^o6G8+GM`&+H{hXRkWBq{>SzLky=S93mV1aF?Ce>BL
z<fk6o{V;i~?rt-ck2p2I{?q<@EKe@Nm8+>^*x4Iyh3RI4%Fxy}0*2$tUl%_q<YphT
zDLSRL)z-d_z#FM9kEX2UZNI`=-9H`E6peG<Z3tY%4n$cqsfs9xK1MZXILSHG9VEQm
zA^e7UGVAQ0qIS`*X#O2_@q6{86XI9Q`M-OeK5KMgi+_F;x}A7Z$hf6CPkAKquUJe7
zQ6IFi-5zFDF!#~lt<roJQ;oyl%yBtL8L}&F5NrFvolL^git0OOrmEc2KW^#mH%+HB
zw;gb#TrpLcr9!f;MQRRO&0hB2U>w8V;ZSxd;Fo(Oe{0$S;YB!%5d0tN-twu<E_~F*
zi@RHKcPnneg1fsr6nA%bcXx*Z#ogU04n>Q*p1ga{{&4pB{R8JyW^zB1Jef&W)-CJ0
z8eki9ZV~c*p^+SIdxqDu2|RXU@wCOu1_sR<HRw`90!nDO5-o9%I>ErR*~z}Wz!cN-
zJD=r#H!F=MOei}Egp;@VgKy~AChm%W7T#C@2=g+(I@odrCia`jCEaKp(&MsDW$Y<w
zBj;cLR8>zTmW=L4J5dob2^KNlF<FTe;QA5Plup;E`c&etXl8bD0zM3nideKOhpd|g
zDfsY5sZS*1C*iY~H@E@h=iiopFZ~p=dFjC;(wR9QA}|@OFacXXy!oFR@jal^v)Li6
zjh8X_Oy4QPQ{iaamx`cx8DcYV9h1z$l5*-^ym>Ad^J+8sa@bY_GDE@Qzs<B-=$q{k
za-Rek_}!r}h34mQue-MzBGPM9cQ@YD(OFpC$Q}p!x!x~L6c_8aP9SHQC)Y)Ph`!3h
z3V;l(Djn}h$BWb6Sz5MJ)yAisyWYc`HuoP~D_jL}J*%2+boUH@@v)Fi!n8XSUM^)k
zZ{77da1TL!lIURwwBW=5v~ZL+hmy_OeKA<+YsW_|oa@)st)Nyn7N^U<8k-$R{aWE9
z_r9|{7D92^+t+_HB1=UNvI=8Tiz-zE#e?w>tNS50P`idR!VX5`ntO;bYI3n&4I5WT
znHn<OZ+vD5$pmZ|2{&(TiQ^=;dkj!vJ-)m9*>xLvd!sd9QR&$h@39j6>WUAuvgs#d
zz@DhF(t74Ocf(L|c1GbaA6sdnMPlXsWo|FA>Y3B(qF0b9ek=vscwbf{u`gaKoXjS;
zM3=7WrAX=>#2XvQg&jO!F_&9S#49AtyeaHdV*8z_DYfSS7~ZkJyxIWjwiEG4_OP0!
za?u{pfOyD5Q`N1lobUW*Ic|+U+B0!cx11AS!}HK`i*rO*+`sHj+xVEc{YTjK-3NDJ
zvq7DX?=P<j<5@({>!&WDj>q3SO{a%0XNHlfo6&fa+F|X<a|5rVlJ|4tMnhj;qU8+n
z!yFTD@9Z)j?=|~Nt<+qX?YBKm0PJAnvFv8uIm?|x;U=$Xa472BoDoZgoy-n&gX1&d
z<f8wf*9K<p-(!Kc%tpoFUjjzw1ZywY@jBU!BOUzDyTLiGD_X&jjOhfr1_l<5nio1L
zn*g*<Vzl6PRpRu+xfmQl2_{GphHfP<ldv0QDXk!S<GaSnv`$zTK|umC*2$@g0*{>d
z#N?t-_>_^!2Ida!n4@prvxs5(E>#nqtx{27dbYDxxdcsAwXG4XmfyNzxN?k~x8|R~
zQ~RbeGwkw|*s=HNnwphm0j0{IGptH-0{0w$EuZyAyhNP1>obUJ+<-vthpI>W_<Iqr
zsbB>>OKVhbkA|ekx>&)HB^BE)rplpn3Kbh(lGG*^K5)Q}kuHXvktRR3qzr*|!A>A$
zsTF~hV;LTC-Ux>WK19?q?omCwFUm*<>04)$wP}4YB6IPM5h6Eg$O+yL)UR2dt&jKf
zC%-_2?_amHI|sorcfjz3fGD7d#{6Hy1!#1OQ@(hSgFoE&6;?b9=emxaZa$HkSxI|&
zQKf#NsKzhS3>C28*<MTRDRh?+M1=d-rVY3SK;+s0h5p-M^Yki~FgzECIAIL0tZi!M
z4ZMJ-N;~lz;(d$B$sWA{^_YDduC+OH8s|a?y~NT?`k%sGac(EkD=J&$7E}*aYHoAP
zeQz|HD=Rvp$bBn$ZoU#bcB7WuMQBo-1$3I5B@A%^u1JDVvTUy%WrgRwqI;C+CK;Sa
zcP}{yIy#jKDd?(NCCZA_!N3h?Hbs0BO$Kbh5@@o2>ezXT7{bkyU<6l|EiFw;y($XD
zh(%P3pm|XYinb>iNe%vfo84xUTfS<65gV6XW;y{<0<ny>f+@pc77m>22cwRTz`v|P
z@B9fr@$xkgc!=BQnWb=N6IcP!_A^9~6ToFC-hj>yD_Vavr_GZP!`cOil$OFvr)8fU
zK{q9Z8lrJCS2B=ksKF=Y+`;lvRc4OVMVIfhA!)B@t-ShqZ##FrP;LNkV`9~+3y*|5
z3eNM*WdD$Kq+ZneTZoL{oh_Ym!QGAbI5WqTk&eb%UW9yG&U5;X&iE1aepkymsQxW+
z4SfBV8F+v&W>>1a@f4%LHVwzX-vUgrE8hlC{X1sx&9gHs^guGX*Z5@cFJG;gI{dI*
z@N-};$Q3^tF1>Om&XGI%hq<yag4DpNp-Vr7#~09|XIT4`(L7Uh-r4V2iVq2bZt@I_
zjEn$wP~SqJRlRQ&i>P=%S)zU>6|vv86HV%oHU3Il#;w7<u)Zff9X7V>suxe1l*VGM
z{9`&{8%4?IJdenZ{P}n2D328equ{))aixqW9+8r=`}?s-^9qvdPz@|-GOXA<rdWyc
zIqt<c&;~!pmBVcWl+bX+;WnY7nb82UDsSjgC$=M==M@=|`8n*#@iTl{f+6?@rzelk
zR6~`ouy6gqSXq#Vh68FlB;%SawE9mGOV>v9dpQXJdz>_}Rw^U%M<hk)uDJ^+Y)mmf
z(lK!C8mwl{Veob<6aA{gi-AYq7Qdk%i?9RZptMDkaE$q-MlmcsX5+hKO!{CHgc#p^
z6}5=*)0@q2kF{oHUYvt3f3c~5`<Fz{L6jH-H_l^Sn0~<}=X_m)uzsJ~M3|pW=!ot8
zLPu(ja`cm-QbWWTW`QdZskwj^VXj&J)X0BRGka2p5>OnA2EO*}C-fIM5@@Gn4Ei4`
zx*;=U8GlK?YE&jzCOs*2DpStYP9>@k{!KicbvI1-j>vxZY@aRO8=5`5XW{)brY+X&
zGU(Mb;OCka0Cw~A4z$qROJ>v#+cFqN_y`HcqrJ}HVn-KqA3MrT8S-WZt7BWPh<-PA
z-8Szw5)=|$<eK_m{gYLU0)nY+huIX}RoxM9YXAM1r{S)pEMW+4v(sdVE2L{*L7HI#
zl0phX*<ujs*JU#0*AhSoKC-jUXW$`K6Ul9w*m9)Q)$?Hv^Mtp$jcJeJA;!vK?+DD;
z?mTfX4vBEdE#ch7X>HEE^+-5>*Gwx<QW1kXC%hZx=gk`~M(^w6cbUrg_u4&i3J4?#
zQu;NyXy2Dv$Xbi0U;_0DeavO!A~4lbp;-qn7re7EP**y3$h1ted@<d4)2eV48;P+@
z=zEk@NE_0)G(gson?vx+4ADq%)bvXA)%KYPV}DBM2i&Q*CH{#(cG!G!YA*%>-^eV)
zjkoU)TN4%2mXqkA{@J(FtAPfjou{>jhxWLVkCR1KS6^#UMq1ld$2IXi!W=-iaWJ^a
z^`^6}$-&M2&&|#gO$QV-Y=O3Ot(`x872kT1Rvh2k|EpTVX0VJPf+Lg{-5dTHCvG&!
zWR%P63cbp}i&mn1vSk|wHx*@YwSZ6}*!*TO=ZESNSZXvs+SJV&&9ZBYh=^u-^5ay+
z)XZ(6zXB{5R&U3|DPJxz<z;2Du67u)zYK3e`!z*?ShV1rcV%Vupq)ve%ql_OY7-k+
z8e>wn+p?;C32Ci9N}*Vp1D?^CV=(df?*n_)|0}+@Ne-51AV#S)uGj{VCbHg4;oGF(
zrCAb+8T;c@<w=JJ9bQx2Fop}kDz1-(iQCF;rQcXfE31x;)@Wi*n2MM;Px1s<H29(l
zj|Y@bR}bK(`I%5=q03~Zd6bu3v()4vN)TVs2JB>ll!Y@R9W4i+{9;Meq0hLH_nUvr
zV7HT|`Mje3x)8-L-m$B@=$gzV@|pe4ESC;aFw#D{dmi>^%{_^x89G>{?~fb7sicn1
z+(S8Pw%(}2Bl~ZPHROluA@Ole5uJ7M%Qxm!*g&Kfh6YU@_Dcn``xlF3VZ<(zoi1S1
z+`&GN3p!DJ5iY)$TGrC>x;1<15>NB*Gf4{uixCw?o>JU%IDG2!x|?0E?dP=Bp(fAY
z%U3#o6KhmFU1U`(sq(#*V*cF(rnKe!8~d5?p#QUN%N5eBrHysoGwsa>Y$}J;BfcNq
z?Q~F$2R3BSfx}Cl5rQlTg*bgei<Weoe8@X&S*oi7)09WAMG*g79H5G=T+xgsIfo1o
z?y?BWmB@mhqi@rquWQYnS%5sq<}i{}rc;lyjmLbF$qRXV*8@ixX&uf{jygJh!x&x~
zUaByoSO1h>Gh<<!e(t0pVD&R}BUW*;UMqX<-t#u+-|p&uCm@z8H4oo7g<qL2GT$Hl
zNHZ5myJnATZPmIjt~8uNR7PRE`cQ;l@SX*&<dx2r;JDd7!Nuz66oxg08nJ@iWNB>K
z!LmkkoaKG(Z&HpD^9==!!Vw|_E3jvv)*P0i${%7jPOr;BLP^^Vzo<1ak(xnzk}yHc
zALSW)&OdBUgN?KCq~kS&c#XP8anrZuiL{g14UR2SC&*11wlY<mz77|!fLH6@e)>F~
zDR@>Gkgi^KBxz_t^Gik}y@N}J%;D;<S4~r^7Z*u%02G=wpM*P~((9-QDgd7mTZKfp
z`f{=cTB&1%l&2h^DZ3prCKZ){NDJAiK;7(nX3p%9#fa?Zkaj|)Y7E_&F^4j!G7^us
zPn)}@nxv?l|EgG%X)geMBErkzYLW7XiRAmeqlX(T)IHP^F5yqX9Xb*ux@kOe*R=4(
zWF=0exux&St+PqctTd$h`4t)EmQV)VOC=fKYr?X$JyZT0*Z4+fgm_9x0>})7;grT4
z2{vPs=m5i@CAm{+s*lGql@-CITBq~2T99e37u8wB^rXUf8}EYW>#gIldOb9sQB#K+
z7chzHCqkXh#49G}P^{C?qAco_7IE{mSMh}88|S4>;fH2!)fn<?VyUbdDo`EW;4<kU
zj2KD65HXb^W*^dzqLQS_v*+MFzrF8^qWY%DfjC4QyZE6RXs9go!)VNlcWaF!F3h%P
zDiU|_TmaK<<Ut*r*_`_qUoiUZx#BxcLIdHplk!ueGEXyY(J1g}ST>F$AFz@bc7Yr*
zn?k?6dXoP^nSVZzNFuivB84P1w(b<#e{GaeCE@8Ujz^=Xt<IyFf;{U}JL$K?%PX!n
zk)+0U_)|?CDL)6EZT{ysnRK!8AVUzCrJyP77qY^1_ol<dE6Y(t$7T)RuR7InR!XEa
zwecjoNYY|}K33WV<d29LllX5jDXhid3c?L0R#Qg4#*q8C0))R9=h1Os8wKTAPJi@u
zV$Gpwgb|A-hkdYAB9j5u_xco*s%o6!CNTmg@WfoL)!d4Z4ovlwDQ-~pS@_@Q=W@ey
zK6+Chi~r=-fPPO<N{rbH%=GG7g);u><a?BKZk@+)MC4e-&GaPrr9QYhxN4q!?IjEN
zp~)s3acPUw+^&<zsK3eZa8h%*AE4k4tsTuIsPMOz?8l6qSUcz>=}svEE1FPnUD!Oi
zf^U_@Zji2~*T89gB{e8z)XSfgQGs&qc1qSZXHq}|nWxpynDaLg*nfu~m&e?guhR>+
zSc(3POkXV)>KtP!HgeO@g=71w1!se)6M*@~LBt>2#{|UMU*izXS>g)F1BZVu6THoM
zU&t$E+FLGnr(pSrbm-YY6iYmer1eEHau?6q^pOvv3%P^3C2wQO#!B)!Q74WBLWmB3
zCBL)goODyL{Pt=3H?h1#n4<d`^gn0HeQQoJkJr}XQ>%sj_*8PZ6gw<h<=D+Kso&E7
z)I`sxk@1jM<90-4*Ecpub+>iVth}_?a=Ij{899nlXI?~d*;U(VTPij-Qn;+hAi0DV
zyR^|gX&fmjftuyi@CT^uyuU_{;tKVB<Of&DHnd#E;hf3de*SI#9!`p41=n&jbiO3j
zdFQId%*DX-UM5k0vWrio12`R@05|0jWi~jN4`2)q6xc@LIa5=tL>sG?j2F2f8|Gx7
z>cLn0iIYOI8!NN=L*Jsj74G#DJ=Q2VT62(_r6k;*BQwl65qMrN5vHPvA0}af4CG=k
zV;|?h#};<sBx7)0AFDb2NJ7og&n3%nFdG@nkt6uR<HQKZ!i;aDqRxn!OslslRn{tV
zi6}v=A-x`DfNK}fuaD{CHoTziEy{bn-eH$XXFgRLB>cfPwlL7x%GOVn>RVPUbBUBp
zq+_LTZkCg7#Tu7(S72?TFF^6sOj!WMK+0P?aN-?sQW67mL-UQW3}0FSMmcR0p|K;0
z&M;9qe(vx-iLft~2wR{=>N~BHqvwoXS}!HVw+^j|-3bHRFpo(7*fDk_TIwtDhM)BF
z%*kI`7)kAwqST9MZ`Rte09S?RX=-k2t97fiH0lQ3-Xf>Jjriz=P0F&NOf04rDI39s
z4CSowyUGk#gMO;Slzhd9?DP8F3@y#dOyV3s+!(_Is$kbX{sysdE2h$B!cg2y%ohBV
zUmn5FuLX4^H)Lkyk3%Hs)J{js%PFazRql7JK_T4B__J3_{kR(VLQO_i0@Jg{r=`DC
zzPV)*wOkyIAB;BL4V_ra@=eAx1E-{JKZk}(N%<IO^2e;XkDUT{pQCb}_*3!!Ou}g_
zw@V-GS*|;~_>`b1=2HJ-|2-`~X%^b=GW#$U=OF7+u4`=n%pn$h1`m~?Do-2)%oIR|
zTvPSytx<$FDT_okzDmOy->$n5sS#>FCo2ZX8g!0`_YLfNR$P4N9aBR%g(xYnfenkk
zTk#vS#}_n6yiskwqf5T99On(9o-6kTM#QD@Ta2kOB_NB9UDA%9!&bln?m#EBqvQp%
z85+#3n`;dzPQ}*4w3wzcwO;I$Fo&0IU)l4P8<<qp5%9{dI}U>om2AWp8boViF{2_y
zFMd&OT&!J~r267{%dr*w%iy@;jNJaXOY!FH!l8UvC)0X#s%rV}&uq3Ixe}8D8M$$I
z!h#sv8d^A2(1+VkCN4F<OITS$h_1I7g;yj^3UoE9V68p%y=bJwT|S(vn|SF-Y6@^m
zE})ylup(-*4r2*PRoZ=xw7!bW6iJe%i^}Rx#4n~BZ5`#a7RL++?#vxEMf~3S`Wf2K
zgQI?cHy+OFsvSyNg;gd=4$qZAifg161DQHeRWyvjkwP{p0yC(or7pcjIavO<v|-Ui
zP<7#<GG!B>WU*xe%lJkiha=1G07IS;cQm<#cr;G;EvBa*A7k@xX^gB++V`B4JXs=t
z3JpNz(~xvh%zAPbhZ#2F?rjnUj??#Y53_UEEe9vAlF>OZvY=5U@!gb@XR8IZ$?jHT
zPr+a4Ft)SyCf<Jx7&!_`1jKj}A>PE2O{%3Hi2+Odoy5cCm=A*O|Az}utJxPBXectB
zM5UHnmznw8AIJ!WAJwtJLK)$%brEPeB3t)tgmqv+Q%!<FM;|7O&%#7$TE#+x3MjeN
zPjo8J4{uuKNB4doT=$mmkn@+^sgX>^Dz_3g+pqYu{DW2VS=Pl%ZE!Wv%+WT@C?ZRy
zdd*Z-fs&qjc{siK2Yt0@9~Iz_^53l?+~dX0V!%Kc82*V}F8&2Q;ID!gLIGP!5W#%K
zUo1Q~o)bf?JdWdO(|i+(Qap0lMG+XYeTs?6;z1kEnBORRyj2&aYWOYF345({-TlP0
zMdHN2u<M4GGpf|+s^xC}9MfDBX`(J_xxD*i{nPs1@=?l=DD=1Lj5t#ZLkOzj_a$bA
z?r5%obS5-yLNkmjk*sprkW;~!eP<i;WE9FwLr1?!km@B9K|?B`DK?ANuW2QJ^vU6)
zNdrL3Q~p4brY5=z&Fi?SR>~xoYlBW{;!+~ypo(h>>B(vV?N0N%ZG}6}SXzLB{qGtf
zQq~3zi@;8oZ<pHUjtQL02<!<ujfkMm@E^=f*CT&MbIgO~O4ku(Mvc^5S!tro6nL95
z0h*Q=G#2q%-e5~6jdlqBkcGzYcQcYru^%jcNLuW)))#x8n1emaVpXFJi3xRIn$~f=
zN<sw{EaNjvKsY5Lb`0hFFU=Urs;7rGO@W|{!{@Y0V<x3g6*yV|JtD*eU1OamEN-*3
z^&q(Ls}*pSOw^}ix10iNyu`g}klX9Ep*!=r$vMA%Z88$XSeur|gexATPe<1R$F_o3
zNh4)2)AF;A!CtQ%Zr`Yv67Wy;a|yx@duno6L4``L3#4sozu<7|P_Dd?ASHBh^cvka
zRT*8@<h2@zO3@QYz#Qodr$geddt|w_Kiv-{x^gnKNX`C+e$YQ;`<Hv`bpDX3mFKyh
zh02B)P@U`?IzOJ@*ec;<YBMicKepZEf{IOYUhWl+!nJK8mk5r>KSvH@&W0V-3U#el
z1$%_rxMtfa_(7+{Aa90{eX5y^$qGb?Kh+gisSO#ChNGSF$q>9pZ>0z}3d>A;Mx}zz
zHnx<bVe6tp4p#8G+vU$)V;gloQkfrBqqEX_d8}-{m_RNQ&_sv$jiV5>K#p8M@1m1y
zHNWKD$ZmYGxi-Rb$~6uI5&}zgzbQ6gK>yJQUlB1M%ZYdzzU}4`LqVFJmr*A_FJcpy
zhJ0VRvr|bJXo`7)XwS{pcnqdcNWZt%O#4?X;)0dW&}(8j8+X0seU2pPxZo71mg%-3
zX}L-mXLI^Yyu#uY<tUCfbc@gMeMf(_g13b(HQfBzq<NKF%hXlyU4rgyxC!~L0W+3^
zrWppl*o{_Z(k%IVm2)X%WbCRd;+AXjRFcp7ziTb_m#(I$5*rTBztI+6PJxr@NEunx
ztyk@(;lo^qHYws)B<}y<`7nuON5aNgjGS_w?-7!3hX1XFm;hz8tJd`6eok&>(SuEv
z;;oDOvjhE@Eq<!an4VGAsM4iMmnv8OzaMJ&L8CqY*K_~-3N6nnmFoZd^8XcTx&>to
z|DRX-FJyL9DPQ~leG^b<VnPK}_4yw|spX*7<Np|<R{(8+|HqJe-q}*s|Gv-v>*oLe
z`hkeuJKYPKykG?^dpH?rDlXcJ%5Vfk!CW*;)_m9EF(kkHr6$kx5QcM|`h3G=w^h3)
z5hyxJl-8!!2gsfC3glrEc%eZcryj1Vi-1)}EfZq~abx!Hi+BgA{%o|qN6%-6#8r!d
znFB-LhldbB^zEDp3n=j(D?dLK<tF;A&l4%NM(e*!@j@DyaEXG2%2{9@;TUP`tG2KA
zc(BbsZ?b;n3!7{gl~ww%e^-rq)T|95#^aN9gZ`xG`^VpcAtV3_T#aS!H#Y&xkay*e
zIRq;o+A^eaJRHzCKjkM9ghcsU1zYzk7o%i<{TMlV{@#tRCx63T2Iz}X3`jL5+rfZe
zUI$|kGB_$RaDS<B+NZ#)fr4Vmi2O5?<n6Sve6a>Iz+$jsoP#X3PD@&A;^oB)5eS2d
zXxaQVs>UOt+f@fYMkt#vXmcRGe4oiGjL;tw?Bl0?ZYJPyc+5>sG@l2d6!l}!T1g=f
z<0?B$zrSwG=c)kJHOsVK5tJt`Q4mmo5ibStcE!AgL&A~@43?+!u&9W4Qt-!sKkOzU
z6E9!haK=0ty=X)q{daDIsyn0f;K5U?ARA5x*5=Ja)`FKEWfMCN{VH9qfM^8DO+UQL
zmLq)SbOTSiRawK$f^D3c_$u-*SC)4XI_dB}#&=uf7$ox@WQtfPQNok(f7F1Jkgb_s
zqPC_k-LPE3T+*9*bj+BGy|tPg{U{MnFW;#K2EoBJA9C~Z&?Td{z9I=Uy|$XzX4B%Z
zgxEbkLhS~jxEwX3)VNX?Dz>h$cft7I!#^mdhcpa@G1Z1}eKzozqihI`yaOANH2b0Q
zb1sCPD!73&1!jW-rxaV*gHrg6#lb2x$wZ=M?;BNwki|VZaVHb)vrW-}>bN6N;X|^3
zC4~FJw1+cZ1UBhdD=a~_7@js@Xz>D2U|=Q&g(#SN4#QGBPgl<l*Cq=g*vzqdbLyp3
z!}t<9@(&8%>kn#+Q8l3+5T-iR>ZWC0jA?$xwiVjyT5L;7)=9Ib@4d<iJ(C-~2!J4w
z$vEDR)k+H4iF{8&L}rv)=JO4L-7^*9r#M15_Iw*ww=V8!q%tpRZfrGc-o|cmT`rL=
z1otGw$#sNg%XKzNCs%Qf^yOl|bn@!7<Vo`&%E@U44~{<!pm^?nTuLL7rp8l~etOkY
zc=gmm@I$Z3MPo)0wA3|0G779iB}DRsS0=J3fHlqO>Hli_4nt80Fnj45jj1u?>7FD8
zL(M>PxC_WCP8!*X47(zMq1+u73tw>xIWP%HwtC-8AM-p71Afm!syk>^rp*TewRlWc
z#9rH-c43FR`BI0ASRuzkZUhrs@(|@yNB`dW(T%N|@eG=eWf7$Oi<rMy<Na8pQDk9~
zCkpO!1GT)S7#LomS2k|irNi|eE9=6TI#q&Y<++W{EOp&1kNWI?<ENXBCR?ZCMkq5G
zbP^k#_Ef9Vp@>^qItS-Oe4Q%sFH+B!O4d>1EAs8mRjx0OB07^K0Gvt`gQ(~(uE&|2
z@lO}rM?`4jdj>by6h!i1Wcb!EPiJ%M7)%PUrNWGus)x?8?0U^TzF(yVV7|82mNi1k
zW<r>ZO2ZkWzG9-L5FT<TFe2&<`NU87p|NOQP<X3nLXr&Lts(~TZ<Msbh1|!~hmq~w
z`dP*?u>(^T8Qe06`euFtahDU0#A^AS0~l(H&RZE5q>BAllIgA0(Bo+S@WJ%9Vmk*D
zlH)H}zbviqpQo}Xd;O#1n;%cf3uj9($0Cu3kT{R18w+q&Ofd3@j-yUhl!bAcxi~~%
zu+TfD_>F)%d?qhxmcvgG9ua~$Zb+PLcBVd*)<vx@5NyrG-4Lbf2S$;mB!Y$%n#eBS
z>ceSG%&l64W<Odg*VLV|deiH<kW-Xa%G)ZB3^QN3Sf$JLkxiITQd5rA=P_IzG<hq2
zBCeG^o$*GDFEe`r3x9~NL~BAn<P=ifCt9YI%tJnTmhk~~!+NTMH?RQSLY8i-n(^H_
zJnDnk$3yK>gDyUX2#aol=gFKSBK)SA^=+}#ABj(TM@>HAVW4qr!IauU9g`|<>#+FH
z1~1GeEtiCiLE&pjc4Bs9dRF0?hVD&n@I(rkn;{M_7^nqVF}&xP^pKx3ukmdda?;VN
z*-uyg_Li5)3?A!NpoYAuhwZLAI7lH5T82>n;LQzvK?BBsKCG`$^D~E-)%#Sb9Q!-}
z$_6V2Jv4n=R<#X-ry^Tu__?i4*K83D6{>_2<Ik)9jb{xx-EHkSSRT`E2wO-S><x~J
z<Ah@V*mJxCdJ8L0Bt;(F|5ZxV-o|c9CpXEG%{Fr<dQm`MAjjp9Y?ERFWW+9-cy=7c
z@XqHnCR6y}yMz*Mano+|-8;cqbtoiK*O~9s67n>|a8^(}iOp<CU&UH?t{goe3^#kd
zQir7`2S;H)wyQcSs}%Vrb{oO)^a*-7X%v26scO8-xp3=jo+6HQt}g5gG=#<`tpc+S
z;oL+GDo4r7%q=99Z4z!qVi~jMnR7026KtmYDav5PkQ)JnkC*jUtyOLSxXB+m|JIdu
z`C(?aUf3!<s!~~l-<flo{PzGb{`;aMp1#~QYPEeCb>~~;EkA<QXHuzN6YM4B>Vi8z
z*W*6KttBe^kA)}7^Fi=Lb3&GqnBw{x^m$G!+-7=b^TD|Ym>{ANcbhxT%W9c(FbwZ2
zGtxgS<2|82$!LZUXMA=J5kI}0FX$zrLMBRrFXvE_tR@f7AP3!xWc8tgI9U^zQ@^+9
z#be>P*IyZ?jl^mS{JdXEeYo^Wa*xlAt92qy43>+9uHS{EOZ%t2?J?j)J9@<b3Y<sH
z_oBpE*Cjq%=LdLdmZwGd;9TqP!VP&J`l~JbQ<5Ene_^+yw&S!bsa;{9fq!CM)9rF!
z7gxQmCpW*PCWl?P#K@)}q1=kGuCx=W@GJbe>Si{3Q}5Gk7DAhcSt2`&q8d{Hl#@y2
zRbFxKqA0~yx-|yS<`9w6|AdzibqRsGHZGr;nTSfrV@Jc3OyE$V^T8JBvz}?=@JL;X
zH{_%WO-&Zby0^hSP;mCW9lDmrqCk{*J9*!T_b%QxfGk~fyzR)0YZZU2wZuhoMOU;_
znC=n8R6mJDemJPrpzxr8;@U{7yUM7<v*y)gf8^3{H9T>ft8g!5Fe{yVKIPU=qu}2{
zoMBl1h+Gs9WUhqQEp7Ti{ELDPg6hc(EuD^((S2I;({?;Un@0rEawY&F#@Ff3rPfcg
zhbRtQ&#cIQjs>uf2(Z!w|EhWX=D5AIdZ_8Dg_j~$0@G?hq&^htMpY;E+@`%n2+Ni7
zsyc5L+R191*(*~f+@naU+%aV0$O*sw3=GI``xh!D98wnN;<r$dLrDc2W&UD#y%6zM
zZJlN{u0Map9{0zFeS^idu-yEKogi7okj|B9j$r1m!33x{#bFrWrt0TSyW_ZS@2E|F
zPAV!Za5YReV14T2IM|o!MAf?Rt}MTkJxYuSA%1pHduS$e!hJ?SdokIS42-WjV~!pZ
z0{?`&cJ8AOoLuO|0lSj+r?-GN`y2P%sNRA75DrJ;sgC=`EUx3lwAFwDGwtgrRSQaH
zQAUkg!0p1C>BNC!g+FBK)KU$Y@b+A~5@gS%(|=osc~If(hlqyppQLhUrzQcLRhuQb
zwIUKrtxtycku+P|u^gY(<<UDrY}1b5%{(+2EYfqdUg$Fp+tYv@iw^l}tZ(&TzoTZB
z5b7`1I8p${7xk1pXV3S1tzBWbsTpPOGc+sTE^J4Xdi!|8pQ7}n;d9^hQjJW1wP$0N
zTupP=ZLm2hNBPr}3Dsh2nerg~rul*$A_*S=YE-JqLk-dhV?T40&y-QSi6s$S_??KQ
zg71kW08m#VO)7z8yz~-?P1@mpg)dpdHb%-^5`?aDOFm8sz@%h}KXX~+)xA3C<ZiRP
z9pRcUT$T&2C>eX!$2Ax4t7D2H9TA5=A`YFB%Khuq)(uP@X%FEPFH2Le0#ZbamC{_G
zQjjxDXxV<zy@|E4GSS2GZRW?!)YeSa&D@;^?N4qkpLlrnY1x14$%`wcynM~EqDr9V
zZM~(!9=_$Dytb0f21gzwvpeSB?J(x{iP>msrZ~4*_f8!N;itrvL28Z;uhwRoF(rQ7
zs61PBDod>*2DgXo!FVR!6|ic30;`v+_lNbUm-xhRjEms^x%1TZKc;WDpuspwhp4g8
zKDEAl2nEIE`ciQN>NM(*lj~RMo7VcW*&d0b7{iNZvBVb>5h4k*DdfZ=^3bfpaFl>~
ztT6;ZsiDpQ*4K@I4NO-Gqoms8;z{>3ehyfHESuXiTrc9G67-P9A*0KVP%l{L7+iDn
z&@&0Ow7EiEX__X%TLHsX^3hL*7=j|dTRj!>CloRs?e!Sf!*7KJP~&*QMq_gmlt{+3
zY51w_UolP_B^FF$my^vk;pS5*`rwGWF($;uR0apoLk*O93On8e>AJTJIk9qHlcrP%
zYpU>L^-?3EGXlgLG%QGKlW4wa8Axgq;~%uNT|z@W>BAlBj<h1|7B+a10&oQqr0R6>
zX{B-6R<d^C5r)|s8)lb`Jj$Bq8J+bU>?hf*FI853Q|_wm%Y}z)nm~%=Rd#K^@-fFz
zBs@Xv1V_X}ZUW<mmMI@bnKq|qu#xh{-;`WOH-=te3<-09i1i=pGyTz|h7GPR{OvTA
z)QJYK4Cf+T5AN`ly9GcqNgZ31&E=BT`V=f^lFasyqY^w>ZR+@9>m;NYtnt;`AogLs
z<%eBD6o{`y3x}gC6ar+%-L6^>#L;Yz9QN<=p7ut?>;Tv3?Xp{U-i_O?GD1h_11v3@
zwKHtO+c4_xU8+0}K|(&F(VimM<PmOL2m(yr9LxjX%+Az`_}9!y@*X!bC)j6_Yd1At
zMfh)FRaIlf?6FZ}@sL8FmOUywoKOx8Omv2#-zVYbhf+R0FO+Z_ckKwZl*1avhsT>+
z>uldz{+w@Puy-!zE;<Bnf?-!q%WT_cN@p=D<TVaRG8O01nUC9A<>v6@{f0IN{v!E%
zm$tPJ(EfG7q*+$}dV-Qvlb=>w+vIqWnMNDbxNpE~2Waz_*#DHdbj(=&x7aiBtWy?R
zkS6Y))7lz_n&xl#E7H@+`sfIN?@q@<V<;PZg%v+16>v#2#{cQjOZ26Up!ba_64`_V
z>ROQL+&g1>A6l1T-^-K8=D@4!<_>ja77&sD#=&X1`OLQx*W0WqHjK6Zs)r-wkkvkw
zQMigQ(&t0`iyS>2<(PL%ta;so;pLZHXM$H0zraXIw%*7N4qm*29b?_<)8btPzCm>4
zlgw%$t}J`7ANb6G5873>dO<e8L377;y{PaCvMzG;#NEZBHs>@Y`PUZ1wTSoB*X(xY
z=%$kD+ouk|{feEElp-O4#S~^BK^Q_zDu`_?RZAE)nOk=+a`E@m-VmyA7kq_AK5Ue3
zC=`T~2sE@QDqK(qD<TySj|q$@ECuNzkHytxUDcDM>(llzYRFpzywA)8b$+)3NFR81
zS*50CQ@i5;7p@8)JfNvCv#1wHQ{ndIWBFH){ky*B4_Wjbz0!NLpgwoC+HO1WDm~t-
z0naUE^~o|T*NWo@8$F`il~-QtaA`r?yUxO@wA(#~kniwTXXj&^<*KcWE9zJDa@A**
z6{C;p^e^2(^+E>lN3KV+jJ<xJ#7R&PxdOi~-Z{W?XDQd6rH!NHl+git<0`Wg8`IB1
zzz{506Ikcb_G^?z1Wk6urV4K~WqMnUOqwW~a+q4qjUd#ZCHxRnSW?HCy_`W@0B#rs
zFlYroS>O)|Y{UUBDHNMG_{j6?0te#^nrJ}^ncQMg4l0C+>`z_VhcfR<L%7rtylA$#
zIpJ}5DwOqPbTa32UIuJN)wJvhYSTJOf;XLgN4(lpoZx>C?x7jzq_o(fL#FwjT8ST!
zQouwhjwGmH3AAKXI4y=ml2Tv@2aQOUdc**%72=HAlU<cUGvgH_wO;T2q_>=E>iTB<
zh|g?-V(G7ONU%F7ftM_So#=ZtkyhJX@nPSIo^`<8Und7ObDM{yw12G<1wZr@4160}
zAD_1DZU6bb>epX2x|6#G9ve>N|GFKKmum8?%*tmg$GMmctiE+y_FI`+Ur~!q9Z>7*
zLVR>pOp@Y!Zsoey7dwnjdS_4Xn%exxo7~L(Qdibd=y9BQ!hE^$=~efZ*Jt{wedKj;
zpMFZ>ey7^9x{`8RxiU$;A#lg>Do=|R;Q}XB<}%+Z-N|_g#0p=9LwCw8h+-lF<MFl1
zI$}%OPEi5hg3QkH8TY|mLN466YAp+{O=ZkF@Jc`Ru5oKWeq;3T=1h{=vB7?8614lL
zBf0Zx{dHMgbZgT3R#00R-$Ms9zmu&uMAI3vf072J+$al4E9WLSp3NE1VcjonUpM<s
z=Q=%BkJ$!=l-g2)KQ1{WOz?um_D*JkBssTNR^*mLv<aj%es5Nn(oK~c2NuWdddGi7
zsmaZy^mO~pJn6p>)9)~M2tG2|FlyV>v~nMh{rh<7VXOy8ub)4;lPOQ^ze>~AwC1$!
zP`|sbbBVaap;~y?N)VEx#q_a_c@)>o`KFN;&F)+k3NQG*<d5@Z^@pN@Zx^~oJ72D@
z?{SJo?9Fs6<8|-atly~h&2W~-PsKf;AOCsid(is$B0+HK?ZSYA5J7*_Ko23+4m=IB
zz;Q87inwIN8c+n;U|@Z47I^#2C#HFW#f)pkrT^o_foFi^^IXlZKw97RKkyM6pb>Kn
z@kEWXk^Ei|Hg15$bSMH%)5e?jFA1#U%mbTlt!@W8E#-gM;Uv}b$$B+!dwVJ8nZlF-
zT$Oj`ROu3{jmf#cIpm0deQqYqCVzAARB7m3EkLgG@Q`8BwrhaK-=Bm}bOt&G+e|jK
zy?N_F><>~{301L#9L^j@Cvwq0hkN6Fq}TAJ5vZF!s-}`kUXk(u7XE-zT?xrj1`k@J
z;b;YnAsb+Ae;N#W7Tbz;3~M)4#~+$Z0xX*ylTX51wwGokw&N7%IO$SRO!4iLOs(>}
z9@0`VX<!mXh>@mj&rFAYnEcMS>e==-2TGtQtvj;WQE5tvl{6}pc>cufWNdBgf(F12
zf!y#DbXY<hHgF}2J1#nOVC7YIJO1uU9?B9?nL7vZglQB?8Wkpp;vv!=(nx6r?VH>M
z{|k#McZM<v>E#Svo%hp^Mdm9;G)@^~(l-~)Xk<(RuDW8YU9h(MA=u@<Sm&$Jc99aR
zh{*(QY1$+T+WQNT`RIA;$}ikGjszB-8nCT!z)gh+h~vMN0wLXWPtGWSiqa3-bAM;x
zj|qWEsNzbj*ta*?PY-!{#8-slpU1p3Up00piVQ8zTZ!k-+TBUBc%gY*f8|~UR-uh>
zArzU?qb-!G51jRG)p@t$T?6GXLKG<?MXF&<(1VgzLsO-3=;ox*j8%Gv+s3i8vIv{y
zzgXn<cB-4-XG`b8*YEYfw=05|9}*JKLrITHL{)bJD8+e&zDW_0ps6mv7b3&KMn*O=
zk=CFzYN4Bgog16xijgEpDJx^*2a6N}ZK3tbZSeOe%b3Iw<ZKaVN|!j75)&}*L@3D7
z(FvncLcl@?_*w>rZX}-2O>JQr)|ODy$PnWwyhp8)P^BKqc#Lh5o_02Ps@fcu<8t4<
zYyChNu#hgH^hvXl6mfr<BzTZ_Ueg5Dc#@=;+i=YT#^L}-7WG6~Xx%rX8ic8%z`Vlz
zvB0cf^HVA}7Ck#Ois1Ri-7g;fL6e@PQkXUun$cEdf=ycriZmmEg1~Gf<BYg`ge0Y2
zrgbG{^2pdB;VbV-JV2k9Oz77NZ0an}%T>oVTu?!ViSw4Sj>f0Js@)6cohP4zgWwLt
zFk$-a-OH11`jOrEd5zuE@5H)_pi4X{QvydXB7^or$GPX1{$+W^*0`do;z_<W+sF0I
zl)uZx{bsPl;D+@4j?-9(4K@*G<OGLqAzZY1CZC!0uGX=Bt!A&Ec|yk<ABoCUn%+lm
z8}IAKHW?dN9fwz({_!urXDEC9{%!vdVm6RDI>wnLv-#r7yxK-cGM7g%0%_1J(~C|=
z#xtU70O+K21eQx?ZF56t2A_>6RMH~HCSCcJyTSDGE4VezA0(`JY!W~koqTVvv%s&g
zOZ>jR_MvmRY;do2BI~Db9mq4Iw*Pv_VId}toazJWd_L03tc>!<R)ajZUY@6pLwtN5
ziJo3G>z6+|W@aak9&2=;4!Ke{1a#}@t#gT9_!oY7)U?UgZqH60n^^}>-`?uyKUbh6
z_F_;V2PVVNzke>S!N)3LSpBfX*)Jha%x0O#1|c2J+w}51e(V#{B5Pz%&I36Vk100I
zkZb>Kvyf0go?Xw!c;5Q})lC?0TRnL~lj?}8CDcJS=b2YK0iUy76|Ioj`Uf{-M{Yv8
z?8Q%Y-1iki*|I2$nr*@5=Un&wgsNKd;iAMT@5@L1xE6PMfy81{sFEIqbmLJClpOJ7
znC0A{xwR=3O{lhrAET}JlWz*!E0vU>LzB#gk<ORjIEA*5e9)imA3ZguZ<t|8?IDq&
zxK`#cA1B|0Cj<)A)CATMuLBvC7B-zn^?uE7FHG1MOI)AM<^bF^@GzjHXira1sj8|a
z(RLS7Eb@HQBtNeiz>utQLAY!oFligQRCps?Y<Uz~$_Ei)GnrwB%aD9bhn_u2Skg6+
zo{?uS1S6qCL<-h+pIVO+%6-T42Us)?GgX`k$WkIl0i8lRPDl;Hui35-08l<6|6>|T
z*(6=sq9xfe<Y_e&tdbRSO>AoGaO)=xDRhMsd(lw2!D5ign)juk=h93c{hwKYQ$nVn
zP|`*dVI=H+(x*R$CUEQ8-#5j#Ksr;UM%=^!(wZI1_9%*_T~qQSr>N+l82PcptPGYZ
zs!<5v2W*w-WZ@Bn&>1e5JT7G(Af~vgu7pYnW7#bX(kX#hp}8}Uc$rQ_5E#+P{x&p3
zbSN$vZDXau6o*8XlV`F2s&rusS?muqU?%u*6~S&9uN<MfQEh2SH675p*W>*1(%i5z
z1jzNYZHO0zlz|G77^N@q*XM?sED-qHn{h3PqEO#PcK<Dxb6H%o)(F2!)o4>UWe6|a
zRgf$nN)%lCB~l<(<&1qWkQIjDEw)N^yJ20C&PWk8J#oOuOc>LWBdnxc78p*#S+*co
zfkpNpgF-+cc5i+t+Uw&ewxoFZ47uHS6msWcW;IX{7#xvXTvr#k)}LlSZ4xgOj0%}4
z=u8Dx*~q=O*3uXaqIg9=KDrq=#1>Nha8u2d(y_Gvo!odJ;S2(}E`TH)(osLrL&>-V
zJGE#-$I!a*CtyORawb+*nNtSAc6zM2z(8<VDo!=>IY<<tK9F-hE+MFnn-S4&Mf>k;
z{wxTyC5yYXn0or9%go=pTd5CI`X?d%{OOPIan8Zu_e^~#Ph5@g*3PQ%5^_%DN{X$>
z(^7Br10>sY{YN|0sA#5lm>DfHA~J|T%06BIrOwc%A=n5{P9hB&Sx%v<P(-Xiz32Un
zw@69($k<zI-Osd=h2VVWYbRetg?Rd6)TPLTsAd{W7yvMwn-FU;SUE}<ieiQD{tiij
zMwG4$Lewx+JUWMX%poKv0akZ8Fmi|+kkVfvLnlU7nE|hDB?iM9zsP3f_JN8bt|}h@
zhC#P2r99{Ng&XP$YOvNAK&$TKYAF{$P4s-MNJjk8WMB&p3w|v)ap;&|O9!%#*8eaR
zOQhBqUnKxj<~8jWA|7QG@fEH*YgpNxfD0ZBm>7Zg+&=N5?6)<8z-U|R@GP(w`lk%q
zKI=p-A*Y+$=w~Jpmy$p}zkj_*rOJ_hw{9JzWzVqn*kZGcV3_&H{(Z%7`<wZ;a`px~
zp80P9;PJvWcmG{GEb{$HaM0(DV6J6LM_0(=OxZWCTTRPw(Mz-4k(3=J`_Efv>pyf`
zjOs)+l1X7ip_k9@;ZjU+ZZ&5;9jMC(>*T?2U+r}IeQ@<CzC%7i0}!szj9T!7E}zf^
zVZm#|DjP?0AwD~blWlW2c+2Urh0LAR%om0AOKF?<Rn^!lN(kd82&rS6nZG))7v^!S
z69N3&%th!vBo<Iyw-pR+&+e3G{Debw(9Kca{*N-KL1MdF%42u}(hgd_y<G~W>EzY%
zrk$cun8#4~A}ETDq_cwlw84a`O{9u^A<pCL-xpe>sZ@0T5s_sUL@iYj#WDBV;_@gz
z(0%<1Nb}vySnX(STW%}$kktXV>A~{A*j=~KU^SbF5S3f1D(dRdkLTsHcYciS%DH>}
zA!#GRrN@2ssgyYr7YJ1O%su({cg{a0f-W0kw^D$C0@wKbG|$-h(_`~rSC5gv_bgZb
z-xr<L@lY;!hw^YVqksG+`9VBNBtgH4al|dp<EcV3(y>*fFh2W^*y-~hju-Qi?52Nc
zM&NM=Hih7?+%PCg{(Sz{`A=O05_s~1{kH!_Pj#0=i-|lsNnl_O(j!h~um+1bDPqf3
zG-zTo8VrHdlek>AxNmWrIH6}Bh@X#;^gTy1;p1<<MV32l339k@qJ@F##(vL#95j&-
zNz5`<w%sb;*=km3(`9;HFFxq6#3%_7bD+B@LOcZd9_6~&&)rCeECP=DsGD{!p#@=D
zz_<F!+pMDBj;@$QG;>+#Q|(it=L*SUkpxqdMY+W--lXkchZ(1`Twwy~63fC~S%kTR
z<tPK;Pc8L;75HE$mU`nND&>%d=ztP}#!AY$EJaV3_(<qN>uVU2W2u9bt<itOAV#s3
z{?Z8hVE%Ce0GZY@+%Znr2Ae^%PiB_+<{ye^9RG!52of7Uy_~Nq9U0c~wg;rmYTZ%k
zarVpFsNhJ8)i<<JxcVC1X=eYINqr?(9xodAh{KsWe(fFH!Gn0g;ZPiXsgmXhUP7)1
zI%VK^G=c#GS+#RZVoDN-uni%rsC)jX<39)@FUlw=pv>@qJjudYWgv-I-dasD=D)s-
zp9ky50ceJ#uC&VM47cX=2(4ty+Hx*M4kW`v%E7x0E5-~F!hw<Vb+Q!ReOB1~T={a<
z=@=Fqejo=Hgdxb^849_i9j(4?lv2`Qb|3*lwB|t^v49L9AYQzBX}mJJ1;#$H1ERgb
z%8ZJFfZCE-tr-YOyju{AT%^1%njzqPJ#s|>Ob|tb=@CrEmsS!>a=Cdfj&HJFt|Fte
z0`af^lUIS2{5(So;}YSOA%V;pE7C@a5JbE6XLinK6>OuPLS^8QroUVdTFDZICw@5y
z2CLEU#;2tOVc4lb`tXuQCJJC@gS?6|Hf5nP;(%2SkQO?K`?Mcd+-#DnD9j6rwgRjc
zs}j<#iSInWUnn3c_CsHyak~P=1PNALVST@af)43$?`4)`m_%En3%(Box`*jo7V6?+
zuY0SoBZap53q^-2w?djDd=_v?aWpJW)Z(H$yV~=}84`mkJ)cbRzV{g(#dSFEFHMUq
z^nOj|#R5df7W|EBqr%I}8?hMBpt@hC<2McRl@BIS%4jwXoxi=r3yp#-tTEH4<^ehm
zzf(avuiG(`S*I!1QC5(k%WEMem44N6QSQfs=+xnSNNZw5E9|b2KwH3n;A}9g3Wm`z
za#ppan*p5dnuAnjevq<|q7F**=bfE9tLIKPZMumr-Fkx+yH!@xBhpt5Ql7l)EF8wV
z#r|gquc@;chLYm(y3Y<|;DUtLacWI0LyO-c9{%iK`gx9$I1Z+}!$kzmyIcA(ryz_V
z0jjG(eQ$Y&IY(vd5DW^SZ0<@|(=8UU*Vt41`_@zrPcn%4LS9ND=i9+Olk-|xSX-N0
zp6}NQ8-oW@OQ@8~LzBdLuhprN)v6YY6Np8~<dkEwr_mJiv?x371p6EM&SG~Sut|+S
zZrY71N&2M>nU9@HB}=o+6y0w+^yS>i8)4B<B&`i)9;}k1m{EA{CWeE|i5dyYY(MwK
zsN{2`6b+1<$18A(W0cWEtQV&At>#?8v67Yh=a(0U3pyo~@yJY2r3k|~<Rep<Xkd^?
zLa3!#IVZtCE`!P6N(}@X_H!m<wx11I&V!qcbsm48SPB@v^vB!pMjO1+!3L8^BnYR~
zI_cfQ=L@{q{PDS)7JE=KqHxMIc3<x$%*+C@^b9_yK&;&10C32tTkKxJ&qExciGz5d
znaAKZ=8ddQOiC|;Ov1Qh10UjJ!nc3F4Srr^@o3yGBlJxCWy}U?DUd)Ss)PTu=pbJ7
z5ol{67EX+(vq=>_-B7F6h_|e;$kP6h_jTBa<Ni0>;LW`L)-%(NG6;UGAMe2%#ZoGA
zKE-SP_vVyKS~KQjC51Qkr#EpzVH115r<((b6S5LMls(B9n(YcCf1ae-NQXAL^P~IT
z)w#-Fxn7GTcm)AOMIgzANeK2cE_5kL>vB`8@-sPmv76t=kZ^*tj@eTVd%Vs1?^>$n
z=Ev^(r{F^w%5#Ba8mvOAzCEOUWzkA9;U`6G(u#sH)!pfwkJ~`Z#7Fqo2klktF-D;_
zV~&T<hUt?c&OSD5{w~z(7{F&ivG<+%WzU<vJl}z=f9%FCyM<PV4-W{mN^S7+<UdQG
z)Y=K3?ipZ;MC~%2wcs`RCgA=aP*z42$9qN-7C$A`(DGyWH-`w#(Vp45UU#Q8)(Eb|
zMN@U}<doaPbI=Rocfp)_lTq_@bP~o{{s)f#M5o98r<0#qqRX0LzPs!jJ19*f#i_OB
zzgGoD%H`iE|5KM3DH?Rw=a`_fm3N-GM!$4Xhtv>b=B@t14=HH=o8BEqEm{Jg#S>8+
zk+lC<95!-X?qDnzmLxkPUug;jO8SHld_J7Z3Y-B5pZsTznkJbJ45)~ub_vbEvlr|I
z-In&=H*tEW3P*`YZjL|ccRp9J2peW8iiDeaOqO{rio<Vud*89jh}<|vzv`Vkj33Q>
zJZ?E)9<LvFmA|P@_*^s$J};#A45d^`I)6qULpXw1@9cX4kWoc9=QbUeP4xfCeO^v8
zzefB&FDy`@-j90To+mVV20#WIp8Gj~R<{*euirkr;Ol&R6Ue95^ZL)BcX-&QglkpU
zK6xydscppJ<%^VD6q*1Gp*TE*d+)18@o49w^$5)uH|<xa=1KZfa96F$B+nDB1CfU7
zn)(Fa?z{VXo!<*}R03^k31WaMOfBFfkR)ZzDU`?^SJXn1enC48q}Mo~cINpP!`=vU
zc1DhlQQR`X6Wj=Ag`MeO8#GxA-eTY-4W8Ow9&wwl?0D7PTdqo{1TR`dLk8sU*3T;Y
z<NxK$V6Z(7vOU7XJ{o-9;_&i;Fy^b>cQd~meLSg4m8H(Q&UkkNAe(#)(ol$eQ2)cm
ziNgB%&u(3C+tL`}IxXfHpi6YJ)clo)AyS?He`q?(peo-cjMLpA-3`(q-Q66JROu3s
z?vxhk?(UFAKw1Rpl$4f|ly2Vp{Ab=7{oruU$UL$4?zPwMPh?K0IGDbMI+PkkEJtCt
z!2_@>--^nVFgQ3kjAp&UX0Mc=G1LiZPUS++QA!Nm&N*so=DRNZtT>YlsFBgj(j^Ta
zcNlUGMJ`>br3^GYAr<sCM|2BcIXd~92o3s_dQkK%Zt6G~Fz6sX{oJP1S}CDm^LGYO
zYNvAiN(guOTniS=3-M=K1#)KuI{E0+30Y+~9A^L&s(0Gx+sqsnGF%oW=C#5AWHKHt
z0&lwFL}~(8$1Wk?2$Ca9`M9aMRi*Q3UBKU#xLUH~R837cVELp2E3FLK@>?Y%)cmA<
z+;Kh+fN14THroUtj`j-(Jzu^Z4`0YwCoT~Xh9;wzm7LlHiR0`oC2QZ#JD3`ClrK&b
z>|!X};sir-Ui}43v%p5Le`Z0!5R<7*E9kEZ7FM@(2Qu9pY<8b*o==#fkbE7KEmdD7
zV%sQCc<Mjrd`b};h~5k}kAtNMcDC=OTX0XG!MZZylkkERVZT4y{nXdX9*^kzm40$^
zhJx2;*2{4DU0LT$+{xO?xo>-Wd!1ltqG~$VN2Mf2Lac~vteVs)ly#8z<To}Zxs;6N
zQ3Xb4@LEG+*+w&XBT`1v;PZKRHNI#+J=~Z_Q%O+TvDehpP&OC&X-erED`b{7xi`%$
zt?p20N~QeWnloM*0GQ-=r<oc;CamlZHN|tMS>FOi)@hb!W@r1evvxNT`fY>9-M+f!
z!)bJzX3H@vW#8q0qDE7iNsqP6f$4PXn5~@r*T#$C)fkH{v)Ny7|HEZz$p}W1qfbcU
zN+l+MME8q5T#b#R@jyTlHuY?j@*?dYk)hR#xTHRl<j-S*t2y&=)1=aTU=bdZimqGy
z04f6$u3^26oCs&<DJ?Wx2FhAcpt=+em*cAF($bR8l$p@3je~>b#oko5+q4p&fWQls
z6Tw)pWZg)=-b{{@bR;$w3w#i&Oe{KfKcWej8QyL1p|kP#ndV3(!|nWb;XTTsSfKyN
zj9_N~>*nx#w2s1K{WOjoX-?B6S9jV7<+%lg<}(b&7J0%P3WB?-*e9jd^D7bk4jZkH
z;wSpQM+?I^o#qUzwaOkkes#2xoT8xYlKURkBZFOWRCq&l*b;q7oG&d#TkEsEF6tKq
zrHWr*A}AJ8UnJ!6bHBX3AKB#}@ugdDHon{LWYzui1v3bt1W6K2dG6QC8h-6Ezes?+
z&JuEUq@KmIBm8I&gYfOxa8TuQYCN$*J_GGYOqxhuCGoGnQKOp+*3$uitMWPT>7wkA
z&#$aHdlMV`_jl{3%Qr1#d`;n<KWXIBe~`BJIWcP;w#|z}7j{GO)aMd`;&?x*YHPOR
zGa?#*Aj4zv6DJ&lL|RXesAwQs1}*PE!)lJ_dBu1cE>7$pF-(N22ts`Np`o7t3Gy4K
zUc~<K!mt6Fa{ccBlT6aE65p3`{bdbxpvnYYIdYWisLtRtHBLOi=<7QIDcVY|eBFPB
z7M=@(t&DyqOpx<EF>I&r-$!iF1pWiao;I<_)$_@y{eE80gYsj6(9Ak>?a9JWmg~a`
zQX~$zGYU-~bUI5)6<o#E`|IoTL5U|^76y>w)AC{o)hQ0WL<B4D=g<FE)w&FjSl8_y
zn{#0SP}mNL1lFLq;uG+xIt7d*07TFzZZ1l4-jXMr|Eb;LfcWP+HwEY9C!awqy+daR
zN8U;HckM4^`yoQrr?N;e9OugRWIdAI9cA>qMj{zC#_Wi-S?3gzBR%&YfBH$uQJY|Y
z(@00@K^@HN2Z+WDQA@N-wdH?-ffx9OM#B0~P;KWGshF5pSf}+$yZd3s`FLwTjZ3-?
zdve7UzLOG!ok{G%?)Soqt)t^^%w^WiS(^IB{o1p*NE|6*(4^34WmGx2fSFzwKo|pb
zOboC1?GRjykl%u0tk4~?d^E9C+Z&Fnmv(d2rgPN>Pm8m{>D8B-e^dpkeW94x!tn5S
zTU(KFlw)tZAH`gEN4I$M?g10A=qnvtVD#K(v=qX}vTcSEMO7ApjO*%jd`#gHsRQbj
zP2|RK9kRshDCuec{K1U#32J6fch|?a^Yw!_sO0{egHqq+j30&bM0RI7&Ms&}>k$l{
zAD=tJoxjEGBx|cP47hB}ca86ki$|19bcezOoNd{`K%;h9t{*GbgrN(llBUmQ&)@Y9
zebqJ$whOq(I~2AeP#6;?tllOgGDhhslejjZ4BQBVL;T&<k%T=a7>bHa(X$sp)tCCM
z?Py>bN>^wF%xV&_2^I0r@f(_(VQca4;ATIMGk)S)c32=~syIBt@taRX<fHcM+KqB+
zeTnT%xxeuf<|@Njq`->hYcqM_<5^eyQ>T4*hf?Tln5hyQtN?>rQ`vYzEM2T%ww|%U
z&~Tav7o+Lona8_8cYcv0(dk{pscEp~NugZI^7gIYQW?f-o{(2@a<X*B)BtM0MG@5*
z$Wbphbuquk(_jAHRu)`Z%Cfb6`@q=ckrGzZ?D<lTMb39{I(Yo#lD|q2FxZ{v>eh-o
ziABq&6b@B-e`h^?t~VbHj=>%04mkXf?L0sdhR>=n6%<D~PvXAm@O4Gr+L}p!1WN{~
zAhcp=CYk4Chpcm^Zy~r((>nHr9XkXyy5`2GIUmo{Qgp=DxkeIgIE8)Pj37pJ!JpO;
zA@@!eew;=wRfBDdfw2hUlXM`bpZ$}Lp1gII&Noz6@N;5Rzf@bIX4GFrHf{}0uf&$r
z7ul4NXrfDM`{i7}TijokZS8ggC;-Rl#k;wp`CRW~M5-|3kbp?-k;D=Qzun=Je`AgT
z_$6?V1Bwq06eG^+I0ymSQ*XaaE-WY}(l<4^PcL<06i5%mQe<iQ3MHdCN#ihaD<h(m
z*w`BM6*91+5Yjdev?6SY>ba^ISx;Hf5k!!?pg5ap4s-s7+omWR7Y_A0kJcU8X;Awu
zIj*o))#LgWeS3<om<7PJtOKbQ3teW@ol$6P?CgqMZ%0Nuw*^ij@nJ`niHe*qh^1Wb
zT{4L{|1lyFPQLvF3P5~y{b41<!t#akCBe_iIwvW=OuOyboc{>->i&{qoc0!X#Z^pP
z&}6k0un%UA5(G$<bl>y0O>SK<J?H3oR@W<@$Z7e-76TWy(Zt+^k`+5bk}d@H0e<U2
zK8C3r(sn;QLP)u7Gni*pbDN0l-uaUR@zoGSGBR}Ho7<ine=0<&A(z%y1vr)I;>>Gx
zOB+pV?WVI{e2SwQ@!B_~X#+77W4Cz{ae(xJwfnOB;m0NPKu_{P{tW=Ef|s+}=#RMv
zH#Pg7*6cD7UIQ*HZ#XFeN5))nq;d2%D`V$bi)Pglq~jS0lV^2?ErTVuAbAH$Lom3^
ztd>+^OmMZPAa=*pqi!pQDWcbLJBO`m<v-JF2`u~0DZkiGwVGO$MteHzhf8hxny3t)
zP-=$AQ6u^G=k&J2o69Y#)635s(n<+h5`wAKHWM70MV7qdNvXfsZpP!f2KBoCVfMAh
zQ@(^CH`Tn`nBe}jtLq#q7{oAQXwXodV&a4kXsZK4M)r_UClm@z=NGs(FN1f9#54}h
zT3#l#&^y5;d6gNCY%(-)yp7^bjT;puNHY8z%Gly~^1f*<2uF5WM2$bO6>m#+xafR$
z=3$yP87F$h!L49T&*SCRoX{&GDQpS~3WIN*uGNhtF(rX--yp!X<)W2VsfNTlZKJ^o
zH?kCqUr4-4w+JWS@R_h&rj@V&3M{h0dy)&C@gNiGnD3(;ODH@hcOQ9geoP#1jvECI
zW?1J2HxL*cxfd_`K1lI}EpJtSMGfP}aB34POI=k$A=c5Gc=dFI7JI+<@}1WPNzZb~
zj>IX3p7G<`H!p5Fyj?T+?cSLmF1-D!{8%hYG5@ry=J)TjF})KeF+4-P*w=~92!LhL
zom6-{_6*D~DuTB?(Bv1;n~LcD@Olsl=Ggd|7y!WO08h^Z5GaAnlR03bAnrStR-yAi
zzV_?a4*;xgI~!)Js9Kg6dV^>oDVWW&!6Ts~EzSJC_TN`c!`BfniZj?%@3%N$m=xto
z=}PGNSnMb};;5bmIoobO3H)7ofhBZf@}c9O{J?RS2uwog;O@jOa+-hAZJapTWR>dL
zS`9b3&tAcAw;8o)(pTeOYf`^(#l4@XKx|Aq-OGFGzg&BMfh~FpBaH&+dE-z>=$`YI
zBnoBg^*gKua|B-TKt|*4b+|6y%U=(-2@23f=RW{>1@TM!%?Ji5A4(9Qu13463nv0>
z+1U))htv5Hy1#2tX#YA{k&-Jd+8}>>Sn3Fq-DXDyFbe138B4L?Iiy-giRaj+l)L>-
zi`(a$e|RF3)cqt4aYK4DRHo!4%v%y|vUds#^KjEHk}lvt3f80(B{7{yJqB@;-_~oq
z_y~J>*+9(m>bKKp?3<YoAFYSiM8*M;wLjCjfk1^LeY%0I@jX5IQX5+VLF};pnos<)
zcI{8%NbMF;_l*+giz$rRFr~*ey?{V0z)LsZF)iaZ3tjt`Osc@Wop0U-m_+5ZMD&Y^
zw|(ff>p$cCqv>r=FFp!*?!8_{uJiaVPuXoP4vGl00i9ocJ7e6|{(q~W0;Q1abI;#j
zRR%5gB%t^@1c-6J=X*Vq##?pE05)@;QW#unULQTnMm?G;ti8S|O;nWc^4J1!?TY0c
zdc@LlCzn#=@D)Kw1>S$!pEvI&jE_wVQL~$#F~mD95I;ZN5Z{ygp)3g<KTJ)uZC0M(
z8KNYRcRss~-`XLoQ2J(m(k^q^)p$Y=a`fMoK*2P@P*Bets4O`Wy}8&T7jzK*>mW^q
z7+eL?CBVm%_T0OEUxX!s62xR|JRrj<-rH0$gQMsnWWg$NbHx7p>xB?tn0x!IppD<$
z5#9dy-4-oqBhfze&yBM4u{-eLZP)YT$qZpv<EcuB)O;(I3&C}Zw5unqOx8o6C3}2$
z6pj?as&f`y^5YL&o<U3Nvv*OnM`_Oqz#V(Ht+x5HnB-Lq1dfKzO$?>LS>3c|3(AZ?
z9_!9;E+HvN$!k?pC!z0Y4esK6b@hvpX*?#wAHQ5919<vDqfv!5kAX@ceD;nZ5(h4+
zR8}i=l(x}dgzyX3>W!<86%OwJm-pG9KIXsK=CkEm|E;V9cqUmEsI<M7{IX3TO)j^N
zR|AGrPRrD7+R&HpbvWAa$uDA=3LP`>h3tU4eNZ^ZF{q?cmneur{~3lW;5kWR-g-G_
zWNT;lo%nkwDt<bbDO}d8YRiF(tvb9-w;H{E<wkE;A~m%Tt}93C=Fl2QYUIm;l92@q
zxhP5NWpM@avSel0FTca)Xhwun=`>Mr>qdU&pfvopZA+>JfunkR{Q1doT=ZO({PbY$
zfl;Lstw<rp?9a~x<`Vtp_k`^1?1JX44y&T$!Ur#509#do<#PvB77fjNe6?MaFS~iZ
zv0{v>%mMi0-+wnJ+m!6>?{BnO=l;_!bd6qhc-1s_adFeZ4)~*NtItuEBU&@o`AvD!
zY$<30Fx6ui+ayby?2KO>8g>)-hcaNZN)brl7-uuN4zOXyiBs8)V|l|@JlAqkU&P%w
z=~An~^%cAhB}|-~13npVVd)ZF4jM@8BncOxnVy1DFos_s_tgp~?cKH*BM16xJUm6-
zLHA2B5vID7WQYzeB54-&7dN`1bY69dFrlwrI$Kq>UCMUn5Crjw%{;1%3Fkoyg$f;M
zRH-*#Oj61`*gV~RNj1&A?4OEI4k8PKU@>ECXnOS-PZ~Lsu2zwQK!6Pa<zphfX<sNR
zx%Zz0=)OTYE*S#3Y_H8v(6!7!K0t8-2XQU{9^%s=K$r1vyJUcNs%mC~<yI-}nV=fJ
z2RTo&3P(qz3TG!l2?4uGCJUl?ggN934(nU@aa5F)G8eVY&~~n2vuQ*_Thas3%AY0S
ziuX3^EXB{?bj&VV|3PayBgf7ZKm#pGHk&~GMkKsi3`cSE#;UvhQeeq2&0Kr%W@V5&
z0GrCa&2&T<OVEb(a!N88-N{y(S&ZFLo8p&6@3ZPqX*yg*injvff$`ViC3-5yB`gE;
zkMSY}?OpncX8HmPOILK8vD$!3HUPWR$WJyK&lWO3$g!BW$08Q<;d3TG2M_z?;xwG}
z>GIlBs(GB)9No7MS$gzz#ErYTKkf?Oa5S1N@;JpJqQBtsxX{Ny%8*;}L6Zn0z%%!k
zKA!957&;!~_UH?x?RCOI;CuDTJjeX#&i$ukQZn05Me`Mrg<IxN^#UV=Op4V~Hnm%?
ziK%%*P5!<dNn>Yf4oCfVqHbr$CW%`IPtu^L6q-CdycP6+FF=z=lU|aCoy7GIMw(f)
zEHp@Upu)$m)C;6O6~BYxk(Da1GsYS~flb9;IPg9tlioT|x}CYvhNL@J1+L5e$B)Q-
zG8IG`S_wpJC!&p>jL|Qfowv=STGaV=u3;{1(aot}zHk8bB{Dc3KUO4{>`QmT<-vTT
z%@i$^b!si~DAY9P=I&mQYz8Qg&{N!^@7UZg=Qfp^S95M2@!?|WpmM%Bmg;%$*(|!C
zv2kYi98{l+&U;u-xo?|uc<$-Z(P}xp2d6Xw{4k+Ou9}*f=)>vvH!edA7$iKQkcxgN
zxer9wSrHgtGLq<Qu0M73EYegKvDDNT6Rpe{n(K9F${wfQVT~^QQbDm8Qy*w?R{!ZE
zaUrq8e|b9omHnKe(y1kolI?R=;dcU3^BLm@+Nx#eAC`KjC&ss)ac*3to*ClZDkM(R
zs7%hkJilzVg9G=)ntja2<X^k5KW96|p1i#k8<#lo3q32})YcGBsp|G^u7&TN?3Kv?
z0s>t`90EM2*uCiB!w5IfA|J{5{Czk>e@0+GGy6ubg@EU(YPB-EG<OkJblTHMZt|I~
zm^ewP<tard^!Ln=5>-DcmJ-?*$mMbdModLZ!dA2KD(@x=j)8G`CAC~`ZZ9upalYuc
zyk;E5kBYE`eN+N2f36;`1D~U~&E-wmF~}GCi<NwdY<J%U-10*~nxN@6Q2Clo>@AI>
zdKKU$8W*><WgH(L2RE<xz+*$*gyc>=1f*o}(I3Bj8JnF?t9ubE>cWUjDVP!!r@>Bc
zDdcMOF3`Voy=8f){ySEP9UM8iw4i{!{zM`}l`W8*A}BmewuiSdDBAs}p1Qohx#Zog
z_lC6IY@)XvwRj^)Ng5fE!GJyHoFXS3ynxRMt@@#z-6VD41R9xef(oYhzZ3$VUysKh
z`i_xje3i27wpA;r!iSf-Sml0eWUVDV$0g8?f%<1}u?aDi!}4sFg~U55>}AaWFQ@o~
zI4{;lvwQ7STVWDN2bS|rU})0oPTq|IqvQ91!{J1A$CC_U%c%Lm;RwbPF=x(qK2L~P
znI3f~%YicQ)=q`lGP8t0G(&H4i;W}`@pw5Qe*cmx^NuDET|eNWlZtR{;_;!a9P%ki
zJOJXXO8eO-9E)_AN#a&%)mHsw%z&Al=3kdAnpSjY+M)tQUZW3Je$gi;l>$elRntds
zk5ZT?=H4eZ;Q`(gith^u?zkfP_!UjI3|p^rk@w!!GJ6`pZEG4f4j$tRT`g>tS5ayH
zsf|iXm%Izx&i^P=S4wfZ=G--c+Ee_ZgqG@8E&v)yAyf8Y$#`xKG3E8Iqy42W7fwFQ
zR{eI^fX-_*x@UO-tMM^~9u-g$IXQUMV-{@noD%tzaQd_f!vK?-^Y#TWsmMoCf7Zl(
zY1|U^nX|n*zXm?aPLaj8c_{rQXLNLw$F_Psvpw}EZ3r8;dIIrBtj_b1LZ#@RH=$y}
z9NbtmaPc&@J2qVET5E!ltsS^DqSHc%6M1&*K#?rA^*FhOdt(2Qlz8Jp`D23#e}eub
znqiuT@E|4${e`HKiVB&u?Lc%+Pno({ffMug_>lMc&eYP<@bI{LC^s{ULAwnuL(QUP
z+FL>Mx#~V|me6%TvMMPm3K^f%O+N3ZVgudLt@O^TrO!#~XGvD-oF&ax$G_WhA(d?z
zdd2}6ps$E;70YDDUI3v_K_i>{VrgJRHY@~cy21TMD?)jx8vgU0G5CD&3Tz)8iC+Dr
z^V^f5I#VCA|Cdq}d$cUr+!7{u5HGzxO^_p`*eg;eF6u9Ar(09lwy3Sq*<glP%u<lt
z#QzF^cGA|Ch4_^RhHgr^zKx?%l-h_*2{#7=3VAVMa6YA?&E#S>h39Q}`NzQSti!ax
zSc@wD`OiAYF4}QA!ZK1)<nU;J4^-UrSgBRAU3Y|Fpdl`~!7o&x#jVM$&S(2?Uw$r~
z<;Qn<M=Hw|=ya?rVi@m9C8;})v0f7*`YjYtl{R~`7%HpZ-YCgx^;a<E^?{A)BV>*p
zy+CuTM^YEt*Xe4$F@nkVUdo^sLzTjD8vaxUZ2?(d5Cr$aBN2^XC32XW)w+IZy+8wo
zw4-yMe#(EB!%=#zhnpi0_!X!gdsS;W&Kag72goG^wuvvKP~4>`Y#A;uXyr{<2aK1~
zHyvpZ48sJI@<XN3!}#D1I{sBx^eO~-T-%q{ccGGrMAzHUD=}PRJ)arP;6Cye>pxiI
z=6u;%NYZ!+OIsE}4Kum<U@mw*J5w<cY|*ZG`%9FnsoqS94fCeB(@t2II>z+j=3n*|
zYTh9GiMce&hi^v$fITfg;>yW$^Jk`RuA!j~Gpy{7AZR^DJpo?fPL`w9_HatI{`Bwx
z+dHeW9FH>__WWYFX+b$FtdXl!-8qhZ>PYsYfta0(87tGXJ->tWG_5TTgQiFc(F<xS
zp-rb2!H@uC+(Ql(IEU<gUVfv}vF!1)UsX27?aAY>+3Lrc#0-|&n=(_DwKQ9!NqI~6
z@^Ty;iUklEM01=7f*^tjVsE+OG%Ss{13s|WFLR7U>{N~=6lNM%8Q+x{x6oEhl%WJg
zp@qhe3hDhq((ot4wA|xx_d4HW)J%wzok!ci3vMSUz<-}qNYp0R(2f7~(`U|mMxz#c
z)6?}{3^IY}-VkKo?fg%Nty{6smZen4ctJ(|P?`ol90kFp*S>$u8}itEdi`%7*x`{d
zsH<N=56z(@rbZx^K(*{IomYn8C!$z7(+3?L90xBUiq#hfgZLUkQ4;|>jW7@MYV&m`
zPg}SsskQp!NfYA17+v9XGJ*&Bg3l`>@;EW##yW`>@)LMSQp`hV?-_1zy!RZA%avZf
zCT>GyL_a~#Pt)c+#Z9qaXv*+?Zb)!kx#SjrxopoJ0*f$5y02XW4ryWLYEL$wB)Wf&
zb&HeG(N12lo|mqDtl*71@BWNOTYi`Iev467oI!A*t{`Gf?1rHG;TILxG}&(^|2^Z@
zi^{xPVT421NeltT%hAC0F!XMJOa)4BdTndU!PyLjc;UT!wD4o$*pRrNO9Zl=l)nU&
z-OsRbQHiw_lykn?vZB~P$<PT$ul8{ua;&fpRa$qIc~OrcY(@t?&BDDw%2mdMf}4j;
z9L3dFF}5>|>%OYkF)`e{R)=Z?a>>dJ;aC*I>LN$|KvOViJ3}0Z6~m=zpE6v}mb3fh
znwt0gpbP_urE7JjAqt%Vf3@?h#xnKx5B*XB7&iv`7fR69Gxkpb%u?K4@|BepsQUeI
z8X-|!VGGlC1sPW0*n8PLnCpv$>{IE<`O0ZR%oR8nx*KNHMd|_9xZZL^QNY1)pj-<c
zw~!3yhp>=i*%}c`w$UIga`bkrNC(As$GM+)DZaX3YEkSR5zLpaJJwvEzt|vYm%E{m
zLNtfu8PWu93I~#r)@EVic<C^_kj0L2(UPBx-`__=Q3&}UOOIr^x|#A6BB!hor~jH_
z)tMEyqbq>I20(L$vOR)7R49FQG}*|sOorEQ3Lwl1HJvQSO|}fzha{fGBfb0Gl-ViJ
zsm$POH8m;Ks)+~Bo$B<Ban;6HN!y^ShQIrp1ECgNk;*b|X=lpahYiX0WnHbjL^W}K
zW>Y5pi*<89uReOD;#;{h<l3J!f%hlv_aYifdo4V$orJGXpCdw-dSAfK)o*!Z`V_F3
zE4eaO^$*$?TX?hNe64u_qudFa|D4Yz`=0MFN3!VESVM)4W(`B|fqSw&etf4np0B|6
z6h|woZnyv6>?;8ihv1~BSyYYPRJHKSONO4n+N#e{9eS&E``@2J%OM6z6*l(OWhEtT
zlCB0uCApQUN!hM*8FK1`c}(9O``}p0?aHm^nAA5atG>{pz$Kpz?detPgJQuo9^$fJ
zl9pJCn)X&8tIc(HdUZ7x(v~ns8~R--?;N1u+QMXPHqN;%$4MDp<DwoHe;0N<jaI{e
z({=N3x;<1uGymci8$0*nY0D_>+9mbZYT1j`g{1}yZi*UqeS?js``k@CBEK9PkuZJ}
zxKGVEXbVjNV%FKttm<LJ=*EK2fzG2-REX3{>>png2f0zl#6{QjnTE0AkV~UhE3qiW
zBu*}^s#l9f3oAqLvStFq(yLd#%OCblx2Q)ui{3w|gqlVYW-9+p7u}YOPEssTs$oEZ
zyIOH%RWV%BV*RtYYkY>KMxbjG9<zX$Wx4gG7|S9^3NM*dF$aO>cBUwf*KY15-T{#z
zvtqT}UL-PAZd{Voi(3wEZ@dq%p9uR&rhCz15~H@+bh-WIWfO%WbY7$;w_}VUISMs8
z%%F1SsoDwDjrTHH<rw6T<ocZ{P4kt2^BT;3Sz~qAZw;J%(#T%7Wz*Jd@9!|$J)LLx
zJPDzz?rDpp#Jws0xme0kpC9s)v_6pP`XfFQ1>C{2NRLvUhJs-9sk%MeOJ)7+q->XI
zqMMOMHsZ!X)mX6&47hFN5r#D5KK8G_rG0%xAWA@sA1QY(LaXc*b<~MXRKh!PT5>r2
z;O6#HW-q9`>*H}$8Ivqai<@tI=f{w*$lLBRVDy3;iHY#w)d$+<*u}KK&rsaf?n{<;
zWsKw0aA7HRCcSWPQe6)K-?_{TPgw6Iya~u#V4x7^eu3?2Hjdo^GRM(9F=ovw$X&d5
z?G~oGplrTkV-JEkg|gDLeLPBq6RRtpjgq}~{A9fuN!ptLTf$F3{wRt8YSi+(K^07a
ztEpr4b+}LXscD8p4Nc|=wadLS*Ak<M#@~YOU_;9LQx$*VF2^KGFSMG{fu}{DTQ58C
z>6YC8VhWMMEcWk%Z`4@fvJb;-vv{EP(2c`;Uu!194X3`3se&x~<H1_dIMKu+QK{ns
zviSz5eXd<=%soPk>KUqu3>SOqtg@!EiHwJB&-9tgM)?G%7ske?k&=jtLv>o;G!w9-
zw_R;mnH^pQ-4Mnp`rUjQ6ukYDq_y-__=jn;q|qAo)5B<z8+r&Py?pRcObdyFQ^FPm
zWv=UrbXb_{1v{Z5XP|khlha&9qdt)SG$bBD6+W}H^f$yDue%F$(O#BiyP!$_i*25R
zB7*ri-$kRrVFZv1+bLDQ`d~v{$C7}nk1BuLwzhFnpXzdYLee_L`FJLU!%!)?INH4t
zH^QSHY>9&x!w1{_c<Cs#pKM;&LQ{l<%w|>`sc5$MH_G<Ihn+?1Jj>g&ZNJ9@V`OH{
zfe<9D?WMej>}I?9&1`ef<mPlidb;}BZMo)OY9U1wQ0ubRdHoOEZ`W1tUBW}$^@^ss
zT4QW4`7n5FQ5<V_yZM@}iNJthd=1eAKHO}l2NuLEXs4MiwRUI0Tpmume1hzVdFP6J
z!+^y)@^4*2v)!wP4=cidmEl`=1|`ga+TGOB(y&*{4(=c-3{-ppDi3Lm9XaEPGHIph
z(?>lshEp9nIP88rG`Ub0A&ifI%ix)`U3ep%a&ant7O=sR6@`|S%PAXvy@(Ih{JL_5
z;FTlna$_?0#h^Deh74A}xiIMt1qnO)!SXJ7|6ZggRwmQS{D@Vcv;HH<L~g~Ts0MZC
zEJ8yqk3QR$>)A^%vkqgfnNlCTGg6vJ<;ct=g%NbSBf!cL4+9seas6M!EdWy+O~w&1
zcaREQH(HxS%dkv+UOn{xp~)dxmLIme=Lr|u7^;p(Fm=->VP4e}%Dg`%q)%lsOI7jQ
ziy!y%qS)43z%*}A%F2cvkpsb0iTBDQBX!S)h1X|+loa_0R`bv|Gb?Mc)tRnj+GJ<E
zvr#c7{u%4^evN7*5_c=;@4qGgq~ye0N-6(G9GL2u_=beM$3x#kV7`hb<&VOp8{jq{
z(8>nba9OR_o_9fSquO1N3OZK#)By9;7nEq`jc4va^CEc@_;~q?pWSIU=NcCQ{^R?s
zvv$~G0+J#**$O`5U1by+Vj&k%n3xz56IDfUPmhkONQ7MIU3Y#<SzBAzN^UQCpwl$K
zZ05heo(%A3wSggR@Z4ETo2c9HVSjV9rf>Kx!_dsku2j;E%TA#v1+&+@GRZ)qgs-ny
z{GFP3i7UuSJ1j9I_mwD{vbIh{zeqKNe3>6p^1~sGx8>XV8=Gp!Q=N*wZ}8ux)T-;n
z((wX}Ayt$2PB`6N`<PUxQ##l;Fzs)K<{e)JjX7)Ze=5u;)n`ID_$E+4(mJ(x(s>~F
zOV+f*e9>^x;ad6|F2-cg`QvCwc)R`YBG~Pzc~}CaFt*#Z^{I@<)r9Mki|&>H@1$g9
zEzZzH6lO+k7KU-bX#P-OSDB(k!XWO^QlR(&k1h6Q@*fL=76GBWHw^-tZuL5F1;dMG
z4jZbxeM<o*Y649$R1-s?UA<Eq3UgJ?CbceRr*UU+j-JIh`oXg>0F#Eb6s}hvDiZ-x
zmtgE-gQdy_2AP<To7A}{_!Ow0(oAZ$p(^*);i3*J;nk|IBgntxfXA76@v{JT*(!Q)
zkdW(6o!_lXG<T?}qK1ZslnU{n^G5Or{_-wLP<A1kBQkJ8JG!h3?M9tPb2IPS*hDJ6
zA)34Usm{Wnlnw3CpPhcko=MUVoz+ir>feZf%MbZ^POFk2RCktRa$J7CHbXBJ;f<#W
z+ehXW!%a=`@#-_B-PuotayG<uzYk=2q%+iEa;!@EV*RgTeH%Q@HKz-Tzb}SD{+z5m
zN?eVLzKQy4c;DSwRrzfKCD<u_mt>=`JTi(dmR?>9!CKCB3nxUY4Ub$4;0Nx1%gPkC
zalhvh{4BE>zww!o`5z;9KCSh7k11X(HJtU{apltez4`oV8JVg}EjLp0{CR#^RaI4G
z)wlU2yBg2VS~Mj@tH8CXK{lBhdOub+p8TRvAD)P)^I`~#OIvkYL;_1XQkVBTl#MSX
z@F6UTgGh0o=58o$2sfFuNtDAr^j==~9Z1j(vK_LcN>-)AKHkGj);e-h2Hcpk4$u5?
z5t?Sn`?qzyb+e_>gO6)SRvE)=Z@pAeh9{9)^2HRh>r~TOas9S5hfrR2soEMPDFwpv
z@&0!PPB3)}Z4@m}2El|xV|r9(eaG!iR+AOg&M1jCEU&kS`2YqOr_msg1Ky7@zSoTg
zyLmk~m2Wr{a1cUG({E<HKfLEp8Upj)c@rX)G158GWY~}j0+4*OuRZL%`cxRXv}EY{
zz=dIJZ*Kv55r59TL01wn6o(vmt;TDyC`A`K6(-JG!GGHJba$Yo?0mIo?K;A4wP^<$
z1wgj?t~O^#_5Afg<TSzGk3bF~SF(aQ;WWPL&(l7$5?<o-Yh7JDChhVUItde7etzP4
zp7<h3Od3!9q=qJ?elO7p*BP75RJk?bb8&8R70Nedi!>ADH`!(|Fq(SsDOstsb`TlA
zD4?<ULlhyOJTE(g1gR4L`8VeAY3uXwJd2JR7PG#!BGpln<|rxttx>S->LnF49#$OR
z|0z5_gM&bxTrlbsE>Cb2znFKY@mIdNj0{;=g72YsZh`!ie4v&;hNz)3+a2bf#@f@P
zev=LEegj6^8!QPzu7C$Gb-|5LtSGaZDY!GOpA|FeTL}u3+yq>Gyz^}S9Tqk5O#Ml)
z%-T%}Es_Oj<?pKZr;Bxml&@Z?xvA>dHdU<w9l{7omwP^udapOzXlah_SnKt9Xi$w7
zjqv)>iv|$bI__8B_f=bgQ*<;}EQ?aiXR<pm5KDNEdfuzDy80!nhZPSAl=TWc@xa})
zOCjK})L$F;X!p)%pP}9JL|t1OAIun--JEWSKHY3Vlc!@Bhd01q(g{;uQ?s4xd;FhM
z2&&-$q+}Y7be*Y8A%Av%R#X&PyZgZasJp;9Kt~=VZTGhS4NZxK?{vR62p-TYQ6wMZ
zqK$w|?gYcxd4UJ=BP8Ima<5&S8SURJ#0+9d_bCi1`elW%n5rwRw-HqZS6<0Hm=@fU
znghe<=#$nd{5p%@uNnvP;;8Z9+|aJw{RmE&G0y>cg~T3K)LU$Q@2gw&LW?_kP9^J#
zKys<}Ylhq)?A0~p*K$JilE^d+Sppv<^ND!@F^7B(g=MN9@gUumS)?SlPI2LUKr3`N
zP<U7l>C~}F$N^~&z0pbQ>ih*wBryY>b}zw@^_L?Hl^@-QcPdx4+$EBXGwLlLYLg4C
zCop!i{jr$xJcMAZbG;d*a89d?AETgiBgrW#Q`6IMlCG}2%PkJ(ph*iF5y0N~XAjY#
z@h9cHIM(-g)d4O#C6@iAX0}%=Zumg8h0m|48RL4U=WmsG*ucIxT26b{>C<NSbud%J
zlLP22q)@A0*GVIEPy?w5lT6@r#P}%@v<PN?{~iU}4YbR)XrJ29z_WXT?%(YHdu~o1
zM1SbZ2}F(!pkTqTgJ-Uwy*(3XxmnPP88nNMnWNGLooU5<E|}k+!F~z9u9*3IU&Z*=
z617J`s(7qP!<!UBxmZWxLK2)TSw{V>GwcK1x8UH)Gc-TU+ZGR=?d{?wMg|5KBeu5+
zx*5c|epE#*F9Qq^bLVsrDeoS0+^B5LI#MZ;*5EddYjdK<b@U4ThQ{BB5E2qvP5d!m
zU0qdvJ>RKG`$=SoyNRp0wbS=9e}4rNB)hu+6l{k`+VM;{d6MVIEl3a<jQtCfV|p@*
z=E$9|Sy}Q#qt;(omxbY8wx*;crBYejOSNRCTofd!m=9jPX1X=@@|8TAlTAxbe?P3W
zr;vSUh220zkTQwNY6zjk#8a+(U6sA(ZHj+zyPLNQ%FM!!j_e!}HM4bZO8@(-?NMIf
z1vZ$4xmyJK4PO{f28I^Fzj2WhLH`;ct`G4?y0nq!?h@{Mt~h<igr+RvnSbYu+IBMS
z$Rqq&QwmkmIKBwl@^c=V700`bK>jQSIZV)$(C4TZH5dFQ=69ioz6HQU#0<`BE_?x*
z@bphbe=;4W_+rhWVg9|Hot84U+n#3EX&7hSc8Dj*tS^RX`Bc}yb%7WUPi6+SIUm1t
z3PH-)y($wrLk%KpI6`~6my3a<9`mS~u@5!%bo_Lw1<?UY4Q4t^iyD;y4o#@^K8msO
zFdNTJ*VJy*_xK+Cd|F5sAHEMM4+~X+c`*(`uF{fb_=&)_irepk1jeL!Z|N=6nB-u$
z^#V+82I=?7v5HI`Z?v)pu@f1yHFxqbjilFaF2ZPaDQeT)qT`q8C)FnXgPE`1Y6moG
z^3QiPIxx9kvW*E4z$0O8E^IO4SBV<3>Ld?pq%e^haz|(gCx_q3>s0U0RJ}T?aflp#
zFS-6%7H#mw%yqQ8;*hu$I`|#a!gt0awz_W^-aXD3hT`$@@o@z{y@GAMm{u`pcVmH&
z=3W1!i{l0pP{GtO$<qYuNoER+Ny?<J6Yoy~CvSha2_I#Ozx((u$6>S>Qc)O8LTu8P
zq#-_3DEFCI#G@LX|Gy@f>li<<^|q%wo%SNnfHwyR;W!(J+2*Sq-u=#pq(-_;@9fmd
z)N`_HTLPBv4!U>NHfry05&{PY?<+78Q3w@8JN9sO=YE0S>3UsnHs3Zq%<5W0O2Fe?
zx+~JY#U0OQPX7HS+w#TmtQfKk)d50!30xKREkhf`0!fAfhFJ4Z<;ijgy$M>(B&min
zBsWfzw4zrAM-f{hBU<?$R`{<XF1#(Nz{d;Cp$o`sZ?NEimcPHAXJ$~CMU*v0cjuv^
zhmWKL$ym`(qTnLNTqclQ(Sntlr(;N#l`(IEjD3_kOgZ)l{Zm@Le$GG&-$lZ6`H)3U
zLt{x;_?-{(*vWFYL{gkssvn#OY$V>8`bSCIA2dmf8S}E-{yvwCX?9FFJFiS`&&+G;
z-rP$Q@ilqazCS|c<6vN0%zbZ2!U(@-<CH(v-s!m31<mz2CmzsJDmvS45^E<b+VPcp
zLEusH_7;*+xMGO6gnANQ>erck?JH>RJIYWE|5c@To>J=<2Cde*{O@_dy&Q<6(g2-U
z<A+Tms3B$FZHZ8jUGh7(IDP%?Z`TRSu}pqwf#!ao+wO6M)&t7NF$&M$QlO`)l^ud5
z>tBH7vLRkm6CU6Mlz?cmS6CD0P%92p&jj;+CH3_Qptammp8k#P^+T@7v9c`kPmQ30
zrb#rn^?34+>Zv0MNNX?~$+3~IRZB<zI0aCD+7v|zJC*$L<NoljLvnKZ{p8)W$`={p
zN41ITso@~*(B$&aft7Q_OyZJ0i=}LOB5fU=R0i`l*4uGbh%FV(ddSg9*1582EL)Ug
z@CCbt98+7aEL#|nTM+fN8+rmle~^nBJGN?owWso#3_BkRZv++mJJ@ZjA#vlmjl|Gf
zk&NKV@*rR!iA0UaM(P1c{Y14v<3g9eNQF++(ehW<aiQJG5)~qFlE=<vY8A;()fm0n
z8P6F3UJDXYFDKlZHk{zyyue&SHr>;Q^NB{&??^Box^Ark9=52?4;v?VJP!4d#ewC3
zNpLsYRa5!jkz-(%d8|-4(D!~)Xt2mJCYge32~FjoW~}72G}j(j<ksu?xw(N%emk$7
z4D0fWih|P8aNwkJnO0&-7};iKV{;`q4C`8b>Mkh>`-IN~oe=}uJyhc;#b9pwh`QD@
zZSWaY*GscL+)dsa5ig5ttJQ@5lVbS1=T;a}9;E*0`!U+!n=M^uI~KQLYR`b$36IUG
z$z&$(_lPYT{gv7NpISw#w1j=BmnS^-3tV?tB%zA(mh-j76Y-{Wqn)2k4*p(73tLBq
zlvjn)RMoXvbY86&$Y?o_E?pjwaQlTWru!kCd{|X|{s=#%G)^5x4+#%Lqgn8L@S^lP
zm1;R!P<MB4w)KKVQ4)`Z;WrTdzZc*X-)BgSL0WeGA{Sko0v@Ip!_v&JD{gT@3yUdE
z<!ZkPqFrV2&EWNGL<gllNa{E>e4HvGR7Lo7?-|BCM)#T0Sy<^RPp-|VsQ@!xApVDQ
zhta<=h;B++8h*Ix6#X2qeITcs*d&~Q=F;7M**S2z&hXdRJM*Eyd8!(Djap!wfjJS<
zUP&1h{=g@{WXNxiWrxumku>PpP9fs)pL+tR;rs;yyuB-gsfFc=CoCwzw5%MyREO@~
zxr1N`BQ`m4f2Qm_gz6y(qzI>1%i%7=j5*3IuYQ@tB`B8&ZbkEs${M{khSR`dP5RyM
z%rlf)jcE_DJ8m`*-EIg^6U;*2*XbmxS4K0+EcUX>G*l+;H^V{r)11RGr)Z2*u|PfN
zZh=FJc!EFLj8H*v%SEq?RzcAAb21rajd$<W8lCcScWQ9Dtcr4GAgfgGfK7zJ&|ref
z9BobfM-Afe58r+@)&F>Vzj9}+@e&1$MiO!8qFHuVA<B8)%*VE}t#CJAgHQn5D%<F*
z7SrZ|;5>_zORtMF=sn5pe~-OBJ5!cmH{?8d<DY=%VMF!6L&r!Koa`H~s4*n2#oL`j
zzOO1obr|h+F?G_oK3My2M&XHy?ftxn^E&XBIIVv~<}vF>0mpC!NPdmh?o#hp2w&{j
zyZ*S_5R4_^v515x)bIXa-a{8BYGz}2?oH{n4jU1K0CllIPxXDbOh56H(){9N(Wlme
zmdL`N1wHJo1zcTke2<!c1CJ0Wj=FBg9C?7A%gW07bNBq^+xBFTmDNOwPYpzX1LkAg
z;Z{sw+$lO^;A4OHGM3hBIR>e9+yNRxluc&Fqx<l`X>a{IZ0jr=4P4r??WP}R)I4W<
z56#^ajNfFS^W`-nY^%`Hdz0W7p!mxK5-}L^4dXLw;HdC^B2bxBI%kQZg#-*wq9Icv
zP?{3p7Dk3L>fi<l1GtX6!X`PYLu^&kBGGFEYL*AeaML}VmS)AhH;JuV$Xyv`XrgWe
z1!78vBrCAD+>@3I2mYXVS5;Lfwdd#OzsYsnH2W>JXg@D)Xm@vcFgR;F^Yd#nFQ|dw
z5r){{)&>XZAh;y5T7?yg;n-*(1?@?xwg{X-+omEw<GdZ~cUIl)6jc{}l0~4bSyolm
zTvE-ArIHx$kl%^YCq~;Uy;PCHAs{4pluh%PE>`l&Rd97}^nELwV+F=@J}fN;^Psu1
zXEo}70|QxL%0a<y3>qt29G0Q>_pwWcc`NfhkS^Z<wQK}zgrfLa6c4w0E{gn86g*A;
z{PSvzz5+*UEp*yus@7P-=VDJwhxoD1?k+7WtP*`@j`@piMZYQIF+^tKi*LO<+kl20
zPVQjob*55E$zH?P0-S@DT23R01$hC-AklBfjmzgK&^Y4#TF7_a?Zjy?H-oi7D6&6W
zqiA0wU&5%RQ-cF-s2J1B`SBYWDi5r$`%xwg9%1e&N>P08_i)H!MU>uiT{4Mvj;)Uy
zT$w{Do<*0kq#Hyr;)7nez^3`M)cv2)vxM|+1{=$ZMA#<mCEjO625A&cX-~fZHIzxM
zQ0DTer8`w`cd%vU$^a%aGZQ){k|RvQ{&(52n_}YJ#>U13SkJ1BzQ%%3A4z2Fhimfn
z&Cql!7KLy!2n<}NeF%)I=`cOOeJU4<6HUS!4m}pevR>gMz>0Vt|GAtuww@}KBllP^
zmBXx&?lv?q7y=oV5-5KrM9+a6*}vJB6Aag_g9+<mkReDSOD;ck+RfCQE;y4z&mpLb
z1`NyKEy?o!Y6Zh?0r$&m5M$#B@bmx=jJvk+iwivz!KT6I(U&5GHz+P1CG63)Lo~QK
zY$lse4?g=RLVM^kl7oIY+ZB%0Tc=xbu&6o1RU{%CFaghZknlLlPqH&z=E!XGUv?(V
z!49GZ{AYmx37CLyZm1#hn6H{Aj@bEX-i|mcsL3<OFd^pdO_sRe&nX?RG;3R5QFTky
zJ|163Q;PT1z{8Lp{1rOh8wyX=R-7kpbbbzY9pCQq4eZB0c@lTs9Y+-IBGct`xcvJI
zgIw?wzw<H>WKiL?li!S}LpqJa0`E(v7s}?8)oK!ttlHkBN0*UShS=NXss=Y;yV8iW
z=;<q!X)~9!{u?0kcHch;Vj)$++}_@PxSa?r?dqbWLHHhyY5x3pjZB?2Cd^uiOD9iY
zYmO)-DW13cbbsQxm)sY6@`)+e$<4n)XP7}XR;3H3#wN?Y>x4{Lgyn653pl@_8h*CN
zk}ZN%k6srT8wXJfh`K}DE<v_G50|3RzfbaSfgEC7ulw$`f}CX;p5Gd$A@zdQ67<v1
zMGDZj2@Hnl%Jy2;PIRq-e9c4Ctb*LZ!5b0~#T6Bg=i(f>TQW%s1|QX19DL(i{8Txp
zz)sV*$7ucgs@{-j9cp_>MzKTR<ELrFz4Oq*p<H8@FuZBLN;LFfWqj6{Q}}bgcxeO{
zzF;n%UG$<*JS`alyJG9ThU6DJKB3k)pA<zUGW17wG0h+-$pV9D5E4;|utdrw2%nGm
z;!nRTDFY$rM6Vyxc<B3e-qMkUyT!?W%)?IMLCk6jr%xwCh`Nz2YZ4z--5Tc!L~q&^
zu}~{5OtdGac_5h9D2mB{0r!!U-<g2N=YqJH2X0%d`sd#c+a1#n-!9a6$g4EM+p8^n
zV^L`ot$CW7n;G0Z8TT_mW_cNtWpFVG599WC_7mt3f8@e|3guv6BQHZ#nUvpyhln|o
zI%+E&4BD+|AubIqf~2?6X+t`{RR}ux2r?*QQLj3-i#4yUm<|wtfS%UPB27Q{>(|D;
zc-IN=9}%Ys8fYTAsdBvq#mZ=&L>`NFsU~>UfG6el>AHJjuEuDkvrfs#G->Pme+pX1
zDV$F7k0dxky9b?2@wx8hJ_x}K=$7X3;M+jx)mnP?jEJeaV}QH|o*z5odr~PJ!M}+b
z({Fq#o%du8IFG~<mL^GlFc(S&Pyd;+;!#%Cl+Ry*uy2yE^OD(?oZwdy7t=fXHziY-
zEDoMWZl-hHGKF#_X&4p-F=I$6Pg`12?5d6<xAwVtGs2_z?Qf_|^oxx0>4q{=$(8=z
z_wARa3*{p|jiV7?aQ_S$zVWa%duL!ka?53mfY#8BuJXM)*(}fhN{Th6?*Yk8T)N`V
z6CHI*u<W1?y5@9{ZWFWoYoELfG#Q|iY19R;MP$w5F;dcpqAJoGW=R(iv<UtB*WOp!
z<Ul|eEcv?v?=&Jg@zt7N5;{jK3=P8L-KDE{c#ge#3FL!WTb;Ac9`9irEbtE+JMR@l
zK;FI`o2}3$0R>{w=zB*;rYPFs;pn-_DcWOZd%|D|m{s2sF&LnF$&(Md9W6okMex{6
zQYS;~*Sd0`%Oma%YX3`*`FkkkWYYO+JO<OrO3^mFWpT0o%yioEi-~;C{F<uOxMmTG
z?fBAoU_*gIDg9=V=6l3cn;p?=MHT-OiB{7@gxGUgcSkQxgqpHdadLqQ5fAKZMrZV*
z`9svWrx2v+lI7Lv>RPB#+-urg?noI+;;$mD@Z=pg=>)bWv-(XR0j2L5>WADww%4y;
z%fFiLG}gBGFxK*If+`d?_NN$n*z@(MswJ%A1O1%A7Zw^k7|@S3XM>?2=>HDkKR>sp
zO;FMtu1>;;$CBhW>4AZPwj2VVb_01VhY3P)+)h?0f(~lCb6Z*jpnLl^1wF>9pf-%}
zSR&C0bTN!+m)A1k7^XktY5uGCtRSo@oBsx>cH;sc=%9m&yI}ex?X3sN8kj#3+RG1#
zV$4N*d=D<^$!ZShaJ^tRlm*|2ka(bh0ml-k@ri{(K(8ncB)E0p+z`K6hfn2rJN*1G
z@qD`Zne&C(+|t3BXdDp7Cad%~JwGht;7fA>Z%(Y>CJLy9RD^Rp?iYOCZTQ~5%#>aK
zhWY{dljIL>A|R2Wb9Bepb$f~5P)Y=>2dImkytz7F#cZNRd?U6?`g9)a=(^#(8^Y<m
zKBSQDv*YIT>jM1f8FNOl{U~->$ZiKwao_p*J2v;xA@AN&RRoPml-Y6jbHFN?m;aJ0
zwyart9{{=MtyY{L<DY#ZddG!N!=NG2OdLc~rcvfuyBd3}E61IC@@<~N=agnN=N$7x
zh9_^4LM)HH{`BN5EduuMKnbAHnXCti6;*6^z2B_uXMa1nt}GcKBz&c&qWI{_S#R-7
z&&J=Z+>%AeXdsOuQx#ukeufsJ{)+jY9DKFsTof$l*wehEf3qQtqFZB#*b{JnG#F1K
z1M`NOqqwjklg%X9$K0F_l*jIJ+P(osv$KXtV^$Pvb?oI!Tg&nWG<ZB1GKI`8YV|-j
zoYi{1?V9L*zVC(xG0(T$?dAI44x01a=M0XU1z=wmFEpT&jpSfJU2*twxL^x_rxp$9
z9mJ1Wp#}%p&Xl(K5)N|jtCE&PD>f?G^j`ZF3Nj84>gc?72IpH+#i499HJa;B3#r29
z(i;AVq1SxWcz?sNhzr%dW-Q}*?u^jRZYR}TQ8#cESf}d#+EAm>3q$1dn1~YRgC*~j
zc(rKlixim<gP7%6Vos1{wX!GkGVZIA88K>(l{8ORPBoS3N)$NK24(OR<ZI>AKZDD&
zWVD200l}Grs|ec~c#%E`1Ig-Zh~n2GW|~Lsw^AKr)1Q50V^2{Q;hW8TT}3meIDyg=
zz9?!dA!meQnL!}!W3epbCjqYnu`_E{&O{4VWkj@T#G1m1uu7~&P`Q1`P;O`e>Z}WH
zlch6a6);oszMp08p}qlLVoNNMLoT1I!=ad0t3QD-1@v|~Wu-iEf0W>$fdTnVK`Wrj
zgZ>Eo9+<`WGkEQfgh{HyH?;j=PtFToOT9mhwPai-Fd)Zr8vNSDA4RLcOD30)70`f4
zQguQo2bLb5rM<;hR~`c9&jA^1F`WxC&WnG6W8KTU;!$xEVYu5F7;W|<Qt_K$u>-Nr
z`Wh^pgjSjQ7p6Q+mXYF5-_!jlg&D-V3Jd#CWtI=XtmC~!rV(QKzm;6&57r!fcQ&@1
z)cEM3QUBr@4Zvzl{!p5Ctd^ukN-txJP$&k~JM`&w{H(@cWpUrJ%8A4FdqMCbXmWA9
zh6w7J5KJxhoY-URjZ{8Q-6}u45$99W)nMvLoAa0skQ9#^-AzL=w=?BaD&AN{WI<ha
z-4h&0!@K3~kJO2$%^GWHapv#edBaMDw|iDBhOh7_t0LtkzF0k#kUXu)?t^E>qWb1S
zba;ekFUJ&l3L-9x^dm3zh5U216WSu%2C1@cL-%{bEL3)`a{FE6l74Jzbpd^_yUoUJ
zxk|o$Kzrj4_yZ|uDT-l#zH)pX&<MQK_Ur(Dv9XV@?2k>(>GI1_UkkQb=J|p?M=OYY
zxYCnA4gn{J_svE`8%X1DrJ+j%Kp^V}tA=hs6~9$lgkC%HUe<7oNMz9+d~JGeZb$PD
z=;5MZFUW7-lTLE4^^eB^N1yPGlyqeW{o*z_rVZWJVYWm+Vq{!4bDpkUuZ_4PtINZE
zej(mJZ#_nfLvgY{Y@p|Kf*tTP_=GbEd$H`)yM!=~!zbOqxRadIpqVqkhdW2ueKV3S
zmYKMc-!Pd}P5Dl4aMI=lyI&+5$1`a_;tWfT=8*8Pg3G<b+8@w|-3@k~*aFLMiVF)1
zBTN4VKHW$>{EG|x1Hx>$g$o<QBUVhRMiOCUIZ8}MKU?%XEnG6g)Zg^KE=j(5lkTbZ
zaoi)3L_cVzUG0lM5eCwLg*^OUI?j|S{V%ji)pMdw?sfF^L@>XD7a}bX91cOitlnDs
z!p)}KXuWyNs7avxv-v=D0SHDNk5|i}vJ0uKWVISi-;h+F!zL0)%`ZZ22EFF_o`xko
ziJM;RH`T=81;~2kxC7mE1&R#lMBD$^x~izGx^4{!0wN_KNSAa=cS?(ZfV6aXOLs_j
zcZhU{ba!_*(%p5|`;GD69dFPv^ab}`G3QegG$9RCj;d96!yv7I5+cOi7#!wUecN_@
zv^d}7@(FzUHu&?_uD}ZykZv1-bS+D5Jn_Q{0Yzj6#Cg~ds7N8N_p#7ew=#GR?=%Z$
z?7EY7S$=GxdDNM53ACK>^B#TN6b_keWsS?<KCvEG-JS64`npo=9!T`G2BHQx2Mf~e
z-#}TFd~YAR7}48W%PRi<Od)W877^?Zt?$~>%O}v75QfKL+a0Ifi^}~y6C`fS*-rFp
zXL~5U0`1y+u%UkdE*{D-jKrEXP-`rTanXWf^(WFpC_yLNYGP;Dg{3YyE-5oc6+t1G
z(E4c56Z)a5_f{5rT9o;^{43Hzj#s#Q!oiB~4~~xy0&0bYN|nnpQk|_xk>{`6OgOva
z6ms2(;!ZMMs=5V4p7Q(kSs3Qe-CHGkbDNu&;XW;XudJ4@{fYzYZFUHe$gVWI@{sG&
z#F4%gfpqqu{=ttZ7n{d^Js`|2{42ZecNB0X(a_TN?_K~jT|r4n%)x;T3Zj<>kNPMx
z^c^NWAxN*47<{=ed)WYV*&m*=G~M!lo}(AeV8d{Mp&k->+!RZ0t483Jn=jR)|C9Jk
zf2HK=>WXfl0;~=uEXJY}xn8wQa@<tCw4d->lKj*LQ5ANywq-@gB@9CQej9g|4^zBt
z64pU9{52QEn#l_KC-X@d0yAD-7!ow8b(1vAA0z{YlJL+v{1LDlOfr|%ylRezZ5TeB
zcj16Fjvd7y59{G%#nsfwb`4O-|BV<fpd}R?LJBjQ(_Awx^Om9)Gr($qd_YISz=cbG
zC3CSP;FqjEY+`T!OWutApWjUW6F#KA3)>FI9A}=-Snl>!-W)^QRMKyfpB}CuxOFBb
z^(y7S?zoN2!XUAdsJc3>feG2OB<8v+Oo?u`n}m|8Aj52UNM&ZFv8Mm7uSEU517g~3
z!>1Ty(M*}i`r;v0g@vjG6HK48bzloH2h^707cPI!@FaeB?z$ygBpZA?0LQ?5?v$eL
z1}NWgbUg5Mfty@2W{|=bEc4fEpae4l3MZ;OPLNZ!3>$J)2NGMRW@aYl=KlRwb-g(r
zUZ^oc7cK|kGg!?;(%MJ_9sLSYz=kBLq=W?^9zUHv05cAPwjCc!N5|*(=8?yZm58wk
zDobRd(94a1?(~EpjBhgN;wu$)nQa4|K?`{t>H1BXb)y=)GEhKJl0M1RX~w*Vb`EH4
zA7HyTFcGovBc|tdgJUt_M6~!vpSz2(EAkaLPjAC5hzM5nZtSb^7;`r{diAk+`^Bpb
zCKf7TlHiN1td`*&gy`33Qk#grtfEOh9^4WntXoJFZ`u)v#pOjKC1a$k@oOl@JI0QG
zro&gTBhvxuH(N13m3YFLP?SgnCSi0}vS9M^WHp1$Y$j3IiAAa905dSpp<-2K(QSA?
zsM4ORflB(fE@?>+bCC3+Nj0+gUe6kLEdZ=OJ}er3cYjBLu%W;3iX+V|r7xbHSDD!?
zC3&ua%!tCo2W8(@X{?$N)^RRTQQ~TLY^=@mvDfpfipp+&1T^5@zxib|nrmTABw~Lt
z-M(<CoG<<r2888n;EY}~5_tG9Jv|NjOD@wtvMKE2HD)RaH<%MAKht{$89{DUAmM&`
zQDLFZ{pDV<T1^O0-qgRI2WS@8%j3nb-=){p0rAX9u^*AYG9)Q7DK^-!Y|*l*8E`oq
zE73jer)k)2dcHecYC{hsY(MKw6-<RX_d^>>dO~Js-g;-*_)MXuuKw$K{_0MHp|>5L
za45Jp9YGCE9DR1UAv}sq+;(bR+-ZvxMFPQi*i++<NW<-{d&zk@>~OBaw-bdw6eA+V
z=L=L``ff_T(!|={su5IK%Q+3mR3gy(9IgWn*>K^K%F1X^hr|Mp3{D|m0lhg}Q&&2x
z=8Y5GyEUX&XqDfQQ7wEjo~*?}<yQw@>DYK<Tqkar_w(y0zF9>IlEH0GVc-SydX<@$
z!@*40iTjDG29R=>db}ctsaU<AsjJ`rQ^a}yFXvado;yjRTXta~qL*1|LqI30Kq?5l
zU;&W1;Ye!#M)$8M1s1y<8HR?fci@u#W*;iNYI+3JQ=3zYa%`EGu!q;Tn-MXC+qWw>
zExnLM4;08aAV&}M<3FAs?|_4xssZi~g4>S*pE|46<NRr$is~kMK80_SNxeFpj{zs<
z1rVQ>18w2waC}wF-r~`^Gg<W+|4|>wPumzD<5FnI{m^n{%_F?$6EwS56xD;zOEeZ*
zd6?hWR%8t!$~;+K*UV1LG{zh?I5s*irhE$($}q90egSH$vOI%c$G)At2?wiFzk`0`
zwQlVeeXr-1h-|$Lymuk0B=!6IMJ6T|AOSG!czGx`i(v}S&yQb~+nt{2zV*`xr|$=&
z)b^d7VaOTs{PdtGuXUEyd#jQ>-R}ci9yzvy7Z4GCl)fzIyXvjf5kdVgH12RxKQCXo
z-?55H0T4hb@VME<w*F~MZ_mUAuwHma$ud~q*?Z^rm&T}g(GoB0T7WJ@ZQ#9S2h$dy
z)9t^gn#nSlctFy8mH^BFg#xV*1Y7aKnA%r)>1%@@BWYacL%arHMmPm25DnA;K-;-W
zgI|+2%m`v41_rBv3Ilwo<$Ap+kj!7|g47O>N+D4&X&6l*WKNR@aLmk>mUL__XYcs<
z_X&Lq!LY8w{{DdGS3bL(luL;aG_$O~!*h<RmJ$ALYwh15@^0JqmO6>N3T;QEt_^hg
z1$2r!FeW}6aSP|nzhTo1->p-<)`nTg5x+?{kDmBO7Z;U+{i8ZHTHixc5rZZ~;*MYM
zW*(mU<0!@K)5xOLC(Fa{UqA1PP<K~ytXNCjxPJ~4Mrf~+69}704rt`2pjetP4uf^0
zo<D1c&&1p#{&6lZ-Y|Q>Ng?rO@ExcAhUmVT*>LvT0t`gVi=x3TKKoTyE0|H%OWnLv
zU$5_Gz9V&rTdl-Wl$E+DdvKBG?oB>B=CPxjh#2`fZs~t2ZG9=iH~n6;FbWfiPx}nj
z<+*hquEFF*2e)t*bpD8&*f>C`GglFNyYaKY-H$<iX!^mpzmNOtb|DZv>+|hF)fP7{
zuuIcutJRa|0B!!p<o(-n5Kafs@||}$i>#G`uwGRo+=Ibyh^*V7EhqX-x7pa)Ay5Yo
zk)?ax^yI_Si6<nB`S9yrEtT;wg_3kU9m!h}Fbri5R9nCTuFvve_IqV;hFI$h+Y5@D
zJrSe3eQM_oIn4|!Yw1p5LB{JWwEn&!S&pTkSPDzT)!qv}-<{4a&xX|TH7Zf(d<E*w
z)xiONW?6bF!6p|&b<y4}xiYQhxkp332DfwsWRG`sv7f1i2O8BrRcyP45>}x?$?;y1
zM}^!Co*P)EU?(PqDL2(@Sj5D-M9V*fewOP0G|DH9BVKH&)+w|u%DZZ;O9Hp(bZkvM
z!`w^o>y+HXhch%LSJM?{{SM!4jvNsDxCQ^DNij*sKaInPTm1G;XEUVH>}Iw(jiB5n
zJ?;NY!h(g<QREj;PM<;co;Z(~C@Da_F?*?tl4&8P<6mT6t{l6$cyGuNJ-z)E?RYr;
z3LPf7{XY}keOG@$eNCDT2o<tyFRnNkXQ^Mo{(dca9K@O%ZU)F|JWs2_@gc4o+m;-0
z7b<b)<<;baqDsSg-T(4<IR#Y5z&pbV#qD%Fbh_R-Y#=c&PesSMQ~J%btm*uXUT?U=
ztnQ{ioESuj)L>`ShH2>vm<@2=jje&4*zL&sYd4@%+5`myP_UHhEY;gU2mz4kS`Z{$
zQ$E%p{#}?)pOe;2Tw32Dp#3s>2s5zb0vhZ5$n_+vqK1QwTUyNEb&1Ru9og>kHF$v^
zU@8so=%PhN$8KKiH|4m7qx6$1`#mDr=X~llg={D#%kNYk&ZB&Xb1PEZX%uN+e&^@P
zyH36FJ4}2pDp<DC+;BoEK$7tb#Bqk@d^u=RQDW}SXZ2U^z0O)wBrDvG<~Ny-LN=e|
zuf_Be{y1FHw^ie4L?v`35NPN@_D<F~PZneCgJ7@yI7I3JLKmSqL1ZJOLj#Ow#AJ9S
zTRE`Nr!c+c0$jJ3?hmNMCDWsYnx@kZq!Gqd0Fi(D_RYGNzzM>A{Jh%VwoQT9B87Zk
zpFuQUfd1%Vt9s)N&yIb6k?}IL)L@4^C<o&VG_YUNxUr_U4lln0@<a2lBV35ddP#}q
z-}rPs*G6Y1qY-?_O#@ZhxKDGv?4%7K+g?adspWPi#_h0Qox+gN{F&$OA&CTjjC`^<
zcrr*<Ty%4s0_#ZdeWoH@i*~ysD>1cn3eGYq`8JF$!KRYRcrXnqW;*2=FNSMJ*WG79
z7Y9nw8VV_8au)V+Y3I)g4R#KcPMW0i;$}ytYCo=v=*k4Kr_7K{BIwJ@(SCft{c!zH
zOC;i$O|bVzXF9BMMJ5>&tB?1>a^6Yfn=>lopx9)!47LB^O1#2)_U@z_I=!#u@T$!t
z>87gBbE$<5s{%I6RR-RwuT=}xM>-CemV?BCpbwu6Ig>xCZ)sN>%geXfhu3GKnW6jp
z>&DeGnfjVqSkQZ6@Wm_G0u6!n<5}+o(DGO5_rHr(2FZi%Shm^#6gj`+1!RMOV!q5e
zFywBP)U5z<Li!J|G>>onWAp%mo8xq{0>rR4T66;~kq;iT-NAg}%ZBhlZmX0dB7f+_
zFnc9<AMKN`<54BzR2m-e#CPndX=%yA1zq27Q_6Q&UG><@4BYf>V+IA0p`?haPZFsr
zt+xHSS0j`<b|rzSwi(7H&BFm89I5VG-+;GSwYBka4i1lp>v}wS)&J811nw>W<q$%{
zApunykiY_cdoO_*@DxtX&SLK!<xZNY)N3~XWK%PHje>7{`w#=w`18s%n@V@CTnBKS
zvZLv5aasI8c-lTLuJi4<c-l3xpu4bdp|{X0c-z}3@#K@w_<A+@n3&_J&$PeX{|IV-
z_MY?)t}%QVN$XO>=tkfBCa7K-Rw-j3Olx#|8>dX?a;3yxqA6n0$urb+PGQG{c$Ae*
zBawY=Wayw;`BIXnKwVv#2^*w4*<{t<`?x3PNn?tF^}LgkoJ_^Y==41NWmGN0G(`c#
zM&qcNh4bGV*Zr&Momb+VzvohatbE`DJSx!OvzcPjs+59UlCdn2z13C$h_JI25Zaso
z#vo|Zo%K=0?`;B+Bge_0aR|oJNe3iJ>*F?~YNbA+*TtFVVCyTQ`QBbhJ?+f#1vI5T
zzwpuIt1|`M<=<39=&%8%O4L+x1tumYHo!*r`z)BP8cI<{kqYl1rch<f8G#p5TrO2~
zaG++2oxR*ti4T>xIx_U#ty*T@4+ljI<KfdNvb2`zOcD`ebK}gcmUMqmQA6>$UU%KN
zLn67|-T44o7RCYx=1>6X6OawDma27c%JJ_}bjB%w-W7()d5)>k@!&K69g0^lz*Ape
zpG3q)Yw`VL`Y}d4!^&RSSegzdg6I0adD{-)z{vF8NF5Qbt)NbX8d8Kuzm^-KmNa62
z6Jtaf12F{0+FHShF)27i&W{OsyqaHri}>bZ5>_Zb?m3l>Pfdw-?td=5+sBfy2yt%U
zBp-V|o=++1$2Mp5_lk}?wE9k>WF%CR;~Su!Ic~nvlmtmEO4dqiI9|Zh5D}_!)#>Y*
z>d2QT=`Ka{fgDqS7UX=fM3Ms?1Von@9#9*ph0TqtR@Q21AD%#`Dl8)M3NpY#{$eN{
z5F3^Aba8QUUP)89ps2M(rY~VTm@O-S$QPlNomas5nFPF-yBZ#M+f9H{|BKal{X0Ne
z_h$k;3e+x?r)0Z2?PkFf6vFRPzc+6M#Sayrr1gLQKGs;T#X7#w8O63N+yC~wj!nz`
zZ1u8qh4VMvYS!$PgaL0r7|!E*?{j1?kz**|G%t319PI#1CfvY;ya}Q(Y4itrAz_Vd
z0Wqs;eWD`Odm+WlUax>S-3vpUu_C>aItWV~Uo4*&3(>rQ_J_&s)>@?w;0|;e^^w5L
zSz>HElPW$rH`fQaN>G3|wMTedk^!qae4^Tvrqi;Rv@}vLzTI0v^jF<LN0rFy!UBpF
zn?qrOSu3|Y@b}V)7-1y^D~PPDtc(CvK^m?HXi~7#o#f4_E~jr+Jq3W6s>Rg;D4CqW
z3=62qDK$_&0SN&r;QWXHjOT=&BB-Nj9WkN!STBP&AasKqMd+L#g@x4^qmMrYv^&<x
z<P@r&lwa7_{Ya&1U4_@2hEQTevSQxLdO1SDvi9qja;4>j)_A3Awhd-y${*nV(7kzD
z6A4!@k6_h1B)V^GHMxerxv6!B{bi08fw5#@&Y$LtUD}Rws^6afuI)k53R0@`M`OJo
zk5wq8*5kRRVHYF$6#yUd58i@BSy_1}`8yB{1MUX<78e&ck;5uoWMuKovM2{G|4wHQ
zzGR-%h~%bRY-8~PUI-#{fm!?a)~*2`qR9i&R1k{}Nx;BNhGewZptnAr4?3s!0%40d
zNUr(0UkJhN!j-#L0z3A+LDh%^bpL?Xyc=<95#i6Ed@BONB-?>|1y1nhczKw7veIB`
zdb$Esn@$n*-3bbB*6Z(>V>q>N2}~T#AH0%2jplvCdq-QZJF^YkT>4I{pZ4D9y0Lmx
zCdyGEROB?g$-=sG(?|2PAQ2W`^=SS+{WW425uB(Rq{q=NB#|wnd!Gz(=(S3c*xk7y
z4dPcRd3Ej4Khs716iwrIoD`Z`Bh-`mAFd2YI34coEnA-)JD75HhLe|DRH!?fyK~7R
zL?weMKUy`Ny@qHSgb}`Q?~#LL5_rmcz3Ax)zEe<k5EiY<dy`{KQq>1kd<h0>{zc~7
z^2qzI?r-4+J;(geoFzXwEO|W0&yKBm;L#rJ8(UZvR5l_tXdo2=u@Gi<TpKFPLML4W
z|227$N{Nu)(OUxM<Qd1DsCcDF6mp$&jOVuupQNvy=n@7={9ms}NhS3oALyxo(={QK
z3I)J(;L3%d&DArHmjwa_|HiVy|CEhkcU1_z?AUT8j6r~928Qo1W|R`@YHD66usLGf
z0XlsU4*<L{RqF9F2vmtuI0xdu%?EynDjf)5?Pe7uwqZH<!y)l#5cP9&2Pm3AXMG1o
zcfqf@J+7Xgw25DzxjLOd(~O=QcxIK>raPG)41LQROLaPSIkuCR${Oz{FQ&FNbR)|9
zu^4UviKquKVXOiqatP$IuiW0iEUbE#?!k7akMSnU!b*HihVssc_%LooT%=M)@r?v;
zwv;SR(a*+DO0r2}(*=6+3wM%WV(!h5VgXohPF`MYHTm8(rCH<93O;^5&iDO_XL)j^
ze7*ig8H30iOM={S+0P|d$2>4-Kt|@%Ng?B>A<=W;OgV3Kw$TlASYHI5>;q7VzEB`=
z-j8+Gw`PH;*CG|1=6VUrZUk7G@V)se$_nLMkC1Dsh!<w^e=s>=4KMZXCdEZgv#q8E
z^bIrYe`t&PcaWKW^TtuUu?tWbzX|S$EFknub+`)ilkfVy-q~MMCt|>bj{bqsfYY8g
z#(EI8Btqi5?3-WldHSUQO$wCxah_HPLeP$CW|J-8S}QCTYma8!Xnr=M{blnur_(I|
z0S#(Z6e71%?_WGf*bKQq$Pa#ze$`$->y02nZ370yfewG924M4YoMZ=;2vF!$84lxu
za|?3kTD9C!fExj#G6TWRHW2ORljm(Gku6~8PG&OIK<+&N2Y381zN~><)7=B@7jAG#
zpo65KzQFyqL-d^$AG-ZLoYM^9LbUZf<T=?GG`W1iyKd{#=p$x|YFJ@nVFUpaVahRA
zyxr<TCR!4~qTIGnAtz`cdD$H;=82KCYajS6jfkD@M5v<iM9?;+*l54Ep~V4&VF0fr
zY<W7e;_;dfC*Zm5;b>WRYn;x5i8-+c@XmOSc-4<~AJY9(2+#wWt<VI--+*afmQ_||
z4+TUqPz#UtMUg?KLP1h0Azwj~2{5$<LfGh1&A%)r6S+P&)ADvzf}A22>11hNHeCo$
znr{ff%_k<y_~672yw*KokWfu3%lR9-aWQNgAku?K0V;vvrGECk&!;zf(PouL-cW%N
zh9Rt#ulC>3oWe2teEVJa`cHd3IUH6?e~RdeUt}{(sOv&ah3@P@d;rtD^08#9v~*Y2
zd9++ukl*LO?eJz5A!oTp3l2n`H>B|4V&t%*1C*jpUN^A^E)5S1Z=M;jy7`-xLzX^1
zNGJ!j^?~A=Lc%c<*?n9{#h7~P4P{LA48zw&Vxz0lH<aE!#MYDS%9(TfNmO~MOg1FA
z-RvxMx`ffBmMm~zIZ58bdh^ttJRIZljY|7eW__jGVtg+g`K3TJXYh+RmIC#{56gnN
z4klnq0Fshf>w}UyK(~S-7c@zK58@S~hYd}PjjjKsgShXBzfM-H{SJs|LSC_m=Zd)Y
zFW|Uw-b-N2$`hrQC~(V{;X?6}={fD9B_OWRoO=-TqdQ&cxQLAZ#SkefpDscnn_XNS
z%CTap>o5tbU)Dwjn_eSs+bxN6o9~xg8t%j9?-?}u0Hc7Q<VWPZb*{=w{co~>qvfQ5
z)CuiY!*M<j(wspqmE%vWIgc=Tx_|=}m6df8j6t@e%0i5h#CWJv2+<)6v^y^>fcOw`
z@$e8g5!Mm)g<}wb2;eIU=Rd*6Z)>>O>ANH~nqwcQLS4OTDB*^q9zwHUh~US7KgZ*h
z9RCxhIFSve`~6I2jpW)X--_B9yQj+q<5Zccpm=nYNT3~%HVYNBWV$P)i{-?B4bD)J
zE|}c1JIYmno=2-aVGYa75O|@zNRgr%g$zi@jRg&jH>B{jT$oXSTgVT8LVj9D^~}u5
zf?mDhI&FKkOpcwn#{?$V?`9ye75bq)KJE%QP5FbvcM~{MoT|Ov7eg9Hh@J|FNkN=a
z_(ms%n&Z!*dxL|{qswv&FxMr`{n_g7M)T#l_11a#_*v7L@DpQGWJ_~sc(6nzj@(ZP
zMnN%LELz2175vY62^z1!>`E}H2$Lmgh%(SC;hF~H1?R)D5VRi^E5AYqKyYM~O_j~H
zq&Gw2GD9OPqIX;;fR}8x?b1C15OxfI4_Q1w09ieCksE_}<61MIutt`x9yB4S>uooq
zNBck=ZCZL*8VV7wGZVnHt!-MHmyKl>a9VSk<0SBy#DQvzYg?M~0;I}-!80^|0ioED
zNW(X}x4)zCZMx8@h_QF0(ziBori9JTb<)Sl#F>rT1i@h!l!LJ6!uvF(3sIZ=yg5!g
z{6lc=wjJUGajqj@U3J-XjjeZb9p_Tpk9-sS+XH%86D{~B;>35wUy{{qgf`-7!b}0g
z@=|^T+-msFV;(r!#Q0al9jx(ho-Rv=ON1>jO8^y6tn+Yv1gaT8<R-BgNTK^r14wv~
z?`EkJOGTp{Tq$cfjC44;<&uuwSSnwERtThTb%HVyO?V8Lv6y@D!NA-l+d+DOkxl{`
zzQ^65HVZ%!0625_Caspb2Fw}-Xz<?H?GNMH=ZDr8yww}<4O}(aZ@l3N*FdN^#1Fp>
zzUBim##LDa3IFN``bla^Ku87D6;9iQd@j&X(omDeC#$)pN_x5ZOhr<`YMnQ52MAFM
zeJsMEmQPIlcb7K`6g=D^Xf1I3DcquUmXK9Mfp5#{b&=t(C=1>Tl#DExn0J4DB<<?@
zH!0J~_y<^Ykn8!|fP@p^XYIgTm-ozPP!NDW)nz9M^s<o32#^k)ASjdzgvV(XgR+rO
z`{jNXnV5E5^^!LNw7-El;oKn*P70{4FfxNabnb9SQ3HGaD8wNMmWFDDE*xkvb^+F8
z2tb!q+X3oOLcWD<7s#xL1PxnB`6Gsm(TlW`M2b~N<qm-+!TW`n?2z8mWdfO{7!>k@
z8q(BC%<uhY9i*Y^xbhQOyw#1I!c&Dlzi!g?`j5nj%<;H=$ridF6;+5#aVJ@SA?NtE
zzoUtb%*d7?N1Gt~*8dcp?s_Q6n3R7XIQqkOva8p~LB_aB*&x3rT1N&OvkB#{Uv668
zwm-xIRb40ka8kDvA(52>_+3+y>(+qhGf(0%T6W-Y)xY*2s8S&WXp-s$1wb1Dw6cP#
zs!5Y;QU#N|Ac+AGYmFAQZ7RQg6#z*KsAq!i&maf`3o9UB2%#NFWH*y7l}86!P+c|e
z+z1)5>R%_j^M~sNHX(}SXkv-UM_tc=t9}2PT6Dm;q+u^Vr5<=^frtjAHDWQr)q}YM
z*|{gczyb1a`Hvmc%sLQavC+kd6*|U7?_Tf4ydj^Lz{r)UVoJcgQP^X#ZNP~-$KNPN
z`$)z#sYVli>ibqc7l9X*9=YRT^;*D-86#pHwe<yTrYo`j+pb`Y51iJSCj5r#nlSjz
z$)2tM+`zf&ay6~8e7O`ANF;Q!(j4632(!wHJWdjaWipazgIlD0goM+GsUAzn@5Tu(
zV5>^6P%0$yhX}mRnPEKele_nZ3L}Wgga)zci6a{w{$78)pKZ-=X!yVcw+ncPSJqwV
z;)y$Ei67$-O#g63R4N9*O;Wx{UF3?>pbJIOY%P5%xR}$6kXKO&48JJ0DBf*Ra(GLy
zG>W;|{MN^7;Xdy3>gQW$h3IQ?tUXM6c~~kpWE)s5AN62?U$IahA5z{2kbsIC)SwCm
zSl^UBcrGn19VS2h@Qd{S&#4?f@H<j?#^=@Hi6Id3dQp$5+UWC4#|<0)dR&COzJFQ3
zKJ*t%G)&`hYTUU6d3@t)c6}s)r=ZV_9NeC3bf5)-tIX_dkXv5w1*t%QhDW(fYq-t(
zm1gJA#B7g^44A$fqoG`d9&)#7fq!TAOmN%B@3FR)H8tBJyXI7!TH-m27}@sB@rFz!
z4#Px2f&8R>>C+;4bxixb`3&0{f0R#_0)l9JhrCZgKlD+0zCX@3(l^YC6ZAkP1^kW!
zX^+tvHaz1C%o7)qn49?w00~55`+WbowGjj*ly63vpXSPrx3UbChy(O0pDAM2$3#%`
zj7HN%J4E0*5I#K+(|nTQJ9w*HtkDj3;Cg-~gd(`L*)&mkO}K{`ArGH8cP=ysj7Amk
zgkEh0A5D92$UATF27}yt&;KT98-iM@)MOs8mf^{~N#AiQ4CL;&l>BaSL_hQHk&Jc0
zjBxInL!==42m4>uCd&=kt^~BY*GNRX!UO)qMh7%DhKION5Yr%?{u$`+O_mV$qrlrn
zjLjmdj`{vBVe*KiJPZIMReC~FnYw5#pSzi1Ozht|3;~pI2-J@nB-N3Q*w&dy4|qhJ
z*2IvZ4f2@dpnieKaI8EZ)H8htAacRa2<_lsXiRlghPj!knE^r&u?<Y{*!|aBGJimo
zD^XomH3@j{TFd1Dh`($kjfVhEMNAB?ZrL$K<OzEr&1t0{KndNDL1N&}9MTCS=z;tQ
znL@t&&nYkH#dGj~au=vD7`WA~d)n*uw!O~<MRVbdRcW#gP2MA&N<xS5n%sO?B+Ylc
z0|CD|Er)Ye`iRSCF6`@g<&24v^o!=W+_tO&;2eVL`~y^U5I+xqs`RT=0ipJEv$_j>
z3ie({kl0pG1Owg0-b5Y}B&-`Ggo+Kl|Eur>us--V_3Ht65W6J!91lng&qis5=gadY
zDDn`RejFG<@Kw;+<pSpxXduf$GS`axjV2UeeFD&6z{Ru`E&UICt{mSLCa9Jn<Q;%q
zf&4>UT|MR1_hQYz?@MZ4#e#Mv%pVE+9FoZbs%K~>r(=!k8JTq6B#0dh;`4vL+x_IQ
z|1E)Rp<cJ3uGsR_sDpJ2PBhri${({|MWUh`Y2GUn9Y@uI-Rg^0tx6B(+9K0|tvE-}
zyya6|;e!1bDusTbD9&~%4S7-%%!{HzZ?N8J6AJEb7qx4LwD7TjmTOUbJ?|nDJYHtw
z*%`&1jcl6_MYeBkP{byU)OA_8Tz^@n{zqEXejxjbhEVwb?{71V9zv@Jq^~8_DPjP5
z70aGXiOPRVj2vb*re?Pp#v$yDP*EeSNtTJQj{2H@4g}>~gGY|@^l;;ZAo|Nk0HC7;
z{*wT>mIHRtpv^fO40WwdKuB_|>lf4=vONJUKUjF+U|qg?^t@hS|6Y4geDcU=R?_U0
zVOLg4zKAn82aJPA(68Zr=m&8^OM-s1z1A)76pUth^-1zn@K5xgJ3$=le2P^=1(9L+
z&TqsFl-`)mEKG2mBVC_#ek1yRP@sOtGlLh-S5HSL-N?9`y5d@LH56DgbG+<666bdu
z<(B*K_)Vn%)D}-O%VtNrcLI!q9-8zEh6DvAhAH&!cv7`$t2Y;C46g1^MV?n>2qe~k
z7qqytHDx<I0c9u{3L-Z$3$YxOZC|rM@`$ZKGFt-xdv`Y-RyHe%Va$|-p%D4>WVVVw
z`u?ENEG$sdGWi}y(KCrD-B6(yMp{RPm3E^Ki|yCh%OLWdxs-sK8hstv%K4QQ2)qtv
zw!A70X0FsK+eMg6{gj=@wK>?jVbJvO;Q50=eujV702bB!XHE^mh)VOmlKMey<&P4L
zX;xW?c+L)E`uao7xxVJ?S#4djMR?HF%IJ2NTCmsqz3Roc>Y~<i?|RbZNJ4XtHcb~z
zRsG8EN}3KdPc~FMuT<_pHoRD9)C&CZ-;?KwHRYeS=YC@`5=+m2R6f<GpSbTjk*ZZT
zM~04$ArjNmoHZzf&2!Q>mTs;Md<>Z8-$dYV1G)`W!k_<0zTNEF;+8C>+@xnb?`7<O
z2epg(a*DcZJ`?Xf-?Ibo$uLGxK|kI`yxcyf*o_S5t*ebPczNEOJQEa6hg|)X4Ye6@
zv{027c<dCAkqQ<ka<@OObJ&~s2ytxQx9nn-Im4Z}?eg3WuRh8?uM0c|+9?`RwI;~9
z?<;cd)`Ko-w?&Iyuh@XQ=FEu8L&L%iKK&YJwZ~~Vc+u)_n$>f>)%JGXN{A^<n4``!
zp2m+qtv740Ok{Gql(qYmh0BIJEgs!^Y{I{I&$LQI7lg((WER=@6$a+8M)_=({yi~t
zNU#>%2)WS^lxp~np6X1FfboGcJgV!L*7-J++sWS*v6*PLgER7X@?>A|V|o-~G^Ok}
zi;kU<%kKX01lPQ^K=AghPT#c^cR_8#TfH2nk{XI44@Tn_4F01(>^4}}97iozzrg8|
z?quz3)5VFNy0|ZSWzUs7z)Uf`=6B*ctNq^_b6Hc&5Z;7cxoC6GprQRWRVmqE&oPpu
z(z35quQcl9G3zwE_|01;AITk2z080=_cJ-SX|3Z*u@z2FO_p)t&oyY+6N2P}(P>w5
z*p-Ksn~|Gz`iH_{jw5~t)z7CVk7=!+Ruft_u*%lT9^+bBT4&6CiSD9^l6LFZd)XE=
z7PPRV8uYQuZ@xg+9#+I3)zCFR%+$4DDH(WrW)MA85WSlFDPUgNqGq@UKAay&NN%9b
zuj09+>l=El)2qJ2!4I}v*%?~;?Aa(EQ_z*|9?G|zEi31`?>!|+-)TY5989QlJ4M2u
zmP9O6wujj8Weu)!=RK}f${q2noT2>K?EU%jusRw?tKz&4qUjPo!Pi!5s#8qc%2?J;
zNPfIk3z+)ZoPXl#alGphu5Q`jkKO!F><R>{_MD$oVs6MicF94VdZW5|wVbFb34G;A
zHxLWl#-9qexx0;c{ugnp?s;HELi*Uj_y~h4@W{0Kw7fbD{v$2)>P5q3bt^w<TOS8o
zhi=?PZ%#R0c2>6_5g`s`t^@+tdICLQ6a(uhHHqKu=-daLeT)V}!nBJwG*sz~hn-Kc
z>svv(MUN)#ATiBm&nt|vRGEr=4A(FA`K#0*H_g-EnOH4(Mo~cL5G~tLiW!Xb!RXc)
zh?+81PI3hk{;^Oiz4Q1-?GXBBa=6aLr{OSXz%Z%W0%w#<W{M^KeF?dkz$Gu}tS`Ep
z0X&GAV`if=muev`R3v-(SGX4-*(_x@K;mk(j9<O24Q$9eM;Z;b{wbD?!61xX?>KLK
zz*(co8J!B_1z0&axw&;~)XuU{-Hlg{K?^7%73_Io^^8pY(c@^=BkZPWm+x_)>~X-H
z*W=WeXsMAJkpv8I@Ju{;XgS#wA-YG~<QxgmrA3^!d<%vw9^X;?1O3b+P!qFzUW$AA
z*;(?BY_rKypRM_#l9IX|+huHyixX`Fe~Fl0SGTB&KbcYx*M@uBuTw>VD@EtBVhjx~
z&W%8ie}T2NuG<62mIAT-QZV776jmHTPTqroX7|nB%@>So#_yCpBFY!^LTK}G`W5pJ
zu6N`({$#f-jfqxsf4_m|42iPjRJos<I#{PNIX_^YvA>k%Vm#sKDV@3s&{;Fax3OIN
z5o*5-M}hg_`h(IqleGajmQebLyJ8cwH=e#qvG2S^w`G%hXrY|C;QwRrw`)3o{?A9r
zyLl$lbbX?hi`@H%poW#u>(h~w7ajiRvxo~oG$zYEKFC$#0xegNovV!M-N2&dZU>U(
z7SGCanmsIc5?WYjMeI`02X1(UGM=aXeRv&SMiYaD)Ej%0Qv8ly{_6@-e#$z7TL=o0
z%HtFs;eM!hcyu%@tb}A}ynT}lI<kZ&KUDXB{|Tv!XU!~|uas>^9-mKM_+K`(AAjB5
ztn|D*J-KUv02^E2KG{g!1ZU@qmtDB$WyCXjnfsC1-4>+)7oKYb?>R!*&*>t=s0YZB
zPGHbX9?WLvzsy1%*1i3^`eezV-7-YUe}<?a^;qS3J$gIndB=D$2zEqOl@(WyHaKxh
zO+rVxVh(P@FP5$wM$i7~MVw;4bZ8fgYYyRJ<A1^DGR*$`^W-d@yC;EA2Ob{&0%z4n
zD8A?5lNiPNwAD{$$Gl}r!{J}3uzM7&1|tpOTilot;b0tmN@I8;F+o_>??k)Q-A}uM
zHCXYCoY;T7aEpYN225*kpL|Twc?X?3Hr?gYSx5=<uaml6JWRaNS^DXy3@D9HLzyvD
zt8njWa$j#?eyHp;LimCsZZHksqxf5waUw^OW90STUD4{rQ_9&BMn)n5#m65aLGpJ4
zWldp_(TR1fPe-r=OSJPj2T|@vkqeUKa%8Zr=%vZBb*VQcC|NohpDKUv6)G%!RZ!Q~
z$1J(%Rv7b<MO-pyEzUEZHZNhPWb`eIKhb`zUELO90KnO^LVB|={o9oWqlwDlt_?4a
zZsj(QwHEZd<F&)JO$Uv7A5mJTIFmD>gKAalW5pFX;J^53NDJhNAi!rvY*a1tWUw_2
zW+F~)UTJ9l{P2C#fGVR<z|Tgu0A@a`bYHdLtHPYKC8;od7)5T^5`5iTyA;=E_EN2S
zYj1#}53xg=!(e#Ld)mMIt2@PVn|a5V8Jgw^F5=~GX3L&a1FtrPMAOZ4P}+tMBmUi3
zV7dL_f?}}>+K5GOT}E(9J{!)iqx??`@UP;AfvB!w6Lf-GWJGs;<}cL_3+zs#n_O}O
z#i#f+yU=5yq)k8UAImbH3!1k2E-HUzMmKQA3hq29KxgDYWup{)JT0zy<^3+^{nJh(
z@)Y5?X2b4FrO0}L*2}9U=P&iQ_zk&9VL`j`q%+j<ax!iH?uq!kU(PO$po<LExlQkS
zLINvVP-Vt`GU#Q;+sLx&?Vg@Smfi1{ZNqz>!Vkjcn@>;p%T7t<T(Ox1j8vAo9uUA7
z2tOue+<g#60PNI}I>rTH^m;&_z`_v%e+#rc|52#B{5K6Qp8xF%V3M;Lu~~J$S#RF0
zU9`*r7tCIQhsnx`FDic@7$*1Zk{yvOy?NW1=J@I&Sk4UyIp;IdD345%Zz*54GER?H
zpN}3P+r=nwzYHkBH)H4b7E*stvVzZJD|11Hf(*=Qz2Ir>gw#{{CL5rB$~a(qX(xJw
zE~|fff4%gp{%X5yYt(bF2yL|E*3arT-)e*CWz#eCKHUSSk2;H9Sl!QV_x0zQ%u<}y
zyhw{eg0k(#u+8MBX1<VX=C#HP{1V12^zhTIW4NtUt|7cmHnU)EQ`OfCBC)pgn%4M+
zVyT#9+OV=8Y<$$YdJK!pU{+jF3QW?Q((LL-Y?x%JXt`It18Hj#)E_+V)+9?2(gUeS
zIV_n!lPj&q?#uiexH;Wm9d35~XGJ`NSDAP5gSyS55Y11+D&Kv#LE)45cSP-rfY!Fz
zm5&;?0-gy1cZVV$X$=*$YzGIO57bwE4*aDRvoU((a(*2;5F9Vb%IHy`bbI3~?+%8i
z^ALOvQ_L0{(rj{$YQEq&Lux(Mfz+HRy(!kqf>H*D;9);m?RfYPu8QI1oV30uDorW$
zoCOIyZ^9di?7t$qZR3Rpxe0tr0B)<w9IJ&;Xs4x4&1T02+o_xX-QO=40;9kb;lls4
zCpU^xfkOtgAm97H7GL=OIS1rS6bBG`*s)``8S!6y0PhF#(nvc5UUpybsq}gZ1fEQu
zS5_ZfolkbT^25&LSX*AH5Nxgo_ukH~I&ic;@DOdjJRBM&_Vj@sc2p$enf2fb+Vl2I
z0Cw$KWu=G7-mQDE)n|6~Y1nVg#-VBCLrY=!*+>J|ry?%=(KMc$Vtv!@6$}<*^Lc?_
ze0)ZOwZB$w$ZH>h-@=9`e#OUi3*#TUY@0(8$0tFcRi={qv*VYI*Zww0v@uzd1_1*j
zWJ0l0KJS@>c^rAUDQw%5P!7axpf3I4qzmNJA-C%j{jnyGa@U1Tq<<9a94Yao&B&Hv
z-Y)r*jeFN0qVL+g<d?JQHsyPg*f_lv&+*(^=SrT04Hqf5X?9@2f(PPt_waL&zU~xM
zf;E?Ju5KIn-h-RC^F;p|vF`tgU5_S&T$Lh;%J)p%7waF^x-;IJ=j{dwnd2D6$@C27
zPcOd{+F|v|JFR-#^_`nM#kZV`AtAgc53QzHq1;~Wc%}C^L_^xuFe%I-`gZy-G=|3N
zFkNo(M!|W}{YBN*ION#y&zlb_)spm%pHl_7+Ykr3TA$kl)@<=Pd2bXN_x^Pq;c9b7
zOJ8#NI+8a-!}(~4ZbjDM^?5{C-%>Hg4*E60oTpMTeU~dah)<QS!IhX2a4`2Zz#UR3
z)_$Wh$ZW|)!O(lj&cG;^waxBS<&<#1;ENycBd^ymGI26fJ5959Xc%!sI9fUCuBSS^
zW2&pk!;YTxs^ZI=6;4{dLAkJzih>ZUaLf~Vcn*?BU!K+s5h?oA=Ek3RI=TFo8&VGm
z2j>l~7%5h(5)^cFxooEwtq*3r6d22P<Vt4Oqe`o1tg?@59IfzPewH6ac{NF`TMSLy
zQ8>9*(Z2^NIbcH%_+S6p6*Q){>#;+@s0l6TzPs2pv%i!GjRGU$Ynl$=SE{IYl__U|
z!F?K1BtkXaPnWGMd`{s^xsX%31ICn}5L3Kv=#s;%W^Qb~?e>61av2xVqKBHjXUf+s
z{r8|Y1jwgHy?)FU9|W?d&MR@RIadbrygJ=5CI)!e3Up3&22VBK*bf^YHip~a-yIqW
z44yo#oScE7AVZ4)27ef<ZwOAz4+gwJOxg2(+0H@B@j)%=cL=-=>Xja=$JYsr+B=|L
z%sG+nx+Qu!Bgzp+y|~T6cIm8Zel1I^qqkIwiu&RDMnW|)Uv$+41+^5ayv{Q{r!z>M
z4IN2}PM2cXfFtunxwxi@H*e@j`vrN85r%N0K{vL@2LN^M4Rk&4_NH!roqv0iRa&M-
zx~{O$B%WKm7*9WO`=M9im7u;Q16LWN!u9ORr#J;!wk6N1a65x$<yyqdWtm#MA{)90
zk+~^i&R5y*spIraBbrxRN;P|FwLKXn{P5dP+HtZj+P;q}l0-zJl9Y`#r_>Ag-}6%*
z^<2J9EjlY!fI)qS&7kpP;FQwyoU%H)h|zQVC@8j1=!8#PT>Nb(Qv!j-HZ160WzDFY
z3_eRTAD%Da-&XVzZZ@B|GXi~mpwZcO8t8x@(|4N80#<)%U6S&@+>Kg&_)gjZMm_Q2
zS<yF^6MzOBBtF5;!F~l*`pLX)F?5x4Ip$n^fVS()@hysmL#Q@x2xZurq<>k{Q&~3G
z8D8sitIgg3^vhYx!@v7<&N#CtfQ+JV_F=c&!mXU<bHpf8a0*przX9di;mOJ9#rIi<
zi^`mw9HYGAN%SN>8<4ZS1l<{RzA@J$i>gKsSHSc&Kaq?+%dY6H0;&AC=|;m0YyAk^
z*B*}}cD;tL&A(!qn-8QuWVHn1OF3zOXMNU;Js>mhaYg6khH11tErKS$h}MZnXx;h}
z?j;f%yddMjulVtsN%2;J=0kmwggq=NV#MmZO>W-R`(|yPUp5W6)nrlOf~1?)hw`EI
zv{YolFZO!^APCi@>#yLSq}MbC_sH+Ne&7^@=fvX0s(GNXi>QoOe|aC^n88TkZSh5;
zr7l?UZ^x8mUhV?-piEiJ99H+jQZ4VwwhaS8%wP0}!7IlI_l9Rx9mNt|JtPY)y1KQR
zeCU%~JkM7XyMvG0RzA0<RZf{biXsYk>)X)vSt~reuaD6l-{_akiJq}of9h)_-{UGr
zXg$E*{q5-JqNRa+^ia-cS@&>4aZTr6^6~HHljYZT+h*IhzaIAeai*>-Fb5s0%Hs#)
zvnP*R9N`ML*%y~8mn`uP7N4EED`Qw{<9Ewxb@|F*>51V^<#DPP!Wp8Pc6~bhX=Ay)
zWh;Eb{%p@<=KW^gv3qL2PG*=1++WR>k$7_tQ-34t8NB+c<?sVf=cGX)c1^FqQ=v9=
z?I4jhk|NqQ3P!stA!-r{RRf~obT1{HC_VLY%^eE0?w;DHJZLn-V^D2c9<#+@?#gd}
zDcE0TQASvgF1GyQM8VhKUIR2J^Ucb`@Rmb~2!4`y5r^Am-RUja$APC(ftM0HybM*l
z)3uu3>4ghzyWWGbla)@_(S+qk7JO^Vxy`a?BhNHQyPZZk)N*ncesYUKjvUnC`Bb+$
z3R+FWANC_Q7s%Ssv1)w`H(iCdt<TqT&-OO46cL8E=_fxq80$|f6~c?+`;2|fXiIAF
zg6Fcg>_%iJMMI+s4WXL1!x)oh7ui@>&*STc79<Hd79<hSX17mxS}z{t!H~LpIX4gr
zX4I@zb<Kgsz5R&f7s#mn`I2JGSTO0d2;m{A51lg*LryBmqaNA|)4$Zt?t%hKqQvxs
zboj%tfuM1X@k1HZ0Gs{;|37_h_M;@nqZvK85vO$E4k>Y*F{##5{lY!io3j5cWmB!?
zB>37c{<{`N>xGk3TJ2-5-F#z~f~B#OWWE8P6`}^UfD>{aY-%hik3zCo<fQ1V6MkvL
zSo!F~%U_~h4Ad`&%hq_Sx5!pxcOp;_MuGo+JfgC$u9xX8*q@uttGHw-<tz8*VVixd
zvc<CpEQBrF^r-@#>7ly>s~4s8Mj8{i(>Tgq8VB7@e5+UCCz~m)+a+#bvToYw9=(+d
zlIyOeXe!yrLmZ2<WFTkn^IZzmZ4}YdLATC7c!iP2HPk0I{`07au^YaJhlfjd0iHEA
z^2Z72xgKXF1N`0wUJ1O-)yo~Llu?%?ox{}4#g?PR9pX@Gtf=0<={6zGts6Y6XRq|%
zq4Hl+gD&|c(54tB`_f*g`?*}VDH<mLwsPm|I#b(l<zCA4^{fwLdI|r>-2qXd1%l4M
z8ol!|uj9V;Y+jVx;2xFMvFbmLR2m#L-M3`*NnPqxtVs5Z*(O;&qYNVEXidh}Lut&Z
zy(EP%S)L_ZyJ=G%@$mJ#`pvf777gg`_EV8q$$k9GS53#h%#;q>;6s#BccEeP4UX8=
z8Q*PX@xz2L{l)O-(9t27tEPF3?C9kYWLx!uCuAOuLA%$c?yJWSWwnR776S9<lnX18
zgX!ixv%;qwSi4%U!|LLDE&ue;<R$w$5n~@<+A0RuP{Dum>q+~RiD@iQ`vtMR9_oq1
zlBjFm4AqcV>3toP>qQO{Ww9|Ykp%t>^8nfKO_l?l#ZyT$+2%Ox16grT^tMNzQN4?f
z8>3a7+#ja;(2NDmTDK-V?Zvt2r)B|{<`}RW7iTa^RyPR2!Ew^MaK3#(x#;P-T$RsS
zWhT%S6iF>GlvN_T$L$cjg(7b$dS#&T{if;mo%IDq#$uIsU<sbN_M6#w5rLPXV|A~(
zY0psuk4v@_k84NvSjIIESFKhLbtfeXV?HBh`9aQr6MW{Gx$4-xdqUFs^>=!_9msqm
zE-TLoEm6(5g+$wO=m|JAoY#Kgg`bU^jtI0>*t;_<#d5o1B}TGzHt`Q1($!mkW;8eR
zML$z3j?}N5Rr0)UdDwQUXl)hX!4V@2++VYQ)%4J81;;KMmC{*rP-@M|7qI9gGcvyX
zjRhe*9}!Wkyzzmt=_jpPh1dm0>3h{lE%li^ZEkUOeQBw}%dR<BKL`30wkZ~G!6Vc>
z<FHFqlG6~Lo7?k6?NQSj(gsLJ6uq4MbWs!EEZn-!sivKO0E5S7lvHkx({m-xyzot*
zfwA$*RB0y7H})6vMLx$9SL1V@sQcT3k7(<IT1=-;t2akg^%-{QgO6#fJMGu?DnAb$
zx}4!6C<_n;)W2*;C`^cOe2n8QLea9f8%2FubBdeD4GaYE4Rd%z7L8wWa}AwrVooJT
ztVQePM}u@UcYIrP&*SFZt4sG*rPsN9D`APNVeA_5XsQ$8E?hk)C7<*6-fRt-;z(_&
zs*x$MnB}uhIE3?PIUtp}%{Lz&VI@dO37&T)&f`%iH2TzwTnm$zRQzS@WC~@((j277
z<K$z>eU^W2_k3w3whyXTt7k~?jIJYhaO&VS=XIw*L2%>39nD!5``7YX19mA~ZF;<E
zX0!|+x|1oLrZoPovCGM}(C|he%@3pjTpVkdW}ZON)Abb2O7a%PkBqw0OOk_y?I+Lp
zz_aaFi^#juzHMhDJjwEQ^R15MyG<NDnaeDR>HWCkJeQ{o*UcmQG)8&r^mipxc|MW$
z=5IPWH2O(;WMlg)lW}WUaGAur<Y~two!3)C7FQTyZn5JHF#Jpls-^6>L}q<>&hsY-
zQ1YYO?2CV<m~y3ec6P0}$o;`uBp1Y$3->WO(fqtoZ4qMrOn);@L)>8L`gbtId01ut
zukB;W0{&DmpRXSlXP;z0OVp&q5aCH9u1$4ckC&__vPnR#4$kZ=RSusfOSP<5#QOla
zlR&o%KzY9Wm4r7+Idk^Q`0%VWNNtkSrqh%y!2o$AZY!~S&b=63oN822>hQz?pO9Y$
zt^HRo&&PV%egPAM?f71;gcIZ%OrZdsQS?o*YYp@*1*xrXgLEIQ)ui>OjHR$XK7?Wf
z9uREaqcgmHO4UI^V;!3RF8A#nS#Rj#9y*Km;``jJ@Z4a#qHq(pV%VREecvb$cJXP1
zvAKUH^mCCC-Yb2c+=Q~9$Qh{+fc=Ky>b|x15B++iAv#eXCqReyfSe=KA^W`La&?dA
zHaO$ndsC|ayuqMTz>X?UW(<FDmOJ)xyIVOt-uuqi_?KZ|t!#0Cjr*97J9Z12iLV6J
zGwvBD#Z;ZCu$k72!wdbs_)J<DR>7AG`9kq(cYgHTxr}`&uNzKpL->`bD>P(rd{y=^
z9UPv--v;>NuPZH+KZl6w4&BCi=}(!Nn1~j(@`RCXa9lZyAjnksY1=;#Fvku3VS?@9
z9R7Qzqu4SK?E5w4Tga}ldsoAg-R}rdQ(0&1K7HGSGflY}1=2U#_-TO$4)e#^@PnM;
z_C+YWub2yDkJe1GPkyotjCICRJf<6kpkp>R+b;dB*@ON{pLD>Z`lf_)!nPqO(o~f}
zC{hS*7tJrnCJd>@(c@p@hUmBMK!qhXm=zb>FR;T~3nBv!gjnCN(qrLlpr@?xCZ$QF
zb-7q-)IJ2*XpZ%~@)w&7N?L(G6&^tx?@DNHyCCyCvLKGcyf<y7SKqCfu_w***K}>d
zJhNMADyXS5LoA#8=7X?sKUwYKiG+FDLKjg{ya@9?AnWz(DU1AXggj1<HjA}(2CMXm
zx*xuchgZQ7q+*7a>#JO-X(e2%g$`LfE>vDMrpBN$g{26F7%IEAa_!ju*w(wKvH!U%
zozk?%3}2{&)05yON?h|!Ndac+GyP^`v}w$(vQslUlpOtj7#aVd_|xNO*7xgDyg}Kx
z=+}w5*+`9rRey9y4C$q(XSff*DtRf)o84V)n(3|lQ01RKB(Y<o;Y8N@0x$5Y@LLrp
zZ;{2D^ql&;3n~e`Xr1REmOQ?)gmZl<wAbcy>~A+kA>|>u8?kJO%#Su#lD|PaeX)8r
z!iL>0y@~IZ%XL#pm-q2EIiq!{!g+@Azf97-t*QJ8^#xz@(?+L;0|9Y;xQt<qXSzek
z^xnU(lrpMk{KQ|9s)vG9r5Iy<ulkr`|Awc2+Q6qax}WA}?xq<*WZq6jTmzUs4jG^i
znGKX%#5FN8iGPu8I}+4l=G+@88d~%GY<;74)f)SW-(F_Y&(5dk8!lfU%C3}Qd+rB^
zmTM#GLEX?UwGnjrr7@9ka(_-SH|s4}O_gWiEA6^}IWuj=@xtJ$V=5)O^fRKd2tDn(
z!aeaHR3Y4bMCM?D9sPz}&t*4-t&@N7-e0yz@uX{uqLuzV0Y3Tt+OJ#1L77{_@^9#a
z52Of%!qx_h?aViX=EBOB8wo`vvA>flEvrVANlcY;ZhO#Ksk5+W0ukl4Sv<{%>rf3=
z>E0H{b1scAGKQAirDA)pehFck4kXPNqDIDpc}sBdoGhV&F&!fDTdDj{y(6#eZ$$)A
z?lXOi8qh_iJH+mPSLbz_$VVoLMnsZRFZFq^J~)Kol<ce;WgnJ+X#M+g>rfg>TI<qQ
zen%^Nk&mF{r@xMdT;B%2r*F>H2<>K%V6VguODT_>BO0t_`TE0$9Fs81f-p;U!UA24
z{f-$sg#l6A03BAy8;{akiS5Jat=A$`xDRmF#+`EAZIkRLB#pFj#`1&S6^fqVionpW
zEcc@OppE=D6*D@%I4^!LciRP5x+j5L92=(PjnpO%A_FlDb(|JOnw4-En(5?dtn|q{
z?a}VpZOvUt$$tIDuY_d*SX|VuIvFN>;|_cxDmUAIWl`fW8Vl=Wii=$)(Hi0>73ke{
zMPVz^QZu?~Y=n~)&`<L<%Iq3y<uPQo-Gz9*B8*h=ziS+csREFUzVPXm#=V_M(MF9y
zJ-u|eYG$jBhEtHG7{JWM|KT$m3b*zmbHM82tICnRNnLH#Ild}T-E*gguAjZxN=^D>
zyJ;r(XL)Ff{WbUQ91&?a3;*%eN2oTF5AeRCeIKn!B^Wl1uQ};aDCLN^VQFM)de?b(
z@_CJ3|6Ncoio=4nE*(PpB0h2w$FIh5_^PV`TSY3Em)e~#;_jaACv6CV{3fq>+W#?k
zmO*g@YokVky95vJ?he6S7I$~|;7+jM4#5HhcL)+7xG(PRt{WssxRbBy`*F^#`~OyL
z6>Qba?9}vh_uKM5kEWj*rWB{K@b27UAP07yq(%%SMiG@gt?B_u9%IV*Ph=WJ0qf(A
zP$@<WZ|#Wa!-$PqLZA{q!_j+SO}8uKhS9oi?ljBd6V7ydFEKKui%}lpi6{^8c=m@Z
zl39nz8Z`v9!Z_;>0lc_im5VMbf+4To)?jLvCyj}zk6NytC26jr(C-x~>0v`pD%W`P
z(Va}4txT=3Vy2(mr7#B{ir}n{^!>iwy<XEfFSEy=;X+afhOb#&{34@geaPGr^00Fh
zgOlCjJ&lH^@02ts<xUQoPtBCXRg^S}&pJ}Xlwv&d@M&P4SqlYHJ@olyiXiA$5!<m%
zZu*&yY_b(D=~|-|mO^$HF|)q1B(i{jV?HnT@la$p!Y+qEL%{lju%$G2YMFbj(bgV3
zx3oJxxAcHK*41{_91btmYo+gug>|Y=S*4{`*h9%FqBnSF01u&7qCCtfKDipb#}ze`
z(;$Z1DyoQz4$(-b5yo$7W(LcMZ=R93J)$A4D=ql0BMxAGIsp}GINF)mwP{CfDB4Y=
zLH{f4WC=sc?m)Z8^FhlAav;54T_xsEn;hD*oAT%jV{3{s)JV{<R87ALRwRf7X*1b_
z@~BbCBR^2yh#nf-wo*@n#+>teeA3x%;|42#*n4-d2V2cTCEJ=Lmd<Q<PJQ~E=2IjZ
zDTB&>-bqIS!5*`|?~!Zw#K+v(@P+5nG^aRn$1wbiWN4>&9r!B}h6IQ^md=(=V}J(O
zk|leLCiMruNRteR)kE@F0`DjR(=CEE$^3+L@s0_kf4<I++W$%0HQvLOlwJTC{Ji`A
z3yqo+j-EDu2BvBML&a=M_gEXH<ATzq9Y{N{Mo0Qkitlq2qa)Do1ty!}mn+CPod^l<
ze5InLc{G2c7U`4kOXh$-XhurYYy2{mjV;oR$e)BKs1ogg$vHydDIx{m^H9of+s_VW
zr?n>>Rv%<hIICb(A<couO04AcBu$!0yGe2VsKz1_HW!US3P!=Sk^Sb8Q0szqr2sx_
zbTXk_(@-cOx!J^-D|0M@#L5Yj5yV-v5DV@$vr?+_@CJTK)1)0hGNSKOQ9$~jkFP~-
z^HuSkpb6#e@C|a{7S3J-BdR~CfV6pK-!xK7tEOUFH-yuMnLaKaK6Z}lfHEl^BCq83
z<sm<B@0|G}N<6PjS)h;L$l*|uJYCx)UN(8b*py^=cOJeh>zV~wGjq7T4N0SiUe=ai
z2tpKN^&&uPZKL@M9)oPRg#QNh`A9%e*pfgn>`cCfw1D4S>2Z4LcyPAPRV1OQUTa7u
zyGbLLeLs)RYOoz7%YJ0o%4MU;u$fst0`gwK3M+7UV?cMa>J)_ofy*<<h!~h+DR*P1
z(-nPJCOOjEnT9D3BWo?H_o_TNr}$#QVFp5pr9q`FniddQvMB|oaSd=ZidO>GkXz1_
z30&A!3<t$#vR1T4wTDjDI7XQukdjXDlz-><<$<){zTnP}5-yIJ4|p3|u`IFL4qaJB
zNO+ON#u<@@4~>5KPV(cWsU2-;TGj9v3wL8!PPVnnkHgJ6nFQr3pzS>se_6q{-Hpkp
zZ|e97ODlL)aWe<^gOOLckGn>LmRF<7Y9_Wy>npDlr{>3wqUsce!|XIIO(z0AR10EY
zcG`kI?H#Rnx|6-9@1Krl=3?(ma!OmqcM%FhorYhFYuKIJiI-22IQdSh$EzS}ryg!*
zILq|J2;Zs>-f7JGk{Z=6^VYHZc-KB}RWs4teq&Y4K6PYyc3MN=t9y`IvhT+41W$Xv
zp7O5#5!zZ12xz+S79ID>sH3Q`Y`9ZUz}ZCM$i&t%<|QQ15mV@AuYI3q?S57exwhCb
zjx6nZQ>x~xGVSwhxxt`SUMjqOTgMnN{b9`p=TpX0&xmz^XYC{v@mW2!<R59I?$uJM
zfDN7TG?t@=&cxQPr~AP+{{XQ~m(#jT2t%FNC}!}kfU%Rw#Zz;jQ=D<W;S}Bc()i2V
zVMRi<^kbaBPAAzs7S7|en>m_cKRfDPw$(~vYu3Br<P5^*xV1kHdGBhi^%t%cUji;j
zjo0%6p%-9QRZ9BLDf`><%NH1NlZyU6^_!1)H`UozVdr0_aq+0YuLQdG@nkScfiDLH
zTB*7C);Ur8%m0Ppl2$@}JpwX{BeBef`BrI)db^EK`SkY-f?Af<7#XRLNH|Y<Us||p
z*DRiuQ1_mSych8>z2Z2!UN#VooFP)V>OL|gqQmoTNAz^ov#sNX{tCO^lrILOa;=BD
z|88DyztZ9goF~W@+$heqr;ml*gw~Ocq~ka{EBRd;m%&9hK(q!U6Bt<`;IK1uIy)5h
z<@Lfhk~AHWN7jFG&chE92y(d#Gg-}t*=_aLY(yn%5Ljf7M>{;=O-TMRHjgAX)DooJ
z@_uCKNDnpkUU-cfg4*z^fw$2GM%nXY?Npxzc8R~PqwV>TE8;b0gyD74$i>+<PHS0%
z<C~7EqaBtMo>4Q7nwVTa^3<wgw}L0=%n730zPf>u2##xE(b+$q=}B~wz8&r6xWI=p
zi^Y{9+7%?58R|IR2~V`VHz^;dvf8%Yi><l`l{}g#%?oNZ!;#?XuWbGacB0X5b<)#!
zq~KX<F(aTer3!aLHQLvoGG9<h>oFjJD=rjDKIY`x<yL{MIZ(AOIL>JDIT{l8q54}x
z&}?QHj0;;BKQ#`D=(I}70DoQOQkV=8G7_el6JQ%UBt5e67_jq)!E|5`+w-No`q#k^
z^ZSaXu^_CcrV};`+-VBnFaIdasMHtLio52EKg&|K>*YkJ51V&0lZZ}p>un<Z@KRh8
z(XOb9GwKL~W&LL6@78XO_aTv|fg81dc}T`?k09mqU=we$VGXqVLIbWPmanth!S@Mm
zsa+gIeUJ7^*Pou5x~lm7ZDhL5!qo=|YpzM}6Elky@S1Y=hTmP!aXY8${qJ4?X$Rku
z(n5X|U1$HU4FT=OmN%fsc?7w!XrWg-E^zg)$5~-b@b%!oXIiwU%(7TzuQl-wekPk=
z*e^VU*xyejxSHQjpLm#?jD0=w7&0_iIRpV!G+|?haQNB)B6IODLu3oKLgb4YDBJl)
z_8t^Jw<ASyW<RxYGu7+Y2t+WFqeHQXTI(60X0xiF8_vEG^5j_R(kz{jv<jos_oZfv
zyz9<(A9YB|4O9gFy|zk?@vK{2Q=65pW>|;c$MW17n2lO2>Bw<xXp_tg>m041+i@~5
z*liqm?8QXg8%pEnBo__4nx81c6|0^d{;tP58Prb=052*pB@p}=sP8Cbl$NC&O;q0p
z#0hAvp|;JHD(|i>X?e#Vw+CmOBN|vw5)R#k&x3ph1o4@<BJEX0z36KUnR6R`upPQQ
z_L`?lDZ0`Hj=US|OJrzR9UgzkNymYJl6p*O^k-vHy(TiHxymQRzK8_`pQ~HG`D{24
zIREO`Y@{zyZX44>how`NKLS%D{<yV^Vc>3!TQ_s4#4Tz#RPg3cC%XS>@4Xj1T;fEv
z2#wJna<lBZ^YmeyQZ;j*`qvOh?GAAm)x9xulQHSF7)VSw525b)Q7U3J89Mj*Ttw>n
z0}~l}EqljqP{=pc@q|;Uv&!v6v!(B?Dnlu}NL-jEMOt#|7hGGOddvu-ZXCZgA<pUw
z?wNAMekkBe{IeggVPO&H+#$@A`?n}};yCrvhSrqdtlf-YiZ2bxjIdf~kTI09zRl^?
zKbbTw`&xCrf9jD-Lt&q<K-=0|_DZQLH|3FsA5vns#EuFJ^dmTO?iVH~4du6U2N5}c
z*YMt5M2)k@hpGA-Oy!a$Wthj1TzySH4!Jy!-<rZvEu?cg?ROy@j+@%9e5cR5HVG8I
zq7-594_LH(L7`rhp$gXpm{EiGf`(QCS%p>vOBhXsJc+QO7OZM3dhk%60fyYQhP8))
zPh(Sp(b3ll=4TSr{x;-8B%l)o-tqtVfQ5l*Ckl87|M{^Y`=0^Ye;pjpl^CP{Uq=F8
zhq8yISpM^}|4!VS0JgaN=Vkw$>vs5Wocdo^M(yW^08RgY-$s;Ko-Z(R`@arm|NsBC
z`!b)P;+sC>B+B^;t6(iiZ9R~X)vjzAU|IN1BG8l~U5WGT372qF6i4S#mBNW59ys8y
z5kp^z2prJ~qhE<sj&Uo0y1Lm+1xcLt3LHgqfP*`0*ZhLxlQCf`6%>$;NH+Zw5)ufa
zptYp6>R$Z>WEEoaxB|Dh5~#oeQgB!wwcNmyM{60XbKEAmd0dqpUPe+8c^(^!1BY@V
z`ggWPwYjhYzCRIeFd;Jm&YM@sl>C|b*TSn_$_!X9)?1bO2`ljK;Aq4>{zWJMJP(Kf
zbKU2MAmfLUsXC+FgTDm;AaN68Vz{hwKM#q*F|clx$8i`w_bM_P(;~0Ve{Iu!;{alp
zbNt+$2dnzB5y;4Ym5slZ?#(J-15yBr^DQc4sWvb(^Jp)x|Fd}Q8C#wDvvbSYTx-M2
zr3lu=GWYnRTk8}do$6!I-U{7Q^i*h)u;s6hzmY~F<~cBENvropw*C2G;I5N2UN`+i
zHwsE%mC0GCWlnqGD#@W$c8;KbE!Vi|4-0xl3+63a*bi$>B^rVeLSo!~QcrsvdYc3a
zGz(P4O)^h*O`aL#0`pHfTJ~b^O#IK6Bk%<X`mxSLDfzWOGllUNn|QQ$j-Z{uFMXo?
z9&Q6O328cpA^fZ%KDNB@hZ9WSC-@PYKp2xtoy+LGoE_KHq(}USaNdCCPz|PV<%GwW
zMmE)I-qr)DfXftQQ^;Hi^7zN|L@K%=epEEgsb-j``%F9{4d<fYYgdh>2ZL=QkteNA
zK?zfsq=u>=6O-J5P~Su3cV4f*T*XV|?IV$@MVT_8fD4)At1=L@6~INI8VQC8C-+zW
zT=E$}Oo&f)r;LRi<)0LnaVzKFUv(J-Vl^{M3zl)L=}<pRHG%Sna99CBLxRvaaat~N
zL-MNrQHof~ycdfxE8MS3GfxHmlXVxhxpJRg0J4EhSX%gkpggj>Be9bZ=*^%pz+}xE
zSeo}<<v8#5d3YUPs`Ezs%Z%t0;FRi=s|(8>BKRv)fT8w=Vw(1t1pY{b(aGSz-q8e(
zZ|dkGewCXN1g0wJ^A#qf3&daQ=t>p6ntN2LO|KV1$a9q&b{uN`ncr_WLFLw9vh(S;
zPgwxn>JMYC=*x8+Vw*UmYfK|cOEmgW^hCO3Ii=sIk6t=aDF-&omkW9?POajHkMB!X
zV8)kL_wHBauS`Up&n}!De0-mU8XCCr(N4P)I+%oqsD#J*4R<IV66B-=1z}!=`7Ak;
zV@9NC=h8bRgh`q1T`kf1sZ(?tFvL}dN3b&6M#$cOLNgghLGK|ECi$(ie6-&5sZZx4
z*NF9Hbi~U3Z^zuQL71(qD*X@swO{F;i1P8BY&Tx**)j{@(L+$EB?1U;DMpa{Cd*L|
zx+zG*YR_zzL_-f2915oA=%RA3=b}GB9|_pEl24x@GHweFbp&A~jf;N+C$onho|rC8
z@XIJ9-NNC1mf!LTne`&w%NiBnLJcM0qWB$BoI|)owH=$(+=fRotox&dkXuu=Lu=i-
zi`86vt4>X8$gNVPwjQ&Dq*7w5F00Hg$OWrjaHs)Mmf|<cQ8r!a9C4HeNP^yc(NAP2
zZ}^!V1*u!`(H#s-fb{l4g9U&bhT~(zkOh0KAL-XYkfwD+vfPi_gIeKVI+(vPMyQd1
z+~V7u89)c$_kTH<t1fvB5MbR{!e7C1$jmm&NF6^1;bZ37E+jNH;C8NdoZaP^5m?X2
zXq`Oo5`qP$!d5hjka#8uSEVYxeWQBn-oh5mJzg2e^$xlaQATtqwiHTmN8^=H!C~&&
zijH0C+u8BM*U<oy(muj{fDsi}LOL4rF^fggI_?lZ64La-kNL{B<dZYYEKz)JSP%kn
zA|%iEdNuO?dpZ|_ykE_KBu{kGNqNeU#7(vNHJ05mkG0q_-anLidMJ^3yOO>sg9?4!
z%Js<Y89I?2%Wd+bHmSqpPEp#G(T~#+{UExF9{Do80JOkmcV{CM`AW5_HNI@dgEn)K
zO>^djz$m|UBex+Vlx%)tYehN>H^iqEtoE}cLqXI;erk5|kH?j9YS-eXV+XB^0%p6>
ziX`q^k{Gv#I^Ef>^^~a4J~GY=xUxf2eHxvO35H<#-p>hnR5>^t)<01mrztqFsBr>@
zxf9fm+^=%BzwM_*$C(nig_RX&_-Roy1Ywc8^GF~{O&)X4ME=qocBhv}UW(s$cn*@O
z>4gUdst;sL2m)j~n1Zr5L5GH4zI)Hi$_VHS9Sgq0o#%Me#NBgSr`KiCmGy9|El(Ma
zCXPBK5ukx@uh0SGrC%_B2cZ(jgJxX5!uC)uS>(76`Rn9u`=xX^4pPP+K|v>l-|BK=
zX2~FmlwYQ5WV?J7H@u=Si+qUPCf7NkDOhXd%@LznDu{geB_W;>DLclth;Xfii&4w$
zgK`C03WK3VY{LBEns<W3Hh%Yg>%O!CZmh-PT1ZUKBvH`=Z?oS#w{ed+pFA&RPwVW?
zepu9`j#g^HL12)_2r8N;7Rpsr*ecPalY4NJg|5K8Pbk<h!dE8gV)&waQvejNRsGE6
ze5G5`N9ZOiI%<~I@BC2AKFHY<trm7^h*bZn4C6a|9@K!`pCAU?CbKtGkMB3k>Ui&Z
zXXdC7jmP=7jmUQi^r-sfh=8_Y#4^qNDynjqS^D*0&-&_o6$UC<j(lRg79X*m$&l@X
zs*)#Nxwc~N-U@aZ#C-dU(#x#ND<#|kJ6VVg`mT3JHglt;CaxFH2)`u0tQ=`SH@PKx
z6fA84VHtEt*E;0WrwArC{$C$ZUnX*U+!9Q#1AF{+O(<hsoCWaam(_m42&$LV8q$iJ
zgfqDt*=h5shVs_(Jxf>$Ved6(m6>Oj988X{m_>#exDhz-MBDRD)(H;yG;O!x66Df&
z5t(m`)-T;9ak(kA1yR{+(XE4pZmUC9&u@MU5a(m>Deb^Qz7{=5jNt(Ly3Vh*#K#nl
zVD<=3I)34%iz4QTfG>6=vDBqVKy<BY%&{zPrAjfXj#!?rKwi<4Bix54Jt$iV#Zlo<
zH-*QQb}>A+Ujh<3QsHi^mOmn#vxXn$u7?x%bh}3KYnj5IHL2RZq70;s{~f?d<*Zh~
zjpf(9xLC4arBoc-4yV%joEuG5MagrjX0R;V70Wo796k5P&>NoC3WhMj#_+s;g41D5
zw{EG49nc(L{Q%`~ML%^9`-`D`QT}OQjSwS;hu@z!v*@+)^2C{XQ4y`rqF9c^ue^TJ
z;p|<>)^x2e0}?vSX!tRKu4_&9Wls*+80L3&=f}iVDTQF%$yNK|9uvV006an@;t3Zf
z?{HQuW4*d;a(I_8ZtgrP)!#z+RaM%N^@!EE%Dlg{SbMR6caY#z%{<s}B~0?>gFH4~
z2|Kenh!&2qm*d+8>s{%IB1k9^3=c;>B;kjtVMU+1z{{5ZKyq00^pwLyAK54Rx;VqZ
z$@v(XoV?vnA{|uo8x^wGDn|2-K1R3n+r+>~&+B;)jRM*bV%Ql;X02i0g$cyW@r&a4
z#58Y0b3+-tQ8P#5m4BI?xhoSoL^m&79*MDBqY;kUxzX7pQ;zcY4v9SF@1z9Ijm;JI
z?G#O}LNJZTP`$M<l4h3qfDbe_YT!-65I?0rs76GHjw_f7*-iw!()clA+C|jEU8ae$
z8}V-moIYnke)L^BY}5ksLcB_{b9>O3*V__r)KSvM^t3r<7^*N2BdkdM?fAN#meq-}
z(*tzz+d*;1AzOv%hd52v4T6LUzg)>*JiRC?N3>-eTB>axR-GL9dw23uQWxV4$`)lf
zU&A>hz81Jod>W0r8o%9OkvmEo5a{Q3TF<$EjklwG)WT~X3k(=n0XJA0TH|S!!et-a
zrrz>sK#{9%D^#0~x<>B$+9f|bbySSXjo;Z|$;S35ty{0$k_9WC*=pRlj~HBjFS_$q
z<EvvU7@vjJw0nNPg6Pq?wy0aHjyP=Mzj}Nx7RjPsg+|v~7t(;NJ}5K)oqwKtE0qx#
zRVl~RLQ3)wzaPX0P_8zBkF?M6?0@adb>y?Gy;l29Wo%{R7;Co%HeYa!qL}gJUw4Z!
zsoqzxhWi~FNJ0zLG~4a{PR=2Lj`p@vax`f~rSK$i%^voRNQT8umfB5%T{PT*;ak#5
zw5ddHY^jdXrL&rC9qxGye%U>*TfSvsexaD8n+c1-Ku2$_+=OaEyUfdd<(uNBB;!ke
zz5EV&Ioi50sh{$Fr|kh{+%R|Rh>*cV&7Tw4R_m@Z{0qu5f!~sP;b@SyW8(cE4bdrt
zw})VT$C*m#QX3JT3r0lTDuF3gQmyGR2fzf_EERHWCh(b^dG>4VhV~{A9I01I>eup2
zhm!P|W5ewEKw`x2b8EwP>|`n0M}LH`=?_jX8{|FX9ZLpy7k@Ado(e7rHuaP^0}!ML
zFvn!5<Q=Vp0{I3l;%DA2zTl7>Pj+A<jtt5oyc)JV&W4&j>tM5W9%q$VK^FQo1$$Y_
zV0vQ;ZUAPPry8e8M}F3rlQm+fo4xwUr9A1bUV0YY)AW>{D47mSmRFz8wv>DhW}R(U
z=Q)B-1PNSe#uY?RRdA0RTs&qMPg0f6_Uu7Ooy@Zb^XX>br%b7CfONCMr}qSVx+Ees
z=U**Tkc%G3V-@fQ2e?X+OhlR($2E^oODHet%;Kkg6EPCFg{EU6YFY)qH4S__^c?#*
zCI~GRl%(63W*}A(D(zC^6oQ!{n(5f944f#r91LItb0tf>GTkQO0ScAMkK}YjoSh|x
zuTFQ?YKHV%U5s$)7R%_mZd#gfV=Sd=0O8e!84rO6wMwLnVfG>a+np4m2v~kpA4Np3
z#k$*&m?pOQlQgbb&aL6t*pg%PPxkFH)KlTu!rF{2CE^|VFt*_XQoV71X$5Z0#d!XY
zhFIDfs|2R1)^PS;)Lv`0o>>0Tng+bmFedhX7ATSt*?eA6k4lkNTZvX<r13dX=ool=
zZr?gm?+!UzClMgvnO#`*Q@m+J;xk&BjLLWbcBk~WS0DUb(r|Cam@_7XFnTl>Sf<1&
zDaBWx=#b8}$DEJ*xS3O?9vjbNK|EU~#b^#WbD4`Hs&<ejC_Qjqx*n#}4`jg~5!^(Q
zb^5ZFQ2i4W&m^B-L%$K6loGfaM~2^oZ0n#_1yW*Re>yq7J73?m=C#UJ)1extw55&v
zJnfbREvTKQ)$|ap$UhptLQ%wK#mYqk8DBtVbN;rF*07eI`?8)P@H3Nc!Oi{0Fm)K*
zFgJryl5gk5rgeu58qN2dFXxdB&WiRaI7T|y4cNbf)=cfY<x8bL$?A}L!kppXk&byk
zNtdeCBw9q&erCAjudNFuD8djo$-AD^OLXxV$LV={d+K=;ae0cW+mSdHcgMO9;mG)Q
zmAmJkyrxslkJA|6lxS^o<xl=d^p^rhFJXjam%+UCzV`5l7Mp5UqzL{!8vCgnn+P01
zZa)`E;;f5acedf-!E``9@sTa<ULUPup%r4@zGYx>ipOxPUK8r-eNOoOGLN3+duBar
zngRW&)9f4Wh?OE;GU{ziSvc)&wnAKgDX(8_jk|_{<67U7_5JE;mJbalKBuMEGI@vj
z<U1(c2P!l-_+%TVbMatGwFYDd<7aprd(br4`sNArtNz_rO_~1lS5a&9@2cw1k#bU`
z0a$|vLe1!LJ^S`s>@P->B<FWc7?pS~h0Pbc4>|0^L_29FFH3J+i4gkdIbU!o#4sg}
zWYaYng1`rJ;9&xE#VqyAds~b@J&pLpoO_+dfAMJ@^aQMJ@q2s5XqJA+E?^sfxX4I!
zO0&nitgI{GjZxBx{2u6RYMKaHk2d+tNwTzRck0!;Wq6-U6oAxcOfKZAr{HViRo&r%
zN@i)+_$NquSRB)2=*%xTKtEl)7OxcUPMs+>tu5h~Vbg5E9&)`kYrG=zK^mX*y6O)9
z`cQ31qyvPz2U}=O!AlJM)xLX$XRe0T73uL=>>-!L79(dFf?Od5t-KpYs#w9<SS`@F
zv+9^y=<xS%@Lkt$@e!J3q|?`wR>VRYmV>WEZNAE#J<k9356X--GbCCNoxDqwY;*rD
z{iIdc&v8TF>mx-uYDY+Z`Y8#i^mo)$^&Mx?;x}D~yr$NC1?ryniDy*Fl2H^%x*nS2
zFU?M5gB2M<^X5RGX+_G=en_8CUiy+W%ymjZdeoX%y406Eq`av5rFc9?9=CUB&p~8o
z7xC*z#CCR;ElA@!U!E>6A|-g+y{qAdnlH2&Zgtt6mHt*~CGmmj-4+g=oExP>$(6xC
zJe71HsJ}Gb^R_50gBiR&n+x~h&|Xe2P7TJ35{>jzZ)ZZuc4d)PW7ya>X%lwk^=u4X
znI9K9i?~riN>d}r_gQ$Iff`2L!DW&YiJY9a0~@+;nlq@CY9-{|j<6q&C48M?6~LHJ
zzXTb(uO2TF13Xzxmh&FIn@u?T<^N$xHsuB1%KSd>aAvJp=A6lR?O#RnK6_qpE;oo;
znPu1_`_Y(mA}_^Y=D}=6zuO{v-0@b>#g~{`RSEgKb(H7tqUGh&icSqDCovL5O^r!j
zjU3A)=$wR^1du2P@epiWgZ`38sl!PX*yQ=9&A5F;6e<r97Y_?@RKRa7nfnTsz-~-!
z>8v{4^Z-MK1&*|}u;r+sZzrm{eGe8|Hfz8l;}4M60MRjddfG`d%U~3<1EzbIJ_XI)
z)p6dYY=s_CP$jxtddhrBbfv$TiaC9ffBnHiXhUU7T~JCT+Qe**$+m;V4G%bHJTm{Q
zs96|aqHADUN?0t5MI_A~RL1c9sFFg-L&`x3+8FFE+8O;tts(=1`XvKZX$j_g_Z3j7
zW)Nv?v6OXl?u!nm{;sy)lOLvf{N01AHM2Y}jx<M^EY7!EM#m~E@M=I)KXh3r(s71*
zvRj=MWf8)1&PRjY_s**v-5R9oNrkVf)lO*3(B)Bs&ew6+D#lDDnG&z7`Qpbg9|uzt
zB3#bye#OI07_;6}Gn$~or>BdFHo&E(O>@$=#LmlU<`sT<pg^^H3z6s2v?O|ZoA#F2
zA}jF-jp`N-nYh#H_V;4#9^=!5{yd;~{kTplC!#q>4L>bYJXZ<X4bN+^s<9Yecpt92
z^Hlsx)K-S;ehCv;#)IN>nQ;6tGAZY$p2iV0sjdOt934>R973QRg_Se^J&2pADI87<
zD&_Q>QaUwm_B^y2#g1OVWAznrxM#f(Yql6Tw6UXHHS4tF2wG0j2db;Dq0NEnw#0r}
zr{$8LNeIenhs&Q>C6*~3ox!=;?L^xaeoXgWkrYZ6Q7G!KWn3La9g~wDE(1*k9jW(G
z!Fri?jk>dZ*l5_7$KAhyghY=Qwe$8)0pa;Y;=7bnp(l;56|4(ud1E%A>a^Xh;Fq4z
z96FlwT$0OLP@YhC9U1^{*Eu_+C8xS$1ZMU&&k;fU3Xd_RoZaseNQ!8aH5u_lA+=MS
z5t0OXq!C|Gb9GfibM_?z#a_QM_4LEPH~f_ZE$PjOS-;%T?Q8RkaU|8{bOlb_-(ix^
zXSiGYTQ*lP`Y=ggyvD>eJQC3cxznSJRHUfSWJp9rG@O|sw26ubR*|Y34Zir7A?jaa
z6Ym=tiggdGE!7hzpy17Q$SF80HSLhjJ_q)Vu0Ojze)Vj&gTm4^z%MZ;uE$@Ff+)L3
zv3%0`k#n`!oiP3Rbz95^_PK1DyIMU=P$Q?z*SIiW<k4D|aFl#y*vdpSm3N&7y=D+(
z;kaP2)*Z!`*d>yx`uh9hlPgzpTd4BY!Cq<-%JA7F_hpZ9PzRwu*)x_&J>eu!_&GE2
z*4*FzU;i$9aVwtZHfUI8S2P4jg$<<S80QBaa%iSCSidsr>nk8q*g84E+YcqR#Qx18
zyZ*<q{k67v-}=)}Op~HVAMSn*$j46xxdW)j!>upgMV&Z|IeY*jT2&R}n{1+UIXPmt
zyj>3VFo#{vCD-eYnl$M$cTr!*rU{K^UnFBYz#iUz76X8&fJG#bNB^<9ivIU)R4Dku
z3M(Rpg2io}<EpT9I)MY(jLMT0t2{<?qV~%w@f0*gg(m1<+9qQ**osopYbZffA@onv
zH*`;aj8aX>7T-@@-KWBjJ{NWizDdstbKGK=YFcEgN~S{a0)y8C_9^ye#6|9I!xj~p
z=9&rzjSfQuzeUOQ5ppx<ZJ9IpV@KjCnl4|`Fu`?bNxH=nVfVA2K@6N{@idaeItoQn
zO4g*~QwksLxo1yA+aXEF>1~i&%g~ue<RePdSqa?XEbhVGEI$x(rCyXAODU*`L9FeA
zjH$g|vagtEHkV@wo@F3|x;#Z3wo!Sb^`2sLIWJ&_6t}KQL#BZmOE*T2gdva=IFrUR
zG_g<HE%jncwj%8RjNx+qLva@lOVawjJB3w4EBIwU*y^ugQZ{bx0%H=&n7p~AT@(mu
z)t<7+r{hdt=$>kRC`+vrA4Picn;lCt2S60k*ViZD+$V2dY;JB=V{9=6NouRemlm<r
zs4+Z%F4u-yH!tD~_$jV3@W2$+AQiih$tk`J7;#^?C3^nw$c+hlyQ)cg<|%)G;iSkC
zv7s?=LW-v;3ad9Qk&tQNQz#2f*PLS9NDd^@l_+vaD5ifquafR^*-NNivTx}a|I@!k
z#DS${F+(ZgF9XshQK-ZQQCFlV@qb}D)3aIRXLtgkO4ajl4Jf4iabt4!YE;k{vd&K9
zk<q)P!IsRhi<OT2fw4hXbQ@?;Vx#=I!O!QJAkV4LkwPMgS{lSA;8dftG7=U#SR3?%
zug8e*8(|$0TFN<j={vtay{bJIKbt^F$Pp82^l-5g)|MR#MavTUkxJSS0Z-HQE5<$%
z9g{yKhgC%`^ZFXh_>dOqtRIWiznuCFg(qhE`K@c5Os%6@)Q&x1CAs8mB1QSC&H7&2
z{(kDWsNr!VIeK;E&#3Q4+S*!+=H73aEItpIazriLHvW+k*Yje8ij<uoN2$j6=zgk`
z{~%#ZaS`O8GxIvgW^}yBvY7<z&KY~2ow~*|mt8q3co|Og803z5a9j&#^`z6Qd}aT;
zsFi-hsjR;7{vi@d@%4A{U%!R7RfP^-%!DG)Bt5kYvqc$Uuq2|p7x{tRvlfV%HkdUz
zTDO>^c{b5$O8eNyH1y+V^pBQ8FJ;EkI|XG%6VGMciT{8m94>?hL`q#78rMIo-u$|>
zb3@PG&a9zi63He33fFS89S#5*v;o8x|G4RZ%evm{OxyYAjZ6>#zP&w}3T|w5YLDxw
z^JtcGYRUcRCog>>+|29@W{LEe9@Eor^mYV0Fi}Aqxl7Hz=k2MANA5?v&Bihi@^U7;
zGWQ+cV>Ae%Nl7g36A=a3O2+P|6o`)D&nSqEp-!|Wl}Yn=LH+MufE*=(mh$5Aq?N!U
zG<<hA3ZKU>Y+b2K;lDMu_WN2NGL7pH2;hDr#9A{ED1RIPXK6lr@yxQ6g=v;7>|5t9
zhOBn-m=~O?UJlrI38w|qqe#~|or^O0jH5XZ<;&pA+Yf<v0y%%t@KYYp`bt>;F3o*}
zXgojwk~63Myxc!+-lvpB%t0&}0=m({JnpG=o1@$7y$7gi!Rdb{*B@rmm!WqG6cL3v
zWrjX;ssPRoAn>s}t#YTOX}*M2lNGiBOH0uIP>+!kp#U%XELzUAMDg0*MbN~SVR{b2
z+11tcFP;2(KjjOFKnehL&X-UORq!HCj;|jZ_13k+s-$J1OE=#H{L><54?ISPpzP69
zy3RpkqaN!TX~xVX-o8&Nl_twiQ*y}V{AFrMmU?(K@NOgrI3A%(Mq887-_Sx3;^Xhg
zO!2GGK+?7HpAu7{==B^J@Z&VAR9vZJzrF#wf%1{Bb?o>IT4=wq8)b|VaM)&Ous|2d
z?%}cKm3D_Ob;v50Op}}ungkZ|bBuuU4Cr7vN%4<~E_h%`5pid9yk|-O6bmuz%g@t$
z-;!0BWZCHZtNl8mxlEOgy91>Uk*c=rVj~PnpooZwsDP`tg(n#gVpShI92k;pFGN2)
z?@O4tWP>WC%D5uZ<S3%e<xNaJL`pzYun5kvR>Y5`{`5mwI?ox)?n9>kt`+MhQ3~=#
z34k5s@6#{WZd&ImCVW<vJA=cFM;sb)MrI7yA{*q;dK&Re&sV?z^25j#-!t3n7f}f~
z{zg>E@xvCs%+{jEy><F5diGCX2(>qycm!Y|y$eA_;_Zeq3@xhrj@4c;#M_qak?A~y
z@F)${^0u^<N#VZ++xEr3({DOt>k$eh>pW1)@!KV3&?kJ_ouT`)_u*|h2?~|m56amb
zOM^mM2F6PKJ0U0Nc<pjK+$IP!WHgm_@t2R8{nJfMVWP3V-C7l7`8@@@ihloLILe=i
zb!1NH+_`9XMWIFgu?ntMUQSvrd_q0WjU(*3h8jHK0X1g)mGz~}AG!Pq8@ee2^rhvR
zjB*s}(hF*jH}+<3dvx<qqvk!{tvZbUM$?q>1B{V)DVonxHhD?*80NlH>`12M_~?#!
z1eE4aPfzq2z@q~p$3Q}xD{!bYhU9T=Drk{qfCQ}&i2r#Fh@LtturQK11yh?iHo4>!
z$CGI1`9){z^L-G4K2-N7#GeDE8>%!lZ%SM_H*cV5(PpnmGEhiqd=Y}h_AZ-oXkm|k
z;{3P_jc)UYPD*BuT;re^*fk8tWydbh?F2HNy2B#u-6_?VYz<XGo`-E@$Si3oYJaQo
z<}D|{gQLS%XB}+W6YaC{Sq~0u4AK!b+#zu8+~Wm59qY9Z{FLCyBd%9HoXSC-4m&e@
zcFo?`&T|;muKhc@#O^3HtAL9!V4K8<>)u|~1W_N!Aya?lX>ex!y%+qJT*Y+*A3j5D
z8&K53hFQNasy{UUg6#9&S3N@KSIxumDn=SYs0cA}pCBtXOdZjdtH`~Twb!sRycKs0
zWT6tlCbfjNe<A*r-Fs(;K~iw#EJ=@9YeiN@0Yd~;^b_KTw`Wxdc<L*7SnTQHk9YOm
zA6Jxc41o}~pbJIA`|vA(QQqTvg-7fSLfP?H`52Ba9%}F901$9~2Lf)-F#vN($Zd;>
zK*}3`1Dh~XFy;3cB>Gya3Zt@UAmf}%{}!-iMpZ|47g%x!Z_6v^_<U~?Tp?2aeRSHE
zISK|IRq%aDjZ@o=zi~6M1qmuXRR8wLowaPx0OiO-hoRp7WIrWk+R0ucTX%ZzFVig>
zgkRqx{C~1h@~E8F{$*TzFMjT{ABmgs^QuMsv@84}vs{8zCEgJ_d65x0@TOZa0v>Sy
zqu9$SrwfiqLyvsWrE&ELH3gc&VhLe{yxc@6Z0J)s=INW+3|vc2g9g1;XEGeJ6IW$x
zSi06OT&tsz?2vayM)$ADA11&6P{{`>q=Nk0NOtEQwm(Yep_$ePDBur=)oHSGXLs!O
zLLZEqJg56_H9;XEz4=$vL}Aeoz^Fl}rADw96-KQNz|0G>p3nE71_rafC4NO!CyXFp
z$Vbv}#f$G+AS3&990X{qJ5@ccOVd1=^HmLJNrGQ)253STwSu9g^Us-e7}29OVvmIK
z!HF2#EAYW)yE7K`V~0z$n)dn>?6fe+W@q9oi21nKPGba|CBXbACl8dWo8k3gdL3Dj
z{pKg_#3V*lW+G|{!HlHqO<gDuBBS0|p0CHr>1iK<WC}Hj*SvcC2eEv5w|Ct`IUzYk
zY39Af==q;^V}y}}o!9wQ_;(^i6twjVSff+3y6qzYMsE~=$~jc}2{i9q0NGu0Kea%?
zyry?-^}4}S9RXnXPBN)r0JdY`MLAuuPuO@@m_o2;SY#v}TW7iP{;cemCQ6T2x#LEw
z6EOg<FV<zg1%P9GJ=S2l3AFfh1sc8ro}qv6&fCYm+SkttE8NPE)AhXsp#(tf_p_Ic
z27%PumH>QKa&O2W-ELT>CTEx)!(+?B&P%02S-r`(={dRl*(|Uow7L6<p7M$>tk8!#
znG*P!6e*;V>)F|bYQrZ=HWm}KU%p3QWvj6fqKI~s>%X@5w{h(Kqv*%sdz<s13$w$}
zP@3cQNbuZmc^^p$)u%>$M}%+97)&XvfSs!h^b&W9${<8c=Q1#Hpr<*mYU5yzrSz$@
z*4ldC-xd54e#S?w=|p%Dnx<q)a8S8CUY4n8uS1e#m!ew~NxN|UEcQYc!Z55fB$E?s
zbBD}qTc1tCaQOj+|9xkU<Jt>pa<e)<3<bQZh~Q_d@eIm6?21SRSyk)Z<h_xXw(<xG
z#Z1Ii?NQzNJ4%4(3~T?omn=*6PlBcHMqI&RseRr`sORu!uHf%L4v&Mowa?pM)|-IC
zY#Z}aU4%Lw$Ma!3sm;sAN^eoqchjuCmGuO<Ma;Nqkc4V&f89=ZfkYABMEcl&oP9zL
zb1CI-o3`cFxM>pK(_iv-1sDI|VEj$yRu+Iva6Oy~|GK-u5aTAcK>c12BsgX-wNQ8(
zpli&+S%uk06U>zDonF<}@(m2woed)`CC|^z=Hji-3yRde=4gCO!YPMlqv~rAVc(8&
z3fy|(3%oq&-^8eupLv^I|JtzauQ~)r-ts0^g2kvm#Y6~(LPd$)ADd)x+hM=H!*#I6
zgc;fXF!ikS7hns|@)p)hrpz3<oW<@pT#s}mx&z7<&HyPP(1H`w!~^%wW_Ol2d=?3X
z(X}mK`~1PW0-V?+hHpdm$@zh<XFve?qbM>l_|~uw#7PoYZuF}QozTjZ3gK1`f&Y+}
z&>kuSR~|#H*Sa@zqbUU1Obg8K98GFGE{r3qko|na*mBkmIV1EMztyKacNy_r@Yz}9
z=KM1dWSo`p{U^)7UAHxv>Ge2o+PaU}=w)qm)rOW_aH7FRIsR<?D>L^%GAs!3fuoOu
z_Z70ek)|$_+DHB<w)Bdyx9P#pR=0)t>%G{UG+eQEQsH(u7kK|VhAQiI-_h9bga)v@
z560p#K|KOIy5(atdLLF~%o^vK-<KFY;SW7;ua2wDF=)iR^+B)p#=1|aKt%K}@%><%
zLvo}9HY!cCyA!gO;<E0ju%bUx+=6cw30dYe1;xdmt_E-P=lu8<5lby^Y0*tly>QyS
z_Y%cjQQna$(jaqLs&(5Pi)fGffrR}^oaM7AGIWD7!_Km#lQ)+p7dFrI*R~av;@Hm3
z$1QAuMia53a{oXR*^y_&UY@r<9|DdgKbS9%BtHj<)zP4mD{8{0#n4A1nrVn#4_k6?
zGT~>+qWM%{*J|2ap?_RxO)CHCMC^d6&_SEZ)#A;w$^Efr30qmMiclyLG`(~Ll}xYB
zdXv_^qk%Y{WEQUXWd5F9cX_6F^2hpE`_ZJht`0Aynl5Jvt}d}Qp=lt(2eQr{9?4sB
zE>A$?Gq-uk*riws>h~rKxK6A#Y!3q(8dijo{Gca0fF+zp_ktnhNKMS=ynXWhO(Cbu
zwnhMuQ#34ok{2K3nwszt|GX8PLM@<32lbvJtQmj41mg=wyVB=HLJ4ue`U~t@C~aSP
z2JCx-fb!k_S1E1yF<9O=whEKo<M0mc8F_mEsIvm*FP7EGh!qXm<Wwh$9?U~*<}?Yf
z5%JN*^foNiG0WeOc^TmK4#%L4%qKzeJQ9Cm;hAEn=$NqaB|(hVo35mQ)@ZcOgk;#0
z`P_28T>W1Om$tODe4O2n%H}(OB50Gk*{!8vJ~1AOK*E8_;&(IuD->~tACf~4a&8RS
zy<c-S>kWf%=Z<U17&KBO;Ic-&@>}X8;Iu?s)Gdi-@9uCvoFInmp>bbGlNvE!>i4y|
z?eYJ8+~XL6mBX(j6Nh16B_HT!kmul)WZc2t?oi01F9{R6dLlJ4GP2Yg`q&=@<vOn}
z>-%eRv{Mcq(MQZ_vpHtV@WCO(kc<=RTPV`Z%0e$7`s4=|{xBs9DMNX??H3l`snqa9
zhFScK;AIk4%hs;4)Kg%HnbCVUB)Dx^VED!Qe2)E*Sq@FNQdI%#c`)rNN$hth#?5Dz
zhuMt=TnE(gj!anB*A_0ok^;o*Db&BumJ&e~`?_4(q26)XTbuVn|LhVuWxwz4%#@BD
z096>#pgi5zw5E(xnq=?J$1UZ(#zjgCd3%r)^>p!NKsq^&9(qXEB4Qz7+cVR;#Sp>a
zMD`rH31b9#f)*>v!M+U_oLY38CYMR-Q5-^^=jStoOy)D|{pB5kU=#TsIb1|@>}Kyq
ztpmW*^BsAKd}O^FE4eAESCRjZ1j<<^G^h}5ogIR|$ZUi-B=2g=^vTVQZYqKB;`tC(
zer_VMC<B03Mgp2=zGfr`mW>U?lXJrfpg6OH|IjzL`bv8eG^I!tc_sZ!2N*?PJiWaa
zV=<-BKCL!d5hIX#V*Z2KL2iQR3WG5ubT3kyc}*w=XWO>Gg4zZ_ll8hg)UTXQSnA2L
z1%5d+iaPwQ?OsrvV-4S`J%@9MhKbOZC?`2JaUmmAiPuZ*5FngKwf*pq#hv$l3eNf<
zA!WEo<Xud?pUk?+(S+H3f1+u8(OaQ6t!lnf8ISyKtA*RcNBWX-7mr2#>Lg2X^@u+y
zE^)l>7Fw`{G$!u_(yY1n?&2g^P$;1)NKh2{y;GWN32bjmez<P7&$XK!vPfZB4$*wT
zVmJTIekjU^uorsq)fq-C6Z>u=7cn8*k9+RB>Jgx{+?n1%IN0hu>-PUMfAlxB^**fb
zrN)V|ZntiY7v>4$a=0h(D3`=zg5#e#9&pCa%+1ZI!7XiWf|DpD2>_9WtaA1UGy=LO
z5dVj;zt*f3fXSJ06kg+$_@CsWg_%C8jdSbX>*dQ7_R8o&*FMaM7nNC1fJ$t}C<0X&
zGI3W^ciTzx4bFb`y$A$#*+hF!zjYqvKWFhc;7cHa1&&Anqg2E@>v@1GK6$><geAdl
zF@##CA`(Q`I<L|$r|z#z5BMwn{~fQ6%MtkWd*gC10w+P%VI0s;{>cpf^PaWqHwch`
zI{j(w<(N*x7Jj+@W(as);(-zQ&ED|b($eR&lz)KWcAk$H9o#V1L-YA>HNFcQftIr7
zd6?(iN#}6`e^r=%8lmbYh?w90fSq1}YrdZw5l57#N@hK>ht6NAqlu+{YkzL};X36M
zY}y}(KCj<o&r|TSPp8*nkEgyL3*+zaPYC#tB>`JvSJzj-nCS-O^_(o#eIQ<PG+qAK
ze;Y{P@RPe>EnI@~<@a&ogXph&<u%~()?3)4k&W&2%O+DJO$mz>+<Z=yq*OCro0fkK
z#l<Ea%HboAz-uv&Cv`^6J|~)XSEW&J8<*VCF>;GUbBO8aohat8TpkhWOQiD_%l;^X
z&u)jVFNd|Hz0udg%7TsRdK8G!bX{u#RKlN-^Wg20R8@%)O3cv3iH@l+#0m_+97%5j
zU0(eW!Y=*5M2E|w!{!hKas&Z}!&#t;0iEt!JYZ4@O?_Th1nO@CF@k4lU**CXA+Niq
zljI#~(5E~rX9KS*qktLde~+8Z_|py;7HPg4!YBJ~u`r?<TBeVs4agcXb}2zR?^GN5
zLZ!}6CQ($7n~tcbu|F{-h~RTpz`f@ZLtDNO|Fuu|cHkQ*rQ02YWe^{#a)I;8cgkP)
zhcl^SW=%;=OL`<qr|!)z5xVH3X*@nb73OR>ALY;|WoQQTdQmJJ`)X1L7QIvK!k(vt
zx=$yinNaRV-le$7j7g>Ms*63;`Ha3Qx))N%QoeN)E8zRENGr(e9Hj_Xv;=L82LRN{
zkC->}slgPAEJKJx38ceVvLwlLX5mNpzW|f+w0>$C!(ee$2>}4sH)wSn^*ZnO2YiIg
zpGmz5(%oe$G|e}E`u;QZ)Ir$_Y%|6fxf2vtc_2;mQ=p%J1xs(llg<H=s4T}#Ig)6v
z=FM7k-b+=QEwemmphT^rq`A3ms~S{6Zq;n?r&rBadm*YBtja{K*s|zbEU0vN1ZbaU
zOBEplX47{F|CBHPfcXDxZbag}K;w6WG;sY#qDWj;7GYy!BP11%M9A9_T~JzG%Eg7P
zG$k`6IVB+h9iLfOyg}D7Ns@c0-gbr(pUoIZ!ud1#vfRhPX3$#QYr_&^0GtkZvT#FE
zRRkGqR2JwE!y|Uj8eEZvjWM!B0mmPdQZ^^Nh0^89J07SrfJ)|YZH9>T(vw16q+c_;
zu`x;Nv)(HUV1ZsXVWNc*@msNeeH>;zFHOT`AM#&~Kn|ct>f1#Py-ysq;sR30!_!r~
zr{%&8*y%^aX~HUNQ&t(}L8e3TkhY7bHQlyP&~^)VidEXxj!^vlfk|9{n^T!C&UMpZ
zMa0pot@VbABT41p0EQz82(ht}E$cdnxMf+OwCEuJkGoz21dbiS*jq25>nK?Yeps{L
zXpVVYjZz{e8g>;!s(8AhNYmbHwAWq%caRk+UEo@z0qbCs5j3c>$TB%?3FHu`w(p(u
z)r(cg$Dkldgb%i@c%!D<D=2~rfBg7SWP%Z>(QfE`&FtINce%@#f^SlBocC@{A=?)v
zquzdybGJGS6Q2o%o_VZ{*?7KpcF+i*yKfZ|1k{CCZN!mQjVAI%O7tcgrCfrgnNLnj
zYBVi>AI}UlFs&xJoo9Vp$cY|pg5;as6r)pPRv=z1DcF^w0?b&&lT+Pj_MACYF%h<b
z5l-b2dw8bKv%S&{Uk2)z;<e}}nwWuY)W-F8zc*ri`BK;p5X0?oU$_aB{t}#i^UR2x
zS?T^@i^H;ZjT$|E^qT%mGG$AYy?*JkcJcSRS#oHiq2+FFFH=*JgR^0$K&#bvKQvn3
zQNX(yFwtsTvRX;luZx)2vVdx>|A^%39o5j5cslSdiF?oR+D|D5$q#}*I0FlobL@J`
z(W<&hoX~L5xaA>^k}%)9?`&(kRNUs2mu?$+#1iE+p<`%e7{?#K$zbDgaT+L!U(%1-
zl#gN{!>KC{y}e!3-E3mygFU8%kHjoKJ;vz0Y>#~~mH*gx-%a)}x*tj)i%Uw<`+^E-
zAtxqyFe^4r5Idk>U0V|X46J%UYXQ{Honv{~qoXx?o-Ff{T%5|~(Rl5qN(-OoBzZY>
zLhYKX!)W=>W|nvq(m|O9)sTnuXW9#LxsUiS>i6cO;`l!$TobPfNDEZ<r33FgX&yX%
zn!?qZFmZ__dKlnK7RjmP7v^K(tf%rPt+w7+0ehSb!(het{4oHWIGhDcMCg(QJ(|_F
zJt_PJe1vB4JM$W<2Y@Zy<~*xm{&|L+qUww)oX*+)f4p<CRdts4=PN)6ebGGU-I=)c
za>y6|fiIt7_``Vp4nk@X<S#$<s(Kjec*Xaw^bdcZ6SFw)bpOqyK?z?EBYb+uzS;pe
zu@8{o;SL~7{KDiMt(nHeAB8x2&wak9Wxj-W*7+Rjy*$)yoWT9L;!S{J{imQ@FwZpy
zH9fBMnuG*a%+(fWmhh9V<;aRn6mhUuIMemn(wIh~NbkR;R;Q({(~|x!`|z!g7cPe>
ze-tj2=;lS7VZ4Az(`&HCl<C=GjpKS-GEG&dhgc0T+|7|h7l5-%Tf^Uv70@No{Vl^X
ze3mGNd*rL46&Rteac;g5q7yRL$CUxO>VaXmIosyy>~Yh5S<QL+_CTf#!&#KY^L=6}
z=&`_a9YS0mYq9y)=jDm7XJ@4UGbNnx?Z8|B<o3hOz+R%7GK)*#nSoJ(sP#PAgPnip
z7peq-{JugnRH1!L88Qxzf<->|BjHv&*S`fOVyPzuKSLNBENWIefD9rO?8WbutzWIq
zT8mL~H+;@oD<p4ZHgb4By}y*F$D?8|e?bk~FEsdez=e5MWyE}r=)}<ABVC-lP!!|;
z3(VDW8u8V^kXg^-ImVS)887aCM_!m2Z7SqiV~E#Cw-RAxi9K(R!|f+SLb~;)6bgg5
zC@iKl>S-ZZ1P;P$rTgzgFo$X4k>uSDLaB%E-Q+<)(5kywy|Vo!9!+9EfNLnLxt7uh
za~P~z@#Dv-aw=>+w1gBro-aN@JfR01CYE$cM#f(Ps+c{PR9a=T*%Enh@F3l?1k4W*
zkj18@1FSs$>mi_YY780EkI&8=*d1ySOlIdB-4BQag)%YyT{4-Q2Ha*~(LQPCh*yM(
zS?9u{nHlIClz67k#Nm6*Y-}i(rnvu38>P<<Q44}%SD7OC2&@e|mGTyk_<BR7DbW02
z=c?P3|E`tk`Gr!i6AT>y!ouR2rh#Ib5_vF0kBTSaL1v&S1;yC6{PQzHX9;-F0`6GR
zA%e@cy#$hYLb0HLpi8}s2FuYD5=?)<am{}_LUjUQ&FcWg3b9_s0p3Y4U=WkTKHA~2
zo*;<lXdl*w0z0?uyaNM2u*-kA>QD!7U$z3xjnd=m<HHAB%JT%2Aoll><R?npf*;kz
z5|ad8KwBAwy3rIbqPV?jH?VxVU+_UJNH`n;qn3uODIQoGn_gM@KQw(~TxNZ=cD8NX
z=497oyQU{klQo%>ZQHgvnUihX#@W5+J>UA>^<j7Kz5i>i3+oD9NCU1c3)yoE9E8)Z
zLPbJ+lv!~7!~am5)ZvWh7*w%`I#V7IXH7~b%<QbcBpL2z92K0?uuCf@4;2>2QpP3`
ztTMxKlDA^5ChiZB7l<z;Jl336MIT9m2^Oxfbn|eD8D>G80<r#524ful7`#%ghat?+
z2V^s0I`L6e;jjy&Myn%QyQjhDbI<gOZc!48uem#w{dc+CzXucPw286=EEX&la-?Cr
zM?Iun)tK%=+bO!NO%<%Qx;@7ga4RIbXun6jD)c_cK;EGmP$^@~Fkk8fvkU5RY+@-z
zfGA%<$soa}{ba);fUwGa+KPnH<_4_>#f&Oc`%VT#aMM0MC`B*;w0}`Wkjb-}$FWWm
zTBdr-`$Bq(pcfwq_l6s}EDl>H>o%jUy)L}^^g<nwLtMwPd))o4_o?<zyeVcvjWow7
zGgjQ+XZEhEHP6R36#_vk2w`wXcOaXG##pP_3{w_|Sv5Ux7Os|#%zpl)9iKVq`bOQ^
zMqyDOo*snWF_a#)4C0o-cCXudee=!!T+CjDo<s&t0dS-Yfnhnj_|`-aIr2!!r-jtw
zVYb#*=#`=j0PGqigs7yNDm01?!aT>sW^{ar@*m<sA^kh`h6My17GDmzSXnZPXP1EJ
zQWeyFhg504>ns+tKMy2`)oIXc*8;WV69u>vOH-nf>eE#@qj;V-6TVfH7X%(^H<doG
zNPSJDKO<1pH6!U6p!_h=0zkHfO@`26d^?17CxA$-+aEwCL{p8|tIOzoLVc0^p?UWk
zB#?@elb;`eNw0Ags^|mf+a*Zd$!{@D(sdQYY`2FhMJBEWP1DK(L#2>5Zm+5+{OmkW
z+8>^AT}AG;7b}Is=R_k*6A`eX4>pKHb}W1qkyE8KEOja=#5|B>vr3DlTf0SfYQHG0
z8xY6ZhQzX(uJqSAaRL&;UVrn4{c$8%>%}%xS+=WSn4N~f#uC@)2d6J*);vU#R+%T$
zZ9WjOm(AQ3LJ<}t%648+-?DQv_Po#R9a@_-*8QsA^`Q7E&VVUBFf=KU{PjWA`+eg@
zh<^okWR>A$^<XVqRof0+*G(Zas+<JF{I5s*jR;&l=K;(&VAC&o)Q=y9(s>~uypqoE
z)m%`tm>Sk__7PXzi|w1w3LnpT;cgeH<~#=}CO35ue!BgCOsYO@8vi{vhCbv(V^>qg
zHk0H3$9(9~zCu+a*QCMzK^z8CF<ugvH*KIi9A~~<nVvtuM=ILtA-02KfOw|e8x!!y
zzwxJ5a}s>l6))54hN*aA;OeigR3P+=<^C`Fx_LlF@dC;v$U@*Zf0T}n<=UPDA^vDl
zT4Nw4iomW1Gy=~P=luM<p|SBgvD$lC#sWUW9?iEgg+VNM?-PGs@DwKwD?pMPL<cRj
z&)w$|#uxHRj&`W7(VGIpy{ub7>R+3;AQ{zyGL!w0!KTm+l8$d2WX5uGvJv#qMjr`n
z?)O2=Wc|L!t~UwhT@s$x3gp}V@~$b{k5el)GN#wb=BLAIuX)se7(iaIQ&q=~$J-6|
zXQ$MVv2k@XXihy`(BFNm(@Lkp5@)|{pCZJ}aICB4DNnXfy^X!ou8IN+kiv#q8rqhR
zHp5pi*vH=%UH5+vbHoe5jGCZr){K-|zG2d^O$|9wG6k>2^;Z9@NmV3x^_bH=I+LeK
z=j^M0epea7f*<SD&r)@iztj2gd4!o(>l1VzhbY{bfr@R?l*n!O`Tkh%Eyt~)AuL@H
zn#y7fcIo}tZbAHQ8kAETvcj&;N@ZqdCXRXuh*h?N+WBaV0g;b`AcasZRwx9Lw#HN0
z?Sgpb{!a@)tvCmX0x%8-Inj+tK$8$StH9659b26o0^l+5j~gDgOxQd?UM)dP=q}Hr
zgO#P_h={*cV_JZmWm=u*^Ud1s4nV=vgM5u$n3s{51#Tn^7U~9sUI1tRmEKi1E+YwP
zOIIR;<KpDBw`g*<REq%@ltGEErL4l|YdI1}37}g48e0PLsTA*Se1okTMXl{Yigv7y
z)|~@68acD08CJ<hz7=AWW|PQ(rWO;>7djZ$^^p7qC;|`%v)QFu6W-Uedi=O5cJs;q
zo^SRUw0E>DRAhm{$`{2SKpNO>RvJ9JMPY!P0K;0~f<4>El9|o-hGYNJJ`kC->dFL;
zgz5SDAQXQf+4n0TfXqA=hP35^k2-W|eLWJz-=Q^wxI7^aV01E~ecpC?(`95wCb&1h
zo10S;5)ys`n1}C~e2t(>a3m@M<fgXyS+!6cZ<Xe?5L?U*tIPG~w_hPVCJ2m-jJ%F!
ze?w3+Ja0yL^O-y^fneie;7Fp7dAiO19LF`!h^6k0r^qcX{`0knyGvI7ht<yC(C>m)
z{^f%8oR$i9=b~rehG}LU3Zy+@Z4i-|F!Z%5`Gq3dL85WX2<05wX0?4r<h|2kNP8K#
zQxn|xNBPkzG#FpQv5k$*H=qv(u(lr!#(jaX9gm%t`#`Xi84x)Qf*|BcAqoS9a3lpx
zreFDVVt<gx=@trZbCeevsp4Bs9~z2(mUo((p3{l`6LrfdvJ|WWFx&m=ZXm{umxhqS
zS*}oHh)S}qa;5*lfw(s2uZl7&lSIjzfLchng(?(5iheE}q@Mz(kz=q%S`lx_{Kf_n
zplDN^r1JJM(OlGQk?@7KBl0?fZEM2<)IzxI4ivNI8DeE&zdXNtJWotC*8vfwd~JNJ
zeT@NCl<PPqINaU@=Gp85n+`GK?LY>lR)-raJU@1==RO<~-w{b3Pu@-;DSaYb;s((|
zrLL5Db36-xKPittEx8wDWn#GFhFe#6a&3O*M+*%Csj8v7DJ@z(Fv)ZEg9qKu^g)Q1
z@T&cdqWi=%ew+~FW5Nc{)Lw8r<ZQowhJQb|7u&S8I`3CgacUf8_wfo_vI?seEx?59
zt23P)^kj$;wYyw@1ikur?9LvSd!&lp0U}BmEz{TljEwK0fvVgL!HTif9~E}rftW4o
z`1lwY(p8v}X_Lpk(m4SL8URWiV2^&ZWZxbbS$UuFlj&)o6>zJIdADjOE=9caHW>xj
zqCa7T-WFgG>DiB$qxB=>>fXAr!P5AV_#?=c32{Z@VLE#@=OoC@VL|{SuAvDKAN!TI
zjsS$R_Xa>ajt7<!cwY4);Iah*X?T1=VbXi!xznaFpxQdBFC)>~qR#E6!p9(Ja0Tje
z7_e|;KjuhxaRi+LH<hwn9c5DaR7|h5lV|oVbCb#&vAAqE27PBhkHY=~!c&3xy`=ea
z&0Aoxd<+Gwrk?$PwJQ3Gskk$a8mW*Aw>WLaKMxio*lJDd&Gi5jvYLjq+nO*mrTCZz
zCiX_tSu+r#7F#UtsSebjzgGR0#T|*|Ox<ADn>F|Q6W<=L^4EUvPUH}p>=o=PZoMXZ
zRf`a<Ky{b%$xx+Ri8Hvn?@AJ-zEl6|e%1(_dOH6_Obz&k%Bx$##0QKRM&Z75iY-he
zJ0<1jg-G*6=W5Y7Q95T|@2UhAn2$BBJVuz#79~4Q$&x}W%%!lzD&bVW)JnlVzB}`h
zM&P#_knR!HeWYDe1^Z*1i(t006HCXiQzoYtzTF(Mx8AU8ilM$7iD*L;OaLkJz`N!3
zbGa_4{?i*usQW{cfc+VVBN%y#Cw<EMqP}9SCHSkQrO6$R&}=3hgC{{4bVu10&Vv8E
zw+6{@{@_I*7R)$bZLa^d&i9q}2y`^sEfwLBj$C!#@4tcPc?gh*9$#D>^uaUl-Dbaj
z!{sx79OWj#W%AQBGlebCgeGOV?~f!s0VG*_X7ixB&ht26#~-@t{M@L%;oL&k*?#6Z
zST~2i)|__55JGOS7N$PHeO%X@d++>*dENu_pZ`5F`DJ@2pW`=Q-{sf4Z^|#`*43Fj
z5%Gr-dtIs|Jj+B9z0uXW_Xl@wKn+flvB4Gj%f2cZ*K8;YEkt`XZ#t1@^F97lTREg+
z(W(D^Zxu}F$+B|DAD@I(CeR5&(;ci1HnSnvhRs}5SsA(A8$NbvGBKKXce$AjBoj7u
zys@1_6TyhOEzDjhl4s+<gi?+co8T=z1eSoxUDO5kVRuhLQ-5}2lb%!%``;O#cWQ#y
z6I?;I@n&@dTlnitEJ>uaeaMjn?&XCu?F`J{)?Q;p(0E`7?hH47FYvk&o+&FQ)0929
zDfg2=L*u;9(U8`-zq^|TYNlML3@z`LFStVw8HL-40I4Y30iar&5(ofG2hofBxz*#B
z;c?Nb=f0oVTCD*>^YVC^9UujW(0HO8&3dB}uSj)tb#HB;r(uUXDMiWHLG1D+BS%w#
zN_k+UeZ|`bL=rkm;sAXPI1fNb<1g`$vi@FoUX0afj>o`&$NDxF7>dpdOk3;sNx>ZD
zdF5e`K2KgL4){kYmwGO=)LA^7LXy_jagx;-vb0jK_rf9Zi7<&t93XK(C_IC(L1AUW
zsJ6tmPBcDu6s25JPUFE>PP7PZAqF(DXHBx_s6&$8=g0f)Uz1k71Av4#UTLs&Z!h`u
zBL01pn}NFsWcSvb17l%{a{kwWsv--hJyz|qw3ApppRS;6Q7fGT##eR*!g>E`>IQA*
z649(+Ut19|ZS$OzZ|(9N03l)fUwBLW5fEi%ut8PXi1=ZeTM0{w!)b*9RLX=gQ`tf^
zdjeG^e%>nls`e}(&aBNvF@{hbC`&iM_snk622MX)R(_|6R;G@ag_xWEJp9i5t;7_p
z*Xs|Jb%QhBK36|ZfBkK?>2n5VhW$!AkHs8H-JG1NBr0`=ou}C5(Hl8}ENqcHKRhV}
z@#|8no(GVEr{9(k6$SZ4<n90kaWWhR<oE9;Wk9ROWq7Mw0Z{ZrC2ZA5gZQ<lgNr>^
zD`TsT1~cq6-%tIjV9}`nP&SW%M3e3zD|9cyU|LCSSo#NhxB_ArOd$oE<Kpg%;rQNe
zwjj1O2VH!_G?=>iXodamJP^<-U$|3X4upDCJKT&=&jMd9i$-Ub@ePEziF^lmV}||x
z4JLD5Bq=-Y3LYDVByE|xShxxTE?)xER6!=cZ5BNhXdaK*QwZvFN<2lVZU{s?(&uNa
z6tU@G1TKzs=uS%3#+R;{E-VFzgdl-Xp^a$7v6k_yl|C3cyK0Mx!`7_W+EY~+nWAFW
zD~Vt7Dx98iE2X~~2sM!ymHgzR==JZ}=`|zmXfP9xR7W9NtQ!`&K^KOR&`me#vOYgB
z)q!uodb|4bsX#CUAO>L<7Z;uxY225*8`n{;N|hS!fMUS`w@;4g8L2qK0S@HSZHQkw
z4vVRE4=oVI^L2HzNb4tXtB}2~x|L*E0clQKTiXq*<^CwAlQZM}R?6@cH+2Roh`_Sl
z>oT8NHe_JQtDYm|V#GVw&kc>r1RNCxyi85-Kb+CqQT*OxQbNCz!2cwW&8m}O=8{p5
z#UIl-9tif=PYKFtYHC)2{aZN^t`}uJNc((I1nQA&l4(P8^WpmDrMD|}e6-+bz}-b~
z>u%I$t=R+~kjnwj(gntM6ZCq|COp`4z{sr1KuJPUbt{NFCkP@Q=|v%qs0dqaX%qV6
zG4uI0d*8MDl@LU?`Ijkhw*OqJ@vGEruKO_nC-@eN%=X_f%2&e?jEFzzI3?5{gl&*Q
zyeRl0g9pM*a`^|b$oQ9$gc0aGk2U`g{o|wTLbZ=)*UU<=w|6S@2LALyv>!NBOMh$}
zDKtUP!bJLVmnr<ZzCOlI%7Y*HEyM<Ce{ikkuH^4#oBTq<?hYe{P30@1huf0E_4S6v
z1{S-mo^Q&n?OqW)cS;MD`eK_8=Lst-4$RRg-R<u#T_5+=j@JX&x5rbSPz4m_8nk-|
zD7|Iah6U!UTnfC$6g5e@K*k`zK`8L<V*p><_*edsqR-1L9?l_i(%r};;dq1#p+{fc
z@BE?A=_t`!I1hospM&YBSj?%%TH;|agiAPi52WO5Nb!H{|F$4zc%Q7cUMw!MI=%Th
zxiWzZFQ;s7dSh=sPrC8|LWETJ^GHwI<;Ud~d-C^}dlTR=^?bg?uCsc!>bfgrj&Uo@
z0~ywW$M!k@0kG8p3u!8)3Du#ZepUd7cTqmy?c;;Qvjs93g*TNTGsJ4t1M;Qu2SR^e
zs}iL*6{SuhWsc#-vOr~>BbZ2_OZz;y#?kxn)ZJp_=6opof7FP!Z;oNp6sRdBxYd*2
zS+!kMLy{?joVJWISL?#VAsuM~<8*}SQZBi>`>a`$?fT%1=~yky3$oSJ>s>%F;rI(;
z`cj=GJGkfdPWeFB5|m1AB;{<dXHAoL&ICM<c-MRJ;kfCGJMoxd_e~jn%k3n8Dge>|
z><lspfCTQ;^dp%7|JIt<<2vwc*8ewxM&$EmU)6em_i;`98LQ}dqS_@Z$>K9f@Ucz&
z;l~U>Yz(?h5yt(&_CVO}H83D#0K?gTQ-i2KL^s4hLZ<?>l;hvf|9iIAMdhS%l^qu}
zeuE;B4-H}2f%|4sVSK*CGuYy0@^<}EEb*ntRv?9Fy_BiNiN}@iZUTLNjq1e6k2Kxx
z3;MdEjOQ*_R_g^X=~5ah%D@OE+1lUtbb)qI?beG`2P4$P(8IlfG+`-hrf@Pl6gGh<
z0Ism8M1I{yTfmw`qaqEovOzpit-vtQ0;Dt7EI6#%H18YrH(1V2^W8dlGuaz5J<hDG
z0Q=W4!eIyhqoK64VKlC!3e&q0lJ{~JK~tsq>&=?1LK8TjA`+t0NY?TQ4ubuF-veU}
zX@BO`UsncgLtv3Yp**<}hZbQE+h{ef51m}n@V8b}TI|ZeXZ!gOSLg$#!TtuNlymK~
z#|y|&iXZ+SqYfGLV;l&M+wxHqsa+B+vQDc*>~6-!mzl+&way~5R}t2V4dAuZ7<o?(
zoH~F^l`R#<*F=qG!AUp%CtfQq0TxI~D-L5>Yq2A4{1d7Vg*wct@w<xME6nOVm2)<G
zniXdy>vD}Wq%y-u9i0p*Sy<;{VgrU8tjczjGERYmuhOr*QW9G6XAdJ<iVObvId_eL
zP*Dm0GH3@BFihCIr4bdV)UM7@9fwV>Z%Ci@3}dMzDzau1X#;=;ClR~=2rU=0&IGUW
zi$rO{_T1UwDXOoJ2O14tj!I4-)$VcK&mx|BH&HP!_|n<g`TFb@p6B<BIC7Rqk${$@
z5e7$n?5n=?^i1pZ9L6L@A78%;ZKkPho~$$mps@P|_22oM>!Oe2U;EvsVQ3+UnC3>1
zNliLTU}^^GDNHs<3wY8_y8cSb8)IsZ0B-JSRXcG1!!A|VJ9Qbb@O}L`=6ORpPoGv-
z=jlf<FKi%;NHHY;+K;z6pFRP>-f`WBFU5H9yeu|i*;)#!5tuZnyI~S*yI1e?%w2z*
znOh@#UREUTsx2`|f`st~xIfR+$AQjyW5dAzaQ$l%#@6V_sw*KqQ$Y5AX%>Z}E=V-R
zRe?zmk}4=qP_6xIHRcN%FukHemq>#ltMI8uGoyw+1W9d-pjM6~oS-tjvVvU6IM_1K
zjw15wUt8@2b`wnQn+FnuGr~Rz>*-)O1?>+3f3zmZCx5i#a2+Gp8)P&@(FIW6du%!e
z{b6jCIo2Amn*2_PD9Qb3NV4R*I+!x|*D#vmq^>fk#2D=qu3eKguS*6c*pieBQ+gJ%
zyfV^T%p19+>i?GOO#hQ4$p_n?Ly>?a@_M5MnRI*faNnyP3Jsq`zlcT_=(8-&9PE7D
zh>i%?+6szdnnD7QXkkOnec<O`ox6|U)433t%bISTVQl_$ol!*$Z#CmE@d-y;nTtb%
z;r{$bwVxX0KHlbtL>QiB{{B+<?Qsph<FNDc$7++!c*nKQLiOgbTsfDDO!Mttyy8t^
zwPzHo$sh|*sy3c4X0_^nL4p8nse}4O?7d6}UH0dWx%=wsh9m&hmups=EmUZKc_+SR
zMF1GF1Mq)<ZGhGLeO&4J0iSJud&PH2hQ5?1>KoX2s>$~I&Z41`ENHwlLG33?f%HDa
zam2k~!va|%1$&k6h#OV^UzuHZ<85OQ1VX+9``tl%z}Myg@F1*Kb2LSYjz3qYvnM`g
z8KQVy$Xu8yy-xb&Q^_}N?w*M1=c|9|k%pQ`ZJ`(4zmFk#O(@K2WkcNZc5u{Z;a)2$
zywhtuKmCf>2Kq*GVDbb>cxl{E3j%soBRhWe{N<I^{z1m{V{VI_+2Fp)!uTXK+@wfd
zE>(-lQYI9KKi;~f$r0-{?&y`^FNb{J7gp9U^5?1%5q~aqgl@mDNLTjKr8Y-XXMfN?
zPI<?~=KdfeQCSmu6<PST)&pSwsX{)2uInDiZ1!x?*gm_TCcr_O#foG9c?SKwn@zN{
z4s;&pba<yo>?{2c>ruNy*4jE8$>bAD>NrYlk|au84vyBegu7BaBm4n=S~0o2r|0Ii
zZ!L{)&wfL^=5JC@=M%0Vc0loV1ekQYyV!VKeg}jvgbm0Kqktx4qu89!Ch`mfW)W!A
zHva6<I^1EHs;)O^07`2Wy~H_qGbI8UQV^gL(#X`-s5Y1Y`V*~o7e;H1s_mI<&v5$c
zj=1~{hyN-wb!%v=hPNlK=2mJ;NHVMq)Q*2pKP)g6dn`nv_RjS954L^I*iHrLq+vXO
zE78~a4in&~^QqK4+K>8Hc4FfEhQTC3V?@DLvXKap{dp#@-Dr>hj`R)U{8n~Q)N)!$
zg{-O2pR4v&?H#C?!s=Ku>agcRwiXnYK`rmi-G|`j*;3l{KMai=??P>NYk8-x3#;|N
ztMl#006NgC<!ZgZCXQLVtyTLOzeialqdefrhdP6T<TP`4HP}!|$*7Y(=j^FjJkKCB
zY}N)TlMO>b;J}gg7P;4ZyS?ADACCwe4k`xLofh`%gBVDHOdx)!*5JM2p)x}ul1MM>
zDvd~=$Y%8ko~#T-IY!mEjT+ik{V|XOB%5Yu7=s!j5X{6evJTJt$+yIx&a`6&zk}!Q
zj2etaW`MAav7$=7k4}2P4WmVggIGFw6KE<u!)(yOlU!0(?ey7C{Lur*IOO9fjAwzx
z)h1>D>N%dmO0HM|bzIzj(H3r=?Uh6!Gp(aagW>k^e0Ur94;UFTsh>srn0=vS;-+VI
z;*1JaXsi2@Rc2;q<29}DO#lD~`X+E!Rk0O9DW-0<!<~KJDp9tu@9VUDb%&6QmKz+=
z<6{O9+x`uaq;#RWw@(FDPgT<x!RpS&qrw_QLt1Gjh>8<iq@FiC0+Bk$$`Uq!HgA<G
z`tLnhMI7nB)kuh3xgIz+W*>E?<n^w1cZi+>16kH#F|6=`H@rXADBo(pW@&3KMDv;f
z#0nUnAZ!GzX*~t?J^{^Qp~}dr$dYS}ONB=Hvn`62AHakGY~hl#hygP}IBT#&YeXa0
zBX~LRTQWA^1Y`X6_`;Q0loGy$WctX)g*Gb*`dHB#8*5~c#|`#yfUm~0h8WcPxBw?M
z8cm=@@|8tY5y!D{Gy8DVA}`S>cTd-8ut1edW_Wy!`!4VyUZ;l6V3L4BVE;E3fq)0a
zhSAH*tDw~EU!H{4sb?aUyi;7VB*+S`#%jkwp>}gicTns?tw&#8-%BXbjLe;BOesbw
zB+kInzg_L-X?NT%KrnPvryJSYrKWuB`1m*4`;I&zgy;Qk_Ft(zhqpB?l|tKD4Qdqa
z55_fz*-GHNj?|(|@Sqouz2!xCG-jo}7=)J1+n>DbPW!u6wc$a@wU{NzDep+7R>r;=
zZP}L0cug6!5y0yB?(=G-X}UZ!BPu5sj_>nRvRdKzyj*ExW30%Tx!!notf1}tzp(?2
z88$vy?S}tQ`=bc_27_C<G<x2iZykl<jR0V;mE)xAB;G)d$2o81Bc4Vw5A{Es%-q1j
z3Q}bfr(m==EeW-kQ6q+d0^j*yX}UNXs!E<Dqg<uoeIFA{UV0wPSn=9;j+-EgK2zh{
z$4Zy)>dZR+3i`TOg0L2BnbpK($9u89o@a@o7pTG|KhN2nU3Mq=rT%F-GZB=_y=*Gr
ziv4JScGSC;IJd9F>evbAUE75NhyjQBaGZ2jmou#$@qI|VL@Q!IBxrVKAsK<wdoo{s
zd~{?qqQX4H^THCvcS6I*$EVd`p*)%6dBUuSMostG)oh$=_^HwG$jXReA{k+K^Peb8
z5`(rsYj6-DGs|nPNQTGF(;HmK8l?rc4W%0WVkcGQvcS=Ngjq+ZvE^8}W#!`D*f0%-
zE)MSx4%^h2o8>9*B%b5lF~bkCN*l(8rbfG)k>VLa3mS5%sA8>gp2rTy=gaJO0y}T9
zT%M99;?rzxC`6p$xQkzcKE05hTkn_VPgmO+t{d(dJogYyHk*TggziWIcqAOd9~_y7
zWHao>D6w%tj%?w_ZZKlU?zye2SMeu<u%RupW0rSXBE9C!^705!IG#WDH=#*~os!r=
z0;O~+YCO-@Zt<FIDz>ZmFaHBdZx_bD-?#z;Gn8^Tyq@CT8~j9z*6-?~iS?9}A);*s
zN*O+bB(Yp9V%CjI9%v%oEKv{uJ-oB|`S@A3AJ+&<c?%W@U_H~FOlzo+17TgVxDXTr
zV0121`UZ=vk+EO}IZG;{bcq3R3!m_)DLPPcWrXm27^)Y)h=C)_OY4K#!<L9yO8xM?
zdx}?m-OKhDR$NSApzs3(j4m2!j69dbJL#_qWhjD7TwE#C0cB+oViB{f<vEe};|Qeo
z9dLvw-!U2zxH=lBy7e<qFbGnLRykMAdEUdvA?>yoDh64jzROlagWOjQ*F^*KXYV9t
zn0>2OuRZ^0FuOij>Y7hb^iH-F{p^A`hc(zC$1-Z3AQ<7JV0nB?4!kwjPunliAwHK)
zL31~A6?g`v`Px)+tc?$=ToHDdfM3A*uT17fBJJOV1rMD_ea)C`b3bo-WIYEds`wJd
zSurp$^j`L-{w*zto46nFLhDd-j)d9@`f-nE5AD&$KfU*l9a-P`<Mdy8zp^NPyM_Q{
z$g~8`<M=xJZT`!`O4CLDRaWL}jquDbDXS8n-{ffB0bu6~Eace<A^(E7OopONPwN%g
zcXN&_)0rtQ#T*}X)&s$t*?Ve@whnEQLMT}jV3zCfoe(*|SIwl5(P`pi;>5QUN(E?N
zy(pD^=V?$wEJ~Y4;Cn(SlxaB}st+9K$V}j;ar^0_)`&x|+>HG`Uk5_7E0@qur-M_N
z5%0@KIV(JqrE<5wpf}>r?aS#R_0uwqa*6ae7HUFvvk2YjsYc8ddjhz12a)A{m%1dt
z?ZwMP43OD?-{R8Hh2J{RHW@M`PIeS9o~+J1ZG~7!;dN;SX{K;blPOvEVBX!XZc3b3
z%kFmdx)rj*SRoEF{i{oH$kC`@E>NkoyZa}_t~e>I&berCE1<}YB^?`gh)0_-<Y}BS
zN=~shdWtABu9Qq86PMBgr*ej3o?$QfUnQh4xxA`^z`dGAl^(9mslT(aMNfdKB8aMJ
zWoXU0k}CaJoz|Z{;9Yfq0P|0y^i><g0p1Moo|pSmaxrFtSzGU0e8Kxt#Q_Rs5L?k&
z^K)yR1Ru?&S^7&y9>e`M+Cdx$IFlZWiP|A1$(SGCC5PQE_<u8O6!9!n!k*GKd~W<o
zfhJ=^&(NXQ%LjA<c8lxXPOk+sCH{OzIRSvH*gzMpQBRHmtFc;bsnyY3#(qH-FZk2|
z3M_z+K^!EZ&kME6`8f-&aSm)6OStap-R`3A8TpmqBH8cW7Wk;jr{xhW2eV2!jsg{B
z1zp)&vM;CU4H>B3L?I(*t{uCC<C*;2S^0Lh)H0$#(`~&_VOf73h-LA6l(zV<$Sp8^
zop4#f6?~um&|CX`>Z4rEUF)B+B~ORxwBbZ2;bRa1jcG`bf)nu>%EAVsQ|Olp-2@h1
zN>;ALC>^-tW!@c4tnZ8uXh!Fp<DJ*7L0itk4R_!TpTGzc@ig@w_T3C^+|OX=z(P4E
zZ^+8IpMkS+@BlHr<mn;LW8aFX<<k30lPXcj6l)qvhpT&(eCIQz#$T_yq+tVfED`|n
z-^ltBBn&;R+6Q$;uSUUqfoXiM9wM3F7NXRY$%ZlznCVPUuUWMRK*3-}Z%@||IIR8v
z&BW$(q_pUQBbN)Y>Uxo5F2&sWda*Sk^m9PA_lfltl;v@_ABV3DJUpkmc##!Vbvjj~
z7}C~bb+zUFVbz3Rt5U<{v#yvy*^n*|w{mHn{jxIXmP(cf%A}F3mMTB4U@0#rPPSr!
z#ujx+xw6e06=fkWFz4VkoNTJ}#e4SYQp;rLGFJkjk^)vcA9OJ0<$*R9O%)?cds9V`
zo>Zs1IZ4*CPZI-DR5k&H*<Ar0eW|2Q9Ynk3ISiFe$Ds8`4_2s%PGn6mM9Ik-I~b$?
z&Fex%u|}fB`P@OMzb>F|AiI>39Vv+wt1Xg=eT=6uZ{|Gixn>Y}<tPA*RGWYZ=-MMe
z<%@&^2n3bIZfs{R8$Ge4`{brvnFJAR=U~Lj+;F~7HZ^*Gv-<4g+2VSq<KtT7&Xclo
zHvnp#)~UZg9zu07J4VVe9kXzi=Ns&nXx&)IQL$!u*r+KP)?h+_F;$)$X2)6K4QtNu
zB<mAphyICaLLhRU@*F<msQI*p1xr=#YY&S=8P{|Hl*R39Y!>$&5d#0L-CzHw1?YZ~
zw?Oo9d(r-B0GT-*d(q#^en9d)Y?@Gd7M99>L{<BD9Mqm^o90im7iwhFiP0gR%Z}SG
z?@fO&V!p?KnYYW3z?VO>A4xC_Ak$|zO^yguNMT>Kf5!xoMEUi*FSmHJvlkIaY7W6@
zilT7u(xkC8m~6KA$WpbREoNiHDpKld>z_s~mmyW4XjPNg{HvDT(kqSpom*2Y@QV^=
zonxTkKzL!9lc=U!mG|BFv|j5R77HuuwVx~41<`yjMEOj}o2--nmB!(m@SD-8>eRG0
zeqUF<q(`|FE+a8?Fe{7cjGD{0Zgf?xnYr_RIh$Q%-Rwe*W1~&w{P$ATw^PmPy^{P4
ziTCceSxvX0)lJX^RDlY}OneqFUAp2Q2sEEV-Jh$jt*;mDty})^JdYW-ZcfV%3DPVp
zaS050bHrj=PvmSK9*l{<P_|e2#rV#PK-82FwkziRmOi+g3Gq*srAs2q9fmqT2S^>w
zK$Suww)5M0oHlyDV7Lf8A%B3kscG_Hg#Mo0(>}?U*xa={q1KKD$F}I5pu)V;G(6rq
zDWfhs1jBvM`LJS*%jyXM)~ng|fGHV8FapT=jxR3{eZi5o_!qs>YhR-u=qQD!HKfYd
z3_8<tp=oMRs^o3v=i`WI-2BWpxkv3k4qvg0GvaM*9Bab9hbIJ<G5>%ptu}6a3~Ua)
zC&4a)Eucwd@mx|`>dGtvP56tEcCIKid8M#qo(;#2gd;UY+(c5xM<?H_<MPuhqpHkk
zZJ4Q;>vt$npp?MWHl)f<3%IDc{_~5VhQLVk0LGqXf@YAy`Xyo`2DVdE3oB)+;vZ`r
z5-@&skeqD)Ey$^3Ep_CiQK)?f|92*`oP<dtTz2X-3RcLD;*S&?f5JC1#sk%?j7S6L
z$~jHcAPX8)9$gaTwBXA_FYK4zEm9KxFnT}^g&IuW9xWUev87$!wr2Ep{J&+(SR#^)
zjXABvDma&zzLPS@#fpjiB~w4M^7dIyu)of9jD}4<P0fBm$RBXq3{bp|AG7m5Z3XPQ
z)G_3jmxq_C7LzGg5wjDqH{10)Z|-_sdJr!(B!H+$cRjve4qa-WQ~fEb*=U|a988q0
z(fYm(-<6Jqb-6@ZbpRz9GmKHLFO)!I{x6BMBQXmVhF<RM8#k$vsCzgUbd-7=oh)Ar
zWl411x8gW8)V}9>S&Y86|Fr9~Vuv}vcx~Npb>CmsLwI6oq0pnJR+uU{zh?@j%BVgJ
z%?L{`^5Dli^j+KTBP&w;)VoE{*Y9sVlS}^F8!2q8^A33q&MQOXPoH=vtn(E_e!jeh
zS=T4-``>$8BmzDk*~%sEITV=!EriTYlni!~v6z%^A_sn}DkU!Jyt-)(ZOQIv7$3>X
zAt>s94B#f9C9uA#H92@CIy8%X;LyQAdtFx_JqucawG^T%7=x6<jV5?_yQ%Smc}mjs
zuC#G4)Il&BNd~b8$O21LRwhdg%g=YZYlnbNXK$GIkLnQu>AEZagtr=5Mq-{zLe0@g
zG<aEHpliqNIf3`nHbxM*Za~m~20gYH@0fsk&#1pB64ePNjNjGJ%7@5n0+K<6E@Be*
zyX1uzDdJp(^`An_lsOdykdTwP+65I(P7AzAK^t1_g{{|5XMu7Jeu+vH+BTFd<%<gb
zh<_Ew8$H<FcD#Tt1K2q7^7^{YZieA*EIkVs9P#;aX=ry=3r}8NWqN*a;c@-rmYLqG
zGSk5Z0sWJyJRtBM%Ml!e=@o8SgH)7Xzsea~_yPdf>;Tjopg9!9?YsbSk(0CYzvboQ
z6&~wh2YDZ#%&rZ>GdFK~-6lc55eWBe{5N8s9oW0$b6{UcV_8Wng^_xmh20kt@U$CU
z@A>S|u;BSJn;j+Wo&PVSAWW0J1673}W-O)0GWC~ME(=Tq#$HhGZ~_<#8)-A14P%f{
zHdM{VlqUY{L#nK5-Sh{H9}-EtQ-mk!TS0HRxoMnl!d?~u+CpVQ9t+eqI_wHx$jl1X
z8U9ONO{O^QxSa1j;(cHINzRS~oOKW@mI)%A@?L8%O^^(RB)wvs5)7S|^$%XOp<l_v
zZr!?JdNCzP5v(2id(E%+t|SteCh$e_0x;HS4SuqG5F(I%J0bm>!V6A&k))aAG!A|}
zlqz(`D4K@VF*J#J@UUDGN~BZ5VRYn=xTi@TH?nzr(?JC~=Q+cWh>39l<+NrxZ{m!U
zLGmaXv08GX!LuYZ475-|s8!t_i5X(hCM*WP&T*7JZINQEo<>W_1DG_5nb>QWhkNq!
znM528NYR4=O9elw+E*j_{s0XkFFNqUx&+~y%XUqnzM1+d3K=vi`LsPBTcGU5fF_As
zJ{knpns|_n68h>ctlk1nF^yFBX*KWnazm0hh>HC)_NhH~Y{S1d0A2oI{HM`dwd$0^
zNRQF<Y`9ma>xS89ikkU_hOx19^$+Lz7_Y?wMh}?__}p7*?uas%bGoY7G)FAhC0h$&
z@r<mDOde!?q)~XV(NM!GtElv3Sc)N{363jXA3>R^GLP4b69^y)6Z*0#M18xFPC#QY
z?8<n)JtU_Du<?Lca;YyF6fi^vpj-j&#u|V_>P8guj&08>wu3EF%%1Fey%+-yGTBwP
zz~|d&IvW#+u_S@<4)<FWFz?UTP5ZZN+q+X%996rQ<IC;4*-ZJplX?2*LGG@}=fiA6
z2SiY7=T%eeKIf-$jmn)cdZyy`c7Cd(#>rZZORNa^)%EzW8%=g+q3OJCOvW1tUd&?{
zhL7d|--QiaJT)?1ESDCpf}Z(r;yWvr+Ro2mAd=cG4{4`QhjLXAB(<0|EDdaNBfI9)
zo10T}@8-EbWi{K7RUfk@eq@^byBPhHj&I`{<>MER(8cbV6q<Vu7kagJIa}%lNZ#;-
zxOVRw>#y(T%O8~sG$qcbkJAP|Cyj9GMCFNd+P`Ya)snmrT%MUO)+~_eDadHF8>XBK
zu8Xlkfqv|+v4;Z`@|~vb;VSl>mb7=T(j;Deu{k8eF0g-4A@Co%cGHf%!Z#-ZQ8*&)
z*0ARxCJ=DdB1Ln+{qqf2rA8oVj)c<>IM)G?;$Ib_EbJTYr-11T2H<#^Jdq;W*&>gI
zKr9C|g}r&`(`=EAU0}-voF)NwksNV)>a=lqLEm=|gIFhP52u%7DAt@-6`zuQK8l;1
z$w)NSPc(8(WvO(Xvl^u>Vscc?tu5jpXi4G|TNu`az${{7M&N?V{&qX6J97mwpQb_F
zgn`$MvYW_<<KRnygKM-=`=Xjss^_5h%=I3AXAC7S59*=_RSfyd>(XE~e(p#3+|2+b
zQrWvi87Q+F?xK6jhnxNq>MJsmpvL=J^8S!6oJU9Cy`+Z>BEzBJ1z>n9fC6c(w;p+C
zTFe}r@wGRPMjBXe#NM#*p9FfY9+>&83I4G`J0BKz_2TKA73V_5X%tB~CRNv<w}wuG
zvv8=AQH9GgceOa{Ldz899$$W{oZx+Gw}iEP{c;3cYPxM#YNPd{N~K2rqMOnapmhVZ
zccqJ`8NdP$3Yi@YVN3~jSI%0#_KZYAUsit#XXpN~yc~~ntBOVdJni?2dXV8`HpaT(
z$!PC-AS@73@*UP$gbVGNFKD|dH<Z98+(Yb-siO0G=Lad#-~-O`pbbvz{~&9n+wbXU
zG~55ug4T`uVZ~}E9NL7{#r(D5N>}E^>?zIRf+sOlQiP;b$4!AF(M-7ud=weL#Lx%L
zfLrX%)J11R89RH4aZ6H}vEayq(6x>nj7v|N8J_2?#ej;F9iY@G6xYR1l)>r{*vcB~
zf~1$V7-{T(28_oXcud>+_hG?b4TYnjMlB@%FdzqIx16D<W3c8I3dd$1#1R7qX+Q(|
zZL0OrKy&lr@YBn@Vf_0H#e&9|qo&WKzK5n1jZ9#DcS6*^)O}pKQWl)*KP;6l0`?5N
z^aw#C%Aw+?84c|F`sQ8m$JCghPQMQ1&4}%5z}3EAfMSgO9YH@b=nkV*EwbD2_>3Ta
zKa#q;yHmn|Ob1XD8i!w>BQ^xCYxXpS851R&kckb#v+Kbu1$=}`S6+`(Sv7g`rIafh
z-E2l<R^bQ>l?#o;^9=`#isgDP%cdXh=s2L!wi|uB>CJy#=%yTh{*dW#uHe8Tg;yy=
z;oD9f%7WaI%YH`7e!nTEJ&{+-0^x$oD+{gv-g#Cx_Cr^z$sF+>$}l~q9rtH*$=js%
zs9RwjNo??Vi%bO#T;ROXnlrQKyR<PUS~xA+Bd?pB2sD>5T(L_{5}9?3^zbc190dea
zDMg=ES!s2%X~unS&B|fy`Uh-zDaEfd(u%>8wc00Vg))L;3I>l_CCY^y!mD#5sgyE}
zbIe>7HX9ZS9P;778?T>Om@(5Tzo8TDLb|#HxqdGBHYUl$Q#GOr-+YViMy%d$&Si!<
zxif0SZ%RINUmLo47?s^Ny2O7;y>bUi^FxS_)fo`Zg%fbihgd9L7JC>M7H5PGM~*T&
zUnFNm$V6(cFDjAH6CP7ehyU-p(2-G9>$V4~mfK^yq9Ytdsf87|$}|m!2Bkq0o|*(z
zn@ymzwaqZ2#crsKrKmDaF#;M|!PW*e(H?a38@1t6ALYfL+(>1{t3e2Q#(5o|$28}k
ztIPf4VrKleAu+D#{!*Hi2Q|Eov=Njtf0=k&1+X1@UTb1C$Qm&HCP^&GpWkD1E2$Sm
z7?$t3iQid1?eASv`E8M)#+<qJ#)g>NP~b7zwoHh-eJ>5}2@fXI1Gv)gZQqE1=VntJ
zJM3%4-B%R?oIzZV;R|mdfTtAMabA7Xbfbe;%er;D?X>l1H`{JEK5uK6xIL!4>y>ys
zogX=$llwC}o#(S!aBu0Z3u*7k{#9aUci*rrLLwaNI;uH)6v#Nt%d8eRY7C_SoN$}g
z8m{cd-Pa01`&5#5j}N4%Zy&D6j!)Zvh8!`x>HA#`kx31XvT!J_I)91HC`dMn=<g-w
z;}$f+an;kHVT>|PZI4iA{TOdi24S_0RfC{|OcYfC{9P9Rc~M|)GkinG%s##%XwOhH
zsz#Gj(lYdvE41r>tS;24bl6YSWc%?K=KmBih}>&E1+?h8>J$(*zZmu(tA8W>)n37B
zKh#^k60z(ex!MW>CHHFWacIvWm_~W}Z;%nS0Or&oI(KBjqegI-9-dk5;rOnPF_)~o
zWjt#a@AFv#`}*?pfl=WtdCJ(d&$YTc6<)|5b~tyB!t>5nXlSw9k<l>Q14EM1^D#L6
z>55uVOt~z`c{h^#OGE{1`S3*`iSbF4y!oeSFuK@v>jkKMByd1CJGUj7-kzkCs2;b>
z#WyVC4RsfZ>0;xs5dW*6D$VmW_v)Yj>;;tn%1v8-09IHKbgS&%uXrLp9!lw+i3FFx
zSaM{xW@`%1M<ad1H;Pq~=SxA#h^OXMZ^I;^%#v4@vw{AYo0mTK^W!E_$3rH;Y`|uW
zqCwsh$|F$#J0NGl=nOAOV}cKKhAu)nMk5d0qwBvhsA-G;=I98e4X*@B>41epL<3!<
zY!(jZAh^(;JmFB4@M?L&vUOQhP?5F4rK;vEUIz^&4X_Ii@x=;1#*!GBc*)lg%EyPX
z;cScJeVw>0k?tHmF$d3+O<X;UzRRE$#)Q_BC#c*cHD`_o(=YP(Oc3`2PnJC(KD>Eo
z?#|@y*hWN}9>M2eJAJGcc9#p$Gbhn?PKS%nR~bS6)rOl2)4+v*&pMZFjy(D-I3&)D
z_SOVHM1!$nm|@p}6<CyC?=MHeb##Ikw!kaa(Ofw&Z&>P#;65Uked;f(vkjQ#cDsE(
z;QvcAlBeJ1@_6|9)q;Itlg-&1vNbee7JiydkO?X?Ke<fEDu)TtJBJa~tGgV*+XKnC
z8RgJpbKoDnG_%LcOO<}18E8Ec8mg8J>z~)gkc_86$`)x^AD>pXkEFn!DdQ$0lHzKW
zs>S3%{zfDqfV!&u#^Y6J%mK4awJB*G!L_k4Md*`diNyP))I*7)S)nCTsIu$9y*BKz
zaZ6NzdhHSFAanUnP0-WPDmC!i!B-Our{H^u#+thZ+rGJwGbc!amA06|lI=IKV3$}^
zNMJ=JwXK;wiXjH%pm1SaPIw=vv4ib(BT+<h7>Gq{n*|Aas5HZUUUqIHQ%_tfl$tyQ
z2w-c2wkona@le?B-4v2;)G(b%kon@pR&BcpX(=xSZ`U#MzVwI(H)G7yQ;xv8Em>fl
z*XIFxURX&iH@UC@^9gg$nhm4e&0i0b=JG~^7cF^+CKq$Lr=dU;t}Nu@NxGO&Vba<l
z(sFKHkdA5ud@Jd_$hCQJ4Ed9LK6Q2S*gky(Mrrx55pa<l|8Yo6<wABsQBgdp0WRfQ
z0YqW#N)7TU$CLZt={A?;xMWT$+Zk}p-~Ylg;LxMcNv!zyEiL)rfUk(MFOm?_=59EN
z|LNmNv?aI33}uQSl_|#<RKhWTr7RSs!gR1fr7P4p4+(_T#TYISCBJ<bl5W6Mw!v?K
zp{tzPw8Lsxfh&MUxXlxWxx8g^lZjzzl&FAIH*?X|f&5cxvpj4iB=Wr+vw;hTex$DG
zQVesU&4ugEwJO{4dGhn>PxDSfb1<O#YGafUN#cC(n29^Ym5B8is7vQ`c5zX{;MKua
zqeF_o@<$u#Cv09#ity1r{Us<x!Rp;gBBQ1!wM3i@vaa9pR8LKYZ$IsA$6_F4Y;rP6
zebhrq8(F@qBeQiqp}YKWrh%01h@&+ehOkRVnfGlA;{*}ZK>y@`Cbwv2SC?ovKD#Xh
z5c2bvJ6YED)$NZk{<;90<163e3wIYOhH>VM7lk2x8qgOyl&8)PRQtsbhWbcOpN7Lg
z{1h&9XhQsoWB9zH`3wa72+LLYP`-W0CZ~=!RxjqFbvZqM?%dEbIqrQ^3vEP3{^lf9
zob}sktvIK~s}L1k?4o4Mr$y1E+ji$8P(x??P^G$LB37AFnQGQ+r!U&^YeH@X$h_a3
zlW+M^2NTqtuy(iWirzb9f=|834#$+q(~FBk|D=c~(gDl%)2kd)YalrCx9NFC#(8}v
z+rwRA*VU(r&B3p~kg;`GRg&qWW@-!Rrm@*gF^Sr?vGu?QiDu9p6hjsO_V54)CbTb5
z@c{3J(*_`8yve-u{PSg|2KJl*#(D0q9l#ISJgijALCvh)*Ms=2Pdnj&f%Y39Ljl&6
zR&{^;{<<talJdscYMVZ<Xo=$w+JRHZWc&zF(vX5}6q^Uvnbs>cQ$)?<5nibE-?u*Y
z<E%q&ZhfezL|%8OMGD!X#tw@QZ~-xsW#1N4ZA~~u;Z7pG`K$<u?jFSVx14e)u=<mV
zvp}{g3e5-KMt~&*^1!}{MLkzeV8O9D&@~gVSmx#PzP7n)G(?qr5A$Zs-)|)mNF-UR
zoB6HP$HUA8%9t*?C@7~CW0FCbVKKo>u&?za;HN_HzQ>50G7s1fVf;*QJUTMS(_wqU
z#)R&VaEZ6wKR(0BOdk}TsYbvqCIZ))A7GFdfDY&X-T3U+H|&zdQwFV?uYDD{k$3lx
zKA(@@Gk9Vt!?7D+-62SV_L!$WX^2~|cHhn^lV^`EO+NZ0GMr{$N#OI=Gd1Ci$BuoV
zt==y_{<v*@{5dk{Ok#bF%3uC@1E{whCO8+b#vP~pFxOtPKL<>c4L9OQ@$hj8g}Qc&
zzt#QZ)#VbH%<OLAuX@+N^h5@3<z%$fXTpH(H@_h+p<aQ<wSIlqk2nCF4oLj+VF7mM
ztWBXem~O8l<EqvdKBP=99`~%<?B=(a{jucwmw$=lX#yTo2Hl#pz^d4*WpSEOyKr_q
zPf~{?0|&!y1rwAxya;{KX<6U4M?B~<3~9I_I#eiHkx`C?BDy-T<9ZCgYu)N*wB-BW
zot1-%N*4*#5bjw+pjX=FnV9N2gDEgtU;1=r&u#ejIYPW)Cnjcka8%HU`h^)_P8NJU
zn>CZij(fWQEkM;I3T6S6MVi8|{b|yG=Npg)>3tVZj3z%5t0|#d`~V{ZxuX(7rTwfT
z0G&qXx<<F$W<U^FWfD&^#V8V~<m;0UEGSq$dK^la<}84>*=cC@!uN|P#RzJ`2r^!4
zvH!A4V=QQ+2V0{}4*5gEgJ_yC5SZ{N_M)9`1MXcNxk`=j_+nk#&M(!;m!hb~2sAe<
z5yhX7-=%>PO;PZ9FRn2ts*K*6kvh`}4Pc-sWcnsE`Aap*ZQz3C)43$Jz8q;q%Re1X
zE1L&*J<qNgF;?=>%<tz7ZIy5$m7*#IO;^-3uXwedp1(D;3l<h!(8CPrG@`P=51sUo
z&$@I?@$4c*u>OCSks~0CwO_LzA}uVau4eXEcz=Jt-hd~uQ{!CG`0+6$^kKh-Y=SUr
z{{661ZYuQgM14^&lC_FR9)!u+Am$KKl$VogX4&&ZK-=gii$T!Ry7-VKI(z0n-a1J`
zA7V*<mSO3Mch`43UqJ7D6Xn_wz#(I~kd9D_nA_c*{g`!c0CMs4kI*Q$egzjZfQp-f
zj1)DFU6Fsikv>)R0?wq?AXHTb%>ayh39<*yCtW}3@&d=}UGjF+vuq4Dwa(P4q#rwL
zeL5T1P@JZmFUbmYoxoZ*k>Uk;gvN%^HS)u{-B!Y%9>3`MJzQdr{dERt_0y>GEz@X>
zEwSEx?IGRJhkr&<J4%MnA7*((Dd!V-+%?t{-?7i537>?-PjZ18?feCIx46x>Q1t%?
zGc(72XA|U%&ut=L=e!i_2yB=iL-rBZ!1(FCBZKwgGEHWQg>-d%Y-<z<M(Vx`OHCS=
zkT<RHe4wtR0uv?ZLb`s+rDP#*49O<XiWPyvs2(U=>VzO-J@HpY)nc!TImvQ>7L@@b
zNOBar+m@}?BPfn061;*{x|)v3^L#qbLVH^lp;0HOLz_C*Tb7(asld|*{pDJmk5fAj
zLzahAJL|A=6R-HkX@xpXn15FmAqB89S;)PfNZnfF@MB)T+N_hgY#*YED(&y|(WS%=
z+MdQ?N?)X5L5zZ>IUE`-O`Rmf^%^XSa%<D;&?@z{T=ZGa{g%<$#C9Bt;J1|U>Iszd
z;Ry0<%S`**38B4_A+FnPctHrBaP!AjIijGwe2mJ3#M(&%eHcw>Ze`g1VCQQ;<c@5>
z{{g%z;j>dbgeGu+c4e?an;k^V&@ful0Ww^#+4k22Bhq`ck0<|^;pMnKN!7o+bFax~
z05iEBFbF_kQ9@#^lfg<}Ndw~cg+#+oJYcJ%(ar}tNb}`>TGtmim)vzX(6TN74UZuh
zdxVy6FD}^AH1s5Xe1R@aIy#zxO}v_{X?qK{J;W)ZKpnW1zWGGu#qZkoGC@dqA*CZ#
zK4|N6x_b9>GfE|F(l}}f=YZ#N)5nKEF~M2{$~DXm@WM^l*3H`HFZ|o)GIq)!fA9W$
zuxYxqTP>Nl(%Sqaba=}B*5)9w%4(V4`c7TR+uQqR(rkSYv2xG-Ntj0hPS)-maWNiy
z63sv+sCKeQAcOqTYBz%xq_ny?WHcW1TE}t++5|b6Fvbvf2|-&?NvTGo+B5-gN{Q3I
z+_oi6(fhp5?NW!KdDEL89-B#A1{3oFYc;7GX7dx+gE<M<C2*1xv*WKmE<Y|VK~ngf
z>2@FSk$bqio)I6G&542aVGYs@Fh*Lu(M7Vmb2n{<<uQey0YZQLlbMhy3aJAvA{zE^
z*c{ALD&tyiBB*d%r^M1~Vmph9G(af|Y@RYbRDCC;#br$n9i2cwAF4moM;%W|2t~Pm
z3b}F=g-#%dm8o|*?+eW~J5sC2^0l?~{AbhD-TihPt;e{ZcPv~;la9r~J{F@h+U@h6
zz0_KdottAVo!o6sXY+2jMI~v~5u-)%XFPj~naB;q?ly<TwC$|iFn_bBEYUiJ<&>q4
zgE5jp$9_cU4Lo@#!E*EUU-XWH%~lq0n(hd9*UlDQXIaj=J+>YyyR1z3YTNRJNpbxW
z-@-Vb+e%+FSbfM`loQumT%`{-ZNG$^IIVD9<Nx1+BX+cqqmP4v8%N;xAR~c}P6lk1
zx}v6-jHjR+nm-5BGPU`(T<NUWz?LkwX$*$b9_N)sLmvUTWTr?X32BNiG&mak)z24e
z8VV*bu%O&$|HKqByw`{vr%Si$QLU*hUc~`uL_lgjKVBUNFf~E0$_?McO#X+xvkHnU
z?7B2A!7XTTcXxMpC%AiXcPF?8cXto&5Zv8@ySvZHRQ)&KH*-H1Q`J;;cXiQ+)6M(t
zv-euhiseg@E)s57u>mIsv-tWzhW84%5w*H~`J|};e%@Q)SyMoSs!b%g3N>A*n1P<A
z5L=aEJX>Vn0{9Gffp$d*oOJwtI49MH4}`jb-S?g5nr?aJvgdSCyFT2>i9)GF5Dplx
zld=-cIwP@4isGbvy_yDqkRObE*8TZ${T<(VTFBYWt-<5g>|bSN)r1or+K}rBjK_DF
z6tcLJfmL+Sy2W(tYx)E^V7U;c8U(!IX#m~?P*_Bq)-NU$Q$>ph0m@R4Nk9I_kpac2
zs`mB_m(#`T_e;NSgt&D;QaOZWHkmn+$>p@_p9|tD?tcV1wKEpm;~YH|qv5Uzps&ct
zU49QKUroAo|NQ6CApE<zwLW}JkqshT>vM?M$TrW-b98>`bTTSBqjBj*O@|~MVoOza
zbbM;NGW_KAHiT;*MYs)Yo;$Y~zl>3Zs!ozwd|T?x(2@~dNxXDKkr$hP3G_kj)5XCs
zXBKJdFM>s>rPq>N?oQ@x!Tx^TuU!da^0+Zczb6+{SCpczW6!-gYwS2qz`YmWSb#+T
zF<|HvN7_Yi5LugGl|Jwxcu3bigZYI{f7vb<-UI=nbu>@U@OpvyH{MH#pfZe4B2#D(
zA~yi<1eOCp@x7k*%>6%FfT(}G2H>s!wZz6DfopQZ^&<!8!f(!qWlA-V`Kloo3)UM<
zg*aMFFl3Br<P^q|7&m?XwecXRe^3BbT_LbZPEJlPP5Mqhe$pv!A<a~gbio*0_d!;=
z7mu{5#ojCr$*sFju(3L@7ne=C?_3Y}S^?NQq5SWjE;ohX%fN?W-KODiuciSCpkS~u
zdQO+BM5KE6MxdGP&bqL46{XPhADK*V_C_?Ve>wjZCI7}5e}B4kJFuJ4`T`g{pRk6$
z0++JGSp7hg*<&|BY|kU6Cl{_{+C_@ny=*@D;Ze;Char=s#sJO0hnWSLzJ7f}K)LAJ
zbJ2>e8j$Ep-oV_GVA5*##1Y9G>dITt>vdpR(*ZU{+k4{Imlt9tO8;>~Jau2BssYK`
z1BBh2kB<+@lsM&tL#_!kmYC=r!xyLhGXj^N@TU4t<7qcPC|D(fn6SKjN<pO*g~oR1
z=e4EXJx>$YJ;&%|@!SY=I-gVs(~pQLq`0T$B2Xh*;7g*XA@W?S*Cfl=8=9xC;>Xh+
zkb^HVQ$&m5za`u}Pw52;a)-gAFS-x%RXvN8!dX;qA5*kA9ub8Nx(W(HTr`N|Tcc_u
zP$we(8n3x<7)SxSZA3oZC90^0>~0$_m>|~eO8{cSn~eVRJgD3Fa54)h!lgEM)Rg6{
z<9mVE@qp9m6H)0F^hD8P<Tz-Q-M5X=zg@Ej6H8^<wkKVGml1%*W46IcIv*{Hf|aZ^
zJqky;ZkWCRcXoDeoDdo_eiS5x-}=UG^NWNZ@VpJ$iWp<Fdk2H3jA|PZjW`SCqt>q%
z!O&QStb=R*6{KJ4s?ePZS4g*}6{ZUQkpxBK{cVi>N_u*nh1wF*xC+|g@q;qfE*HY6
zYv59kK;qHQayfyyB1x-R=sm~{jkqLVdGj?@ru9bTbBJUtmC0*}2txjcf&#<mHGzf`
zz&<jltZ33Gn@6S$R{WhlU)5YzQxhZT;lB(zxs9D5JUgDQS}|#9_<WNL*oQ=T0e|k>
z+o4;%)1f@TMWXok*rE!cAnk(_Jb_#aT)zAIbTzj^H|WU-(rliB6MWhBr;8qP@2uq^
znO)=KR<=1c*9!>g6%t)whfN^wPsmf-_T7ySr1lDviBj2V4A+wSc@5GWVFZ(<+fVuC
zhRqId0%=Hu!Re^k^fi$h=)|=0_d%eFRzTu6D^yJV;BYp;1X6YuxD>VeYpbax_azII
zxtWqJqP24;ycsWaite;(*&qZt;j0QNhH^STeb5`LwSRZ5GR)Z}CY4PB%0S$(%3$QG
z{e4rzoTN+oF`wzvqG$v@yAc(dk8dofi_s#Z!*_kTO9ma8SmWY|L)A+1$S+`<>h3(y
z8UActF)=c=12252Ns8!gu)=u=CI}ktXt`Ppl=W8@Gz!efF!6-mC7Em;)FYaUH5&t;
zFMaz}o&tGxY6*l=bwfr?GDFg{i=rcx%4*kW)EZl5GdU{Qivh@ip8BDvcC*V31v&3S
z<b~48QV>8DAG2Uat0j_1ZGPS8<V67D_?cv>Vg`T=qZfai(h2O`%_)>5kj_{!T%&#S
zAE2lQY-4?nhf{tLRNsZ4<S4pAtWIdbzzv8)2<|imm})PubR}~jRyX!!emroQ-=596
za98v{za0vU5jAX>{MJ4ulp<a2`khbij};+1Jv2QtBc%RI(}{Yl!NLcFT>JKLD%O7%
zSg{$0RzM!^@5Mlz0Cp0}4{qQa@z3%WuROH3BiIA;HyBMmTyVEoF5~yd%c+v|b+udP
zPhfF{sbT{v9}~dM1yZS5*H;GVV;dKT#|an%fyozT5$9=T1Bm&hVO^zGLnx_~k!~C#
zRSvI*V<WRHO&32L81_k(=I(ozz7~vVqth1<?d(I<grtt+m1@K4NW%EYv^n!ys|9N1
z<R9k?OSJ%JEBziSLUnH<D*GGHj0XTX9<4h;v)46?G7f}+8}7d5RIk!JR=1<I)^KZ=
zqNcl-$`phvdHHexC1%Kr3?)To3!VgRY!8dXPq>CU(l^RkE~^^(hb(y#!k7+2iO8Du
z_j5`H@`ps3Y#Ngb6rl4qUPH9mK3&1UKO0=z(TZq9HH5&}x4NW#-RQ@2{bK`UVr24V
zvzxTWu(^w_m3lnpnR;cANlRVbc(uv|j^%`;8mc`hE0O=JSA{DUzj4;CYfbb`=ZOFc
z@xN*2*$!JO82D>fSe-m4^QSJZcXlGAzMZ|1UxL&^c^Rg;AMa@i)e`WQ^mV<j2YCUt
z4zTk4m=VLQ14x=BF>C!|%le$!>N?kNztusfDd8d*@i=|)aMeE$^H}ipTOwsuYsIJY
zco~i9O1f{+lN=t34>MhGQgvOi)nH@oHNlZ1Xh9-coOm5Zz(i*VQ35aa$DVtUe1QoO
z%Oqz|8jCq`JZ-u_ECqY@;GPS6jgA~ZRb#^gO+r>pGFlBlLlz4vN7pQn3cAm3k?}4Y
zu-0;!zWAm05`$cA7oFH-98ZeDk!NobTDU0Huwt^G?ox*Y|DKnWuyD#g(3Y`by~xEp
zK5q$+%Y{QkcF$WPn1GEeJiGR%b~HMq`TR_*P_40vx1|6&UVWtLVC9@=Mp#Vm?=7FN
zjY?TOwun6mNlV~)=euTdo2R2xi1_v&%m_t0U3J1L!WMedRc>f-w2?+1Gch=Hgvj__
ztf)?&1ognRz&TB*<P>m~%8<=$sFJQ*zVB~oQ#(C=0n+7I;A`1yQu$U#eX03^gdtWk
zh+jnEz)=z^E&rs4{zEor`aG<hEpHx<DEv5Q+cN$W2XHeuVzirnDt_l3uUO~){<4R=
z_4YOA>8a%RCr9`GNkt9NQvmuHz(4%c-IK9`!_ct|Noy_Oa;!5R28uOke8yyOM6B4W
zo;2jZ*UL{jfsT7ukdNn_F4V^jtX?JiUsK5%zCqvCUV&L_hr>ujUAA(z&g@;$>w7AO
z@G}<`lhk^?cu70h!T#l-H3y?!2NFU=OPqT>*cbSSxbOjR;Z0>_Lg^R^x?~9@^g^i<
z)EWF>_`eH+OQH4m2G0$SRrvm-CPbheB4P=}w*DEa)E`xa?dCkfeBVzGF7kZU326px
zN?RruK0e~4n<h)hd^fqLKB}1Pa2MC;I^~PMtX_T!{;+1$8#cT)khbnHeDm{%Hl~#v
z`#77D;D1xv{4t=^@d<v&MLD#EM({GNe{-zxNoPEK2G`SUJC$%KM~g#bBNczTQ#Z_R
zI73M_P5wA4p}@`<YDKmmFX4;CsXY5n@c~ps^`9TRfZG@r=mgM<aECSa94mg7Hg?tW
z>4FvOpTiV_ZGTwczI>sQ3a(hNd;8AnueELHSS(K;n@>wwCK*Sns^f^KaHZy7+srn~
z3d;JF(_*Fa7nJtVjW{q;YWQ<0Iv=SN{98SmN#s|%#rb%6A(D|Mayiwt$e?`cJChK0
zB_ds@az#6y)|5i&K`%8<k#FdWC`h*Wk20`EJo*@20`M-4Jq1!j?=q;2p+OX9yD|Ce
z2d^);woE@(F1l1UV`}Oa7Gr}iJLJQW!AY2&?aYK3T!#@SgC0cT1GR(dtQ)l|q$;Pe
z(7!@1zRzjrHjznzWysEspZuBsJ(3Qc7)d*fUX3!5fgNE(l_vaquWJBjZ1S83xhsfe
zh%kik{!CECBK7K#)~wQnj;E{CR+9GsvP}9<Mq16i;qmFQspgVmF#6s7-4HjSRsrk~
zK;BcH)pDZc(nUX8@9GDIkdnTf#BZtjI>~=a+x0Za09Na{Pu<M-iU)wFkXpB@6Gkzp
z`&PiDsp5r!L1Nc15M&{l7z~V&>n=-YvKx-q+`1JsYll3$O%*IsxT-0pjFpsVv>g`+
z*6+aWI`0u>S;VYM+}qz{)<7s7ae2Bh;fTn@w!dFzOpt&3mW2Xerb->>PBR!uJoX*s
z4>n7Q!%$jD9iRmQWbr|R)8i+q<*<u5z!jKi9*f5AKwG-(KLmCwz(rdA?O>cSOD}~A
zY)eo*z_$fA?Hq84I#I`YW|=i$2h;)`jvoL>>*j9|QaxbL{4;<wS_ovNNZe~a`?doL
z*Wc&OYyWb~v{IC$X5C{a)0q;DW3m-GyO)BBU{{~Gp$&=64HcXPg+jinG5dn?02wQ7
z?d_nPC#3~D&r)-&U>@4m>~`znNzB?z+ScDH7AZEs$Mj&ft8)Z6w5<hxWW)D_x>&Eb
zKKOqdVs=`T1uoq(IBfo*gzg?5m<t<%Xq{vUp%-V1&2XbSDJm_r`o#?9Q&0sURp-9I
zqGpLmM_OusOd;7B_7_i(`da*hlt)ofei;KW2FBFM1Bvi>X+l&&1v7;u=Fosh*?6tl
ze1<!A<QbR?07Nu`x4+QXJg&40R*^R$l%t~?m^depidJqeb*m`|zajfClhkUkmJ_6v
zlFrnkU9z93)zH;%htOgG3)CeTkz#U^nV$fUr`z=0R?tvgd4^bMbEtrg7mh8YxL}ob
z-pZf4p{B!<oNIg=!?HhLnhFykFhZ1+3%|QQr{+s@QaS@V`{|S?&wG(TwYn5^n3A_3
zPCLrR?O%|$Qia<{xEc*+_tFWhpfn?qP*4D9UA>YG7-LzoXZ*urZ#S>m8J}yHe1}VE
z2CdBC^<YM$kdN96s&8^AUBQxs{YT{hy2ZNy4tlK!Vf;4ppQaob+?OVtTwcaf8xvNa
z4o(Hv)G408?fG{J<qB#Q)*kicMhg|eMpqgDg9DqDj@|`4G%%qG*TRG{4N&sy4#Z<T
zX+J-KpI_;FafDLZakPfV^?TZyaBCOqfAktbBoHqFtMGN@_d1*^1bgmdlSb&mR9gO+
zFuxYsNh$^2Z8?i7tvYWEaoiYz3(+K5_uGukzV44xfH!M@p`<!1jaMRzB@o|az38P=
z|7Em!{$1SPhskTRc?#+^rq{hwyjGODHa`+1hU*1u^T(aAKujv33Zq@p=a~LRv?Bgb
zSWY3TshXyWcnVVV)f+&GIlZ<OBu!@TH%!%Wp41|KlUsMS>6CI)W$*Jd*5?h3gW!4$
z9AC*|tl4%A<jDTcVQ;;EdNtoxNJ9e?NH6>6`0{<@0!h6h)cqr3iB`h;-Q~XNAMWjK
zuKNvq+Q`}3Rw=vqV>{M6+-b^30BF1kv~0NIuGXwAF#57pi&6Eigz{f>UA9n4WaITD
zd>E5(bm9M4d64<D$l6=e3GWPBgcmSQlPNkPMHIC$2BNXD%129srb)~t*-}}GTGsR1
z)}WKPhPB(dWW13AKIE!oyg4B*-hEH6fKF5MYgYw2`f6uNx^uvf&5V&#+6DBdx-)L(
zjg4<)(8u>)e&L`HIH!Mpe8m;QE<<P5fI<I>q}w-$sW;y7d`tOdLf`v{0hIGNnB)4x
zY3owmknVKJF1KJcqBeo(o34$|3AXReZqC2Km#3Fk>nWz&?OmeIKa_rihWg4AUc2~x
zklzQUKg~=XfZ4=}hH4sRaQ|pMf4iyn`e$f8K0Vz%F3C9ngoXd|4r#UOgW>3Z_n_2P
zio!_Uq$95h&N)uU^_ka=V!y>r5zh~=Ml&xyKeRIX$CH*vx{=<|xGnkyG?p?IrEB9f
zVHVb+fvhnPCKiISgHa>e)H9c-;c4$5_;V><A8bhNsMgUb%Z!0BW-i8|(dvLhiTs85
zsT>GqzebV!1tufl&$rT$Pe;xO!?YRsaf7`JXi%|$8V-?4>|B?M>J-FNI}}3KGP+zB
zsZg~Ov_8<Vsh{~xn>W54$o|c#;?i>gb?whKsU%lRG>W9q?TG?>ABQ@@y~}VSDdhNu
znuj0QpfDUy>QaHtkgH!!JhHmFvOwSox{!on8W->G_+zhZUN^hWv9_MVgj$;x$vl(5
ztdDChA(V2XO;;N&X}|wQCW2_vjs`d^)X85sw!B#Y=*B&oV07G9QI!4diON_~vg$Ey
zna-IYQ`WD?TZe?SJYLOts3mbI%z4w3F^pA`FE4I{h2pq!d0D2hs!+CH%)YTk1H)V6
zs%;8^&gh0o?y<V$WI~4eU-NP_G4Jo~VFYFqtI5^XVgCKGoC7su0V0c^hh~L~E{uJc
z#+Z{81mO>{{u1r$)(u)EdeF{@KvH)kuWYF5nUfxEqbrEIwp6J*!y{?3CP|03%%zbZ
zD7_GnJ1dLtv4`@{(zO6DLE!T*=OXq&*G@R)e}e}eyB+~_@x;G6(V~JCSfv5)<mscB
z#I*Ux+jE`zJV2wuUP>l~1sIj4vxQRsgs%X7Kon{h@HC`xI^@?FDvO3-_k^R6Z~}Yo
z^SQkrZZC=dzujy7T81)`SQ|(YE|uvyv6wOW$p>OZBuPpd0iN?EOQoWame-ZPk6Uc4
z5*u-kCK4=o-~WBs=>Sl`E_HvfNqlob@3Rt*#p~5F&lJx-l+9qKT3o7IoK}+Cc&29d
zn|Slh2z<=;AqN4PQMa$AV}G1V!g&G1pV90yu$>mPw4`qLhbq@Bm64BXC2+DFEe+iO
z(<CK`>0U);E2=-vF>GKF!kQyD5-AO43pk=`qC6RonGS6-Ly}UDZB3L#zqoyHLfCk_
zouDU(DazEBSC)WaAdwO00F-J?4?_e>4#|&aT{mtTtpZE;u?#B-I&vF5#%EIazR|D0
zF{;oZ#eZ!C>b6(!CX-Dz<gaLlHO2=Rgj7V&0!yD_43?7p0V&2(s5PK@30dSKh5Qp-
zY)72jkb7d!sz_NLk#WTTxz0|iA_ymYK@lxK8VCs>A^G(`Nu%$nijwTWzSHE4X)qM_
z2B8+^85<1!43_nm;{HR12hg9K={ZC7H+?WOa=b(PIF(j$V>16W+K615vECuW3lmps
z<>hre@qyMW_}{!R{K)fXAVmaVn~5A-Ta~LV<$QCPXX0(>3{Z|-af%VwvDizO7IU~l
zkzn$gRC?RN?A`+sTWlKQ@PL3H0PO{a0A8-0XDFx|7Qam%DYi`>#&B8f5S$KTqh-HA
zTcz=PEmKvp(oR{#pim4;F1;_~>y!wq>-5|l+QMKSqKV?SFjuU<(^_-Dg@SB4<5wY}
z@}95QDo9vU$Ctd2L=cJOQG@7mR?>A_M<|(q0X<7?e;V<IR|Gvlsh|6K>@f#zsa-(>
z0Vo)r8j7y%WhZ|t^>ZCo&Vl@5x)+@GI9rR=;mpalogyp(c33jfhm}$=*GGv{z4eW~
zajk-vsq^am$bZ*ZG|UzaJfi{RfA}9qk#mxMoPD=N!*lERr>$k*Zk$sn+FTAmuKtKb
zKEGG%i(7r!n-74#7>=Nz6CVwHABx?1<<RQ~@>n7&$o4DOKTc{kV~a-CK5`QM`}Tc!
z{qvLl=51}l^$gR$S?3B;0Yu+cGVxFUXHj=;u`z0T%1GK2K#0m@`vTf=naP4&G*Daz
zRpcFr#nJpes`1GIz#1V{iqDYH9r(PeFJ$IuYDnXZ|HS9_69m?P%T2=dAkRP_Daa9O
z<?bZTmOJwjeHJ%eO4yv<-n_=;G$=QtY!yj&z>MC<_D(X=HngvVlF+3H@_laNXzw)E
z^6tSTM@_14^FRe8<%bxACZZ+S`TB_Km|TV4_+v--*^c#niIaNu?Lf@;R(QMijO!E;
zV#Wd&<|c$7yf&!Gct;LMB9)W0B};hWzg~&0>+(WN)%WGfX@9x}!Ta%7l+hr{3o3<H
zPq?7_ILqblR`>h2>W&-fY_AKhB;`!^H>vqooUX4_^nT=|-Cc)z|DtJ9ov!z~{`nHU
zpMHcQej5dr7OX|KOaU986hMB%WOk={fX9?tT}>Z~L~!AV9<mlzd;avjYZc{%y#rNd
zdNF*+V^xL^dX`4Fs2yK>FITjpFu>AhCZ#z-<4;$La~EdpqJvc+Ssu+Jk^lUGD$Ikg
z{ITkSNLe|0(WyKa-1#`_IZt>#1U`c5K+1wVH#q3%msoeHC4-8j@p(q+_+8Jz>8_C5
z)ThT45GGrj7nyexn~nMEQB#XcW>+l*O~tk!>0dF4<1WD!mzq04yVuqViceK7&K)^j
zp`DvDo!0|W7Yc$phiM{&6s;Lu5iARQvak5MXl~OBk_vkaLWDASU57X}yx}U;Yan*d
zH!5BE%iGw>rnx1aWJ8;{T@QLU5omc_d1J3aX3ZxvCj+46quMdfJT~7B5;Yslni|%Q
zH>7|<$$+?_t)c?03eNRn9TBS#{qF8e@HCUO3?QTxKuD}W!o>)k;uQt^sX6FpOmGWp
zHZIPePUcI%n76yGT3#NsomLxfw7KAUoq1O-P?C+@7VC%gXhR3CdtG({A<5X*8xHbm
z%*lWpXFyAKWHCyF>bN^dg9S^&f8L4^)H0sA_r;T4nYz(o5p%%`niy0^a1b1u8Fo;N
zkBJ!Tpi2;<NtJX>QzVdhiHlh-lOm|@98D1S<13!W^7g-<bDD+X`2bI|aPzD2bSAn$
z_Nx15p%Ms<aQw<iF1rz8KQ#0zsI9eQGSnDO8+HjicCWuEhAuIvcHg9)@*JA<4<@(*
zJ)B}JKE?7je64~OCYc36#8N!q#?iAdy@qf%wTFjhAV=ryqUa@)Oh$~{Iu@T^h1#%-
zK#JwE`Wnkrc9_7%(~oJ(BY8=bm_H_;R^2WRx$C_~FU<iA6wncLZSV^<J|K`QeN`?~
zeQPL3o;BjFZJ{_Jfgj37Ff7ur^bl8H3-iD8fI~{a9IEx4u!>^Tv`k|&<^D_VH-u&P
z`-5A#taixp0oyzu`R?(KILaF7)yq?ANXC`Fdt*i3LPU&gS(pbmQ#kQS8}S>4xj-o#
zpTSl0siJ%bgarn9nGh3}ZE9hIPSZm&CIv%7QC<<@mo*#~Gg%YGcK*_PYF;TJV<}?7
zgIGDrLW?RdKt*DgkPf4m5SUp!Qo2aZDwg<fXvlN%o1|O;{0gbV>~DSJm$*NuU$!%;
zeKwH7_fScQL>T?9i4P(mP={NOu5wzg!MrK*=w!1lUV^W1h?fY_LJqCsFs_DnW36>k
zWmUav*l4xu-CKrHs`~`(0&jCMk5U6Hw0f|F(*5a#87!nKiITrsIt5e3VVT$P61H-G
z?_IIiy_?0w!SZ~i{~42Bmm`pS_#W^(RfHnuhkgCwoBhUErGtpJz+_gs@xk)sgA0gL
z@mc#=eQnHQ7ba<&zhL8hdg0Dc$OJEi9L<Mldy-q@o~NnitpWUyHI|3>q1_soNT}d2
znI<yv;7*c4Ri%6k#-#6ghX4;I;e(VFN7{#@-+h{b501YmHM#t9j??CAmu{EShrUng
z`QyGz*Zp40*@H#GhNf!rl9n&QA_X#@<8*JkuKqI*kvKf=Bu1*AN9yI-#jnq#X-U;a
z0%w`kHqSg*DC5ZK=2s>fQN_~`V$vNj8a>ZASRLY@m&FKZzO6TZV<>%U-t-Xv+(m%a
zfN!Nrom5@6acDg!rA!5^PTXL=;7LOo>iy+7G4-|o8)uC~PBK>VR)G0^nf+o-WfEmJ
z1*x}j{^<9$eC?Bs%6I?)bl1)V(ohAmdZP|P^G3n-GrFdww^%8d72|!al=nyJ)Qn}j
z_7-MifP_8&v)uK2V90R2!FC-(s{q_r{IRKz*AhSOay`V)*CmMQ{K6moSq=5k1%lIs
z*a1}mO^vIAj!<%FLXqgOT#!&A=8D(oii!sDz9%NcJBE7X=VuEsV3MNAu3@3-lDS5=
zn%gsZ6>_RpKbk{|+GfE0!mn-3>rM~b-@Z*s96+zY{*iYjK2AhF;xHsFuwWY#Qo8HI
zsMOb2g8KV$jVlm)d&}#_b-HBKmHE{t2Fi5QYwy*rV&z?gA;=<W=}770Q}=&&on$m=
zyqOWBX0XcY(6J1a@~@(6BiI-Ky4Dd$d0`@zN(mu$AV#%%D47s->Odpuug};{x7g;&
zY9wBP?%~bXOs|UEFQO1q?idV85TZS$SOk;@hDqVbBhu<A-x`V=GO^pySVn$1^JIxm
zoHUNea*vG9y5<w7TVUynss`pt?*0^oq^hL;g5DUfp|;U>nECECLD0RmdAP3ALRlEO
z<M`}rwa|80H?`UQzSe#Cm?rJ$QZ)mmjj7m1K7c3Sz0`-L3Q6tw3C<%~6k&#xhUa+P
zMIaDd70P+&ewC+NsnZ>gCeK@h#DJbZv>XXduqq}CWlf93*CDtuKF2-0XzK=^0=ytV
z!S5h_6PKcBUWeFRY!Sv5$LMDLahG=fxd*z2R*w-@b&#rrKLfY>My<Mgn<mp?2%Gn@
zb!rgPu%Iag$>&JKW}}Ub!B-84;>j8s#MI9`6-qAb>T$DsM{HSllzBj{C`Vl#L$|Nt
zYAlD@ba->1NX(xYEkS1@N9F6-QL8bzwS6?)$%}I1*}Um~)9GN?G;J%!x7y){f9A1v
zO;){x-S?@+75*xHE>K}yxLF}srP4MeY3lU)(VFAua``1&#O~|yX?h;a(Bq9o)FjcM
z^QfLF(?M-k7Dqtsi)J-^q;9)9N)k~MG7SthO46?J=aWP8rY{8$mEkvA_uccl=-G|w
zAF|Bh^}J<{6#=M5g&72oXi>79{t?v&usi@?g4ZtiC7+raM#c4d>Ped!^T^mqTlHh>
zZ`QhTOfMxnstV8j$x06S(=}h(tau52@+4{1VwhtIF7tf)H2ZFVehEMxw+^)^_}aT&
zIa23cCfs?=0JcwIF4x8KuEjy9EID0dq@-|xW4*!Olp4RFjIL`_-}@W=u}Bm0$_jQy
z%WrKn9p|?8yFgcH7bv%hiw5l97wzk;m*@`;4jvRc`1DL@n43iT3V!Ay5vwu7zsV!A
zVQj{xsl+6K*#<{~QgN2E?W`ZWB=Y%7!1)Yit%^F_fJ^+37J%M;6GhbVD0VW|fE|AD
zjImu^Tp>AIzTgI*uSPDiM`Zshi3JqL+&7tpv@-sHu+qVd<Pb~67#mdtl63_QGWX2o
zC}pf1^n4^+i1?g8*Mg_YoDZVdLrxi!9J4f1ko9NsjZu$>MA1@}_oiv{tor^Bc`iI(
zb@7n;j~36*N0wO6x4Bk4AXcdR(n$t4(J8&HBs@eg)eF8J<>2!f^OLN*7{+ui`=xof
zuhPJ#NUwdC<rG7ksw%~{no&f(g4QR!OvMvZZZ8<N%Fdxtv>ucm!+jN%%=|Olgh7pf
zw;67~jMAf?QxLy0R)C9RQMJr8e^U5~8D2{_K5LlR2fs>Qdn<P3nV0P=hGUF?b(nw7
z#R@O{i3d6cwKfh87P!kH`2E1{NGMLjsn?ueVo*{GmUvIU;4CiIuS$b2Jw58L?P8FG
zt3{g{gQ<8(Z(-d}=$nWU%b6(#NSC?t8_nvjq$<9bE&4MYt*Fqaxs90!CQ}{aD#2kz
zc2!y@qs=XSc$I<GOxbuOrGzkn|8cuY66b9S=998YcjBO|ER&k|$);5kzIP$|jH<2`
zcREubn8Mw9vUV?Vth^}<m)icRrbBxfkfyoW-16Xt20i-M+k_E|5vS$-+V7_163nsQ
zy^1Yyv7pKDaR)2OaBPw1RS|^XZ90q#z|;R@doWQ62ZwBscuT&`nF}m=g@nLJeDi_N
z0YbG>Ne`PAAD%PY82#Mrx^XpeJgwOGl9j(CZJ!<T<sj>zUggnL?|JDogML}^{AaPD
z?>vj<Je)2*aS{Q2+f~asx(##AMj#5_JnglZXpk3(zMbO1Bmkq@pAMF$h#K`M=l{5l
z&FkY1b&XkU@+n0yHsF1jUm&S?Z<3}gxaObL>>s1|=3)}owd})TDOr4Fv|)wH9MjXH
zN>;SgHkK5{e7(B24w%<EJ}Z*XS13)YP40b_B);2>oQt#WJIVjuHs#JKh9O0*Q(s=}
zzP{YE_37s~b-vXeG|GzSwsZac#X42-`Rx8B0Lk0`5+DGu_J>NX-K^^SejBwCjyls^
zmcqqT)4iUa46BnH?6ctD-_P_?A|vk-v!1A<z=IawzdWHowqG{yqQOlWl0B~^Ywxx@
zljIJQ%bK`fYLYv|bFq48AV-@ybYxeX6xho&e!ST5zRsz+96qu;TjWkPZRTa4_x2k@
zvPcxcj%LQ-=SB^f98x$AU&<DEDfs-8^YOq_W;s1Nt=Uh?Rmle6_c17oEY`{>(w(Z2
zYOdUO;eRFLYrz&UVX}q2zP4UHODNaHO!vKT>yUEiM9(iq-ltK7aaN8S`ZQ-5`AfsE
z^t)JM6(q~vE)FdpH}$nltFEm6zHiuGHG(a#b4!>-jZxHWDx6%+#Jk|0LB`I<PI=Rm
zt(sB$EV$8bUWw&+pROaV{3UJ(OByw~?JZmvQB#>F_sY>fiko$oh2cK-`KIo?(xGjQ
zP2k^j8F_mQpJnb1<FWCc{rJY!fN0cNaHry*Y&MBpeB6-^-qF0g7D+uY9~PxR78|de
zj@}t;tX~yO`xr4``P*9fVV=2}_sp_-eL2i`3}5V8@F}SRPhJ7Hlhj22FwaG>+5wqS
z@9^!4QO<wfi()Re%BuA+&4^m3@!40d@+A&(F^PdYEH9zG-88svWhn*^lrJpoJLFww
z=Qy01>&MbpBvF>V*pC<eUmC`XgWm>R16xE;)1{$jH~aS9AR5K@k=sWNWFaSKjzsq{
zpA|k&8D`@e0$mjf$KDq@<@b(ItphMp4ZS=*-Di|Dw^e`H!Q?e2Do0}UOMErK@sRyv
zrL!D_oJ;-AkZajVvBFLvP}5*=Gc9sJ^qg5i0V&V>Y(n3;m#V~BIm20B>og}dT<5Z8
z2Ev}@jHj06RqizRrnWm0US3{mH3lT~Wzp%_o6lYxgfMW&1{<^Dps0L}*>rha#0$(F
zeJ+vrT9hi>gK-F~x{33Cd~CU%?iVl{j`^nA)Y}SM<GnB@$*FhKz_6Fm;+-ryusndG
zu=S@nF$P={mAnEjYawL(q_5Jl%F8r~saV6N$#$%6m@jeTC(o~p43glUW}ZtzatYlf
z8}4IK`tCPnR79O#_xiYg&Ex<(;HVt{(Yis!m&<RHhgU&zwhbUGTruveo<=x^QQ|Zq
z>Rc;MHv!T#f%a)M5e1D(WGK$Awgm9B4g}j}Mp2@dOwIo?4<K@@(KjWCsVFpbNzSah
z!UH;lh-mDot$$3aETtqlYECxw-Lz?T2~D;_cq6VD(Gu<XGULa-W_tFW2vfj}y;zq3
z^nqB75pivSkzFroEMv}xW#b4UM(R8n4*Pa+6<l%O@oNfv1vZU2&oK7%TRms~&$pt}
zAMSc@^>A3vOy)U{ymrr%-G4jH$8@(u7Jlp$qj5flOi`R999EsJDND~q&*(9-Y{Mo-
z4(OF7Uy4obd*MY(z=w&fHbd6;jCh3AK@mJM{1%x+z%Diw&$FKJdHg`qX6t45uvyUF
z6@g_}!evD^qK}n#GC(a>cIUOVh#)trB*rL+6&19MR!+&{`O2t?br_7VraRNd6x8dV
z#pCAqhbRn$F-<m{i5U+ON&2ZbS7qdUiBB|I8YKxDkGzj2#}B6p8ka9xZd(54ppki+
zo6*{21*jR51@%~6azN)@VpNKYEdvWqRC&mbU1pc(@z0b6=s6b@Q<w3ptd&f4Rj}cR
zXjdd<h)uRfLJciT+h{XbNb#;1m4bRv9k@ml?4%_tnoU$87@|M?6J=EvYucqaiAAES
z{dEh`6|vwWHKX8UQ-he#t>I6=&Qt!J;(VOyzwfgKD`6jOfFim)hLcgZcD60WZPE1b
zt2Js*hs`XJ>Dm-IMWPlpbX*DAs03-R@~6Ow(PU`|@{3v}-oil=)T(nomy-oDkAWkb
z6=cxkY5qxjsa0i#Z!qERU2k*w<qWb|6_K#Tkk_=zVn3NkwzphKzR?%V6yZ-J4jvI)
zM7`2@lb5$g1s9wkGiF$7amNiMCql?*$f2!uaDYTb5zE`t*?4ARgNZf6tPj8YU`pvy
zTWV-2_)s>;pIh7G_twH-#_>F}xL;#juF8$fLp;pdN3!S24#Ez0>fee^M9(}iwE!tm
zGN8UVH}R}Z7;KY;=)quybkXIMnr6NnQm}x-4ZeF+Z#pjvr&l*7pfwbzHiQ#T4+D*=
zBSc%_suM7nq}q<G3xXi<c%m^UPd)&mAaG;Z0aT1+ZLrOwo8O0U$P-b3xj!P0>z=pj
zzdMN|FuA0sw^+ut>IJHOuPG|>ToE~+&l)Uvh#sqRh$7Y#28y=_9G|#=i1y`sC{|f{
zr2bQgey(jp?&Z1tC-^`9IOwkfR<4s$5{F5L?GVga))f{j#!`=SYyJKooO^|d^2EVC
zrm3BVQh{1U=!sI3*z+DGE?bsWuj_PK6#mNP8s4T;$+T@i#UST?X_+3X0puWfB&zx9
zEH!sb&aLjJet!9vmCFOzr}70)ll&BTocS2Mdp*V%Fxn-wEKH(Dpg^z!e~xRM&ZxR$
zUiGRV)c5s1-<5uS$=dgeA{wTM1+}Ga94RYS)fPe9EXcK_9@<SQE_3?jyg+5bXC`Or
z@crfc<L`zfw4cLR4XK7BTPkX5HG3H9m|5Ez)I5=_#;VYaSnRa*iFsMO`KkH?hZOpM
z8NDDFxE_lJ_O4C0z5a=a0O|20_bIKs!22Ell0JUF)h6IIy~CH1ri$TzvG^Bt&;_XL
zch-Y<V+G<dJN`mi*fyA3j<CV-mV4U>y<(fLD>GCT5^s8>TlrT_B;1nalsb#u!0Mmp
z><Cgd4HU>)%gsg@%kzfnzfS1y0?OdMV+B9%+uK__*JVRswJ0y&i4#FQPL^A6$L9+_
zb46;fHU!Nf-gWJZ3Bxz5_d@Xx!(Z?9c@*dgVdx4s$9oy~X38j{`B@+q+HccX2c*Xg
z?J~LOsw}x&wk}{t@McqLt6r|bI)eoy^g|@G*da7dF4ov(26bA4^B;)XSr1MufY@eC
z$nflUIT_&gnFN+uNk(XrZv|&Beplw&e#23^WTn0&5Xo3#nQO*^nK3rwm=3%u6k&*0
zgCytD563R}jWmIHbGAAmlyzjqZC{Dle$y0GM7G2-(B&Ft;&q22EkVrvQ(f(m^aih7
zi5Dl$qGbw>C@+$%BP||W9a^c{N-c$}e8+aH$nJ*UXmK>NjX_yY+PQ|O?fpEUzWq$T
zVV1teZ4^GDZQe_|tkSw*Z{CwxaW!U?+^RF7IDd$o=Y1WV9GAB8Y~>j%Jhy3dPAu1=
zb<{0gf{nfkN*VT?*$DIJnOQ*CRb!N88DnWa-r0P(R?)(a0|XdUh^gctF9yyq%&;9g
z2HR;s=9mq|XHeGLFZI~)-@F|ZSe2;V`FeQ^Nc?0fHqvomr^$oLyl%O%Cx6F2Id4}!
zP=9gvQ`P%#C3Tm=6py|(?%P(X{H`pr6vZUVT3q?Vm}2gT{4f%iV+n&_p`6{w(=TB?
zikJ6j^|vJjzrR;JNxKi<V&B5&!k^ocF+MJL_SAvO$S6qZ$B6ZZ9Bg~)RRNP+C%(Wn
z77L*)Ycx|~ocn?d1C$_S$hk0;EZ8De;FzC^Br@2L_9XYaSSumF&YvaMZ^OJkB&qt2
z(hP#<?6^$geUvjtT)y`R-H_8>N=>#87v-nv*B4Ao;s=KujzeX*^<bk*r8{}Z-bWZZ
z+SVuHEcnqfoMkJ8y(t@T?P<XF{yZR<iX1GT_wz;uHz+cGsrz8*wI0V9##Rtlg!&QN
zShQ$Tn!tBEIdv!Vw4ZR2dNwNdXq0)C%W&F;@Col?v>ajtRX94|(7|nNeciL1TIT_L
zegSXu+^ctsYQ4>RQF~5)RnO9mkrz}QPB*-+*T8d2+Z?ICyxv+mo(1=Ts;U;)gYLfY
z8OzqpqlPS=6tmo=Mcd?pW}#bxO&IvwAxKmUL~`i9WU%gfIM2!swNs&hwf;TJ*(aOw
zvc!)Sk08gf86`6IIpuQvDAmiu@CHFDd7nS#J$yQRAo`xHQJQOUha6(mK@%%;8+8B7
z5$lY^@`W~tDzfT0zfnrW**cuxa9TAhZj6jt1x2_g?ENwHIQT9rK!ih`|F!hc@y1&I
z^i_08Kv{a_*2OKmg!;f?SIp|26;c1wIcb#W%PvtsIc<Ms)itTGzlWP-_sT={;G}tM
z^|d=dM-DsuW_rnbxGsd7BL@cLnx~_M29p^mwGrn<P=`G<K}~}<l2Nqg2%7d*#bdN1
zf|}Aq%>*w~>Ctgps_JeVt+Tw1o?W61u+>AjC5k20S=iPVlcOUD!eozPDTiy-uG0t_
z+UIpa7&hhonlUR3?v48u#JQLCxx`b}B)kn<4p&shq)VWf?oV_<bI^2!mS<UJgjk1{
zL|*hInOb(96hpFH$O&dT&Pdc!F8!PRqdVT%aKUnoSJNauv<DJ$DYXbp+1Av>VKgss
z8D@`2B#KH$K%0z)0@)=Y8(V1o80}>{IvDhoa>)L<krqU-DAVc@x_17Ib~2W<VsAK=
zb=;;s&B5o0=+}NbJUU}nApD9bciZs7d~)*A4W&Nar}%yOY@txvI>V<kJz`*!#(JFi
zvh%<mTMaJi{8uixG+8(f8e93}kDsF7BB<k32SUGK&^XtN?nju(GSZR$gw--Eq8_w+
zXPn_g{AMd^+z|a8j-9PS#3Pf~AQdadidA}_F_rqlEN0+)_rd+3(;pMaL`%n~hj+50
z%HPt6Mj0eV7T{$=87{ObKI!Tn;mqR~g+BuZK2nX;rm21=q*)U0n}N$*zFrryulR7v
z^Old+!-~aZFJP~V_K;@vqfa{KuE*pURYren`XT60T~u$XJoEJsU9BnJY0{l*k&>4#
z6f{nR$Fho%QoB2x390~|X@<tMl0N0xf!l<-h;j69U_XvbcphdTVmYX-MVVZ*0)7=M
zkr`2&uqgwnN=IDaY3yn3=)=z-b`!V<JEHA^l3BHwq7K?9gD8{&gx!(fkHSins2*r=
zWJt2o7HYzX!PEMFQoeMf1L~t7iW%wbjjF}Os9t%bCO`9*2dy-k1?ZFgL_-#Ju_HmE
z@m%W`5z<4^L&p&ie8HM$HM34FcvmM3IH$Tyva@c4t)m%Qf0R?hS+2RR=h$~ER*1yi
z|Ji}u@B(+mEuCa3nxbhD=b)$AXDg!?k&Z`pq@2XM&1V~$WO1-v9863rL@iwo;~6V`
z^KOVoPWiH+&0e_`zJ`I_7=-szhss7fJ)>MjPhCxn$4ZKdNh9epOSojrT5X9fQ8zKu
zYl2<oM#tmQWA(2h$WFCSEM%544LvT0^JrlI5lJ+@LH_pm?&Rp#WRss`DJh^qzPI<9
zGt)W!Nq!r@*LTXwe(oE7@FJJ~bez~OEe+QsKP#mFnhPs8n7e$p&r}>uCv<q}eKTHk
zkBn+onpVmL&U1F!YP&XzS#s~kI9(^c{n|l{BBLBCh&T+X^?+;a{|kol`*f?^g7sLq
z7H-fYm%izov&`{#XfBhY^mzLsGU0l=%xG(qIp)d=JTw}it87I(#+oX^RtRx0q%R8c
zr~zd{&E(K`b#E7+cltA&@Lh$db)>S_X7^e%A!Le%*|g+O3aDTxK@3h78p8%ZNO=R&
z1x1WBA6Qw$S0jl<Jx#7tl&Nu{%g#_tt1d9Khj_W|Yu?Rq3QOVam(Huq5eRa=$nxbg
zH?)dYHNE3|`>nJKS1?EtRFPnySh2fTE-n(A)8*ZIc4$x+RU$93AU4sVt+z2DITBzX
zU}>5Jd|RLLWXG$`hX5mZDnYvQDvoKZcDMkc<is~mQXAMw7GM<deOK!a7*d=|dY7Y4
zVEKixpMI)`slX}IhdD*3K%B>f<$SEdXeKKbl2>dNH&w&S>U!_qScEs$j7wEa1VX2+
zg$^}axQC-+5M}@_n!{I;pAD<hMPQ*zO9;RGtc(iRcHG%)#;0f=R;qPb-6R+hyt!rn
zX`}q$$dBbLPDxGf%yLB25__XYd~|g2S6O+%d?n?5)>!D@4y)RnO)bcar5n$S)fA8X
zGj3|%eRFXSId4)#g&;7;znYSMad5Z()+TyRPAuL-2^Yqee{(s()p#9tC^Amhw8E^)
zsWGRGzV(q<m0HJNhv)vn*%YGUG=Y=7&pmTX$1yUtIPC!OXIDe>=wZ5j@shk4%=*Z=
z+NkL%e~s$mT#_>r7Wn3V_vAh^2Rp7=A@P#KKK{Yk(1#n7d{|8c%}kWz3u+Wqo}F>b
z-$L}#geBHs6mg9w5OQoICJ-f@?-?OQYHD&pwE|!&NOfm?OCCSTdeXHBo}1*GYcD(E
z4RF4o7T-6TEr;bZ4R)lB=TS8f|5n*Nkh|B``cm}`RSYHRj=3)S0xUC8Hb2jBu5HdB
z%ZwxpWhBK7915h9SpaoRT=f1+md&+A%G)_DN`tU~y;ePr`BafO;n^Iu(z&V95;nYM
zc_r(Y=|U*9Q5W7UcB{`c1lXNFC{es~SpgKoIFxzPFUCKw>650kG{zf$4HdfGIW~E~
z&DXNTtY{(_z*~eUvoE<ub^aZ~n)8bFxb%wsJDjBoW^s40YbAANFGl<<SOumnMqS`Q
zgp%KfgWxM#bu!$Y2@P%<VQa9gN~-TzQf9X~?{Uk@a(+T~Ykn>GokZpenU}kanc;{Z
zPdzFd9egnfa^g!NYQ&*|mco}lg-GFKQvRg6HL%lSsmtAQY74XOxMGT)N(`1GHQ`_1
zxKkNK7E}ofBa>!Uj~kOYGU>R!Z<EX+>@J>sJPuJRk1~(Go9aK;zOX6hcY{JQ8Vf$F
zu4L@5WvMT-wj8p$^aRAa3e+<ym1UOD4Fxl`Z4vA<X#8x{{Y6G2Z)C?brwFcM?+G3Q
zo7fxP-ML${{QDFJal-||uxmfN*OgrsQGjs&2)!Y-`{f2>6zhd=$x6HcD~UJplq;C%
zbh&4{G(LN1MTOP{n)-E;EJ6*z<8uBy8BfOaIi;h<YRnuwTuz4e`L{Z>80XRihpast
zuBc{I{quu!*Wf6+mf^Fh4~Q_8Ytd1XmCsRc(r3oJ8a>V1MBpFpb1#1HS*8pfIK*_0
zDh{=Hg0J97*afANisELG;8Uy1N^sCgad+eC^mpGB;`@<6=UM26Ef`%ca*L_^P3pcY
zvTW-lr)hbfmU$~AgYttC?K?S_rgg)XrkC;7CQe!L@FB?#*D$geId_BCx;)D(jFqSO
zEg^7P9k_T9r21@TRkWMA<!tyk=M*W6Tc<24X%IA+h8$mS)L9;EwyBawS4ZUi(N*(T
z#?+GH#J~@gh)1%fMqki^R;Oltz_^|#k#p<q478S`YV>c6MZk+&^Raw75!A$hjireU
zkB~V}vWW+S%ovV{OxM7_g@dM7e?yZCo^HKzbNUltSH7`XpuOENoL_-`_B_*3O$Z_E
z77Zh^sBn&NjJZm;b!xM`Jd9g(<7uG&Yw29KWVjSBq4w;o%|@CE#M5_zl>;PdA`kk$
z5h_WGE5t>Ud9s=bj#rHV)@c=c#sc0#CEMcxoh?gPK`soxK{*7rue_IjBdXTsmN)t5
zg%HD%4juDvw!SlG>-if+>6Iw*L)5%aN9wWJo*ygKW|zi}pGW40d@lat*EuHEO&W&l
zI2j7~WaTQX0(NV25-<dnf^MzJ3ze7sDz{pduJFyVd(#ZkJfp)B4tRcELlTItGmD5K
z_|aKCn_8Ag3m`1<1TJpVRwz=(9iZR&A8Saz`w5Rfzl@wxwSh1xjQboO3M3aX<g`87
z5G>4gHh#SyHx@F}7ea>kkDK8Fi8S4RJO@1df9dH4g`yN?tK6iQa^(~KWaSp>qD>`!
z0X-_024$<C$|nWVWo_P@Nsl_CkHdUjE&nk``S-BW1wXoE5yD+d`na_xtw&6{>^8s+
z{8Ea_77QP3Q#Hm9?Mm_ajc=dxzt4k^`}mnkSu!)+zOC8^#b48&;n71s=#J)&kD{CG
zoA<F~<rDv8&Hp;9Lhhs2rH)oqMNz;Z`@i~6f$Ic>sYrL>-GKj1+rMWCxCBqSbOQ+^
z_#+JWpOYU#7zH}~WM!HnbY*FCuwjrxu#$_*kYr_hb!i`6s{8wQzSRHz8*7_ARHdt8
zl4kpRTzcUA5u~Yfy11c!mvnSNKT-MLyyx|R(`250N>ODcDsvw)rRe@24mq;+--$kf
zDVh(x@r3yf<DFW$?&-sIwiKX{w3I`8NwxLqBu(}0|8mG@T}a!Me=n1iLn|R(Uz27_
zfKLjPKuP+Y=?C0Zm*d;3KN59OGYAq`Dj(wh*Z=$)3L$3;|2xsjFJFHwe#x;XIb72I
zc*gp@=bt41--n9)1nXo4``;(74=sSQ>A!z<uv+Z@fBHX8%m4kH_~gS8R`Ms{8lM{a
zqJd$_Oh;}Af@t@EsJZbXkGU0h)Qp<b3l-eB)p-lN;v)S1Wzk3sQl44z`kojdc_G==
zQR6R0P+HZSa;frMlh-{p(8v$63ZcEf73Mgp+6C8B+K~f}c#SUuX8Q{wmKlTGLoq&>
z{6PU)R=UW!HA(0&eQ72?0Sp6%TbU}z5jK#RE#h?n3{j3$5gbZRETlj0C;z%ZxF{HC
zQ15P8nG`9xY-y<?B{9lRvJ}6sKN;8-eJ-vG9f#T25W{+NM+IVsjimOyH#`tRMTHAs
zgFn2-U?KjvAqo=T?;nj;&%E<ZbP=$a_2dfpxmCaSCV45BcF7`2sAcQA#R~e>AWDd^
zs1UFPOmEpgdc#9SUtH#S*)ldZ&d<Nw->513O{n@Bmsv#k_Zdx#hK?wEgrui&2X210
zAtTblB2cR4H|7e?;!Bc3iesxbx)Uft<!h1F;sW^<XC~SG0qfbk_^Ei&Rxhp)cju$q
z{=8qEbEcY4*NYurPgb2Cx$jBVM}N(hs63}$(kZDt?uspi(Ug-7JD<1H7|TTqqrgz2
z7cd%4j!#5kSQQZ)F#7mi4z;d$DVBC2w$;_N`z~!!TM_QS_z=WUQq$JeC@fFV!+(*J
z8_a`W4&SmYqlTfh4ekCPEx<k6)>J-ok*TR_AQ?;34Q2Av<dU6<ZWU2X9Zq9?BYtC*
zBK5Y9j49iUM<o><E139U_nxte>eF<-l&bpf8dFo1MVd$=jA5`6s5RZDSCk!A3%)6B
zbGamEhrOkh%#S%wb=W{VNh-srKCjrtbYw6!H53s+y0k3Ng$?tM4EzweU&@hKBvfRT
zO%Rt%L&(BCSR!l9@KE67iLi^>Ny@><iqeL%U0si+*^SKt8ZZIC^~q$OtmrW)y}FI(
z!-}AoVhXE@Ub0|Obw_$(3Ll3M^22DbXyWM8Dapt@tp!%%qe$Y70EozFSFv}>pIy@y
zqmu`|jV(J>jxl}~mhStrR;|k}<AbM*+l%&O^TwuL1@K2uxIMm^xqy470A*6O3%-3E
zxgz5E1SSQBivG7OskW_3GQnpU=>`kDM<yeb;XHWGOs)$P=(r_ZD=k}wPaFHo_U^Hz
zuVb;XqO6{S)_m#KlhsXTHSKTf8;`B>DJhmI`}-b+j^|Y|>NIc`P<vgK(}nfMEtI)K
z#^Ia?X+J0<JVHwq4+(0%>-M2dLMXFf3^0P>NNzmU`b(qC5^)*`U6TKxKoBtCL%X0k
zz1x4cEPB-4()?&CIE5Q~)ukBL^j*JMqN(esnbgD8blhCIq3=HEC1xCmQ;d?Ljl)8*
zoVF3H!v_)hX~=;Q&|}#6hMsPW#cY6lc5-B0_J?Jx;P^oz!p^S|cedaAqF-YtjfCy=
z>YTn3zksQfwWo0KO{Xftfm<RZ?vdT=J2P35^7|<kiFDb==h>@WO<l`YjWc2uo9&wN
z65Hl$ha1&v0tGnFIZNp}B9v7h4MRKLEb>{~n`vMnw}*4tDXT7tV2eC*P7Vc0{ygeK
z4}9qRX>)WgFJ1a6kY}q~C$@wzLgC8S2wueoVB-l%w~)dk1j9W9v7*JJhK&I?nxvW<
za0euG!ar#?A%G?Lf01=oQFU!il*TP^aVJP{cXxt21ef6M?(R--hu|*3-QDdFoZ#-R
zz4?3eQ$H|xVUW$vS+%NWRn4!*p39hz(5On6Wu#I~#gGP$Hc<)qNQL$PR9uw9lA8T8
z)fPcolpZ7gL^qTM?K~tTYhI<)%gUTligeSpw~gI}lLd+9qPnx4(`UecqqMQL*Q--<
z1jRRQ&Sz+jm&6gg+j5l2S>G?qgt}jLX(GJWc|qB1J6Bm=-pyEq0f-4~NJ4&sF1qL<
zhtS^C<jT@q2c;(+B!|aFYAq#=L);Ros&95F3LEPeZO`mOw|(968izWlUHJ2<C0171
zD0S_oSgv@6(Qn+my(OS$f7x@Awk|@$==fW-05i9mc<78N?r7DwZ=^$#kxyMPLBf9B
z_0#!Q3iW4|?H{(k6Cup4q`%@*9VkhoVip$Gt}rP~ewp?b03j;*JLQx&eZ$v{DP7Ac
z{)V5uS@BIoCIhEeD{--Tw(f3YRZ|7o5QdPy^-G>Qmh;&p_wXkw&9>zq46q=_C(m~^
z@Uv&*&c?sI{Id7%t@Nb6{;c|>yOodjp=3G4Yw5ytSi%VLVZ)124;cARm!){SfC&PR
zA2uy2*%<|Rc$2^ek^E|t%=0K|LP{(nP%QXV;G=ilRoO8U{KpLpU*RIc00%!wSw&TB
zwW(p_mvW&E;6--&tz{&FGHZcW#+#tPiYa+0#vqNuew%g5xc2>$?yD}6b9-Zs#ldu-
zDAMmD-2Y~H0G&NIS}M6u1?B*f`dRsb&`Nd#g(41=5|8oQyfJBNT8lP{JrajM+2(RK
z0EUP_)grh22b#yUK9ht9b4GNjWe_vw&Q);nm{YX;-r6q5@5ZUK#Z*U^vp29q6tjbp
z>MDGr=ho748X!cu=ss*}Tj9nHds$0nJK-2TbEGMcMOHCwd;}r{Ucyl+)9fV7ie^i6
zHGf>?{<;lUf!3MB;qPgaOvd=xfBxsafjJJk{>GwN{(PV2K&M;&43smJM((69(Oi8N
zd$#}AR^WIj?3{D`?_5>p5=7V&>cdLLV5<jA)sCI-gWC~^mpO1bO!oNKq=o5>W_yuj
z$4<>gZy~%j4icB41HT~Q>FJ5n6W4l`k|}if`d`79PEcE}S4xddmXo<|^rhP0hl%Rn
zghzPeA!)6##ZGRN4VC2x2Lu~lz?j*ykB6Jx`p^8;66a9N_2<f()rJPwTx%QKx<Cpt
zJmkD^X>`>AQMvp+3%LdG@o=~9J!Eh(guL=11XBZcRx2$q#egx$c8*^ILEhKJw9BRR
zC*nxRp!xT?yKdWc4%*zqd#rkKDP6s+WFP0Z5}VbZ4zr^?4Im9Gi`AwJbzQxSUDWlK
z*}L9eeWBGRAKdm<)253-oEL{;d~{P2<@$#!_npSUf$(qfw}m*`_|7;YW1*ci!c`Ez
zzZzs@gj^#1yy?>$bcA|T&0h$_tduZaO=bw>oSg3br%s~D^dHXNOEi00JSBb6wco|>
zAEtw82YIpQ%cnN7XTJC%0^*FBa1VDc$+1$)8o@y`4FADBD~9tvy-IvA;5Yr{$B!(n
z1W|m3EBz&R)E_W?z520Ikz_^7M?;1Gz-ll^1^1@`gQ~L1y$9b98!G<nQYqAR74IXN
z|BG${di;WqL4x`R=BU~S6xcRurlsPnO5m&w1EreIq{ppz!$0@3U@`AXhwR5iGBGtk
zS|Y0TQT{nKPur>86S{6D;110ZGoah3(B<xRagw5eTHLLy6V6V^RHl<9hks%<`Qmk6
zY=)g``t<Es5^qBqMY)w5rY1>~`jVEAG%!gV(cV$>s6UqI=%O_nG{a5t>P3YyL=2hb
zm|~7N-;2xKTP6d!f8gKtusA;9fE7VZaa)@(_h;x%?a94P>=vAKa1+H~;Z7i%|43WT
znxttiPN5j`SH?TI?n6V|iN|)WGk=PfDh)6}Be0N1xB0mcvQ-MBsz<t^pvdtBR;8%)
z0DmQ(Jo^aPj&7(s#ZxFPM=1IuVl}+|E$AOu_*)<QPB_gMu~J&B?C<Ix!sQ3Br!-MY
z7Y5AowU9E>s2AEvN}XFxHrgWQz$0kTP0v^cDM$S6A(NevHh@i$S~Wb0#QAaGrlT{E
z2BRQ4EXouqnIL7XiWXIL?>MODyXOst)>vl_9SXH#R7qLJY|GFb;{ZohaXAFcTRaX=
zzmjHh+KF3DHh=u1$hpZnxV%p7&O*Au52ujF<Qnpy%w$s_d=Y_Dxly%CKkt-KkLzeS
z4R+#OcSZ+kO$oeMXBYqpOtd(uz*~fZ6X1-ftcnRliO4Ejy0gQfrHeHm){+fh8o+jx
z=kv=|Kb4@lnZ<d7C3~hM-O64J?5?rKCuZ8`+({;BtR8<3hZIxHLk{b`uvC&n8@uDa
zu?Bq0;~+;&lD+^(vDokwt;#bk6kNbWaC!0jm6~$h1WEP-t^0+f6f<6oMRXqsNr22S
zGR+cf7WKzBa<N0*E<^^3=&D-P3+=S3qv)Zh_|^JZl_rB(lOTVTRX!qfIl42i?G?H#
zCK#d{)_yNTL1w%_<ltTN444B7nc!|%&+oa)`4$=0LI#LYQBj?}Vak>fm%mxBf#CgS
zmaS;0M*QCoWQIm9j{Cq`ad`h@kRkQ71-K&3m6_iqktetm#Chk*PkEku^KazR-y_uO
z&T@zxCs^Ykicii(Q#8T<PO+NrxAD53?a$oZ1b_Q6cj_T4!D(4$a$zQ@EkgzIVC0VD
z{^0k1)A1jDj^>lnn{m{w`~bg9I5<97_FIW+;KyaA-@tD+0xa0PA4I~29B2cT*f6!b
zX~|u=cWZ|o-Rg#RKXv2w@tE%R-5YX9X+0t*bF4VV1~%MTJj`80AyPCtD4qN+o^|u{
z#aMf<{9gA~uS5wH)ms@U)8o{si?Yj?AKLtg5_=1+3y&s)aC}Aiac;`%eS=yb=iT{Z
zGFYO9|3~><87-l|@;v@lP`RL&`tUZDY1vw0>ekiip*odPbn-#_=GwHj{x~C+kVb#d
zOQy$pQUn1s>wZV2%5r+)B4Q#)jPrXxTM>y|edohWAw{<Ic!-rScu`WkW6r!?Fw($H
zvWRrFQp7rg51hJa)G9YABMJ5A*#FhYnEj{B3!k(cw7&OZ1Pm|TjbCr}qwT3m!>iJX
zh%5)0-|XToC`VHzCwDDVnXF8|yiX}-mumf~MN_Dvr2Q6$ao@wio;(K802WR8AHVq{
zFp=PJhh4KXLmI1ZU;u8$j=psLPFu6vft{tGyYzai@WCdoDyzKh#HkZU$VTAPchzwb
ziR=JOQP%@zOJEEwTM)p8ce>_GsxncI;vze}*m6ByWjmo5s&e>(lanS%5k{HDo(^Ep
zSPQHkg8|*fnJKL`XX&UZntF80)^820Lt7N|F<&0rM35{u?Elf0g*l->GQ;d2ga)S-
z8@X!XX&AZIo^#Ce)a|;15%H^8B&bD;ME!wTHy>iQzfrd{e0dW*t-Ai4{Q%2(J2fki
zy+6Mi)SBj&xX#J`!uvE~CNTfy^5<<nw!?yw#GRT<8%GNUE+&8k)^5I8tnqEEi9>-X
z$#%s%$n_H_7%kMmnQYSy0?jW;Q8S6by~otl*{nAGOna*|<BwvU=0o?dbHXXOqSodf
zM0<8Txj$R;zM1TVUB4P8ng-UV<CvFaOn!g8!-@f&cl&e4!@*<-t3lDpQy}icp00If
zejfaZ$;qofshx;aYI$09t0vPd^BN-2EjYE(4ok=v|3dL?iH9WT+1xKu8|reO(sTa-
z^cZ6<0>j!<25F6v-!>v+2o2`fa~wy(HJHK&TMl_;crgWVspu+AzvYxDeyK`P=A<(;
znqgbG*!+RWbD`odW{;wp-=6=r(l9`y{W@sa>$ds!;B?lwbiGc$S(H8w@%-<ZWV7Rp
zqN{!n|I=m{uwu8<pJg;IfD-IGjBKKdlr!m}Ah2EWlj-&%LtEZ{tk-*mQdPZL=luv7
zXY9_>5PLse_shbB3FC+!ZogDYv?iPg%vdHX%vV%KuwieudE@x}+PRJT+&7THIHILD
zHoE<5V7q67Ba9UFds6K-W_v#O18T^0+;de!#4`O+Y=%5bv_zjR1BD<B;Q(UN-hVcw
zLf}OSOrHOQt3l2kGd+_RW1d0<HT`ht6{>hIj`CNj(FJs+vU;WUV?LTn5VzJrx?Skn
z{h8n1i<_MUbx>5~f#3&UtoyULHym5jP`gwe9Wt9-f82#%d}fy?E7*{+Lhu3K0k?^+
z!ByAf3wn_f&%e(8l|SHLz#GWbJNBCF{4(iRyTYs}fh1O*|Hv3jp680Jyj;9uMmud9
zFWEWe3|0hVZSpCR#jgkl2$wP)cr~s2c<0?8z8p3y1FZ)+?E@lypv+b7VV5Oc*q$i8
zS3@AZ7f|DCg!<<&$i!}}TuycIoAVg=RyYgO?!)aP_T$Z_dt7tcZ$rfibE{kL!J`56
z3G<MFmSydu1V$93$5&2s1#jS|xDN<`{9RPY#2^Nwxur^X+HAKu1uUOR!3C)6I?s}B
zNH9`qqnBFXcoY5cr-7=??s1cD#?!i*O0|L<KY_xrc{ajO)Q+MK1My6cvNxIYoz6Lx
zA8qo-YTo<kO~ucU6VzDot2SD@R9!PS)<F?5yPC>9NG=_oGcLb-_6~2<8*y=JC~2=W
zKWW-nj>$EGT(wNA)H-tck9S5z=xC-oiUz3UXuOCL#A%<gux5NpeQc-@5KzOb1s48r
zn41Q#6INLhE?{a9ToRMegN6AE8Xuqj?%|CDx3MMi1Dw`}#(&<kY&;HeCJ7)+$`bIf
zu5y}g=y{#py`D~Y1$*wV6^IS1s_PEoGx|X?V@PB=%G+$cxV`%W1x^95Y4ITtHKO9j
zneLb!Mx><=7;R)Z$)|WF$$rCed|Ef|<LtVl-+0-0xZZep-=sq7JuvzY#wUA#F1%^)
zy<XY9%`b)?$Exg>F9Lc?m`<W5;SGiSdDPlZPp!6)#8QMMo<`td+3IOQP<JpdQs@4e
z^Sa}PVMHc4{`+sgeb$cw_9dO~Kq$Rt{9pl`XEz9^)MTpJXlWKv(7mq{bGI|BstvdZ
z9G6W}9j3UaS6?p7rhjjB5p7OyzD{o1|1}7^N@;%O09*1t|MGc!2t0P@iM~UB^=^Qm
z0AEFPe1lae&scj?i?Y^q*NIP8|Nm80jl1w)owcxy<<B!5HYa7UkmA(WpEL2ll&sqy
zas-C5tMJdCdy!jq#%Lp$uNz{*uJ$>*)2DyFDTrbV_k(zwuCosK!fG~O?>1O1Zfr6=
z#<e#-9i=6y_jItB^J8+fXlepmc;aYY^O##WhGqWS)fm_7e6m{ennsSo)Naz}95ODs
zOsgF;MP}C%L9>**{3WeUmfMYNLwqD^Vx<U<4Y`0RY{3V=0<X&7^R&$5^1|>^dG(V*
zyU63h;+R{~7&&T%C6=`Mbp~^-{mho4p)++crOFNLlJVnN?&npc`X||TzUjXr`lh+h
zSGo(h$(vrJrzmg7UaU_QqbuxhM==&#WNajkXpO@PJ_<M4ZFZ4S#yQ0TcfcQyn|%Rn
z`6=|7bvqI2lwC@TRd}M8zr&v-)0Y-d;%G1s{{ai^P4F#jzbkCqDfR`qkOz9G7J#|+
z7f4fywnuULzizf?w|94Ef38p7Oz|QYp#vKb8T;ygiA0>)f+7cJA%<`=gvG*RPinez
z05A)55rza|y6)*T&r2(Jv}X2qzBUHN?}4y{Ex8ZfAKiBzy{0>tZXTLV)=z{0{AXL;
zcTaun?Wim@GY|tUk?S`Hma?H+=67T6TqjPb1KhOYV0@BIKSY**N;OaHj6L~#?xET3
z5vA40wkF`#8F}T2k}*0K?mL0sarkeC69>^9u#}y?KU;mtBHp=5`=P*(dBG3hvMIrU
z+;~5dwt6eo_bB!G)t$RCVJOW4Kmk5pQ4jj;YblOxt@K<CXOE=NYyWL>;Z^tBum9Yu
zg38^QC-PZf2F5BG|B`yxt~L!CDb?53_L#G6_rekx-|SBoCOzIa+wE1>=QPu;t>pym
zTmM&RTYWs#0KAT(X`49@2ZrhHqNH!>q9+u2B!yVo+TQlED=7>mpiaSe5()kr$p#jH
zkeuEyfCfeRD0(-b5LQ&lYNUtVnE$0HKhiCi-TSmW&gr3v{?Kqarn>d6iI?jdH_U1D
zgb^mM?3z!J^ZWbDB&Sn<45xf<*w}{MX6M^~IS#w19ov{STO%d>NL6&IMYsxBb-;+h
zEeTjkaB(@FfbSw@%T4te9Rq{n5<7_M{BH-QzftTnhgr4@7LYV}Z!*NpdhJpx!4m8H
zr9R{#8*tBqDjPpnIo5z6-(WKi9QaWH<v4lr^MMD5_7_!aA;YhKhXsZJFJ6bKt8(>D
zgae(akXpQz5aLRMKW_4xMq%O28O-sPnf#v8f7=T&D6M&NQc?;XaqPUoDjvtwyC$?Z
zpF!Jq%6%Hs`wR3|T1M>Do=G4eE}*`S=Dt!+@!iOtw;m7@x^IC4aJqj2*f!D;{m6n@
zW_&xi)@v=F!>@e6qvQmcIq4^PFnfVl{UR$RQBLO(#LxY%IlC7j$NHP<Uc}dMywlRq
zRYuF;0t0*#1_Ovw?HSqvzNBsCpWp<Ke^FBxaW@QJ3_kBSdRzbVhil(W+uOu9Y5|NE
zxdF>WPeyv;gNhJ-zR@<L=tb}2MD@?C!!X)IWr^$J)#;MqD#p$`sJZ*?tmEy>4t9&2
z(PNhI!K?p{^SbMyGtMZM`_qU1&Vz9CwwP0Cah@lW-brtRvb|YtZdt6@oe?7}VQmO)
zp+^H+XY4m@n937KClJ^Z39aSP08P)!3aWDx@U~^(64ra?4>DorMZJIJ`<v0lzu9_b
zEAae4^iTJ~*Nlir%}uu%kup0~72b2s&v&eh@}@0HZ<q)etcI|k;x=Q=s(Fm+_D0rj
z@{NNfboc+Ha?#Pz-Fn3meLN0-^0?-=@W*#1u)wE7`$1QsGK~L{e#nC#-uStY%yC-(
z?TP)$>qwS;&FjQOnuU$L-}S2NJOV|fE>OutAxJ6F7n0ILCYZmtU6xGsl8>419pA|}
z3xmBigm1eWw|Qr#<MwtSWiTMa;LiPV@ls<vPd#iHv%O+fvy<)jT=FS7^l=YEW9#%t
z2wbDa3cmlvWaMx6IajvRbz8@NihAe>WAwes#f_oRvG0ZUzAYfy3r3Ii8rXDuwd;QH
z>NwH8*zPBKXVQD0i|GKo(sqk$yhGGGuQB7wJagcbAyktWAP`FFNXRPHWOkvRwkRvp
z`gZ(B$U^vQrS8yQ^qI&GPc8k1j?6CaCN~F%!^<qaPb6t(V+o4?s5BY&li)wCOzQ1{
zF1%r@Fer2EA!_{IPBb+%^!f0Bw3-7<MB4l6%aY?Ld1_umpPUv9`Du<@PFwlqzlcq5
z{){Vm%33NFcx()#UD(x&zL+64kRw14FsLf3whBGjhbO7^ee+CGUIYAM!QNmRtE~W8
z%X~Fu_d6q*s+!Yoo#5aXAjeuO9n5=vWiQ6fGdvzgyH6`MJ!U@o8NTwBNcDYnL9^$O
zBB%8qLAwO0ZQEUi^|CLesxo55t9M3%AktPf{i^#%)5A4y_YJQm$^j{}c-LKwcBzZn
zwu^!C<*%wtyMyV?HwNF22R|Y!L>jT^!1iVvu_94bsEx$I^-8^N!UF;~oAthji?yNJ
zKh`S^@L*!2v&w;3v_-WD=g;r!*6l{DDNq~#bfZ+cfA*xTc7N&7v+m0N<c=4{D+&PH
zAy2*8Zc4P@odcvT;NZWly`O0>U$0LVf^PD$JK;DT$kmtLFQzJIgF&o%2lgtz?{KP?
zuX^4VOs+f6v_BAfoD}@j7!?)Bl9=5$m5(B8*6jHy#G*MGc2MPXAGn9WWTi{v|F%83
zzW*CTb^>w`@R?YNRwNbf4>cR7v_TC1pf-r2QtA9K4AR+Vs&6zR1@tug@o`FO&=u&a
z9DuRHUB>yp+P-EF2e*+n7g=|d1M$P({UNBxYt3Lisomb}ZZ_?yV^jYEcar2vP>pu4
z^;-EJZU9TKit085|0=h&Nxk>fZd1T?O2>It%zEvwzKllKn1*5aLeuf8iOECCw58TQ
z4jM@1b*ScpLj9))`R3cX+tbj}d-CV4RdM5fGp7DZ09Y6~=bqvmV(x3gUwA^#mjJ|h
zofI5gjv$?{J~z6p3a|&I73G38(kR~l{ooYL3Q9}EM+9WurwpD0TsE4p`uVY~Y%{K`
zX^-Q|!R7!QbU9Tc8J)y7JbjhqoiWO&_m5@X3#RsR#}7Y{me&PfpG1);MG_yMBNd<+
zlmtbFA_Of~_#NzGnjtUmA+oOsyC~8fyyi|Bf`nKVx==>R(?6btJ|_59Dsyy#+pDJ0
z*RQhifjcpHM0J;f=i+8}G^_q$*{9~(cbxVbYBzotJ`ly|BSUj%7}eyYcQQ5UB!P{C
zqu##PYpP2MAI)g=;5!ee>mL9-`~Ctn?AGjZznke?8zZ})hhVP_@I#M8lSF=N0MTi8
zCGcK!B-Z%eRos>zrGr;0<5ezpZ8i4go9A%G0j$zQ9615Ko-`;Mp(osz$i*tY>z85w
zQNW|!cG;(XL*c!k;h-GKA-VZTF76Fz5){_ep^`w6ibv}MK2Q@m{P<Q?^?z?KMgjW|
z*dBmfN|h`f{gt;#M%d0qhN$J^{4n<X@ueR}`xjWUW<vY?5S+g7cSwLdg7HPYdi&}i
zXE(c*wY0q4aJQ6M$LC44@@H2PVPhvXMam12ckoGdd#uHLF|+f?j5X2Ge&Z|^?o(E7
zewxPpikV>Sdmpjq!}%oO-DqxOWAd=>CV^nqXcmqCfot{Zc2rlq%CMwY_d&MPxFDhn
zA-rO$4ap-cj*3+uQmD(*WN&j)9K|UogC|5Wb*)4)I_iI5CNQp0-Ideq_kbXkm(FAW
z@d-gN5|0BH5(^aM|Da@OKSs;(d-w2XII(Wy`yZBe;H%r3@Y{$d7+N+;X^-F+!qS@u
zZt_Zg!18L%e%=M>r=X_~KEj@n#v1*{y^CVo{pKpZ7-zT%d2i6O^;OE=M}kLIHkZf<
z306L%XL0Exz98<6ZrJyqMl?I$$LkG&!>+;S@2D2+Up$ubdkC1%KZBmQCJAn*u$Vq3
zF#t<a7=kqjj}(!%5F<PBZRuP@ep!l?bCf1bW7&{Zi@;bMH5ero!$h6R`0=78#o~Nc
zfqCxpy5HP27W3iqRj}<Gp}O;43#G*kymstN$lYwFlK+15qQyz?{ltvZaYh&x{PNRi
za+uI-ra(Q1G)^F}%jZQ|UH#x%?47?Iu$o-!2t^Bk<2$tMe5<lM2=PjlIAh5Ijv|!(
z8$wT;%ZdBP+r!aP4bi999y>dGo=$oI!^O2Am(D8PIC>ZY?clW!O<?)A2w@sA2K6s0
zMsQ6eYkwC_(KB!9b^A1o^IEPY1a~W)x}vaRgMD5jDD4*)mo5dnpQzzR8<%tj-KO~Q
zsgVr8D07_&Qh#3OuK5ms-LrY}KxpYum;U){ZW&$A=`Pv(244>xQB{?MGTWUlRnHZ*
zdENIV#qZ*Mq|44W<NL?E4-ShvSWWjmw(IF;*la;4@3*OY!&OQ$xT6!QcBMd+zGSwW
zodTQvtE1RBwl?9a^S^Y~)D)#SY3T{%N+~;W+$pMVn^O5?Sfj5g9&lfb#A7JJMR<64
z!UN;UzhD=jM<qr2y=<p;K$J!rnnhoz!>bI`4d;eqq8ZT}r1j&-VC?btkP4s`-wvbw
zyvmOe9$u+1qEm*^pdC5*Lf~{-r(L=%oU+jQd|Xinas)P=Oz32bFbr@KlBCJZRpKmx
ze`!SfVuk*DrCdihr{rqmLS5CNpU#!sQmGwJk>kI7i2pYWkPO0Z7@AwwETmD_ud<;G
z7r|oCfs;0DFxct|5LHUTY=C89`~Z$}D!`@s>JQ{P#4;%V;d4X-DJ*UghD-|GZ{YX?
zh@N7(orZb&@%O1co$MUMhU*Deqi7jL3iz7NdpsaEjNr&v%rTy&t*)%N-z}=^qVXf*
zvKWUexTaVk`^(5~x~|%M*0E5+JwP&GmFk_605~|Iiv6m<1sa1texOSwTyA(>m`GDK
zZ*~*eyF6{+m_SSA^K0zqL^KR(6>ZZ%!iMx_uxM!L>YgQJYO#9WngixUl#zpm#XQ=b
zp6vAW^uY!rYpqVGB8qfa^;%|t&$Bl}UcHL5H3J6DK{HV(caO(o*pS@mckn)*8&fN^
zCM0+%H)|~1ygf=a%p<z0ubCd}ZFje&NRpCpPN=M=ofFZTjCFYi>{~teOmHLtv;7kD
zm$r~1aK-D)F#xT#_M}vI`C!fmAflB25VDZD$uhOgabmqIfT_Redik`h0+UVP+FEUv
zaWIoexxB#JbE8G^aIWmti2ZnVSj%0T1`rnJ?rsdqa1glHb54wofKth=-)`uIAN3V5
z$=0Wy1%ZZu17)t)jD?>Hmz3X7_w_gWt{!fw4BecYPM_Acvg`(}+{?oel9X)L!0GX#
zl|^=O!&jCeeR!!#@$f&9;}j`W;eabC78vi{MO+h>6>$>DZ-m*F6}(??&1FjNskEj=
zjBVh=tkBw(=#_FfZ2H7OVMt=-1akB@nyuc{4m!rqJ`axDSCquTdw}hn>w|*WX~)eJ
zSTupE0>Cf>UiznNq>joUP-#rynq>vZq2*Hv;j2USf@>uhnneHxVCiI1Goad=Dagb1
zc1|!_wMq#6t9){my)PZyMlc29apF%L9Gzy=V3rdUdG?-dN27k>YxXO}C1>^5TgySn
z0+PUoqg5wXI0%N>VOuniIIphm*Z#g~3ZtI<#YQLK;SSIWA&tvFopDX1O{tp`Bu`_3
zD=3aE31jA;)lx{Tdpua3`YCc1(IBN-)xW$h9}5L(@4Iom!nX3B|K8mC3?FlO?Tj8A
zye=2e73Nn~#%$Zqr)GM;*X8IU8MAMa(|8q;NBTT0$3apO99oTtxTqpi_*|pB1Mc+o
z{r~aiCE0ycnuY8fI|+GMXO&+E_~jI~vlrg76!FEEpu>^orma6u3)b>hnJ+cYR_~i~
zkfG{$EUUgAXl}X~^H>4yo;$7Q85O{mqVJBcExo#qEA=M-0|CTQ78=dcva<(D2!Y=z
zcFX-S(N%@s+ZZ)<5(HuZ9qmaaGt|#Xv*>?qR@M6xcHY@D717OYaXJI=!(Yc6)jy~*
zFZ9{^innqcT*yO+rIf#gl4HZ{_JLK3!}kiKQ`UMPy-fJJQ2>_rH%^*CXb@F9h=<Y#
z%|{yMFJIEvuQ9kL3BeVDm4yu1U=&VQbCS$I66&C5gjE<CHIVhc2NwzHAg(1U!mD^(
zN)WdI@-^3%Kjh<eYS+Xmcl~D-*b0#L>d%UegKphGT7~hN_?3l*mwCOPo?&1B#RQQY
zv(xj=5b$vy#dravIir7lA*Vomq0?#z2VQEl5VXcx3(H&kADdJW7Ap{4*X<r7_!MrB
zmW-Snk}5D8g2x`2l7a=4K97i^A1z??{%9GB(ryED`qbOI_TXi7Q;MR$@*qMe^8Uhy
z_AJ(~s%@}W*wffND51=-3z7*@gKG4q0b##C8?d2MZpk=)Ipxi%sO~}~u`^v9OXhpK
zo&S7P4WoEzhaPZP`?tF#qj{awr2n`!ou`fkYXA^{S&m2O%XvE^zT*GXr%#rdTAWi<
z_rt2>jV<T^$n^k71>`5NZ525$IW#~T{|mLxM!$LBdDfz!jdPIJdYvYmRYw=zcI-C;
zk^{P>nNNfco%Z|~x`MzN?qg(eF@?OqlW~$VfQcCVp}-drP4?Mb%#O%}6E_&~lEM1T
zzv`{>MimTUgI1U{AkM4JRsq-NZ&((*(ja-xftI=QoIbG)sj@4COiHiDJ)~=mZcIEw
zAsiCP<(dd%gXENX0Ff6$F$-VQanJJUlC88s{v93tJ29M?<ZONYcUqt8dFAAq{QA=D
zd{bHj&If|E(=6vx3@vf(nb0LYpO{uBGspzp;pn}>#JK6xq`8POfZg}q$cXd?n^8mt
zMTEua+`3O^WUDub+hZ&V(Hm8S2_P=+Pv<Jk2YTR;*J5R0Pf^AixqfY?)#Cy!D~l;5
z)vcAtXO96%W!mc8Kmz9Y;bym>G<7h75F8!DQ<tp~{sqT`Fib+gcUY&fqY{&QFA=!&
zt`}RdjLb|6%?AGj7a`@FzNb<}X761|d7h)fXeqm@JpdJB2&AqZt4v`BKRp8?&?$Ko
zCtu&bdK;ptFeq3f(h$uaO{?#bHaJ6v?Kq3zySt`ATai!my&xu1DujaNAJarPgW+lk
z2EIMGVUZ<QPL=&AnVXvnOuh`GBs>TwjU~Q9c4UawXtzZ|=D$S|#`xX2xbkyHu=lXe
zsAD~E(0s;R$UvIB5b>ArZud#74b5rswAu4DuU3{a3N|~xmDSP0iBPMX6R1x;e9yD5
z5Ed)|V<8S_NQyFB)X9nU%RW2g%-llHUh2;Y07)o@At}%QBpQq&n=8N%DW8dtugPRb
z_=%DP?EK}AG-yojgU414yf3*WgYAi{Yo~ne#;7Sjm1b}(D<rh>wPn_)fk*yvA*emG
zZWWr@+13EA*yUOSNP0P>YN0v~6)|p+vL;_dFp<1he*t;4a>v$*oMydT0Vu*!a|_Ec
zy3@+WT}ST1qzx@jA#FMBmuD758}d1k(z$8<gap#>WO$k85E!jgLR_f}A2;zJ&7;Uj
zktzRro!&Bk0G=vX3z(vR!<=cZjs@70s5_V%fRFP*FYJZcMu!KTcGE-@Ectj!wUG4a
ze{V0>PeuZ;T-#BNDL(gv)3-<SfHsO-RaI5_voquC%YBi?DOn^$w2P2oKDw5+wuj%9
zVIXn7o-REl!Nmv}kO45+6TbsGT>teRt!AAew9~<qNviI!ewEBv(eVMs-)Aq<Mo%fs
zC2CC|89}mL+Zm+=zVGa%>P<Z_0m#!x;0Yt{?_D!8GS<kOQUD?)D?oUbe&=`ew{UZo
zaC;^F4mk$7pb>Ss3H_`2>j~BMVGuH_+tEwM{XOD-7k5xy!SU0xrrCVGnS$qCsmWMb
ziGRo!#*XmZ_g#vg-tS`ft9F9<S*y0?cqy1hcd_>|7}u(?1Lxi9BDLdbb*Bs6-9|@N
z3bOL}5GTSdV~=LmWWi!Pj+qVw?LQD95AHC7aU`Z8XuvAvDqowhU{aljz*aOP8|}=E
zHg4?f2fl^{fMWu7c4<9#OS-A#gxbI>dROq<4tTkM%@S#L%$7@q^uEpuYB^uA>eYh9
zD1sk}%)dWs>BZ}R2J-mp2+DNdX@1;gTdNl)EvK0PSjWjlb+eav5#!+hs37xjnMdR}
z!expW#`Bp9JFDn_5Pl)iXmlj!_q-JnMO7}deyiTaZhs>aN201BRLT@!<<J!i!o(f;
z6{t0O|9$ukJ4yv@mt^eLw>XkS2uhxNH;I5mDQQ^I5w4iXUj|wEZExCdKg3LL3%KD+
zT7Rblk!I)AE^9Jfk2ewDuH)A+AwOI&sbAoD445(cQ>2^0hXTP?s%B-=g36-(<v<WM
z27-#Js*<|;a8CdvKoT;$3LHdS_n_+I5t5XK2BDs#$X6v!2@gdR^oI!AV-?ug*fcaX
z{Q=Y^K&2gC0X6cyzYKHlEU?0v1jnMQpm<yRgrJJC{mzUlvXofMf<t{&z_LalL4YWW
zHz&Jlg0!^xMOfFm@Ah<Egi(HTd#u`{566s1(1Ko}PG@ws!hyx$nd_~J)Y$21R^7{c
zOEmH#qdwO)`)kChmB1hk4VJ@B9-B@_B&njf*i6^e$8+({X!&}v!H({2aT78chdb2e
zg!7N(9Zx9Y*U{Ifn^ARDv-{B1?>GdrVsQU-QVZ>kxS;!ZgPey2&8#y^h235^#dj%T
zgixp>yFiN*%k}Oa`>O-eES9qTNKSav<nRCiY<F=<?>qbvj~X?T_zfWDvY5{hS2oIa
zzA!A+KH@1$1~d&Zf9D%#3eU=xH*~&#Xu`?di=y>}v02w2MfhmOxINSiY8qOmJBGsK
zA@a>OFE1|#`(v!$r5J*#C_PrmX%C5I0nH<&6J}P~4)RrI227=^J8sb;2VL_;*eP?^
zLVn0Qa5w>s0!3|}`$)Kclpm45z7b=Kv$3%~8dr-A%d?wQRTw2w1iAsaK5z<Wv{->(
zkxw<2-~vRA=_y*d^Rvz+0X7b{f~4&#+>4~;OEnhz9^yixiNk?>V>q?=Y0Ouil!F|f
zoJ0lgUtF={Qv80has2lh#f9Z%Wuw(r8K;CS15aq2ob3aznRZ<f$~3%Py+N>MOC(hl
z$)U=T0FUc4@3f=7f8fk6V<G16a`xk?gG(YBN=}ahY2~<Y!~gWy!>nwq)M})GOq>_C
ze&#FcmTDHX{>ucg1B3{MZ%>z5wTbkP7uwA&Envv&)Lwkx@eU@<jNi<GIF1odBCdsq
zMVeGEi;hr%w9eXM!)!d*&DPPU=nqQJVIY%26rm9*Pg~_e#)n5CA(e;&+gkXc;Brw~
zi&#T6M4qZQ@mWx%&}oVg46i-kXmbI~BY-1-JcJj^I~y<uSbUe%)uk&NT3G)Sy6F&;
z<+4YU6dwI2?3+N@yh63^G@aeh(7*(QH37g8kR*7oPljO&NcD?bThj+9qL*C|zlS1^
zS!xM$gYR0R!2@-txt*QaV0j}OC#O+mFI<GU{vaq$7*1^TL7QeI5$DSst=dl7uL9Rm
z%dgbw{f`d}vysXY@YL)gR1L8ajvkRk|G{Q#t(1%`O|`t7ml3%2)j<gd*P^&3!NqG-
zp>AO9ZN8svYIJ#X8{jws)C<P}REd@qPr{CmurP-ShfKU{ji{?eV>*A#*ZYkZ>fNkA
z=8u;K|0S{YamQ%<1yVN67JCvcE5N#3MmbU#C%P`0NU#f$0@rMqMr1VnSHDJ=A@vXr
z9)8A=-cqIz&4mUqt?=tG*lp8i>1~4y5tnVoZ4$4j$MYr4_wD)e{pF$nVic$<KV9{)
zvNcg4bl<BH-C}j=y_>P9Z<tlhO3u7j-nBaVR&i8N@?yYAM$}aUQ21^y`*AJA;d*-$
zDSnyDp@g1$k;IFjQNpu~1SlUSh<_1Ap;i!j!oP6&yhXs?J2@?eRS92kjdYMR#~peH
zQ$^(1fzk;&5jo2r?C0}*lru&Y0oxVW`{Mx7=ikTJ{fsToe}?hto&4$OP9dKKr$QwL
zxIPIqRO`xh#h+yKper6Kg<y7Z1m7-~F-WcW#Ynoyep9uCCsx!IV@)RFuxkE5z^xs(
zOaS1D!cd)GTxQ@w2(D`u#N#P7#%FCvV01bSX*yj((<?kZjW@}F{zZ_du=!P^wLBTb
ze)3t&i4;!@wGv6cAd*04WE?1|mfpiFHhWvW#7ob$D+sDpTbistb8rd*KC)3CI6IZo
zpYz6DA8!{oz$_e+nE}Ns&xRUsV{^D-EL&apFu|O+n(=&MxBw2^gbKu`%l?!1hE21`
zQ(aktu~&FQsM@}|suHinZ<Z4m;Z}jf^{J^VJ(@PKBq=fNb4M2Hx*k>Os5%T*2=zT(
z-9P52Kq-X_NQ6W{^*HT&U!Uc5-s=5aTJZP)0K(@>WdC7zKSKMw+)p0A{u_3M6aMiX
z^6llJ#7tK=wcpCpTpv&wZ%=1~+Rr-@m5hd*#Rx0+>I7c{gdt&5np;P~6dtE#uo+FS
z^2J`4<T$}+YV-sxEiFsS%6?Z@#{xg(IBe6*tUvq}oS>)5<;nfy-@gM|-33R-o91Dj
z8F}gmV)5As;*ERh@4mzaeZK~HIcy8n<zn2loDY@@+Q1+{O^x-|i<X}E=R0H3+?+y|
z$DTp)m0Iop`3v)2BqfueLo`sQzC*hE4N8|=c`1#yo-Z~NVf7xnoZvF)4>NRc&XMT3
zADOH<k5@4atCjBmMOWA!`e;iAhlH_zbr{xH-j*A*_|oQDzbdU^Re#`dW3&n)%sYRi
z%Ew<gj-AZMNoBEO;nZ-7Zt*&APgNEWp%o*bqypBC<`(8;8B$55q;F4FQ-6Plo&RGG
zUcZ?wwzC$>5b@GeucMWb(RzJN%k_L%@X2ivMnQSb&?>W8xwp{X68X#Gqzs>)-8|mh
z3Z+#Ff#88Gf)f=Vs0d9(0^jeiJh~pHT(<inrfx(=W$}%i#387V3Ps2LUr5K}TD>zC
zFZ>Ov7!uHa;{yxP@vkeQv&(dK$Az4f(4qVF<wl<lAp|WJH<1iJ03&!TNYbL_3&!=L
zK9n>~<$a~n$rL?;V=&D(YI0HG|ES5Nr6F>+D#i}72lGIp29kLyLj(;VQ&$=O@(R}b
zU-D?d$-8|h1xP#&QX+{OBABgjq!EEx^&>KgNpNnJTr$|y<opQi7*S%*)*}ZMir;$3
zK)G(=@@#$557`JkaKSDQ*QDS&Z(u7WVf>I9#-KTID@#wVL10sZvFq_$P3Lu+lbu>5
zYJVuH(k;~jiX`#?yjzxZnXG(u<9=sI6mvuojXD@<coh7n2h&#Y`w)&w9+RY9BQX<b
zj;$tybJHX2s{K@$7=kVm2Ub)Do&tZ$dAJF4>RS_EJgcq9c_Pn>d4n3%Zzx<eBkt|X
z>6wOImxR1*<!uTA#@D@R6?OIb3U$Wvii)Gz0@%*y`(8<a690t?cT%o-Z2~~Da_`1}
zKRi7x?Cb=bQ(%7RYD;)|@y5i&AmXrv4`6RrLuL6_#RlUaG^3oPHJf$Ic$Foh%X5MI
z8_9D`X*L;hpYNhGgXYS1{)Cu4`$~tzM`n2E9It5IXEOzareLXH%%+q5w_hp?XWMHl
zFJ0?vSQb)|jWf(#p<c6(E%>2o%8dhs8UagVj53#QuI{C5l?1+Lg;vf4yHfPM5aBRd
z_fJDvQMy%iYba2|X9_e1!QF7v@%&hRKa>zO1c)6_!a{W#qHw*a5(Kk@)%QeSs}&M6
z3n7Zken9lqwHDpIHFE?@&*7%R8OJhVx7pwTnJ76KD2Q(eN1gUhpW~utUZ^O|h(ASD
zqmdKM-impMy1E!p^OzMBVu<*oVO0vlYd1ZKphG1XD${8%YnAXQoCVA?k8k21F=I!G
z^)ZoMmREgs{Tq;yk}{#@x>&B-;Bns$%HWFqG@`Yzv>eMh0?@6sJF&d3N;3FHGZZlY
zPMK!$Fh$LEn`qfASh{%I3V7d(YttP6DDV=GH{L`Jh1@*u4W`a6E9LVm0*~_kcVp_4
z!B>AHbiZcChb-tW7~&lZT>Mw#dhegwYTYLmct#j~{G}F;dYr>{$Xu;bL)^!QU!Ub$
zy4MA|or#st)V%Fau(w=(QCvHw{1d7Lu<{CvHLNgP-AjfhmIO>&86YKA)J}ZUfD?`*
z<X568>|9?g)$Q^o2eg}epNx9nN*%NLJ&@M({W`c6^kBY|RITso(M5g#<z{13b&=Wh
z&t<nBwFeUNg~dBzxWQ1Qvjq^y5HX3bUs8IsqNZyK=;2NS$$9-Il<+{Jf`FX%syI-`
zr867BJZydtKY3$<02+?om{m=uxUMce7;rQ2F?QZ!Qz!@w<c5u?z^*RVm-Qf&5K7VW
zquNeTyc((cs=d082+EpjUq_tBMlzl6n{ZBT!)c+}-!C4qgoxFn%EEF$>va_Ltu$Ia
z4K15p-RONp9<Stp>3k&OXXpPEQk6qy`1q$v@Sd*VG)~`a-vcfmLaxHyT-WcP&GnS@
zgX>=~`Xd0pNSlm!f4u#X<*}bG+i8Y(HYhJGHSv6RqS@w54bWjsfd6zljaNdeRgM|p
zq^aZO=j)Y5XTzLLeDD6yHSJypgvQS{mX(#!=TBN$nC^`w1Lbe@-gxQ_poB96lrG=r
z`CMOr`d113^YRz&5Xsg$LbfLAfPunh_@vw7yrE8qRq2)9#J{wiw4#HXU{v5*?Q2Bt
zOlzjQx5wr2I#e?bVYxLee6l*$u)oX7T%R;{Y`*oP!t}|Vdk`Z@n_n73*s!Ic#m^L=
zpguy%z{3o!POG6I1E7_m5bUfljD!kbcg%-P&KL#nZS)Rcx})O_gmS;)#2D>zQ!vJ3
zyNYH(qWY#f6lRh9fh}@={#U%<e|N_>lSc?*Of{P5kFK-;B!oT9U46*}G(g}lQ}Bs=
z{zsZN&y@;%PzxGmx+L`;sYwk^f^|@?M<9`FH&xY9hnc0!C=F2ZF5zA{=B;e_cd`Ea
zp&4~u@RSn*0Sa<)GMg?|q{_>W7x~>&Kh1|Egwta64vOX+r4|Ds6h5b|@tvjm>i*?u
z7&=yY7JV`Y*$UbKvvhEhA!qs-o$ELMsVQY_Xt1qCF{10>=w3qtCNsLZwt!B#hGt<Y
z18l$Zl_&9sQkwYjJw-U@ER?o<rC?=)m_a-_)g-4?j6fmD=*_Uk3`y=)diyb|G6;3v
zvA{m&IVWLBVlpXeaX4X@uCYbOsBg{kJOL~q8_wkIm*BE?9GC*6CngA+a2267O)_Qh
z)}8<pTYoCJBFFA|7z2VGahkhde~}Yd9N!!?k?o~U-l@yX^)CMAP7A&e0H1fXA=-`D
z=Lx0L&WDX8o~YgaaAgTfC5%}vGX{=XO%YZpG(0>!a|;V$Eh-mh$1QO5tmbAp#xRrH
z!@Aq^rOf--m^9HJbXXSV)<ZNh-@ildxUGJv6nW0zhEabpTg4r6mPozaimrF-57Ml5
z%<0Js772|cW2PdX)wtjEg9rlMZtv@@```YEe!9UnbwQnLbF!$SqW&>htUr|nQ!+u5
z%woTnAH^_MDOf<2uR+&V=tBdVq%bCWg+jvT>03OLVFR>AKxDvwTv_TG`W%KU<Ge_B
zWu(mYW>La1X?&-l-@HV~CZh(le7|0tn8r|%-w4#>DIs9tlZ)~U%PLR=`a28R2@jik
zfpyHvE^k0sQaC<7roun~;t3CV@_*D6VnEH_KQ!eEs?t5t*u=~>tVE)zrPJ+7Xj{o%
z=M7bV#T%!s=|~7*P?la<U1jYdty;f}i>r<hxn#$kFU+TZ`J)bvh1qhFIUbUfkfv<5
zEmWk4mO6$2TmkA?O4Z5XB#30<!@&mr5!qwhxzp`tb6oxl8fObO@4%}5?;W&ODu@DI
zWf8c_W`Y0u!E6Y?v}DoDHpz6+a7S|7tg3QV><oAsiY}T^sJTfS*JP!5!-#yj0O7l!
zwJ2n0NIVz?Xrw?HAp!*qr?T`<uQT>w6vVGQ7{J&WcW6@+2ch?+FokSdPpv`!+F$7q
zVv&ZL5c3YP(|cKj<Jj}RDc{5s6$HufP+Lcxbm*A9N9=t)Pv~D?mjJi-&3BHk+00fc
z9n%VSt?3_bkN<9oxcu$1-srz&v0Dk`ngN8uhUVta{f63A+jj8UrJr8l#&0D_3CXnT
zj)%X)OkLZO8zVoT35#A!6p(j;jm@nw+M2a4|DZUkwbG#nX~XgQAvTBSCBvv5b&!}>
zeh|urBd@4cN!Oy5*>R!s8j{7=xa!XmA{cU<in2JwpNttQfb0x3Z&TzNRK3~Yi&%f>
zLl!YQ!U<^mWD#CDbGvuVT`vCt`m5<oUcj=4l?`>WwD&%1?!Wom_l9OC8a-r^yt1-i
z%GCP`ex3oq8^mJ#`5AueZKzaX8k;B|gz1qYN=~T-*X26Iw06;_@Mhk<n-b@b|0`|L
ztj0?C5>%JhP#CR9QQGq>ueR1U$g8^B=bZcXk&^SzO`b5FUTYxGW%@i<=H^?R&2Pwd
z9ak`ulap)Ack&c+83VhciNN-w4-UU)gIgxbZDrNoRC_TCz+wkgYL7Pqqs3raOVz#l
zn(e+{&0bfrKP&Ra#zs-40f1riDQa&De!y1hbVlZMTDC8oYHc<J*!kXM0`%cP9R>-8
z)Pkm_6rcbFSN!qgJJ9ATtEecbcKb?tETOEc8<Ug(bi<TXRYQdGV%{}a><|>}=lv0M
zNNx7^6LMw2%e?ukP~>9uwWW~NN;!24oz5197Qvl(&l=Ea)f)}jM|zTLH9GT?%XI>}
zI!|!e?7^(G)s0*N?{=pbY1I{zqYfA6QK^&4)BJEH4rO$`gHZr6dg?ftmPsmFF;Yl6
zrK35Va1q+F{7YdYdXby{m?e!`gV9jSN_S27I}Pk<GvjB{s`qiMerRsD<Ay9a@YCxw
zhnp*4(EN<fY<={dwC45$NK_xF;~?j_t)Dh*CI`!Tn{p3f8rzKOYcW0e6V2hW9`2tu
zKfsas?=UJjt%;@35LzLM3}=@o$>F_DyLbV)8|WN}N8|K{VE+tnOxgT+f9L>OPSJ$?
z8Pt@y5IqeR^XAspqSB*4TYD04+dq!?tDytBg;&m1eqPq_6D!JvVEvz{1M-j`9PVT9
z%tHZ!H<{Yz7S<xvu#WH~aDo2KKm8Kzd=GdRKzDswGNn*7sr|>n#0aI;t~7;I%7=d`
z1>UY)FnFnUoel@s*XE>~f=8j04l$rWhIP1BI-c21W5o&=Kk&+Q^*jlWo{G_Hh~fS`
zYY2hB?b^Ol)j!Qac@|%Ofq?a<kh<5j%GTt?H99<XzFVDx@^1TZ&QBwkc@qHSmOsYA
zJpW`;OQb-z4@bT%<2`W!F{wX!{ZoH>jt2E7kjC*P@ZbN1BoQqxo!E0UuSy5nz4d0(
zQ&Xg!AT6tWsz|4+q5n4vunjDgOjJ(-gM(l~#Mv?8hF?*h4=;S<7zX84Dand4(7)A(
z_67vYgV(K3D1Qw0g^%4+f9*-euOx<JTH!YuY!OA_xm1G-s5Nmd$4C&R)TjUDc;O*1
zW+MH2hEie8t6WL4&}Di(YFe-#nEyQ`1h&&uu&BpfHh{l_MR7l9JDS=I=+3tS{qZc8
z%7TGO&h?3b2y|RdR|4RXO@+rHs#|U&8Kk<UGdV>Q8l|+X^sK+i^ZfflCN+zEDahrV
zf23r8kq&vZ=2iRB!hG8L!tj|g0hH%Io_Ba=><v4y%e%~Q5N=gAv4(X)%F4!*TEl>d
zVGS)U{d|80+4GJrAUwT7c;2Ny%kf*1a#1=zc<UzjgD+0--5q9a5d<<N9v#NkhMA&i
z?*q=^S{HNs`=H@3Zvll;2ngYjKXlGL+4gt!35B~>J}p%iXGx<VE!h0fb4{7(W&}~v
zB9W7U<@_NB&B5U2y-O&>=C1bXzNB1WdTNl20-=yp;L(#pTVNE~jyEkdiRC10&X>{$
zOF{6>xp>0`8<>_ZsH>Y@6vAN*Er=5Ua0n%)M*bqd!FmAi+Nt)7E*XO1&-UmqQVGY`
ztKp!T0V&FFig_4wN0q2(H7F<dVQ0YLm6r%=pzDqd&C03{!-1oHYxvMUg}}{|5l%~g
zY&0%ku;AN~g7?QCnbxiCsPva9-rZpR{Rz!A&l5j%C_o!Hod`$0LnPpWv)_$qh>={4
zN=bQISc)?}UlZ>F42_kvwPjpf*zLOc6M^<f=x2!owEQwZzXN#{NbOZd6xv@^Ued<0
zbwbVi7b47&P?X$(<Ax!1FSofqL3@`%cUP@<XAoy0q|IkJB0x0TT=PQ|5sxh%V5b5(
z%GYJvL>CTShG2bwNOF6qq#yS6*)&XXpDc+T>_4(Uz)VRv2&6htccT5VA`mD95Uf}%
zEz6JA+f#GBG<~lVi6EH(wbu7_=Y)Gdt)TvmmOLI!x=0>Eub%s?U72<rxZq;2^}K%p
zWn*HE_nu70%Q>(4SF=^g1H5=N)#kV-k!LRJjB*&sDmLYit>RfZ^5v8Q$5#qk#}zDU
zHYhY{A*!s-LWdsHGNkvN*Bh{w<1r#25(D6r9C4h`Qvf#z+}gk-a892b=lXzxU0Zs&
zH|7rD^7_{86Cd&gTUk$UI*r+gZ+zyg(@oco(0j`O=!Z^6?p1)59w|j7MB@8-qdO4H
z&&~A(!@tx!A31)$_N#Va+%)q5P87zRJmPn;1XA;+E9KbIlJ9mz$zPzO#Uip1iAno8
z68ChyN2kS=NG_8y1N;i0OW*-4tFm~OClF1$E1{{eZCNMCz|pARro`(9i=gwIt#Pnc
z%AprKCQj=X0J>EGr0aGJ=Z0bK+r6mdptfBnaE=uK9rJLupzLHZDjgUrMAg~IU=A7$
zBU~8_A%*^xRMVQaH(O~kaQW}{Y(C7*Y|6N%^8rTxJd2xv%fV2*vZ7-wbJB<Q+C)L%
z3hF@59vu7qz;CX;I8+Bi>H5Z(xZ)-4>uS@4R_Upywff26`Qt&Ptzrc)M-&|{ToR8F
zS)ejnVj`-9-#!&X+qHBbQ+|E@e3=6nF3hL?1oZaSm&eQgh{@yZWzj$f$ZH;f483)K
z4a%hcsnc-A3eC~IsElr!g_#Sh=IW4(TYw9Y!uO-DpYAhRVN%YonxsPrV)}O96Pb{Z
z-|ebQ%_B`-UI;<Wr}(SShXX%f*I_T2tmvDR1!-MKjo)6ya$#wF%{7+j+%E%Bb)hNu
zCpM5F3fvKlGw$QA@Hbj$5t;^O&27fI(7)!!d9})1@l&*d(mdeg>(@*@V0E*3WD(5f
zbZQ8d7^TGa5sx$MQbo$6qv;_dO_-t#brw)?)F>jTFH%49*;_?0)ez85din<NIFC@?
z2|86K;Zm(J0={AI{@IC&q2M4y{H_m`<QzEk4MO!@Pzs0^=}hkgZ-vNq3P&ICBx&=1
z<N5W6@P?vs`_H}3q)?!pU0w54MurSac+~~Lr^yz&c3kht6BLMC$c)_9F;S^1QI>;c
zrAL8+eEg5ecpC2#xS_NptPV5^Gzm`70>lhemRK=u!rIqWBaA5K0ECEd&@2)Y;?Z(I
zE<=Q^>n*@2@PZBId4DREE8qj<<roOifVfou{(J8F9H2xRFaI$CG#MawG(fT63Xagv
zYShMM`Y$}{Kd@<uR^;Cde}aX3j!Lq{{|`;)7+h!DcH!7|)7Wlot1%lVjn$~JjmEZZ
z+qP}9jct5)o_D@qo#{-{&ffdJ_I0jvtz!XZAHZpr_dSXLkKL6S1b#5;D3a&=x4*qb
z1~+t)$Bbs1P*XfkK$Al?0wH@kAQ_}It6#$vRS*hcNQB}u<a~hut*LUxft15<xbgAj
zeSR4NRtYWOmpozh6<a`!^@(S%q`qRjgN*z4uR!mm2~c8PuH6Y&!G@0?S>P$LBZwA<
zB90o5pgFW3OhujnK{EwB*`vV2<pS&?ip$FOXmy{YpkAArnoMP`yp4=TyFw+_Vpwbb
z2IIuk%>gWrAWDfZ(ft&4(5#vIFme%zm`M*GJR~Bh1;V?W_oLJYRYg$Lk}}~4>G-um
zv9v$9qy@;SpTHid!|=Bzod%htxXE>M6sY{)g#wr)vC;arzEW?e#li(tc_)_670RRt
z120fj`!(EWdn=#&y%FHwVP#_@E-<?NJENgwaDTc2IP8F@)!fq3XfmDa(<GnHeIN=K
z2!>y9OQB(d><7);ru7dnNOe&fpk3bm!Ry``e3hi8fIx{|UDX82=(K)(hkX8Vkn(iC
zHoy48e&Zl1&qL`PIMNPk$Eu8g4^+AK!e~GUIt#gUD%BJQH<sp7Icd^2Mz?W>*Moy3
z&?FviKp}bRc{*aTKcoX}e@nHvhkNZ*A49~JuA3?=lr1yXf!hXQ`uy>->D{eve<rEI
zetNozeEc$*^hyH%a~?KuxyZwk?C`gTY|xkMF7Lb2QW3WO*Q3uJ6<O^Vx0dGnp_a|>
zZAwb+u{O(1az-=gQc$Q2YL*}*Lg6g04s)~AF^0BN;cUN(a}lCst`q!w9zESASqN1e
zIo_2Ot}MH2>hbhDq^9==dXK%YTbD=Yi>Rf%b3c3?y0HBQ51}eM@4t_yunbWO(?iKt
z(u^)9Nx2+UQ}$U=$`@aMKp4T!Sh8*x_-<foG%iWdpTB-TDW@QY`ubdM1|o_>Z)Rn4
z;9dEI&+)zt+8rni0F<0%GeU<jQNT7`k91Tkv`4OnJbjg@i9^9C@I6p836nu(w=ts4
z5%8Il^#NPUUPR^j0#LOZ0`h$8w3w^*(-v}IQVxmB2rUC2KY&2H^KLPV1(-9APfjvq
zi|txiRd#~w_21kgdG43T-+u$H&hb>Xu+K{ZusNG9k|+G>^<?}B)Y(p8w=EZ7M~lBf
zzl`ZLOB*NeE)xkXvo{I2Lc$ubf|-}v1J{JG$-#jD{meOiw){q2xuZ$+-M*afrQ3JH
zZ-O52Y`q~{cNx;SpR*BI&^^d>MFIAvrZK!bUueQY-7YUYkINoGjS_Otl4$$MH--=U
zHGJ!@i*^^D5B8HS)`aIS!(NpA6rS0-jZSBp&lj8^74&y`gViQyZi!Xu1KY%XuXXsw
z4Qn(HT@!<aaf|1VMG^n|GRV(ospk*1BSR}S3P$k*;>URT(l`fx(<Sxc6mn=ky$&!<
z+V8MzOZI)>s&0WiJ6ltY?^tCsUbhw}*LHqMOUj8&v+rjekntMT@=pmJB&yZ@HmRF_
zA_VFIrX+1QYQP#&0Gv9&Ayr8;1qAw_1V#M8zYT`tW{}e~ECM&8XoH&Joj3|12H`wZ
zDWzGcUAbN=c}2=_Jibt1xQi=;OV?2?S%=p=BO6eH@B&fepJXLA4=`LfP!UuLt&s?|
zEduxX?W0hhu{xR^^HZM!gN%X#j)Xh(>kPXPR3%m@CRF~NC1uiZSR6zKO_j&428~$F
z4-z>vgQ`ck-fFtC#MuRO<<LoknFR&OP&q^_DO1}o65P`G>iW~-dBRDT7{Z({3_|QJ
zwY@7GSkmO71+mhVxTdsJ`3`YXUGo3ps)S)e3(!LDRaFTG_UVknT<*IXj6s{gn?(zI
zC9F0MS;xhx{!%%i1%N8Gh7>iz3gIvU7FJ-w+5!ERKs|{BnAs5W=@RhP4ky0YfL#u4
z37mHe?+Ho-cK=FJpW!t2SFR7VuKm^FCJ$#|#a^MjhQB7-Y}#2StI3_}*vQhGuq`BE
zN^D^TMgp2smkMM4lA$@;az`YYj%ZTa%e5Qa_n9B_j3DYU6>ujLihlSdlCXfOBxOWY
z`L)>QZKtGO2bbG{>hA82OcLmqY46ULjKcx`BI(I*qRwPu6vXHG*{zrH^U&+B0hBKt
zE}#V798g!^p6?{S#7rIi`}>z_t=`k;L$KrrS?H*bB2MeW=t2fe^M33Y&3vXGATAB=
z4qTuQ)hH`aKuP-{U}V@*>KCU=N%}_`CVta_>uylf?F+=#l~4)|873|up%71jutE$J
z_5~h+KR|K152q7s{q1sDS=kud=zETS>*El3Cn~MH8E`=6$WMV~WZ;`suzZ!+Tj-LE
z^-+K+2ftK)fnHY^c9%vVA02JGeS+`Bt7$Z|i1|A%B?1L1EU{Q+0sOg0Mxb8zQ-g*%
zRtDp+Mm4+F2XG`U4W^Q{y%84vG=P(NYmhj}0(6s^%}u%e7dE1=AM;B~EFO=Sr$70D
zl%~#U-fWE!Lz3a;c8|xga#k-+AK<Suv$p=P`27PMX>gQA_b<<_*Ry1z<$vDUE<Rzf
zB@Qrk<;s+p!=3s%ub0K9&H?_t`|a+IAyU+UZofp=dU~D7^@K5VOY_6_8}DwwkF&EH
zB#I7Ic_w3@!TdWxL`u>R2BVi!)Nrr-Fo4o*VDNe!e(I*X57`!|o=2y8`gT`HOUnWR
zPa{}9YW)gsX+V2qnic}lT5bv(4;I0Z<zJO`W99X9qHDh$bw2>&4D?tZrMn~Ll9W80
zU8Ijy2z=OBcRVf9S1Pyg_WXXhb>}=;+p_y*K<siE^QX9?!g#hPjz(@M4z5N5^R?<;
zzk1#ELFe_wb?XK~`8(sf?KVjF-O-Y^H~z~-m%w-YH!dfF?b9elG|WH@N10;b@H{>e
zpTR_v<L7%Ji3FV~#II}A6ih58orv3lJj9+831^f)p4ZjaS@z)%sHjL+EnSC~`S4d&
z4bd@8OQ$h>yene_QDHVXqz?X`N;LlswXYEhRMA@1!{O-Eu4_&Bco<+w=J(`Q5AFpy
z@m$Qv9lV&+gyvNV3QYQVMFZL5dcgB?zwpHK?uAu1{JWBu>gDN1>eITKsi!Tse<aYB
zz?66Vg|BpT!CVOrpR)Lem}O33Yd5yMKm;(4q5l>Xvo{0>q;i9S+JaBRwpj(&-)w(v
zbGVaHz2;ODLXm6PM_kj{kF!^-b?3ASct4zJ3ABd?_Bg`w(?CI^pl1GZ2pf3!bJwlC
z6H~7w>2Dp$0D0>3mCAa)?J?8kxf1An@cg82Ftps-u>*O>o;aj~>5)IC4<azlyQ74k
zMj{G=N6Dh~J5dKKK}gV#cfupv`xR6Bk%FW-z}MdSg5Qk_{4y-Nc%5&0aFY}v`p+F*
z#-s7n5(SUvUEWa^l^vPDF61*iiUve@peJb#o@v`6P}4WmyQh?Bk!3E4i$pdm9`A8C
zvQ+^ezYqo*HvX9!B&ei8jj!~guc*|(B`PV(;#zwkXh?{8>eB8!i7!bpRbwVak+6!_
zdTgxpXp8V+5Mb&9KOifOpT$;QZ``#3eVtpST+|$l#7`st6^MdoySxK3Zeb0&{`7FR
zmh?F~e!@HeQ^}?V&X+#%j7LrOL_tYmyy9f6FofLb<{zTNKH3bq%R88;Pd?r5m|v&r
ztMI~2N7`@e8p!^RkCN4Fdh0xcM%jk^`ppJur!T5)7s?-J2CNz&iO&Fn5~uv#{3>-w
zm}m}3r1anX{d-JoMJ4X8t@`4QTV@OD8VW#%&cHwWaX?zm9zfG=HuTFOQE`!408KU<
z16V-7t_*BePb?KF^{#xAs$wTr8FF+oVe2bzSb3P$B2vYM(}<{y%6Ib3f=z6Sm%~8>
z^Pd6GM$m1Bs-Zw4J_o&z&sp9OcoX<FEX}s2#fdhuo&IhUpFV_m*7ckJ^2QvpHCe+c
z1)(O7dfQQKI%CX@$B2<)vg?N_ErWAtw8_s|S<}nK$I-8#ou+}%EE)}!eamk%TnnYI
z8PbmJ!7(-Ntf@m!kXP^}YGe|Esrg2fso?_TfDm|_X_&Zk;`>FP@S^zNaXzoO6e=ni
zUfd!<2*e>9FIv~3mD-x{BeZ(6ZdwP+y*%WjAT~H+<0(AYK8J~)chH?r9mh8I^2sn#
z|300|3>hNb?zjZU@Noy(J6Q1dXo+%gTtKNX+menq_@d@#yEt7gZ}W$b8u#c??E|%>
zHsDb2HVa1AHJs8Uu>mcOgP_K2@xZh_{ly_kD!GfcpqQ!d&PXW6P)xIA#|#!HNMr|t
zk1~u0N}N+FtT6(t6aOugshJ4ih$*UmC*`a3c2L(JJv9a)Q*GYYEtUC<&Ge<V*IC9h
z3&KdJ;{^O3cecoo;$Yy?8-)vGV<N(P0;^Uf6<z67fCMGaa3RNvw9F<W&md9>mV7-@
z3K?EofXl(_RtCMq-2-7gJ40$<m!Orf^g}ap`a_sc^14<3rfFf<k`7*<3v;k$2pyU#
z1DT<DTdCUx#K^Iv3eXIPA}3Crdr-du9zEcAq}Q5;>FEuG=eo=Wp#`r_B#T7Nxdy3w
zMfo-S8Q@#il<*-F$zbFxY{B49*-Q9X+W1Y+g`pzM_!jdrgA*?6a_PG@>%0%+wr}rY
zJ6fWBorMi()pL+|Ne&HLUUyU9Y%lv^gB?13#0s%Wxrt-_Y_gtYKS^3%wkPx&I*Hy5
zC_aCgb8JUix)!Cpmt)FE7oT1X*swkBguO1$?3<5w$IK&^Y=?d0IIl0KjL0B}A+S9z
zJU-i;>{bSnW;ZuCjRF5Eux+O}=kd*p<V5o&wb8!yhyV1P{K4mKfv?HCPVW5JxB>Vf
zojnEEI=zDtUx9=~WF*q(PsqzV1q5?!y7)KgdWZd+;R88sdgz?Ta%29b&R2xz%?z|*
zRjIeTFx73kCAnXuPX64xH|q8Q4JftbbD2l|Zui00b>MfAH3gX4|DB#1%~!h-v|b1S
zJyZ@Lxf{LLZ0my=YV61>knO8zwhM1Iqh#9fweE8D5#pLvBMaoU$ud^6ke6_?Ho(J7
z4t0kQCq}av2*@jazyfF}<~zPvO<f#dJ*)wg(praqN+y%(|M3KXR5$#w+XKTzryoRu
z=CD5fMb`&^$J||cfqWLfw)0to$3^=Mds$vEini_3K{^0;z~gqUtqZ$Z8v&sgq=fZ+
zzTG)%+M0y<U8Y)*+uE86I5K~aQWs6PIooss6tb&>Z{A!yne$5Ld$3xd%$BKV(lqah
z1Rt?fX{A&mTcF-De_9)=<z!X)<0np#NDIo+s33tkYQl248B{1_mB3N`<m4VyeQe6Q
zBhV3(fzbDO)w=WY^=99$g<*>);irbV^9g;^^x~p83WTVlXlR})H-swt(Bj&eCcHk;
z)2dq6E`b{Sp3_ExG(+dI)%Qe;;-VLM?>72|bIjd`aMn!xtv}V#Tk&7#@Ym%37oos0
zgds_arWt@Gfx%<Gp!W-J_#B7>q$go)IEsYDBrW*8Z)Z+B)(BPt&VW6@E+82hFIVfQ
zd###-tRGlN;uzwn=uX)V5uAX~XDNj4JsO^Be>5&@=26llX{pJe(!@g#W=D*~R?9u(
z$AyuoR;mjzgu0y0d|#>d;`>A3Mui0pk{FxZFFZpb+MFXuZM#;f1Xn<^s~$&&G!FLn
zlq$9gt}POs9X3o_&%z?wH!k;Ktwz9L$5JM@4S{R+Srr5xi*6XL6cBjbow_O-fa#6)
zh2c&=i^)*jM7jmY!;%DS1M^^yrZga*0R6^UT}B5X>wtHFS@00(9vdm2Qi|GU6><TX
zF*DQCJ`IV0q0h0TDh_KyPC$!x*{W#+rY~CHbq8790`Ml77a~cMW)Q&<F9S7?b@m@X
z8|}L$3^B9bRkC5^4qzzE0gRtOM)(HEvjCL>0=&xI`9tfAi;2GK<`p%!ue26neD9BY
zD@W={#e_UTfa%9q8->KD#CiSS;($q)e7KWS2#jvsa0$MWxTdqE2r*ujy4C29%Ewdu
z7pnamOCx-};P`w#YBZ{c^@8*|W9cLlIAtq(MU1Q4Ir$ra*pfH`uRpw<BQmo{ZC>uc
zf>J_O&@7ryB4tYrt}9B!Ui1D!H0a}>aX6ldILbE!%meLD_gd|5=Q0fKm%<Q6_u1Q+
z#hcF<L^Zha+xI*dzx%aH@VW^($qyHMlkFm4FfjNO3M3VBNNOY#|1x1c-=0Y8=->mZ
zx$$Hse*-0wS+EkY`&CbQ&7}(CiB!k(pACx>CzdUo9)kPKRGL1GRCW?z64`21KY@f^
z^wg2(SJld-q^TnTf`Y^(@+A9U6BhQ@cxvUxv--6e&AD1VP;|=Eix-fiKmo%bL)(-7
zT0oo73#_Zc5WWQbkc=L~C?L5C#!C%3IX(W_#BJegn0p;t%CuyL2;vg=>-O>{?N?ZO
z8tq0QHdBbJ%jFUyvjt<QnS7l<i=c_utrRsQcOBF#RqXG7ZzBhGm5X^zIC5yFJ2`J{
zaiz}$p0_jQRXl(Z&lOnXw%*z&z^&hUbF9S2*9S3*ny1K?Y`DH|6wTmoTz$g_BPzpJ
zgc4<TjQkx5$?^I)6kb9&K#+G*1KO`16oJ<>u=EpL>N*KK)AYf%E!L{IHUqVnn!Z3(
zvL`Q!_ms{AGP~n=7MC#yB<;=fM&5cKIA);o1PFrO8qpdp<XA|xJc6aG7f>y*X+><F
z&O2%YP{?q+uI6-6Dyf8jDlO!Jz(UlLWT0tzIgL%3X*Gcs%pacopi5FxDlTz)k)`v6
zb2*!;YIV|8^U+8N^oPEpdwn$(_}DrlwH0lW5C4E*F&(GV73|ue{CG6FcTHky10plC
zm~SlSsZ>Xj!}v`bJ7hupWftiEFf`{<r_7CDHD_cw0;1{~SG<zWe<Oo_sG4gwmd`(9
zyMZ@N4X?r+DJUW(bi6<vNhwnOXk%kHS94l3bVf??paLQyqzEFwK1n)9K@Z4B#*Q7R
zKW>t^-Io7*X8ch<!K|?2-1X9T@wA`s{|45RFM_<2n6)u#bMZlM1xyjQ#ZWt5H<RPe
zjOucHR-tKOV|H6HlXIZ+AEp-*dE{JUEK^%Ap1Pt&n_6j&u>oX1O2xhs5fl3&NWj#%
zoaket>47@6t(`dzzPPPh)w!nG^9~sf?m4-S_lPGa<`nT|4&op$<BXfYLKOc9ON{L&
zk)okjdDt8ewn-7#{6@lJv!Q93Af<NT;vmJrLdwj%Xydx|;F{TMdo+pfv6Iy)!S*8|
zq{IE;y5Y?E@}O#cQukwGXt~Q<04O`(iy0y~sGsBb)n*WN1m5~8IdVh9i>^zvrb8m)
zD;lPtSHhZkx4(3S4>?}@Vi;&|Mq#gP_(P*lQj#oMG|q93t`qTO3AKHQdyVNza7m-d
z2~qH&1r70TkUo90kNVM??=AB^CSJ9WqU<#I5T`HyH_t$D9hB*)WH%9^SsF-&3}h5~
z--X%l;tkH;N4?#jD|`P!`AZh|i00Xq4wlGK(gcg?WLZqeV)pSUaAYKamK9!El4l?C
zo$sVdiAl6$vE&!9gMy0NYl5vXY($H-`g`KIsHu2#I&h$BiH|A@`j*J(7wP&9aw^(E
zC@5;%aTGpS;)cVR)&_L4CrJlzI#3W0<$EntZGHeA>Wc7V&pV$^`u%+_YbOv82T~0p
zAxlC;`qpDjiIBef;uNRXMh(6oD@-EWiis7cg*Z>FM~kD$qEeAI7uVWJlEhA*>A_I%
zqTnovnz~#`S00SLoFDmaINM%Cp1rO>f3%;Ru7VtIxSx{}S+2DrmW%k%AL)LQWt#1m
ztobjG2xsBCa`l!fbpUO~la=Z6)KnnNh*ik#{P1SE!2;X)q9sY7;|Ux@hp5UN_k7d-
zo1)&FYZ|BVo;jHzQ{c`y-@{Ia+CPdD&Zqha%S|VK=M6_}=Z&^M$nQ&9+~V3&Aw15f
z^I^txT20mzQe-I0?N%&+PoBg1wAnvRz~M$l!G7w^{c@zvaZMG_On;?L>%q-M5@#|R
znE12F_USQZo8t;6zDZEmaH@VIv`t)Ue|LO6w0Wn<A1)GjiPC<lJz*TI<|H_<hs`S-
zEJ|K+jEqNDu=-A5C|bVLJ!?S8r-A$DD>VPr)_drc3oIdc+xe;_-TaG8dTsaOFvwR@
zv|Oky|L9{;O-%4Hf;lQfA_K{KFE6ZvA{U6NS2EBVq%1H+y4b_ZJFv;jdD%$*#~$Qb
z;2QiTD<|=dm;{AV2t(lhw8b6S`K(>K40OxG-)mEpMvYjqSUb?BUWX<WECQ=1M?95!
zAd=o`4f_K&gew`F1gnQ+-TQ(-7B^p#d7VslHZ5NWFRot4OK-|kLdJ5%gMflEFfx9g
zS!uFbC!I}{Fy<9RgAT@$-=;3}rFI|C(e;6y1D)l{=wt@Z7wkkUEq43slp^Ixwebl=
z&V%^4;E`&>2L?WHNS6~Pbs4z+{g7~tVEFL_%>8`{Em20?2=2V6yqI2~UN}d|Noy@r
zP_2Ltmd;mA+nFuvWncHE0{VQf$A)kGMc@)U-Ene7UAg+m7uw~q$A)xf@Qb(V|9b)K
zH&;v*t3^Y7ZzlOefE;Vo%XvrGC#1}WUaL`Q>1RZJS4WbNKIh&1T{`bKiCYLl;7i1l
z^N90bejHVJp0?Ii_#DS&g^j>4N7s|=`|XtVB8|GmN|OyC2(s79369~i#~>8Lssj`n
zHcOcL#l`$=jT(2sKa?y&SXdaNl%(vdE5<=1D0}9RhY#J4l$Y}sXn4zw)`YkvB-aJ!
zF2AjpjxF%+!hi_2htqcd7U67-T*ja|Ile#)3vK-z4bjpDGD47|DHpOZfdRQgND9yv
zDm_Dih?AK;h2J1&EU~PK6h2YYAmFz7>X)=#kF?tLj$heD`CqX=5FJ14k2^?y>zzt+
zp(dh{G1Xm9a1}4J`r7c{!e}wHCr8&ak_}}@3{LAw=Ds(`qH~^oIl4@l@{DP%bRChd
zJYUQZ*A!zaG6ZER9E)3CG+m|w@7pfi&5hwUH3_Az>4j*HN1P2O4(>6m7O!e*jMp4L
z2;XLGwjGIfMTprMflL-&(%$sY^yyE%yLs|AJ}IyFIKFvuw6Du@hUUBD=z2GO8@#>7
zj4x61@Gh>*dE@;+@`)`kx_AAxGDodMWyo?D?JvDeH+qdq<aVe$p{)#p$;N0yBAO>A
zMLBKnw5x{@dnB5YVvHW_qMb%OVkJ(3YKBt;NoPoJWQGo&%jE<A%?&B*d7hFROm(es
z)9VGF_x_P&5p%B8pT+8eqpKx}Ws(yS_3`ridn}NN`eJqi*8A;!w(8nm;K4rK8)$OT
z+5xAF;3aH+>bkCU?hwJGD;&(BHg`|{dyU?HdayO1*45}A#?T4jV<E>A(eiYc?s>nm
z{<{w`{$Stt$DQ9Ee45>(2GI_Mh%%G1ZxhsS6k>Ip1VmEt_;A}H_dm$s-{MrKm<M-b
z;@38Q_=v)_L7aZ?<qj=RCqv35i~g;UrVkI2Rl#d7_B_x1=h@@A4SEOVEY_j5^z6hq
zOa6g}*5l~+nimbxjPYQjS)ec@S>8hn+i@1<t^`KLKqyL0h0%@_S~6cI*55cyJ|y9@
zuQW+>nv(`7#}w23bSdDUNf87qUuj_?RLk8+o+f2PXpdL#S2n_zze7}GwPkcp-5J0T
z)mG~fZy!k_ebD}zYC+}O%SA>`xBWeDQvPpA|4jWHwOT||741<qu|j@;i+o+RPfBbd
zPCPj=#PL<KiN2744elW{I2NHN#&PP=PY)&lDFz(~^YNCb<??dJp;#C|3opaBm~V(4
zOyR?_?s}B*x<2(zr`~6<aTzD@j_z^c?1>B_dmApR#MPMq1;&Do<0}-1e9TtoI%jg;
zX~=VuzE_#?>t|PcXa{5B*D7yPqrpi;o94`|$A~SWg8p-So)*&woeT%1;;)CLpi5on
zcbs`>k~W<?;$A0BUef65ZB8fFMysB*6Y&PS1xxJ^ltCEi&emI#^iAu+;V~QlE|&L1
zV+v@QmacM@Y>LQOyU3@4J8nH+wmoayx*h!bdy$DC;;ECPBgMJnMCTEunQ~09QE0cd
z5-78VZ_1Fj<P|gd_2s7jLe>z1G$<bYHpkkbIAd5y(i2e+75Fb{O|1d2A$JL;tDH1}
zVZfCzycfm03i*W0d(SxnO`|3#gZs!qGX+DW829B{+kL_MRYKJ<Q(;is#g+&jhq@$L
zQLF)|F~@w5A^i7_X@z7oQ@V*+oSxlL{=5d77}Z&DGqyt6Obqg1)Ne6LQXbcOfdU`X
z6Q}DHQwn2LSDS{?#_vy!zwqu-f>V05kfaj^S%8BX$?I;(!9X>UTz$x_+^Ut@Zks{w
zB)5t;7y~>{JdWeR^Nk<OH|$$$@)9skx4&#$47yKwKgE6Y0yZ^?<m+en;v{se)alEn
ztXD07#JKc!m}cX~5|JfgM}Wlik4ZRjKo->yanbk-*UVnmQK$QDb81ja3)g2Lb+f~r
zDNv~TlQ2kk?j1iwu=1S`S6?<zed(<enJtzmKoj~vF<2BzED*x1{b<SiZRrWHsY`^%
zJhUJ2X#=!1bouH%kU8$Kd~L^Nxx?5X*uY@a=Ht?%8*gtH<Vl>)_q!M1h_^vU9n1?d
zU=*|*v9NK0--IY<d6l<mj(Gs!eDjX`oi=<B-I+H0GU21yp&0YhVe7ZYODs}}v+=z3
z7sQ+1H}ozSt|wYsql0r(L{j^hTQ@YDz>g9e_q-wVk1sv&3rjeJjN~Sh=J(ezc!U11
zFl%38ty-9dV*cli6h-;d!LUMqo)z#TS)Fd)IEH_%cD*tA`PRBrvrKllnAPigq>IoE
zd~>dJ9GxhvPvSeD>RDSO9;MqOee1aAE~i$<S$OX5hcP2rUL9xti>67tjnc^i1d<h8
zT0P=ah4X+Iv{{Wyz;8VP6$FoiENkP}Wy9bqThV(Ux9@%X;@vO*J|4xShC?PMp>eC4
z|CK}DH0S;C!{eL4lcB)4Rn`aHh4;scd1_%CrQ>9c=>yWH$J^H*gNrzel{#Yp9W^;C
z>&S27$nJ2(ln|pJ9)-fP4jr^-`&5+E(ybYNBvx+Yk*kX3ty30(zHY_*3%6?gPQ{+7
zJcfUKBvT9&ID2WiRTMf>WAI}-tD~IZsUyJ&LOPZTBx-xeXyaejZ-`^`ac<>k`M*kw
zDVbrtxzRw0_e(*iu?z?d3QjZ`nI+M1;)+Wpq}aqY<(K0-<hlmMu&sK7ms<Ri8FCif
zY<ZS`&&hIJ!gpQ5)6wzvKi_Hgue}2yQp>%lkH?3nv#oclbc^4M_S@aSN>$AZ*P{J7
zsA&MTxXSV+h3#2G(Z%cW&(`GUdbGRKa|?icUKl@|SN(q2(A@*9PkS{(1cr<_m%tmI
zba%1{?(p)OFEU6L)UT~aL}Adjl`)xFkBKYWNnj5%a4eadAB@g2L7$hvfJhBE2Nk6u
z3rTjaL2?b^HMacuF}!7FBJDgZTTS`N*Iv3lXazsW?l@r*%DA&)rdfk&YXXfw-rq1(
zMySN^d4CH8E?p-?lL*!Cy6~1Vfk6hj>4(g6-R*#u?(~oIFA{*a@2;LC)6%5IP#_QH
z)lhh8+xEjf@A<l5;G1HDz^`!7NFy|4H8Cq%k(lu=8d`R+3(9T}eKxc2rSAc_aJJ9W
znMvH3^Q`j^OJ+j~c$hTC%zh|V4+dlegx9gDSL(-=8sV7~0(lQ@l{**62wVY1K;;@A
z4y?a=PRUm|RGEKfgPB0$C6%pFZ^}w{=N$UHZ;1^+7jnNnwpO-W_QL0wb5ErkA#^>m
z2=ryWtz_ALygm3oAe57bi(?pYK727%4lVey;6{+X_x<C|dpnQtjS8qy0bylV<NQ9Z
zIA^8RXnX(9)IgYIP_V8!*k!7kPSQ~pl+%#&<#x{XByF=FyYmv#>3m%hCqli;1kq7v
zZr}YP_hxEHoROHxVbtNVhVWUR|0+ZtAhzF!h#E<aV`8RdwuE47dj(4IM*J$x-Byps
zf7XoJ(vU_=_-mFLWR=ElKQ*}#45P;XU`JH<Pk6WO=RMr}CGrOfIy&-t-Jcv<*MWwm
z8uu3@(Etlw2R6u{eS0>q&IhINV{~vFru5c#vPNEebda{rE0DZG6NQAwR^p)UMEFKj
zgws_|j)(J=zGV)CvCsil6JEG-{BjXdoQSV*OHmk^*1@3VoQ%JAbXa=-q_#O->`&@A
zJDun_TtR+np5<CaOYXpw&UsX=iau*BHyt%(2qdx8H8pqbvtIB3It?;O?;n9z{muKv
zCdDN~?U;r^d!FQ4hjgd~HKgXRw98~3kEBhExZe!QcUqHlq&2uQod=B90W(YLd`xi7
zcE_`9%uC*)`#&vYFMpuTMj??LK<Gk+1WW!fwiR9BM^)5-jz7Z`Cdy$B4jFd*)|MFM
zd1{zplbbBMH^kd+^Nq<hqgxcwg#Y@IU&)N<MG=<?qUt)R%5|fRi3Fq4Pcxn^3?)@m
zf7{vn!TBj;>j__$&ZKei6ZV;IpCNM`7rD<Cvp_B(bH5TM5&Cz2vhLqH@Vy^lvmNK1
z78fELvFumM%F2J$i>;IJR+~+*<a0^r_4m)y4e~oAGJi314K;Z>y`*Hp{ZE&Hu#_y_
zG<3F@p$Ee5KfyR-zJaKeFTvD!Hx^6wZ-_R#dxRgHF9Zv)p1*L@6Tj_R_5cWyfkZ3p
z-Mr|2$IWo7-LatnJX?d0{<V%lXl*9Ftvz^AWd--C!<%mg)*F)HcKTCyHcBIh$ew>i
zJnqVLjhFZy^hb$JZB~2lY&=K@duN+X7`36rgsy?ahY?#H-Iph$Q|~B(4_p+G@qQOx
zxsw{>^{a*|6p=pTB>q3nA4MPiYF!>YQ2KAr4cgvsckTPFNuTtmbgr4kRSsFT_-UxV
z2d|g4Q(f)tTG}d*Y^2NvizU0~e?|7I=0*6i)(g)au|w{$2XaC$)pFOzqP{Z0b#-~r
zbM(-y5SZhYRAVgqb6b-d(v(<)yn$?F^~G}rKQd(sJzU(<6(lHfb6;?OBQyUKR_UOP
zVT>Ov6*V?FiPQAGCn!UbIvAfYrKR4az=3-kuRr0^gR#vNP3hXI<%9@{8PyCAmMaY3
z>CTI~o)$sADv|f_`{ZHT)t0;vI2hhWP-cx+z0ZETicM+{Abgu4oCrZCa;Z-2<(lMs
zvFvq`F{aN81W$KeHJ=X)v+Oz|03PNM0FX7@^^rq3dW#;$7PSAhpR3np+!G)V1GltZ
z;GKf(2K;MM<rbMQr=~Xmb~qF;D}48CmSG5x;^tssRzbWZBawYc-tiSLkuKTJp03ZH
z(kMrv@$6KUo=fA&@sC|GzptW(a>;@zPGZ|n#ZsNhNoSN0{2}=eqK}heoQCAtyT+wE
z)#$mJEC1_G9n`${7pfw)%J$5uG8st#&Vn}DNT#OLcT7^tRR;p#atQ^LM{T;l@5{dj
z0QR+N;I3K;(DFeD+)tSpTK3WIdaI*jtCz#6pj%dTcSGG4hQ5AR4l%85tdU2JatmXl
znBABBZAelR&49WOB?q2IQEMaQZ!KLzvC{8H`0s&cEHP}#6ws+K0uKlUKDd6iJjp0Q
z4sxYJ6qwU`2Pi5lt3dA2of(Vdn#pHV1Lr|6!l-|;rX)rEGubFVkvM{?am=qN6*cV<
z$0Y5^Ezh00Q{dJ`!Uv`8xS(RW*`AV1r49_>_HH#ngpWPmW56z9!j%6eMXvL-_2lOy
z@OlS6blzA=cc(g@v|LTt0D5}-8E_t6BD>-JNAUxeGD*&x)XDt^YP?Z@2v`0viSwyS
z)VI!QO>GSVT0jfHS6uG&bT?5=nXIw^;z2lC+bnx@LES-d@;fS3uJluGljOkVsDf42
z-#+s5pZ>)6ikG`x1Uu@ThU2YtVdzBn!1P!oB+$ZFy1XaM|9p7r4f_Vx{oObx-RW6;
z^K$UyX*Q{|zaZ(S)ERhDh8NV<!(!7;2XKYrDA_{d_Q8s)`Y!c|152!=DGPY2&bE-d
zlbkn3lDclmWhA1(0Akk{ErwPfvw*Ib`7QMCoo#aBbpAVn9gRw5=x_6sS?=Qo>|e6T
zpQeqRwIbHZ*CV2XF#`9(nmmzUMlrw^t2CxClgx#i=o+jt?}z=<=XK+_U3-TJUDIDK
zI<n%yD0~T*Ft}E1#`oO|Xp!Zp=nW%SN$<9=4WkAO7}INaFl3n9pSRjmzimAOcekcO
zX#76)+|7p^B!p7XHs@0lbtLoDps$<1MVhSVSihG`VFWDyy|Iosi<Mx9W=%w*t8ucX
z2P$rk8!*Aq?6oj|m?(zp^IBj2ti}mVAFk{$<-`N4uCC60(?>7P7tYPeCJ{$W_Bb^p
zoa;pOY|OCFK+r!?zRwq7_GjHvu3=l3Fe|XQCqaqh+zX%kM4qj5xO3>f9jnzCORA`7
z+pb_*uDj6WGf@XvU(c|{5H|md0LYYEP}^a;?Fm;d;SGt>-#n4hII^J8%mxk9Ytp?p
zx!T`VwnV~X=%+lORE$ebm{c^FdG~(2pUK+nO6xu|17G^tm6SiN)0(UfT&LU#3PM~2
zdQG@wWB;<Q;RquP1%e(SpnU!LG;0Bq2l+WSz%z8kyY1M*S%f(HkDZ2o)&0Qbw?7pB
zY^adqI>7k(ew}0}lM4%5Lrx4kVHVz;lWM-k>%=3hwjf{Ze#^<*Dmugr!8G69-TQH$
z^%4E6&&`kavB0?#81TC=qVz&T>9MVM9zOm?Jq~~8_cQIaV;5UE&`8bSn*k=DvDa?`
z4>zo4TlN5zc|LH2Z#!@3%l9K2srg<o!`7+RSfe7)751w+uLn_|I!5siwnh~smOoKq
z6=W?OnBAJ2SQKc1jws-PsOeGh-6-JvwtfY$?)B9{r5yrbMq(}<@_wLn;=Cr*O!?ii
zkbl)87UZM+W6-rqGqCK?$i3NbSdR>Ret=lq5)50JujzRMY2UpqkE0Xd(KpZ+8nXTj
z>H-Ea-gakH1Ka99@?}d_a170tO7Ghp7klM(lOzjnL$CAlZ@ixcczH80h6u<4uZ$lz
zQ6FF+DmA{&@6R8j0HJZY*_QNVDQlFkt!u<pi{pUA6h%4u*YC=0KdY$&N-s<JDBf$B
zY^J8EwyNrf9`M(_qOSgrn`*3dzUyJqPm=y;In1PUXWZN*=N(SgZ36KVbq$!L-4{<9
z6Ta8lTOyFcezLfN`ZSezhv6H-6i?)ePy-0w#?2@ew#XVUjKtr8dAaHuLM208c!eOS
zgI?>9{H-VU#jx_se^l%KUe))6Ftwi9%IGT{m&=7WWO!BC;N~I+v!!=g!P|J8-C->1
zryqC{J}RHLT3|DYTetqBPzq+{M-kpX+xh_a*n!vG9%q@_4`-PK*<P1Da$Gn(W-0&`
zCL|HlaP)mD!u=y%-hi(-+$k}L$@MSS?Ll&W>R%~#Ck!DvKLy`rHfb~pNglTv!WVez
zt4q?TV`s_r%R-Ku_lK^m3?Ap}(<1?Fn(wtS6lNsRMmSHXOcaMlM<qjulj<G_2-=_w
z$Q4P02zDeg$TZLmT2k$1D7**19K$a!4c${{gYeAml{wgMI4eDNRNA)(zV*8ckRRmB
z{82PxRGE1?>&Gz<=G|Mjcj;^4*iNUoHt>nDVbwR09b&9jkqnM8wdP0mX|h!GPEBNE
z`-N*9SUibI4EghUG>Nv(N&(g26fB7eN=Zul{*~6{-Z-hWrS1U%>R0kAYlKvx2I2$-
z6OIM-s9y#4&u+-6R-8D`N>K<x6|Ztx2S`O-e@Zx1KCifW<%%ocTDqEq1F>%{sww(h
zsJ|+}Ent~!d+IVLAgA0KGfi2q9>yn*cG3;0p?xdrbTG@JQt(qq4zdk?EU_Hz<-#-U
z_EsEO+@;vE_83+e;_)tZJZj2FAjFa~u&dLB^cR2G(=K@!LrrGj%Y;kThAm=&n3J>y
z86m>qC89=?jOXm{ot*_ML@_EyZx4J6!%(HPzbnq6l9H7wC>gyhW=`$*p@k#ZY>S!}
zEqDBNo`xqX@((Umya{ys#AkS?PdJeNX%KvCdy>bA$Y9xSQxySAa&_``cj^P}6Y-k`
zv@gXNwR)9RlQW4*sqGBKz&GmjSL(F6oLT&w%1$1$0#Of@OTAGDrrx&GDh=Tg`4_LD
zl}KKc?m&|_z@(b~Ox-ULtXN<sO@AukGFnUIA}<lri^H~pxzin6>o~=^hhR+J0+t46
zFioq9fjL3)**38qm-R|E$#vX%2+#v?45>oFEY45?#t#$=i<}q_s+oMAVv86Psv$kL
zwg+ljqxFR4evoTiG=>Cqv{^k|r-ss2s1fHLMuL#*-9vjVZ6=CjJ|_h#j=GSNpwfOX
zeHwEaJIY{K>UfAG<*1~~yYbXct*P!ZIjGQ@H~QD$m{}YpCD8cfFud~mO*Q#<h@t&G
z3%?`xuA2lLK=Lj2?AK?sw$rJAQS%;BNyhF_FlMQ^S8UXVhCQY*=+_VP8<>HyfE%4X
z;J0=~&^yK@2%=sAA`iTZ*6<8U{hElrPK$(KN107uzOBwl1uV7S->nWEsxWv?7HB`G
zOSf?rITV1xrM>iCcmOoy`r0v9!npi+FKf-G!V*VPSKh6f+@3$#u&L-UR^gnDfz({h
zVdIUi8?K2f7hERjnmIb;ahUbb{3dz0it_Skh(E{9UQBKv#5${I7;mDT?9I+B;1M8E
z%p~H2<5W<a{@UV@v8IR-p~g^W@ey%=pz0&cj(Z@+kE4RM@9|yiY`wJ+CdaVy3^y@$
zApxqUq@s9sC3=->Us;H)FpBGt`)1$!+sjJx)sJ$>i#wh(oAUjz<<il}p_o$U*9-4U
z=>O=MpL-DixT7;}1(f0psHoKd{_syp1>jpv0#=7sZ)0<An~J^Nj^&JLnECPa4~7=T
zm=IPN62u_XveF%^m$Y-MbC*@QAgKJFK&E^#2_f=7R^Y@Wl7fdw4R-_|X!&g0N|C-~
z!2~ifd<{qs9i9Cb|02;X0@n7Dv!g>a+~8Ij)UfXPnz6t8#Ru;NjdBl4lUA%Fk0Pah
zqv5|Cy4Wjh>cA%r>gBpW-x%vd=I{4l8)`A-Q*Qc-UO*izD@uYORgoIS6l*Yl{7;3E
zTl=R_jgpuc3}_kzYM<Hs<%j2S+`8isLwW;H2N2@@JdXW1B=6N4R|LlW)KgGE?x|~G
zP|n48D01;ZL|wvvQ(fNtsxfuIlJ$XS`(kCuYe>5F=2vzph`+ElhzZ&C3g0E8ZV4ml
zUt;6=vKEhpOlBHAYQSAIw-TW0E+ZZQ4vkc8!nmK`KsiH4$ApK4P#S7CUp?*OVOhW1
zfp7FAN-Wp(1-(-hIg{+!O+l4rg$+sFkg5V`pEUy$&8Zz_l_23FUO-ZvmoF=&Ic_av
zjbxnd$BMOMmPn=znwLIKVRHaaZPn+*sjjaNh(c7k2d60@inRZYB7y*G!B&hXJRMV5
zJpYAhH*W|xuV)tmLS4D!)S$eN>1y6DSQ;vh*`AdAb=kCq1pq46&J(jbil8i5%SA|)
zNsh-R@+;~wDHhB4)>Rrp`nhF(iQD+?(JQX6Fo9Z^<S$gMB#c0I@ud~43x`UyT9px0
z8zQOvdE*I}6pJ#kz^#A!-Z7*MY3}w(<Og9x`Di?tB#A7`<hur&@qE9sC7AbBj+T0p
zHNKc4P_CIQ2DFAt>W0UlfPH>LLlbO8Nzza;poWf5Uhj#=eJJ`n&k^kY@ngC~_I{v-
z+hg7F=`mP#Jdz9QBhGriEn-8>PC1`cKA94@6xwKtj;2Bg1wFFi$~x2CnB25EI{@n8
zj!MfacL50Ozt>tMNaBC@Bg_@3sjVgosvbKxx_S?hI_X`;U;B2VfzN=FRJ`#@gK?Yl
zMIqKyA${5%S5)jvYybKx8(joj$z4>|2Q>dp%djlRrYQa(LTH0%e9bA1%Tn9sg5+!v
z$|JCjwiP7TBX7!*=SK)8o`sSw0;@v{89_}$<%gBU$3cLisZ&S#@)S@KBznOlPPx<9
z;y8a-d_xm9V_-LnUh{|oXgWRFWN0u#FCc{R#>*qIIG}lwLnv&%Yy?p?;?>s0Xd@Ml
zvP@(l+n)X`tE#0ohGZ0o_Q5fEm-BABW_%k;ISGNF#UkB_H>}jCBK?Q=D(JnKk^vO#
z0Y*&5BTDEsmD$B`#%M?GDqjK>51@<uL0YV$=Qrm7eU=k%H=Htq*+B|HrFdbBU6c;g
z0!{BmB!LP-X-^1`!Cl>{Q`PEaRvg7JUG&@9(p;9T_og_rnr0?BjJ{NmP=GMqqw$GA
z#!R<1q_p6HVoY@+Cmju=Dqj3}R^ay(Fe?P56|guCii0z*v0y>IYC00p^h}dAJhwf*
z%tl6+2CcDy?<=;j5;%E~28LjJz35+*w^3m8gD3yEm-E+`x>@DAiN7=a@{=x%OEwI5
zP4Ca!W=D}NQ;$WV^!PyT2qYtrCn+ozg9Y0&kx}{KT9zV~2CJVLpx%*WnXpozDgcdB
z&W5mL4Z@3vs+>G7J-YTU_WEpHrwJc~B^Uxh;2K?EIO??@8BFtajYhF}`c)LKPNy*E
zqj-%q^_YhNEOj<fXORpSBJOPnAw=?3*p67-2vwxwM6o=xT{W^u7GkAA&)5#mv3^?N
zQjKHFFNSzg@K%~Et~>ED?~3c6?R@Y;$$Z}Td0dfOT`AYUaeA8W&0i5^%dGE{_RPaN
zAI@x1Jf^o@!-C~dl%yoUOzD6RBv7LEanR`9nn2-Ft;q$j&CcDoy*A$_AEsFm0UI<E
zlep9AvXBab%l*KaD!H22*K!qcL5)Op#jeCFmvXl*d8Wu9ENvsMQc;5my-XYLJs~(-
z&OItwU)dl&=MA2{6;KDIND!njmdG~e&4rQEhU@Q4rKTFny&}jL2W^!u{7{IRA!NrW
zF7KFX*Gb+_Nz1~tbzirN2{EbGC@HjQ)XmAJDVl}D$0sU4(E?GSl%`n;Btg5>S-PCE
zC5IG`?<*1*E%=I|l=N|piBt*$P%+e%m?bR11O1rDM<{!j>lRgy#Z+Iu90mpVeRB&k
zpq3Gem*~j8*0L9yv6_NfB6khiEq++CEIuqb;dVD22?y-Q=xc8dPY)DFIzY}V0tcbK
z6*w^Qmc|nbREG!~*Z2NrGe<tJ7k`h*F%Rg`>z?gJvlvCYa<eU06Boav3Bx>D&M3Qo
z{ll#gOuU$>-?<ddM3xmiV<<MYq|KH6v(%Ozz88n=Z+1(yJ?kD5mWVM#90Hcm@vwpv
znb@IrpkDmWUMN3JE1WH3AFI=>KS)KPB|7h}DEJczQ1LXScf{A^Oo733u_~sB5k9mR
z`EQolQY-G6ooW3U7EHY;WL_{oghrZ54jdv+9R2CujEjo$|GfZIXno!f58hLgZBFJM
z3{YS(&h0N5T-~}xk|Lh$v5JPyK0HpALtIJTYql>-)>E)NQ}-FIr}tGGlb;1wpMAP`
zUX&2R)*PO27?s&{Gz8wSw(l>xZwCQ^xTGp7j-6U_FpVZhI%_qvQrK8}bi}N|7Hj(;
zXip8!@3;bQq+P*Ztp$B*bdB^DqYd^+37=r#;o-|Q>Y#jv2))w*^-eDKMxHr$S^KR}
z!ocL!I9$|{ZR!vn=&j<G@O5L&#Y3pwhi^Fpv$gt-g*XWIY*I>x32Je<SEE{<nDT|D
z5*`lv!m)fBD$FqdOB8ibA8go<fN1pUK~BUU1Jie1g9m4_u6hK>`6w!@>@UAp0iyjf
zX^ZGm)X>IrWvN7ksi{V6DwHj5`@O87-OR8aqo7Y~Nkarb!oD7^8UV&B+zyGSr>Ffj
z*k@2(l8~L23o{TuamwmT7HCzXyt0;K{9eybQ$ZIrV{FCcbs0CcdKW%Iz>w>$<vGJX
zvS4DINV(qzfjboF>*k@?T3s($C#ycRE$kj|tw9T)uoZdZ_r*AuRpXnoQ^PWddhGDG
zj~$-h#Ej+WBX=~AloTZFz7auG!to)Pz7wz~R1OImenUgJl*V3OFT!0aA?u=P=@ejv
zXgYIt0eU5<8eoat(QgJLkX1;e(*dY~^wnVerE1UBzaKdVza+VmbV^e->=v8TpkjsS
z%&gxsFG-Ps9q%T?Q0p1jI!S0wCuUgrPLr2Gq%g`Bp*S}FnUEyc5pah(IwM5xj|^1F
z9VMB2S&_QMmEqvMlaSJI&gW%#arqIgaR=vFWEHe&fux^#;W-usLLNZ!(B~MLh8uqS
zHq0ajqK(LkvXC!k;fuBi_h+8y^hikFo%c?|A5c|Ov!|Dsq!GVw*B<Fks&#oG=Pfff
zc_Dt4<4+k_7Bok-xKi|IJ+kD9a=&PYR#xw<y^#Z9Pq#yx>UtLICvnas)(dywryEG+
z!o2`ZpolHa{s}o_=Z?`9tpq}=Zt;Z;(OLmDf)Pn`z<yA}bT$}jto~TLY-!C1!b}*b
zoJu$tWA@p6i7YYFHV801)0ADDCksuk7B>zN-lHBx)8%j>QPHr~`|tu;EYC2NvD~O^
zqsj7*R6$@eBgrqZ`1$;bt!PLk$tkQRYgDtrzS@&IFgslg#V6L(u9Ily>8ah6Iq@M@
z(R0zb(Yr@odtPh0TM~EOD$DLpPhC>dU_kiqzF?zsN=3lLWSc6kC!^5H4Di3H2@%n)
z*@zI~&A}B)<Y#9W7UrLNd1HJ>7}q;|5G|6$z}Pr((r`ygx+v>&T${d0G^q?5oA}_o
zVff)yFyg2H>G<#xg4UBC5P@PaE!b&80oBB(%CuX%|0)Wuy+$ajph@jIvakebm>Wa)
z89X3Vqkb{Ev%duMYV(v}9C0lM^p$K%C|JpCB}zsT$YL<iwHVfA>|!V&!{^A<hB2cG
zqcv<=F^fO)s`IyOa3{OAJz_-%Z38$rpw+m9FOLgYi%ocoaC6V)e=!E?*KqkXU@7DL
z66nPU@`NTJ!OK})#SuZT8}y=PEujeP&x4N=j495`3^PKL_ELPATDPsQ^%<m)g@Bgr
z{bFb)?5{470SfR-)4O}keCHPZw`;`620;|#==*jQ!l;xCHd-BrB%KsuezQnc4A-wm
zk{bkXDy!P3P#aetv@@64?Xs>2W$HnQO8MCok4kx#_wRcCoy?eG_J}WFu2IUY*Sa&L
zO()kh6DD%KbL;cLM=J3oBEJh_6=wuD#wmm#rl3-}Uf?Q=?Mwlj=W=*|uKvuqvtbrQ
zg0>e?Nu>@dbAR6hSDHM!(Z_iL5sxxI<*Q82gE3SJZ~hfMRr4$mIY=7hPs~)$b~tC>
z*tu~&v-fxBub;oc6%MYRCH;$Ujw3{DWHKGPlQEL53KWGeH@CPgm{A~D$^FfJ%AHT0
ziPO_9cH!cgmtmk)?W?=IAqEx!G!jTypqs_SfxmSD1mgIhLlVA6+GQ%28uJ}8xA^bC
z%Syr*%wGX$`ht4qp^DRc{XIB6F$dH$+s^@O=r^V@V0|DxV3YCa$Feq!jiEE_Fz6GU
z34RSaeZS05z$m!&@Li&dPs@6Lr0z5-43@sg2`wzJW0V}%9$R@nvI<$YPNcM`TtDl$
z&o4S27+dF3A;YGRS1x(7Z!F8>TQ^luH#4rlIx7ra@An*zf=@7EY`pgcv|C=kQ-bh}
zZ{3m|tK5bUq?~;5oo+z05@03DhoaT=C>c`wbrq<%=p1DTZBCd>$3umoA=liYR7Xn2
z75U&9MTzaHR5m0O%(N?$7;Gnu7@lW(vRKjnG{~WUNHfX%f_F#XdJV^a-$Yo1heQ?X
z_sIjxrDEj2;_$xVprfa+X>J|{;E8sjuHVNKtU%>4W_lrcLA|FpnrV(<qev<jn5|MV
zqLrEtF_)%s5+vOwEPmR{IEg5cX;f1yMW$)P#uABH-UswTr_ZEj?1NkUDV_T&l0E!q
zb`-JE;a^CqLx5R*EER;*p55&CzC<JuHjx;97g5ED<vj*n{uFC9ma$#pMd%LV+1i5)
z2^jy9Um3qN4kk})McC}m*c(`tA1t0OWTSAS?z;POkNTLnzJ-9QBt~ewyPrKt{9qX?
z{9ZBlf}6sBQFkIE-Zo323L00c?TTSeufawQ#cm#$Jd(Sss?D!ewkvtY<bVHMHO{Vr
z%vU|=M6DYFl?K)RFT!T;g8#u51gMbooh+*~=9ZH#CO|skzu~QHzLrXG7#l~Iz1;5&
zYCi4M5TdwRP3nwKMmt%mItrz&6clv<+E||e{O0KEE3du1y(UIX0qUCPuN5WYu_by2
zhT7`t>5<k!8~_9XzjcMbW!q1|U%#nu1RFE1!1ET);xhK&N!YaMMhm=UhJX@axm@iF
z1R0n(IKHc>jPf^PrTt)@ev#md3>P;f;Hl06E(YNPF~Ekd^8@vsuRZCc()G(x=9A1U
zC#o5h#3O5Abg8sn$gr)UsVu3nZA7yN^F{?xFg*;aHq&^RW=-!(ciF^)FH?dWp21ya
zeOzaEOXFWsY(Ya?xq7gzRXLrqHPzJ9#apLYhdVy*mJniUQa&;|Kej~kiM>7c37HB>
zr(8xXzjxWQ_l9II!kgBXtZ3Crt`LfN$h+)t(3D?UbK!JSz}+Acid--wiw}P0SFZVt
z3NO@Y(SgO^J+?2f7Ss!9xBsFH5#V_d!hH&q)JZ5Cm=z}uTP6b>+4QTM=yTJ~|Bt)3
zj*7Bv+ehh=Zlt6UP#6j6l2lS0N_yxN32BfnK~RuJQ0eaOM!F;<6@;O?+1KFneDC|M
z-&*^>z1H4))*5svGjq>%U)Nd3ah}JenRV1ShfL;*q4Fzcy@$^ENMZ3KkQj!Y(wP$a
z9G(i3CdY6xEDytqTBPoeA6*x!iwqRTait%9!1NME4s6GKC3G(hDGZeu&iYX`L8&;q
z_{3QxHtV_%O!!!!uZZU4QtgjdE_TXUNWE6l)V8=A99Ml{`eloYO=F_>T7^GfIlm%;
zL6#s=F;c<aMbth-Sl{mD$o7|h-1;b`RFX(L(V}jM42q@Qb0saF{4qD)1BIil<FFuc
zu{p#950KV@NfiIbUtkkoxOq1m2~%7m;hQT)xUH1IU8P-Qwli+RopXMC`+2`@Xo`LI
zDmQe{nA3NkJ*&xF^fdHRKE@f%do5wFz!Vpfr}@@!vefVM_n==B=}g}`rs0zZRZ~q5
zFQ{mfpLzr=Qcr~ZRQo7ns8L7!{AJ<723s{;wNWK8F=n9}%Fxd{BmHu{D6jM3TLm9l
zsbxrwIcEvvdRkj5YqMsj+xkZc42_}^$Fc3t$Y}2R27{t`@E!@o+ht^oNQ3MdLypX6
zN`B1u+Ob_MRXlnZZCWTmBGydDbw6ijHTlav4Gx1^QtXJO%o2yv3!?8;49}mvZ+DPa
zS|*$Fc{3FqDaWK@P#;s>uxFQ=zHlg@Ib(1k0^u#v<t`%hfBG$$Ni>95BvHoZK@fGG
zvjJ(9U0vwG;5^1~FHY6Pgtu7O`8i8Yfw7Ne4CoS2A$yL~_Z*`GEtWt~Bc=;omd2FD
zo4p{l2QCq0$vJDT+idnN+)e}%<D3xdE~(R9ZWvlF&MGe>YGRy9gYG#_Qyn|bMlm)|
zXqVjZ`e-e~`Il<s!SI*v4;Vv{O=`w2`?hyaCl=0?l_F(0-H4clY{RDmSQvYk{b8FP
zMU$n05txE%@qAF(-;Ef&u~FP9iK7Gwu_U(j!^g#`_ooeXoLXG(;KrFAp%zn9k+OA(
zR&{f6BxhS}y&Elq%|He+Y<13sM)G7crE>N~Fx_a`!jT(1dS{7=(Q?_R{odiXGX=hW
z(7Wx4b8?LqNX-}tiF8ucF=8)uL)oaO;raUdA-ro7rY0Qte4+Sm@Vj1Jfvx&5oKaKy
z30>JMK0F1^Il6K($(`0Wwj=x2*@HGs<hd`?AWj6doAAk%^&lUa$(4IwDXBiroJ}>U
zQ_-MuHyz>x|1S9c(l+yaqhnJe<!}efj<{=H;}@Ey$?NvQs8`o9jujWvk4DjRV4DrW
zT`mKky_U1$$2Q+co3KmMsQQ2ZELSzQf0wf}kC<jW^cnRZ6%GGDY&{;zFa1FZvCz?;
z;@~Sm)_`2-D;mq^`JFP{^E9NC(DkQdRIE>i=6vd;BhNiLe(^)1czA3W{KL}r{P8h3
zTUg|uhH>;45DT(9mJ{QYp2k=VrUv50n~yt(`T8-k`Qe8U-hG-+Ck>HXeOp4x-XEJ~
zRafP}qV$XS+W|);gz+7iouy?>+)3oMBr_U$@+tJ*R+UUnR;(8%ZFH!SZetVQ$mDRK
zUfb#~S0y~C`g3l+gcs^bE<gCbzGr|G;3J<^MO$9P7kMj;t}m+Eh<rbm3i+daQ;Uw$
z>uf0WLY&n&iczZ^r!24=DTL!;!h`49QE_^2z`-vi?m#w<E#xMeWN0{9^A(t#gA20r
z<SR@xRj05Q4s5jEH9}6cnYm2s3V49UKvekNbXkP=C#`eidRdjzH*$zEAxZrd((?}J
zul3Ot?I$LBSwHMq3EZ!z@SM(*-e+T*ut^CLkr3QN-W9wfo;#&srIKa-*5m&ERQiR4
zk*UtzV~TTalNaL&e@d{QH$4;(R;tja$;`;$z~I@}@zr}}`}G%JET;Q%MkQOMN?VaP
z_lGQ#Q3WL{uB6FiClB#M=de(5<@ZVtO&}j^AVG=KsuUd?#jRm}x6LLBq53eryCKn1
z`o4@Wu;rdHVf>byHHYwo8{zUf^VXj|l8nTC_qG(1DE6d0?ob|!wFrae;rt|?`Ad3=
zol>)2Jtab>3@h!I@n2WNv@F8~pN~uClx}zialVyN>dRA4=Xw$=sL}ao%x&nN%9Xy)
zXB|aRtvH(cy5ou016r*<1;mpx&LQ-lQa{BJzq}g0UiXXJF2t@cvLc?uX2sUTp*ave
zh<w0&x58XS*<Fk-z2c{<dkx`hw?SItQ?=~7Z1@47+03X!+sm9)h-Mw@&4*SsjW71T
zqgS-qkRVbw(MLxMe9x~{&x~^~>?4jvmavxOqO*I*48vr;<KHpg(9!Xs*>aF)3=7P{
zdw>zr`6yM~zBmrqT2AmO4cUsLDLWfUBMZ9JoCZC)GTJ@otD1rD^W#Twa}P6N!^f<}
z0W<-|7D;F2kD}h&sj4Pe*XgoKjb)+T&=ka9?To-&zGN92m8TmkQKoCJJS`lin=+cg
z4cS8B_`!-<vTt#9;A&f%pshh#U{&~v#DiA)-2=?pbKhleKG;O;l%cqnL9YDpaP|H+
zF(jItuUGwe>M$^n={v5OG#yJ8c5~*Pc-_P~!?9~6F-;0a^S%pP>#XFjcdV(cc(U&C
z8AZHnxSpYBc9cMmpfTFMKXX7rf=Q{a#y~3hBy-Ryox!eC(kPP0+z&mNQUzUwq04=a
z18d1y;X}fG))2I8lULYZ%H<N@i=*N*n3z518d>=P(s|2Z!dE#Qn@z$Ofl><x$AVb?
zEKx6f)KWTio>+^<X?$s~U6OpJt9w6pRB7aZsh9biOZDYTreQt>sC40e0BMkUuci=w
zepKA2pvBTb<7anRAqIM1^EhlX#`%TFzv-z{p)zK!3Ciw@vz2dtkD7^}pCgaYEc5q4
zPuhxnLCT>aD3unPox<+9EI$Ifo6jmzaVaE<dx4Ix(p8wW^2gCKA+eeDF%^XM`FKSH
z?}4Nwq3!6`czDydg@e|W=WRkECn2hnLt%>r?U>Pll2FYKQplO-HaoK=1qDT7x=2`p
zj)^#hTmdBMMYv9%eJ!amr#q)YHNIqI>1;eCm=I+ftCNh(=#dCt$LG-ijmYx#m4x|_
z<H}G%t4E%ib*AC?6>N$Rh+5AWWYuG0)$YFimD8P;_d1{1OF*M(E<>4|F7N&yoCopG
zFpCa@p1AO7!08^+9~Bp-Kh^N|zQb@#@*9L$M9y*_X6-&m;@+2)!4<@8!r*<tKx)zS
zvk&)3FqmCdwLC<REmOB$hqwgKbb$2en^`fzH|YwL{T?g1u%hlT><98mXtKt~NwZ$`
z-^-9RyRizpGt`vKO0y$yalhio+!^MXab@OUv|uV=D(fiu@Nm<SYw^_&TNffKl3a*T
zE?NDw`V-SgsS;c)Qp#YRN4}Lt9M?j4y|zPSAA@Fe(uIs=hP9}4<sN8kycj!>e_dv?
z`s*Y}o(<*vt7dcn5mQ#&cV1~m$lYkF!GWdZ;P;sxA(W41IsAyc%yF^(BSFJHV^=6d
z`a?^i0+;f27zSLk+2@I9c#}rTm*X@_rnti0xXyS@n<H=3W6BaMu7sS{e3_Py@gMn0
zI27}{nUz^z6BDF;4;$dyaZOFUm(hnic+Otge=oH4;feHmDb8+3ac?W~^0cREaABqG
zPXiB_i%Gok<HR22mkNQO?KF*PA;{`x81En*a;VOeI1uZ~a>%{X7A$6hAfbk10B+rr
zMebdaK~)=JEBNz^XMC?@|1FXeG)zhX5*w?Q5KEZJnmFUh%c5X)JV~nV%A*!do#G^z
z-4yV)K$@5#NI|$w{3|JI!SYv8P8;0&xenU4e;O)g7VJORV(SO<D?I)BdCIs=J4&PZ
z10|_VD1_K6-UH3n*h;#iD0n0(QQ4p~_a;JHQ09?r!l77snl5L`Mf)-D)_K$HHjyK{
z0G#M6cXTMMr9w<|b8TEvOeHqBf7Y++p7E&p$@!j`2Mr#secL`BT>;B7U%Cn2WYt#l
z@iRqSb=~WtkzZ5R;q9qNIm*gm{o?Ot_*r^ix4zR#w-D;E8j>n*pHNjIa2E;g-VYK}
zQLC{pu++)o6H5IVYqqipUH&i!V^N|o{J?LvV(P`t%B^<ug%n5ONhGbc?eAQQr-B)3
zD(OQ->~9l~CuVJ>$UYJ&qNQLaiBh<4u*s%zInLr|n4J^q4i#7}yf(9F{kWqkaV=ie
z^XRO+aPI~2#pj^`hWI15x}{tPB98)bZA<B*Ebgar%1?wOK4bICu9qmDRTggktfABR
z$Ran+Q>us;U>zDAlZD;!LoMEV+w!&PV7Et?p5uf`P|nzXBYEIkYIY_D>~9Ys33?+A
z(Qm&Kjd#IhBS<vHhmDnpX046}x1q6C(JuqmQNoqbTQsGafAtY3M+<}*thHlON0WCi
zCl$wE397%Jeu%Q=E(FG;eZDYf<ecP1(PYDqjY`qUVcL0^PEt?>C3;S{`C`N!tI)PU
zs=$NIojz2)=Tz&-PlZT#YQe}4+mB%%K1zE~o)RiBsK=u-<S5o_KiLwy&r(p?7p%(w
zA#*aZ;>o6?Rvso-Z~h{LH?}TAa~HyFn-)abtJsu62pLqNba3~E=-`wp+x}jAD&zfq
z#w(un7yd)CXmt|fN_vfWmaNdgVI<W?KQF=$+|yqN)TU`aMDx=dKGK&8WmwIO(81VK
zm&)?TAEUo<a`gz3{Ghb85Qf`}e2o79?or1r(IK=w@+Bk_NonJ07oTPv3DaGa)Tg>G
zTj4Cgvn!vR2*z9P*OW7>D%(opb`X}>BYi(6**dazBxR&2e-glVRQ&DFp{za|E3rOn
zf5lFA@e!%9{$^`+Z?k_!C$@iYjeHOKYm`v(7P6c;lkVd`j+;EnkvfB!sZY=SR*OyU
zQ5;vIGrL0dJ`6r~3a1)ZoG%=gcUByC-R{T}j3U7yuJ>n@UG3^UwD?s!<TzhVr^<0_
zX4I$&vypp~WGe2K*qHp$Q~db$+uhQGD(1+8@j*%)8%1$r;_sh?YeN}`k`+SnklOQ@
zgC`b}Xc$&!7-XS6WEK^fY?&H5Oe6v8Y<@<>Uom9N%1!!k7EHn*s~hj9CH1L?$<4-L
z%S$A>h1^;vbX%y5^4;2SN5)092`)Q7=NTsX^yuLQ>U@H-OW}vlJl>4D1(V)mX}z1;
z^9@UR*mZ_7K)X;b8Zzou=-2g?3BRLzfO3syQdjyZbJ%#{B^OT)hQ9AB_L|W}@6X3f
zarBYzo?_Mo^2ZalXMFQ}O=vUc7M|In8*q~Le)OZBfl1ezBRb_L%Qm<D`2>|~M%L6l
z-TVt;-Y-frRD&j8VfI#J&8(<##&-xUe&Hx)Zq}ADGs;*|zN1LY$DnCFhu7R2IY!rs
zw3-ey9E&o~@)uftD)XU{itXfGqHRUz_pdOIE3(P0lGoC!uX@X;v4=E5U}`&$g2Zj-
zLHdN8oDjL9fdu1Uyw8h1!&zbWnbfR(W9plmM=Rw$cRlAbJvcUnjZ^j)_4cY+y<Cxm
zCRW}l2X~$_<<QQ{DbtDev_vOqX%o$l#sWq^e^UHJvwZEFNFzAmcdQ{%^{13Z<E@=z
zf11OCWz9s#F+8y+cbUwwwHXs!y1k=SO_}8cIeL6Lk<)F^Y3z#c?#bLOKqkV>^-~+3
zcwOv|b?jV-s?=ky&&$wHtUv7jp;~D;ZOJ_{<}h-IXVsiB$@{`9(}au<TTGMM8Rapg
z_>CYI$Uc-=BqRH#opHpA68L^#^Kf92nb7V34rjGtn`c%H*$DiC|Kp9Fsdg5$6=Ee3
zm$y}S@Ic3vN$#)%MLK07G&Qj^k!roQ9UWz>Y%yUg=Oq6dllFrP$4nzLSMmZ<=Lwft
z=tbb<Mn~&rZJgNN^Q@UrD6j2)%Ag7h5gG&fm5;Ih$#e`y%8UyPtB?I%*r+nQ5hReQ
z%tZr<v*-@%AJ@YZn1X!gu?xoY?V$pts5?>Hf-Oo-Am2W}7~bkXJT|k|PZYk`cqcgR
zdg5--t(!cQ7=Ywt|1(d)61m3h^q^3GA;Z-*R^dfK$vi`GXPJw|-K@A~=U>lbvzvf$
zQeX98)c~*5(Ry|NZxxSt&1kxikoWw-*S^QwqVq*{O(!?-i>o5z^!rZt-jl?BkdjoB
z0=0?E&i;D=yeJ&kc~e!*le+3J?Q9%gdh0b(MCjUW*;OtFdXy^U>e`U06Uw~c!pFrX
zHlajCQy84~R9@J*f^Ag}iiZTsQ13A}rF*AryW>#<?Wv}zwS%Bl-#}D4+80~DPGWt^
zw%uVJN<$cd=g(BKcaCdU<p;KTL~12!BAP6ZNy>Kq7(V0u^4N#x$K6xzChAK<oZ_{-
z;|0`#&L&=A8;hd$sktzvr&l{4(WA%IuZ5`+_j<OB2FgMybEe+db8{uYJ}e8D(+!=T
zNHx>h-o%YaV79m1oZ8Rcuz7|TPKl;mtA4a=UZz^Oij3egGyHtR8W#GWM^8$&+B2E|
zdaQ~D=`#M;<I&pvablwXJYs}YGvfdA`q0qt?l_hI^?Di8L5&=i|9S&Pr!?sY|9L$T
zQQ`rq_5TYlVavFSvf+eO_U#EB(i+WRt>IqG?Nx2+H=JxaSix`^&@$OGXSbFc*}@r(
z&L6vx$_5hhbb%g{t^~t9s=rInlSX=#M_^<)N;>o8$rFTfdWR!m3zu;r&!D`#T@}{>
zPt~L>BvM%^D0IsyRRwcLOfi;+pTBc>m{3+-jT9Zcj#)9&=<iK1rT?O4LvW60@InH^
zQ6EWTbvP=ilk(!M6Sfs^?B2fPF^4*7B}FJdMPv|R&jF2}hN@WH!JtTVOiax2NwpLH
zj@YC%d(iFm^)$(4Owu`no{E*mZckO6cxPeSeNl%y8|GZ$x4*EdOVe$a6SpRmsuI%M
zt0)Hm@cDzfsm3VqreuD(x%Wr;Si&;}4y~nm{@%jr$L!-X0pXd8Km>XPkWHC~oMM@g
zv+B#8J97~4*t6YV7O9dwCT`z_Eno~in5|>S8Rh$ymv`*GMxzs73Ag`y>wD%^$>!-u
zyCG_ZEW)Cq-^a&EC12XU5vI+{JCCDHzkRFuuVd@)&8Q@)pY!YMHX0PuKDLasYR_+V
zvE^rxx?Kv>g-ozqHYkN${giZc7#p1O)Vevki2pJ@;im|7-=U(SGPbtHf#J5dx7%50
zs*}3yU^BdDRum~K8Cz%n@pofShakSwzNm$p8|9AJppYL`sty$T&$gdd4ru=Nftg{*
zX<3jpY6bA!V7OLERq!c+=IZ=BYu^tMh-(e)X_5jf{VT=hjsr}|)Atqag9k4jAWlKq
zMbg4(0K>#)3rFDa^<Yt+b(h8Mwab_;7{9q9|L>C#WBB0z_VC>-2|oUp3qbV$n}hiO
zv@Mw3QtPK&iG)0+gf$KOiW*1O4pNr4QXc5Qltz8Uw<P9=eZ42R*M?>P-j>^+zot5e
z#&VA^7PH=V-`sI?z%qVNLguD?b2f$a@?xmD`bPYJ-?tz3LAqTDj*Dk4As5kjyLyXG
zSJhs)Zc;bbiC5i+U^2_~#ecuOj1NQoZ0?cxVl>i%Zv&+pd!X7ey#q%4$qqNB55}<X
z)zbg@`?!^nkZk0WAi}0<wl}ts?x~Q~`mThX?cS}wNBMRef_=y~^X`NOGsQSki0lv2
zf!_)0ok)m!Gqnv(P5I5W$Xl`&bC?p^cTBp3#u36c38Wh&qe$$V36lTyeY~e0id0Ht
zIox503?bA+=Ef|-Vo15;MeC+`W4DHUGurIyegnChD#l`9ulQfi!SKg+QEnxZFK&uM
z<g{d?`K0{$kT06{obR>QV&P5Y6Vm_sm>9!1^}IeMWYRUnHD$gx*)0y6i(+Dn-#iz8
z{MVh3k@36H&Hi5=r)Nd_{{t>4nKuJSa1f$Q>K0UPIdxGMR8qoRPp^R;Uw(DF7kutV
zLZZ*9FU4r2Kj(3zIBnTEEl~2VlsZLavd=6`2Q;U@Ir8+_b-6k&y3%UGgyFiU!6I8d
z3Bc$upc}v=EL@jO0KHu)({zLv`aD9qS)Pu{-v7D_9&nQE|N7YC_H&NqlouhEyVXUI
zU}i-y7s1;%Z4LMVq$;~vIht-}Md{+Hb|HCE*+Y*;*I8gE>v7G^Z|1LEs@SZf^78&T
z3vUTzvYX6S*=<%Y_|Q0Z5PAL*{JWWtU7T3QXJT4a3R4@!-M;1C8p3&d9zm+hLJ<+Q
zu*^`PGprDcwj=i8!-vRs@6_X2CUt>F>3>vE=`5ToFrJi>V*8{dbz}!_C4y6q(VYGh
zyM9?PD~}^Q^E0shYeFy>4DpG!;M1e&if^}gPm{GukZwTG(k6-zPs{%N5M}pmv{6=^
z@G}lwq$z<j(9a1!r`|7<y-i5Ic#26`BP%Nl;K+hej)O5~R8~nSJbZi={oVwwkz0Qr
zpA?Vn?4}0kN~mIL@d}iUtyg5gpiqRs518w>wFmoapr<JY64h*8iC=B1vkh3rb#0o6
z1n%aX++NV;JB+CW+6IE}0G8H$7H<|vs4(`ij7xEhbd3gRG(BKTb2Aa?n)OtTWmZyA
zL3y9G{skp*y4sPZI@+T77@jo+7u%?9edQ?6#ALx4?q6C8*{)rqLwFll8A6b{lE!6R
z-SaZxpEZ|tdNh95%U&LCiC0t^XGUeFQzcvdcjXJF;6$=VLx+M<VVRa8HbVlf<m1f|
zjVD0P=r<r}BS7=IpAW&BI@jwj3);(*2wYf-*O#m1&u$I=SlFEMP3<DtX4jsZhKr0{
zmJ9B-piWm<k?6!jfP)5D<P5&>-{b3U3+}rI*MN4MQ~*Q5iI3dKUsqQ>GtsF#kW>!t
z7VP@+49OPEi2-67qg$>*7n|k1Wsxy4L}1qiJ4Fdb&0=G!MRU$U1<eY7F1hWdip_u8
zvgYX6b}l*c5KD0aE480V(HL6nbCmzfT?GoQCky{4a-booOyZPIM)7(j6AlSx4=<UF
zW$rD8s^4D2MNw;I8Ag2thVKr!Y%37fSU>XkXx0lRXihhIiy+#A+|xc}S=Fa$#xjFl
z^-d-GN!pquZORsV#a*wthNdr8IJ<EE@K+XV=LfEZ{0r+~Km<P`Ol<AxQMefTdk<vj
zW>fA|V%+I!Jv$sA%N@8{<qy)ib@tq)lY>~#FywFeku^wVf4N6F;UmHUAoqb}sxyfJ
z8`$|X{ZoelE&ic?-{`kzRCILaV4-;1N@$gzS&GlntPO5%I~)>9*Qc(&@hxANhQAn?
zU*+a}F*|bJ7yzT|FSDHzh97w-DJzd`xz=8v!xhIWu9TcEQ&qCTAN9`bqsK>??0i0I
zf9=BKfm$`2H{Z8^Sl89w{>!mzK*r>t@LLS}$IT(hLpnAfK+2fiB9>LZ5D^M&>FL2z
z!@REw1X^?|EQun3ZmyeK0@#osg5&D1dQx!aG*~I6a-auphv+ALbk%^j`cP;zPHiT;
z$9KQ)lw(mCDUI7&PQGt80%^Bk$Fj}?E3G@_Q}7Y^5onhbK?oG;H2a7>o_mEle<6_4
zf6yRSyjl~_Z72>~4z#Aa-CZ$1%e8`ZMK}bK+ZuJ0Zi8r3ht)>k@k9SqDqKg5Cxu(H
z)hj75YwK`k#hIw0Z_hrV@Q8~K#y@yDv*ld>)0NZ!>n*dQd7y5<+wViXuN|~9G)W@y
zFciw`dtEK)%i9a47y7qtVe&BYWEe;PYq`(A8?(8d?TfMvIU(fR8ct~3f5jjs<m)+)
z$l7^7;^UL4o=x4;2Smp#yshxe;X~|yF0?|6MXJ15!Oq?{b@dDp2lh1C$uoQhsH?Sf
zqFhqe3#e@NnAOxDk3NmGwh?)Hyp{_vbK$r#;HTa|sHjobX=|(fw^dgyxYCuVd2#ye
z-1cCV`5U1t#tV8qiGt~_MGunz7<%%FCUj!E9=k|RUj8iG_nZxgh__uV045tkQyTnZ
zZeAX(gxjzbX64(o|5?Opct80+yY;U(_31p$Qz@i9b2tBe8?t_P2}sqg7m2qo>1Pid
zF3L8GG`6(F@&o5}|HSv^MDnK&sq|;|aQSy6wsOY<KBBXOKE6W+1e>LsmYWNUHudH9
zQ-fW12Dcx2a*50gExMD+j->4@3U(n5BslT2Lyc>o?WB|22f_{d3z2J1mm_))8n!~6
z1|1jQVphT5^27g01AB_Ljxxx3U{Y3yb&YA9hP^wUomfkn<x_SX1Y75iYdsAZ>;OI>
z__1Ov$OxcNbtu_jt!wHGD=syzqj@nDJsRI1!_a2FKEEVOC);}W2j<kIM}eP%7DY9G
z@s3I=Pt8_qWv@9sd~>(t3|}}EUnhPk#ipB>&h>{el9G}V7+<nV#Y9ISxQLX=&e&f@
zwNQC$+e1aG8ac+3^x4~;k@KW)Pt`GcfbsAUi9KG^l&7jXDPShj(_4&W%6?CN9L=nV
zgmADoKYmaGEBn=|Z+KX({zfp0zH+N|cdfMTD=XnlD*o42wS1j7dDF*6p?s%Jt@gRo
z3nQD|^%`v7i+GdXJM7l3dV*Vuw@}8#)c>f;cDfB2lcoh&i`t{fmpZkt&1M^qfIwrs
z#MPLDT{XPXVyP{t`IX!KTQ9f|w=Kr0yjvb2K^Ayxk&1teIL!WB_pqLyYItL58ER16
zz1$h07#oO+FZdbH8i)$EgOih!7v32`QEu8x3{t=u^`LUpueA5WGsWgxPw-q_w(_3k
zt&`de*5>6=(V7Lo2;Y!fZ*f%2qzW6mW@{ci{((m>O9bvmG6R}NBOUPd;G6fwae@y4
zGs<`b1mwUe7%CN#Sfi0uf1hP_am9b}HxP``pXGc~nEH<H_CQU~;$2t$)OpW;dbxPj
zIwSxW{ASgcDH{Po5WACMV4GeLnWZ@`#+q>@HTw;!Nn4mE6mYbTFTLxGt;5O_7VOK9
z9d+x=V{EH{wOG)rlFZ{05NJT5)PMMlq4n{f2<+sifBj<aebok%&BNr5@Jzq!`&1`?
z?#v=7na+%asbQ)qvrjmG<xhnww;bl|h|+35`93YsQbrj`z<gCy*)uMsMbjNf*droD
zAWtie1p=B<w>^{X`W^A`Or!Y*cS%}n{%&CQ>#H;owE}bJfm@H&v%;2b=<&ix2?=%&
zNnDcD5~HZzs1=nQM|4u#7HK8_oOix`G03`{9`z&3p$&Ob($`$nJe-u-y)|jK!wc6#
z-7oG#Q3jq|!RKfL_$myCH@Ijh^OPva)VTAK$hceAY1jS7eg713T#K-iZeuIg3ZpSG
zK4IbTcOku=VsaFN7V~XqAm$Pt=XmXfrA-~<;;Fs%8RrF5QJ61qh&Kg~a3Uzy!kjI!
zlAIo8y$5Ta64!<Bg-5h9{8bOMYJ0TrjxEg;!<1Ubaltb_nris-xbn%8yQa3h9>YdT
zpRvqn={l=!YKf_*E^_a11D(SoCfrCUSHvndqs@lLfNp`4pbfGHW}tdxCHbfEHRC>i
z2F&&|Qc!hZRt(L|!7EMj{wI@2^h8lZo}36%zGoyI<XGj6*YeyrGRVhUV=*ied2QFN
z;NAHw+>m|BcXLmh>l=4i>{oAET#T`%sLB@^ZL&leH@@VvzK!#kewBNN;t55NnXPdL
z>-1vjAYN34A29a&5XnFE$ySdw((^kPz2CYN`J|s{!*`+gT%#Qtzx-WAQ+nPSIVepx
zrAEjj<r-pHqElj=fX8$WIS$^6WE=d2kq{%)Nog|CMuTWIX|ku`d6_+Zt~7>eYj&X^
zv}oKR9Kgy1(3zfy!LX-l&c=2wSAS?`Xz#o!t$7?vOXN|dE1ypZ{*IC-&7cq{F>yWX
zE5ol%+Ahd?X(JCKD=eO|s}p35f@QJ;wMC`IbUs=X8=h7|ts}B99q4dj`1#zquR7nA
zdjdWsneFWk`!2>Xo~i|fs=nu$JQn$Y(skJWh47wRiyAcThpgZ|RNQ#;dmgjXvSU9?
z;ltnMKKjvOWuAXRi^W_>$67j{g`8er<Q-qR6vkmj^H$M|^IzcBTtmj?6)w&X1>Ovp
z^!7#aJ>5f7l3?Y%a>GNJtn*1QgKMG6461#k%j|aeTKBiU7vuecH@p}-W?vL@qa?J<
zFYjv~uUFdA6ImVT3o|J)(Vxr4(&*?%36c%jpOmFG6E<BS-?Fu!V@t}uGKMEHTStE+
z)@DBEV;5pfb-x~jRN9F!1Pc7=D~l8znl;K7+Fs=yxN3DNPO2<UTC{%kfE$qzNpOsN
zjAOMC@=OwULmX+QXtk`KOH@G~hGAPFn@OodaigOMJMYg=nc1YR@lK-HpcJhS?0AUr
z-P#*>FBUf?q=xRLjTbpwExM8X9ppJrKTGVlQa`uJC}_bP^^qrsLf<IhMPc)q*V$OV
zA~oWjD!$Phb=Kw|8lr?)nnxxn@VJ=p3-%GBy5lQ8YNX$R&(gk=n@h~Q=Lq>Yu;S%G
zZw6hn?i$Y~N;>^cgmfW@q5JutO_pB%W&1Oa&Buk%ck41_cVa)HD?Vk*li$Lt5ADNL
zfcc-CmVvF|U+Nr7%JrR|s-1KUt})%pcwYsW)l7CQhH+!fl&TB#nG);uOW7@5E%!8%
z2ayR$NlAN6tuVUHvb&cy$15o|0h}-;>PP(PsaLcW3(CG9jj3agrY6Gmo=Z?1J;`HV
zU9zQCUveyBGzx7$OwteVEsy@2Q?8P0sAKB552iQny^~Q?(uo&@<@$He*2!V%NOgCT
zLc`t!=i%ahqxBj=-H4d!D`Qq9HmaMq+6d%Rp1l~5bq}zC>)OoLS?(^^nD5JdIa4YR
zcI5rWs66IcX!1F=VPtMwMJ$%|YUSSh3(1*vwbSqhuF%m?j?=og18pMQ4TL6lx>n%R
zTb{qtySJbJEhv{oTL~xmxqdV(@6!CT;mUw!Lhyug)QCs+^FCQz@xo3&j`Y3OSPG$1
zN9TGCZHxOJI@FA{g7zmQ-H?o_fU>V2vPc{0K-z_@p%b4#Si)fB_(E1HhojH_Qy-<5
zdPS<9Yc{RQ*wABiW3xL-qyvBcSR1efy5e@@S8J+Z`q|ZxL4VZH_mQCstQ+~iEofmT
zW6i_#N*JwP!ANd%d}V9Gd+xZK@Hg<rFQLmrDxwp+rl8Q{`N=|tayS)UcM^Z3cRu~>
zbcp`yeM8=puby@K!ZBvOu`CUTEZj>aE78UE1xBC8@*G^N$2V2-R(;ABS_!Rp%Vk@e
z^<QIWgYLvH3)IqA7F*kIEb#KUUNjkqOrs8}ya}tlu6oiryKtE0kq-64;xf*5ndFZ6
z_fRU0#h4bRLI#W1={Umu{n&7t@En!=7N1G~hBLCG?H7dXxh!g|3jGvi+K_Q(W82gA
ziXo9Dnx#*)E`Itw-^11@d(3!t+|>K`XVuxg5MZ$lCcm)b=pl;3!NI{Bl?@z*++MZ}
zl+gs`ZMiyYScG2pRN%v0oEu03x?^d@{H|+Onuc@DWjukByJ?G@eSOth(yUgPYc#A;
z?BS8`Wjk2kCrcLhnLC^F&ss@yy`#Gg4HA<%P#->0b1PGm+Sr(3$nQS0n|H=UXz%KX
z`Xb*fa1O1!S-76`*vc`26KzeFFU7w+9nTi;q`%pDgBa03;>Q#vY~A%k^1EaY8On>d
zh@7SYNHnt0QmxY~bv}5!poI1I3BFhbt)aHF)5EjxEJ%8<TG0_*zY(j^ug~uo@GxEC
z$iB6zubXo&sjNrV!rF2S9)&U9$&d)2N2c7pI_P<(0NXvPgS}|NZ>3}V2Lw5W44g`>
z>4~&Clz8T_<Lp1`K5t>t`Q)ziiQO-^@7Q~waJ{*i>sdI*n){GYn4t2Eoy6sC`z4s1
zCeqk&XV{fR`G^&pfv_8*(Wsc0*QVQF+D+T;kA4R@5D{Nj3N3vXvS=k`mFVm8_$LHn
zAG1Y&->>&Ba!+$Q7<haDy6poE(pFNKv;&-?x8Ki<oJ!H>d2ofKZm;2Fpco>@=l06~
zIVILGpDRB?&a{m+bwis~+yyb4fr!21v0HSj-(M7BH7fi%Pi>GmC1?>o@+&ov+k0=p
zZOD`R*A@LGQV9LN4DQw4HcEeNZJV6XM8U&X*PE$3CjD10-2l}dwZ$qswF`cI0mnka
z#h@3R{Bd3iF@oU>?OTcNL{hw3|Muw-s0P}SLK@ueJ9y5{*8ItgsM+u1JnLW+Z$*ql
z0^PU-EeqFUvNTuJ^3~n*SZGSbl^3XPYxN7g)bqO-U2@&7@OpNo-s5cnIyWOuUm#RD
zc*pQm+Q8%dPU+;~n&~NMzb9~<+u<oU%G5WIsmcDgmXFDA3`zzD#h@(oWo#V@q!MC=
z%P6)@J9OlO^h_LLgJ^k~Uq8LUZdF@F2dd}$&J>zDi^Sa>NfRASbYeBe?HC4*gPixn
zQF5LX)^FEEa2C~M8tmk8%((B!s+T$M`GC4nCLLUBVg)PJDUgtoo_;NH$cvQ$O|#hG
z9VF|gJx%wPu+Bh2Gi$>g07GB{VGE`%%C7EiAuy|d*8ONe=yCaGvd+;!kBrG!6eKdz
z&SGo@uWLffM3a2mb62Y9tOZlM!d4=sIJA${L_UWDlyg87(%aihHLLlzNZLa~PV*6&
z;v88SOV4Bd8zP6*^O<hu5fvT{mhGb4Fss?<j#`(?F?$yO`s~{cN(<cZ#zTCcfPS^I
z*!#hxf2VYUkoU5cs!;3)RnagK)l=oBy>|{Ei0?WrLp>Y)ZTM@z_mmaU^KW=LePd)|
z@;kEK?)b5w9U(51EL6|wEgenSvC;g&_8qrvr}TbQ0?`(bU(^7B3^y>`z6MM%p00D@
zd|72{4CX%_SuD3SltPYY9VD8(e;%d0UhchG*?t3?u!tx>s$FD!b)^4}{?;DEh-G3|
z<>gsZaANcvq--_Ph4ITcO3(NA9Qgle0a1MyWt!)*!=p!h0Da@nXKk54R_D1Q3Xpy4
z6!j$i)o$eO{cy${(PGgxJ;c<MCNS}HR58|L$prR!pZOu<>G=;JbK5%PbIhB`zDzab
zh5>Xpk{s~?(6%<p<uexu9r)9cmj}G7bvn@RgXw_IboaFCAr?jHcr!b~Nv;yy_h$`1
zA4<+|#V})+M*MN*jVwo_P!DVwA9oG+L~TD<<>m7&Xv%ZqfH^yV6deMYFy0K3i`C!o
zO&vlWj>8eq&x!S_f1rAQ>gJ;OtZS;6`6)@uWb?Y6H#L^m_2|&SVK!}NM|hkIp85o>
zUJ<?S)z5`XJd1P}a#g+h_!U(_<S|zDu?vBg<eNo;)5C7@;Cv3-8?pXR4}W`87zj2R
zXbP#RQjQhpp4(thYZUvUA2o?x5%!gscjC-@T%XOA`iMRI^w11+ekgGn4lc7vB!U#t
z@9KCq8c}ZMo80+jJxReEpDng!m2y6}W-NMz)K9#-2roSf^yu&mVnK=#{tE<~s)F>D
zj#iuW#Z^#5{m+MH>X;g!eWje{0|f{KO>x;$c4_29+6!LR7vcN)o2}SMeA`cZ<hCr>
z<*7<43u==oP6aQk@pDj%C*YGPMvUE8FX>Hz+-O41cGo#L3`7M>p9AngyO_rX7;~jl
zf2>8qS7_MW^b>Q~)q+1dBC~}O6!k=bUG@6SX^3pu4gzq6*suC8ZFvZ7<QN2gzEGT_
z*U)yC{vD<tct_2Y>WSi0D$&8t4gl_wO}9s@L0x36mqX`|6;}t6wcfNGdDuPHKDfHA
zn)J+(U{Tt@vGD|R7Wr?kD!nd0q%%KV_E63rC6SHKZ+3ZLSidY$bGhGtc74*+{if(b
zr8%OHq@ma*(WlTK`|_uRPbfkT717j`v1IJe4VhfNsKn}eANlnX{;9cT_cG?X4oa6?
zCv3Dq&w%<2s(+ITKTU2vqZS9!ASYAAu1Ciy$p-EHcd^pGVQzP;cZ;X0dL3(^vgsLm
zP0@wq*B2v6qjw2nrP)7*xw%!~*XoJ+lw{Hy1g*b%Sds4I^{*+E=eM!q&(H)FhW&yT
z3{^AzMte=}_M{eUwv?7j_XF)&f)t<^CTR)-^JJ+{N1}aXJ_>?^+~sgxy#{JBU?6S2
zxw9B@fu4?GY0c~3Pl5caAh1R!Ki|-IQEu14A2xxBXF%C0`%$Smzs+Rf?*t2)eyEnd
z?8gpLs1Byq<gV|{4hwJB2|umc08pJ=ALP_pFu@ae-C^ZV@OJj`yg@YCX7Q0(VW-gJ
zl?xu^mTKe#><n5dZ77+IW-(JffW~oaLuvjfyI^+_z6t*0l0D;62~v+Nk&5x$TmHeL
z$L_e3XuEWuI=s&M4VQbe>7j;SDK>_?<l=GMHpZe&ZcqspmwN9v-8AmRd_u1!5QP63
zqGZnBh*(W=`u?&akfPzx%<C8&ygsdlSR=SA>xv1-XZAbG?B<;vNdkN*H4gA6UI$4w
z(<R2Ji2f=d#R5Uh=0~D_Y+t~Q8)@Pnk=VyO%KVMl=)2^(+2<BDBX)Rem=`?B!Euf^
zDmjhp&&I*|u?v_p_UB79D~dFaZR(udBVi@q?R#Wn8Sy~^<L6MQ@P8r;UEP76rq$w@
zHKBP*I-ggZ(AW~YS%H+3%jz}p5CB*3l-lz*<{1I8*}2np>a$)|d)AL+h_z2-6HVso
z*Df`Yt*4s=m+1)ZjtLT?2pz_j)&mB&^c*zMqMl9<`NXq_1N&O|vidccYg+DcWbIE7
zNnexF@QFw#1beYRoH<<RWYcLzEr*KfhqI2*8YmpWi<{FB1L@ajvKG9dtLRMyp_y1D
zi(~0^34&2Nr6%Y@Ze6@R`~IL>t5O)=^2#E#j0-zsIYi=|NCJ%dM{=9UH+c7sDNj;N
zXhCQ9Z6*hMy+%hI+i5Jw`|NA7@^LG&ZsETkz26!5_Ws=$u?T6mh9%V-N`jheN{MH*
zf3$596SRBp3<{>6G_IkflnTPJ4yURf5!NO;_Peh&25U7elY#J=)3IvXZU5?|OC7f*
z+Q*A`$78Dg6!gM{J0TD?$AbZC027U=`ZBDg-^c@fZUR3Gv5VRUxc*Mt%z3$0EQ%4&
zN2oVPQMz5{9-&LE2ByZ8{)t%A%@wr}7!Zcg0Qv#&5`^{wm<Q^M*#ZWE`U<X@SuE)_
zx|Wq=pu8}{-aB+p9F8t)i%jm0`!_=W2+7(SwVwQh0&0g%dECnP_ZuD-dG)r*MrfDY
z<IT5hCEKq__?|xTdEKvkG#=68I6l(c`hZf{<X<kp{_lRsByo7kNQSYkpGu0+g%-yA
zn-eiTkKQ%6-e(4FVD#B;9NRk+6SzVq<qF(Bhue4W*=DsHHVFxd-pv-A^6B%k+=svl
z7Ei%`pN3c6xnu4s%1^87o>p#uba%4YaHx~QFwzA-gyEeDF@-ugOBZHv>_>{`(@xPP
zY*kB8jZdC;y|M1hWCvj?G+)!{bSdOuUDY><!)+gEbp$@<iCg>VSBei~s&kjy;QDF*
zDU*Fe=mK>G^R_~5TD0%N_?xDQs}?vL!NbKTd#x*xc2n7}{USE7DI9sU35YPJoVkuZ
zeYbu}4%=u({q!)a+KyNQL(diqIlRqjtgw22%fUC<0muaFId+m2=-0=WB}3xYHk&Q6
zi^X;tJg|$9py(2{ds!asvYq8oVs-O|BpgBK+*=_!8NN@$9Di^S9vP-nN`xK^Nu+;?
z5)KDnq7)QE(~f^J{cu<^K{9YZ?w=j;592?IDr}b}g3f;U4TMqC>MZhWz5mCgcJsjw
zlU{vWXSDukrkKMKK1zbAs`IOZ!N|5nNS+&x{AvYA-0=qszJAf^O!@0#MaV-|g-SH2
zc!u*>PcH;WHh#|rx&;Pp3?VdaeG?=!Cj}P!N5#IWf>HMa*r+ykdsThP(%1;>LO@Z|
zWeO_sK4QhKONbwC7DPB>UkFBwCIBo1;B@4-5Gn|kBt9@>gFa20`l+NLVCzX)Om(_;
z)k3y@=3ZYAtcIzIHqhBwDv8ApkAD$~%t#H0K5;_@)msD*yfAP=D{a|Lcgrp4nIrn#
zi~&`|dOP9{d+|c1OX_#4wwKP*1DQ#6G8T1Ty;^Sg!Ua@k(#9&tZ~f#!qR+ix2$z1D
zA0~axr>jp7U@Kb^zX4gubRrL$;<1r?xksNml$!i_AGnpk(o%+^n?V8{u~HP9CGQ<?
z?(CG;IGFUWv39(=Z;%#;7OtzV-)i|WxkZ;&H5)$plWjezyNtDO8(YR6x}4A%vAQ?6
zCi)>l{J8Y^^6^hc%$5F=;-N2jugL3Wvngq53%mO-kkA?un_7k|3UyFhHA}`|c=VY6
zF{(M@IU*@}c^lD_$=e<{41_OVH9Z_ob3YuBnT#>h-)@4d6gBKWEx?vhSBm+~tlszS
zlUEc%N;WQ^ZTE!{o;ooJuFO>ose776kjLec{>j`xYY3gcWBhN<E}@ObX@fw_Y^EsU
zcghlgUmt<Ex0NL-B%6+b0DV%&=X`8-#-0nI<e^4-^h46`>SQ4ubknNGrx9{gS?D+L
z{>orHU|}c|q~6;dcU2Mn{q?-EY%xmt&F*eHfkwO;HJ@yp-4uFqwNWJ4u|$30KnaY-
zf-9z!6e`QkSx_p2;Q>X@pgUS>Wa}?;y)-)Te$3~8i(;5==UQ;0mBjbXGvE2dl|{cY
zT=8_;Cq$D>^kMjJ6Eoqt4Gx71X8h2_(sRR{b7LK&Yk#6D@rt0UW5bS)q=1?|`NxL~
zB&T3Ywk1_^>IgybGZv5E!G-p7RZ(z!ZjPGU5qSjyT?lx&gSz-tLnTnnhza#0C8kTw
zIKh~nH(Kkbue9<8N<=Ie(oz-xb79*PZ?6+3tiO5d`NLP|XRfU8yI+{}(@%vTDWMC(
zurk_=;5%Q>nlSp3eI|NWrc1y5TdW-~5CbsyyeB~L!ZVL6^SHuRNLO&@DdUUvI+H5l
zyYm|4uZB`j2ote1JTi^2%%n=<8~Tu=v0b!D@;vzDq2mlHrnCb_XEsIG<ec%*<$&VF
z6|A8u{%xzudh+W0>gMLCVTE?5MJo++XCb_E>DkcC#8K&OXu0UQaA81g4TW}sq;-g7
zl1^!p3(*{{Q;jKZl=`h_{~Sk}J}2Cxk<;s*E76;4TV9aBPegvsrMv(9O>pX2CdKn<
zkJJYRt-&503V))NaUZ@%o|Fyj`yedOJp^niSO1JkmHEZMNzq?^pebO_wNOaPl2WtM
z&#6ShqkrU|+bPE9J-SlFTP*y7!U1dTxXQ)lg~j(xjj^G;<vQu;<>_MGLJKj!+aYne
zBJaL(=8Gm{w3wIQJOH#RFj-l{Ve==~&1S#*;((7Vwv48Ha^e2@XVwXw&H2jgyLMUH
z4dTLUrP!$Vzv*Bk(nygjlSkF%3)lS_;?J-kvvj@M4W?A$ZnnFWj0GV-bU%M}PV?9{
znY*Dm5^Mt!&Chsw{-Te6fBp|Qv@Dv6M&d`o`nWRzzi-uo`gmkNZ&w9ek{>P~hZxB!
zqdrbwtnM&-Vo~|{BJpafK@&=*7z@~vHj@QQAYEJoPYW<Q95L&AOQGX~iZC`z>F0>q
z1BCN6(*ZSmz`9^WJVR(*DFI)v+Y{iUfPXUYDSmS{RK4R6B@sAm-%O=LDQ#qG`n7QZ
z_+DCh%R;-U!?MXL8wdXAWIF;^7#<!TAU`OP%DU}X6c|p@gxsQn-}ONm48%y!KO!%v
z%-BqJ7fwwS<_%BE`WRj1O>0nL*kr$>eq82twtw6>a4_$=I+Q*HL<eT<>_E4Cw)j!*
z)=ZD@KAUe_^86tyoyQ*r#B&}9U{<9U=WVFZ(rhkTN97Auh0yyXIC#TfA_yP=q~w7>
zTI(0C?5i9W27xHWbd?=-U<=?=P*PE~-i=s~HcGMi2l+RpA0y?Hd`XBG%!72WHk3`m
z(gE!R0Gi_S=yps04EFF1|DF&kS%6j9Esfh2-%r8VAqbYfC8s*l9PCB=(%3uz%5@JF
zf%KK@?4s?_#sna+V!gE~stvXP(~P&82En{X`-VL^fM1NQgA02L2^QI|2@oApL~vgm
z2O)fX<M!o%O*3ohU%cw*1W#eAib9ZLD(ILJI$UbV&Kc5N>~hGco}{=ctOPC%5<ZLq
z*4hjKK?C~Jjw8>XTS{6??H5Kr$Z+$I;#|*Bc7LNBtzrJF7~dlcX{^KkPyv~{4R(|;
zke8!a%I7>b);hSZZaOLyy_g<a1yfKL`#6hjxsQ3fc58>cqX<zP-1}ell;QeOfZ3D{
z8_?oq82tt1U@g@w+RWV?p?L7Qy3FB>X{Q}*K!)Q}iRDOU6}hzpoR=q9p3WUOs?v3e
z$krM2tI}Gj)W0O@i-cCm{Poeho8ynVN{m`j5E;_xsUQ;*)77%AaKYV8fk$17DV{G9
zCK@K^2Dc}zg};VpTGlrK_Dyt#4wDT)4O5&)740mh3ajV)z_mXpYC7L=l^7H<0Mm(4
zcL4<^=jU4U)oe3>tJ_P=@$x}15bh6Z>`oDxp-W3kR9ZX&Yh%vdEQ+H(<ZdrYr@RD|
zHOL?8^&l`~pNp+-K!KUP$2oLRb&;<#t8{w|SIMWoLa!9bhCJsFPTMG&B%vc&+G93a
zZ=JFXVIxffr=Gqgbo=uM^f!C-G`^R6TYOjvMQo7}Ts%BPP=o|rZS-D!^F=)+tY)*U
zzvg@+AEAtZ3}{b$%4gp##M=>4QwSNa$?k;(bDelf*i0{xn(7BjR=F3@&hoS)F7d+*
z0sx*ayW$nK)(uX3yhId)_Le)3gtX%RsUrOP<WM4LC5L<N8r<bf5m!k*wUgC%;MJAc
z>8Qtj`SQ#o(9ns+NP}m9t`zc;x2fd*pQIQcS#?os8M$~fP(%RmS#hM2M#G7o_tC_b
zXSC2NrDl;{FJdfy9yA{qSNh9EOaw!TP@0;Ze{FiR9-!M&ni|gdQOkb9et=~oN!!+`
zILECGm*m2YbSgM!Fm}SvG9(p{U0J+hwbiKpo0ZVK7$O6^#^7RR3do79?UqqAf8HL)
zp~EGWm8F@=02Bd1i<yNB<;F+#B>lSm=oES_AOqVwu5E4Ur)*DsGmp?Du00%mc2>A>
z{$P~vx5Le`12{zlUW=%_;+RJj?;^OH5<uOoSnSgLZs#KDyVe=ZtoC9MCk#-k3YN<S
z=5f{DTwmw_6<jlDtT3j}_5|eL@Wwa*ZpokYDtC6vq=1DB%Z^P`;|CwpAvm>t5mc??
z+uPOmHvt<$R{d4~gX%?$$$jg|?fv;`En)#J3W08~PBJZ|SHN&W0KcZA18q<UQ9J{@
z^fD?fCV@`l@&Ul~CU+W`(GEagSk!I<?h3z9Fbexh&7L_|o()5UN3+zBJ>E=Fo%EEk
zVr{}JI>09B{_rhhG|}_2#T$kj0m@6fL!uw<Japssb-spY7@~HRt|Zwswa0-0*q`+F
zb@DiNLwrv|8g@Immups{Rsu;>T7;%B^nny9kc>`X6I)j-FlekRj6fJsV4H?Ytj1WY
z?0@#1AHjuwy?;VUO}$dmj=v&tb0Fb<T%JC#6he=$glQHqD+Vfgvx|AT4jV4bww`x*
z+lyAO%H_$#PZtA#QXeYPT<N-iX0cv(<H>ygHD^3YZD`(i$JV#0*K>ORf_U`2A~IH8
zgoE3Q*7cR=#uK0t39P7}U=*O8eRX{F1<P8v1mjL-DxgA^arGXgAZ#Q;na0l8yb#bJ
zNBPXarr!#E;)g&sc*A-942mB!l~~KY;-fVma5vSF!+I+im6(*os2E!%QgC(&cJBmv
zuv*B=C=We(swDUGquNz>aVFA=$>sA@TSppz+%$?ZfnP?oFEJYbBXvutsqK{(GZ?CM
zwQdLHa}NW@W6sm5e(tqHUeypED;J<gTWWGCiLf`wbPWuY+7*`SBHN9LbdD2Ht-TBa
z?@+|G9r>K|w*ediv6}Z)-Erh;9MR@I?d!J$iW-k~5t^<D(qa@iWGcez%qoIQXyjbl
zNW=EkBjYd1yXggN?oz{U3&3Zq$wFtpJ|Nh5Ak_fpwC<-pwci8XiX2)h3E|15&%a4T
zcNdvbu0FEGPzXJ7L<8HI_uuUd0qnI!s8@iET@Zp~nNY_L@XqD8m~@MNsQ4|q{Y3IK
ztw4cFVZL8MZ`@UczjLOjivT6XZ8;2V5s$8%cdVyYb>(GT?Vdxx1|ZddG5#|!G$7}v
zfuvn-uIxDng+hRnMIabvpY$im!`#!x)}3dH3iy<P8kU3a#Vlgl-Dg^TLBS|gzr5`-
zoT`T^`Ghqf^G~3Z$f1?^5g2fIr-xMm7-vxKsH{jTDqa=EXcxN0n-gKuIqj*0C6lj4
zWGNhvpADKGtSVn_!*Bcujjk_=$iLIC-&A4xv+#(;wd(eR^ja3_dY}6y76EXWo0^)+
zJy>j9n{ygM&kD#zfInj@B6vC6zv_=?Y|em=l-B(9aW${7aGxSuY6p0twIG%QFpxx8
zmMtVMs0M`q&@j98m}1I%%f=7}>b!bxOF@xQQDwk!0UFt?FotFJi}9yeM9l{x%|oDQ
zps*Y$+;G6tZB6x0Tf#-cs&!-Qa<AutLC{nKTmK_3C`7c$A3@Sgaoo5sxTanEnx(bn
zXv%P*sCL3xq0@2B)f@c4P*C0*M>BjD&t(kCN)GZU0xRd4?A9XL05^_W5%~2qHgL#m
zl~>y)Yz53={v(Sh|9;*or~_>+I!#_eP^cr=r4aDdK=5=Qfi=fcr5Y()x^&aNL7aAc
z;Xj{+Yc@h`2Y$4%hks5Vao;J8S&4c_UAvlOqkS|=#w*+bOy0(uIDRc90a-JWI=mtw
z&7rwB^<*f26Wo>Wl$O>3K9X@+M4$A1gdA?hszEQ0#ctG_rerpWS~_g)sfR3+O3}|3
z8r*q(_b7Z@nW9BeT0S*j{;mqJ*<G$`y$FCi@E<+N9$OL`a>pNXA9MmkK0p+>e>UWs
z2K?aw5<*xHL~$UZ(kU~=9`e0Da0hise5DXm<)3?HpBf_(HQ{!A5!>u+aav1d5g-xn
zHu@or@2Z35PKD%O#(R&r#Vh4()_c}nE68@Oe2*TrsWlTi4aQ_yiRtp>MkK?cgo1Z`
z);vDr7vhWVN|tfiJV^sFS!^JvZ@mIbkLqHf4KDEqEZVFyoJ`vW#1cTsyqfm@k3dP~
zT|RMf6`Jk>phZ8XDv=6!pRBB`Ablc5M^rR_0aXKHr0i7JyPR4@XhvRMDTs>05qzo|
zM)U-03mP(@9fcrx(f~>wf<gJqecsiPpXN$>Qb*wOP@w2)(B%iu&vnf+kG(&+xev{A
z#)5T~Yt~Y1s;j^3MgOI6zj5!rxlFgM*~)IX@V(*(8bOcc>Gn+Vg??d2M&TdqPRiG8
z9jCu;$9r7zYDwB>=fo%aOSJgi&yAK2Z?gC|b6k0{=-!Eg6Z~?1pM3r>>oM~GRoIt@
zL-oIZ+n18Pk9}8&LX$0fg(&2kB|?^rvNPG4NY+G3)<IIpS|<BWB@9zB_I(@1Zj8Z%
z-+R>edVbI6`Qve2T*tX)&biKc-|u^Q-LLz;KLR+UtY+p!g2w&>kc3lr94~3z+%P<a
zZ)JgdRihPH<ywB~X=b^2{c8BH-?91QkqG5#T0{QmLoX&fjO^22)6;&d#R|rEI)Ku^
zMMjs-)(?h8gO+69yuOP>gY}Gk%iwxm+Pox<EzQ!^vc^jU2s$P1ts8!!ha4v6_AP<P
z1sybiywe!u8ys@%RRc*Un=WN6U^j~Ex8)mVQtX1ku%rJt=K%jR0DP?z+|8?Z6J)ht
zli%{FYi~J^+0ZWV&QXR+<Na{1h>n|P{>jramPwAH^as7gAj_er%Xr`|ezm<F5c~mY
zALN?spl~em*<ilh=|8rh{BX%bIDnkL(>$C#Y7GY0gMYE{cZa&TsL~4t2&0t`l@D>{
zhecUeJ|W9W#^vuom~7uz3<?vfb|xP`?$sczRF%1A3zaAUfT^QYDeHbcOh%+XH>L6x
z-ALt^1*<o-|JBy|hW<ZRd`l>Rl>_`p8LQ7Rxo~GQ%&PbNYYQKge8)%cS>2R!?(+KF
z5W8`l0$J{4x_1F?csWmNYl5tJU*H}Z2lGD6xCjK(%nYbLb@FJUGy3enXCZGO9$sE5
z+!I&up5qe}$zQPCaI*A4s}nyvvQYM3UM}%_WRHYnzCGtlPfKgzZkO-iZzty<Kt)|$
z>lFqB7$BK=j-B4Qlh3hmD@fd7OusV@5vRrb^=ZR>z?Cd&RvcJdl*~$E`}XbI9k8xh
z;%5vTx+>ff{Xaf%Yg72E6@z76r#p=`5bZ3FMW;TUrMeo-Ha=C?Zpd1x8vbj=sul2}
zX%Po|3*HdAM$5Z3A5=2Dm$XK|R*qtwWoRVh!VH<hwKfDJ{=mU(duRBV%rg>-S{)t$
z@-<-0lfC%|oJkHbuQ+p@L$Ioo(>!RkY@i-qot8HL2<_qB@+dz)kYTxMXm`JK9%K?c
z%KuRVj?YL^IGA>|P`6w*3pk3PnxIz=kMb+AV+c3QRNQO<$&W^75cWf=$f0o$!;D>B
zjB8vKG^^~jv_*vr<{UXaKpsX0swYO)?ZGEk?|f~&;`uRPdxB$h|J?V6bDxsh`C{U~
znumjo`bD)XW@a)rb;(I!R8_T|NM5@bicf&r6^J*`1Z(~qIgv5nk6hQ+9Or7*NeRu(
z_!?xrzute|nOd^Tv>O*?j#RJ}L(g;Ru;1}`b^Sn0(we5!hVR0OckaA2?z1$7VO3f-
zpC9apUGicrFme^CdXryfN{GEh%~f$f{rLl(3R9IwbGZAr&*%tt29IPF+&fB(&a|0G
z@>kyT=p;3SwXdRjdr@oC(Y2HC1>yn7yMB+av-+?B^_Y75g{Y3YEcVfWo?u<yr&O&F
z8n(Jkn*r~HP7iL^L2G$BG21;EjtU1-Lup9Yda+81uD8NWp%+`)<<JoohcVE?yMKv;
zb94}OgW|xklEm3GIVPZW#TmHwDZlPBpo^;>gCb0o!)eVPFkP<8y}Bt<Tt5KJUo_y<
ziLdy4J|Zb!s?pYSX2?*o60|qGA=~(6sjq;Ci7s9@&o7}wm|+(X$2&NT)mn#<&MHCM
zl0tjm^%&Y;_yc6hy*3_1uCO(uAVj{DJ$IGU=}g~<)-W}+iZ6~6wD?oT)&M&?`l58W
zDji$=b8LDj8TY-LN|uQ`@te5zwU^=u`26W_v+_#gk2v&z-rR3ji8`QcKMZRqx%|m6
zZD6H*Z|o0FuWK`VL~wx@pX2mljuS0UQ9aF*0oV0U827lh;!+oWyDxf??)G@*?~KqW
zdu91VbART^tQ7T?bKUG{qO&(|DdS(w;33x|neU-j#P@l(37va2IxRDEc405@8<(uS
z^_@4|iAD7^?P1P2p(!lzjR=_w7ubW$oQoujJ{Bx#`QBjl>-Nx6h}{HGeIhmxbZS#W
z&!W;HtDE;iHg$S)G>x;d)(rtkdP1SeSEHE?AED~9Y5-rZ{i8ZGul9yK(^aFa%Q~Do
z;IdoqWkX}@*PHkziTZtyW{kedF^%FqG%_!C1MfW^&xBX6sA%cfxUr$02omG4t$rkV
zq|+BXp8v8B2fkeer9Ex%ZFVz%Uj1)DvvoliJJ!+$%1YHN1-WC_Yd@hoAMN2ApD#Ng
zpzlj)Pv!`%UUSX&Ok19ar?*Jqx!IR28F#kVwf>3R7nKclFXb&6v5i^Rd3dnVUvqyd
ze{}PqOPP`rL%3h<;IcjRn#r|vbFoxFv%W9YJJtW;Hm7C6Ycs|snQqDDy3N_mn)8)x
zG}?FJa&d3{J`dQ**}1zv7iIuyt%B^JI1mzhm0S?xrlEY8;Q$4#G{Gp=5Za-|A>W9a
z1>c_a{A4@-w{uXEFhc<F=E=V!!X=C|6*8IJd$g>3l(U=MvP_cRcc%JIitXFqoFct@
z?|3+5Vvzb%h>4pydYcq#ZSpczM>Po|`TaP^s}<=FO=<AjZA!wP(Y{rhw`Sm`_&6nf
z^?^~b2lv2=CvtubVP$AHaK6-*Q-|+?@4ae)_tK-oNCWyN9ys?EMkhw(rOK4Qd9r2H
z_Wc&ay;zSI^9Yc(M#YQkpS`Tmv<^X?d@0(|9jZ1NJ=C6VoShKc$bfD^h$$r|Bg73B
zBh^o&f^H(8xC{}`Ch-yp<M9&)3&9+DwHZ0k%wGm4PFOF=4SvvE&#=8M!1|gwrzk%-
zEb(<zVstluE-k@Tu`7gIcaeVefsAv)MOjL{`}Q_`P5TWe%EiwY85KC2>9@<p1N(n6
z<VlSlr+Mzw&qFr4PW}ktHdsO%e<TlRt2jSa3j>kdOYv2ba`29fHtWs!3xBY!1Fq-z
z{iNu1DsHFUo5sRspG6EPUi`yBdQi2Eb59>qbd)F=zxQd$hsf<K5Ely@aswU07Q*Bg
zgk8)BgdzvUtrlALFtN$BCLXuuKK5*^^}PLqiO*d|4qID}=RoKF!%Blfqp3f)a+&6u
z|3k-_WXu*c#h1*C)wVx#VOxqCup+;!ceRZ&<XfAu%iRp((6)kUB<v@&Rn+yNRiag*
zvHZ4{GpA{71zhF>nRUXYPi(0bJfJE(4;5V-$X`Fa1anmo`()?nAzKc&sUEWnF6$}0
zfTpzNE27ob;*O%RHyyWm7$u}fqn{J-LxOjQk)+-r*An;SM+g}7c(zL-p~I2FqiGNN
z>(cm+O~xN5P%osY?ny3MIqcq^YIE2+Dc*eWdQfBU^M`9L!L=eabtET`Eq}aqEH%1-
zsSw8jhVVYreZD$<>jI(V7~U$o+IMCxAN!Lxu>(jX*r2EF6TnqMC1)5O!?VTX7}at_
z(LCj)ubds-)XJ_DT>AQ6DljE!KJc{0X;Y-QW17uy;D_F?I5ta6zSELqp#fojj?<}C
zQUJimz!ecp1R$osSIkNQ30_ZWejBO5@T6)*7uh60Yo01cGm_-c<J>>q%#!Awd^Ns1
zCY!$Ths5Ww8m_Cq#Q1j@qp4?VRc59s)L5EjHoIJMmY0{AqOT%g;`&rSQZE0YIt|ZI
zv79VAY}Y)jeEeMcM#-Zo(yfJ$5nG{3h$a4VsW&E=v*LMQrcJ1_g(_0j1^;>>gSj4I
z7=cPmmQpcKDQA=jAi9dxfJ~sKlCh_<-Y=n-;*7cXFb*ui<a)Yt=8wm9n7`gKe!d7=
zvUGJi8bctzF1bhKoi**kBV@V(BOq%|8Dx!YV7)!T2t@%G!y`9ejh@W{EDo1lKA9rO
z1hnM#Oqwv0Jh{kjJDPg`K%>)63$Z=2*Vtt;67^78TAE+Q=V^DUY&8kPfA@HGkrD;x
z?yjMx?Um8fO4=zx)+VYmx`O4HczxGrEuOz3_3`m?3{o3%YWyS!=Kt(`s6C5Ya_<fB
z;v4&y3Ceye-1giroy-i0@MdENwouu#JEB3_Ln-TdGMfkdSlzH_SQM5vz{x!NinoHc
z_b0Y@zUU8pRkUD^W?-t5WwawL(3GbtGfYg!P0S5%^@oOXj45=)%+}Z_3Ap%niz$*;
zvhR@)N|s%63{*36B?hbmIrhi3vixVZ-wN8_5vbla7|@f_OW?Qn_D%vdHC<gNKrehP
zg#vT`Ul0C7`=<n>+I<(^ZAo>VtB^-Zzs!Pm8lInID8t;9!atFW@$Q`-^lUh%-Yg&F
zS}ysg$3|_hwS1zPz{I6iQxi@Q6>M!?)Mf=N#b{;$mx$K@`qh(yPseO9od2Sf^Mgey
z;w2*a0e6)cTQ;VJf`U~~|C+XGF~FCFs#{+;ii+c+j161*)h16hWZWtayx@6**W;;G
z&t`r%iy|qTxvI{IArrQ==D82cWKQd<d+JdA&K^>pa*y8Cgp(~=-h`W-UZhdDjaa<*
zwZiUH;R2eMiSDUAVOSyV?X5qA?2C>iG?U9u<(aAQ5-Ccd<1EC_J72HCt2;?MAwv9S
zfx757g|?1#6M`1`x6lUvI}rrU*8NlbHKwJ87K;;2Nli@$P5`iH{F-n?g-qZiX@b6G
z50Bkz$qCWyNDy#T<_8Ra+Ol)V)a{Yl+^&nBdaAy9n&~`&R-a*}t>3PngShI&E96zy
zy+5jh-Sr*xO%3<sw8&FESFCDmW|_j#v0S#E!t1s(tUkDN$8TP~j<llVeK$^SE5fDU
zoU?LWR{B5Y78ZdAgh3VfZ)N?LrtkT~fz2!mmH8vLc~beeQZQ3u6#Sezu|jV0f{VOj
z86FIo4SE(SX;}A=2j7}bS^n*O;*+&Rr|{gv?iE%A^!ng^zM+;_S*bt<$eQq3>l6Sh
zmknEI`@**Q(U`yEexMPnp7NxLa>w-C^?Kvx3rmB^0+)bM8~GL9<87b!G-sW}5srDp
z({lDRF6E?K7rSEXukA7W@r%*pFQ1L0?_b`<2|8ZB<Vsa*%Wb&Up2Jj$fYD?pnkjG<
zPxWTBT)+-}azv>5Cw%>?v(`woC1ZBBp#(DuBZPA9R>bksLB3NyOkt1w`AyAI_wJL1
zMjnr1fw350h5`k=GX@5&g*Huvu1uhA7GfS<YL|0U&D1ejEYm#>D5c2RHWNMyg<O4(
zt+e4U3`Ne_fl&ZdNF^gW6d}OJZ95r~t+98>gwuO>!@e*=5{<uObFcWUVOfNb>q7oH
zr6;+l*Mn+kIi~RE6fQyV(gosnOzBeX$Fj2wWfSky1#2<twy^k^H>YMBTFN6zpTFoP
z9=N<qu^<?}6$Z$=ny}&RfnV~VI9)a8_nZU|GpH$QBWEBeY=c)7fTv`cLixhUQq_5I
zWP}$u*$#()+R+TFnuJE&&0CF3t4u5($JZYELW+7OJr4o{Yzg%r{5IRwOAULj6=G<t
zTwPt6=%~YDe~FB){x}AV_-V-MHM6euh6(3(kUB7CAfZ4Y_)^%y_nZXbyY?EyS~R8C
zppC!x{`N!NJT`4kAif|EnR~~ClTWPGAUfSu;)q~PI102^tyl28jG`L(Y2CN%zjbu9
z9aHHl@ECcJN!?(S-WHUqgB=?&v3sQZ2-9ktVwO8YP638YV#z~`T<?*1wMa@-BkWN?
z*@8HW?xEqX#PN~_cZ}>V;6gJ?vl~t5UG^U76}+}MMX)|tTA6p2=g3WgC0sI8d9=t%
zapT4f;3CYD;K9TX$JtF&5DRE2SbfKHCwAPoj7pd1W&)805MV&iM18M_e<fkO1pp*U
zFizYb-?1LV6=?Dqe_tWo{PtON;h`}RH*xc}-w!~LNSNUjduNB8HubkHTWvb*VQ3i%
z4dBZ)N0vSHTazhq(`&75I8pILxaBma35`%`2ine9V^93v(f(jK&NJDG$AnX!_c`!g
zBO_@MurJ=Ikr69CjX!{$pi4cwuNzW_>1UZAeZjk$Vcle3f>qGyNxI3K5H=UeF`w@9
zC1qtZB#memA&+iYvf=Y%d1a$4V$%fcaE2~IxNsr#1Ql4Dz-)-7W+E&B7^=Q?q6+ZQ
zSTuHpDRTDv3wD=s4)tf0vn5IHP|iRN6$>*<G-m!R^}DQuewRlP?-zbo2f<a#PP?p%
z3oi@S^OQP4le0*i#T)K<k8;zsEqi7ObFYlIKv2FFA@tkMp;?>tV^dRaTN{;SiZ_7%
z6(vtT35$%3bZ@jcHX3^(K5puBixjk0Lq;bqtQ}x4LUISvA<kd{-b!BY<3!`Z0sv(B
zFHv8yYnsx%29qr82ESla_Y@n*uMv`V$q>YiB_$;k8@scKcvPpS7l28<wiiuaisw$=
z7<u!R7+O)1f$f~_Vl(a2Mw{2+gWU6j@gvaOKBI`j;^7jMeS38$K5Zg@?`DQ$VEAOp
zLlCIU&pUxzeCg^kP_<eE7(0bsXlRzWJ{VV3QeIwt`W$d4nA%;Kkr<47BmMES1ajCQ
zrGGNIncgg$S7fYUw%*eYBUk&yyp;r!1za=G-5@mdAmf$hF|(Fm-0=Ia<Yrf@RFJWT
z0Zs7_7Z=#e6d;&mD_kgdStws6lEgd1Byns}^4hG^(PmOtEBLXBR1(gZQ_swP+>TZs
zE!oA6tSUVYcg>mj%3n7auv)caV@f&kNNYg}=DY_d+)lm6^dwA}0o3yAkN)RL-p`oY
zS{2*C`PY8gC1-D<$bJuqMO~#$L9KB@%3mk__JJNeOxMv+YN?5oFMa@YleA_I^jUGL
zWC*P1rj=>R8S;F^)C)Ea_V(f6HAS`CtGfItq)==)SoOp<0#Z{)q7c>w>=o!YfZTuS
zd?f+U0Sv|rj9G44uNJupIz%BfLNmIhyU7C%<dlq?C&7qYKx1&Sd^vcAfiF_Jn2_ZY
z-kRwpJ5lmfB~p4+r7@$nU9ITl6TZAYoKHL<Mj79zSZHNgivktgV9&s?z22ZOIQEve
z95+B8L6yMI2Vbz7M}g#eRRE!PS5)=q9q;#*_s_?2m&V;NS5+9OE9+X#ZSoxYSiUGK
z4yAH?L9di@NEl5a#Prz*1_vc>WV-#w^&Z)h&%N0S%>nfb5i;EXq#R#|r%J}{;w+lu
zfUB)57BcDV<U~u3t3TV&gWF4|rs}tG)4>9`hn3G5L%J&7NyaJuc({;q&f4<zLUFs0
zg{|s~k#nCjE;!d<sz#;G*JRq2WN}Gv7Zd5YnJrSxYRq^S8tP;mZE3CSwPM@nF|MJL
z&LFm(_WE86!`6iCT{PkR)X*@d(Og#-mk!hh#Udd{4J7!|ELDFQ9lIVocdDiBIhxBF
z4^DksZt#wfo-AHg8|K{n*o&=<{*cqtHl76aW4=Q>F|-b2t4$Elald|FsXywPpB^|M
z_9y%~@c-`_{|OvI)vxq9RJ5N;4tcbSApJbhU`bkjJ@VrFTH&eKl?3v^2Z6r_|2;GD
z<G$Ikg8uvG;=jA0p)%j!T2|U2(OfHoxm*>k&r=kyP$eG}`29Wj@0nmb=)(Win~iO_
zOXm5%&h$4$TS0f^A;H5aqs(F1RpMR@N)%Gxj)J36wwI8enZxM8<;1jMMhK~WpEw=V
zm_FQGG>Q>Q8>ZhO3dEr3Ao!k@<n#k<(4ISV=jOP1lLOSGQKKx?co02Vn}a=nQ5v=K
zB3<19deniky@;Hx`zO{dC;i~cLG##2=#DCMps(`!eOKttU8vsx%6742bkny3rD5WE
z6b;>pJqVB$B{@Cw+csW4!XCDu4)f9laB9TiI)O$vDA9Ld8MLK%q3C)w3k&~=LZl0n
zruv>|AEqYpx}>MG{N~-jh7l<qccLbcIau+Fc|KEIS+I3V4VvH9<K}S2fHoDP(Nt^@
zPmv;_cOP3$+1x}dkh%B@E*S6u`P(?`o^Nj>^&Kh=oARSeA5F`TFhQzp;g1=AD`CfM
zT1bT>ugY{pN9~~1l#fsDnqbEgDKs{`ADw5f&5)%M%CxG^TTVz9pcJGz<n&_Mu;X|d
z%g=gmuFLNc)h-)$EI53fB8RrH*?CuaLjpJWJ9Gb=R_J31`eClxic|Jbe+nU5+trc*
z@M>6Ah}TPl!}@w_HGO*Dx1@&C@9ZJ2K@(tW@vybw{mqM*)R=0M%GAj+C6fD;Y=5E<
z#bH))2(|KkX!Y9W!)GW?NWH+W{*=<P6E&PE;N8Uix>E}7%WP8}%d0xfGLX9zhfEu^
z6j*B<qCN&Zkt!QSY&xx$ka7h0=dMrhf$UUUKQ7DGe~sEm8GAt)n^M?>YM$xVPvXNs
zSHG(eH4(7D#@$m>JNui|EvZ>BN9>q%%fq7ewe5o}{yTTApS`JDT3n|pLmCZ=CaHg%
z%pGuVad6(N49bde%e|nEhfncL{evA#-~TWeeizr_a#z%1pUd>fD}}IS*z(oLxop&E
zMA209@rMc9IpJVFFXr89Umo=2B_t(Tl%B!XNG+Oqb(usaOyxZBCwOB8enI0u3D@Uv
z8nnL$_dks!eOF8yrfwl-!O({4@4<GV-%;&nPZE$rifjmlHLoJ-$3lBmuxSmd*RHuh
zcW$>>V#n6kiM27-9aK})E#<1%36_!ylP$&{mDlH7aO@D4TeM$1Dd`rOZIWz$`s&^H
zN3B>C=BoxgXh!SI1=&MGr&7b4n<ufNKDSo=3kji93B6mZgv)H{4p0%ZH-s0kHUDMI
zQMuH->5<0Wa*I*+^lkeoID{l<_90#<FN1IKmJQX3v!nW>Q8+AD7(O_#O%$@IFr|PU
zgAlgFOhr=3f`{j5k?ySpl@Uz4Hbu*1SyV;#Ehm)eiDt?njj1$gf4d2=Mu3O+YskW#
zeUHVhb9>NdD<CzafZoX-TEDd{zF`Cj&P>-aePyCihXxx2hTG$4qG*^ykg!<=g0vbY
znujzSIhU7@;><K?tZ~VUtSVjzSdMS+_!^@A70wE;nFZ54`CD)ULV)(U7e2hJdE2wR
zK5SkK*4xM|lQCSsY^KTFKo+V|!nq9*$nxOPsDk@-AO=~(OTO4<Y9R$=J0YesabWox
zkVF!<*~1-D7R>ed<{r>b>5ZoXBsL6z99{2LAvVC)pxCi2f<IF^rBEU123Y--@Trl^
z-y6zU9|(!h^!X4y9DHu|KuscvhhRS?OO4JPt_I`{rV!?UA(Y0c-}k0u2hayxQ|Q}<
zxH`Jum_~J*q~&ju)J<;CBL-!i(}K5;4vA2|43{fIiR*oPv8aq@Mm5vN>6?0{gIp{&
z&#skqsu1JT)v45OnBkT<H~3m+7p=rAq#z`LDKwQ{&x*p7w%-uFPp^#}^6_=Td$kr#
zuU?xWq>@Mly6S;m4z0t~HQyZCNw3}mU+h?Q&x(Cf<^`WAECQy}C`8`=zo715M@61j
z93=1jG>@JC-FGd?vU+m<4<-KCV1Ykn`v2;5_V4bT|Bv<fSMyECzq@t*e=k5fdJ+lG
U^lE#|O96iLb&Rf+U9o%eKLi6)vH$=8

literal 0
HcmV?d00001

diff --git a/applications/液晶屏读数识别.md b/applications/液晶屏读数识别.md
new file mode 100644
index 0000000000..4a41afec77
--- /dev/null
+++ b/applications/液晶屏读数识别.md
@@ -0,0 +1,577 @@
+# 项目背景及意义
+目前光学字符识别(OCR)技术在我们的生活当中被广泛使用，但是大多数模型在通用场景下的准确性还有待提高，针对于此我们借助飞桨提供的PaddleOCR套件较容易的实现了在垂类场景下的应用。
+
+该项目以国家质量基础（NQI）为准绳，充分利用大数据、云计算、物联网等高新技术，构建覆盖计量端、实验室端、数据端和硬件端的完整计量解决方案，解决传统计量校准中存在的难题，拓宽计量检测服务体系和服务领域；解决无数传接口或数传接口不统一、不公开的计量设备，以及计量设备所处的环境比较恶劣，不适合人工读取数据。通过OCR技术实现远程计量，引领计量行业向智慧计量转型和发展。
+
+# 项目内容
+本项目基于PaddleOCR开源套件，以PP-OCRv3检测和识别模型为基础，针对液晶屏读数识别场景进行优化。
+
+## 安装环境
+
+```python
+# 首先git官方的PaddleOCR项目，安装需要的依赖
+# 第一次运行打开该注释
+# git clone https://gitee.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR
+pip install -r requirements.txt
+```
+
+## 文字检测
+文本检测的任务是定位出输入图像中的文字区域。近年来学术界关于文本检测的研究非常丰富，一类方法将文本检测视为目标检测中的一个特定场景，基于通用目标检测算法进行改进适配，如TextBoxes[1]基于一阶段目标检测器SSD[2]算法，调整目标框使之适合极端长宽比的文本行，CTPN[3]则是基于Faster RCNN[4]架构改进而来。但是文本检测与目标检测在目标信息以及任务本身上仍存在一些区别，如文本一般长宽比较大，往往呈“条状”，文本行之间可能比较密集，弯曲文本等，因此又衍生了很多专用于文本检测的算法。本项目基于PP-OCRv3算法进行优化。
+
+### PP-OCRv3检测算法介绍
+PP-OCRv3检测模型是对PP-OCRv2中的CML（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。如下图所示，CML的核心思想结合了①传统的Teacher指导Student的标准蒸馏与 ②Students网络之间的DML互学习，可以让Students网络互学习的同时，Teacher网络予以指导。PP-OCRv3分别针对教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+![](https://ai-studio-static-online.cdn.bcebos.com/c306b2f028364805a55494d435ab553a76cf5ae5dd3f4649a948ea9aeaeb28b8)
+
+详细优化策略描述请参考[PP-OCRv3优化策略](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md#2)
+
+### 数据准备
+[计量设备屏幕字符检测数据集](https://aistudio.baidu.com/aistudio/datasetdetail/127845)数据来源于实际项目中各种计量设备的数显屏，以及在网上搜集的一些其他数显屏，包含训练集755张，测试集355张。
+
+```python
+# 在PaddleOCR下创建新的文件夹train_data
+mkdir train_data
+# 下载数据集并解压到指定路径下
+unzip icdar2015.zip  -d train_data
+```
+
+```python
+# 随机查看文字检测数据集图片
+from PIL import Image  
+import matplotlib.pyplot as plt
+import numpy as np
+import os
+
+
+train = './train_data/icdar2015/text_localization/test'
+# 从指定目录中选取一张图片
+def get_one_image(train):
+    plt.figure()
+    files = os.listdir(train)
+    n = len(files)
+    ind = np.random.randint(0,n)
+    img_dir = os.path.join(train,files[ind])  
+    image = Image.open(img_dir)  
+    plt.imshow(image)
+    plt.show()
+    image = image.resize([208, 208])  
+
+get_one_image(train)  
+```
+![det_png](./imgs/det.png)
+
+### 模型训练
+
+#### 下载预训练模型
+下载我们需要的PP-OCRv3检测预训练模型，更多选择请自行选择其他的[文字检测模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md#1-%E6%96%87%E6%9C%AC%E6%A3%80%E6%B5%8B%E6%A8%A1%E5%9E%8B)
+
+```python
+#使用该指令下载需要的预训练模型
+wget -P ./pretrained_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+# 解压预训练模型文件
+tar -xf ./pretrained_models/ch_PP-OCRv3_det_distill_train.tar -C pretrained_models
+```
+
+在训练之前，我们可以直接使用下面命令来评估预训练模型的效果:
+
+```python
+# 评估预训练模型
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model="./pretrained_models/ch_PP-OCRv3_det_distill_train/best_accuracy"
+```
+
+结果如下：
+
+|   | 方案                        |hmeans|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
+
+**1)直接用全量策略finetune训练**
+#### 修改配置文件
+我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
+```
+epoch:100
+save_epoch_step:10
+eval_batch_step:[0, 50]
+save_model_dir: ./output/ch_PP-OCR_v3_det/
+pretrained_model: ./pretrained_models/ch_PP-OCRv3_det_distill_train/best_accuracy
+learning_rate: 0.00025
+num_workers: 0 # 如果单卡训练，建议将Train和Eval的loader部分的num_workers设置为0，否则会出现`/dev/shm insufficient`的报错
+```
+
+#### 开始训练
+使用我们上面修改的配置文件configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml，训练命令如下：
+
+```python
+# 开始训练模型
+python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model=./pretrained_models/ch_PP-OCRv3_det_distill_train/best_accuracy
+```
+
+评估训练好的模型：
+
+```python
+# 评估训练好的模型
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_det/best_accuracy"
+```
+
+结果如下：
+|   | 方案                        |hmeans|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
+| 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
+
+**2)基于预训练模型Finetune student模型**
+
+我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
+```
+epoch:100
+save_epoch_step:10
+eval_batch_step:[0, 50]
+save_model_dir: ./output/ch_PP-OCR_v3_det_student/
+pretrained_model: ./pretrained_models/ch_PP-OCRv3_det_distill_train/student
+learning_rate: 0.00025
+num_workers: 0 # 如果单卡训练，建议将Train和Eval的loader部分的num_workers设置为0，否则会出现`/dev/shm insufficient`的报错
+```
+
+训练命令如下：
+
+```python
+python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o Global.pretrained_model=./pretrained_models/ch_PP-OCRv3_det_distill_train/student
+```
+
+评估训练好的模型：
+
+```python
+# 评估训练好的模型
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_det_student/best_accuracy"
+```
+
+结果如下：
+|   | 方案                        |hmeans|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
+| 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
+| 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
+
+**3)基于预训练模型Finetune teacher模型**
+
+首先需要从提供的预训练模型best_accuracy.pdparams中提取teacher参数，组合成适合dml训练的初始化模型，提取代码如下：
+
+```python
+cd ./pretrained_models/
+# transform teacher params in best_accuracy.pdparams into teacher_dml.paramers
+import paddle
+
+# load pretrained model
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# print(all_params.keys())
+
+# keep teacher params
+t_params = {key[len("Teacher."):]: all_params[key] for key in all_params if "Teacher." in key}
+
+# print(t_params.keys())
+
+s_params = {"Student." + key: t_params[key] for key in t_params}
+s2_params = {"Student2." + key: t_params[key] for key in t_params}
+s_params = {**s_params, **s2_params}
+# print(s_params.keys())
+
+paddle.save(s_params, "ch_PP-OCRv3_det_distill_train/teacher_dml.pdparams")
+
+```
+
+我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
+```
+epoch:100
+save_epoch_step:10
+eval_batch_step:[0, 50]
+save_model_dir: ./output/ch_PP-OCR_v3_det_teacher/
+pretrained_model: ./pretrained_models/ch_PP-OCRv3_det_distill_train/teacher_dml
+learning_rate: 0.00025
+num_workers: 0 # 如果单卡训练，建议将Train和Eval的loader部分的num_workers设置为0，否则会出现`/dev/shm insufficient`的报错
+```
+
+训练命令如下：
+
+```python
+python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml -o Global.pretrained_model=./pretrained_models/ch_PP-OCRv3_det_distill_train/teacher_dml
+```
+
+评估训练好的模型：
+
+```python
+# 评估训练好的模型
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_det_teacher/best_accuracy"
+```
+
+结果如下：
+|   | 方案                        |hmeans|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
+| 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
+| 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
+| 3 | PP-OCRv3中英文超轻量检测预训练模型fintune教师模型 |84.8%|
+
+**4)采用CML蒸馏，进一步提升student模型精度**
+
+需要从2) 、3）训练得到的best_accuracy.pdparams中提取各自代表student和teacher的参数，组合成适合cml训练的初始化模型，提取代码如下：
+
+```python
+# transform teacher params and student parameters into cml model
+import paddle
+
+all_params = paddle.load("./pretrained_models/ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# print(all_params.keys())
+
+t_params = paddle.load("./output/ch_PP-OCR_v3_det_teacher/best_accuracy.pdparams")
+# print(t_params.keys())
+
+s_params = paddle.load("./output/ch_PP-OCR_v3_det_student/best_accuracy.pdparams")
+# print(s_params.keys())
+
+for key in all_params:
+    # teacher is OK
+    if "Teacher." in key:
+        new_key = key.replace("Teacher", "Student")
+        #print("{} >> {}\n".format(key, new_key))
+        assert all_params[key].shape == t_params[new_key].shape
+        all_params[key] = t_params[new_key]
+
+    if "Student." in key:
+        new_key = key.replace("Student.", "")
+        #print("{} >> {}\n".format(key, new_key))
+        assert all_params[key].shape == s_params[new_key].shape
+        all_params[key] = s_params[new_key]
+
+    if "Student2." in key:
+        new_key = key.replace("Student2.", "")
+        print("{} >> {}\n".format(key, new_key))
+        assert all_params[key].shape == s_params[new_key].shape
+        all_params[key] = s_params[new_key]
+
+paddle.save(all_params, "./pretrained_models/ch_PP-OCRv3_det_distill_train/teacher_cml_student.pdparams")
+```
+
+训练命令如下：
+
+```python
+python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model=./pretrained_models/ch_PP-OCRv3_det_distill_train/teacher_cml_student Global.save_model_dir=./output/ch_PP-OCR_v3_det_finetune/
+```
+
+评估训练好的模型：
+
+```python
+# 评估训练好的模型
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_det_finetune/best_accuracy"
+```
+
+结果如下：
+|   | 方案                        |hmeans|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
+| 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
+| 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
+| 3 | PP-OCRv3中英文超轻量检测预训练模型fintune教师模型 |84.8%|
+| 4 | 基于2和3训练好的模型fintune |82.7%|
+
+### 模型导出推理
+训练完成后，可以将训练模型转换成inference模型。inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+#### 模型导出
+导出命令如下：
+
+```python
+# 转化为推理模型
+python tools/export_model.py \
+-c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml \
+-o Global.pretrained_model=./output/ch_PP-OCR_v3_det_finetune/best_accuracy \
+-o Global.save_inference_dir="./inference/det_ppocrv3"
+
+```
+
+#### 模型推理
+导出模型后，可以使用如下命令进行推理预测：
+
+```python
+# 推理预测
+python tools/infer/predict_det.py --image_dir="train_data/icdar2015/text_localization/test/1.jpg" --det_model_dir="./inference/det_ppocrv3/Student"
+```
+
+## 文字识别
+文本识别的任务是识别出图像中的文字内容，一般输入来自于文本检测得到的文本框截取出的图像文字区域。文本识别一般可以根据待识别文本形状分为规则文本识别和不规则文本识别两大类。规则文本主要指印刷字体、扫描文本等，文本大致处在水平线位置；不规则文本往往不在水平位置，存在弯曲、遮挡、模糊等问题。不规则文本场景具有很大的挑战性，也是目前文本识别领域的主要研究方向。本项目基于PP-OCRv3算法进行优化。
+
+### PP-OCRv3识别算法介绍
+PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2205.00159)优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。如下图所示，PP-OCRv3采用了6个优化策略。
+![](https://ai-studio-static-online.cdn.bcebos.com/d4f5344b5b854d50be738671598a89a45689c6704c4d481fb904dd7cf72f2a1a)
+
+优化策略汇总如下：
+* SVTR_LCNet：轻量级文本识别网络
+* GTC：Attention指导CTC训练策略
+* TextConAug：挖掘文字上下文信息的数据增广策略
+* TextRotNet：自监督的预训练模型
+* UDML：联合互学习策略
+* UIM：无标注数据挖掘方案
+
+详细优化策略描述请参考[PP-OCRv3优化策略](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md#3-%E8%AF%86%E5%88%AB%E4%BC%98%E5%8C%96)
+
+### 数据准备
+[计量设备屏幕字符识别数据集](https://aistudio.baidu.com/aistudio/datasetdetail/128714)数据来源于实际项目中各种计量设备的数显屏，以及在网上搜集的一些其他数显屏，包含训练集19912张，测试集4099张。
+
+```python
+# 解压下载的数据集到指定路径下
+unzip ic15_data.zip -d train_data
+```
+
+```python
+# 随机查看文字检测数据集图片
+from PIL import Image  
+import matplotlib.pyplot as plt
+import numpy as np
+import os
+
+train = './train_data/ic15_data/train'
+# 从指定目录中选取一张图片
+def get_one_image(train):
+    plt.figure()
+    files = os.listdir(train)
+    n = len(files)
+    ind = np.random.randint(0,n)
+    img_dir = os.path.join(train,files[ind])  
+    image = Image.open(img_dir)  
+    plt.imshow(image)
+    plt.show()
+    image = image.resize([208, 208])  
+
+get_one_image(train)
+```
+
+![rec_png](./imgs/rec.png)
+
+### 模型训练
+#### 下载预训练模型
+下载我们需要的PP-OCRv3识别预训练模型，更多选择请自行选择其他的[文字识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md#2-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E6%A8%A1%E5%9E%8B)
+
+```python
+# 使用该指令下载需要的预训练模型
+wget -P ./pretrained_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+# 解压预训练模型文件
+tar -xf ./pretrained_models/ch_PP-OCRv3_rec_train.tar -C pretrained_models
+```
+
+#### 修改配置文件
+我们使用configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
+```
+  epoch_num: 100 # 训练epoch数
+  save_model_dir: ./output/ch_PP-OCR_v3_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 100] # 评估间隔，每隔100step评估一次
+  cal_metric_during_train: true
+  pretrained_model: ./pretrained_models/ch_PP-OCRv3_rec_train/best_accuracy  # 预训练模型路径
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  use_space_char: true  # 使用空格
+
+  lr:
+    name: Cosine # 修改学习率衰减策略为Cosine
+    learning_rate: 0.0002 # 修改fine-tune的学习率
+    warmup_epoch: 2 # 修改warmup轮数
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/ # 训练集图片路径
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/ic15_data/rec_gt_train.txt # 训练集标签
+    ratio_list:
+    - 1.0
+  loader:
+    shuffle: true
+    batch_size_per_card: 64
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/ # 测试集图片路径
+    label_file_list:
+    - ./train_data/ic15_data/rec_gt_test.txt # 测试集标签
+    ratio_list:
+    - 1.0
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 64
+    num_workers: 4
+```
+
+在训练之前，我们可以直接使用下面命令来评估预训练模型的效果:
+
+```python
+# 评估预训练模型
+python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model="./pretrained_models/ch_PP-OCRv3_rec_train/best_accuracy"
+```
+
+结果如下：
+|   | 方案                        |accuracy|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量识别预训练模型直接预测 |70.4%|
+
+#### 开始训练
+我们使用上面修改好的配置文件configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml，预训练模型，数据集路径，学习率，训练轮数等都已经设置完毕后，可以使用下面命令开始训练。
+
+```python
+# 开始训练识别模型
+python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml
+```
+
+训练完成后，可以对训练模型中最好的进行测试，评估命令如下：
+
+```python
+# 评估finetune效果
+python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.checkpoints="./output/ch_PP-OCR_v3_rec/best_accuracy"
+```
+
+结果如下：
+|   | 方案                        |accuracy|
+|---|---------------------------|---|
+| 0 | PP-OCRv3中英文超轻量识别预训练模型直接预测 |70.4%|
+| 1 | PP-OCRv3中英文超轻量识别预训练模型finetune |82.2%|
+
+### 模型导出推理
+训练完成后，可以将训练模型转换成inference模型。inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+#### 模型导出
+导出命令如下：
+
+```python
+# 转化为推理模型
+python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_rec/best_accuracy" Global.save_inference_dir="./inference/rec_ppocrv3/"
+```
+
+#### 模型推理
+导出模型后，可以使用如下命令进行推理预测
+
+```python
+# 推理预测
+python tools/infer/predict_rec.py --image_dir="train_data/ic15_data/test/1_crop_0.jpg" --rec_model_dir="./inference/rec_ppocrv3/Student"
+```
+
+## 系统串联
+我们将上面训练好的检测和识别模型进行系统串联测试，命令如下：
+
+```python
+#串联测试
+python3 tools/infer/predict_system.py --image_dir="./train_data/icdar2015/text_localization/test/142.jpg" --det_model_dir="./inference/det_ppocrv3/Student"  --rec_model_dir="./inference/rec_ppocrv3/Student"
+```
+
+测试结果保存在`./inference_results/`目录下，可以用下面代码进行可视化
+
+```python
+%cd /home/aistudio/PaddleOCR
+# 显示结果
+import matplotlib.pyplot as plt
+from PIL import Image
+img_path= "./inference_results/142.jpg"
+img = Image.open(img_path)
+plt.figure("test_img", figsize=(30,30))
+plt.imshow(img)
+plt.show()
+```
+
+![sys_res_png](./imgs/sys_res.png)
+
+### 后处理
+如果需要获取key-value信息，可以基于启发式的规则，将识别结果与关键字库进行匹配；如果匹配上了，则取该字段为key, 后面一个字段为value。
+
+```python
+def postprocess(rec_res):
+    keys = ["型号", "厂家", "版本号", "检定校准分类", "计量器具编号", "烟尘流量",
+            "累积体积", "烟气温度", "动压", "静压", "时间", "试验台编号", "预测流速",
+            "全压", "烟温", "流速", "工况流量", "标杆流量", "烟尘直读嘴", "烟尘采样嘴",
+            "大气压", "计前温度", "计前压力", "干球温度", "湿球温度", "流量", "含湿量"]
+    key_value = []
+    if len(rec_res) > 1:
+        for i in range(len(rec_res) - 1):
+            rec_str, _ = rec_res[i]
+            for key in keys:
+                if rec_str in key:
+                    key_value.append([rec_str, rec_res[i + 1][0]])
+                    break
+    return key_value
+key_value = postprocess(filter_rec_res)
+```
+
+## PaddleServing部署
+首先需要安装PaddleServing部署相关的环境
+
+```python
+python -m pip install paddle-serving-server-gpu
+python -m pip install paddle_serving_client
+python -m pip install paddle-serving-app
+```
+
+###  转化检测模型
+
+```python
+cd deploy/pdserving/
+python -m paddle_serving_client.convert --dirname ../../inference/det_ppocrv3/Student/  \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_det_v3_serving/ \
+                                         --serving_client ./ppocr_det_v3_client/
+```
+
+### 转化识别模型
+
+```python
+python -m paddle_serving_client.convert --dirname ../../inference/rec_ppocrv3/Student \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_rec_v3_serving/ \
+                                         --serving_client ./ppocr_rec_v3_client/
+```
+
+
+### 启动服务
+首先可以将后处理代码加入到web_service.py中，具体修改如下：
+```
+# 代码153行后面增加下面代码
+def _postprocess(rec_res):
+    keys = ["型号", "厂家", "版本号", "检定校准分类", "计量器具编号", "烟尘流量",
+            "累积体积", "烟气温度", "动压", "静压", "时间", "试验台编号", "预测流速",
+            "全压", "烟温", "流速", "工况流量", "标杆流量", "烟尘直读嘴", "烟尘采样嘴",
+            "大气压", "计前温度", "计前压力", "干球温度", "湿球温度", "流量", "含湿量"]
+    key_value = []
+    if len(rec_res) > 1:
+        for i in range(len(rec_res) - 1):
+            rec_str, _ = rec_res[i]
+            for key in keys:
+                if rec_str in key:
+                    key_value.append([rec_str, rec_res[i + 1][0]])
+                    break
+    return key_value
+key_value = _postprocess(rec_list)
+res = {"result": str(key_value)}
+# res = {"result": str(result_list)}
+```
+
+启动服务端
+```python
+python web_service.py 2>&1 >log.txt
+```
+
+### 发送请求
+然后再开启一个新的终端，运行下面的客户端代码
+
+```python
+python pipeline_http_client.py --image_dir ../../train_data/icdar2015/text_localization/test/142.jpg
+```
+
+可以获取到最终的key-value结果：
+```
+大气压, 100.07kPa
+干球温度, 0000℃
+计前温度, 0000℃
+湿球温度, 0000℃
+计前压力, -0000kPa
+流量, 00.0L/min
+静压, 00000kPa
+含湿量, 00.0 %
+```

From 9cc37f213a60bf0eb0cf565693e597fb7dc12760 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Thu, 16 Jun 2022 10:52:35 +0000
Subject: [PATCH 019/134] add case

---
 applications/液晶屏读数识别.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/applications/液晶屏读数识别.md b/applications/液晶屏读数识别.md
index 4a41afec77..428e317b9d 100644
--- a/applications/液晶屏读数识别.md
+++ b/applications/液晶屏读数识别.md
@@ -58,7 +58,7 @@ def get_one_image(train):
 
 get_one_image(train)  
 ```
-![det_png](./imgs/det.png)
+![det_png](imgs/det.png)
 
 ### 模型训练
 
@@ -343,7 +343,7 @@ def get_one_image(train):
 get_one_image(train)
 ```
 
-![rec_png](./imgs/rec.png)
+![rec_png](imgs/rec.png)
 
 ### 模型训练
 #### 下载预训练模型
@@ -475,7 +475,7 @@ plt.imshow(img)
 plt.show()
 ```
 
-![sys_res_png](./imgs/sys_res.png)
+![sys_res_png](imgs/sys_res.png)
 
 ### 后处理
 如果需要获取key-value信息，可以基于启发式的规则，将识别结果与关键字库进行匹配；如果匹配上了，则取该字段为key, 后面一个字段为value。

From ee22ebb49ff73934cef94ac9fda6ba6b3ef3ca4b Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Thu, 16 Jun 2022 12:10:12 +0000
Subject: [PATCH 020/134] add fleet tipc

---
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 7 files changed, 371 insertions(+)
 create mode 100644 test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
 create mode 100644 test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt

diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..5fb87fae18
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..4af814036a
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..8275254f3f
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv3_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv3_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..b76048813e
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=100|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
\ No newline at end of file
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..3a7c487173
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2.0_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c configs/rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..77b69f6c3a
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2.0_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2.0_det/det_r50_vd_db.yml -o 
+quant_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2.0_det/det_r50_vd_db.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_det/det_r50_vd_db.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
\ No newline at end of file
diff --git a/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..d486d37668
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2.0_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]

From 9832834c3f8b7fc66c6e8b6f3c8c3e39b4de15e3 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Thu, 16 Jun 2022 12:53:39 +0000
Subject: [PATCH 021/134] add case

---
 applications/液晶屏读数识别.md | 91 ++++++++++++++++++++++------------
 1 file changed, 58 insertions(+), 33 deletions(-)

diff --git a/applications/液晶屏读数识别.md b/applications/液晶屏读数识别.md
index 428e317b9d..6b1ea24306 100644
--- a/applications/液晶屏读数识别.md
+++ b/applications/液晶屏读数识别.md
@@ -1,12 +1,37 @@
-# 项目背景及意义
+# 基于PP-OCRv3的液晶屏读数识别
+
+- [1. 项目背景及意义](#1-项目背景及意义)
+- [2. 项目内容](#2-项目内容)
+- [3. 安装环境](#3-安装环境)
+- [4. 文字检测](#4-文字检测)
+  - [4.1 PP-OCRv3检测算法介绍](#41-PP-OCRv3检测算法介绍)
+  - [4.2 数据准备](#42-数据准备)
+  - [4.3 模型训练](#43-模型训练)
+    - [4.3.1 预训练模型直接评估](#431-预训练模型直接评估)
+    - [4.3.2 预训练模型直接finetune](#432-预训练模型直接finetune)
+    - [4.3.3 基于预训练模型Finetune_student模型](#433-基于预训练模型Finetune_student模型)
+    - [4.3.4 基于预训练模型Finetune_teacher模型](#434-基于预训练模型Finetune_teacher模型)
+    - [4.3.5 采用CML蒸馏进一步提升student模型精度](#435-采用CML蒸馏进一步提升student模型精度)
+    - [4.3.6 模型导出推理](#436-4.3.6-模型导出推理)
+- [5. 文字识别](#5-文字识别)
+  - [5.1 PP-OCRv3识别算法介绍](#51-PP-OCRv3识别算法介绍)
+  - [5.2 数据准备](#52-数据准备)
+  - [5.3 模型训练](#53-模型训练)
+  - [5.4 模型导出推理](#54-模型导出推理)
+- [6. 系统串联](#6-系统串联)
+  - [6.1 后处理](#61-后处理)
+- [7. PaddleServing部署](#7-PaddleServing部署)
+
+
+# 1. 项目背景及意义
 目前光学字符识别(OCR)技术在我们的生活当中被广泛使用，但是大多数模型在通用场景下的准确性还有待提高，针对于此我们借助飞桨提供的PaddleOCR套件较容易的实现了在垂类场景下的应用。
 
 该项目以国家质量基础（NQI）为准绳，充分利用大数据、云计算、物联网等高新技术，构建覆盖计量端、实验室端、数据端和硬件端的完整计量解决方案，解决传统计量校准中存在的难题，拓宽计量检测服务体系和服务领域；解决无数传接口或数传接口不统一、不公开的计量设备，以及计量设备所处的环境比较恶劣，不适合人工读取数据。通过OCR技术实现远程计量，引领计量行业向智慧计量转型和发展。
 
-# 项目内容
+# 2. 项目内容
 本项目基于PaddleOCR开源套件，以PP-OCRv3检测和识别模型为基础，针对液晶屏读数识别场景进行优化。
 
-## 安装环境
+# 3. 安装环境
 
 ```python
 # 首先git官方的PaddleOCR项目，安装需要的依赖
@@ -16,16 +41,16 @@ cd PaddleOCR
 pip install -r requirements.txt
 ```
 
-## 文字检测
+# 4. 文字检测
 文本检测的任务是定位出输入图像中的文字区域。近年来学术界关于文本检测的研究非常丰富，一类方法将文本检测视为目标检测中的一个特定场景，基于通用目标检测算法进行改进适配，如TextBoxes[1]基于一阶段目标检测器SSD[2]算法，调整目标框使之适合极端长宽比的文本行，CTPN[3]则是基于Faster RCNN[4]架构改进而来。但是文本检测与目标检测在目标信息以及任务本身上仍存在一些区别，如文本一般长宽比较大，往往呈“条状”，文本行之间可能比较密集，弯曲文本等，因此又衍生了很多专用于文本检测的算法。本项目基于PP-OCRv3算法进行优化。
 
-### PP-OCRv3检测算法介绍
+## 4.1 PP-OCRv3检测算法介绍
 PP-OCRv3检测模型是对PP-OCRv2中的CML（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。如下图所示，CML的核心思想结合了①传统的Teacher指导Student的标准蒸馏与 ②Students网络之间的DML互学习，可以让Students网络互学习的同时，Teacher网络予以指导。PP-OCRv3分别针对教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
 ![](https://ai-studio-static-online.cdn.bcebos.com/c306b2f028364805a55494d435ab553a76cf5ae5dd3f4649a948ea9aeaeb28b8)
 
 详细优化策略描述请参考[PP-OCRv3优化策略](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md#2)
 
-### 数据准备
+## 4.2 数据准备
 [计量设备屏幕字符检测数据集](https://aistudio.baidu.com/aistudio/datasetdetail/127845)数据来源于实际项目中各种计量设备的数显屏，以及在网上搜集的一些其他数显屏，包含训练集755张，测试集355张。
 
 ```python
@@ -60,9 +85,9 @@ get_one_image(train)
 ```
 ![det_png](imgs/det.png)
 
-### 模型训练
+## 4.3 模型训练
 
-#### 下载预训练模型
+### 4.3.1 预训练模型直接评估
 下载我们需要的PP-OCRv3检测预训练模型，更多选择请自行选择其他的[文字检测模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md#1-%E6%96%87%E6%9C%AC%E6%A3%80%E6%B5%8B%E6%A8%A1%E5%9E%8B)
 
 ```python
@@ -85,7 +110,7 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Globa
 |---|---------------------------|---|
 | 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
 
-**1)直接用全量策略finetune训练**
+### 4.3.2 预训练模型直接finetune
 #### 修改配置文件
 我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
 ```
@@ -119,7 +144,7 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Globa
 | 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
 | 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
 
-**2)基于预训练模型Finetune student模型**
+### 4.3.3 基于预训练模型Finetune_student模型
 
 我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
 ```
@@ -152,7 +177,7 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o G
 | 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
 | 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
 
-**3)基于预训练模型Finetune teacher模型**
+### 4.3.4 基于预训练模型Finetune_teacher模型
 
 首先需要从提供的预训练模型best_accuracy.pdparams中提取teacher参数，组合成适合dml训练的初始化模型，提取代码如下：
 
@@ -211,9 +236,9 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml -o Globa
 | 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
 | 3 | PP-OCRv3中英文超轻量检测预训练模型fintune教师模型 |84.8%|
 
-**4)采用CML蒸馏，进一步提升student模型精度**
+### 4.3.5 采用CML蒸馏进一步提升student模型精度
 
-需要从2) 、3）训练得到的best_accuracy.pdparams中提取各自代表student和teacher的参数，组合成适合cml训练的初始化模型，提取代码如下：
+需要从4.3.3和4.3.4训练得到的best_accuracy.pdparams中提取各自代表student和teacher的参数，组合成适合cml训练的初始化模型，提取代码如下：
 
 ```python
 # transform teacher params and student parameters into cml model
@@ -273,9 +298,9 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Globa
 | 3 | PP-OCRv3中英文超轻量检测预训练模型fintune教师模型 |84.8%|
 | 4 | 基于2和3训练好的模型fintune |82.7%|
 
-### 模型导出推理
+###  4.3.6 模型导出推理
 训练完成后，可以将训练模型转换成inference模型。inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
-#### 模型导出
+####  4.3.6.1 模型导出
 导出命令如下：
 
 ```python
@@ -287,7 +312,7 @@ python tools/export_model.py \
 
 ```
 
-#### 模型推理
+####  4.3.6.2 模型推理
 导出模型后，可以使用如下命令进行推理预测：
 
 ```python
@@ -295,10 +320,10 @@ python tools/export_model.py \
 python tools/infer/predict_det.py --image_dir="train_data/icdar2015/text_localization/test/1.jpg" --det_model_dir="./inference/det_ppocrv3/Student"
 ```
 
-## 文字识别
+# 5. 文字识别
 文本识别的任务是识别出图像中的文字内容，一般输入来自于文本检测得到的文本框截取出的图像文字区域。文本识别一般可以根据待识别文本形状分为规则文本识别和不规则文本识别两大类。规则文本主要指印刷字体、扫描文本等，文本大致处在水平线位置；不规则文本往往不在水平位置，存在弯曲、遮挡、模糊等问题。不规则文本场景具有很大的挑战性，也是目前文本识别领域的主要研究方向。本项目基于PP-OCRv3算法进行优化。
 
-### PP-OCRv3识别算法介绍
+## 5.1 PP-OCRv3识别算法介绍
 PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2205.00159)优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。如下图所示，PP-OCRv3采用了6个优化策略。
 ![](https://ai-studio-static-online.cdn.bcebos.com/d4f5344b5b854d50be738671598a89a45689c6704c4d481fb904dd7cf72f2a1a)
 
@@ -312,7 +337,7 @@ PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2
 
 详细优化策略描述请参考[PP-OCRv3优化策略](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md#3-%E8%AF%86%E5%88%AB%E4%BC%98%E5%8C%96)
 
-### 数据准备
+## 5.2 数据准备
 [计量设备屏幕字符识别数据集](https://aistudio.baidu.com/aistudio/datasetdetail/128714)数据来源于实际项目中各种计量设备的数显屏，以及在网上搜集的一些其他数显屏，包含训练集19912张，测试集4099张。
 
 ```python
@@ -345,8 +370,8 @@ get_one_image(train)
 
 ![rec_png](imgs/rec.png)
 
-### 模型训练
-#### 下载预训练模型
+## 5.3 模型训练
+###  下载预训练模型
 下载我们需要的PP-OCRv3识别预训练模型，更多选择请自行选择其他的[文字识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md#2-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E6%A8%A1%E5%9E%8B)
 
 ```python
@@ -356,7 +381,7 @@ wget -P ./pretrained_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch
 tar -xf ./pretrained_models/ch_PP-OCRv3_rec_train.tar -C pretrained_models
 ```
 
-#### 修改配置文件
+###  修改配置文件
 我们使用configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
 ```
   epoch_num: 100 # 训练epoch数
@@ -414,7 +439,7 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o
 |---|---------------------------|---|
 | 0 | PP-OCRv3中英文超轻量识别预训练模型直接预测 |70.4%|
 
-#### 开始训练
+###  开始训练
 我们使用上面修改好的配置文件configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml，预训练模型，数据集路径，学习率，训练轮数等都已经设置完毕后，可以使用下面命令开始训练。
 
 ```python
@@ -435,9 +460,9 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o
 | 0 | PP-OCRv3中英文超轻量识别预训练模型直接预测 |70.4%|
 | 1 | PP-OCRv3中英文超轻量识别预训练模型finetune |82.2%|
 
-### 模型导出推理
+##  5.4 模型导出推理
 训练完成后，可以将训练模型转换成inference模型。inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
-#### 模型导出
+###  模型导出
 导出命令如下：
 
 ```python
@@ -445,7 +470,7 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o
 python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_rec/best_accuracy" Global.save_inference_dir="./inference/rec_ppocrv3/"
 ```
 
-#### 模型推理
+### 模型推理
 导出模型后，可以使用如下命令进行推理预测
 
 ```python
@@ -453,7 +478,7 @@ python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillatio
 python tools/infer/predict_rec.py --image_dir="train_data/ic15_data/test/1_crop_0.jpg" --rec_model_dir="./inference/rec_ppocrv3/Student"
 ```
 
-## 系统串联
+# 6. 系统串联
 我们将上面训练好的检测和识别模型进行系统串联测试，命令如下：
 
 ```python
@@ -477,7 +502,7 @@ plt.show()
 
 ![sys_res_png](imgs/sys_res.png)
 
-### 后处理
+## 6.1 后处理
 如果需要获取key-value信息，可以基于启发式的规则，将识别结果与关键字库进行匹配；如果匹配上了，则取该字段为key, 后面一个字段为value。
 
 ```python
@@ -498,7 +523,7 @@ def postprocess(rec_res):
 key_value = postprocess(filter_rec_res)
 ```
 
-## PaddleServing部署
+# 7. PaddleServing部署
 首先需要安装PaddleServing部署相关的环境
 
 ```python
@@ -507,7 +532,7 @@ python -m pip install paddle_serving_client
 python -m pip install paddle-serving-app
 ```
 
-###  转化检测模型
+##  7.1 转化检测模型
 
 ```python
 cd deploy/pdserving/
@@ -518,7 +543,7 @@ python -m paddle_serving_client.convert --dirname ../../inference/det_ppocrv3/St
                                          --serving_client ./ppocr_det_v3_client/
 ```
 
-### 转化识别模型
+## 7.2 转化识别模型
 
 ```python
 python -m paddle_serving_client.convert --dirname ../../inference/rec_ppocrv3/Student \
@@ -529,7 +554,7 @@ python -m paddle_serving_client.convert --dirname ../../inference/rec_ppocrv3/St
 ```
 
 
-### 启动服务
+## 7.3 启动服务
 首先可以将后处理代码加入到web_service.py中，具体修改如下：
 ```
 # 代码153行后面增加下面代码
@@ -557,7 +582,7 @@ res = {"result": str(key_value)}
 python web_service.py 2>&1 >log.txt
 ```
 
-### 发送请求
+## 7.4 发送请求
 然后再开启一个新的终端，运行下面的客户端代码
 
 ```python

From a0c33908d500fe893d8e79e11399a5ab665f330b Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Thu, 16 Jun 2022 13:24:38 +0000
Subject: [PATCH 022/134] add TableMaster

---
 configs/table/table_master.yml                |  138 +
 configs/table/table_mv3.yml                   |   39 +-
 ppocr/data/imaug/gen_table_mask.py            |   95 +-
 ppocr/data/imaug/label_ops.py                 |  315 +-
 ppocr/data/pubtab_dataset.py                  |  126 +-
 ppocr/losses/__init__.py                      |    5 +-
 ppocr/losses/table_master_loss.py             |   65 +
 ppocr/metrics/table_metric.py                 |  115 +-
 ppocr/modeling/backbones/__init__.py          |    5 +-
 .../modeling/backbones/table_master_resnet.py |  369 +++
 ppocr/modeling/heads/__init__.py              |    3 +-
 ppocr/modeling/heads/table_att_head.py        |  136 +-
 ppocr/modeling/heads/table_master_head.py     |  276 ++
 ppocr/optimizer/learning_rate.py              |   43 +
 ppocr/postprocess/__init__.py                 |    5 +-
 ppocr/postprocess/rec_postprocess.py          |  140 -
 ppocr/postprocess/table_postprocess.py        |  160 +
 .../dict/table_master_structure_dict.txt      |   39 +
 ppocr/utils/dict/table_structure_dict.txt     | 2733 +----------------
 ppstructure/table/predict_structure.py        |  138 +-
 ppstructure/utility.py                        |    3 +-
 tools/export_model.py                         |    2 +
 tools/infer_table.py                          |   86 +-
 tools/program.py                              |   42 +-
 24 files changed, 1693 insertions(+), 3385 deletions(-)
 create mode 100755 configs/table/table_master.yml
 create mode 100644 ppocr/losses/table_master_loss.py
 create mode 100644 ppocr/modeling/backbones/table_master_resnet.py
 create mode 100644 ppocr/modeling/heads/table_master_head.py
 create mode 100644 ppocr/postprocess/table_postprocess.py
 create mode 100644 ppocr/utils/dict/table_master_structure_dict.txt

diff --git a/configs/table/table_master.yml b/configs/table/table_master.yml
new file mode 100755
index 0000000000..9dfc0e2746
--- /dev/null
+++ b/configs/table/table_master.yml
@@ -0,0 +1,138 @@
+Global:
+  use_gpu: true
+  epoch_num: 17
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/table_master/
+  save_epoch_step: 17
+  # evaluation is run every 400 iterations after the 0th iteration
+  eval_batch_step: [0, 400]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: output/table_master
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/table_master_structure_dict.txt
+  infer_mode: False
+  max_text_length: 500
+  process_total_num: 0
+  process_cut_num: 0
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: MultiStepDecay
+    learning_rate: 0.001
+    milestones: [12, 15]
+    gamma: 0.1
+    warmup_epoch: 0.02
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: table
+  algorithm: TableMaster
+  Backbone:
+    name: TableResNetExtra
+    gcb_config:
+      ratio: 0.0625
+      headers: 1
+      att_scale: False
+      fusion_type: channel_add
+      layers: [False, True, True, True]
+    layers: [1,2,5,3]
+  Head:
+    name: TableMasterHead
+    hidden_size: 512
+    headers: 8
+    dropout: 0
+    d_ff: 2024
+    max_text_length: 500
+
+Loss:
+  name: TableMasterLoss
+  ignore_index: 42 # set to len of dict + 3
+
+PostProcess:
+  name: TableMasterLabelDecode
+  box_shape: pad
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: true # cost many time, set False for training
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: /home/zhoujun20/table/PubTabNe/pubtabnet/train/
+    label_file_list: [/home/zhoujun20/table/PubTabNe/pubtabnet/PubTabNet_2.0.0_train.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: True
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          use_xywh: true
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bboxes', 'bbox_masks','shape']
+  loader:
+    shuffle: True
+    batch_size_per_card: 8
+    drop_last: True
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: /home/zhoujun20/table/PubTabNe/pubtabnet/val/
+    label_file_list: [/home/zhoujun20/table/PubTabNe/pubtabnet/val_500.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: True
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [ 480, 480 ]
+      - TableBoxEncode:
+          use_xywh: true
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.5, 0.5, 0.5 ]
+          std: [ 0.5, 0.5, 0.5 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks','shape' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 2
+    num_workers: 8
diff --git a/configs/table/table_mv3.yml b/configs/table/table_mv3.yml
index 1a91ea95af..9159addc31 100755
--- a/configs/table/table_mv3.yml
+++ b/configs/table/table_mv3.yml
@@ -4,21 +4,20 @@ Global:
   log_smooth_window: 20
   print_batch_step: 5
   save_model_dir: ./output/table_mv3/
-  save_epoch_step: 3
+  save_epoch_step: 400
   # evaluation is run every 400 iterations after the 0th iteration
   eval_batch_step: [0, 400]
   cal_metric_during_train: True
   pretrained_model:
-  checkpoints: 
+  checkpoints:
   save_inference_dir:
   use_visualdl: False
-  infer_img: doc/table/table.jpg
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: output/table_mv3
   # for data or label process
   character_dict_path: ppocr/utils/dict/table_structure_dict.txt
   character_type: en
-  max_text_length: 100
-  max_elem_length: 800
-  max_cell_num: 500
+  max_text_length: 500
   infer_mode: False
   process_total_num: 0
   process_cut_num: 0
@@ -44,11 +43,8 @@ Architecture:
   Head:
     name: TableAttentionHead
     hidden_size: 256
-    l2_decay: 0.00001
     loc_type: 2
-    max_text_length: 100
-    max_elem_length: 800
-    max_cell_num: 500
+    max_text_length: 500
 
 Loss:
   name: TableAttentionLoss
@@ -61,6 +57,7 @@ PostProcess:
 Metric:
   name: TableMetric
   main_indicator: acc
+  compute_bbox_metric: False # cost many time, set False for training
 
 Train:
   dataset:
@@ -71,18 +68,23 @@ Train:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+      - TableBoxEncode:
       - ResizeTableImage:
           max_len: 488
-      - TableLabelEncode:
       - NormalizeImage:
           scale: 1./255.
           mean: [0.485, 0.456, 0.406]
           std: [0.229, 0.224, 0.225]
           order: 'hwc'
       - PaddingTableImage:
+          size: [488, 488]
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
   loader:
     shuffle: True
     batch_size_per_card: 32
@@ -92,24 +94,29 @@ Train:
 Eval:
   dataset:
     name: PubTabDataSet
-    data_dir: train_data/table/pubtabnet/val/
-    label_file_path: train_data/table/pubtabnet/PubTabNet_2.0.0_val.jsonl
+    data_dir: /home/zhoujun20/table/PubTabNe/pubtabnet/val/
+    label_file_list: [/home/zhoujun20/table/PubTabNe/pubtabnet/val_500.jsonl]
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+      - TableBoxEncode:
       - ResizeTableImage:
           max_len: 488
-      - TableLabelEncode:
       - NormalizeImage:
           scale: 1./255.
           mean: [0.485, 0.456, 0.406]
           std: [0.229, 0.224, 0.225]
           order: 'hwc'
       - PaddingTableImage:
+          size: [488, 488]
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
   loader:
     shuffle: False
     drop_last: False
diff --git a/ppocr/data/imaug/gen_table_mask.py b/ppocr/data/imaug/gen_table_mask.py
index 08e35d5d1d..8d139190ab 100644
--- a/ppocr/data/imaug/gen_table_mask.py
+++ b/ppocr/data/imaug/gen_table_mask.py
@@ -32,7 +32,7 @@ class GenTableMask(object):
         self.shrink_h_max = 5
         self.shrink_w_max = 5
         self.mask_type = mask_type
-        
+
     def projection(self, erosion, h, w, spilt_threshold=0):
         # 水平投影
         projection_map = np.ones_like(erosion)
@@ -48,10 +48,12 @@ class GenTableMask(object):
         in_text = False  # 是否遍历到了字符区内
         box_list = []
         for i in range(len(project_val_array)):
-            if in_text == False and project_val_array[i] > spilt_threshold:  # 进入字符区了
+            if in_text == False and project_val_array[
+                    i] > spilt_threshold:  # 进入字符区了
                 in_text = True
                 start_idx = i
-            elif project_val_array[i] <= spilt_threshold and in_text == True:  # 进入空白区了
+            elif project_val_array[
+                    i] <= spilt_threshold and in_text == True:  # 进入空白区了
                 end_idx = i
                 in_text = False
                 if end_idx - start_idx <= 2:
@@ -70,7 +72,8 @@ class GenTableMask(object):
         box_gray_img = cv2.cvtColor(box_img, cv2.COLOR_BGR2GRAY)
         h, w = box_gray_img.shape
         # 灰度图片进行二值化处理
-        ret, thresh1 = cv2.threshold(box_gray_img, 200, 255, cv2.THRESH_BINARY_INV)
+        ret, thresh1 = cv2.threshold(box_gray_img, 200, 255,
+                                     cv2.THRESH_BINARY_INV)
         # 纵向腐蚀
         if h < w:
             kernel = np.ones((2, 1), np.uint8)
@@ -95,10 +98,12 @@ class GenTableMask(object):
         box_list = []
         spilt_threshold = 0
         for i in range(len(project_val_array)):
-            if in_text == False and project_val_array[i] > spilt_threshold:  # 进入字符区了
+            if in_text == False and project_val_array[
+                    i] > spilt_threshold:  # 进入字符区了
                 in_text = True
                 start_idx = i
-            elif project_val_array[i] <= spilt_threshold and in_text == True:  # 进入空白区了
+            elif project_val_array[
+                    i] <= spilt_threshold and in_text == True:  # 进入空白区了
                 end_idx = i
                 in_text = False
                 if end_idx - start_idx <= 2:
@@ -120,7 +125,8 @@ class GenTableMask(object):
                     h_end = h
                 word_img = erosion[h_start:h_end + 1, :]
                 word_h, word_w = word_img.shape
-                w_split_list, w_projection_map = self.projection(word_img.T, word_w, word_h)
+                w_split_list, w_projection_map = self.projection(word_img.T,
+                                                                 word_w, word_h)
                 w_start, w_end = w_split_list[0][0], w_split_list[-1][1]
                 if h_start > 0:
                     h_start -= 1
@@ -170,75 +176,54 @@ class GenTableMask(object):
 
                 for sno in range(len(split_bbox_list)):
                     left, top, right, bottom = split_bbox_list[sno]
-                    left, top, right, bottom = self.shrink_bbox([left, top, right, bottom])
+                    left, top, right, bottom = self.shrink_bbox(
+                        [left, top, right, bottom])
                     if self.mask_type == 1:
                         mask_img[top:bottom, left:right] = 1.0
                         data['mask_img'] = mask_img
                     else:
-                        mask_img[top:bottom, left:right, :] = (255, 255, 255)        
+                        mask_img[top:bottom, left:right, :] = (255, 255, 255)
                         data['image'] = mask_img
         return data
 
+
 class ResizeTableImage(object):
-    def __init__(self, max_len, **kwargs):
+    def __init__(self, max_len, resize_bboxes=False, infer_mode=False,
+                 **kwargs):
         super(ResizeTableImage, self).__init__()
         self.max_len = max_len
+        self.resize_bboxes = resize_bboxes
+        self.infer_mode = infer_mode
 
-    def get_img_bbox(self, cells):
-        bbox_list = []
-        if len(cells) == 0:
-            return bbox_list
-        cell_num = len(cells)
-        for cno in range(cell_num):
-            if "bbox" in cells[cno]:
-                bbox = cells[cno]['bbox']
-                bbox_list.append(bbox)
-        return bbox_list
-
-    def resize_img_table(self, img, bbox_list, max_len):
-        height, width = img.shape[0:2]
-        ratio = max_len / (max(height, width) * 1.0)
-        resize_h = int(height * ratio)
-        resize_w = int(width * ratio)
-        img_new = cv2.resize(img, (resize_w, resize_h))
-        bbox_list_new = []
-        for bno in range(len(bbox_list)):
-            left, top, right, bottom = bbox_list[bno].copy()
-            left = int(left * ratio)
-            top = int(top * ratio)
-            right = int(right * ratio)
-            bottom = int(bottom * ratio)
-            bbox_list_new.append([left, top, right, bottom])
-        return img_new, bbox_list_new
-    
     def __call__(self, data):
         img = data['image']
-        if 'cells' not in data:
-            cells = []
-        else:
-            cells = data['cells']
-        bbox_list = self.get_img_bbox(cells)
-        img_new, bbox_list_new = self.resize_img_table(img, bbox_list, self.max_len)
-        data['image'] = img_new
-        cell_num = len(cells)
-        bno = 0
-        for cno in range(cell_num):
-            if "bbox" in data['cells'][cno]:
-                data['cells'][cno]['bbox'] = bbox_list_new[bno]
-                bno += 1
+        height, width = img.shape[0:2]
+        ratio = self.max_len / (max(height, width) * 1.0)
+        resize_h = int(height * ratio)
+        resize_w = int(width * ratio)
+        resize_img = cv2.resize(img, (resize_w, resize_h))
+        if self.resize_bboxes and not self.infer_mode:
+            data['bboxes'] = data['bboxes'] * ratio
+        data['image'] = resize_img
+        data['src_img'] = img
+        data['shape'] = np.array([resize_h, resize_w, ratio, ratio])
         data['max_len'] = self.max_len
         return data
 
+
 class PaddingTableImage(object):
-    def __init__(self, **kwargs):
+    def __init__(self, size, **kwargs):
         super(PaddingTableImage, self).__init__()
-    
+        self.size = size
+
     def __call__(self, data):
         img = data['image']
-        max_len = data['max_len']
-        padding_img = np.zeros((max_len, max_len, 3), dtype=np.float32)
+        pad_h, pad_w = self.size
+        padding_img = np.zeros((pad_h, pad_w, 3), dtype=np.float32)
         height, width = img.shape[0:2]
         padding_img[0:height, 0:width, :] = img.copy()
         data['image'] = padding_img
+        shape = data['shape'].tolist()
+        shape.extend([pad_h, pad_w])
+        data['shape'] = np.array(shape)
         return data
-            
\ No newline at end of file
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 02a5187dad..a55869a641 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -443,7 +443,9 @@ class KieLabelEncode(object):
             elif 'key_cls' in anno.keys():
                 labels.append(anno['key_cls'])
             else:
-                raise ValueError("Cannot found 'key_cls' in ann.keys(), please check your training annotation.")
+                raise ValueError(
+                    "Cannot found 'key_cls' in ann.keys(), please check your training annotation."
+                )
             edges.append(ann.get('edge', 0))
         ann_infos = dict(
             image=data['image'],
@@ -580,171 +582,197 @@ class SRNLabelEncode(BaseRecLabelEncode):
         return idx
 
 
-class TableLabelEncode(object):
+class TableLabelEncode(AttnLabelEncode):
     """ Convert between text-label and text-index """
 
     def __init__(self,
                  max_text_length,
-                 max_elem_length,
-                 max_cell_num,
                  character_dict_path,
-                 span_weight=1.0,
+                 replace_empty_cell_token=False,
+                 merge_no_span_structure=False,
+                 learn_empty_box=False,
+                 point_num=4,
                  **kwargs):
-        self.max_text_length = max_text_length
-        self.max_elem_length = max_elem_length
-        self.max_cell_num = max_cell_num
-        list_character, list_elem = self.load_char_elem_dict(
-            character_dict_path)
-        list_character = self.add_special_char(list_character)
-        list_elem = self.add_special_char(list_elem)
-        self.dict_character = {}
-        for i, char in enumerate(list_character):
-            self.dict_character[char] = i
-        self.dict_elem = {}
-        for i, elem in enumerate(list_elem):
-            self.dict_elem[elem] = i
-        self.span_weight = span_weight
+        self.max_text_len = max_text_length
+        self.lower = False
+        self.learn_empty_box = learn_empty_box
+        self.merge_no_span_structure = merge_no_span_structure
+        self.replace_empty_cell_token = replace_empty_cell_token
 
-    def load_char_elem_dict(self, character_dict_path):
-        list_character = []
-        list_elem = []
+        dict_character = []
         with open(character_dict_path, "rb") as fin:
             lines = fin.readlines()
-            substr = lines[0].decode('utf-8').strip("\r\n").split("\t")
-            character_num = int(substr[0])
-            elem_num = int(substr[1])
-            for cno in range(1, 1 + character_num):
-                character = lines[cno].decode('utf-8').strip("\r\n")
-                list_character.append(character)
-            for eno in range(1 + character_num, 1 + character_num + elem_num):
-                elem = lines[eno].decode('utf-8').strip("\r\n")
-                list_elem.append(elem)
-        return list_character, list_elem
+            for line in lines:
+                line = line.decode('utf-8').strip("\n").strip("\r\n")
+                dict_character.append(line)
 
-    def add_special_char(self, list_character):
-        self.beg_str = "sos"
-        self.end_str = "eos"
-        list_character = [self.beg_str] + list_character + [self.end_str]
-        return list_character
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.idx2char = {v: k for k, v in self.dict.items()}
 
-    def get_span_idx_list(self):
-        span_idx_list = []
-        for elem in self.dict_elem:
-            if 'span' in elem:
-                span_idx_list.append(self.dict_elem[elem])
-        return span_idx_list
+        self.character = dict_character
+        self.point_num = point_num
+        self.pad_idx = self.dict[self.beg_str]
+        self.start_idx = self.dict[self.beg_str]
+        self.end_idx = self.dict[self.end_str]
+
+        self.td_token = ['<td>', '<td', '<eb></eb>', '<td></td>']
+        self.empty_bbox_token_dict = {
+            "[]": '<eb></eb>',
+            "[' ']": '<eb1></eb1>',
+            "['<b>', ' ', '</b>']": '<eb2></eb2>',
+            "['\\u2028', '\\u2028']": '<eb3></eb3>',
+            "['<sup>', ' ', '</sup>']": '<eb4></eb4>',
+            "['<b>', '</b>']": '<eb5></eb5>',
+            "['<i>', ' ', '</i>']": '<eb6></eb6>',
+            "['<b>', '<i>', '</i>', '</b>']": '<eb7></eb7>',
+            "['<b>', '<i>', ' ', '</i>', '</b>']": '<eb8></eb8>',
+            "['<i>', '</i>']": '<eb9></eb9>',
+            "['<b>', ' ', '\\u2028', ' ', '\\u2028', ' ', '</b>']":
+            '<eb10></eb10>',
+        }
+
+    @property
+    def _max_text_len(self):
+        return self.max_text_len + 2
 
     def __call__(self, data):
         cells = data['cells']
-        structure = data['structure']['tokens']
-        structure = self.encode(structure, 'elem')
+        structure = data['structure']
+        if self.merge_no_span_structure:
+            structure = self._merge_no_span_structure(structure)
+        if self.replace_empty_cell_token:
+            structure = self._replace_empty_cell_token(structure, cells)
+        # remove empty token and add " " to span token
+        new_structure = []
+        for token in structure:
+            if token != '':
+                if 'span' in token and token[0] != ' ':
+                    token = ' ' + token
+                new_structure.append(token)
+        # encode structure
+        structure = self.encode(new_structure)
         if structure is None:
             return None
-        elem_num = len(structure)
-        structure = [0] + structure + [len(self.dict_elem) - 1]
-        structure = structure + [0] * (self.max_elem_length + 2 - len(structure)
-                                       )
+
+        structure = [self.start_idx] + structure + [self.end_idx
+                                                    ]  # add sos abd eos
+        structure = structure + [self.pad_idx] * (self._max_text_len -
+                                                  len(structure))  # pad
         structure = np.array(structure)
         data['structure'] = structure
-        elem_char_idx1 = self.dict_elem['<td>']
-        elem_char_idx2 = self.dict_elem['<td']
-        span_idx_list = self.get_span_idx_list()
-        td_idx_list = np.logical_or(structure == elem_char_idx1,
-                                    structure == elem_char_idx2)
-        td_idx_list = np.where(td_idx_list)[0]
 
-        structure_mask = np.ones(
-            (self.max_elem_length + 2, 1), dtype=np.float32)
-        bbox_list = np.zeros((self.max_elem_length + 2, 4), dtype=np.float32)
-        bbox_list_mask = np.zeros(
-            (self.max_elem_length + 2, 1), dtype=np.float32)
-        img_height, img_width, img_ch = data['image'].shape
-        if len(span_idx_list) > 0:
-            span_weight = len(td_idx_list) * 1.0 / len(span_idx_list)
-            span_weight = min(max(span_weight, 1.0), self.span_weight)
-        for cno in range(len(cells)):
-            if 'bbox' in cells[cno]:
-                bbox = cells[cno]['bbox'].copy()
-                bbox[0] = bbox[0] * 1.0 / img_width
-                bbox[1] = bbox[1] * 1.0 / img_height
-                bbox[2] = bbox[2] * 1.0 / img_width
-                bbox[3] = bbox[3] * 1.0 / img_height
-                td_idx = td_idx_list[cno]
-                bbox_list[td_idx] = bbox
-                bbox_list_mask[td_idx] = 1.0
-                cand_span_idx = td_idx + 1
-                if cand_span_idx < (self.max_elem_length + 2):
-                    if structure[cand_span_idx] in span_idx_list:
-                        structure_mask[cand_span_idx] = span_weight
+        if len(structure) > self._max_text_len:
+            return None
 
-        data['bbox_list'] = bbox_list
-        data['bbox_list_mask'] = bbox_list_mask
-        data['structure_mask'] = structure_mask
-        char_beg_idx = self.get_beg_end_flag_idx('beg', 'char')
-        char_end_idx = self.get_beg_end_flag_idx('end', 'char')
-        elem_beg_idx = self.get_beg_end_flag_idx('beg', 'elem')
-        elem_end_idx = self.get_beg_end_flag_idx('end', 'elem')
-        data['sp_tokens'] = np.array([
-            char_beg_idx, char_end_idx, elem_beg_idx, elem_end_idx,
-            elem_char_idx1, elem_char_idx2, self.max_text_length,
-            self.max_elem_length, self.max_cell_num, elem_num
-        ])
+        # encode box
+        bboxes = np.zeros(
+            (self._max_text_len, self.point_num), dtype=np.float32)
+        bbox_masks = np.zeros((self._max_text_len, 1), dtype=np.float32)
+
+        bbox_idx = 0
+        for i, token in enumerate(structure):
+            if self.idx2char[token] in self.td_token:
+                if 'bbox' in cells[bbox_idx]:
+                    bbox = cells[bbox_idx]['bbox'].copy()
+                    bbox = np.array(bbox, dtype=np.float32).reshape(-1)
+                    bboxes[i] = bbox
+                    bbox_masks[i] = 1.0
+                if self.learn_empty_box:
+                    bbox_masks[i] = 1.0
+                bbox_idx += 1
+        data['bboxes'] = bboxes
+        data['bbox_masks'] = bbox_masks
         return data
 
-    def encode(self, text, char_or_elem):
-        """convert text-label into text-index.
+    def _merge_no_span_structure(self, structure):
+        new_structure = []
+        i = 0
+        while i < len(structure):
+            token = structure[i]
+            if token == '<td>':
+                token = '<td></td>'
+                i += 1
+            new_structure.append(token)
+            i += 1
+        return new_structure
+
+    def _replace_empty_cell_token(self, token_list, cells):
+        bbox_idx = 0
+        add_empty_bbox_token_list = []
+        for token in token_list:
+            if token in ['<td></td>', '<td', '<td>']:
+                if 'bbox' not in cells[bbox_idx].keys():
+                    content = str(cells[bbox_idx]['tokens'])
+                    token = self.empty_bbox_token_dict[content]
+                add_empty_bbox_token_list.append(token)
+                bbox_idx += 1
+            else:
+                add_empty_bbox_token_list.append(token)
+        return add_empty_bbox_token_list
+
+
+class TableMasterLabelEncode(TableLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path,
+                 replace_empty_cell_token=False,
+                 merge_no_span_structure=False,
+                 learn_empty_box=False,
+                 point_num=4,
+                 **kwargs):
+        super(TableMasterLabelEncode, self).__init__(
+            max_text_length, character_dict_path, replace_empty_cell_token,
+            merge_no_span_structure, learn_empty_box, point_num, **kwargs)
+
+    @property
+    def _max_text_len(self):
+        return self.max_text_len
+
+    def add_special_char(self, dict_character):
+        self.beg_str = '<SOS>'
+        self.end_str = '<EOS>'
+        self.unknown_str = '<UKN>'
+        self.pad_str = '<PAD>'
+        dict_character = dict_character
+        dict_character = dict_character + [
+            self.unknown_str, self.beg_str, self.end_str, self.pad_str
+        ]
+        return dict_character
+
+
+class TableBoxEncode(object):
+    def __init__(self, use_xywh=False, **kwargs):
+        self.use_xywh = use_xywh
+
+    def __call__(self, data):
+        img_height, img_width = data['image'].shape[:2]
+        bboxes = data['bboxes']
+        if self.use_xywh and bboxes.shape[1] == 4:
+            bboxes = self.xyxy2xywh(bboxes)
+        bboxes[:, 0::2] /= img_width
+        bboxes[:, 1::2] /= img_height
+        data['bboxes'] = bboxes
+        return data
+
+    def xyxy2xywh(self, bboxes):
         """
-        if char_or_elem == "char":
-            max_len = self.max_text_length
-            current_dict = self.dict_character
-        else:
-            max_len = self.max_elem_length
-            current_dict = self.dict_elem
-        if len(text) > max_len:
-            return None
-        if len(text) == 0:
-            if char_or_elem == "char":
-                return [self.dict_character['space']]
-            else:
-                return None
-        text_list = []
-        for char in text:
-            if char not in current_dict:
-                return None
-            text_list.append(current_dict[char])
-        if len(text_list) == 0:
-            if char_or_elem == "char":
-                return [self.dict_character['space']]
-            else:
-                return None
-        return text_list
-
-    def get_ignored_tokens(self, char_or_elem):
-        beg_idx = self.get_beg_end_flag_idx("beg", char_or_elem)
-        end_idx = self.get_beg_end_flag_idx("end", char_or_elem)
-        return [beg_idx, end_idx]
-
-    def get_beg_end_flag_idx(self, beg_or_end, char_or_elem):
-        if char_or_elem == "char":
-            if beg_or_end == "beg":
-                idx = np.array(self.dict_character[self.beg_str])
-            elif beg_or_end == "end":
-                idx = np.array(self.dict_character[self.end_str])
-            else:
-                assert False, "Unsupport type %s in get_beg_end_flag_idx of char" \
-                              % beg_or_end
-        elif char_or_elem == "elem":
-            if beg_or_end == "beg":
-                idx = np.array(self.dict_elem[self.beg_str])
-            elif beg_or_end == "end":
-                idx = np.array(self.dict_elem[self.end_str])
-            else:
-                assert False, "Unsupport type %s in get_beg_end_flag_idx of elem" \
-                              % beg_or_end
-        else:
-            assert False, "Unsupport type %s in char_or_elem" \
-                % char_or_elem
-        return idx
+        Convert coord (x1,y1,x2,y2) to (x,y,w,h).
+        where (x1,y1) is top-left, (x2,y2) is bottom-right.
+        (x,y) is bbox center and (w,h) is width and height.
+        :param bboxes: (x1, y1, x2, y2)
+        :return:
+        """
+        new_bboxes = np.empty_like(bboxes)
+        new_bboxes[:, 0] = (bboxes[:, 0] + bboxes[:, 2]) / 2  # x center
+        new_bboxes[:, 1] = (bboxes[:, 1] + bboxes[:, 3]) / 2  # y center
+        new_bboxes[:, 2] = bboxes[:, 2] - bboxes[:, 0]  # width
+        new_bboxes[:, 3] = bboxes[:, 3] - bboxes[:, 1]  # height
+        return new_bboxes
 
 
 class SARLabelEncode(BaseRecLabelEncode):
@@ -1030,7 +1058,6 @@ class MultiLabelEncode(BaseRecLabelEncode):
                                          use_space_char, **kwargs)
 
     def __call__(self, data):
-
         data_ctc = copy.deepcopy(data)
         data_sar = copy.deepcopy(data)
         data_out = dict()
diff --git a/ppocr/data/pubtab_dataset.py b/ppocr/data/pubtab_dataset.py
index 671cda76fb..105f28db42 100644
--- a/ppocr/data/pubtab_dataset.py
+++ b/ppocr/data/pubtab_dataset.py
@@ -16,6 +16,7 @@ import os
 import random
 from paddle.io import Dataset
 import json
+from copy import deepcopy
 
 from .imaug import transform, create_operators
 
@@ -29,33 +30,63 @@ class PubTabDataSet(Dataset):
         dataset_config = config[mode]['dataset']
         loader_config = config[mode]['loader']
 
-        label_file_path = dataset_config.pop('label_file_path')
+        label_file_list = dataset_config.pop('label_file_list')
+        data_source_num = len(label_file_list)
+        ratio_list = dataset_config.get("ratio_list", [1.0])
+        if isinstance(ratio_list, (float, int)):
+            ratio_list = [float(ratio_list)] * int(data_source_num)
+
+        assert len(
+            ratio_list
+        ) == data_source_num, "The length of ratio_list should be the same as the file_list."
 
         self.data_dir = dataset_config['data_dir']
         self.do_shuffle = loader_config['shuffle']
-        self.do_hard_select = False
-        if 'hard_select' in loader_config:
-            self.do_hard_select = loader_config['hard_select']
-            self.hard_prob = loader_config['hard_prob']
-        if self.do_hard_select:
-            self.img_select_prob = self.load_hard_select_prob()
-        self.table_select_type = None
-        if 'table_select_type' in loader_config:
-            self.table_select_type = loader_config['table_select_type']
-            self.table_select_prob = loader_config['table_select_prob']
 
         self.seed = seed
-        logger.info("Initialize indexs of datasets:%s" % label_file_path)
-        with open(label_file_path, "rb") as f:
-            self.data_lines = f.readlines()
-        self.data_idx_order_list = list(range(len(self.data_lines)))
-        if mode.lower() == "train":
+        self.mode = mode.lower()
+        logger.info("Initialize indexs of datasets:%s" % label_file_list)
+        self.data_lines = self.get_image_info_list(label_file_list, ratio_list)
+        # self.check(config['Global']['max_text_length'])
+
+        if mode.lower() == "train" and self.do_shuffle:
             self.shuffle_data_random()
         self.ops = create_operators(dataset_config['transforms'], global_config)
-
-        ratio_list = dataset_config.get("ratio_list", [1.0])
         self.need_reset = True in [x < 1 for x in ratio_list]
 
+    def get_image_info_list(self, file_list, ratio_list):
+        if isinstance(file_list, str):
+            file_list = [file_list]
+        data_lines = []
+        for idx, file in enumerate(file_list):
+            with open(file, "rb") as f:
+                lines = f.readlines()
+                if self.mode == "train" or ratio_list[idx] < 1.0:
+                    random.seed(self.seed)
+                    lines = random.sample(lines,
+                                          round(len(lines) * ratio_list[idx]))
+                data_lines.extend(lines)
+        return data_lines
+
+    def check(self, max_text_length):
+        data_lines = []
+        for line in self.data_lines:
+            data_line = line.decode('utf-8').strip("\n")
+            info = json.loads(data_line)
+            file_name = info['filename']
+            cells = info['html']['cells'].copy()
+            structure = info['html']['structure']['tokens'].copy()
+
+            img_path = os.path.join(self.data_dir, file_name)
+            if not os.path.exists(img_path):
+                self.logger.warning("{} does not exist!".format(img_path))
+                continue
+            if len(structure) == 0 or len(structure) > max_text_length:
+                continue
+            # data = {'img_path': img_path, 'cells': cells, 'structure':structure,'file_name':file_name}
+            data_lines.append(line)
+        self.data_lines = data_lines
+
     def shuffle_data_random(self):
         if self.do_shuffle:
             random.seed(self.seed)
@@ -68,47 +99,34 @@ class PubTabDataSet(Dataset):
             data_line = data_line.decode('utf-8').strip("\n")
             info = json.loads(data_line)
             file_name = info['filename']
-            select_flag = True
-            if self.do_hard_select:
-                prob = self.img_select_prob[file_name]
-                if prob < random.uniform(0, 1):
-                    select_flag = False
+            cells = info['html']['cells'].copy()
+            structure = info['html']['structure']['tokens'].copy()
 
-            if self.table_select_type:
-                structure = info['html']['structure']['tokens'].copy()
-                structure_str = ''.join(structure)
-                table_type = "simple"
-                if 'colspan' in structure_str or 'rowspan' in structure_str:
-                    table_type = "complex"
-                if table_type == "complex":
-                    if self.table_select_prob < random.uniform(0, 1):
-                        select_flag = False
+            img_path = os.path.join(self.data_dir, file_name)
+            if not os.path.exists(img_path):
+                raise Exception("{} does not exist!".format(img_path))
+            data = {
+                'img_path': img_path,
+                'cells': cells,
+                'structure': structure,
+                'file_name': file_name
+            }
 
-            if select_flag:
-                cells = info['html']['cells'].copy()
-                structure = info['html']['structure'].copy()
-                img_path = os.path.join(self.data_dir, file_name)
-                data = {
-                    'img_path': img_path,
-                    'cells': cells,
-                    'structure': structure
-                }
-                if not os.path.exists(img_path):
-                    raise Exception("{} does not exist!".format(img_path))
-                with open(data['img_path'], 'rb') as f:
-                    img = f.read()
-                    data['image'] = img
-                outs = transform(data, self.ops)
-            else:
-                outs = None
-        except Exception as e:
+            with open(data['img_path'], 'rb') as f:
+                img = f.read()
+                data['image'] = img
+            outs = transform(data, self.ops)
+        except:
+            import traceback
+            err = traceback.format_exc()
             self.logger.error(
-                "When parsing line {}, error happened with msg: {}".format(
-                    data_line, e))
+                "When parsing line {}, error happened with msg: {}".format(err))
             outs = None
         if outs is None:
-            return self.__getitem__(np.random.randint(self.__len__()))
+            rnd_idx = np.random.randint(self.__len__(
+            )) if self.mode == "train" else (idx + 1) % self.__len__()
+            return self.__getitem__(rnd_idx)
         return outs
 
     def __len__(self):
-        return len(self.data_idx_order_list)
+        return len(self.data_lines)
diff --git a/ppocr/losses/__init__.py b/ppocr/losses/__init__.py
index de8419b7c1..0f208007b1 100755
--- a/ppocr/losses/__init__.py
+++ b/ppocr/losses/__init__.py
@@ -51,7 +51,7 @@ from .combined_loss import CombinedLoss
 
 # table loss
 from .table_att_loss import TableAttentionLoss
-
+from .table_master_loss import TableMasterLoss
 # vqa token loss
 from .vqa_token_layoutlm_loss import VQASerTokenLayoutLMLoss
 
@@ -61,7 +61,8 @@ def build_loss(config):
         'DBLoss', 'PSELoss', 'EASTLoss', 'SASTLoss', 'FCELoss', 'CTCLoss',
         'ClsLoss', 'AttentionLoss', 'SRNLoss', 'PGLoss', 'CombinedLoss',
         'NRTRLoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', 'SDMGRLoss',
-        'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', 'MultiLoss'
+        'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', 'MultiLoss',
+        'TableMasterLoss'
     ]
     config = copy.deepcopy(config)
     module_name = config.pop('name')
diff --git a/ppocr/losses/table_master_loss.py b/ppocr/losses/table_master_loss.py
new file mode 100644
index 0000000000..50a773dd9a
--- /dev/null
+++ b/ppocr/losses/table_master_loss.py
@@ -0,0 +1,65 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import paddle
+from paddle import nn
+
+
+class TableMasterLoss(nn.Layer):
+    def __init__(self, ignore_index=-1):
+        super(TableMasterLoss, self).__init__()
+        self.structure_loss = nn.CrossEntropyLoss(
+            ignore_index=ignore_index, reduction='mean')
+        self.box_loss = nn.L1Loss(reduction='sum')
+        self.eps = 1e-12
+
+    def forward(self, predicts, batch):
+        # structure_loss
+        structure_probs = predicts['structure_probs']
+        structure_targets = batch[1]
+        structure_targets = structure_targets[:, 1:]
+        structure_probs = structure_probs.reshape(
+            [-1, structure_probs.shape[-1]])
+        structure_targets = structure_targets.reshape([-1])
+
+        structure_loss = self.structure_loss(structure_probs, structure_targets)
+        structure_loss = structure_loss.mean()
+        losses = dict(structure_loss=structure_loss)
+
+        # box loss
+        bboxes_preds = predicts['loc_preds']
+        bboxes_targets = batch[2][:, 1:, :]
+        bbox_masks = batch[3][:, 1:]
+        # mask empty-bbox or non-bbox structure token's bbox.
+
+        masked_bboxes_preds = bboxes_preds * bbox_masks
+        masked_bboxes_targets = bboxes_targets * bbox_masks
+
+        # horizon loss (x and width)
+        horizon_sum_loss = self.box_loss(masked_bboxes_preds[:, :, 0::2],
+                                         masked_bboxes_targets[:, :, 0::2])
+        horizon_loss = horizon_sum_loss / (bbox_masks.sum() + self.eps)
+        # vertical loss (y and height)
+        vertical_sum_loss = self.box_loss(masked_bboxes_preds[:, :, 1::2],
+                                          masked_bboxes_targets[:, :, 1::2])
+        vertical_loss = vertical_sum_loss / (bbox_masks.sum() + self.eps)
+
+        horizon_loss = horizon_loss.mean()
+        vertical_loss = vertical_loss.mean()
+        all_loss = structure_loss + horizon_loss + vertical_loss
+        losses.update({
+            'loss': all_loss,
+            'horizon_bbox_loss': horizon_loss,
+            'vertical_bbox_loss': vertical_loss
+        })
+        return losses
diff --git a/ppocr/metrics/table_metric.py b/ppocr/metrics/table_metric.py
index ca4d647420..17f3dc92b2 100644
--- a/ppocr/metrics/table_metric.py
+++ b/ppocr/metrics/table_metric.py
@@ -12,29 +12,30 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import numpy as np
+from ppocr.metrics.det_metric import DetMetric
 
 
-class TableMetric(object):
-    def __init__(self, main_indicator='acc', **kwargs):
+class TableStructureMetric(object):
+    def __init__(self, main_indicator='acc', eps=1e-6, **kwargs):
         self.main_indicator = main_indicator
-        self.eps = 1e-5
+        self.eps = eps
         self.reset()
 
-    def __call__(self, pred, batch, *args, **kwargs):
-        structure_probs = pred['structure_probs'].numpy()
-        structure_labels = batch[1]
+    def __call__(self, pred_label, batch=None, *args, **kwargs):
+        preds, labels = pred_label
+        pred_structure_batch_list = preds['structure_batch_list']
+        gt_structure_batch_list = labels['structure_batch_list']
         correct_num = 0
         all_num = 0
-        structure_probs = np.argmax(structure_probs, axis=2)
-        structure_labels = structure_labels[:, 1:]
-        batch_size = structure_probs.shape[0]
-        for bno in range(batch_size):
-            all_num += 1
-            if (structure_probs[bno] == structure_labels[bno]).all():
+        for (pred, pred_conf), target in zip(pred_structure_batch_list,
+                                             gt_structure_batch_list):
+            pred_str = ''.join(pred)
+            target_str = ''.join(target)
+            if pred_str == target_str:
                 correct_num += 1
+            all_num += 1
         self.correct_num += correct_num
         self.all_num += all_num
-        return {'acc': correct_num * 1.0 / (all_num + self.eps), }
 
     def get_metric(self):
         """
@@ -49,3 +50,91 @@ class TableMetric(object):
     def reset(self):
         self.correct_num = 0
         self.all_num = 0
+        self.len_acc_num = 0
+        self.token_nums = 0
+        self.anys_dict = dict()
+        from collections import defaultdict
+        self.error_num_dict = defaultdict(int)
+
+
+class TableMetric(object):
+    def __init__(self,
+                 main_indicator='acc',
+                 compute_bbox_metric=False,
+                 point_num=4,
+                 **kwargs):
+        """
+
+        @param sub_metrics: configs of sub_metric
+        @param main_matric: main_matric for save best_model
+        @param kwargs:
+        """
+        self.structure_metric = TableStructureMetric()
+        self.bbox_metric = DetMetric() if compute_bbox_metric else None
+        self.main_indicator = main_indicator
+        self.point_num = point_num
+        self.reset()
+
+    def __call__(self, pred_label, batch=None, *args, **kwargs):
+        self.structure_metric(pred_label)
+        if self.bbox_metric is not None:
+            self.bbox_metric(*self.prepare_bbox_metric_input(pred_label))
+
+    def prepare_bbox_metric_input(self, pred_label):
+        pred_bbox_batch_list = []
+        gt_ignore_tags_batch_list = []
+        gt_bbox_batch_list = []
+        preds, labels = pred_label
+
+        batch_num = len(preds['bbox_batch_list'])
+        for batch_idx in range(batch_num):
+            # pred
+            pred_bbox_list = [
+                self.format_box(pred_box)
+                for pred_box in preds['bbox_batch_list'][batch_idx]
+            ]
+            pred_bbox_batch_list.append({'points': pred_bbox_list})
+
+            # gt
+            gt_bbox_list = []
+            gt_ignore_tags_list = []
+            for gt_box in labels['bbox_batch_list'][batch_idx]:
+                gt_bbox_list.append(self.format_box(gt_box))
+                gt_ignore_tags_list.append(0)
+            gt_bbox_batch_list.append(gt_bbox_list)
+            gt_ignore_tags_batch_list.append(gt_ignore_tags_list)
+
+        return [
+            pred_bbox_batch_list,
+            [0, 0, gt_bbox_batch_list, gt_ignore_tags_batch_list]
+        ]
+
+    def get_metric(self):
+        structure_metric = self.structure_metric.get_metric()
+        if self.bbox_metric is None:
+            return structure_metric
+        bbox_metric = self.bbox_metric.get_metric()
+        if self.main_indicator == self.bbox_metric.main_indicator:
+            output = bbox_metric
+            for sub_key in structure_metric:
+                output["structure_metric_{}".format(
+                    sub_key)] = structure_metric[sub_key]
+        else:
+            output = structure_metric
+            for sub_key in bbox_metric:
+                output["bbox_metric_{}".format(sub_key)] = bbox_metric[sub_key]
+        return output
+
+    def reset(self):
+        self.structure_metric.reset()
+        if self.bbox_metric is not None:
+            self.bbox_metric.reset()
+
+    def format_box(self, box):
+        if self.point_num == 4:
+            x1, y1, x2, y2 = box
+            box = [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+        elif self.point_num == 8:
+            x1, y1, x2, y2, x3, y3, x4, y4 = box
+            box = [[x1, y1], [x2, y2], [x3, y3], [x4, y4]]
+        return box
diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
index 072d6e0f84..2b5fd9142e 100755
--- a/ppocr/modeling/backbones/__init__.py
+++ b/ppocr/modeling/backbones/__init__.py
@@ -20,7 +20,10 @@ def build_backbone(config, model_type):
         from .det_mobilenet_v3 import MobileNetV3
         from .det_resnet_vd import ResNet
         from .det_resnet_vd_sast import ResNet_SAST
-        support_dict = ["MobileNetV3", "ResNet", "ResNet_SAST"]
+        from .table_master_resnet import TableResNetExtra
+        support_dict = [
+            "MobileNetV3", "ResNet", "ResNet_SAST", "TableResNetExtra"
+        ]
     elif model_type == "rec" or model_type == "cls":
         from .rec_mobilenet_v3 import MobileNetV3
         from .rec_resnet_vd import ResNet
diff --git a/ppocr/modeling/backbones/table_master_resnet.py b/ppocr/modeling/backbones/table_master_resnet.py
new file mode 100644
index 0000000000..82b4f37a74
--- /dev/null
+++ b/ppocr/modeling/backbones/table_master_resnet.py
@@ -0,0 +1,369 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self,
+                 inplanes,
+                 planes,
+                 stride=1,
+                 downsample=None,
+                 gcb_config=None):
+        super(BasicBlock, self).__init__()
+        self.conv1 = nn.Conv2D(
+            inplanes,
+            planes,
+            kernel_size=3,
+            stride=stride,
+            padding=1,
+            bias_attr=False)
+        self.bn1 = nn.BatchNorm2D(planes, momentum=0.9)
+        self.relu = nn.ReLU()
+        self.conv2 = nn.Conv2D(
+            planes, planes, kernel_size=3, stride=1, padding=1, bias_attr=False)
+        self.bn2 = nn.BatchNorm2D(planes, momentum=0.9)
+        self.downsample = downsample
+        self.stride = stride
+        self.gcb_config = gcb_config
+
+        if self.gcb_config is not None:
+            gcb_ratio = gcb_config['ratio']
+            gcb_headers = gcb_config['headers']
+            att_scale = gcb_config['att_scale']
+            fusion_type = gcb_config['fusion_type']
+            self.context_block = MultiAspectGCAttention(
+                inplanes=planes,
+                ratio=gcb_ratio,
+                headers=gcb_headers,
+                att_scale=att_scale,
+                fusion_type=fusion_type)
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.gcb_config is not None:
+            out = self.context_block(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+def get_gcb_config(gcb_config, layer):
+    if gcb_config is None or not gcb_config['layers'][layer]:
+        return None
+    else:
+        return gcb_config
+
+
+class TableResNetExtra(nn.Layer):
+    def __init__(self, layers, in_channels=3, gcb_config=None):
+        assert len(layers) >= 4
+
+        super(TableResNetExtra, self).__init__()
+        self.inplanes = 128
+        self.conv1 = nn.Conv2D(
+            in_channels,
+            64,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False)
+        self.bn1 = nn.BatchNorm2D(64)
+        self.relu1 = nn.ReLU()
+
+        self.conv2 = nn.Conv2D(
+            64, 128, kernel_size=3, stride=1, padding=1, bias_attr=False)
+        self.bn2 = nn.BatchNorm2D(128)
+        self.relu2 = nn.ReLU()
+
+        self.maxpool1 = nn.MaxPool2D(kernel_size=2, stride=2)
+
+        self.layer1 = self._make_layer(
+            BasicBlock,
+            256,
+            layers[0],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 0))
+
+        self.conv3 = nn.Conv2D(
+            256, 256, kernel_size=3, stride=1, padding=1, bias_attr=False)
+        self.bn3 = nn.BatchNorm2D(256)
+        self.relu3 = nn.ReLU()
+
+        self.maxpool2 = nn.MaxPool2D(kernel_size=2, stride=2)
+
+        self.layer2 = self._make_layer(
+            BasicBlock,
+            256,
+            layers[1],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 1))
+
+        self.conv4 = nn.Conv2D(
+            256, 256, kernel_size=3, stride=1, padding=1, bias_attr=False)
+        self.bn4 = nn.BatchNorm2D(256)
+        self.relu4 = nn.ReLU()
+
+        self.maxpool3 = nn.MaxPool2D(kernel_size=2, stride=2)
+
+        self.layer3 = self._make_layer(
+            BasicBlock,
+            512,
+            layers[2],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 2))
+
+        self.conv5 = nn.Conv2D(
+            512, 512, kernel_size=3, stride=1, padding=1, bias_attr=False)
+        self.bn5 = nn.BatchNorm2D(512)
+        self.relu5 = nn.ReLU()
+
+        self.layer4 = self._make_layer(
+            BasicBlock,
+            512,
+            layers[3],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 3))
+
+        self.conv6 = nn.Conv2D(
+            512, 512, kernel_size=3, stride=1, padding=1, bias_attr=False)
+        self.bn6 = nn.BatchNorm2D(512)
+        self.relu6 = nn.ReLU()
+
+        self.out_channels = [256, 256, 512]
+
+    def _make_layer(self, block, planes, blocks, stride=1, gcb_config=None):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    bias_attr=False),
+                nn.BatchNorm2D(planes * block.expansion), )
+
+        layers = []
+        layers.append(
+            block(
+                self.inplanes,
+                planes,
+                stride,
+                downsample,
+                gcb_config=gcb_config))
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        f = []
+        x = self.conv1(x)  # 1,64,480,480
+
+        x = self.bn1(x)
+        x = self.relu1(x)
+
+        x = self.conv2(x)  # 1,128,480,480
+        x = self.bn2(x)
+        x = self.relu2(x)
+        # (48, 160)
+
+        x = self.maxpool1(x)  # 1,64,240,240
+        x = self.layer1(x)
+
+        x = self.conv3(x)  # 1,256,240,240
+        x = self.bn3(x)
+        x = self.relu3(x)
+        f.append(x)
+        # (24, 80)
+
+        x = self.maxpool2(x)  # 1,256,120,120
+        x = self.layer2(x)
+
+        x = self.conv4(x)  # 1,256,120,120
+        x = self.bn4(x)
+        x = self.relu4(x)
+        f.append(x)
+        # (12, 40)
+
+        x = self.maxpool3(x)  # 1,256,60,60
+
+        x = self.layer3(x)  # 1,512,60,60
+        x = self.conv5(x)  # 1,512,60,60
+        x = self.bn5(x)
+        x = self.relu5(x)
+
+        x = self.layer4(x)  # 1,512,60,60
+        x = self.conv6(x)  # 1,512,60,60
+        x = self.bn6(x)
+        x = self.relu6(x)
+        f.append(x)
+        # (6, 40)
+        return f
+
+
+class MultiAspectGCAttention(nn.Layer):
+    def __init__(self,
+                 inplanes,
+                 ratio,
+                 headers,
+                 pooling_type='att',
+                 att_scale=False,
+                 fusion_type='channel_add'):
+        super(MultiAspectGCAttention, self).__init__()
+        assert pooling_type in ['avg', 'att']
+
+        assert fusion_type in ['channel_add', 'channel_mul', 'channel_concat']
+        assert inplanes % headers == 0 and inplanes >= 8  # inplanes must be divided by headers evenly
+
+        self.headers = headers
+        self.inplanes = inplanes
+        self.ratio = ratio
+        self.planes = int(inplanes * ratio)
+        self.pooling_type = pooling_type
+        self.fusion_type = fusion_type
+        self.att_scale = False
+
+        self.single_header_inplanes = int(inplanes / headers)
+
+        if pooling_type == 'att':
+            self.conv_mask = nn.Conv2D(
+                self.single_header_inplanes, 1, kernel_size=1)
+            self.softmax = nn.Softmax(axis=2)
+        else:
+            self.avg_pool = nn.AdaptiveAvgPool2D(1)
+
+        if fusion_type == 'channel_add':
+            self.channel_add_conv = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes, self.planes, kernel_size=1),
+                nn.LayerNorm([self.planes, 1, 1]),
+                nn.ReLU(),
+                nn.Conv2D(
+                    self.planes, self.inplanes, kernel_size=1))
+        elif fusion_type == 'channel_concat':
+            self.channel_concat_conv = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes, self.planes, kernel_size=1),
+                nn.LayerNorm([self.planes, 1, 1]),
+                nn.ReLU(),
+                nn.Conv2D(
+                    self.planes, self.inplanes, kernel_size=1))
+            # for concat
+            self.cat_conv = nn.Conv2D(
+                2 * self.inplanes, self.inplanes, kernel_size=1)
+        elif fusion_type == 'channel_mul':
+            self.channel_mul_conv = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes, self.planes, kernel_size=1),
+                nn.LayerNorm([self.planes, 1, 1]),
+                nn.ReLU(),
+                nn.Conv2D(
+                    self.planes, self.inplanes, kernel_size=1))
+
+    def spatial_pool(self, x):
+        batch, channel, height, width = x.shape
+        if self.pooling_type == 'att':
+            # [N*headers, C', H , W] C = headers * C'
+            x = x.reshape([
+                batch * self.headers, self.single_header_inplanes, height, width
+            ])
+            input_x = x
+
+            # [N*headers, C', H * W] C = headers * C'
+            # input_x = input_x.view(batch, channel, height * width)
+            input_x = input_x.reshape([
+                batch * self.headers, self.single_header_inplanes,
+                height * width
+            ])
+
+            # [N*headers, 1, C', H * W]
+            input_x = input_x.unsqueeze(1)
+            # [N*headers, 1, H, W]
+            context_mask = self.conv_mask(x)
+            # [N*headers, 1, H * W]
+            context_mask = context_mask.reshape(
+                [batch * self.headers, 1, height * width])
+
+            # scale variance
+            if self.att_scale and self.headers > 1:
+                context_mask = context_mask / paddle.sqrt(
+                    self.single_header_inplanes)
+
+            # [N*headers, 1, H * W]
+            context_mask = self.softmax(context_mask)
+
+            # [N*headers, 1, H * W, 1]
+            context_mask = context_mask.unsqueeze(-1)
+            # [N*headers, 1, C', 1] = [N*headers, 1, C', H * W] * [N*headers, 1, H * W, 1]
+            context = paddle.matmul(input_x, context_mask)
+
+            # [N, headers * C', 1, 1]
+            context = context.reshape(
+                [batch, self.headers * self.single_header_inplanes, 1, 1])
+        else:
+            # [N, C, 1, 1]
+            context = self.avg_pool(x)
+
+        return context
+
+    def forward(self, x):
+        # [N, C, 1, 1]
+        context = self.spatial_pool(x)
+
+        out = x
+
+        if self.fusion_type == 'channel_mul':
+            # [N, C, 1, 1]
+            channel_mul_term = F.sigmoid(self.channel_mul_conv(context))
+            out = out * channel_mul_term
+        elif self.fusion_type == 'channel_add':
+            # [N, C, 1, 1]
+            channel_add_term = self.channel_add_conv(context)
+            out = out + channel_add_term
+        else:
+            # [N, C, 1, 1]
+            channel_concat_term = self.channel_concat_conv(context)
+
+            # use concat
+            _, C1, _, _ = channel_concat_term.shape
+            N, C2, H, W = out.shape
+
+            out = paddle.concat(
+                [out, channel_concat_term.expand([-1, -1, H, W])], axis=1)
+            out = self.cat_conv(out)
+            out = F.layer_norm(out, [self.inplanes, H, W])
+            out = F.relu(out)
+
+        return out
diff --git a/ppocr/modeling/heads/__init__.py b/ppocr/modeling/heads/__init__.py
index 1670ea38e6..da09e25c0e 100755
--- a/ppocr/modeling/heads/__init__.py
+++ b/ppocr/modeling/heads/__init__.py
@@ -41,12 +41,13 @@ def build_head(config):
     from .kie_sdmgr_head import SDMGRHead
 
     from .table_att_head import TableAttentionHead
+    from .table_master_head import TableMasterHead
 
     support_dict = [
         'DBHead', 'PSEHead', 'FCEHead', 'EASTHead', 'SASTHead', 'CTCHead',
         'ClsHead', 'AttentionHead', 'SRNHead', 'PGHead', 'Transformer',
         'TableAttentionHead', 'SARHead', 'AsterHead', 'SDMGRHead', 'PRENHead',
-        'MultiHead'
+        'MultiHead', 'TableMasterHead'
     ]
 
     #table head
diff --git a/ppocr/modeling/heads/table_att_head.py b/ppocr/modeling/heads/table_att_head.py
index e354f40d65..b64713898d 100644
--- a/ppocr/modeling/heads/table_att_head.py
+++ b/ppocr/modeling/heads/table_att_head.py
@@ -21,6 +21,8 @@ import paddle.nn as nn
 import paddle.nn.functional as F
 import numpy as np
 
+from .rec_att_head import AttentionGRUCell
+
 
 class TableAttentionHead(nn.Layer):
     def __init__(self,
@@ -28,17 +30,13 @@ class TableAttentionHead(nn.Layer):
                  hidden_size,
                  loc_type,
                  in_max_len=488,
-                 max_text_length=100,
-                 max_elem_length=800,
-                 max_cell_num=500,
+                 max_text_length=800,
                  **kwargs):
         super(TableAttentionHead, self).__init__()
         self.input_size = in_channels[-1]
         self.hidden_size = hidden_size
         self.elem_num = 30
         self.max_text_length = max_text_length
-        self.max_elem_length = max_elem_length
-        self.max_cell_num = max_cell_num
 
         self.structure_attention_cell = AttentionGRUCell(
             self.input_size, hidden_size, self.elem_num, use_gru=False)
@@ -50,11 +48,11 @@ class TableAttentionHead(nn.Layer):
             self.loc_generator = nn.Linear(hidden_size, 4)
         else:
             if self.in_max_len == 640:
-                self.loc_fea_trans = nn.Linear(400, self.max_elem_length + 1)
+                self.loc_fea_trans = nn.Linear(400, self.max_text_length + 1)
             elif self.in_max_len == 800:
-                self.loc_fea_trans = nn.Linear(625, self.max_elem_length + 1)
+                self.loc_fea_trans = nn.Linear(625, self.max_text_length + 1)
             else:
-                self.loc_fea_trans = nn.Linear(256, self.max_elem_length + 1)
+                self.loc_fea_trans = nn.Linear(256, self.max_text_length + 1)
             self.loc_generator = nn.Linear(self.input_size + hidden_size, 4)
 
     def _char_to_onehot(self, input_char, onehot_dim):
@@ -77,7 +75,7 @@ class TableAttentionHead(nn.Layer):
         output_hiddens = []
         if self.training and targets is not None:
             structure = targets[0]
-            for i in range(self.max_elem_length + 1):
+            for i in range(self.max_text_length + 1):
                 elem_onehots = self._char_to_onehot(
                     structure[:, i], onehot_dim=self.elem_num)
                 (outputs, hidden), alpha = self.structure_attention_cell(
@@ -102,9 +100,9 @@ class TableAttentionHead(nn.Layer):
             elem_onehots = None
             outputs = None
             alpha = None
-            max_elem_length = paddle.to_tensor(self.max_elem_length)
+            max_text_length = paddle.to_tensor(self.max_text_length)
             i = 0
-            while i < max_elem_length + 1:
+            while i < max_text_length + 1:
                 elem_onehots = self._char_to_onehot(
                     temp_elem, onehot_dim=self.elem_num)
                 (outputs, hidden), alpha = self.structure_attention_cell(
@@ -128,119 +126,3 @@ class TableAttentionHead(nn.Layer):
                 loc_preds = self.loc_generator(loc_concat)
                 loc_preds = F.sigmoid(loc_preds)
         return {'structure_probs': structure_probs, 'loc_preds': loc_preds}
-
-
-class AttentionGRUCell(nn.Layer):
-    def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
-        super(AttentionGRUCell, self).__init__()
-        self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
-        self.h2h = nn.Linear(hidden_size, hidden_size)
-        self.score = nn.Linear(hidden_size, 1, bias_attr=False)
-        self.rnn = nn.GRUCell(
-            input_size=input_size + num_embeddings, hidden_size=hidden_size)
-        self.hidden_size = hidden_size
-
-    def forward(self, prev_hidden, batch_H, char_onehots):
-        batch_H_proj = self.i2h(batch_H)
-        prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden), axis=1)
-        res = paddle.add(batch_H_proj, prev_hidden_proj)
-        res = paddle.tanh(res)
-        e = self.score(res)
-        alpha = F.softmax(e, axis=1)
-        alpha = paddle.transpose(alpha, [0, 2, 1])
-        context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
-        concat_context = paddle.concat([context, char_onehots], 1)
-        cur_hidden = self.rnn(concat_context, prev_hidden)
-        return cur_hidden, alpha
-
-
-class AttentionLSTM(nn.Layer):
-    def __init__(self, in_channels, out_channels, hidden_size, **kwargs):
-        super(AttentionLSTM, self).__init__()
-        self.input_size = in_channels
-        self.hidden_size = hidden_size
-        self.num_classes = out_channels
-
-        self.attention_cell = AttentionLSTMCell(
-            in_channels, hidden_size, out_channels, use_gru=False)
-        self.generator = nn.Linear(hidden_size, out_channels)
-
-    def _char_to_onehot(self, input_char, onehot_dim):
-        input_ont_hot = F.one_hot(input_char, onehot_dim)
-        return input_ont_hot
-
-    def forward(self, inputs, targets=None, batch_max_length=25):
-        batch_size = inputs.shape[0]
-        num_steps = batch_max_length
-
-        hidden = (paddle.zeros((batch_size, self.hidden_size)), paddle.zeros(
-            (batch_size, self.hidden_size)))
-        output_hiddens = []
-
-        if targets is not None:
-            for i in range(num_steps):
-                # one-hot vectors for a i-th char
-                char_onehots = self._char_to_onehot(
-                    targets[:, i], onehot_dim=self.num_classes)
-                hidden, alpha = self.attention_cell(hidden, inputs,
-                                                    char_onehots)
-
-                hidden = (hidden[1][0], hidden[1][1])
-                output_hiddens.append(paddle.unsqueeze(hidden[0], axis=1))
-            output = paddle.concat(output_hiddens, axis=1)
-            probs = self.generator(output)
-
-        else:
-            targets = paddle.zeros(shape=[batch_size], dtype="int32")
-            probs = None
-
-            for i in range(num_steps):
-                char_onehots = self._char_to_onehot(
-                    targets, onehot_dim=self.num_classes)
-                hidden, alpha = self.attention_cell(hidden, inputs,
-                                                    char_onehots)
-                probs_step = self.generator(hidden[0])
-                hidden = (hidden[1][0], hidden[1][1])
-                if probs is None:
-                    probs = paddle.unsqueeze(probs_step, axis=1)
-                else:
-                    probs = paddle.concat(
-                        [probs, paddle.unsqueeze(
-                            probs_step, axis=1)], axis=1)
-
-                next_input = probs_step.argmax(axis=1)
-
-                targets = next_input
-
-        return probs
-
-
-class AttentionLSTMCell(nn.Layer):
-    def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
-        super(AttentionLSTMCell, self).__init__()
-        self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
-        self.h2h = nn.Linear(hidden_size, hidden_size)
-        self.score = nn.Linear(hidden_size, 1, bias_attr=False)
-        if not use_gru:
-            self.rnn = nn.LSTMCell(
-                input_size=input_size + num_embeddings, hidden_size=hidden_size)
-        else:
-            self.rnn = nn.GRUCell(
-                input_size=input_size + num_embeddings, hidden_size=hidden_size)
-
-        self.hidden_size = hidden_size
-
-    def forward(self, prev_hidden, batch_H, char_onehots):
-        batch_H_proj = self.i2h(batch_H)
-        prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden[0]), axis=1)
-        res = paddle.add(batch_H_proj, prev_hidden_proj)
-        res = paddle.tanh(res)
-        e = self.score(res)
-
-        alpha = F.softmax(e, axis=1)
-        alpha = paddle.transpose(alpha, [0, 2, 1])
-        context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
-        concat_context = paddle.concat([context, char_onehots], 1)
-        cur_hidden = self.rnn(concat_context, prev_hidden)
-
-        return cur_hidden, alpha
diff --git a/ppocr/modeling/heads/table_master_head.py b/ppocr/modeling/heads/table_master_head.py
new file mode 100644
index 0000000000..acd1a9145f
--- /dev/null
+++ b/ppocr/modeling/heads/table_master_head.py
@@ -0,0 +1,276 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import copy
+import math
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+
+
+class TableMasterHead(nn.Layer):
+    """
+    Split to two transformer header at the last layer.
+    Cls_layer is used to structure token classification.
+    Bbox_layer is used to regress bbox coord.
+    """
+
+    def __init__(self,
+                 in_channels,
+                 out_channels=30,
+                 headers=8,
+                 d_ff=2048,
+                 dropout=0,
+                 max_text_length=500,
+                 point_num=4,
+                 **kwargs):
+        super(TableMasterHead, self).__init__()
+        hidden_size = in_channels[-1]
+        self.layers = clones(
+            DecoderLayer(headers, hidden_size, dropout, d_ff), 2)
+        self.cls_layer = clones(
+            DecoderLayer(headers, hidden_size, dropout, d_ff), 1)
+        self.bbox_layer = clones(
+            DecoderLayer(headers, hidden_size, dropout, d_ff), 1)
+        self.cls_fc = nn.Linear(hidden_size, out_channels)
+        self.bbox_fc = nn.Sequential(
+            # nn.Linear(hidden_size, hidden_size),
+            nn.Linear(hidden_size, point_num),
+            nn.Sigmoid())
+        self.norm = nn.LayerNorm(hidden_size)
+        self.embedding = Embeddings(d_model=hidden_size, vocab=out_channels)
+        self.positional_encoding = PositionalEncoding(d_model=hidden_size)
+
+        self.SOS = out_channels - 3
+        self.PAD = out_channels - 1
+        self.out_channels = out_channels
+        self.point_num = point_num
+        self.max_text_length = max_text_length
+
+    def make_mask(self, tgt):
+        """
+        Make mask for self attention.
+        :param src: [b, c, h, l_src]
+        :param tgt: [b, l_tgt]
+        :return:
+        """
+        trg_pad_mask = (tgt != self.PAD).unsqueeze(1).unsqueeze(3)
+
+        tgt_len = paddle.shape(tgt)[1]
+        trg_sub_mask = paddle.tril(
+            paddle.ones(
+                ([tgt_len, tgt_len]), dtype=paddle.float32))
+
+        tgt_mask = paddle.logical_and(
+            trg_pad_mask.astype(paddle.float32), trg_sub_mask)
+        return tgt_mask.astype(paddle.float32)
+
+    def decode(self, input, feature, src_mask, tgt_mask):
+        # main process of transformer decoder.
+        x = self.embedding(input)  # x: 1*x*512, feature: 1*3600,512
+        x = self.positional_encoding(x)
+
+        # origin transformer layers
+        for i, layer in enumerate(self.layers):
+            x = layer(x, feature, src_mask, tgt_mask)
+
+        # cls head
+        for layer in self.cls_layer:
+            cls_x = layer(x, feature, src_mask, tgt_mask)
+        cls_x = self.norm(cls_x)
+
+        # bbox head
+        for layer in self.bbox_layer:
+            bbox_x = layer(x, feature, src_mask, tgt_mask)
+        bbox_x = self.norm(bbox_x)
+        return self.cls_fc(cls_x), self.bbox_fc(bbox_x)
+
+    def greedy_forward(self, SOS, feature):
+        input = SOS
+        output = paddle.zeros(
+            [input.shape[0], self.max_text_length + 1, self.out_channels])
+        bbox_output = paddle.zeros(
+            [input.shape[0], self.max_text_length + 1, self.point_num])
+        max_text_length = paddle.to_tensor(self.max_text_length)
+        for i in range(max_text_length + 1):
+            target_mask = self.make_mask(input)
+            out_step, bbox_output_step = self.decode(input, feature, None,
+                                                     target_mask)
+            prob = F.softmax(out_step, axis=-1)
+            next_word = prob.argmax(axis=2, dtype="int64")
+            input = paddle.concat(
+                [input, next_word[:, -1].unsqueeze(-1)], axis=1)
+            if i == self.max_text_length:
+                output = out_step
+                bbox_output = bbox_output_step
+        return output, bbox_output
+
+    def forward_train(self, out_enc, targets):
+        # x is token of label
+        # feat is feature after backbone before pe.
+        # out_enc is feature after pe.
+        padded_targets = targets[0]
+        src_mask = None
+        tgt_mask = self.make_mask(padded_targets[:, :-1])
+        output, bbox_output = self.decode(padded_targets[:, :-1], out_enc,
+                                          src_mask, tgt_mask)
+        return {'structure_probs': output, 'loc_preds': bbox_output}
+
+    def forward_test(self, out_enc):
+        batch_size = out_enc.shape[0]
+        SOS = paddle.zeros([batch_size, 1], dtype='int64') + self.SOS
+        output, bbox_output = self.greedy_forward(SOS, out_enc)
+        # output = F.softmax(output)
+        return {'structure_probs': output, 'loc_preds': bbox_output}
+
+    def forward(self, feat, targets=None):
+        feat = feat[-1]
+        b, c, h, w = feat.shape
+        feat = feat.reshape([b, c, h * w])  # flatten 2D feature map
+        feat = feat.transpose((0, 2, 1))
+        out_enc = self.positional_encoding(feat)
+        if self.training:
+            return self.forward_train(out_enc, targets)
+
+        return self.forward_test(out_enc)
+
+
+class DecoderLayer(nn.Layer):
+    """
+    Decoder is made of self attention, srouce attention and feed forward.
+    """
+
+    def __init__(self, headers, d_model, dropout, d_ff):
+        super(DecoderLayer, self).__init__()
+        self.self_attn = MultiHeadAttention(headers, d_model, dropout)
+        self.src_attn = MultiHeadAttention(headers, d_model, dropout)
+        self.feed_forward = FeedForward(d_model, d_ff, dropout)
+        self.sublayer = clones(SubLayerConnection(d_model, dropout), 3)
+
+    def forward(self, x, feature, src_mask, tgt_mask):
+        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, tgt_mask))
+        x = self.sublayer[1](
+            x, lambda x: self.src_attn(x, feature, feature, src_mask))
+        return self.sublayer[2](x, self.feed_forward)
+
+
+class MultiHeadAttention(nn.Layer):
+    def __init__(self, headers, d_model, dropout):
+        super(MultiHeadAttention, self).__init__()
+
+        assert d_model % headers == 0
+        self.d_k = int(d_model / headers)
+        self.headers = headers
+        self.linears = clones(nn.Linear(d_model, d_model), 4)
+        self.attn = None
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, query, key, value, mask=None):
+        B = query.shape[0]
+
+        # 1) Do all the linear projections in batch from d_model => h x d_k
+        query, key, value = \
+            [l(x).reshape([B, 0, self.headers, self.d_k]).transpose([0, 2, 1, 3])
+             for l, x in zip(self.linears, (query, key, value))]
+        # 2) Apply attention on all the projected vectors in batch
+        x, self.attn = self_attention(
+            query, key, value, mask=mask, dropout=self.dropout)
+        x = x.transpose([0, 2, 1, 3]).reshape([B, 0, self.headers * self.d_k])
+        return self.linears[-1](x)
+
+
+class FeedForward(nn.Layer):
+    def __init__(self, d_model, d_ff, dropout):
+        super(FeedForward, self).__init__()
+        self.w_1 = nn.Linear(d_model, d_ff)
+        self.w_2 = nn.Linear(d_ff, d_model)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x):
+        return self.w_2(self.dropout(F.relu(self.w_1(x))))
+
+
+class SubLayerConnection(nn.Layer):
+    """
+    A residual connection followed by a layer norm.
+    Note for code simplicity the norm is first as opposed to last.
+    """
+
+    def __init__(self, size, dropout):
+        super(SubLayerConnection, self).__init__()
+        self.norm = nn.LayerNorm(size)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x, sublayer):
+        return x + self.dropout(sublayer(self.norm(x)))
+
+
+def masked_fill(x, mask, value):
+    mask = mask.astype(x.dtype)
+    return x * paddle.logical_not(mask).astype(x.dtype) + mask * value
+
+
+def self_attention(query, key, value, mask=None, dropout=None):
+    """
+    Compute 'Scale Dot Product Attention'
+    """
+    d_k = value.shape[-1]
+
+    score = paddle.matmul(query, key.transpose([0, 1, 3, 2]) / math.sqrt(d_k))
+    if mask is not None:
+        # score = score.masked_fill(mask == 0, -1e9) # b, h, L, L
+        score = masked_fill(score, mask == 0, -6.55e4)  # for fp16
+
+    p_attn = F.softmax(score, axis=-1)
+
+    if dropout is not None:
+        p_attn = dropout(p_attn)
+    return paddle.matmul(p_attn, value), p_attn
+
+
+def clones(module, N):
+    """ Produce N identical layers """
+    return nn.LayerList([copy.deepcopy(module) for _ in range(N)])
+
+
+class Embeddings(nn.Layer):
+    def __init__(self, d_model, vocab):
+        super(Embeddings, self).__init__()
+        self.lut = nn.Embedding(vocab, d_model)
+        self.d_model = d_model
+
+    def forward(self, *input):
+        x = input[0]
+        return self.lut(x) * math.sqrt(self.d_model)
+
+
+class PositionalEncoding(nn.Layer):
+    """ Implement the PE function. """
+
+    def __init__(self, d_model, dropout=0., max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        # Compute the positional encodings once in log space.
+        pe = paddle.zeros([max_len, d_model])
+        position = paddle.arange(0, max_len).unsqueeze(1).astype('float32')
+        div_term = paddle.exp(
+            paddle.arange(0, d_model, 2) * -math.log(10000.0) / d_model)
+        pe[:, 0::2] = paddle.sin(position * div_term)
+        pe[:, 1::2] = paddle.cos(position * div_term)
+        pe = pe.unsqueeze(0)
+        self.register_buffer('pe', pe)
+
+    def forward(self, feat, **kwargs):
+        feat = feat + self.pe[:, :paddle.shape(feat)[1]]  # pe 1*5000*512
+        return self.dropout(feat)
diff --git a/ppocr/optimizer/learning_rate.py b/ppocr/optimizer/learning_rate.py
index fe251f36e7..d96ab51896 100644
--- a/ppocr/optimizer/learning_rate.py
+++ b/ppocr/optimizer/learning_rate.py
@@ -308,3 +308,46 @@ class Const(object):
                 end_lr=self.learning_rate,
                 last_epoch=self.last_epoch)
         return learning_rate
+
+
+class MultiStepDecay(object):
+    """
+    Piecewise learning rate decay
+    Args:
+        step_each_epoch(int): steps each epoch
+        learning_rate (float): The initial learning rate. It is a python float number.
+        step_size (int): the interval to update.
+        gamma (float, optional): The Ratio that the learning rate will be reduced. ``new_lr = origin_lr * gamma`` .
+            It should be less than 1.0. Default: 0.1.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(self,
+                 learning_rate,
+                 milestones,
+                 step_each_epoch,
+                 gamma,
+                 warmup_epoch=0,
+                 last_epoch=-1,
+                 **kwargs):
+        super(MultiStepDecay, self).__init__()
+        self.milestones = [step_each_epoch * e for e in milestones]
+        self.learning_rate = learning_rate
+        self.gamma = gamma
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = lr.MultiStepDecay(
+            learning_rate=self.learning_rate,
+            milestones=self.milestones,
+            gamma=self.gamma,
+            last_epoch=self.last_epoch)
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch)
+        return learning_rate
\ No newline at end of file
diff --git a/ppocr/postprocess/__init__.py b/ppocr/postprocess/__init__.py
index f50b5f1c5f..4a08f1531f 100644
--- a/ppocr/postprocess/__init__.py
+++ b/ppocr/postprocess/__init__.py
@@ -26,8 +26,9 @@ from .east_postprocess import EASTPostProcess
 from .sast_postprocess import SASTPostProcess
 from .fce_postprocess import FCEPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, \
-    DistillationCTCLabelDecode, TableLabelDecode, NRTRLabelDecode, SARLabelDecode, \
+    DistillationCTCLabelDecode, NRTRLabelDecode, SARLabelDecode, \
     SEEDLabelDecode, PRENLabelDecode
+from .table_postprocess import TableMasterLabelDecode, TableLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
 from .vqa_token_ser_layoutlm_postprocess import VQASerTokenLayoutLMPostProcess
@@ -42,7 +43,7 @@ def build_post_process(config, global_config=None):
         'DistillationDBPostProcess', 'NRTRLabelDecode', 'SARLabelDecode',
         'SEEDLabelDecode', 'VQASerTokenLayoutLMPostProcess',
         'VQAReTokenLayoutLMPostProcess', 'PRENLabelDecode',
-        'DistillationSARLabelDecode'
+        'DistillationSARLabelDecode', 'TableMasterLabelDecode'
     ]
 
     if config['name'] == 'PSEPostProcess':
diff --git a/ppocr/postprocess/rec_postprocess.py b/ppocr/postprocess/rec_postprocess.py
index bf0fd890bf..0d01b34210 100644
--- a/ppocr/postprocess/rec_postprocess.py
+++ b/ppocr/postprocess/rec_postprocess.py
@@ -444,146 +444,6 @@ class SRNLabelDecode(BaseRecLabelDecode):
         return idx
 
 
-class TableLabelDecode(object):
-    """  """
-
-    def __init__(self, character_dict_path, **kwargs):
-        list_character, list_elem = self.load_char_elem_dict(
-            character_dict_path)
-        list_character = self.add_special_char(list_character)
-        list_elem = self.add_special_char(list_elem)
-        self.dict_character = {}
-        self.dict_idx_character = {}
-        for i, char in enumerate(list_character):
-            self.dict_idx_character[i] = char
-            self.dict_character[char] = i
-        self.dict_elem = {}
-        self.dict_idx_elem = {}
-        for i, elem in enumerate(list_elem):
-            self.dict_idx_elem[i] = elem
-            self.dict_elem[elem] = i
-
-    def load_char_elem_dict(self, character_dict_path):
-        list_character = []
-        list_elem = []
-        with open(character_dict_path, "rb") as fin:
-            lines = fin.readlines()
-            substr = lines[0].decode('utf-8').strip("\n").strip("\r\n").split(
-                "\t")
-            character_num = int(substr[0])
-            elem_num = int(substr[1])
-            for cno in range(1, 1 + character_num):
-                character = lines[cno].decode('utf-8').strip("\n").strip("\r\n")
-                list_character.append(character)
-            for eno in range(1 + character_num, 1 + character_num + elem_num):
-                elem = lines[eno].decode('utf-8').strip("\n").strip("\r\n")
-                list_elem.append(elem)
-        return list_character, list_elem
-
-    def add_special_char(self, list_character):
-        self.beg_str = "sos"
-        self.end_str = "eos"
-        list_character = [self.beg_str] + list_character + [self.end_str]
-        return list_character
-
-    def __call__(self, preds):
-        structure_probs = preds['structure_probs']
-        loc_preds = preds['loc_preds']
-        if isinstance(structure_probs, paddle.Tensor):
-            structure_probs = structure_probs.numpy()
-        if isinstance(loc_preds, paddle.Tensor):
-            loc_preds = loc_preds.numpy()
-        structure_idx = structure_probs.argmax(axis=2)
-        structure_probs = structure_probs.max(axis=2)
-        structure_str, structure_pos, result_score_list, result_elem_idx_list = self.decode(
-            structure_idx, structure_probs, 'elem')
-        res_html_code_list = []
-        res_loc_list = []
-        batch_num = len(structure_str)
-        for bno in range(batch_num):
-            res_loc = []
-            for sno in range(len(structure_str[bno])):
-                text = structure_str[bno][sno]
-                if text in ['<td>', '<td']:
-                    pos = structure_pos[bno][sno]
-                    res_loc.append(loc_preds[bno, pos])
-            res_html_code = ''.join(structure_str[bno])
-            res_loc = np.array(res_loc)
-            res_html_code_list.append(res_html_code)
-            res_loc_list.append(res_loc)
-        return {
-            'res_html_code': res_html_code_list,
-            'res_loc': res_loc_list,
-            'res_score_list': result_score_list,
-            'res_elem_idx_list': result_elem_idx_list,
-            'structure_str_list': structure_str
-        }
-
-    def decode(self, text_index, structure_probs, char_or_elem):
-        """convert text-label into text-index.
-        """
-        if char_or_elem == "char":
-            current_dict = self.dict_idx_character
-        else:
-            current_dict = self.dict_idx_elem
-            ignored_tokens = self.get_ignored_tokens('elem')
-            beg_idx, end_idx = ignored_tokens
-
-        result_list = []
-        result_pos_list = []
-        result_score_list = []
-        result_elem_idx_list = []
-        batch_size = len(text_index)
-        for batch_idx in range(batch_size):
-            char_list = []
-            elem_pos_list = []
-            elem_idx_list = []
-            score_list = []
-            for idx in range(len(text_index[batch_idx])):
-                tmp_elem_idx = int(text_index[batch_idx][idx])
-                if idx > 0 and tmp_elem_idx == end_idx:
-                    break
-                if tmp_elem_idx in ignored_tokens:
-                    continue
-
-                char_list.append(current_dict[tmp_elem_idx])
-                elem_pos_list.append(idx)
-                score_list.append(structure_probs[batch_idx, idx])
-                elem_idx_list.append(tmp_elem_idx)
-            result_list.append(char_list)
-            result_pos_list.append(elem_pos_list)
-            result_score_list.append(score_list)
-            result_elem_idx_list.append(elem_idx_list)
-        return result_list, result_pos_list, result_score_list, result_elem_idx_list
-
-    def get_ignored_tokens(self, char_or_elem):
-        beg_idx = self.get_beg_end_flag_idx("beg", char_or_elem)
-        end_idx = self.get_beg_end_flag_idx("end", char_or_elem)
-        return [beg_idx, end_idx]
-
-    def get_beg_end_flag_idx(self, beg_or_end, char_or_elem):
-        if char_or_elem == "char":
-            if beg_or_end == "beg":
-                idx = self.dict_character[self.beg_str]
-            elif beg_or_end == "end":
-                idx = self.dict_character[self.end_str]
-            else:
-                assert False, "Unsupport type %s in get_beg_end_flag_idx of char" \
-                              % beg_or_end
-        elif char_or_elem == "elem":
-            if beg_or_end == "beg":
-                idx = self.dict_elem[self.beg_str]
-            elif beg_or_end == "end":
-                idx = self.dict_elem[self.end_str]
-            else:
-                assert False, "Unsupport type %s in get_beg_end_flag_idx of elem" \
-                              % beg_or_end
-        else:
-            assert False, "Unsupport type %s in char_or_elem" \
-                          % char_or_elem
-        return idx
-
-
 class SARLabelDecode(BaseRecLabelDecode):
     """ Convert between text-label and text-index """
 
diff --git a/ppocr/postprocess/table_postprocess.py b/ppocr/postprocess/table_postprocess.py
new file mode 100644
index 0000000000..4396ec4f70
--- /dev/null
+++ b/ppocr/postprocess/table_postprocess.py
@@ -0,0 +1,160 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+import paddle
+
+from .rec_postprocess import AttnLabelDecode
+
+
+class TableLabelDecode(AttnLabelDecode):
+    """  """
+
+    def __init__(self, character_dict_path, **kwargs):
+        super(TableLabelDecode, self).__init__(character_dict_path)
+        self.td_token = ['<td>', '<td', '<eb></eb>', '<td></td>']
+
+    def __call__(self, preds, batch=None):
+        structure_probs = preds['structure_probs']
+        bbox_preds = preds['loc_preds']
+        if isinstance(structure_probs, paddle.Tensor):
+            structure_probs = structure_probs.numpy()
+        if isinstance(bbox_preds, paddle.Tensor):
+            bbox_preds = bbox_preds.numpy()
+        shape_list = batch[-1]
+        result = self.decode(structure_probs, bbox_preds, shape_list)
+        if len(batch) == 1:  # only contains shape
+            return result
+
+        label_decode_result = self.decode_label(batch)
+        return result, label_decode_result
+
+    def decode(self, structure_probs, bbox_preds, shape_list):
+        """convert text-label into text-index.
+        """
+        ignored_tokens = self.get_ignored_tokens()
+        end_idx = self.dict[self.end_str]
+
+        structure_idx = structure_probs.argmax(axis=2)
+        structure_probs = structure_probs.max(axis=2)
+
+        structure_batch_list = []
+        bbox_batch_list = []
+        batch_size = len(structure_idx)
+        for batch_idx in range(batch_size):
+            structure_list = []
+            bbox_list = []
+            score_list = []
+            for idx in range(len(structure_idx[batch_idx])):
+                char_idx = int(structure_idx[batch_idx][idx])
+                if idx > 0 and char_idx == end_idx:
+                    break
+                if char_idx in ignored_tokens:
+                    continue
+                text = self.character[char_idx]
+                if text in self.td_token:
+                    bbox = bbox_preds[batch_idx, idx]
+                    bbox = self._bbox_decode(bbox, shape_list[batch_idx])
+                    bbox_list.append(bbox)
+                structure_list.append(text)
+                score_list.append(structure_probs[batch_idx, idx])
+            structure_batch_list.append([structure_list, np.mean(score_list)])
+            bbox_batch_list.append(np.array(bbox_list))
+        result = {
+            'bbox_batch_list': bbox_batch_list,
+            'structure_batch_list': structure_batch_list,
+        }
+        return result
+
+    def decode_label(self, batch):
+        """convert text-label into text-index.
+        """
+        structure_idx = batch[1]
+        gt_bbox_list = batch[2]
+        shape_list = batch[-1]
+        ignored_tokens = self.get_ignored_tokens()
+        end_idx = self.dict[self.end_str]
+
+        structure_batch_list = []
+        bbox_batch_list = []
+        batch_size = len(structure_idx)
+        for batch_idx in range(batch_size):
+            structure_list = []
+            bbox_list = []
+            for idx in range(len(structure_idx[batch_idx])):
+                char_idx = int(structure_idx[batch_idx][idx])
+                if idx > 0 and char_idx == end_idx:
+                    break
+                if char_idx in ignored_tokens:
+                    continue
+                structure_list.append(self.character[char_idx])
+
+                bbox = gt_bbox_list[batch_idx][idx]
+                if bbox.sum() != 0:
+                    bbox = self._bbox_decode(bbox, shape_list[batch_idx])
+                    bbox_list.append(bbox)
+            structure_batch_list.append(structure_list)
+            bbox_batch_list.append(bbox_list)
+        result = {
+            'bbox_batch_list': bbox_batch_list,
+            'structure_batch_list': structure_batch_list,
+        }
+        return result
+
+    def _bbox_decode(self, bbox, shape):
+        h, w, ratio_h, ratio_w, pad_h, pad_w = shape
+        src_h = h / ratio_h
+        src_w = w / ratio_w
+        bbox[0::2] *= src_w
+        bbox[1::2] *= src_h
+        return bbox
+
+
+class TableMasterLabelDecode(TableLabelDecode):
+    """  """
+
+    def __init__(self, character_dict_path, box_shape='ori', **kwargs):
+        super(TableMasterLabelDecode, self).__init__(character_dict_path)
+        self.box_shape = box_shape
+        assert box_shape in [
+            'ori', 'pad'
+        ], 'The shape used for box normalization must be ori or pad'
+
+    def add_special_char(self, dict_character):
+        self.beg_str = '<SOS>'
+        self.end_str = '<EOS>'
+        self.unknown_str = '<UKN>'
+        self.pad_str = '<PAD>'
+        dict_character = dict_character
+        dict_character = dict_character + [
+            self.unknown_str, self.beg_str, self.end_str, self.pad_str
+        ]
+        return dict_character
+
+    def get_ignored_tokens(self):
+        pad_idx = self.dict[self.pad_str]
+        start_idx = self.dict[self.beg_str]
+        end_idx = self.dict[self.end_str]
+        unknown_idx = self.dict[self.unknown_str]
+        return [start_idx, end_idx, pad_idx, unknown_idx]
+
+    def _bbox_decode(self, bbox, shape):
+        h, w, ratio_h, ratio_w, pad_h, pad_w = shape
+        if self.box_shape == 'pad':
+            h, w = pad_h, pad_w
+        bbox[0::2] *= w
+        bbox[1::2] *= h
+        bbox[0::2] /= ratio_w
+        bbox[1::2] /= ratio_h
+        return bbox
diff --git a/ppocr/utils/dict/table_master_structure_dict.txt b/ppocr/utils/dict/table_master_structure_dict.txt
new file mode 100644
index 0000000000..95ab2539a7
--- /dev/null
+++ b/ppocr/utils/dict/table_master_structure_dict.txt
@@ -0,0 +1,39 @@
+<thead>
+<tr>
+<td></td>
+</tr>
+</thead>
+<tbody>
+<eb></eb>
+</tbody>
+<td
+ colspan="5"
+>
+</td>
+ colspan="2"
+ colspan="3"
+<eb2></eb2>
+<eb1></eb1>
+ rowspan="2"
+ colspan="4"
+ colspan="6"
+ rowspan="3"
+ colspan="9"
+ colspan="10"
+ colspan="7"
+ rowspan="4"
+ rowspan="5"
+ rowspan="9"
+ colspan="8"
+ rowspan="8"
+ rowspan="6"
+ rowspan="7"
+ rowspan="10"
+<eb3></eb3>
+<eb4></eb4>
+<eb5></eb5>
+<eb6></eb6>
+<eb7></eb7>
+<eb8></eb8>
+<eb9></eb9>
+<eb10></eb10>
diff --git a/ppocr/utils/dict/table_structure_dict.txt b/ppocr/utils/dict/table_structure_dict.txt
index 9c4531e5f3..8edb10b881 100644
--- a/ppocr/utils/dict/table_structure_dict.txt
+++ b/ppocr/utils/dict/table_structure_dict.txt
@@ -1,281 +1,3 @@
-277	28	1267	1186
-<b>
-V
-a
-r
-i
-b
-l
-e
-</b>
-H
-z
-d
- 
-t
-o
-9
-5
-%
-C
-I
-<i>
-p
-</i>
-v
-u
-*
-A
-g
-(
-m
-n
-)
-0
-.
-7
-1
-6
-≤
->
-8
-3
-–
-2
-G
-4
-M
-F
-T
-y
-f
-s
-L
-w
-c
-U
-h
-D
-S
-Q
-R
-x
-P
--
-E
-O
-/
-k
-,
-+
-N
-K
-q
-′
-[
-]
-<
-≥
-<sup>
-−
-</sup>
-μ
-±
-J
-j
-W
-_
-Δ
-B
-“
-:
-Y
-α
-λ
-;
-<sub>
-</sub>
-?
-∼
-=
-°
-#
-̊
-̈
-̂
-’
-Z
-X
-∗
-—
-β
-'
-†
-~
-@
-"
-γ
-↓
-↑
-&
-‡
-χ
-”
-σ
-§
-|
-¶
-‐
-×
-$
-→
-√
-✓
-‘
-\
-∞
-π
-•
-®
-^
-∆
-≧
-<underline>
-</underline>
-́
-♀
-♂
-‒
-⁎
-▲
-·
-£
-φ
-Ψ
-ß
-△
-☆
-▪
-η
-€
-∧
-̃
-Φ
-ρ
-̄
-δ
-‰
-̧
-Ω
-♦
-{
-}
-̀
-∑
-∫
-ø
-κ
-ε
-¥
-※
-`
-ω
-Σ
-➔
-‖
-Β
-̸
-

-─
-●
-⩾
-Χ
-Α
-⋅
-◆
-★
-■
-ψ
-ǂ
-□
-ζ
-!
-Γ
-↔
-θ
-⁄
-〈
-〉
-―
-υ
-τ
-⋆
-Ø
-©
-∥
-С
-˂
-➢
-ɛ
-⁡
-✗
-←
-○
-¢
-⩽
-∖
-˃
-­
-≈
-Π
-̌
-≦
-∅
-ᅟ
-<overline>
-</overline>
-∣
-¤
-♯
-̆
-ξ
-÷
-▼
-
-ι
-ν
-║
-<strike>
-</strike>
-◦
-​
-◊
-∙
-«
-»
-ł
-ı
-Θ
-∈
-„
-∘
-✔
-̇
-æ
-ʹ
-ˆ
-♣
-⇓
-∩
-⊕
-⇒
-⇑
-̨
-Ι
-Λ
-⋯
-А
-⋮
 <thead>
 <tr>
 <td>
@@ -303,2457 +25,4 @@ $
  rowspan="8"
  rowspan="6"
  rowspan="7"
- rowspan="10"
-0	2924682
-1	3405345
-2	2363468
-3	2709165
-4	4078680
-5	3250792
-6	1923159
-7	1617890
-8	1450532
-9	1717624
-10	1477550
-11	1489223
-12	915528
-13	819193
-14	593660
-15	518924
-16	682065
-17	494584
-18	400591
-19	396421
-20	340994
-21	280688
-22	250328
-23	226786
-24	199927
-25	182707
-26	164629
-27	141613
-28	127554
-29	116286
-30	107682
-31	96367
-32	88002
-33	79234
-34	72186
-35	65921
-36	60374
-37	55976
-38	52166
-39	47414
-40	44932
-41	41279
-42	38232
-43	35463
-44	33703
-45	30557
-46	29639
-47	27000
-48	25447
-49	23186
-50	22093
-51	20412
-52	19844
-53	18261
-54	17561
-55	16499
-56	15597
-57	14558
-58	14372
-59	13445
-60	13514
-61	12058
-62	11145
-63	10767
-64	10370
-65	9630
-66	9337
-67	8881
-68	8727
-69	8060
-70	7994
-71	7740
-72	7189
-73	6729
-74	6749
-75	6548
-76	6321
-77	5957
-78	5740
-79	5407
-80	5370
-81	5035
-82	4921
-83	4656
-84	4600
-85	4519
-86	4277
-87	4023
-88	3939
-89	3910
-90	3861
-91	3560
-92	3483
-93	3406
-94	3346
-95	3229
-96	3122
-97	3086
-98	3001
-99	2884
-100	2822
-101	2677
-102	2670
-103	2610
-104	2452
-105	2446
-106	2400
-107	2300
-108	2316
-109	2196
-110	2089
-111	2083
-112	2041
-113	1881
-114	1838
-115	1896
-116	1795
-117	1786
-118	1743
-119	1765
-120	1750
-121	1683
-122	1563
-123	1499
-124	1513
-125	1462
-126	1388
-127	1441
-128	1417
-129	1392
-130	1306
-131	1321
-132	1274
-133	1294
-134	1240
-135	1126
-136	1157
-137	1130
-138	1084
-139	1130
-140	1083
-141	1040
-142	980
-143	1031
-144	974
-145	980
-146	932
-147	898
-148	960
-149	907
-150	852
-151	912
-152	859
-153	847
-154	876
-155	792
-156	791
-157	765
-158	788
-159	787
-160	744
-161	673
-162	683
-163	697
-164	666
-165	680
-166	632
-167	677
-168	657
-169	618
-170	587
-171	585
-172	567
-173	549
-174	562
-175	548
-176	542
-177	539
-178	542
-179	549
-180	547
-181	526
-182	525
-183	514
-184	512
-185	505
-186	515
-187	467
-188	475
-189	458
-190	435
-191	443
-192	427
-193	424
-194	404
-195	389
-196	429
-197	404
-198	386
-199	351
-200	388
-201	408
-202	361
-203	346
-204	324
-205	361
-206	363
-207	364
-208	323
-209	336
-210	342
-211	315
-212	325
-213	328
-214	314
-215	327
-216	320
-217	300
-218	295
-219	315
-220	310
-221	295
-222	275
-223	248
-224	274
-225	232
-226	293
-227	259
-228	286
-229	263
-230	242
-231	214
-232	261
-233	231
-234	211
-235	250
-236	233
-237	206
-238	224
-239	210
-240	233
-241	223
-242	216
-243	222
-244	207
-245	212
-246	196
-247	205
-248	201
-249	202
-250	211
-251	201
-252	215
-253	179
-254	163
-255	179
-256	191
-257	188
-258	196
-259	150
-260	154
-261	176
-262	211
-263	166
-264	171
-265	165
-266	149
-267	182
-268	159
-269	161
-270	164
-271	161
-272	141
-273	151
-274	127
-275	129
-276	142
-277	158
-278	148
-279	135
-280	127
-281	134
-282	138
-283	131
-284	126
-285	125
-286	130
-287	126
-288	135
-289	125
-290	135
-291	131
-292	95
-293	135
-294	106
-295	117
-296	136
-297	128
-298	128
-299	118
-300	109
-301	112
-302	117
-303	108
-304	120
-305	100
-306	95
-307	108
-308	112
-309	77
-310	120
-311	104
-312	109
-313	89
-314	98
-315	82
-316	98
-317	93
-318	77
-319	93
-320	77
-321	98
-322	93
-323	86
-324	89
-325	73
-326	70
-327	71
-328	77
-329	87
-330	77
-331	93
-332	100
-333	83
-334	72
-335	74
-336	69
-337	77
-338	68
-339	78
-340	90
-341	98
-342	75
-343	80
-344	63
-345	71
-346	83
-347	66
-348	71
-349	70
-350	62
-351	62
-352	59
-353	63
-354	62
-355	52
-356	64
-357	64
-358	56
-359	49
-360	57
-361	63
-362	60
-363	68
-364	62
-365	55
-366	54
-367	40
-368	75
-369	70
-370	53
-371	58
-372	57
-373	55
-374	69
-375	57
-376	53
-377	43
-378	45
-379	47
-380	56
-381	51
-382	59
-383	51
-384	43
-385	34
-386	57
-387	49
-388	39
-389	46
-390	48
-391	43
-392	40
-393	54
-394	50
-395	41
-396	43
-397	33
-398	27
-399	49
-400	44
-401	44
-402	38
-403	30
-404	32
-405	37
-406	39
-407	42
-408	53
-409	39
-410	34
-411	31
-412	32
-413	52
-414	27
-415	41
-416	34
-417	36
-418	50
-419	35
-420	32
-421	33
-422	45
-423	35
-424	40
-425	29
-426	41
-427	40
-428	39
-429	32
-430	31
-431	34
-432	29
-433	27
-434	26
-435	22
-436	34
-437	28
-438	30
-439	38
-440	35
-441	36
-442	36
-443	27
-444	24
-445	33
-446	31
-447	25
-448	33
-449	27
-450	32
-451	46
-452	31
-453	35
-454	35
-455	34
-456	26
-457	21
-458	25
-459	26
-460	24
-461	27
-462	33
-463	30
-464	35
-465	21
-466	32
-467	19
-468	27
-469	16
-470	28
-471	26
-472	27
-473	26
-474	25
-475	25
-476	27
-477	20
-478	28
-479	22
-480	23
-481	16
-482	25
-483	27
-484	19
-485	23
-486	19
-487	15
-488	15
-489	23
-490	24
-491	19
-492	20
-493	18
-494	17
-495	30
-496	28
-497	20
-498	29
-499	17
-500	19
-501	21
-502	15
-503	24
-504	15
-505	19
-506	25
-507	16
-508	23
-509	26
-510	21
-511	15
-512	12
-513	16
-514	18
-515	24
-516	26
-517	18
-518	8
-519	25
-520	14
-521	8
-522	24
-523	20
-524	18
-525	15
-526	13
-527	17
-528	18
-529	22
-530	21
-531	9
-532	16
-533	17
-534	13
-535	17
-536	15
-537	13
-538	20
-539	13
-540	19
-541	29
-542	10
-543	8
-544	18
-545	13
-546	9
-547	18
-548	10
-549	18
-550	18
-551	9
-552	9
-553	15
-554	13
-555	15
-556	14
-557	14
-558	18
-559	8
-560	13
-561	9
-562	7
-563	12
-564	6
-565	9
-566	9
-567	18
-568	9
-569	10
-570	13
-571	14
-572	13
-573	21
-574	8
-575	16
-576	12
-577	9
-578	16
-579	17
-580	22
-581	6
-582	14
-583	13
-584	15
-585	11
-586	13
-587	5
-588	12
-589	13
-590	15
-591	13
-592	15
-593	12
-594	7
-595	18
-596	12
-597	13
-598	13
-599	13
-600	12
-601	12
-602	10
-603	11
-604	6
-605	6
-606	2
-607	9
-608	8
-609	12
-610	9
-611	12
-612	13
-613	12
-614	14
-615	9
-616	8
-617	9
-618	14
-619	13
-620	12
-621	6
-622	8
-623	8
-624	8
-625	12
-626	8
-627	7
-628	5
-629	8
-630	12
-631	6
-632	10
-633	10
-634	7
-635	8
-636	9
-637	6
-638	9
-639	4
-640	12
-641	4
-642	3
-643	11
-644	10
-645	6
-646	12
-647	12
-648	4
-649	4
-650	9
-651	8
-652	6
-653	5
-654	14
-655	10
-656	11
-657	8
-658	5
-659	5
-660	9
-661	13
-662	4
-663	5
-664	9
-665	11
-666	12
-667	7
-668	13
-669	2
-670	1
-671	7
-672	7
-673	7
-674	10
-675	9
-676	6
-677	5
-678	7
-679	6
-680	3
-681	3
-682	4
-683	9
-684	8
-685	5
-686	3
-687	11
-688	9
-689	2
-690	6
-691	5
-692	9
-693	5
-694	6
-695	5
-696	9
-697	8
-698	3
-699	7
-700	5
-701	9
-702	8
-703	7
-704	2
-705	3
-706	7
-707	6
-708	6
-709	10
-710	2
-711	10
-712	6
-713	7
-714	5
-715	6
-716	4
-717	6
-718	8
-719	4
-720	6
-721	7
-722	5
-723	7
-724	3
-725	10
-726	10
-727	3
-728	7
-729	7
-730	5
-731	2
-732	1
-733	5
-734	1
-735	5
-736	6
-737	2
-738	2
-739	3
-740	7
-741	2
-742	7
-743	4
-744	5
-745	4
-746	5
-747	3
-748	1
-749	4
-750	4
-751	2
-752	4
-753	6
-754	6
-755	6
-756	3
-757	2
-758	5
-759	5
-760	3
-761	4
-762	2
-763	1
-764	8
-765	3
-766	4
-767	3
-768	1
-769	5
-770	3
-771	3
-772	4
-773	4
-774	1
-775	3
-776	2
-777	2
-778	3
-779	3
-780	1
-781	4
-782	3
-783	4
-784	6
-785	3
-786	5
-787	4
-788	2
-789	4
-790	5
-791	4
-792	6
-794	4
-795	1
-796	1
-797	4
-798	2
-799	3
-800	3
-801	1
-802	5
-803	5
-804	3
-805	3
-806	3
-807	4
-808	4
-809	2
-811	5
-812	4
-813	6
-814	3
-815	2
-816	2
-817	3
-818	5
-819	3
-820	1
-821	1
-822	4
-823	3
-824	4
-825	8
-826	3
-827	5
-828	5
-829	3
-830	6
-831	3
-832	4
-833	8
-834	5
-835	3
-836	3
-837	2
-838	4
-839	2
-840	1
-841	3
-842	2
-843	1
-844	3
-846	4
-847	4
-848	3
-849	3
-850	2
-851	3
-853	1
-854	4
-855	4
-856	2
-857	4
-858	1
-859	2
-860	5
-861	1
-862	1
-863	4
-864	2
-865	2
-867	5
-868	1
-869	4
-870	1
-871	1
-872	1
-873	2
-875	5
-876	3
-877	1
-878	3
-879	3
-880	3
-881	2
-882	1
-883	6
-884	2
-885	2
-886	1
-887	1
-888	3
-889	2
-890	2
-891	3
-892	1
-893	3
-894	1
-895	5
-896	1
-897	3
-899	2
-900	2
-902	1
-903	2
-904	4
-905	4
-906	3
-907	1
-908	1
-909	2
-910	5
-911	2
-912	3
-914	1
-915	1
-916	2
-918	2
-919	2
-920	4
-921	4
-922	1
-923	1
-924	4
-925	5
-926	1
-928	2
-929	1
-930	1
-931	1
-932	1
-933	1
-934	2
-935	1
-936	1
-937	1
-938	2
-939	1
-941	1
-942	4
-944	2
-945	2
-946	2
-947	1
-948	1
-950	1
-951	2
-953	1
-954	2
-955	1
-956	1
-957	2
-958	1
-960	3
-962	4
-963	1
-964	1
-965	3
-966	2
-967	2
-968	1
-969	3
-970	3
-972	1
-974	4
-975	3
-976	3
-977	2
-979	2
-980	1
-981	1
-983	5
-984	1
-985	3
-986	1
-987	2
-988	4
-989	2
-991	2
-992	2
-993	1
-994	1
-996	2
-997	2
-998	1
-999	3
-1000	2
-1001	1
-1002	3
-1003	3
-1004	2
-1005	3
-1006	1
-1007	2
-1009	1
-1011	1
-1013	3
-1014	1
-1016	2
-1017	1
-1018	1
-1019	1
-1020	4
-1021	1
-1022	2
-1025	1
-1026	1
-1027	2
-1028	1
-1030	1
-1031	2
-1032	4
-1034	3
-1035	2
-1036	1
-1038	1
-1039	1
-1040	1
-1041	1
-1042	2
-1043	1
-1044	2
-1045	4
-1048	1
-1050	1
-1051	1
-1052	2
-1054	1
-1055	3
-1056	2
-1057	1
-1059	1
-1061	2
-1063	1
-1064	1
-1065	1
-1066	1
-1067	1
-1068	1
-1069	2
-1074	1
-1075	1
-1077	1
-1078	1
-1079	1
-1082	1
-1085	1
-1088	1
-1090	1
-1091	1
-1092	2
-1094	2
-1097	2
-1098	1
-1099	2
-1101	2
-1102	1
-1104	1
-1105	1
-1107	1
-1109	1
-1111	2
-1112	1
-1114	2
-1115	2
-1116	2
-1117	1
-1118	1
-1119	1
-1120	1
-1122	1
-1123	1
-1127	1
-1128	3
-1132	2
-1138	3
-1142	1
-1145	4
-1150	1
-1153	2
-1154	1
-1158	1
-1159	1
-1163	1
-1165	1
-1169	2
-1174	1
-1176	1
-1177	1
-1178	2
-1179	1
-1180	2
-1181	1
-1182	1
-1183	2
-1185	1
-1187	1
-1191	2
-1193	1
-1195	3
-1196	1
-1201	3
-1203	1
-1206	1
-1210	1
-1213	1
-1214	1
-1215	2
-1218	1
-1220	1
-1221	1
-1225	1
-1226	1
-1233	2
-1241	1
-1243	1
-1249	1
-1250	2
-1251	1
-1254	1
-1255	2
-1260	1
-1268	1
-1270	1
-1273	1
-1274	1
-1277	1
-1284	1
-1287	1
-1291	1
-1292	2
-1294	1
-1295	2
-1297	1
-1298	1
-1301	1
-1307	1
-1308	3
-1311	2
-1313	1
-1316	1
-1321	1
-1324	1
-1325	1
-1330	1
-1333	1
-1334	1
-1338	2
-1340	1
-1341	1
-1342	1
-1343	1
-1345	1
-1355	1
-1357	1
-1360	2
-1375	1
-1376	1
-1380	1
-1383	1
-1387	1
-1389	1
-1393	1
-1394	1
-1396	1
-1398	1
-1410	1
-1414	1
-1419	1
-1425	1
-1434	1
-1435	1
-1438	1
-1439	1
-1447	1
-1455	2
-1460	1
-1461	1
-1463	1
-1466	1
-1470	1
-1473	1
-1478	1
-1480	1
-1483	1
-1484	1
-1485	2
-1492	2
-1499	1
-1509	1
-1512	1
-1513	1
-1523	1
-1524	1
-1525	2
-1529	1
-1539	1
-1544	1
-1568	1
-1584	1
-1591	1
-1598	1
-1600	1
-1604	1
-1614	1
-1617	1
-1621	1
-1622	1
-1626	1
-1638	1
-1648	1
-1658	1
-1661	1
-1679	1
-1682	1
-1693	1
-1700	1
-1705	1
-1707	1
-1722	1
-1728	1
-1758	1
-1762	1
-1763	1
-1775	1
-1776	1
-1801	1
-1810	1
-1812	1
-1827	1
-1834	1
-1846	1
-1847	1
-1848	1
-1851	1
-1862	1
-1866	1
-1877	2
-1884	1
-1888	1
-1903	1
-1912	1
-1925	1
-1938	1
-1955	1
-1998	1
-2054	1
-2058	1
-2065	1
-2069	1
-2076	1
-2089	1
-2104	1
-2111	1
-2133	1
-2138	1
-2156	1
-2204	1
-2212	1
-2237	1
-2246	2
-2298	1
-2304	1
-2360	1
-2400	1
-2481	1
-2544	1
-2586	1
-2622	1
-2666	1
-2682	1
-2725	1
-2920	1
-3997	1
-4019	1
-5211	1
-12	19
-14	1
-16	401
-18	2
-20	421
-22	557
-24	625
-26	50
-28	4481
-30	52
-32	550
-34	5840
-36	4644
-38	87
-40	5794
-41	33
-42	571
-44	11805
-46	4711
-47	7
-48	597
-49	12
-50	678
-51	2
-52	14715
-53	3
-54	7322
-55	3
-56	508
-57	39
-58	3486
-59	11
-60	8974
-61	45
-62	1276
-63	4
-64	15693
-65	15
-66	657
-67	13
-68	6409
-69	10
-70	3188
-71	25
-72	1889
-73	27
-74	10370
-75	9
-76	12432
-77	23
-78	520
-79	15
-80	1534
-81	29
-82	2944
-83	23
-84	12071
-85	36
-86	1502
-87	10
-88	10978
-89	11
-90	889
-91	16
-92	4571
-93	17
-94	7855
-95	21
-96	2271
-97	33
-98	1423
-99	15
-100	11096
-101	21
-102	4082
-103	13
-104	5442
-105	25
-106	2113
-107	26
-108	3779
-109	43
-110	1294
-111	29
-112	7860
-113	29
-114	4965
-115	22
-116	7898
-117	25
-118	1772
-119	28
-120	1149
-121	38
-122	1483
-123	32
-124	10572
-125	25
-126	1147
-127	31
-128	1699
-129	22
-130	5533
-131	22
-132	4669
-133	34
-134	3777
-135	10
-136	5412
-137	21
-138	855
-139	26
-140	2485
-141	46
-142	1970
-143	27
-144	6565
-145	40
-146	933
-147	15
-148	7923
-149	16
-150	735
-151	23
-152	1111
-153	33
-154	3714
-155	27
-156	2445
-157	30
-158	3367
-159	10
-160	4646
-161	27
-162	990
-163	23
-164	5679
-165	25
-166	2186
-167	17
-168	899
-169	32
-170	1034
-171	22
-172	6185
-173	32
-174	2685
-175	17
-176	1354
-177	38
-178	1460
-179	15
-180	3478
-181	20
-182	958
-183	20
-184	6055
-185	23
-186	2180
-187	15
-188	1416
-189	30
-190	1284
-191	22
-192	1341
-193	21
-194	2413
-195	18
-196	4984
-197	13
-198	830
-199	22
-200	1834
-201	19
-202	2238
-203	9
-204	3050
-205	22
-206	616
-207	17
-208	2892
-209	22
-210	711
-211	30
-212	2631
-213	19
-214	3341
-215	21
-216	987
-217	26
-218	823
-219	9
-220	3588
-221	20
-222	692
-223	7
-224	2925
-225	31
-226	1075
-227	16
-228	2909
-229	18
-230	673
-231	20
-232	2215
-233	14
-234	1584
-235	21
-236	1292
-237	29
-238	1647
-239	25
-240	1014
-241	30
-242	1648
-243	19
-244	4465
-245	10
-246	787
-247	11
-248	480
-249	25
-250	842
-251	15
-252	1219
-253	23
-254	1508
-255	8
-256	3525
-257	16
-258	490
-259	12
-260	1678
-261	14
-262	822
-263	16
-264	1729
-265	28
-266	604
-267	11
-268	2572
-269	7
-270	1242
-271	15
-272	725
-273	18
-274	1983
-275	13
-276	1662
-277	19
-278	491
-279	12
-280	1586
-281	14
-282	563
-283	10
-284	2363
-285	10
-286	656
-287	14
-288	725
-289	28
-290	871
-291	9
-292	2606
-293	12
-294	961
-295	9
-296	478
-297	13
-298	1252
-299	10
-300	736
-301	19
-302	466
-303	13
-304	2254
-305	12
-306	486
-307	14
-308	1145
-309	13
-310	955
-311	13
-312	1235
-313	13
-314	931
-315	14
-316	1768
-317	11
-318	330
-319	10
-320	539
-321	23
-322	570
-323	12
-324	1789
-325	13
-326	884
-327	5
-328	1422
-329	14
-330	317
-331	11
-332	509
-333	13
-334	1062
-335	12
-336	577
-337	27
-338	378
-339	10
-340	2313
-341	9
-342	391
-343	13
-344	894
-345	17
-346	664
-347	9
-348	453
-349	6
-350	363
-351	15
-352	1115
-353	13
-354	1054
-355	8
-356	1108
-357	12
-358	354
-359	7
-360	363
-361	16
-362	344
-363	11
-364	1734
-365	12
-366	265
-367	10
-368	969
-369	16
-370	316
-371	12
-372	757
-373	7
-374	563
-375	15
-376	857
-377	9
-378	469
-379	9
-380	385
-381	12
-382	921
-383	15
-384	764
-385	14
-386	246
-387	6
-388	1108
-389	14
-390	230
-391	8
-392	266
-393	11
-394	641
-395	8
-396	719
-397	9
-398	243
-399	4
-400	1108
-401	7
-402	229
-403	7
-404	903
-405	7
-406	257
-407	12
-408	244
-409	3
-410	541
-411	6
-412	744
-413	8
-414	419
-415	8
-416	388
-417	19
-418	470
-419	14
-420	612
-421	6
-422	342
-423	3
-424	1179
-425	3
-426	116
-427	14
-428	207
-429	6
-430	255
-431	4
-432	288
-433	12
-434	343
-435	6
-436	1015
-437	3
-438	538
-439	10
-440	194
-441	6
-442	188
-443	15
-444	524
-445	7
-446	214
-447	7
-448	574
-449	6
-450	214
-451	5
-452	635
-453	9
-454	464
-455	5
-456	205
-457	9
-458	163
-459	2
-460	558
-461	4
-462	171
-463	14
-464	444
-465	11
-466	543
-467	5
-468	388
-469	6
-470	141
-471	4
-472	647
-473	3
-474	210
-475	4
-476	193
-477	7
-478	195
-479	7
-480	443
-481	10
-482	198
-483	3
-484	816
-485	6
-486	128
-487	9
-488	215
-489	9
-490	328
-491	7
-492	158
-493	11
-494	335
-495	8
-496	435
-497	6
-498	174
-499	1
-500	373
-501	5
-502	140
-503	7
-504	330
-505	9
-506	149
-507	5
-508	642
-509	3
-510	179
-511	3
-512	159
-513	8
-514	204
-515	7
-516	306
-517	4
-518	110
-519	5
-520	326
-521	6
-522	305
-523	6
-524	294
-525	7
-526	268
-527	5
-528	149
-529	4
-530	133
-531	2
-532	513
-533	10
-534	116
-535	5
-536	258
-537	4
-538	113
-539	4
-540	138
-541	6
-542	116
-544	485
-545	4
-546	93
-547	9
-548	299
-549	3
-550	256
-551	6
-552	92
-553	3
-554	175
-555	6
-556	253
-557	7
-558	95
-559	2
-560	128
-561	4
-562	206
-563	2
-564	465
-565	3
-566	69
-567	3
-568	157
-569	7
-570	97
-571	8
-572	118
-573	5
-574	130
-575	4
-576	301
-577	6
-578	177
-579	2
-580	397
-581	3
-582	80
-583	1
-584	128
-585	5
-586	52
-587	2
-588	72
-589	1
-590	84
-591	6
-592	323
-593	11
-594	77
-595	5
-596	205
-597	1
-598	244
-599	4
-600	69
-601	3
-602	89
-603	5
-604	254
-605	6
-606	147
-607	3
-608	83
-609	3
-610	77
-611	3
-612	194
-613	1
-614	98
-615	3
-616	243
-617	3
-618	50
-619	8
-620	188
-621	4
-622	67
-623	4
-624	123
-625	2
-626	50
-627	1
-628	239
-629	2
-630	51
-631	4
-632	65
-633	5
-634	188
-636	81
-637	3
-638	46
-639	3
-640	103
-641	1
-642	136
-643	3
-644	188
-645	3
-646	58
-648	122
-649	4
-650	47
-651	2
-652	155
-653	4
-654	71
-655	1
-656	71
-657	3
-658	50
-659	2
-660	177
-661	5
-662	66
-663	2
-664	183
-665	3
-666	50
-667	2
-668	53
-669	2
-670	115
-672	66
-673	2
-674	47
-675	1
-676	197
-677	2
-678	46
-679	3
-680	95
-681	3
-682	46
-683	3
-684	107
-685	1
-686	86
-687	2
-688	158
-689	4
-690	51
-691	1
-692	80
-694	56
-695	4
-696	40
-698	43
-699	3
-700	95
-701	2
-702	51
-703	2
-704	133
-705	1
-706	100
-707	2
-708	121
-709	2
-710	15
-711	3
-712	35
-713	2
-714	20
-715	3
-716	37
-717	2
-718	78
-720	55
-721	1
-722	42
-723	2
-724	218
-725	3
-726	23
-727	2
-728	26
-729	1
-730	64
-731	2
-732	65
-734	24
-735	2
-736	53
-737	1
-738	32
-739	1
-740	60
-742	81
-743	1
-744	77
-745	1
-746	47
-747	1
-748	62
-749	1
-750	19
-751	1
-752	86
-753	3
-754	40
-756	55
-757	2
-758	38
-759	1
-760	101
-761	1
-762	22
-764	67
-765	2
-766	35
-767	1
-768	38
-769	1
-770	22
-771	1
-772	82
-773	1
-774	73
-776	29
-777	1
-778	55
-780	23
-781	1
-782	16
-784	84
-785	3
-786	28
-788	59
-789	1
-790	33
-791	3
-792	24
-794	13
-795	1
-796	110
-797	2
-798	15
-800	22
-801	3
-802	29
-803	1
-804	87
-806	21
-808	29
-810	48
-812	28
-813	1
-814	58
-815	1
-816	48
-817	1
-818	31
-819	1
-820	66
-822	17
-823	2
-824	58
-826	10
-827	2
-828	25
-829	1
-830	29
-831	1
-832	63
-833	1
-834	26
-835	3
-836	52
-837	1
-838	18
-840	27
-841	2
-842	12
-843	1
-844	83
-845	1
-846	7
-847	1
-848	10
-850	26
-852	25
-853	1
-854	15
-856	27
-858	32
-859	1
-860	15
-862	43
-864	32
-865	1
-866	6
-868	39
-870	11
-872	25
-873	1
-874	10
-875	1
-876	20
-877	2
-878	19
-879	1
-880	30
-882	11
-884	53
-886	25
-887	1
-888	28
-890	6
-892	36
-894	10
-896	13
-898	14
-900	31
-902	14
-903	2
-904	43
-906	25
-908	9
-910	11
-911	1
-912	16
-913	1
-914	24
-916	27
-918	6
-920	15
-922	27
-923	1
-924	23
-926	13
-928	42
-929	1
-930	3
-932	27
-934	17
-936	8
-937	1
-938	11
-940	33
-942	4
-943	1
-944	18
-946	15
-948	13
-950	18
-952	12
-954	11
-956	21
-958	10
-960	13
-962	5
-964	32
-966	13
-968	8
-970	8
-971	1
-972	23
-973	2
-974	12
-975	1
-976	22
-978	7
-979	1
-980	14
-982	8
-984	22
-985	1
-986	6
-988	17
-989	1
-990	6
-992	13
-994	19
-996	11
-998	4
-1000	9
-1002	2
-1004	14
-1006	5
-1008	3
-1010	9
-1012	29
-1014	6
-1016	22
-1017	1
-1018	8
-1019	1
-1020	7
-1022	6
-1023	1
-1024	10
-1026	2
-1028	8
-1030	11
-1031	2
-1032	8
-1034	9
-1036	13
-1038	12
-1040	12
-1042	3
-1044	12
-1046	3
-1048	11
-1050	2
-1051	1
-1052	2
-1054	11
-1056	6
-1058	8
-1059	1
-1060	23
-1062	6
-1063	1
-1064	8
-1066	3
-1068	6
-1070	8
-1071	1
-1072	5
-1074	3
-1076	5
-1078	3
-1080	11
-1081	1
-1082	7
-1084	18
-1086	4
-1087	1
-1088	3
-1090	3
-1092	7
-1094	3
-1096	12
-1098	6
-1099	1
-1100	2
-1102	6
-1104	14
-1106	3
-1108	6
-1110	5
-1112	2
-1114	8
-1116	3
-1118	3
-1120	7
-1122	10
-1124	6
-1126	8
-1128	1
-1130	4
-1132	3
-1134	2
-1136	5
-1138	5
-1140	8
-1142	3
-1144	7
-1146	3
-1148	11
-1150	1
-1152	5
-1154	1
-1156	5
-1158	1
-1160	5
-1162	3
-1164	6
-1165	1
-1166	1
-1168	4
-1169	1
-1170	3
-1171	1
-1172	2
-1174	5
-1176	3
-1177	1
-1180	8
-1182	2
-1184	4
-1186	2
-1188	3
-1190	2
-1192	5
-1194	6
-1196	1
-1198	2
-1200	2
-1204	10
-1206	2
-1208	9
-1210	1
-1214	6
-1216	3
-1218	4
-1220	9
-1221	2
-1222	1
-1224	5
-1226	4
-1228	8
-1230	1
-1232	1
-1234	3
-1236	5
-1240	3
-1242	1
-1244	3
-1245	1
-1246	4
-1248	6
-1250	2
-1252	7
-1256	3
-1258	2
-1260	2
-1262	3
-1264	4
-1265	1
-1266	1
-1270	1
-1271	1
-1272	2
-1274	3
-1276	3
-1278	1
-1280	3
-1284	1
-1286	1
-1290	1
-1292	3
-1294	1
-1296	7
-1300	2
-1302	4
-1304	3
-1306	2
-1308	2
-1312	1
-1314	1
-1316	3
-1318	2
-1320	1
-1324	8
-1326	1
-1330	1
-1331	1
-1336	2
-1338	1
-1340	3
-1341	1
-1344	1
-1346	2
-1347	1
-1348	3
-1352	1
-1354	2
-1356	1
-1358	1
-1360	3
-1362	1
-1364	4
-1366	1
-1370	1
-1372	3
-1380	2
-1384	2
-1388	2
-1390	2
-1392	2
-1394	1
-1396	1
-1398	1
-1400	2
-1402	1
-1404	1
-1406	1
-1410	1
-1412	5
-1418	1
-1420	1
-1424	1
-1432	2
-1434	2
-1442	3
-1444	5
-1448	1
-1454	1
-1456	1
-1460	3
-1462	4
-1468	1
-1474	1
-1476	1
-1478	2
-1480	1
-1486	2
-1488	1
-1492	1
-1496	1
-1500	3
-1503	1
-1506	1
-1512	2
-1516	1
-1522	1
-1524	2
-1534	4
-1536	1
-1538	1
-1540	2
-1544	2
-1548	1
-1556	1
-1560	1
-1562	1
-1564	2
-1566	1
-1568	1
-1570	1
-1572	1
-1576	1
-1590	1
-1594	1
-1604	1
-1608	1
-1614	1
-1622	1
-1624	2
-1628	1
-1629	1
-1636	1
-1642	1
-1654	2
-1660	1
-1664	1
-1670	1
-1684	4
-1698	1
-1732	3
-1742	1
-1752	1
-1760	1
-1764	1
-1772	2
-1798	1
-1808	1
-1820	1
-1852	1
-1856	1
-1874	1
-1902	1
-1908	1
-1952	1
-2004	1
-2018	1
-2020	1
-2028	1
-2174	1
-2233	1
-2244	1
-2280	1
-2290	1
-2352	1
-2604	1
-4190	1
+ rowspan="10"
\ No newline at end of file
diff --git a/ppstructure/table/predict_structure.py b/ppstructure/table/predict_structure.py
index 0179c614ae..17ec909582 100755
--- a/ppstructure/table/predict_structure.py
+++ b/ppstructure/table/predict_structure.py
@@ -23,6 +23,7 @@ os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
 import cv2
 import numpy as np
 import time
+import json
 
 import tools.infer.utility as utility
 from ppocr.data import create_operators, transform
@@ -34,32 +35,50 @@ from ppstructure.utility import parse_args
 logger = get_logger()
 
 
+def build_pre_process_list(args):
+    resize_op = {'ResizeTableImage': {'max_len': args.table_max_len, }}
+    pad_op = {
+        'PaddingTableImage': {
+            'size': [args.table_max_len, args.table_max_len]
+        }
+    }
+    normalize_op = {
+        'NormalizeImage': {
+            'std': [0.229, 0.224, 0.225] if
+            args.table_algorithm not in ['TableMaster'] else [0.5, 0.5, 0.5],
+            'mean': [0.485, 0.456, 0.406] if
+            args.table_algorithm not in ['TableMaster'] else [0.5, 0.5, 0.5],
+            'scale': '1./255.',
+            'order': 'hwc'
+        }
+    }
+    to_chw_op = {'ToCHWImage': None}
+    keep_keys_op = {'KeepKeys': {'keep_keys': ['image', 'shape']}}
+    if args.table_algorithm not in ['TableMaster']:
+        pre_process_list = [
+            resize_op, normalize_op, pad_op, to_chw_op, keep_keys_op
+        ]
+    else:
+        pre_process_list = [
+            resize_op, pad_op, normalize_op, to_chw_op, keep_keys_op
+        ]
+    return pre_process_list
+
+
 class TableStructurer(object):
     def __init__(self, args):
-        pre_process_list = [{
-            'ResizeTableImage': {
-                'max_len': args.table_max_len
+        pre_process_list = build_pre_process_list(args)
+        if args.table_algorithm not in ['TableMaster']:
+            postprocess_params = {
+                'name': 'TableLabelDecode',
+                "character_dict_path": args.table_char_dict_path,
             }
-        }, {
-            'NormalizeImage': {
-                'std': [0.229, 0.224, 0.225],
-                'mean': [0.485, 0.456, 0.406],
-                'scale': '1./255.',
-                'order': 'hwc'
+        else:
+            postprocess_params = {
+                'name': 'TableMasterLabelDecode',
+                "character_dict_path": args.table_char_dict_path,
+                'box_shape': 'pad'
             }
-        }, {
-            'PaddingTableImage': None
-        }, {
-            'ToCHWImage': None
-        }, {
-            'KeepKeys': {
-                'keep_keys': ['image']
-            }
-        }]
-        postprocess_params = {
-            'name': 'TableLabelDecode',
-            "character_dict_path": args.table_char_dict_path,
-        }
 
         self.preprocess_op = create_operators(pre_process_list)
         self.postprocess_op = build_post_process(postprocess_params)
@@ -88,27 +107,30 @@ class TableStructurer(object):
         preds['structure_probs'] = outputs[1]
         preds['loc_preds'] = outputs[0]
 
-        post_result = self.postprocess_op(preds)
+        shape_list = np.expand_dims(data[-1], axis=0)
+        post_result = self.postprocess_op(preds, [shape_list])
 
-        structure_str_list = post_result['structure_str_list']
-        res_loc = post_result['res_loc']
-        imgh, imgw = ori_im.shape[0:2]
-        res_loc_final = []
-        for rno in range(len(res_loc[0])):
-            x0, y0, x1, y1 = res_loc[0][rno]
-            left = max(int(imgw * x0), 0)
-            top = max(int(imgh * y0), 0)
-            right = min(int(imgw * x1), imgw - 1)
-            bottom = min(int(imgh * y1), imgh - 1)
-            res_loc_final.append([left, top, right, bottom])
-
-        structure_str_list = structure_str_list[0][:-1]
+        structure_str_list = post_result['structure_batch_list'][0]
+        bbox_list = post_result['bbox_batch_list'][0]
+        structure_str_list = structure_str_list[0]
         structure_str_list = [
             '<html>', '<body>', '<table>'
         ] + structure_str_list + ['</table>', '</body>', '</html>']
-
         elapse = time.time() - starttime
-        return (structure_str_list, res_loc_final), elapse
+        return structure_str_list, bbox_list, elapse
+
+
+def draw_rectangle(img_path, boxes, use_xywh=False):
+    img = cv2.imread(img_path)
+    img_show = img.copy()
+    for box in boxes.astype(int):
+        if use_xywh:
+            x, y, w, h = box
+            x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
+        else:
+            x1, y1, x2, y2 = box
+        cv2.rectangle(img_show, (x1, y1), (x2, y2), (255, 0, 0), 2)
+    return img_show
 
 
 def main(args):
@@ -116,21 +138,35 @@ def main(args):
     table_structurer = TableStructurer(args)
     count = 0
     total_time = 0
-    for image_file in image_file_list:
-        img, flag = check_and_read_gif(image_file)
-        if not flag:
-            img = cv2.imread(image_file)
-        if img is None:
-            logger.info("error in loading image:{}".format(image_file))
-            continue
-        structure_res, elapse = table_structurer(img)
+    use_xywh = args.table_algorithm in ['TableMaster']
+    os.makedirs(args.output, exist_ok=True)
+    with open(
+            os.path.join(args.output, 'infer.txt'), mode='w',
+            encoding='utf-8') as f_w:
+        for image_file in image_file_list:
+            img, flag = check_and_read_gif(image_file)
+            if not flag:
+                img = cv2.imread(image_file)
+            if img is None:
+                logger.info("error in loading image:{}".format(image_file))
+                continue
+            structure_str_list, bbox_list, elapse = table_structurer(img)
 
-        logger.info("result: {}".format(structure_res))
+            bbox_list_str = json.dumps(bbox_list.tolist())
+            logger.info("result: {}, {}".format(structure_str_list,
+                                                bbox_list_str))
+            f_w.write("result: {}, {}\n".format(structure_str_list,
+                                                bbox_list_str))
 
-        if count > 0:
-            total_time += elapse
-        count += 1
-        logger.info("Predict time of {}: {}".format(image_file, elapse))
+            img = draw_rectangle(image_file, bbox_list, use_xywh)
+            img_save_path = os.path.join(args.output,
+                                         os.path.basename(image_file))
+            cv2.imwrite(img_save_path, img)
+            logger.info("save vis result to {}".format(img_save_path))
+            if count > 0:
+                total_time += elapse
+            count += 1
+            logger.info("Predict time of {}: {}".format(image_file, elapse))
 
 
 if __name__ == "__main__":
diff --git a/ppstructure/utility.py b/ppstructure/utility.py
index 1ad902e7e6..05452c23b5 100644
--- a/ppstructure/utility.py
+++ b/ppstructure/utility.py
@@ -25,6 +25,7 @@ def init_args():
     parser.add_argument("--output", type=str, default='./output')
     # params for table structure
     parser.add_argument("--table_max_len", type=int, default=488)
+    parser.add_argument("--table_algorithm", type=str, default='TableAttn')
     parser.add_argument("--table_model_dir", type=str)
     parser.add_argument(
         "--table_char_dict_path",
@@ -65,7 +66,7 @@ def init_args():
         "--recovery",
         type=bool,
         default=False,
-        help='Whether to enable layout of recovery')        
+        help='Whether to enable layout of recovery')
     return parser
 
 
diff --git a/tools/export_model.py b/tools/export_model.py
index e971f6cb20..fbb2201e39 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -88,6 +88,8 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
                 infer_shape = [1, 32, 100]
         elif arch_config["model_type"] == "table":
             infer_shape = [3, 488, 488]
+            if arch_config["algorithm"] == "TableMaster":
+                infer_shape = [3, 480, 480]
         model = to_static(
             model,
             input_spec=[
diff --git a/tools/infer_table.py b/tools/infer_table.py
index 66c2da4421..58e7455cbb 100644
--- a/tools/infer_table.py
+++ b/tools/infer_table.py
@@ -40,6 +40,7 @@ import tools.program as program
 import cv2
 
 
+@paddle.no_grad()
 def main(config, device, logger, vdl_writer):
     global_config = config['Global']
 
@@ -53,53 +54,74 @@ def main(config, device, logger, vdl_writer):
             getattr(post_process_class, 'character'))
 
     model = build_model(config['Architecture'])
+    algorithm = config['Architecture']['algorithm']
+    use_xywh = algorithm in ['TableMaster']
 
     load_model(config, model)
 
     # create data ops
     transforms = []
-    use_padding = False
     for op in config['Eval']['dataset']['transforms']:
         op_name = list(op)[0]
-        if 'Label' in op_name:
+        if 'Encode' in op_name:
             continue
         if op_name == 'KeepKeys':
-            op[op_name]['keep_keys'] = ['image']
-        if op_name == "ResizeTableImage":
-            use_padding = True
-            padding_max_len = op['ResizeTableImage']['max_len']
+            op[op_name]['keep_keys'] = ['image', 'shape']
         transforms.append(op)
 
     global_config['infer_mode'] = True
     ops = create_operators(transforms, global_config)
 
+    save_res_path = config['Global']['save_res_path']
+    os.makedirs(save_res_path, exist_ok=True)
+
     model.eval()
-    for file in get_image_file_list(config['Global']['infer_img']):
-        logger.info("infer_img: {}".format(file))
-        with open(file, 'rb') as f:
-            img = f.read()
-            data = {'image': img}
-        batch = transform(data, ops)
-        images = np.expand_dims(batch[0], axis=0)
-        images = paddle.to_tensor(images)
-        preds = model(images)
-        post_result = post_process_class(preds)
-        res_html_code = post_result['res_html_code']
-        res_loc = post_result['res_loc']
-        img = cv2.imread(file)
-        imgh, imgw = img.shape[0:2]
-        res_loc_final = []
-        for rno in range(len(res_loc[0])):
-            x0, y0, x1, y1 = res_loc[0][rno]
-            left = max(int(imgw * x0), 0)
-            top = max(int(imgh * y0), 0)
-            right = min(int(imgw * x1), imgw - 1)
-            bottom = min(int(imgh * y1), imgh - 1)
-            cv2.rectangle(img, (left, top), (right, bottom), (0, 0, 255), 2)
-            res_loc_final.append([left, top, right, bottom])
-        res_loc_str = json.dumps(res_loc_final)
-        logger.info("result: {}, {}".format(res_html_code, res_loc_final))
-    logger.info("success!")
+    with open(
+            os.path.join(save_res_path, 'infer.txt'), mode='w',
+            encoding='utf-8') as f_w:
+        for file in get_image_file_list(config['Global']['infer_img']):
+            logger.info("infer_img: {}".format(file))
+            with open(file, 'rb') as f:
+                img = f.read()
+                data = {'image': img}
+            batch = transform(data, ops)
+            images = np.expand_dims(batch[0], axis=0)
+            shape_list = np.expand_dims(batch[1], axis=0)
+
+            images = paddle.to_tensor(images)
+            preds = model(images)
+            post_result = post_process_class(preds, [shape_list])
+
+            structure_str_list = post_result['structure_batch_list'][0]
+            bbox_list = post_result['bbox_batch_list'][0]
+            structure_str_list = structure_str_list[0]
+            structure_str_list = [
+                '<html>', '<body>', '<table>'
+            ] + structure_str_list + ['</table>', '</body>', '</html>']
+            bbox_list_str = json.dumps(bbox_list.tolist())
+
+            logger.info("result: {}, {}".format(structure_str_list,
+                                                bbox_list_str))
+            f_w.write("result: {}, {}\n".format(structure_str_list,
+                                                bbox_list_str))
+
+            img = draw_rectangle(file, bbox_list, use_xywh)
+            cv2.imwrite(
+                os.path.join(save_res_path, os.path.basename(file)), img)
+        logger.info("success!")
+
+
+def draw_rectangle(img_path, boxes, use_xywh=False):
+    img = cv2.imread(img_path)
+    img_show = img.copy()
+    for box in boxes.astype(int):
+        if use_xywh:
+            x, y, w, h = box
+            x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
+        else:
+            x1, y1, x2, y2 = box
+        cv2.rectangle(img_show, (x1, y1), (x2, y2), (255, 0, 0), 2)
+    return img_show
 
 
 if __name__ == '__main__':
diff --git a/tools/program.py b/tools/program.py
index 7c02dc0149..17079cb86e 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -274,8 +274,11 @@ def train(config,
 
             if cal_metric_during_train and epoch % calc_epoch_interval == 0:  # only rec and cls need
                 batch = [item.numpy() for item in batch]
-                if model_type in ['table', 'kie']:
+                if model_type in ['kie']:
                     eval_class(preds, batch)
+                elif model_type in ['table']:
+                    post_result = post_process_class(preds, batch)
+                    eval_class(post_result, batch)
                 else:
                     if config['Loss']['name'] in ['MultiLoss', 'MultiLoss_v2'
                                                   ]:  # for multi head loss
@@ -302,7 +305,8 @@ def train(config,
             train_stats.update(stats)
 
             if log_writer is not None and dist.get_rank() == 0:
-                log_writer.log_metrics(metrics=train_stats.get(), prefix="TRAIN", step=global_step)
+                log_writer.log_metrics(
+                    metrics=train_stats.get(), prefix="TRAIN", step=global_step)
 
             if dist.get_rank() == 0 and (
                 (global_step > 0 and global_step % print_batch_step == 0) or
@@ -349,7 +353,8 @@ def train(config,
 
                 # logger metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics=cur_metric, prefix="EVAL", step=global_step)
+                    log_writer.log_metrics(
+                        metrics=cur_metric, prefix="EVAL", step=global_step)
 
                 if cur_metric[main_indicator] >= best_model_dict[
                         main_indicator]:
@@ -372,11 +377,18 @@ def train(config,
                 logger.info(best_str)
                 # logger best metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics={
-                        "best_{}".format(main_indicator): best_model_dict[main_indicator]
-                        }, prefix="EVAL", step=global_step)
-                    
-                    log_writer.log_model(is_best=True, prefix="best_accuracy", metadata=best_model_dict)
+                    log_writer.log_metrics(
+                        metrics={
+                            "best_{}".format(main_indicator):
+                            best_model_dict[main_indicator]
+                        },
+                        prefix="EVAL",
+                        step=global_step)
+
+                    log_writer.log_model(
+                        is_best=True,
+                        prefix="best_accuracy",
+                        metadata=best_model_dict)
 
             reader_start = time.time()
         if dist.get_rank() == 0:
@@ -408,7 +420,8 @@ def train(config,
                 epoch=epoch,
                 global_step=global_step)
             if log_writer is not None:
-                log_writer.log_model(is_best=False, prefix='iter_epoch_{}'.format(epoch))
+                log_writer.log_model(
+                    is_best=False, prefix='iter_epoch_{}'.format(epoch))
 
     best_str = 'best metric, {}'.format(', '.join(
         ['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))
@@ -446,7 +459,6 @@ def eval(model,
                 preds = model(batch)
             else:
                 preds = model(images)
-
             batch_numpy = []
             for item in batch:
                 if isinstance(item, paddle.Tensor):
@@ -456,9 +468,9 @@ def eval(model,
             # Obtain usable results from post-processing methods
             total_time += time.time() - start
             # Evaluate the results of the current batch
-            if model_type in ['table', 'kie']:
+            if model_type in ['kie']:
                 eval_class(preds, batch_numpy)
-            elif model_type in ['vqa']:
+            elif model_type in ['table', 'vqa']:
                 post_result = post_process_class(preds, batch_numpy)
                 eval_class(post_result, batch_numpy)
             else:
@@ -559,7 +571,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
+        'TableMaster'
     ]
 
     device = 'cpu'
@@ -578,7 +591,8 @@ def preprocess(is_train=False):
         vdl_writer_path = '{}/vdl/'.format(save_model_dir)
         log_writer = VDLLogger(save_model_dir)
         loggers.append(log_writer)
-    if ('use_wandb' in config['Global'] and config['Global']['use_wandb']) or 'wandb' in config:
+    if ('use_wandb' in config['Global'] and
+            config['Global']['use_wandb']) or 'wandb' in config:
         save_dir = config['Global']['save_model_dir']
         wandb_writer_path = "{}/wandb".format(save_dir)
         if "wandb" in config:

From 5d8c9f47cd2e4a14aa367e4725e986489fa26da2 Mon Sep 17 00:00:00 2001
From: MissPenguin <lichenxia1991@163.com>
Date: Fri, 17 Jun 2022 15:47:49 +0800
Subject: [PATCH 023/134] Update ppocr_introduction_en.md

---
 doc/doc_en/ppocr_introduction_en.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/doc/doc_en/ppocr_introduction_en.md b/doc/doc_en/ppocr_introduction_en.md
index b13d7f9bf1..d28ccb3529 100644
--- a/doc/doc_en/ppocr_introduction_en.md
+++ b/doc/doc_en/ppocr_introduction_en.md
@@ -29,10 +29,10 @@ PP-OCR pipeline is as follows:
 
 PP-OCR system is in continuous optimization. At present, PP-OCR and PP-OCRv2 have been released:
 
-PP-OCR adopts 19 effective strategies from 8 aspects including backbone network selection and adjustment, prediction head design, data augmentation, learning rate transformation strategy, regularization parameter selection, pre-training model use, and automatic model tailoring and quantization to optimize and slim down the models of each module (as shown in the green box above). The final results are an ultra-lightweight Chinese and English OCR model with an overall size of 3.5M and a 2.8M English digital OCR model. For more details, please refer to the PP-OCR technical article (https://arxiv.org/abs/2009.09941).
+PP-OCR adopts 19 effective strategies from 8 aspects including backbone network selection and adjustment, prediction head design, data augmentation, learning rate transformation strategy, regularization parameter selection, pre-training model use, and automatic model tailoring and quantization to optimize and slim down the models of each module (as shown in the green box above). The final results are an ultra-lightweight Chinese and English OCR model with an overall size of 3.5M and a 2.8M English digital OCR model. For more details, please refer to [PP-OCR technical report](https://arxiv.org/abs/2009.09941).
 
 #### PP-OCRv2
-On the basis of PP-OCR, PP-OCRv2 is further optimized in five aspects. The detection model adopts CML(Collaborative Mutual Learning) knowledge distillation strategy and CopyPaste data expansion strategy. The recognition model adopts LCNet lightweight backbone network, U-DML knowledge distillation strategy and enhanced CTC loss function improvement (as shown in the red box above), which further improves the inference speed and prediction effect. For more details, please refer to the technical report of PP-OCRv2 (https://arxiv.org/abs/2109.03144).
+On the basis of PP-OCR, PP-OCRv2 is further optimized in five aspects. The detection model adopts CML(Collaborative Mutual Learning) knowledge distillation strategy and CopyPaste data expansion strategy. The recognition model adopts LCNet lightweight backbone network, U-DML knowledge distillation strategy and enhanced CTC loss function improvement (as shown in the red box above), which further improves the inference speed and prediction effect. For more details, please refer to [PP-OCRv2 technical report](https://arxiv.org/abs/2109.03144).
 
 #### PP-OCRv3
 
@@ -46,7 +46,7 @@ PP-OCRv3 pipeline is as follows:
     <img src="../ppocrv3_framework.png" width="800">
 </div>
 
-For more details, please refer to [PP-OCRv3 technical report](./PP-OCRv3_introduction_en.md).
+For more details, please refer to [PP-OCRv3 technical report](https://arxiv.org/abs/2206.03001v2).
 
 <a name="2"></a>
 ## 2. Features

From 94d5d00a53da017a933b9d58e1ceccc00881df06 Mon Sep 17 00:00:00 2001
From: MissPenguin <lichenxia1991@163.com>
Date: Fri, 17 Jun 2022 15:51:57 +0800
Subject: [PATCH 024/134] Update ppocr_introduction.md

---
 doc/doc_ch/ppocr_introduction.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/doc/doc_ch/ppocr_introduction.md b/doc/doc_ch/ppocr_introduction.md
index 59de124e2a..bd62087c8b 100644
--- a/doc/doc_ch/ppocr_introduction.md
+++ b/doc/doc_ch/ppocr_introduction.md
@@ -30,11 +30,11 @@ PP-OCR系统pipeline如下：
 
 PP-OCR系统在持续迭代优化，目前已发布PP-OCR和PP-OCRv2两个版本：
 
-PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身(如绿框所示)，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考PP-OCR技术方案 https://arxiv.org/abs/2009.09941
+PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身(如绿框所示)，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考[PP-OCR技术报告](https://arxiv.org/abs/2009.09941)。
 
 #### PP-OCRv2
 
-PP-OCRv2在PP-OCR的基础上，进一步在5个方面重点优化，检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略；识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./enhanced_ctc_loss.md)损失函数改进（如上图红框所示），进一步在推理速度和预测效果上取得明显提升。更多细节请参考PP-OCRv2[技术报告](https://arxiv.org/abs/2109.03144)。
+PP-OCRv2在PP-OCR的基础上，进一步在5个方面重点优化，检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略；识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./enhanced_ctc_loss.md)损失函数改进（如上图红框所示），进一步在推理速度和预测效果上取得明显提升。更多细节请参考[PP-OCRv2技术报告](https://arxiv.org/abs/2109.03144)。
 
 #### PP-OCRv3
 
@@ -48,7 +48,7 @@ PP-OCRv3系统pipeline如下：
     <img src="../ppocrv3_framework.png" width="800">
 </div>
 
-更多细节请参考PP-OCRv3[技术报告](./PP-OCRv3_introduction.md)。
+更多细节请参考[PP-OCRv3技术报告](https://arxiv.org/abs/2206.03001v2) 👉[中文简洁版](./PP-OCRv3_introduction.md)
 
 
 <a name="2"></a>

From b473e92d092e9ae1bb362bc3191fdc512cb71c0c Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Fri, 17 Jun 2022 09:12:46 +0000
Subject: [PATCH 025/134] add tvm

---
 deploy/tvm/.gitignore                 |   5 +
 deploy/tvm/Makefile                   | 129 +++++++++++
 deploy/tvm/README.md                  | 100 +++++++++
 deploy/tvm/arm-none-eabi-gcc.cmake    |  79 +++++++
 deploy/tvm/convert_image.py           | 102 +++++++++
 deploy/tvm/corstone300.ld             | 295 ++++++++++++++++++++++++++
 deploy/tvm/imgs_words_en/word_10.png  | Bin 0 -> 5738 bytes
 deploy/tvm/imgs_words_en/word_116.png | Bin 0 -> 74135 bytes
 deploy/tvm/include/crt_config.h       |  26 +++
 deploy/tvm/include/tvm_runtime.h      |  55 +++++
 deploy/tvm/requirements.txt           |   3 +
 deploy/tvm/run_demo.sh                | 152 +++++++++++++
 deploy/tvm/src/demo_bare_metal.c      |  78 +++++++
 13 files changed, 1024 insertions(+)
 create mode 100644 deploy/tvm/.gitignore
 create mode 100644 deploy/tvm/Makefile
 create mode 100644 deploy/tvm/README.md
 create mode 100644 deploy/tvm/arm-none-eabi-gcc.cmake
 create mode 100755 deploy/tvm/convert_image.py
 create mode 100644 deploy/tvm/corstone300.ld
 create mode 100644 deploy/tvm/imgs_words_en/word_10.png
 create mode 100644 deploy/tvm/imgs_words_en/word_116.png
 create mode 100644 deploy/tvm/include/crt_config.h
 create mode 100644 deploy/tvm/include/tvm_runtime.h
 create mode 100644 deploy/tvm/requirements.txt
 create mode 100755 deploy/tvm/run_demo.sh
 create mode 100644 deploy/tvm/src/demo_bare_metal.c

diff --git a/deploy/tvm/.gitignore b/deploy/tvm/.gitignore
new file mode 100644
index 0000000000..faeba235a6
--- /dev/null
+++ b/deploy/tvm/.gitignore
@@ -0,0 +1,5 @@
+include/inputs.h
+include/outputs.h
+
+__pycache__/
+build/
\ No newline at end of file
diff --git a/deploy/tvm/Makefile b/deploy/tvm/Makefile
new file mode 100644
index 0000000000..cf7d375b7e
--- /dev/null
+++ b/deploy/tvm/Makefile
@@ -0,0 +1,129 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+# Makefile to build demo
+
+# Setup build environment
+BUILD_DIR := build
+
+ARM_CPU = ARMCM55
+ETHOSU_PATH = /opt/arm/ethosu
+CMSIS_PATH ?= ${ETHOSU_PATH}/cmsis
+ETHOSU_PLATFORM_PATH ?= ${ETHOSU_PATH}/core_platform
+STANDALONE_CRT_PATH := $(abspath $(BUILD_DIR))/runtime
+CORSTONE_300_PATH = ${ETHOSU_PLATFORM_PATH}/targets/corstone-300
+PKG_COMPILE_OPTS = -g -Wall -O2 -Wno-incompatible-pointer-types -Wno-format -mcpu=cortex-m55 -mthumb -mfloat-abi=hard -std=gnu99
+CMAKE ?= cmake
+CC = arm-none-eabi-gcc
+AR = arm-none-eabi-ar
+RANLIB = arm-none-eabi-ranlib
+PKG_CFLAGS = ${PKG_COMPILE_OPTS} \
+	-I${STANDALONE_CRT_PATH}/include \
+	-I${STANDALONE_CRT_PATH}/src/runtime/crt/include \
+	-I${PWD}/include \
+	-I${CORSTONE_300_PATH} \
+	-I${CMSIS_PATH}/Device/ARM/${ARM_CPU}/Include/ \
+	-I${CMSIS_PATH}/CMSIS/Core/Include \
+	-I${CMSIS_PATH}/CMSIS/NN/Include \
+	-I${CMSIS_PATH}/CMSIS/DSP/Include \
+	-I$(abspath $(BUILD_DIR))/codegen/host/include
+CMSIS_NN_CMAKE_FLAGS = -DCMAKE_TOOLCHAIN_FILE=$(abspath $(BUILD_DIR))/../arm-none-eabi-gcc.cmake \
+	-DTARGET_CPU=cortex-m55 \
+	-DBUILD_CMSIS_NN_FUNCTIONS=YES
+PKG_LDFLAGS = -lm -specs=nosys.specs -static -T corstone300.ld
+
+$(ifeq VERBOSE,1)
+QUIET ?=
+$(else)
+QUIET ?= @
+$(endif)
+
+DEMO_MAIN = src/demo_bare_metal.c
+CODEGEN_SRCS = $(wildcard $(abspath $(BUILD_DIR))/codegen/host/src/*.c)
+CODEGEN_OBJS = $(subst .c,.o,$(CODEGEN_SRCS))
+CMSIS_STARTUP_SRCS = $(wildcard ${CMSIS_PATH}/Device/ARM/${ARM_CPU}/Source/*.c)
+UART_SRCS = $(wildcard ${CORSTONE_300_PATH}/*.c)
+
+demo: $(BUILD_DIR)/demo
+
+$(BUILD_DIR)/stack_allocator.o: $(STANDALONE_CRT_PATH)/src/runtime/crt/memory/stack_allocator.c
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) -c $(PKG_CFLAGS) -o $@  $^
+
+$(BUILD_DIR)/crt_backend_api.o: $(STANDALONE_CRT_PATH)/src/runtime/crt/common/crt_backend_api.c
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) -c $(PKG_CFLAGS) -o $@  $^
+
+# Build generated code
+$(BUILD_DIR)/libcodegen.a: $(CODEGEN_SRCS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/codegen/host/src) && $(CC) -c $(PKG_CFLAGS) $(CODEGEN_SRCS)
+	$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcodegen.a) $(CODEGEN_OBJS)
+	$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcodegen.a)
+
+# Build CMSIS startup code
+${BUILD_DIR}/libcmsis_startup.a: $(CMSIS_STARTUP_SRCS)
+	$(QUIET)mkdir -p $(abspath $(BUILD_DIR)/libcmsis_startup)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/libcmsis_startup) && $(CC) -c $(PKG_CFLAGS) -D${ARM_CPU} $^
+	$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcmsis_startup.a) $(abspath $(BUILD_DIR))/libcmsis_startup/*.o
+	$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcmsis_startup.a)
+
+CMSIS_SHA_FILE=${CMSIS_PATH}/977abe9849781a2e788b02282986480ff4e25ea6.sha
+ifneq ("$(wildcard $(CMSIS_SHA_FILE))","")
+${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a:
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all
+else
+# Build CMSIS-NN
+${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a:
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all
+endif
+
+# Build demo application
+ifneq ("$(wildcard $(CMSIS_SHA_FILE))","")
+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \
+	${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a ${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)
+else
+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \
+       ${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a \
+       ${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a \
+       ${BUILD_DIR}/cmsis_nn/Source/FullyConnectedFunctions/libCMSISNNFullyConnected.a \
+       ${BUILD_DIR}/cmsis_nn/Source/SVDFunctions/libCMSISNNSVDF.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ReshapeFunctions/libCMSISNNReshape.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ActivationFunctions/libCMSISNNActivation.a \
+       ${BUILD_DIR}/cmsis_nn/Source/NNSupportFunctions/libCMSISNNSupport.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ConcatenationFunctions/libCMSISNNConcatenation.a \
+       ${BUILD_DIR}/cmsis_nn/Source/BasicMathFunctions/libCMSISNNBasicMaths.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ConvolutionFunctions/libCMSISNNConvolutions.a \
+       ${BUILD_DIR}/cmsis_nn/Source/PoolingFunctions/libCMSISNNPooling.a
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)
+endif
+
+clean:
+	$(QUIET)rm -rf $(BUILD_DIR)/codegen
+
+cleanall:
+	$(QUIET)rm -rf $(BUILD_DIR)
+
+.SUFFIXES:
+
+.DEFAULT: demo
diff --git a/deploy/tvm/README.md b/deploy/tvm/README.md
new file mode 100644
index 0000000000..ecd635cc42
--- /dev/null
+++ b/deploy/tvm/README.md
@@ -0,0 +1,100 @@
+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+
+Running PaddleOCR text recognition model via TVM on bare metal Arm(R) Cortex(R)-M55 CPU and CMSIS-NN
+===============================================================
+
+This folder contains an example of how to use TVM to run a PaddleOCR model
+on bare metal Cortex(R)-M55 CPU and CMSIS-NN.
+
+Prerequisites
+-------------
+If the demo is run in the ci_cpu Docker container provided with TVM, then the following
+software will already be installed.
+
+If the demo is not run in the ci_cpu Docker container, then you will need the following:
+- Software required to build and run the demo (These can all be installed by running 
+  tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh.)
+  - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
+  - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
+  - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
+  - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)
+  - [CMSIS](https://github.com/ARM-software/CMSIS_5)
+- The python libraries listed in the requirements.txt of this directory
+  - These can be installed by running the following from the current directory:
+    ```bash
+    pip install -r ./requirements.txt
+    ```
+
+You will also need TVM which can either be:
+  - Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    - When building from source, the following need to be set in config.cmake:
+      - set(USE_CMSISNN ON)
+      - set(USE_MICRO ON)
+      - set(USE_LLVM ON)
+  - Installed from TLCPack nightly(see [TLCPack](https://tlcpack.ai/))
+
+You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.
+For example if you've installed these in ```/opt/arm``` , then you would do the following:
+```bash
+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
+```
+
+Running the demo application
+----------------------------
+Type the following command to run the bare metal text recognition application ([src/demo_bare_metal.c](./src/demo_bare_metal.c)):
+
+```bash
+./run_demo.sh
+```
+
+If the Ethos(TM)-U platform and/or CMSIS have not been installed in /opt/arm/ethosu then
+the locations for these can be specified as arguments to run_demo.sh, for example:
+
+```bash
+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \
+--ethosu_platform_path /home/tvm-user/ethosu/core_platform
+```
+
+This will:
+- Download a PaddleOCR text recognition model
+- Use tvmc to compile the text recognition model for Cortex(R)-M55 CPU and CMSIS-NN
+- Create a C header file inputs.c containing the image data as a C array
+- Create a C header file outputs.c containing a C array where the output of inference will be stored
+- Build the demo application
+- Run the demo application on a Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software
+- The application will report the text on the image and the corresponding score.
+
+Using your own image
+--------------------
+The create_image.py script takes a single argument on the command line which is the path of the
+image to be converted into an array of bytes for consumption by the model.
+
+The demo can be modified to use an image of your choice by changing the following line in run_demo.sh
+
+```bash
+python3 ./convert_image.py path/to/image
+```
+
+Model description
+-----------------
+In this demo, the model we use is an English recognition model based on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md). PP-OCRv3 is the third version of the PP-OCR series model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). This series of models has the following features:
+  - PP-OCRv3: ultra-lightweight OCR system: detection (3.6M) + direction classifier (1.4M) + recognition (12M) = 17.0M
+  - Support more than 80 kinds of multi-language recognition models, including English, Chinese, French, German, Arabic, Korean, Japanese and so on. For details
+  - Support vertical text recognition, and long text recognition
+
+The text recognition model in PP-OCRv3 supports more than 80 languages. In the process of model development, since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current model.
\ No newline at end of file
diff --git a/deploy/tvm/arm-none-eabi-gcc.cmake b/deploy/tvm/arm-none-eabi-gcc.cmake
new file mode 100644
index 0000000000..415b3139be
--- /dev/null
+++ b/deploy/tvm/arm-none-eabi-gcc.cmake
@@ -0,0 +1,79 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+if (__TOOLCHAIN_LOADED)
+    return()
+endif()
+set(__TOOLCHAIN_LOADED TRUE)
+
+set(CMAKE_SYSTEM_NAME Generic)
+set(CMAKE_C_COMPILER "arm-none-eabi-gcc")
+set(CMAKE_CXX_COMPILER "arm-none-eabi-g++")
+set(CMAKE_SYSTEM_PROCESSOR "cortex-m55" CACHE STRING "Select Arm(R) Cortex(R)-M architecture. (cortex-m0, cortex-m3, cortex-m33, cortex-m4, cortex-m55, cortex-m7, etc)")
+
+set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
+
+SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_CXX_STANDARD 14)
+
+# The system processor could for example be set to cortex-m33+nodsp+nofp.
+set(__CPU_COMPILE_TARGET ${CMAKE_SYSTEM_PROCESSOR})
+string(REPLACE "+" ";" __CPU_FEATURES ${__CPU_COMPILE_TARGET})
+list(POP_FRONT __CPU_FEATURES CMAKE_SYSTEM_PROCESSOR)
+
+string(FIND ${__CPU_COMPILE_TARGET} "+" __OFFSET)
+if(__OFFSET GREATER_EQUAL 0)
+    string(SUBSTRING ${__CPU_COMPILE_TARGET} ${__OFFSET} -1 CPU_FEATURES)
+endif()
+
+# Add -mcpu to the compile options to override the -mcpu the CMake toolchain adds
+add_compile_options(-mcpu=${__CPU_COMPILE_TARGET})
+
+# Set floating point unit
+if("${__CPU_COMPILE_TARGET}" MATCHES "\\+fp")
+    set(FLOAT hard)
+elseif("${__CPU_COMPILE_TARGET}" MATCHES "\\+nofp")
+    set(FLOAT soft)
+elseif("${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "cortex-m33" OR
+       "${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "cortex-m55")
+    set(FLOAT hard)
+else()
+    set(FLOAT soft)
+endif()
+
+add_compile_options(-mfloat-abi=${FLOAT})
+add_link_options(-mfloat-abi=${FLOAT})
+
+# Link target
+add_link_options(-mcpu=${__CPU_COMPILE_TARGET})
+add_link_options(-Xlinker -Map=output.map)
+
+#
+# Compile options
+#
+set(cxx_flags "-fno-unwind-tables;-fno-rtti;-fno-exceptions")
+
+add_compile_options("-Wall;-Wextra;-Wsign-compare;-Wunused;-Wswitch-default;\
+-Wdouble-promotion;-Wredundant-decls;-Wshadow;-Wnull-dereference;\
+-Wno-format-extra-args;-Wno-unused-function;-Wno-unused-label;\
+-Wno-missing-field-initializers;-Wno-return-type;-Wno-format;-Wno-int-conversion"
+    "$<$<COMPILE_LANGUAGE:CXX>:${cxx_flags}>"
+)
diff --git a/deploy/tvm/convert_image.py b/deploy/tvm/convert_image.py
new file mode 100755
index 0000000000..747ab29e4a
--- /dev/null
+++ b/deploy/tvm/convert_image.py
@@ -0,0 +1,102 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+import os
+import pathlib
+import re
+import sys
+import cv2
+import math
+from PIL import Image
+import numpy as np
+
+def resize_norm_img(img, image_shape, padding=True):
+    imgC, imgH, imgW = image_shape
+    h = img.shape[0]
+    w = img.shape[1]
+    if not padding:
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_w = imgW
+    else:
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+    resized_image = resized_image.astype('float32')
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+    padding_im[:, :, 0:resized_w] = resized_image
+    return padding_im
+
+
+def create_header_file(name, tensor_name, tensor_data, output_path):
+    """
+    This function generates a header file containing the data from the numpy array provided.
+    """
+    file_path = pathlib.Path(f"{output_path}/" + name).resolve()
+    # Create header file with npy_data as a C array
+    raw_path = file_path.with_suffix(".h").resolve()
+    with open(raw_path, "w") as header_file:
+        header_file.write(
+            "\n"
+            + f"const size_t {tensor_name}_len = {tensor_data.size};\n"
+            + f'__attribute__((section(".data.tvm"), aligned(16))) float {tensor_name}[] = '
+        )
+
+        header_file.write("{")
+        for i in np.ndindex(tensor_data.shape):
+            header_file.write(f"{tensor_data[i]}, ")
+        header_file.write("};\n\n")
+
+
+def create_headers(image_name):
+    """
+    This function generates C header files for the input and output arrays required to run inferences
+    """
+    img_path = os.path.join("./", f"{image_name}")
+
+    # Resize image to 32x320
+    img = cv2.imread(img_path)
+    img = resize_norm_img(img, [3,32,320])
+    img_data = img.astype("float32")
+    
+    # # Add the batch dimension, as we are expecting 4-dimensional input: NCHW.
+    img_data = np.expand_dims(img_data, axis=0)
+
+    # Create input header file
+    create_header_file("inputs", "input", img_data, "./include")
+    # Create output header file
+    output_data = np.zeros([7760], np.float)
+    create_header_file(
+        "outputs",
+        "output",
+        output_data,
+        "./include",
+    )
+
+
+if __name__ == "__main__":
+    create_headers(sys.argv[1])
diff --git a/deploy/tvm/corstone300.ld b/deploy/tvm/corstone300.ld
new file mode 100644
index 0000000000..e52b23da33
--- /dev/null
+++ b/deploy/tvm/corstone300.ld
@@ -0,0 +1,295 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*------------------ Reference System Memories -------------
+  +===================+============+=======+============+============+
+  | Memory            | Address    | Size  | CPU Access | NPU Access |
+  +===================+============+=======+============+============+
+  | ITCM              | 0x00000000 | 512KB | Yes (RO)   | No         |
+  +-------------------+------------+-------+------------+------------+
+  | DTCM              | 0x20000000 | 512KB | Yes (R/W)  | No         |
+  +-------------------+------------+-------+------------+------------+
+  | SSE-300 SRAM      | 0x21000000 |   2MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+
+  | Data SRAM         | 0x01000000 |   2MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+
+  | DDR               | 0x60000000 |  32MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+ */
+
+/*---------------------- ITCM Configuration ----------------------------------
+  <h> Flash Configuration
+    <o0> Flash Base Address <0x0-0xFFFFFFFF:8>
+    <o1> Flash Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+  -----------------------------------------------------------------------------*/
+__ROM_BASE = 0x00000000;
+__ROM_SIZE = 0x00080000;
+
+/*--------------------- DTCM RAM Configuration ----------------------------
+  <h> RAM Configuration
+    <o0> RAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> RAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__RAM_BASE = 0x20000000;
+__RAM_SIZE = 0x00080000;
+
+/*----------------------- Data SRAM Configuration ------------------------------
+  <h> Data SRAM Configuration
+    <o0> DATA_SRAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> DATA_SRAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__DATA_SRAM_BASE = 0x01000000;
+__DATA_SRAM_SIZE = 0x00200000;
+
+/*--------------------- Embedded SRAM Configuration ----------------------------
+  <h> SRAM Configuration
+    <o0> SRAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> SRAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__SRAM_BASE = 0x21000000;
+__SRAM_SIZE = 0x00200000;
+
+/*--------------------- Stack / Heap Configuration ----------------------------
+  <h> Stack / Heap Configuration
+    <o0> Stack Size (in Bytes) <0x0-0xFFFFFFFF:8>
+    <o1> Heap Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+  -----------------------------------------------------------------------------*/
+__STACK_SIZE = 0x00008000;
+__HEAP_SIZE  = 0x00008000;
+
+/*--------------------- Embedded RAM Configuration ----------------------------
+  <h> DDR Configuration
+    <o0> DDR Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> DDR Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__DDR_BASE = 0x60000000;
+__DDR_SIZE = 0x02000000;
+
+/*
+ *-------------------- <<< end of configuration section >>> -------------------
+ */
+
+MEMORY
+{
+  ITCM       (rx)  : ORIGIN = __ROM_BASE, LENGTH = __ROM_SIZE
+  DTCM       (rwx) : ORIGIN = __RAM_BASE, LENGTH = __RAM_SIZE
+  DATA_SRAM  (rwx) : ORIGIN = __DATA_SRAM_BASE, LENGTH = __DATA_SRAM_SIZE
+  SRAM       (rwx) : ORIGIN = __SRAM_BASE, LENGTH = __SRAM_SIZE
+  DDR        (rwx) : ORIGIN = __DDR_BASE, LENGTH = __DDR_SIZE
+}
+
+/* Linker script to place sections and symbol values. Should be used together
+ * with other linker script that defines memory regions ITCM and RAM.
+ * It references following symbols, which must be defined in code:
+ *   Reset_Handler : Entry of reset handler
+ *
+ * It defines following symbols, which code can use without definition:
+ *   __exidx_start
+ *   __exidx_end
+ *   __copy_table_start__
+ *   __copy_table_end__
+ *   __zero_table_start__
+ *   __zero_table_end__
+ *   __etext
+ *   __data_start__
+ *   __preinit_array_start
+ *   __preinit_array_end
+ *   __init_array_start
+ *   __init_array_end
+ *   __fini_array_start
+ *   __fini_array_end
+ *   __data_end__
+ *   __bss_start__
+ *   __bss_end__
+ *   __end__
+ *   end
+ *   __HeapLimit
+ *   __StackLimit
+ *   __StackTop
+ *   __stack
+ */
+ENTRY(Reset_Handler)
+
+SECTIONS
+{
+  /* .ddr is placed before .text so that .rodata.tvm is encountered before .rodata* */
+  .ddr :
+  {
+    . = ALIGN (16);
+    *(.rodata.tvm)
+    . = ALIGN (16);
+    *(.data.tvm);
+    . = ALIGN(16);    
+  } > DDR
+
+  .text :
+  {
+    KEEP(*(.vectors))
+    *(.text*)
+
+    KEEP(*(.init))
+    KEEP(*(.fini))
+
+    /* .ctors */
+    *crtbegin.o(.ctors)
+    *crtbegin?.o(.ctors)
+    *(EXCLUDE_FILE(*crtend?.o *crtend.o) .ctors)
+    *(SORT(.ctors.*))
+    *(.ctors)
+
+    /* .dtors */
+    *crtbegin.o(.dtors)
+    *crtbegin?.o(.dtors)
+    *(EXCLUDE_FILE(*crtend?.o *crtend.o) .dtors)
+    *(SORT(.dtors.*))
+    *(.dtors)
+
+    *(.rodata*)
+
+    KEEP(*(.eh_frame*))
+  } > ITCM
+
+  .ARM.extab :
+  {
+    *(.ARM.extab* .gnu.linkonce.armextab.*)
+  } > ITCM
+
+  __exidx_start = .;
+  .ARM.exidx :
+  {
+    *(.ARM.exidx* .gnu.linkonce.armexidx.*)
+  } > ITCM
+  __exidx_end = .;
+
+  .copy.table :
+  {
+    . = ALIGN(4);
+    __copy_table_start__ = .;
+    LONG (__etext)
+    LONG (__data_start__)
+    LONG (__data_end__ - __data_start__)
+    /* Add each additional data section here */
+    __copy_table_end__ = .;
+  } > ITCM
+
+  .zero.table :
+  {
+    . = ALIGN(4);
+    __zero_table_start__ = .;
+    __zero_table_end__ = .;
+  } > ITCM
+
+  /**
+   * Location counter can end up 2byte aligned with narrow Thumb code but
+   * __etext is assumed by startup code to be the LMA of a section in DTCM
+   * which must be 4byte aligned
+   */
+  __etext = ALIGN (4);
+
+  .sram :
+  {
+    . = ALIGN(16);
+  } > SRAM AT > SRAM
+
+  .data : AT (__etext)
+  {
+    __data_start__ = .;
+    *(vtable)
+    *(.data)
+    *(.data.*)
+
+    . = ALIGN(4);
+    /* preinit data */
+    PROVIDE_HIDDEN (__preinit_array_start = .);
+    KEEP(*(.preinit_array))
+    PROVIDE_HIDDEN (__preinit_array_end = .);
+
+    . = ALIGN(4);
+    /* init data */
+    PROVIDE_HIDDEN (__init_array_start = .);
+    KEEP(*(SORT(.init_array.*)))
+    KEEP(*(.init_array))
+    PROVIDE_HIDDEN (__init_array_end = .);
+
+
+    . = ALIGN(4);
+    /* finit data */
+    PROVIDE_HIDDEN (__fini_array_start = .);
+    KEEP(*(SORT(.fini_array.*)))
+    KEEP(*(.fini_array))
+    PROVIDE_HIDDEN (__fini_array_end = .);
+
+    KEEP(*(.jcr*))
+    . = ALIGN(4);
+    /* All data end */
+    __data_end__ = .;
+
+  } > DTCM
+
+  .bss.noinit (NOLOAD):
+  {
+    . = ALIGN(16);
+    *(.bss.noinit.*)
+    . = ALIGN(16);
+  } > SRAM AT > SRAM
+
+  .bss :
+  {
+    . = ALIGN(4);
+    __bss_start__ = .;
+    *(.bss)
+    *(.bss.*)
+    *(COMMON)
+    . = ALIGN(4);
+    __bss_end__ = .;
+  } > DTCM AT > DTCM
+
+  .data_sram :
+  {
+    . = ALIGN(16);
+  } > DATA_SRAM
+
+  .heap (COPY) :
+  {
+    . = ALIGN(8);
+    __end__ = .;
+    PROVIDE(end = .);
+    . = . + __HEAP_SIZE;
+    . = ALIGN(8);
+    __HeapLimit = .;
+  } > DTCM
+
+  .stack (ORIGIN(DTCM) + LENGTH(DTCM) - __STACK_SIZE) (COPY) :
+  {
+    . = ALIGN(8);
+    __StackLimit = .;
+    . = . + __STACK_SIZE;
+    . = ALIGN(8);
+    __StackTop = .;
+  } > DTCM
+  PROVIDE(__stack = __StackTop);
+
+  /* Check if data + stack exceeds DTCM limit */
+  ASSERT(__StackLimit >= __bss_end__, "region DTCM overflowed with stack")
+}
diff --git a/deploy/tvm/imgs_words_en/word_10.png b/deploy/tvm/imgs_words_en/word_10.png
new file mode 100644
index 0000000000000000000000000000000000000000..07370f757ea83d5e3c5b1f7498b1f95d3aec2d18
GIT binary patch
literal 5738
zcmV-w7M1CVP)<h;3K|Lk000e1NJLTq002n<000~a0ssI2qEZFV00001b5ch_0Itp)
z=>Px#1ZP1_K>z@;j|==^1poj5AY({UO#lFTCIA3{ga82g0001h=l}q9FaQARU;qF*
zm;eA5aGbhPJOBU{=t)FDR9Hu4Soc$+*_u8`GPP}YPv>9)0Tsjul0`tGf+8xSAi*3l
zwrwOw#nk4UMa(@tGds0gcYoM_XY1bGzH>WF&mQmHs=HN7o%((_!1>PeKF|BS?>PvT
z=V#QnFJJ%qm-uKb`1(5%iH?a&`26L26gukbAD>ZZ1O^iw6&3Zzcku1o_ix|9AAkJu
zw=ZA6!_V&keh>%*{1$#tNW|w)f4QF!k42;4Bz%JJ-@g3qGXUQ|efoq%B0qoq2LFV@
zM1TAKJ6tC^I_9sRzQjgHBYyw<DDmExzkNcZqW<&0{yPed{xf{R?;s}rUTkdiAHRP?
zg5NPwh%aA0!$1D^^$S2?zDFT{`}!5W6pKWHZ=YjOATAp9+o!+A#>OI1(SQE*+aCZ6
zw+MLdoBY1@SDMk3>anz{5nR<6xxq`Q^HEGQRC6G&IZ)U(ThKmB?_4Yy*{ksXSikxo
za{ouM=T_*v;XAKO+_xg%M`7TXQr}0p_k+@V%kJOEZknNWEEIOk(L3g46Niea;|kBY
z%yv>bek5~TNSzlV`-R;5W0~*Qa_`S9>o&V*t-#_h?p+iOZ!|3YQW3aS_<mG{{zVb^
z$Q?W63~vfXHn{^!634#8b0Tt`2|O2k?+wcy;o8m=!JBg5nas9B>j<!|bMTkEk!_jl
zQsle>Lf3h^!3lCJK!yxt$pAr$AXfm229u{nlU3*}1t7vn37mY$GZSm4Xr1#a|CM^?
zW5w(*^593I=UU*s7WqF&e7D+#UxjbCABs9bMmfmQ0BRMwpf;s)SUS3+_FSmk5%I*K
z)Nux}N$giL&j-2pr;@SA6Zr_DpdR35F|@iA)qv7*QXBfH^j(*XAL9)k4BHIyYh#&p
zF$`@%hhOeKDi2<ldd|iE4}9lEvF)@X^g%qiM={ysIQ6lt`X9udc%5A^b_~M@WX@B9
z_AS5{gH$FU7K3Cqz?Fcs5|F}2By*8(Sl~4iyk>wTE}TbXm7#JgO9tLo_`<s7U)6IT
zh3-?X{jkIvkq0ho0#^klXB4#pqzXV*IUq>^z7$(f4GGbBA~k^<`P7Nb7M3{9#kNba
z^IGEiAQ(S;A{zqK8t{e>NU}tZK|H*!o&Heix#SIPK9O5NiU7P|g7jiIo=KVo?^ZRT
z4>Iq$%y%PpUyB@(%HWM`VjovE2=F4fXOLa-LOCdzJdwI@fXaQ6sfAy#5d>B&MF3J5
zC?X$`UIJdxWAVJWG;VY%8=cHVB(p#&2f+D^(&X|^ZvTAU{B8C82chGLJFzSBgezxm
zbiqivsuvJSL54IoPld@=fJ|9bwlc5LUTHtl&;F=#L1q3gwo=Dcng1t+=k}HA9lEdq
z5M&6l@^P`b%6`x=ecKTFP-Z*AY2Mzc3CF{d1xOMPkuAy6k1Om4D*t7<?^@xxR=6)<
z6sp}}QuXj%C<$Ujwko+|sLURQvINzEGit*yNM|9^i_k<Sc$4>-UWOAJllk>Y?5Y%Y
z)f+~|qZ|pAzyqlakj4fiepI%c+%Q_@JJrtJ$Xv(5sRKErW%|0_bNq~5176cnBmt6G
z5=G!+DYB?cNvfi=Y;@DS@RP=OCAXhJ-k>HWt{bW2x^yy}DChv`a`1u;*^Z~xswZ~q
z+-D7eOVz|4PGto+K1eD;;flbk9DrlrXVubLeVV|fCU{%!x=^_;q1tu6^GwY!GP4Y1
z$fGE#zY9Aou|<G(_Oj42g~;FmJRJ$);TZ)rT@AK%vuC$qa;3?;)$HHXjLa44`W|!C
zAcGImAb~7gRbQ3sP(OQH={{FL(Yvp-zDs#u2m@99Ea!fX6iebG3B~bLDZp``@){(A
zOP0l-YlF86M+Dl5%ykWo2a=|AU%o6K#N^k&ZNbewr0X=}JErNY=Gl+cwgaN7Hws^j
z!Z9CYaij4xfMY;-c(qj<xTv4~36fLcI<N6uHux?wD+duo$(<~T(&RF0rQ-~~2-JkG
zxINS8TnR`n0Lg{%*+Q=AU7dYH=h~{ASl0P=Yu%g1&=DM&wLO5q0b~In^6#;<vbSsc
zz)ijHRylfNn)%lT&vl;08cpPaH*g(V0!8?+Kn8O-Ay*ESoTD4n`Ocuyl+K9Uc?nTT
z9GB2npm~z&CPBIc-~>=n_w%aC2UZ%r5wrhV=QyV8$3Y4UlS)Ir$hb!+1Zjl;&%9Tp
zX0`dD!s`OJ6}Hm`|22#PRWpLgkb^fo=zK}CF7fcD+It1Gb027J?kK7lq!mHcCFDv=
zT5NjPUY+MaKC!B>Z-_=0l&($f^nR9M5M+x0wG`kun1V`i|B}IXt#e+f$HHx)Kh;g1
zk;E2M8XY{R#F98vrG-}A5l0rm9|K%5p~9+~*wIbj!1O3{o<n$$J-O?w%yFEl8v`)p
zX?z4;n2@Vh_Ai^f=jOm=z5STcFdmi4j83CtUy&bW(2=S6cgjp<B`Z40$9A;7YmNP^
z&V61xb&#tWMia&0H5;;*B<+yA-Kg|l0NujJ!d4GDPXXx$DfAeMm}{P>o!nB5Z`900
zDrdq9*N%GnuqJerWgG$7LXge|={!_UnYeG!<hyFHU(`)Tx@P{pyl3rEZaGNG2hYe4
zb40SX301G_X;B5t78v^%C5^>hGc}$w=$7TKJ6S>-P<kRN*D<+a{LX$-`B8Y`{T#K@
zx@7iUSp1g;`w2@wflB4X;EH0Cs4sH3uh~*`1`9l=qI2YoCU1@NOzS?cbB19QXmw*K
zxJVKgl_^b@wU<q7SNX3%{o;@GRyQ_J4scA6M2jJp@+^~j*MZh~Af4DMb04TYhnndV
zNF7mU1vz3s<bbS_=t8A@XvrM7&^ylz_DEac7goIsmBz!QGf}CG*X)X_K7Z}o5Wdua
z$$;L=iZ7_7)Qzfad%DmEweJ!pgX*s0g)5zh*$tzB2<a_G;RO$Jt19|entj(6|24$M
zG1$>KUM!9opO%+Q6ADdT>CzeirIB2eB=3|At<^Zf5SzgjW;fdJcuVDBDT-uyYo%i!
z`VG)8+%Q^f=<HIELW8nQ%#}&nU7exx?)M)XgO{yKKew;^qMtr3>GQpm7(h0x0bD@h
zB(l}Yk$JsmzrlUf;5gCQP6)E@D7X)D$vv`&tZZtWS~WU03Tt}fvSEtzpj*C?v`U6n
zV9G&9fyDqKRJjh-&OK`V2!bd?;P@DV=uv*PdT_njcWd$8Ho&&S<VL0QWAN-ad|@g}
zQad!oGj=89iJ)DfsY<GDs%&^$?~1?;@XW3|-e7>pl4n9A#8&0M0LFz2UWYT9Qi4b+
z#HKS7DWan4-m3n2?bwFFc5HNp8*KZf9o|%_2?GU5U;+{o;0j-hp;^z>+Sd%;gBr(<
zxNrUuOM8a_$A&q^F}!UL9>T$I8b=cOup|I{$$f^F(c-J~UqE=U7^ppwD$h}sdpEb?
zZB&K;i7$@+KiJIP4|>NX-{OaWF*r_a8V$#hHH<ox)^}-QEiA$SFML_rR?;~S^P+C*
zu%y|GOur)^5C!)b)yk=@3ePEMUW`bro;XS|G}(ul+=O(_%YrgIuYn*izsS)(r`A5t
z(>x^0kSR<E9u`0(k&eou=he6CeXG^Bc~j`1a%wrR@ht+{2A&I&84nAU%Kqt&&}mEH
zuySbjhoVYsRw+m;MrD_0X-8oZH_Uyk_Fu!^N*g?_^X=0bN6;BUhz)}mBvLgh>spiN
zgUNHNcU_gVhOik@R2mnXN=xNPEq1@b?qwUT2{hOV3S$dvA$6LmeeLAFxFdiemBNY&
z2qoBjRoT?0%6kghmd+&Iwzw>ITsrM#fhdv0j)x5jjvb#~{2*Nzm&}b#;l-qJ&?!YQ
zmtX+|FESJ8;!>-vW9hiowbHV9TrsggP`5*mg#`_jUQ803n_asd{-Y-6PP^}rQ9Br$
zDFw+qkS@h5`;?<Q#+e`M{5PtpgT~OAK6t<|k76>RCV5C)$-V3fMfak`b8Yfm8{C(I
zwwdTm86u5?Or^cyC|bP>Ey4Hl-l;c&I*0)}&}(6fpgquF-!8XKL-9vv-PI>du0n0!
z()h!mXC+)V;7Z8lqi}hNRPF-`E1pD)O3KEh<VL;9#3W}!%rQg;NG*UZ3nH`lKC4nZ
zIIXs=TIM5l?rl@>P(8Z%n4!jImLQU7PjaO)%XnY#xIb`g^&WS-4(j_Ce#lb(K&iyw
zq>l>fBpnOQ(;tn2TfHw*@7-_kY!|oMKsxtMx2fF3e6_SI)E2rmdQTf+%ikG*^HDHl
zM9yP|qQSn{I=$Tz*yWkWW3r&*GanVGNU|ogV@=uXNa0t4lzfm%kIj{7#@4k^lh&0}
z^;`dgLQzy&E|y5c67nD9upZ}eo)z+*<}-fCq1~qz#Zy`Fd4ea!YKpF3*gp-EL_d9I
znY(HXo;G^-U_E@0C&#3-pX7+rSQQfepyl1XX((hIoHg{#s+ydysFhF271$I}EU_#{
zHC#QiZS+J8p0L5QtM_d3+w6eALuNoT<HYAx2s`|(p$nt;xY2VU>hd5ng|LH%ZPP<$
zdF{l4(X-JK*jEqFkz~!#mSKf>PAjkKca^pbr3qEo%pz<S_YqB{o7{%61AVJu7>`%Q
zs(Tc6LN@0qO~BH%s{1Fj!?yZ&cHM}rZp>XbHLY_l8G^eVtJlLve=^QSnijA0ffLK@
zS?$y+vCIPP5tT-ZNiR&upeHjVnL-Vbt0J%~h@7gV0>#Tb#p7&g0!b7@6u#u>U~ZWG
zr$+yg!L!ro+ZJ><5hOk;Qw)zF(b;9ZW>?G1d86mZ=shg$_M)<28&wQQ+y{)Zs_{94
zYt8K0hW((T+y9tW37!|+BMWjX&Dl!*bA|+)$&Ak8-_KKMMpm`16ELtIsT%P;E|Ngf
zFoeP%@{3iy<L}lE$2N|w^XvT!Tm8%XgKJ^yTBP-TxN+u4=ZAT6+O%-4b?vmwot6&-
z67!{ZshFA{N6Jgc%zHp7d`PAzkn$5VXpb@&@#(aCL}na;b)QlapH-5WFJ~Fw8hv|>
z{yhlH=-&}_xsjOybe04pu`pyg%VcYwj{F0gxW|VkmqO$aTVheUdTh4Qw`p*0nOwV#
zj@_KfZfrUqHg`{G(q}AbJe31|4o%`>vqXyi`KqaXFtmB0otSybl*MEeA#r)}Sw+g8
zvB~Z5==zcK=w@Q?YIq|8x$Rz!bgy4Q-rANgjiF=ct_I(3bLfy~9)Sjf!WSabs2~X!
zL(G5!lSY6e8lM@1BS+)1V~KevT+ZG41R8vOW`(KNwqOo!8G|rg_ey)cD2fP8k%AO9
ziY#N8rYzHEde0#wR&4d7v&DA}I+2q=FH;QrEz`St_jcXnn#FS{>F_;gR>x%VW68Y4
zd_gRQg(4N9GkIvTsBB>2j<>g)=N9kEE1vQ}ZgD)Bl}Kf)`=?xoSL54f!|SI5tEZg{
z2hE|~=DCCR<w)D|1w;i$*V=XO>ZQrMNmiN=sd>n>Toj&yCXk*p8E-i3XLQ;tCi@kW
z^PCQ!Jc3yDoGFOSrh`;6NXvdDP%4I;?TZKInO)1wf!OMUodYIY2C<<ja;Dj44u=1M
zP1+yCP$h`W5{T^{O;Kuf8Uwoq-=1b{)#N_Xj;&>tx1&kLXd){%n~Tn1ATyXKaxsch
zS~j!<4?bXI>!N;Y_I_S5)FcdTY!+S7{m!;~F|u|#w0<_UeLcK$)4z4yyLn^Xy6s%O
zvdo>dEuCA}uP2Uwt$H^N#SBdd(h^O|W$PPj$Hojcd+obX-S}k9$Y@0$bdK@LfpLbu
z`6){X@Dz|vy-zEq8@rlkw_v=QXCZ(9hAPHV<p9<(Y8ku9VG5i!xDQN$V~7o#BmD=q
z_(HjOz^8SuH+Xg+wtB}NgvW1oJYiMB^Fb7e0n%x(uRv1-Py(<Y)Ix0Uwl6K-rH6%r
zxGX00l6VS3-ZcWTLDB{{&WE;d`!=sSm&33OcCKD@tXwwDAGa->_O710&i+l(HHjh?
zBJnwBJT;zNDC@Ge&&~}jueZ+3wa+e?y)$)_E|VwJI<s8XKb9m^MrGt9NCmL;r^sqG
zlcARR!<IP+;Gfu{bIQ3bE~EeCAJ}9A)6u!IC|I=!?3hBCa$;U>TdB5hs>jx9rgoZw
zXH}ys6iqisg{X=EfriLp$K=Vdd1Ynq)^)I-9Nj*zdh344RYa5M5Z<GLVoCeZyN#3H
z#l6Axi@uE;NKx<F)!%`gcP>Rb-k<fYUqUjw77x>9b#d8jNHutxiWe#yT{EqrrOvrE
zi+@4yoYp%-diQL-bJh}EsU8pFWIB*u2tyl{$$rEt<F<@IU@%hRfdDpN2CqY4caP0e
z7Br66dk<lCY49Cz+Z_+s8jw*Ol`TN!h=c?FD(AY^y{{TwS5Isi{b8MRPtq4imgoV2
z4uQoN$w5X*0<(tSGY31nJ8Tt$4(Kx=1)2c`ODf=-d&aiH<Ga`W8~?=Cy>ijDe1{Ek
z*tdS(vvjPT3O%3+z?%$2S~hrrBS@;uzNNO%M#t=qC9r02zt?-0^xhSnd#Po9U+-JX
z(RD}X2%+P_jyxt;oTsxw{vkG@)r*41i)1Fq6h-H&Xr{@gxr>&CE7Qy=#C8v!lJKm%
z2A#qe^!rry)mqQKdTP6Ja;MI92(eX;y-!y*flLm(Hvm~2kW9xGs@YvL@T>^Nx6Z4F
z-5E0NQ+nwWn&<^nDrtKQlkMHs1>~)N;|j85UAurH>{*F)FQ0WUpIVp0t)U(DJKqbQ
zA})vZj3s`;5V1`?Z8JOVv-_<x`=-FQ!M~{wZ0Q3#I`1YNb)F4I^Z3)^+Gu#ofak_h
zW#p<Ztz*5`wZ>_;#nIuVm;my2pIHNW(|Hacbx?bB^H@Ax1&CZERfx_P3x|W{Q>z;H
zfyQy5p4_Xl?ZJbX&a=mCnRqIwhNRtNDUrGIAA}}e&jM`6!PsVGaOu!66@tU!T7*yQ
z{I2ux!^r0O|HBrB4O;ig2~=ap{BBclUEDJ&>KaiF+hlzcM(=Xx{BbLk+04P6n+10b
zGkf~rzAmr}?;~25&b6LxVXu$fWGk#6;aO~&@#T8=mT_jk+Ofv#bg)|NoHi$~!>bxw
zHBN`2CJlij<>(^KG@7d)WHgT!HIJ81EtlKZDqVXT_mReNsGi!baqjDbM{oshmyOvr
z&h56-T5Zg(pkiVlUgUt$%~RX{<;eQ6egDQ6{$*<S3XaMB>%OHU=p_)|z}h(+{i}CX
z4i4+eargTJ_#D`ZbS)qDtRDAoo_4(7?_CRbzCUf9JvC1sHclTHrr}lGetmGSZhFtW
z5N=<)ZeO`-nhQ72o;FP%x6PdPEMFUf`_1o9E%UGw2)D1?w7iE#b!nVAGtZtkPKV8N
zXRuYM^X|Z<EDOglMzz5`m3v3!fY^@Vy&Z&C?L5@^k4&>CTHj`UaK{9@i}`ci%z3r{
c65eG04^*k6S@ypEXaE2J07*qoM6N<$f{v~+=Kufz

literal 0
HcmV?d00001

diff --git a/deploy/tvm/imgs_words_en/word_116.png b/deploy/tvm/imgs_words_en/word_116.png
new file mode 100644
index 0000000000000000000000000000000000000000..fd000ff60b0a7a87c8c16eb1409aa0548228e3f9
GIT binary patch
literal 74135
zcmV)aK&roqP)<h;3K|Lk000e1NJLTq009yJ0049d0ssI2#*Zq800001b5ch_0Itp)
z=>Px#1ZP1_K>z@;j|==^1poj5AY({UO#lFTCIA3{ga82g0001h=l}q9FaQARU;qF*
zm;eA5aGbhPJOBXyj!8s8RCr!|y#;tw=k`7PLLkK5-7^`NC|ZiULr93?ArN=>1gA>v
z*6r4*x73Atp{}&JTYT*7Z@uSCQtrL~-*=yN=NZz-WOCNod#}Cre#a!Hm5dSp#~33b
zK@g-z{0|?IJ}&(y{BMjv<C(w&D~%b8fAWuy9fSX~v15qw&lJQ@CpK34DmF$C@L$IA
z&m1$x&i1YNH)dnUj<du6#*W92u!{e)2xMjB`%L2mJIAqO#?hEDf_1Eg@ZIca%y=Jf
zUnggGcTX=DH%})=7kdZC@#DucJNsc_tQDWYU(8rRe3=4%-Wa^1(HA4Wq%jOHMEphY
zQbt}Ke@^17<1c(HzH|(JEb(7v<QHf7=C)r)`sHlDg!J9U@NZ{(qwpnT$5_c~8pFSj
z_)Fr)@*g2R7x)k2&&AKjuO+=K{>{hYKUwiUlE5Dsf9uEMm5v>K&cDV^{E_%0h;KSR
zViiBY|6ceI;ms4@aQux%J}CWo{0aDv=0Aadx%gFh-S{f;cgJ5+{5bzy>20+AhT`8B
ze{1QV{M*`o5$T<j-bZ|rmA_fyHGp@Hjj`kJ65c)Wa^XJ*k9d)I4bscC;(g{nLV7*Y
zkCxsO+YiF`7<u#Y#B4T;|Ktzi&%;;Y+Ov+uzxeOPpG&`&^vX%R%*5Zp_O?rxr1Vqp
z@tZzh{JqAFzxicISGD*B|GcEpUtIjn#7{_`R3Ra0V!0u;RF_<=N!+0}Y**^G;StYO
z`a-#8t4y^SkSjN8Rhx9m4g68P8PICBXf<0k+O10UW<ajltk7;z=(Z{Kg=&1Cq(XV(
zRvQcjio|Woq(W8lc6G`Q37XUrZCa@WO=_7swL+d&8<*A)n_3^8To;j87ou+n($)uR
zoBg$|KI%3vRj0RZSLnop^0`l?t$A~5-shQ}L+f@@{(jnY2d%$@a`sdDPFm4POWSBk
z2QBHO#a*<dn-=%*CyTmiZX3;N6M#7#!a{s}CoSxx1#OmPJ+!=smUanCI?YSlO-nnB
z%Q{WVJ1zJoOFL;<7cK9mmEE+wQ&`qvUD9S=-fmsqVFp%oT2^*gR&|Q#GNpG8uI=ar
z(%WyOw_nR>zn;-?HM8^D+V;yCZI?6JFJ^XJ0J1wSj^J{3$JLznt6A+=vfD1LYrB}$
zc79#k`K-3H>)KCewV%ptJGrjy?~Jy;GFv$^TK)#sHQOH7H6I7En*UhW^ead6FY6kA
zl#tc%U1lTjZAR1AnN443G=2rFZTLL1{`1U+BO*Ro*YNSWh7ToV*K@3^wIQ?S1CFZq
zfsCqmGAa&dR=y3at2&%h^-gZh;k=sn@@n77u6ir0=B=FSx3g>B#Yb}MK3re-QC|H=
z8|pvKZ}@P1{d;*0@2;;uoZIkbem%#A`oqBby0_QYy$R&kz9HhZ^>wcUc#>E9GLTzy
zD5v@*Ag@Zo%ej>=0y!1W=a#>ai${)U^U9ykEqeyYDSIlf^xr_ve}RAeiM+DM^Gkt8
z*OxuEzWkAmWshtse^f$V=|j1tkL2>l5tRKa_a8ivTXr+<%PBhuFt9~JVp^F*8*Ne%
z*aF&sFksGq(q^+txk;nks8(%Ot2W_Lqu!#{Y*A^pNSv`jTOct<W!Rpe-zq_FD3BWp
zK^sj{krsr(V{)-Br9_`vDgn>IB6V7cGObvVQX)?-j!P<zNi2@i7f0wzV^ZoOlWK!?
zwSk(t08LY{u01q)XGq$evPEymSA3F`{oSqG{#sOZWmT(jT@U5#q0HU122ALng{`!p
znHIIu(hg1x@MO_0TGT_}%z`csPMc0)Zkric&}LfHVOa`#?4lJr`R6YI*+zipi#usi
zhX5?;v`PSxz%F1#mvv>AWliVs>dt{R9RsV|`&YO3t?B4r)7hKR*#`=77Ij>eklA^4
zT?cS&UB?XxneA5rPA!pL8SUpqoL$>?W^Mba^tO{gM(fG7EvJBtX5a*ejYY>ZoBqsb
z`Yp5Z7a*hI=ZwamG8=x%X#63g@w<$sqrlq6Z#Wvh6!AIdkI0-6fIl3Wbsw**{b*gy
zCwR=P{y4Md!^|ohd9tcL0J3Y|&#C<&zgD8h2U*qcfG_|)2+riyeUe}QDX_lbNPYun
zPJYAtU`}4cVIa5u?F|j@t*?J~LmjyE?)uty@;QIr&Z~V3kDOPvZ{*g#npg9BPR%Ph
zRWC~bXGGc@0&*%|;IxrY@w`Zz=R`b{SMqF5>CLcFD6dpxPKm@EPM(crkAXSBhKff;
z&ip5Bz@Pt<HV@>K-pgr&i*@wc#S>i0xS)a2>mFBsf@W*HYIB@&69SGxwMD7gq*QKE
z@$iBm!=V6Av;}f)0X|a#p38Iv3HojLKBG|TcYsLhMCnK1XYmN5O)DP-e=(_r+SILp
zDrJi@d8<5mOG46CSyDkl@^)EDaa?j~Y*JZNVr96tB2-=OuWay8?s3=N<CFYw*e%Z_
zEPgX7^Rrtv{Wxp;@${B~jCNDjPRiX&dHX19FRkt2;bLLCb$+L?5QOOzfQ4OF1fr!N
zlEk7;0smaKn*$%2-(dy2B!bNcyE;UOUkLt);LJgck^tr);H>N%THP_Us(lc^6R-#&
zNTSf%_I?B+un5TPkT4*kFQdH|Slf1uqx}k^lZcDyZ5Ptp&#h@YyQb|lN6Q&tZSx3h
zEIN_ibUeN3kBr7Y);9hMq&NJuw*Dvn*!a`hrXSWeeFvmB9$VY^t%z^d)PI#;|3!NJ
zXCgiYGDKw7N%#zp8MQ|;YCn-ElvVd}b}dIv-N(80pV;Wp@G;nuQ~x2rKiKeT9@x_O
z8L+<T^ZZ6UKa$_{NnYcJd5!PqHHnd@@dFTPL&N)c9EnBm=GKEh?|@X`Pj=nwxixQQ
z)xMTp{VI@CWkYu5E4dY~<W#&QA-5bD;ZK=Fp67E)p98Wblt~m4$y4@JPVr;8rH|#6
z0FQt<5)gbwXd{v5CjR^%F^AKp^j?N5Se49g{OXkA(MNm)*FA!R7%sMf1~IBlpbck^
zx&TN}fg$+#KTznlgC}xrVZ0W!*@nk>?KbfvqbS7ZfH|Oz1XXg0I;B*dTBb=Wmw;4F
zm$FTlvRQ&Qb+aaAlPZ-%mbd{FQl}MZr<7|aS8FC#t5U1uQW`=H-Tu10KI&bb>fPR&
z{eFpe1*Y8}Jo(YYHSbN#`FwWaughxBX0-R`?lk4@vSja}jD56fFRj?k<H@p}!lG`=
z{0`&XHsg}roEvl7E%PL*?4)^J$mJ|}GPeiM2}vC=x0AC@WD&??0qZ0z>K2wD!Rxjl
z;H>C0uSD|JF}$K36fy#9I!&BE9Ybr{htfNS!6E?ZC6LiED1kGoy<bf2cy=e^3b3aA
z60o}MybUeq)&R{H_+!gC8-H3(r8oWs=B#P_b4}y#>Gi)!Sl#f;nx>ycjPggMP{Xk`
z^<RTIBjf>pY!pHYw+^WLa`Z8~?(^)rBNB3IkL1>W3UGSV9|2EB*fK(xhA-APerZF~
z=Qi>*eUjhw5ob>0`|BG&k{|{mF%}__tA8)2_MM!%!$5ZJo7uH*XVttRA-n4J91%9!
zl)sz<RJ@c|aR`velL!8A=5XZPgwa66v$*0XICF|6=KMpRys}6C2M_0!J)Bqm;7#~f
zUbzi6+9(o><cUSfq+(@q39e)5QIS+E(-q3}+vSEGc#P8&M60&MsJ7rSR#O<GEyUx0
zAzrs5URM~e*&eGYh*KBhOYj`b;lW3{nF9rq6n_%81Gb--QmRTTQ%|haOsWF36DxG7
zMY_}-h&dc-1qeIpw9SAftw57jsGYc7GpUFta>*s~#FBV@X>3YUL~?7Wz6JG<pR&nE
z)#9yc^;UQI=y!YT_jo7X9yH}X*}~^0=YFuL?5EXD=W}nT^!>DQw{_JXA$=dM*$?tq
z=d}+n?6J)2G|y}`&T2Ew?G&jZ5@!A`n$tsbx@lIY1(@As1)b*iSm%LnT^tL#%@V*s
z00D<5Y)IQWOk(B3fwBig0>B>$>0RcvogC>MCQhV|VK531Q#+pBt?sz7y5kzKs{Jys
zvi%~ky6xiXmJ1?mPbBj25Y%{TP1A`rO@FQC%=vS5!yh7k1B+HS{R*sU{B@K<86Z-_
zcWWAs0_hFki1-><TmN-d<F{Fj|M8gJ@O5_Mmm+yS&#C_+r~b3tI*z>h&!yOs*Ce9x
zi~Oc9^P9fD318+nf1cNLBo6_n`J>#X4*>)p33-hli}*0N;R7JM?!D}~cd}{^XVtwe
zJ+7<eRa17=n>m#n*_E#WSrxD5lmjm#@&I{w<jfIK`XZ21CSim><vf2IrOiz-=P@Z~
zL)nu9ls|eC9?mHr#e+HJH=*KRS!MSEEWxk?v;jaB3DQ5o0T2fN6^~#72ot5;9HlOZ
zP;U#@0NcaV1>vf~a1|a4A{1Mr)rHY2d|E-2aw~u*v8rvcYR(y%W}{5A2~Xld9$w@W
z%9FNpBo`}FOBEB#l#?ozldAyL<SM8Fnv@D8Z>V!LsYUA4?dsHRs<dsoNjo@a(srm)
z3c)(?IYGZ2T$Lr4#ivxnCs)NJSH~pPMkUokNC-=8@>kb;E9>3mb?%BrFKuVg#C_2-
z?v0)Iq-x#A^6W1)>yJ$;IliFz`kH-|eK%#_ORI0Al?Q0)K3cekQ)MnVGYT+gC(Z5=
zfH~b333I#5^SVtEMwzp)%e+jK4MgRD%Ld@jiVk5VlDE!LSkgO%^e!tN*LDl(9Tt%}
zrZw$GV0GId$3M)uv8w$lu(IuvgjJj{H{*g7dD5HCrZ=5VZ$1g;@W|8n*P4dopb$sX
z@6sc(xb&vqIGTSMN#1zyvDH6f6|}DDSa#D<2|3NjfZQhF+noBZM9zGX+wf&xJ@B~=
zjbDH>B4I{3^KE_;M_%Js;1d38ZsVu9jUNL!4Id$G6LYyw!I|8Kk0j*Ozn@k6?g$QZ
z+K52d`G+<YZ)8`zF+!VH^D1AJXfw*3yt0>o5f=SZ0-^rND}Oep_^I5ICv!_42S$pV
z$Megd0CItf$8N$SxfP>$D7WG!R6dwhc0Z?$EO7^rkW?f=`e(dhd#r979`QU*kML3u
zsoD&LtG5O!wgxEixGg}wB|u)_pRhSFVRMjdb8y0zP}wF4VX{r(37aDoTcZ@4Kp2^3
zy~v!6@w!bi{brd#s#UhhQg$Sy7Re`-C?=Q7r&IvS$<@k<b?VdxRcbAAFl9=GBBe}`
zTBJzbu1v8(m0GMqj7lk&*aa3TQwx--h4>tNrd0n#qy9;$j!J8Yp41XKxh*WU%}-zN
zmQZOISIZQ8neJZuw8uPVy%4(MJ@uyVZ>_($sK>Z?r)Ak*T6sGyJwOX~)9h}JSzybq
z5$qOb?-u6lw9f9a%<eJG?luB*dQ5YAjPtsU^LtFdf*up(1SoF+YMx~fAvlF-MHj8?
zv|&}Zu)2#@cL{5{iKD~1rV~PfWp#%UkSd~89sMHufR*jnffemnfR$}mR<&FnfkYlD
z_>9CHUiq8?dD5Hz27kohb9`mfpDUY=uWCNN2K7(NpKF_e-@%;p=ARLEGMaxB@jV_h
zo4yBavYNk>klXxyZp(Lht>5Rha^?UdWB?x|d@%y)pWkpa9F?AcV7U#S=QbY6Y4`+$
z0ie`Ld7T85Nm5;uUH3s&?aj;qg+$ua9EMUSA*)iPP5J9$v;l2i=lF*?FXtizaf~v@
zR{Th`$u4<XRNI~a^2?qC)|Wl8zT!z6ZT?U4{2#Q5L#mTpl8}U)r-(-$Lm}?s$k5`n
z+vD}cv4+w}Ls^Kf*k8B9SHHtsU+SPPve%T1R~G|g)!P9(Wr3Y)>o|46c-0nr5ss>@
zF6wQrnnF*_wopS+gticf(r%B|?TFDA#TtrZlW=jD#HJR<rIpAgl_{risHT>xrdBE@
zHYn1X<!OxxsdWh{)d?x(^5jy)B}MWMK$%hm+CXYUm_nY$KbW*dk+dlxX=6gtW?9nK
zxa5Mkq{7(b9TA2d5s4*{N#&8rRZ&Sb;fYnDx=KH7om*m$Q_5`)Nw+b@P8Qe6lzZK$
zJs!UFb;G7_Czl?d({y=J*TDSF!FgT7^LxyQJW{rX_yaD@$3J&*=FI9g1GBr#5_1r6
z@W}we&e9&6LI9l_kRp%7nN=cIcT#!>t?lGk(`EZ0Vi8nqUJ<pMcu^!mqR`5AFsE-t
z+YJ%dR<vAS(Ry`N3&8oay7{sce9|Sfo)-%qUh_!7XJzBx5>_>H7Ns};wU&c3XKl-`
zna#j294$XdSl9ehRue~d^UpwD%TGXl>yLRYKjbxkCt-aP9*?eXK8i=MC8ueGFgMXA
zx8ZBh1{qyW!>6E4cKyeqz|G|Z8)3j1$P9Hz&qUO`FDY`IDz-<7Hg8I>(Pkt`L$r|q
zc@0I5Bod4Y1~)yiX>L#mvdjLRTmH0!yo!IXFMkTiEq^kn;z>O6m@|rhg3ry72hShM
zDt!O}hean<$x`@z9mNK;X05J3qb^V?3l-|JXhnIjs>WB}=&WyG+A3zKW|Laklx{Y2
zFI#jMTYcY{jQiQTdsxODEPW4Ky^F2rWJ}xGk~X%agUxSc)9YDsIg=H$s2$^DOMSIX
z@l*Gzr|*wV?Fdp=#ilhzr`ANKltrf&B~09*o{R)c3>s7FWK){sCpN~V)*|&vNGwe-
z6ia-TCvD>~CvCfY;`W5J!uYgp@hRKn2wBP7IaLz3#UyQyO4=TsyaR|yDFI@WO2E=M
zT}iyQBu-NrsVxgIRQe=U`zF`<CN=sQn*0oHzPb)CZMUnY$K7zid(wTubN(H__Hfd+
zZ)ep0HNWlZy4z{(0a^_eVJ|J)LyPuu{>&4zxLI8^vs0K2{jG;%UI)!-9bVdHUeRq@
z*<)VaBLIjk9H32yu)2+aHSM%^6bM9OUr2()F(h?J?UuF~m$eT6gSNqCZG$UX`&YE}
zf<G%;ZCKrUO~RU%E1=LCu&C)mdgFz)O&5WTmW%%sKa&2&mA4aXMcF!|<#<NR9~rH`
zt!n{(74chE>z`S+M$qqQ2jw*Xnb-Dbe*2$UO~2$c{RCt;exK9uJ(@z<_1|SR9L;V(
zE%Ysatp7$rb{z^KUJGT_ewtnLX;#fgSv4PK*M0<mCz6CMsn~!-6{y%!FNn8;Y7ghv
zypvb`HjrEOmZS%W_(xL9232aHQ*)rhlV8o7JZSG^SG|~3^-5;Np^WmE@CX$FpP5zh
z(z=T0v&x<WvdcMgY<MQ8{F%JUXL75=$BKUgBNTcvyY%twQc-*>L;dq;e)(e(B*Bfv
zB~{Cks}xCPDnpT0zullK)@w>Nnkt#PHdtBjrfRk~bh62N*!;VkbDr`lf7@@@F}FQO
z-S!>xxa)i0hkg!y?yvZ_&Z<7Vocv{<_R~uVZ=Z~P<&Vf`e-3*18_zpG@#ucXZsTKY
z$(<~@nZ=j0=<+f0dJk<|KvIXVwjn67F+8;{0s#TR%}}75STb=&y&|nzo>nhQsl}^G
zNGu0!Q2K*535kWs>*Ohg328RY#HAL*CKn(pL$u@_9+5K=qEiZEQi_1MB;=gMGJTOu
zyCYt+10gg#acfBO*07{)5lK5D^u^)2vM^0WsHP@N*BGR0@z-_wB=-2G9q^lWchKwy
zgXcY+wDE_@+fU7{yTQffy|jD}EkxP|=}hF0B(gzRTa3;r(lbcatwW32hgWo4R_zql
z?4s2>X>~WP>f-cR+0Fsl00=k$BG0mRE@bo0><EzBAxtf8HH*v{g-9Vz9*H?NNc>sd
zBBJ>!N8=R${8`&{IlTo>F7o0>tbe2^Br#{C=fjz^jzc2PA6YFY)-|8VYC5s5@vp3=
zzq6Z9<hJ~s*ZNmh<L_BbzveXln%(d-2XCkTAmY1q_21f1_szPxud{2v{vX<K&ir4r
z$*q1DoZ;Y9ISjU}7m;7})(CAtm{HoioLl``ZuKjiGiZ#8v=Q+V?~v9UT37RGX4PwJ
zD-UH<y(~%CSyhMDRlc;Y{6+95tMUZ^%^wlZ^X6&g3%OO#0Xdb=a@vf9Bat?|5VCbk
z#VQDpggB;5s#F+CmAVp@u2`!p)#<8Ky1ICMi+gm9yRO@3_JjTzFZmX{<5T@<$j%>w
zAN<?(`P24qpZ7S@7xa@Q@&v_RqWEhh>m}6?$@?jxm(;za@28|bO1w_VmuT`?(jOfP
zfA%-$_BUDXgKYM0W@u)L8WvmWs;m!5>WG=x8JF4`m)H<*s8bl~H2Nx?z5-C`%kjbz
z3?;bkarJ{0vB}833gc3?1N?d7w)n(tJhx0Nlo|N*7(+pHVnIyOwy5N-F{#@DTjWVB
ziZ<+sHWWn}5Rul$B<02>=EWthk2h?L(`|;}5v$oAtuBsGl?5qE{1eK&WaaJ&m2L^u
z&WdJ7-R;g3AMv02QuOK%v>U&jR&iosXaA}Lv<8Ztg#EN?7p;Vi0F_J!t=L7&cM?1Y
zBfU{{ia5XxTS$>U4RJ;UXamBCRFPnn{0HF72yI3Q!vl`QoKZ;pS=q_~=B#SIzS@SX
z>CHR@akO3%VawvuThFGqo*oH1QsfcC4#)A#mXqkEu4_64@^JpNoXBZ8nbZ3Bx~AXZ
zLI86RcGfliEFrVuhjsPGBxKfo17y^Fm01gXAxYQ>IB57t$f@JS3?h#$_|)-eBL*BX
zJ*(vA1DpS#`fzR)Le1M;KtnwP4UOk%ujf>w8Opmld3EBG+E+&&U(Ks~C9m!^0BxUj
z)vtp&c*Jucr-rwHKpp_W2apU0B3`h;Mw@35f503Lw0bz60Hj*zKWJk}EZ6Ev)S6<o
zu3V|DiBr{ut8vrqcF^qfn)gIx;SqWF-|=@}2!HrW__O^1Zx2rRvTwp4Cci5bX{8vV
zFd{#J+>PW-<SCGciM&kYYoS0Zg_tSCNC`x0B111F9w*iNm&5Nn=2`W+OZJm&+HRJ%
z(>}S|J!N-r+P;X?y%D;eD0zcHQ<ti)NK%*Uw52L-nM_}dawsx!M`Usl5S0XMk4`L%
zF%&|ph|_N4$(sSX_SSgA=2-nEASQ7$5Sy|!I;8+%Co*|E0!w&OQFu~Ggkc*9laQ2|
zkd&2>l#`H{Co|;5>o&w`H=;2mDuLUgRNEufr4gF)NKIw1uHH9guWK5wh3s_u#_4Z!
zo%T@J^4FDlpHD3Mbw<^hC2f7_Jp^}8)-KA~MOk}jxo9asS9E?me}pMv`CbCe&@H_g
zqXvb=?ZVRkfMpq+KJBIv7?-shm$wfs`#-R}eL%v>wgFC|mK(sTW|2QFH`cUW7jZSc
z^=f+CWf7Owwp{=~o{W~W8Lg*9z<+Qe6A(Eg;>;*{vRY1pJfoSLM4okxAP?tHM*a60
z4dBdonRQ1qYQN2>{Tcvskgb6^S+$>W=7>Ps2MwQ*oQ)SY95o+hSAPiRtgDik^PY`7
zoHK`WDxn0RssU%Xe<!yF{m{c`jPl-S&6^U?A(d=9>l@wz&@)Bnhc|p`-^{6d3#J}$
z49pQ>a~?>Bo&R9Yf6zu`P8rG`sSpx5Gs+xMwq_~%Qk|w)t0`0K>*CZ+fr=J)Ww%4(
z?ID?mLyEo)>pB&D&sE<i27F(q&@U+T4+`xk9~1dl$kR%0M6N{6L{1}5CJ@=f?rEV3
zfSDZ3WN$J%n=O83iZ+vOfb?fb`L$X8)-{)&&mGDSvvlz14yM~{uio#cJ19%KGfCAv
zSy`2;u1ZvwtF#rUNusnxVY<S|q~eI=;)uke2*dU${q|@si}UV~Lcdj^-z3*<OwewK
z*X09o`VAb38$q7Pq=JZ~!mz~cAxS%elZwN6fXY)OW+@VLB--Ggar*pN9U{->czuB=
zyBCQ`9Z(XjuLwwP@}1P_J9(FTYPW-~ZJfG&oVI5|;sJa8ot~2)i(2$r;<_(qZ2NP0
z)#dC~Q~qwsK=2WPS_tNYg<=+mMi5FM38;u>bwgC(K=X7FYMV9@Hu6}Owo5Qc);R8t
zYad$LF$(^XQH;_F)B<@>{48(j19%7$wS-k|K<}E?-hU`0^5>@Db6$);HVTdS4|qB!
zk!Nkw$+b;z!6EqI@pxu4!VX7fBlq&GZTN9*!}l5WKd!C&4oI*2765a2+Sb5xHbfpC
zd=PmWc+Ur(18z7#^Rym10piama2_E3!1<O{{oX&7505<Wfk+Z~6sqQ)xBOb}E#UTo
z+IIknLfoy>@ZN@ocK{d>vg>)<C%gWwQAbZsJ&#Uc4zfEy^6c=)Q}s{4k#e@|3SRI4
zFuaW>Y&Vlf(&AXMu0*RTQR}MZhSo4mmz!#*WAc3y<~|!)^;_7k^Fa>`xIZ)O^oG^#
zEAsn`0<Th-h5W4KVI;>P)40K5yJ6!vi+Ma8sbmp>C)RO9;1B2|kqQ98{D{JdqO25g
zg~E<g#CH_+ktyJb^W$reuw_rN)Vo>YL9fKS5;Sdkb*)}guhG>h^tCc=b+ooT5?63i
zMO0!bPqg&gqjX#o=Y3Vp7KI+p(T#HLh6GKXObg`Tk&{QeK32CeO1~*mUl6V@4AU2d
z>Py1W>M?AV>(|Tl8+nG7xG~y*PuLWW$E1Qt1hK@TSoEy)D2F&=4dvd-9X^K2fV9TI
zi7kGq&0fi^o{1fvNj-LP&EsXA4)Wcun%n)8ABw*9dF|rENo&8HT5@h$)%Cf}M#-eJ
zWEb}n%x*WM3IfpinJMDd9`mg|oH=mB-CWq%o{t#b;459+WnR*0THG}X=_4@p;E@9j
zBDhX%L~0ve+BPTwi~^A)Is2b3YCGI<+=%dBA&5tzRvw7Z5x%(=f*dC`QZrjlk6Pd|
zoBo0Yj-&Z^?tc>@^5;hcAI_ipqX<D@4$s``zXo}D<f;D(%n@xnpGnPA$$!8DQSHa;
zs*$`wJopIl#}<O{k*W`JAR$zLm|OiJKA2zoac=F$l2uf4#I3J`Rp-ME_1thE8Fq3T
z-_2=w2h0ILq5MV=3Fd>hcxXaiSN+;3XLzXt{cR*wv(aWmEdXsqtLU@P+@!FRUk2F1
z4%6vN725K6LsMkZE-%Av_R04;E_~H<%eR4D=L7yV<o^nJeL}83lG_<_yFu=K6k?(X
zGeroT10f<AJOshjYVp7mktdE;VS>>-ZqQ^mV6+<;8avo;H#9K8Y<3dJ1#!tj4g+L=
zmHdV%=sJb|P5vJXt`D7I^`EocSJ{F`J(9b^^-a-<?J-GhG9a-zLEj+L*UAB11%d!K
zB51e7YBzxuaeCeeO3<R0Dt3p|xj=#@2Z&eY#HsS)R2xKdLjh8+2z61E3SV9f#A%TG
zl|T{zmjVsN{)xo_$)y3w<$+1%AxRZs$(3mP#HPZFT_?MxIc7>jcxrW!p~6Q~>ZPgh
z)>Qjw>VuOz{Iu;J@)qZqMwjSjx7apMS+|>FpPlBQ{lq7I=e-u6^#wYr^INa2+GkyR
zCoSG>UcB47aF=ENZtL7#mf5=j0dh8IvlzMDC_1^dl)FnsrGdM6ATxAZ;Y*Mpd3hu|
zoP<#dE$%P@5F~)b9Y$bT=dk4N5t~BX-NOw&9o%X9pCM=@k%P3|ELjjP@^Yy8B3QJl
z@!ZPB(^5t^Dkq34HV0gAu#ifC7mo9%{wL0##-F78ZEZcTc_hsMnj0^E>c6zHsQxQ%
z;o&6^ayg#PN&X!UF$!@DTn(4wa>Wnk*7E2iJ#wpP;}QPY^vU{0?m2*SRP-FYms9^P
z3<sbMC<G%M3~!)~l)0fg;yOX)D_IpUuPZ+Uoj{5?T!Is&0!eBE@+x1*tHk4T5_v|I
zYmq!pGr6HMR)@AmM?lJLh%;>3(@r^G1a)2tcwpG~CBgF(a{iv2PLcZnXN4D0;IK7p
z*cfIqg_upjR<pm=>?4@H$?8FzIqqU$aTCbZ!r^QsCo2eJaS{Z3vt`V%b&QGZ1Rjg*
zG1$;c><qb{qL9BR<ZE*H_xW*MU$R?v+D+<qp1jL@^6v0Sdt=geE0Vj_hIXB{QKzZZ
zs7saVZD^q)t%^0|M;kUkf<tE(M2gXDjMZ$2)#S%(auIN16ge@9yjbP>SjFa8c|oka
zFgBq$Hlb9etc+7u$Ej<gH8tVd>R^4PzrMo9P#!R`)<3PrH>JuisnS2OGBBw!B&j+q
zsX8j963w58#IguO1zJGSsdZ7w+>0Hms}9yw2dir#!9^vtMkjYf7+M2#9j^La<MsQ-
z8TL6O-{Cdmk;oOVXg3_0SoGtXy#uTE4z1WTylk&=$sWt%y_QA0t&4UGD1<QGFlw^4
zIY(jY!Hfd}RJUmD5rN(-9=T=vKLHAfM6$vo2m3ZGb_+X<3pz~zVwC9g;RYYc-7~_X
zA=E+~wk*z;%YjQPo6%9_eWBH@ytTTr`NE3EbITjftZ79q$J05{OL}TG@2K*osze^d
zAIV56k!LMx9$WkogHPRe;14f>8d3ki!ZSjo23st$C31D>^Vqxy+>ucGd3Ft4JYN7g
zwO`_&H{r;JrmtXl6FqO-IV!nGc`vmA)>4r+sDB#X19L>$9EMLva=1ZYfYm^vO=j6k
zl1R;UY8!2y&n^cf+HlxtBRX_OV$RbnSXB`-sVg8Ax5@qEZ++Hl{kNfQ*CHMy|L4i~
zBMSJPg07ICkrM`#z+(tmB4{{*%%NlsA!`5$eq{9`s~cHeIe5T9RFQHs)I8&@7Qn$O
zAPOPWIGM>20SAmiTq1H1_(xm@$>}ONoFd=vES~rL%r-pFX5Yr99k8Fc*C%;TY|<XJ
zwkt{1mL#vy%geNiZ3*g)(VCoa{krhPykHoAlB+|sMLx3i!Rk$6`pqG#yfAehvacx3
zhKPjxDA|VCgslnk?Fk7b32|l0gj$uNQKfE7&^E?s>%%lvf!a!6ZH14n%v)dPV<_`W
zs_;*$2uLgoNUR7ph-@)bpvZ~Rl>_Mfh$xBDmD-@^qYV<m4CVewb)Lx$Zix*p`UWRm
z^8{_Hy|xqOklU1df)@N+o%_+Pm47Vny0LbjIrD%36%I|I<-2)!ncHfe+b%#hfO&Mm
zF5VgfPiAz{^bWeElcwWI56y%r8^9;G3jjJrpcyJCG?ZZ0hNlk9L1+Stcz}Xs4pmSW
zxA}00GK1)pGmZ9DQ4y_d<5<y(rV#H7NwpB?k0@<Rt<{yyqoyA2NRVLjVoQynCef78
zz`dmK!mX<Rb#24%qx|8K$A%wx{ZoGwS==Zj2~MhoMw>s#=tK%}@Xl#HtU8kO0A19t
zC7aI(KC?Ad#nx&T3<ve^NsxSQU=h+bkv4CwtA3L!Y!Z2Zs#he4@rUPZJpMqcE_+T=
zs7D5SMxQ(@Iow!qQfv6s{oa!gj+^w*gk|r#ZU5b;r#J8ka(mU{{)N@=6a@{EKS3xn
zdC;I2^@q@41Pz6gF_erUWDXz;XyZ*5PqMg^AVnLIJc0{Zok>7cL2Pjpc;s=jaNr64
z3C^Gd!fYOZ;Y?;G)^Bw<YmGQYZjYQ~^&hdk=h%$9SmJK~#QloI1A0}DF1|q<Tc(n2
z!~6oGO=NOj<fKABU3pMyi<h?2QO2za9)@imiCcV93tUy%?y5X*Wxij+`oMTtehQ-H
zr3s44=&0hj*b14vO0KF+(ACB#*2X5+As>rKstijm2SSodLK2IE6SoH$3WIgqBlT6$
zy6PBSVpXEbiqVwBXo?_MqXh&VFHW}=5;li{BRX+QX!4Grl%l}oQvW0{uo8V$pX6rm
zln$@too>l{T$69}m~?N*+-KBl-<!JayV;c|7q<<dKEZ4cI;v7r73C0^Gqb}A@n8;G
zt9xkPKAOFUX6~XHh&gQ_j5x(0rf^`+>{ipP7Naex>*j<RP3rK3`?AGEZgGbNDV#(h
z+kj9fAG6~`Y6Fb6#}Qp_nXNT+V|gq0C5X7ZqV*z7aU87|0FXzDKirhS?QkdH<yqCh
znIn0z0cnuI7JnLlO0WOnW=OGUq!yC2Y+Dy73rU>_aE8+co}GrH((_S9-4vZ{<oPe!
z{Nr=0hO64f97)c$F-Ihil(va@3A6!3pMlL`@E_zUdk#4p17|{Rxyy6fgU<6`_FVtH
zd(~ODeFMJFk@H(-*JH+jOG20+gpkpndVT1sFI^3%!AOqbNHT?s0Vj|If3o`Xup{^q
z`EZUQE?FfaA>aUrEiM92-oPa%vQ8kg9btIb2-uMYih={EWwyBWQxufC@5te$%WUVj
zEaOQw`M_AiK0nRw1pPjhyh9ORr&3g+IEmJ8i<!7BW_pchQk(CzJ6$L4vP)_ln^w!x
zN?FntmbRH0a#_;)F-aTkbQ{MftanV<<fYgiqAJy<G-wUg3F^|Agzb@WTOwqeqvYIp
z8mZV6soof+*%YJOgh*wBep@uBO?`~E7R-s&RKjW?R!7_w4ri+9vfBt1Tb9HTpOhb+
z1U_$!OxnUVhUD#0sil!=HQ^JR0umd&wT&){Mh97wgS^dMyT^ax!N6&cL@aqzmibv)
z;jeRRFDz~EU$xVO5R`E{t-hUB+(t|Gb89J&JRppiszH>4u8pJ)-BoGW4w3^Oc0i0m
zYBy2>&4ovMr?7xeMqo6Er*O#M0Et381a-qHXPdEuFM;#NHm{0sG%QKp{88)?N!A2z
z?P<FPUz`Z;D_t$NgCs}X>gKbnn$E0hI<>O#<myHQA8yCyE;#NV<v<F@U8D`aU<zmy
zl8`VeC-8De0#_6`;6=!8IGWRVYy{kwFk(u8%sqmJFQ7MMHGIY=?Jzu7_rbc__kqmX
zch=RslUWUio9J@`c|;)WNaT?)8hO@L+Q=hK#ff+^ui_@~Ifdt0%#6E2rrqN?{c+z_
z9|V`4@oK)|eEX2sOXU2a#q(El@C{2i0fRww9SEVmC>n~Tp*S+ekvWDekz@@gK?D~G
zLPbC-0G9$pE}`H-{E-9$JaHzA1DVEi!VF5VqakN9qBQciS_4eva-GKiP0q*2`L(O8
z=PQ=~BAazTQ}6Uoye&erTc+QoN$N>RYQfbUlUfxwbEjhUA^nywl1jfcRDGsx`Z&7&
zHUEmI0_q-Tw{B&(Zeh1<VUsp7!v?0zWAV8xI@>v6qj$naKl$bmWOpzhB=IQ;Wzvpl
zIm)5@SY>{^IzK^|FVjPkULR-JBGD#V$7vJC1@STn(+QgG(m;e<$91`c#QcP08)Ef2
z96IRQsJk}9^~2qFDX5ETB2pT{Q(8ijTLSeMZD?|n)jPyBj!)=vO1vxJ)+eGCzoc0G
zt|9-6X~n<JsXeozYhd|K<H|j>av!b0bi+Pc0&CAMTGGQggQn2DHUU78L3Bcr#%Z(D
zHql`7Xp78Q*uzKhc-n^it<z#t5kMPp2pkakW3!w}#%(?gZhJ&tC-DcZp}yrEy?|sb
zl>lq@2y?ipbXD{DQSz*A=8iZNKN3bH1StezieWSg@mOS&<B-TTpoPlkQqf429wiqK
zIFsA-9e^h{(FRG~e=&#C1_Ry_Z8B@#%cy>rPjpwkjf4%%5e;uP^5j&b7sLk>M#;nZ
zBSs#88{S3+OY<vV`Ue#+=2yJPWVaj)n{v=+`cpv}pGDPN_UjyUz1!sV3VD1^-p7UD
zerp6-!e{`5L1Gq1LIN3;fM6D3O|S|w3&mSVW)(pcM-)pGBUqzFMg<C7ZSxg)U4;6_
ziOlw797jWAX<#f3fin|m*c0l4QHUO-=wWi`BX*fw&QZW06!`vt&%ge%FL{s6zLy#H
zJLwN#i-6*m2NEXU5uxphN$N_N`9Sj4@20k1n|HT)$<ws#FwOhGdh4fybAK34{OCmF
z^Pf-H^KVwPk7YKoWi@Pm1)EvMH0zmW12b%KPA&4CSmK|!!&kA<CpsG$7pCj@P@QfI
z$bc{cr7lLd1%*w7p_->^`ig&Oqr(_qAwF5`8u4WbFp^@5jzhN*g198^@o=Qyg0?Id
zwUbI?61gun40ayCP#FQqA-T~{*Xozl?U%gED{+^*ZkL;WzxTv@{HH%ETk)nL=d+n*
zf30lqUALFAup<Gz)E;3a2JCin3j*(`c31%}5@4<lX49Plf{bX&=73EbkNBW1Z9}Oe
zj<o_9TwNks8Eoj{%MwP7+oGt=1#X_sb#bid<h`S19RtfcY~&fG5R?NjM|u>88f=d2
zRn2F?91$l){XpUm9|w{K8Ke~MzY@9sr4al(+yw{1i~ti2P~>=1h~umO1Om>dw(0Kr
zkFXa(V$RwsC<IceBSIt(=Z~%0`6u#lBLW9>YHooWWzK(5h{-4288zu(;4M#wuKhfw
z=4w#4(epmR_YLy;ntcBjLI)^<giyg0W;I0<DG6PFtBOo2vZzQ<;UiX3fHM~UD7TVa
z;I;rDo-A>KIZChui6S+sBVQ7{$?QQU7am@QoM_OIj9`%qK5nMqVTu`~NSF+uBACgw
zp8~E@<Z+7qjQk!v#kL=2Gag`*?s1v?Q24D+MNfV>Tz7Yz{-A8e^NJ0>DC=(M_ZgEP
zrIa_xaD<Y6qnUj)<pxdZr5U|6{|3!DWiovHm(K(5jBR|3t!ZQPYS_f>Ot+D#HZkos
z7k#mxzQj9jOF-P_P}!D9`8LcB#cH`(J2J5-Jh2Rbl7=fDesB!N$#kU&x?-i?20r<N
zsU!?sBo)HuqcId~^+hV(4yAUxG}(&DJnsF`7eMrj)^Ck6a3F<?O5PTgQXH<Yj7V$<
zPihX<H~Od=JmifYidHvOmru&=0h1qynEi}=*;`53pH1EJ!>r=JXIGrVTsN45L@pB^
z(>r)y6_Nm#0T5MC6z#G?;GVb3GIu9D+I+MD0SB3zO(78HQb%bcF^5k&fJLLU;d@K|
z8S`G&ZAC5zi=C(?3@`5-T;AEgqGJS|eJk5wBn4+iVKegl?>K`bUQ3yq1aXjCf=z{!
zglwsi%FzT1&yW908!+ephc;P_e6>hsJqEcsbEL2%PAI$&fHr)R6w5<I=7=J}n^NTA
zoPq5?gy?*GEx-De|3>9Y>#GhiMdCjB#Csy9|2tyUC$Z%h!*`kdAEdy;<o*TuoS?8l
ziXz^n2{TiyKngLK#E}J21pqyyC-_hCg7lwwsNvCuEYVyuF!^w8+UiSIKQaedOhM+M
zK+_=LXCC(Bassb~P!aiBDL_DK$7#bo2jI^Q3Oq}3e^KPeR?quRu^k_>6)&;rPkY?*
zeCU+t!wpZ!bWg=ieIsGxxtMxW++K=#h@xJlz>kFB-$-es7z>3A2mx36!Y}pf`e>Gg
zRuknBEjvxqjtoUUaLm5qIkxI{c1tVMRk4JUv2mr2F(sbyW&ZN2AXN=)H6e-hp~*E+
z2tt!#%qfCR0sBB?hLQw*8K6q4R3=sc3VmgQzCvcGh}W0Pbmba-jaFL?e*>IBN^P-1
zSA;omIDim#WXYT5saq2!Zi-Lc0AF0VZf%GvGeWf?QdtnK-of<-Lv>Ulm$M_1dqR?T
z`Rn(%t9QAm_PXot^q=@p_{`^{7QUs)|2eVXG&-u-wgdj~6b{zxy|id2E#5;*U`^OZ
zi}uliJ=VE9`A`sOgAgQ*qw<MVP)LkUsEYu}fhx)0l8FwF5+pid4gw)=IUae~F^U*k
z!7%`=?1J*vyNYLPJYho%bq)Ln00y~R(fK*g(Z($ZaKUl3om}1W_b8<JBjs=ZkOwOE
zNG}M?S=VF>HGKG|S$y8i2kHJ_wE1EkFK<4Tu&!R(%Y%UhKC>`74+rMt*1>-b@^Jo0
zIsqRet>xw&(ZBN+N6qWN`f3q1ul|FoSJqd(!V>g*<WugCobgiV%FiRpE`{y2_&-5`
zAClu210E-h;YQ8@FelQ?Gckmhco7m?<Ri?Hr~)7&0sQlMwPOw>lRKH*$mB_8A2J1+
zjlm{kka0M0XgI)VhHnkA2viYc5r4!#ML7<-9Jvn)9#^d4*C_4;g`&{9?<_0&oGmys
z?$($5Z+Rs&<@spCvk{Zu49h+pTx|~BM<Gv;|2yRT1-YEy#TD#7K34wCBZ;$JOOzx~
znu!(=EilsDOEl*=&HPH3_}WFy{XfWee#{nhvbmjXX2-bc?c=7jkDJ(Pm(n~wslk0x
zz1PHQ@3d<FwCZ5=aFQ$FSCARX)X7!aq$*8frP@#-*O!4auoT4T%VfGrI2&Y|3YkW%
zf^?iYh)7@#I<qm_{3vyHv^FcokR6|#r%2t1B|ZpE@v5y+@+}zb1An446;ax{FjZr?
zrZe2IJ3Mh;aN>S%U5~4(-A>ufly|e_C*5Yf8JYf7;<mqM)ZSRsZbpd$Ely(5>I1YA
zAqWO-?&}e3qut%&bBuF~5;p!5Y9UEQ;Lk^v_<&K!-=HFJApxX<A$6yKjSJYju(Zn}
zK}y?3k~cFXI7z}r7id*0x~L=iwJi#9*?`Y-x7jGPszplMMw2(O`P1~bv<m^?ocV`2
z2sl4xHUGfpe_DP7`19r;FiV%wAPR1_RBbd{`(j<=XX_em%mI0_n>a9DikY7g=J3Hd
z4!B6`-dk7mwghP^PTFSxW1G$J#v>2s&*+>2f)6Of;?nMnyY+GJDK9Iwor`KThTlWJ
zFOus=1HLDWK?7zG05pJQ6Tt)*pGx3>HFFd^PeWp672+%cXoF~j&yNsQ0dN44fG5w&
z23=?XkDw6miVA`|>^TB?3!E*y^07r5A(#ks!q*0HISyS^{(Zs*DD+QqdGVU>J(ujZ
ze#zz@@|gQ-;EWgiQy=!5_LAS)KSHXlA-9wFQ{?zAv7^HHbNm~6Sy51#L#)Oqt1;G!
zjL{k+pyHxf6UCcIHB34qB@a>BAWi9`>AkeXL}|ZY4}9SZ*7Y<iJjk+cV{2|_%kN@S
zcCu-E$IaO9IAf37EgjyI8-vqoVv@^a)Z1bb)+Z!v<om8*X47s(0)|nAsMM0!NflAa
zCBf>g7$BD?l_`_U<J8;LhB7&Z7gXE85WZqT&$pDo!5~-c0A$K-GW8a;arpuh^$xkZ
zA}*miKA|yI(Gsm{jnFrSCRYcfmbg!<9h<t>ZpuTBw><7J<w=K$PkT&zHEiLh%Jd&6
zZ@w_Aq<=*#<s6{(2PyY1N=HWsDgs8kF{au<Gq4~I6G1S@p^@56x3&v2;grJ=tJouw
zCcrWKJ*#VYE+5fDQDk1U%YqdPnDmw!snW8ZQ7aH15Q3Y#?IyI-at`j`mK><i5+GzR
zYrnCy?dtN@tIOLkS|`;%(xYuCRT^ioO{Csb_CR@qe@C39`#qyc++{V=0s@`*I=H6q
z(L&{2)P`?4#6@sxZCl`MyWsdbI7A+9fs-PS4SYzh9^0+B;3nZe2D3%M`ky|}KOG+4
z8)cCb?~0oFl;^b9W!tZYw+NvRkngMH`dOdvxqgHN+^RWq#2gI8j9L)1R?-P0z+GGr
zR936PYKb?Cm!^1khor^{AX36f=y#-E2kU^7bqMM+S~}cDz+E=H$pc~_$e?E$M3xZ2
z5@xl84s%fsLYyZJgJwJsA#R+dz;DU_h2imyKeN?uu{qB>On=IA>Z4w>4tcKnEuh@&
zzn@&6pt0`}`_9bH3&`}m1WSN*IKny{YaK?F6-{QmFw|*bWCrjA{~Se>ARuyTh*GUI
zeVArnrs+RhlHR$Ze)@E9=Qk|p88%}to4MO=@og>(_Igfl37A+DKdD-oT!uiQ(G}>D
z`H}<7mWHbH{pI<=+O4o`L#BhbjSoIxLQb(gUb#c2DoRinBQ(hk<*1QjG}|K-+u{^O
z_^(*H5To4y{SEynkVmdYbGA$&t5PJ?Cn#&;bY<{;ho-Lgn^@qST<4g&(<yn6qkgZw
z@_>WtZtuj$11BHyNq#AC@>`K}k0{stoVNACoa$kWB&^;|EB0`m4P}fNd^l|+EEI?B
zct;f!LNB!wkHW287I9ux8usQ>t9)fbrx^=<cv5%M%D7Qa(6ToA2km?gf&>24cFD($
z@drNX0IT@YR<wX_aQ^{j=_H8c8JPv*ogZlzg4px<Q%c)LW>N9a|8OCURzBhiMGn!3
zAT5GB^1tL+SARrg&VM7L_M?ot4<$&25J%m+kgcUClv%@zA88mNvs!A89%2z`w@1x<
z!fX2L@}jGeZ4~|>1#mrJIOH;S&`9YP{FiZ7P8(cU|8NH68KDg%HJQZ<+91?OUNwm}
zxag5?0f;RgW<FMc5d#3Fi;KyMn1ir@PeUpdOjb@CK?pYSFUJQxXc+ndDj(!pURNmi
zFY<oJ?0DB%w&M%7-~~4IaffM7Oqlkf^Rl14iw3=R3677@n8S!RrZH#HekC8l8Ymbe
z1%#Sm=o7KL))Es)iy6Tu)*?h(xvw<bN}*P3hy}dD-N6($Y}5=|rw-9V6QvJRPCw;c
zqO>=zxOBhEwmrs{?_pEw*_1}F1^XlC?Ul{wmZdeuYKvpkTSdieQ<Q4G8hU`bNU7Q(
zJ<1h@c%nmIr`pb4)rJyGp+_1@LUlWWw1q&3c6*4fAXvX4*pL&Floghk8L7_(h2oS2
zGDWdMQK3|p%QZ!aP8bl1PTB}RYq+60Ok0ft%uiPDBWv=NcloOJxXJfA$qzUvZg<q)
z<391xz_~BRu6{?6{rS|gbF&-zSM+cRVci|H2Af92F;zZ`+9Awnx7^xhmLd<DLi43%
zJv^iP2U1>#j7}<tMl(8b<&|*ru#hd(AlkHBxg&vhQF$YEg*c{)M;H<KP=jdRxp|@+
z<PpcY&mehQ#T!17a==HprTT}Xg)>K5A%Yb=2tHVZ0L=hlN8F;wA+3O0+w_IFf)tM&
zD0q;$NxMiz^4K7)u;T0B>iJSdNC+Z$&j$_z&YZf#oH?~`0hu*?j0ZDtHrj-z?T?=M
zxX+B&l_l4sJ1FvD3VxgXjtLRhFj2#45JA>Ri!eePt3(?;aey`eLXf0HgEsigFyWt2
z1Krshqc}$L86uwn^6ewP0rEFe00M|vs%NBDsU;8zm0%4OC_pSoz&9_Rq#2<l@JXHl
z@;pV3M~ogX54+rPo~{3Y&3j?|tQXj%7ws4S;=Zllqs!{}pmp5a#J;tTJIhCa@O?rA
z6PO?)6VyMzR+U+-atlaK@m2zJ@Pb1u{HJ>j^?DE7@a?}5+CL~Cv?dPI#38z6fR>tQ
zaW5@6PgB1$1wZl~D|n18*n`N!Cbo}DYVk;H3`wbnNF9cKJ(^8e#i&si%c2Tm!Zyfa
z3l#Dl@$fV#vD~x_^Rrm8gY_WrVh5-Tyj2B0y6pi;B?0<v0fsFBiR%NB*99b}2PUr$
zNlXvdXGLq)Bb`f7ZRZTs6somb6}t6+*1&T@EOSKf2u-9oZ8a~1G)-ZNU4e#flt3=(
zE(cWy1{}t#JI88v`_6nma_NVL{2!*3oSaj0c~SGgs%~or><AbHf|9nImWnoQUh0T!
zS<p%IfOeWA0?(ydNFtBq<DS!P+u(z61lt=7qatk19cj~*#F>??faM=F8&|gU+s09C
z<J_ZzbbN|VYMV;RH6jnHAW47whdj3C>2KJDfaGnIJfr3vaRn*%_kN50y`#ViD^aJG
zTB5ebsMH?4iB4iJHzFbM&QG1ROvgr^QCAzxpbYb2F|!`?o%x2k^jcg8ML$B}hbicL
zitZ((E?h<<D=sgHW*V~U$f~!JK@c&*9MDG2uS?#m3AYL%Vxf<4!K1`5*`E>|PLT5n
za{rrrPE*ijiZ*glK?+Jp#(c@>V>SAtb%QHd%<Clha-Gcx{mq*M6e{jF0AoO$zsTVa
z@;SzXp<UY_Ea&41OW$KN4v$^@i(5gTTZh@{0jtAXWOtMt&r*;`T{LQA$eciC1qqPO
zQJdkBCz%%h2<aS+Cgh(8ae-vPYru~SBJ`<vEyuIXVN-ZtfBZn7cE~izOf!r$;}Yq<
zvUuJ56Dxj=&DhTrbu7NZePVa)j9u_|MrpBPt|~!M5f-s6G#poaIUEU!#44C^A~ie0
zw0tE<L~?0JVrihR#7|S=r!RxyIzYcY(6A{mF)uJ_9XJ!1yfQd>WvD@94#poeThLck
zX$n9am3E_Ao2OQ0sT3J<<vO`47yTuy<HBN-Fl}j|s?=Xo;jgRnH#A^Jf~T(DO<m`x
zL_2loc;)R5nuCt|`#dK;7C!G)L-sef>^Q!*+mx}JvJX)19fW-e@b#cpncu?qWg{6|
z*nxH`-<QqN!L13iIt5^6m&LYzq{lRKr)lm^?3e3b(%r|=)xWHB5a2qs&9EbKrbUpj
zs+m@|(#jUZoPLo!(om{6&W&9N;w&9(*T~$^HU&6^Y|gh+l6h3>|Jck2wnZWcL6ZL(
zE3AGNw^(8S=utpg;cKH1oYxKCAOzji<UthTM1mcgGw0^L*dlrUH}cr{gP2qOItx$T
z9XIO{|Cz6A%C5(Ca@s_^OCdi}ENFwPN~~>ANsD5X0NQ{pK%xi)9Heca4X#Vn29N|m
z8$=9TvV5Qb?HBYezqhili5(&Kk=5>FtJ_fuK1q=``1ce`AaG;I$2tVsfNv6QkeGpW
zxPpVo2rUl$;i($zIoHYK1bKZ%6YjapioRp3kFdF)v8BJeZ0&b%v$)((E^m?3Q9k1k
zYT;w2a4;Y}Q;-D$23{usJuTVF0ovoAlAaEpz}X-cROD?nxtj*uO?}=pfS-;k2o8xb
z>JR9<7IbA$(?`<=DESYHe`Cn^j^Ek3r<k^xX_}l<d;C+{BMeRPX&s8GdqQ;09`R)!
zvSLqViHEAlMY+{Uz7cTO6!;~V1*O%5q}2r{Hv}3Qg7np)+R`xHwlMw1Fhh1oVn#^f
z>X4+BcnmdUL>RJTFdP9Rg0_IK<kl3Z;7d@hOHgIRtJlS7vY^XF>NbXJw*+AkmwJ1k
zy2M{y>W>}2s>(oRHQ=YHb&hLtN@#bMcR3_<PmuLsK*&XNr)%PU@k`%QXM8ew^Urgt
z&SI=>b+;K~s@R&1rYb5TjCmuk;}jBsScHrY8x$}gG^58nbEjnnunWvFF4)<>u%~Zv
zH_*SNYj8=IE%NZfrp>avRRC7B5wNP21Mz3XKWejJOD=4_>=$$0VhEC!@^A``n0F-e
zsHnn;5?ssQYg!S4Zcg6B{7u?oWm{pzw^(6?6-V<iP)HIFuuC4@($UA8dZxB?4k2h%
zzlLVOb2bT*W+1B7EHY(h{OpGVX1}H{zZTa{(GO9?VG8?!;;!=?Mv|3ngU#TH@L#kM
zz$HW>i8i=HOa2-}8>DAW!)C`z#J(hkO8t<PJ<-P=y~>`u;q;N&^G9=3AD`QSpE{6+
z1JKgpO;t`Cq-Ow+6^LB|F5IC&GUFrg=D-7Le~uiElJoOs)_9z)`GPI}hAlnryrs{z
z#o~MqxxP-$Uz5j43N!I?AFoG7RwY@%mQl7Rn+Zs?atevvTmB2;lY`K5MdT63yzwi*
z8RScYp496^7rf{OUMKD#_+<Go#hs;??<wtj3VHMl+w>xvx|eA?$7s8K40{7o4g}17
zz;ph6&NFYfn}XFPO}rMGT+LF8S;~&F$;A%I<!&h$PiP8C=?F^d4A-^AsxgRO2B|F$
zb~tT*v_30Bp8?K<7}kX*=3;&hz0o*5mOK{AHBc9zG+<<Pb42pyu%yjFhOL460)Jhh
zzqU9uu_{bo6QZsTLf<L2A~?D{II<!rs=_b6DnL~qnutyF-GPQK^isz~)w7T)7TGi=
zzT4GsFmT4x@)hr-=6`eRj^AgMpIy>oTGGnPdcNuax;9oLU?aBZ>%rEmdDtK)2?;wf
zX~!vq?$CmsVG*W<-6qIxA|wl(fRS%r{6JolHYngZANhuDbOfUF4dOMQ1!@}vf20s3
zMIp3<Z2kk$Im*EU(OJ%)R@?jnC?tj;v4z_FM|uliWHo|+c3EK&Hs7Ksu2AI4+f9$7
zbOM)<*NIC>zmzH>Nk{;s<|)tIM5%y9rF6<>KNQSqb2YwAi1`<%P3ZR&dyQhuTvG-l
zm5GNJ!E7Kw3N-+(MbJjgX=9P14es2y>~PT|I7}Ec+n)z#1oi^4hb-*Cb++dcyZ5^D
z%O;;MEKwIId;p(p3FQ!^`#tI+-15P?NDF?tHO_2|u?~dO5I!RM*~sn+Is8sOpU{~7
zms!rYY~_z^*<TJ@`dpeUE_Vy=&;Y+6@4qM<J{@t@fn0ofD8TS)lO&iZ$wEjq@f(3Q
zc-b=Ogcgc3bD&8Y&KDsFJ~Zq_y&iPKldI>wUNnr~3xPM>Op*Phyg|NS3`Bio40z-e
zD}0O1xSyrmGtO`y(;b|!^o4-!M|54kCGY<uY5z}}-JeIdydHot+NEu5b{(5o!4k_^
zV%>z)4v&d@1JkfU0nEWTdR45t1mua)Y>w1z2s7k|=+_4&ZVXK<2*>gew1c$Oay8mP
zRpGj#5U6U&Met}N!SmB@2WNtG<$m&_AbCZoygEu=7n4vICu@j{tB;k{geMdSB^3I}
z3Vh<X`6d+mDk^;xHC~Da*MtU`O6_8s#>TdcPv~*h-R3p%e&1W3O56O?q=G-D7o!lm
zxUi`g!`Qsni5(;RX!$;|9fW;7FlhHk`kPg%cBIPZKhVjCAm9Mvn%m7?)X{lw+r9*G
zSpu3r;?`_XNSa^ZiJUkNBw0vd9u@u9mzTF(Sl)VmWy|?ht>XMb>lsjJRm;hh%_pEG
zU>Dr#7EwG9F}e<pvk3W{ZHv65ABYs<usy$-JU7=uqY?sS126}M12BiFr|*rw^`4N~
zujtDz#WtB@|3&^UTEf1km@DFCgiA`yZFrJnHl=c^+)NwX$dLkB<)Uu@5<-y0>~G<r
z1_J`)`iAXJ6MKi)<HWkIv5r1gbAj#dV+XH09_owulfteO6z3os!T<v9_qd&hVPsmo
zIup<xg7#n)R051}7*gO$Gz4qsVRAZ6-d~aTb2Op-DqH_ITlUA8jW_IT27UiU6P`AC
z!rbsXMGjC5Xv1GTPtri|6o|`sgqUF@jfsDAbe!;eNsQwE)k;%|LuaCBbc%*eAvBDc
zA~b4ITp@J?b#XfY!{M2TH&fI_3OqL8_tv$5M=m;49A&d#V2S@`)88DQ_l;Nm-_f^S
z)jm3S%bPU+Q(AG1^8Th3-<T%8@OyC2;R)NGU~3Mtx%=6S19sEzaL{%*#Z(7r+mzGx
z#_F2`<BP+=pNTcmX%(R<rNM@hAWd<&q9Rsar;@j*)Ge``Ib|X0VjxsqhLaayHjOpZ
zMkp(zFcu-Nj+WJcEwOPm@v$|r@uiUon<Er?;qv@2`G(+xO+oT4{_;X_3^}BGLOHhc
zxXBw_lr7GxcECx~?WjB8oN~}>>cfGvpNU%jhA#i}8I>oNVu*U5fPGiem^a@m4>?=J
zTr`5X^P1*~?I1|o_#ozm=3t_s-73P&_swH26+YD6wlf;gq6KhF5NvNS@2uuWs)!>D
z9VWhDQG#vF;vl>{;&xK*BIPb@X*+5Ah#i5imcIbtTn2tLLklm2I9jokhwtTC+j<<3
zR`W<v2<sI2I#O{XsYIuAaU%kNSs-y7=o=}W8%^P)`Uez}Vh+>H*e$#D?yxy8>x<9D
zHFMeozGTIqf7}&*4@Lk1ZKT3LV;;tpmq?bIqRl_FK>~wjNU+rc6$=uXo0Y~5SnSWz
zn0IK*W5n9}S<^5pzQ$TjWA+U=K0Orj9fe#F!cCT7G6wS*1kqa{b|i88S6fMMrX;hV
zw;+xgF{cnLUb_flUf0O`2XcRn><;v?vh$z~TYtl@W(c%#dfMs(58EFUGsI~F9^Oow
zG=X!2#~HCyi!})$H-<b<T($q^{Mh%-IKFc=@H0~WPAQiuWte{&#6xsTN0UY|9zrkx
z!=cLyu+RCO8ut2Xz~{Ao*V`|$%}3e1kJ+q`S>_K;^_PQgH%C5VRlh;0U(hYT)8fmt
zk!bB8EjdlIzNe(OdPDC0*{<Pzw)qfSa34$R8JoDvXYw6EX}3ezjGekacG_+?Wx1EG
zJP>>2Cbh~^JC*uwwY*0mYm1lH#3?J|RaFX2H6T+~#L9}IVhiKtB^(K5akA1lSy^mC
zS&R(B5$Fr$$Ew!FD6^uJIgzToaOL_?<>nyOmOxcufT|dUke|BJM_cQyt@qJ2c<Ngm
z)lJU2PM5?U*OWbODF-~K-WN3c$?zpF$F6=)lmG24<)@dl53kuPt)CaYIXk%)hb#`(
zo&_j@&_acN07XIEW-5ZqZfFP1>>QfeHZVHKAk8v}Q$QTz?q6;p9eo@j5(EbBg@ch4
zE^MiJ$`>rcg57ZykXFlcbcjbY*o1`tkP~j^4_`+rZX`t^#9OE~IZh-J@1=@DHXlSq
zFNmjeqecWQMd0MA!zA2COh0pX!i>Aa=N{4&o``QAj=P^i4pG>*oHnp#O1?~~nATa0
z5^a*mnkd4+%Z8hx4a6}kuS~cefGM;gD@u`Z@6mYBre%;-o7m<XW2*(bcC*6+M!!!f
z;4cBqO_U9w4K7O12HrOA*+HDK@Kh=lMTl9@SdB7JOBle-9`NrYuiwcL_0qp?u+}SV
z#qTWt2CEwKeSln^BEL^4;8%(pp$*zKh=d3!h$oXo+9=IDl;M|(vIzc{`^OzS&0hYQ
z-G7wb_A|TtwA+&dk?)cAcS^oN$zo$F3etszdyDY-(B)!iC`ugE19Jj;&0c?5{J*8(
z*UA0%Ay#mLEj`9E{`6`RBJZJ)hbZ(#iu#yjKa=hvB@K|KmoL78FJaOR(x0N(qZISL
zIrRQh?$w`<UH1x`_XJD3AG;jJYxl9pIwq@yOvhq&u*mHbBFcTD>LOyg6!D!}MUzHW
zp^Pa}M;Ga0i;@ybQj{gBsuFGd7PV}PB4JB{VhiTLqm_JTccd~uPPtB|Tq{#%;)#ts
z%KT8}`fyD_sCHYZ4%-yUg7p<ahRQ%)mA}5m*HGh^(uidvo=HtE+Qtd0M#$OrnjX9=
z=j8kSXFMCT;yr!-vB^75%&NRRuWpbZwuSwPkPtu~^mat?8iRDRTd{b=I<vz(s}o*q
z%bcC&dAmiMDIY<#8L_#k6ldDCTLGTPVHQX<l1h)#HehLx!3JsT73a_Y3^Ld##AmsI
zQ8_{K#o0`8Z70{X+JX;CAYKK*e86|aafpgG_agjPM`)y%%3TOteB&|a6P7f0k9@|R
z(eq!HZ~jBpVvM<$LJv{MS3=xHP8-~naUDa60&VnG(?re|^Z$!Bf~YAAs0xswAttyA
zpbf9!?cSvcNYa}6Sm`j!?i*K3_Vwg^7kM8hpI?NKeyIev0ub&HXYdmt&EaMb+MssF
zqs|JqwI$rtA3-LrdSTl1EIEHm?4==gdp|Byw!W8@4|?B6PES~UKcs-4DM6%-s0vsT
zMrbqH%4s7PJ*g1TQ2B)PT3mj)&R##o9y-CAe`K|Pu%?sa@4V`J$QXT$bXPgqz+wLR
zCP0u`jqx-TNxc}OfClJe5fJ(UPEgcQ3VW3%>>Okp{$?xw@T#Y%gXI4&@_Cm0KBB;1
zDeN*>M?psN8^kgO3Ws7bOcA{lcY)%+rm*LX9=lJDEBtyw#>auVU&U_tI&|&ZvMuk2
zW<T$;_~6*7T};<JPTlIR?g`fR$P69u4Iul9iz!k>7iwb)bunA@(Hji$8}+gc>V)-j
z1s0FwMX7VcHMx<RY>_sZ@ruk?MP{^OU9>zqQjv?a4Wx?D6ks<uw)yb2KDy#yZ3&cd
zY#<3us|}yn5IV6jAi3VxQ15MM_R@DCg_|Jn9E+{laXk~%w|h-`EMnfvvgIEp=l?lr
z%efiF*XC9aE^5N@3WQyxYf%WnT8eB9ORc($06Yia40b1Q+X3dcM<7v%i*=%r^k)3;
z5}$u41R+86VskroXWuCJvUzN`8rcpfSxd2$M><+d@{*>vofsvLn7m1QeK8R<vWrKQ
zv#}0N?4|N{Dj)u_c@YqNz?>s2ZT>#R%sb*1y%L-ILqgMV<Xsf@68Rr7MxG&snHK=K
zHzV3WYfiM7(s+gj+92TI32Fw*NVLJ7SU|ho-;4$bcQHE)G<Hz118v?DCOko`;RY-0
zXKQ*L3dyO4ocELG%j9*G0#V*@6+wVSAWDn_sFXBh!jH3chfs~6(*ipjnNX#GSFvJI
z?t50$=$^H*&F9$W0anuMb{9E3V)A~Ee1D*L)HVnaJc3z}r-AbzfJ7VI0>E(mr!b?{
z|FX&D#~Wi`zsw%G#I_w}TYhK7=f<}8I%3V$D}#RDnnO=>%>bXDV4(z)AUB!fErXE)
zwCO=e<j6(6u}cw4u>PQ+kICVIAy#va%{exHhc#d?`8+`0hsggc3Od8LWFeg5Zi4a*
zixwbnxR_}C6|(z_T)rjucgW?j;W2Gz#uok-R&#mse!BHeYuX*x40oJL*#CWK+edE2
zFR-=uvl+WtViSugVv&VoqPM%rO0ftbL0hR-m1q*S>SUXB2^%!>e7Q0wPMw2Y3t`&q
zaBVgyl%UL$De{r6VGa+7Qf$B|t90xN_Sr}2`My#t@j+)PR8thB-X5&k0a-jmR|=&e
zNLLZ0uf^J7EFbpKwRmV+Tvg4eg&Y;Fj*51O+Fpqd`cHZ;a@Omr<sYZzeLtu8<jSVr
zj4n$SR>tk2wV3qAm=Hg#PMFtjn%y?M2*+>1l#Q+P9Y$bohlx)CiC)tGT`lsTMEYk(
z4~(Q@{NZA^$e$4kb@bn~RD^FtfEH)-Ww(!zr~S+tkvum3@cPHrP8~I3k9y&x<4|qn
z0e>356LYpN*yP3gm9uY`EqXOP<D2+~!H@&ykQd17{l1{%!?NN3q79NXQ?l?6ZNMBY
znMFxJz)%|6?Y?F+T9a_oh_oTlX8ea1=ckD^-e3g-Y(>B0R&uQ**FEI=9Jw5^c%QQ(
zuj4bJRtu^H+}hP-($WwbM3Myo;En_V$)Qy%m~i3782MO^A38#R5qqCl-4NSuV%x9T
z?-lGG81OhO_#CxH-{49bLQ~}b&<3{v)I1@`&<0FC7mThy4o-M;!2XdTw&O3h={zgE
zKBmEFcVKAT;{#3~4txJ@!H+|^g%_r@U>?B|Z!|?<{6Q?V#My81Mf8J8@dX5lUbgEp
zTlL$R9aguU<o+-6e3RcAd@pc0!BM~*>s4mG%uLtDT8AC1G+~&=Un2Wo$ni6BgxJ`9
ziRJ&qHvbjeJQ%abH1S26@($hd70vjOrX3^0$9=Mw&cr_Q2h02yTX;L0(!mnz$Ledn
z66-KUhk*r!a=TK#RjJslP;Q7<u8-E_MQGPYajU^*nR08qq5z0hYz3l4aJMxMAkyyO
ztK}2RaI_;fU`6YSvCJ_NB80jK2vdPbB^=uFFnv{MVr^&=?>Ge*YJIg;o{9=rS(RIS
zn@7SPU-j()hI_*&KNdUt1<kT|^y!}@WqmVg)6cWZ&M$2lLOTe)1OQFdW&8R1dB_8p
zVwlru0;Ch{Y->vYf5{_GG)U(?^21U`jMzp@(TTGRB09vgA8ls@@dLI-PN%z)-XR|9
z$d!XL=_ALXN@t-;M$(&%*njY|gs~C(rn7`?<Y~l^+}CW{vID9)x5<~j8no*3xcY&B
zJwrjySzX?`;`P(@*a5M#A@B^13%C|zieUWzp$(#qpTz=SGmkbnSR0qcMRNMg;(=?T
zwU=!jV9N$vHj{f9xpkBMqcrC2L5Dy3eFrTea99e~Sj@WdI%-Hm18T8H2}pVuB2Thl
zLAX;Hhhv5<!A9~N=0l)kz9+VqSUItRORRf{9qe^^-Qw|;5lInf1DOG%shAGJ_ema!
zHu#MYYCszd#syy|-#^Ly19E$kSWO?>+{<#WvmJx1Yl!`8(EeSE`|pB(?{J8vAKn}o
zz2zn$Zb(3#g9r)Ez?O>Wfn|$9SIOrGa(w-=!-LmZ$+fYyWY<IX50Ud>ay)8sKFf6l
z2<z@NWKY+}S}!_SplI`K7H%W=ljQsrIUORle~1;GVMUjGI;_EWlK*2C-`7lm-&mv0
zki3_)uxt-e$_>)~MA2_pTn_xiw!h3)+{32#fILpJVl1yh|3@w_7X8$Pu(W{{091i!
zS)qie_-#N${I>A8g0T34h=jrz6+eyv!$O#PfMGONTZ-8sK0c&_*_3|-M1sWur-1Rj
zJi0<mk_U1*VN0N9Yp|{WC%A-Y>O$15A?o%3WrvTV+gq^<6!OvC<Dq@PC-te&nQtmr
zex0`Q@3|HIE1GE)lCd5EMGjPK2@njx9|RxUM8n8-xY0Fn($+XYGLrIxR;588KE+_>
z=$58{IKZEgrEub2xXyvqo&3n*HJ#Vjc3fTC3Fcf(?>xV@<1D95`zf3yoY{7UuYmiX
zEpS;af8p>2X{mfx^KT;uasPywg>_9w*|Zf0)pPDvE_pd**$0a1tD&7YBAzh2zkJ>6
zo4)AloHnR!aEFG53)&=GhmoX-Vj7B^|47ne#Y-Ri0HhThxTNfmhg|0UN1s=TtIq}f
zY;mvWdh*&qo~<<DK4LHSjy-n6^BVUDz)}fSS1h8;IvUpURLv$ua8m+CWk{GLvKQtz
zm;k&iG=7*|&yd~y2smP@);h#)8+3bx+&;GiUf`k}XfqmZK$|pCfX0tNVI#Vsd70#Q
zf?VFWIz39P<TA^+z*b&j8~evJna12XH1^pcyN|AWoFDKT9FXQF<VK1c<mn&cl^6QC
z7QR;mlt6dY_XIf|x#s+GKWn|t8Y~bD?d~JDxA{^8*UP*s<w7@H>8dN;aOPbg6M~hO
z#p-m$>JHW93mWsBNSm^&$TnSC1(*F+&*#YX1B=^l<b92N&D>>&9m0sQ2^fQgdkF99
zE7v{l{ne%JeaFnloo4L!PHYR)G{h?F<K<Pcii#*@S-83^L{sh?kHOVa&-h}%BYwNP
zY@55Rz%72OTl{7(c|o9(8*{M8iVw-DD`J%8(Q<5)D~?rS17j%|#e<KsFf3tnh-^c6
z!X}>GskcOEwuGoQ1}X9a67oV-+n|_32M^a_wyyz4K?dr&0`<Ft5)XK5ZgW!{a8cat
zZFnqd#_PJ3U!~;zI=kV<?Dm0$JI&|<;rKc{LX}>$^Pl~R(iRbn)=BLk=_v1iXans|
zWR4hiMCRCNGbAzxyWj-rWC#AF1BY#yM>Z<ru;J?xb3mK59jDVfPJ%h&aojkU`{LTR
zbNo<lu&DJUUx$rD7DkwZm?MTAzC2#^92{lIOCQiKd{DLEImOC%l^cFk)tyt{Z&tiP
zp6?F&{6^t@{1Sogq(Y5uwxn9j=y>Y*m5n5dKhcOun#3+hVUjzBCux{%Hd&ouB>RRY
zJa2V6U}YPxu!WagR+xSADZGu`@1k+f_D%R|(B~qB8Yu!sHQZ1k0B{Y!WUyKj1pzLY
zRDmmCz(n475zwzhz7%G$@|Eo<5>AuXt2DM}fNeU#>dkB?xjsXVAM(ma^t@q^dq_?u
z$ql6E<{(ZRQ7ebx+8Ssg{|n^)nc(u2mF+mka-hpyWh*b*ZMo{zHO%MU4qH6`u=-r1
zXe5yEfLnOON3!GuSw$a^g)q_WKTLkt$@#c>{Kv!Wz(uy(GOo+&atHanN<K%)^*pa?
z{0P&m{csXtj)?Cr<vV?hVBLf(Wd9T74a=B=#I{~w+Xq}4$^Rbme34u}F}t6%`oai{
zNQDWdVbK&jNWQ(~eSuuCD)pn`(ANk2Z~KL1JjD#%?wZaBb!WV;GfLCwud4CX)q1A3
z*(bHxPwKIs+B0ry*SP7eV{dI@Q!80=AyeivSsshd855oFrPvXvuf^8<s91>Cr4X`X
z6g*Xn(4ae17_8VD62CbP>*?g%<O$nlh*1gK!ev{7WLpDd>!Y-J5xOkipEYa@Ox%VJ
zEKU#$&=v=(i-J_8!HUYDgxbJ_W?xymkF3X4zCUo*q0nXT$#cI>-uC;Hiqo^3uP)nZ
z#_<hn?xf{+(Bl0tBEWoLyrp|+YS-{B9VR&97GMdzxQ8AZ9a^YxVe{2+;-SS$y0O}K
z7^a?OJ%g)vnbYw>app&|^Q-`Q+Bk()x8raj6E^ztgM>M4POt4ek<s}l9?|gOEg-S^
zBj#@iK_}O?bHR=GdHDHRkP5JT^mt|iUqPDH_Dg2_4=i!ngStfzX%_xlyX+0^y3chb
zf9m%QYO!1J4XgJt3ce~1+wgllX-y`iNgxc|G!n$;5^by^Z7|IunU^_j?9GB>FOC0}
z9I(*i9^_G%*}O9@i*ER4n`2wa_fB$pcF6H7livmO7J0b>+Gs@D=q*B$)tX`zCR!|$
zE!HVkVX`1h5oyDt4K{n2d3zLMwAVSK?|T-z+pe>1Cs`eh?IQOl$qBSM$MYzxbc0?D
zEkSBD@i7MxAUUK07!=_647f&~N3E{UTUg0SmUn6F>MIk{uX}783hkokhbiz?a{1os
zj@UKC_a!8lNsdQRhzmt-hAhh}TDi^S)kot`nb;Qt?2)Tu?y}f-S={%d14e;I$>WkV
z95Ia7gFUAB63_+<t|o}96dd|#{GY_$wAejFZ2J{fWOQyIpSwA2oR18-o;E^tK=gKj
znRSRfk)7g+nVkE$oWQ;|xnu72!E<8^-)D&j#;f-DDffk`_XcQp`X=r6pLut{s>h>>
z-ivPdT-kk0weP6>z?bm{K8f1%c1Y)o<2K#SGP>BTN~SMhiUNCenTNK<58fO_8BSY3
z3pGMt9>NP7oPJk|(_Hw9Mp=P8z7PO&;t~p@<vValYh*%tgmQJbDgz2c2*w`(?Zz<8
z`Ut3bP$1+Rq7yboBou_nih>i$edP@kWF1cWJA9`+h~u{+7aodReOQzGdE(~p<~3cO
z-`=~Z$BY5*<p=nts|9<-iJ)d-ew%o95EhS!OG(9p8~8e&#W=Wbx3FTj6~tTIJ-D#F
zZ(j4jGMs%P0-7E_Q%J-rpv}6feVFgU?%>Qx?-T=0=h>AVmzQ(#`tr(-ORG9Atm-_o
zrsEU>5tx(F_Lp=J1D0R&lS`UTWj5i1zh|`n!gR~-)hxMRzvwB=l0(W>AILZTpzgS;
zdW-^Jwt9Rf_@Cjl;l?eCkYE}@Q9iPKgPT?)@<_Ddoe=Jk!5{#q4eq0$jZ+_u``+q!
zn7ok%6<=ZVPmG^;(KB;cqK)%&!_MEB5N%Maz#o9*2X}PI9Drz(EC^{<%OtCfHq%CF
zBUr+XCeQ|%kN;&$@D~K(i<O+=sZl#0R&)A*M;jay5=sNHGz{AC40FT^1pXlX!^hFJ
z<P&v6<o^qKz9F#Y3vAN`w&WyRc!K3!bZ#>F-cO#d(3mg!9Zp-YQ-N>i5hp{XE-(r?
zlq--8z2S-gZLrw-w3&T9#9lCty<ZsHG32;g@OzQ`McUvqCE8%H2czKJU&mo|LF6%!
z^I5Y05WQMrmDgCgX+k5p-9hdmZCp+bOSD1z7|{lYph!6|R@aX|Lt~Gbv2EDz-~F!j
zKeCArj?q2fuDZ`heZQyv0nbSf2QE7lyyg9vJC4Uad0ziof6Dtq$wvm$z8XsX=0@t5
zms37HuX**CpgZ5SD}9iy?q;(aSZX;_Y-RF72TiG8QZ;t@au-6vj#&8)%rD4!v?-9u
zH^<2~#>Q`mja?7K#pcOka^mB1;}b9-#CL6rXC7~cUKcOVk;$`Ua;|+t_>NF!hp2M{
z)th}}+kF&OIFQ&!)9j^e@lbbr>JRuN-y65!HTm+760^RaT5x7w<-pP=@xaDzu4Us`
zVzgFK*kHkxxVzK>@<4*)%M&q&hjUQSd)+4uK?qB6eCp0&JTB|D*!G9uOfGD+5~qMx
zw2vP1$T!Nvf3Uo>Z&_C#_M|TFy0W6{@~W<jt2@uH?mC;^iHLM6qvJG+AC8u@B2H%_
z5Oo}9nx(gEmK-!Jc~rObIpwmq6**sN8ZKz>7s8$uJm0bU{ciL5b2qek5DAau!w1C8
zHJm(BlJ;M;!3}i+a`!<R_lwo>13s}lw(<&F^7r_~mq457X7am(T%I$#ePczw5oqMu
z9f}QHj$8vk-$vj`T8h;&QLrM^0Jn&=0Rb@E1KQw|`RMfxiui$CpE9xPv#i0$S_J1u
z$l)FGJIQxdhtd%Gq>$zeR$kRec^deGAc^)%6b++wgZ56y39N4?w)-O6ah5kvm;TPy
zpJPo!6Yiz)FAD6l>-MJxQ3Zi5sJ^&$O?10KF%KlG5A+WN9HeT4gq5OWjt;YzjqKiG
z)_iqL&k(c_EC6!3z@=)@PdxxLQ7~qW#CEVaBF^XVJg;z7!2M<DjlHbaWY=VMxQ$$1
z5bQr5bUiWTW0X`oD}IXPz>vnsov+aNpRCx+<@>zJt>YA%`fs><9CXilrM?<C>s@Sy
zh~D~ZSn~<D2TnP>aN7ON%RV1n4?AX*ogno^(p{o?M018{>M7EEeku65AI9x@lNH><
z7PYYCVx|MiTnv>#x~gbZc}#p^bkru$CSI|HM-O>kY<zB9Y>qr8R~eI|jmy`{)~n?k
z6$-esx1!k!Y9+`vDC6_g@mWe)rXnFDUXc-naHPoyRb}%-P;}+7n(}Z(Sx|h1Uu>ms
zT%%7yr;l>Km+CHW-NS(spNpCEo?*>*w{AQ&rx>9Chi_X}>=ahQ7biAVC7xhjbun~p
zG+4zaSR#j2aw`sCMSb7WUDuZGy0K(u?~<N@B|Sq(<d$?>Mo)Sa4|^Pi!2fH{(9-UK
zr9J)2d#<nOzPhse%F6CboIG4SIFHOtIuQsbI%c$>5IT!2j;WUH)-64dxa9t%rBCaY
zy{5_hEV1&m;dWEZW90XW;Cqz9ukvdPr9wEr&P?dLOV$C2JRpoT%7Xre6m9$j%ro;m
z1V;=w57D^eCVS-Q&j_sU5?gcHY2^*iY*R!7c^@FxX9c&f1kZDP$210SU;v9GBeZO=
zQJhc^pq`D=W;*Y^K!@N%rcp*d)y!vs1}ORu@;GE>&7e&`s~>WFfF`_2-hYyRKhiK1
zHej_S(abwGPz87pvtVQh<VPR^%*ZXHFHs~Ex!X^$YG}}ZvaM%XT`$`wu!o7gVr54M
zoX#1;hj}Yj+65==O9<w^YzwYtzO(><S?;{m?mH7ZG{EleWsMhDPrt`Y<bjp&=XtS*
z-!6g%qpbp5s?zQ_ad8i}l#p*fdHg_L7)|Z$V~r-(Y_dCGwSUes?!$hUzXyGdd}$U;
z;us3T%>@$;n500~hxq1roLs@T7p(T3=h)OkEa_F2^1k1KALBP(*4GJ%yD9EL8uwx^
zd+#zka>MTXA$QC>!9I&PY@}#$wH>nJWGqn_Al+#SJaWPL>7%UZVYX^7n^?(IMXu@!
z>{Y~JVG3oDs9(c&umP=7FehG?tBTFl#H>@tuGGY>P|MO)37N`-Y?XYyTDD0WzeyXj
zK_8Q^jmcNX<t51SV&!=}ZBwm}mu*dm+nx|#EaN7iidaPrPF0H2cSh>EgVbG~vJThy
zc6Y^Y-{iZ)r#%uq@A<gpZ>w^@oL2hhV$9_1GOpccfjbUDHZ~=IJkrV@C<pxDARLI&
zF|=sUjm3K|FWPl=(XQ*5Xuvqog03NKk>?W-q87)u_KZw-^Zjx?*j;K~xYM`@1naSp
zXL--n6;klwnH%@)h{pl(T~?5=!6l|%-l<)>D{1N7smmTuUG{2H#wW=;{!HrZmEUU#
zMw9zv@<%W41~=8CFCI$6;08pqc)n+lQw3MKL>sBL;m$FUHj-i4)rb__IR3L?TtIeh
zS6TLXr?tKA*~XA+^4v{sPw}-<UZ>GJ=YwMS=C~h=E;2rhfXD#aqzcwaTtgFJvcx|T
zD^Q%om@F8f<73Qds-S0r3$6?C0k`Tthx;G|3tqpXM#tUV$k&Egtp*dP4V*zJnUam>
zBrEE%p#<tj%z=j>VSwakXu>n+*`B{x<<G3*IBUPi?ipq;k^Sf7@R#6q-3rkeY=NGJ
z-x@m>qr_PtT=$+>XC+oRV7|*m8vnC}y)wk^>1E9q*`6T}$a+vj&hfMgdwgQaC=>Xc
zY`8cgA$cSC9y!de{!HHJLG9{gtwz>nVz*hwJtvI+(CGHJ8QucXNMXc!N0el~*6Tie
zNh<l9$^9(3e^1U3uXkQybKhsvK4CMz^U1svQ$fm|B)^-2o}vk#5JSK}P41#1C+;@k
z%~l+JhYS<n1R5qHJbz=cndCgh9J%gw|CemzeQa_SQxrNWii4G<(aI8Rhl|%0;bfsm
z)z;AX4WW^_p~34y16RgGEsKg-5E-*5GIn`%{2ETDgiYFnLT&U`eH5@w6SGw=+Zrd|
z9HrU_%THYF=7i{N@|ePe*dn>C1XF#mb;s*!V)S*G^bSy0c_}MB0CklIj`h~Ha$nq(
z`{9d=T6##4{*Gb&=aUM56j^i~BXQVTicLLh?x2;oad$Rc2upSkA>b_DHLz&s0GP0_
zdw71g5nJT>IJda02ZP`LKpcet;9$TRDp)zP7zy6a!DTxKq|9wa=MB;Jb`|DPaTe$b
zm*YCFGX2U{{ql~KrTZr@|JRh|FHBnf-oy<*BvzhP?;DPJl>A?Zo-24F=Ycnb=LUla
zJGfq@B{d^qN3`GHL>r!bOSCc5*uRVp-&lR$rU|<TP&156zvjGlD5!!wI?4S}LTc`H
zocwyhHw1^l5W0aji{$plNCGr&L>P%SkOXkKOC<`n<lwfAObI8n#a!Yi@UXoXcx$Mx
z*X2HJ!4<rJCf_T=;n*80YJhM(Pqc6?4q{-+uyrC-L7rF+$_Nb^BbV0)Deh<Td%Mr>
z;q&Z)3+(Y;_OjLSEAl-<kkN4AMj$bRL@rw9VA#PG9KZ**vB=6s8(fjDymx6uanBwY
zWbIejzG1gR<n#rK5MJ-09gOj89+t%^CTaMO_lWVX4v;(QtB;87xxqSyS*MxZAxwCQ
z96lk>zj<U3XKal;SYb9vxZyAK^YIGghgZn!4|2g><Bma=^&?yL2U~m2ZS#;%HTiav
z@7;*!<nT3tGf=LvC>YC3pqoRKgLIB?CoOyM87BWB{!`%hiR-1QW-|OqLC^g%rsf&8
zsDl|wS;97_SRCt67nfKUpHz=WT;dVxn$U!bp!n^9@@#+kDzEsJo^h+Z;?@Sr@*)$q
z#LJ5mF{SG0QcYC3Dz;P}zaw6OCK8IDEtv6EDvAI^umojMtg0|dS%8u9K;`BD^%g(f
z7GJ|w-^2p%#I2r5Tip#s-YG5ai5<?GHhX2u1Z9V7;sO6@4~EWpMwRh_KJUv(JAR*2
zcVThokd(Y}AdwS~PvyHy#hq2s*(aPZwvi$0oIQX=gmvy70_KZA9|@<Y;3T3YyNtX*
z>K<6$&DX(6<{im}y{2m*y>oyit!_$O(UQ7s=d@+_OkeiYv=y&SUiW44)<2Ru29>aN
zqL=@c(d8!#oK)gk44C|F-hN!Q{02O-GffhprD!A3hPy~C#Qq#|{EmV?A*VZxY{PlT
zyW`gM`B#u@D|tLf-YDOH<8l`KHJGCerR#CLd_n}{w1IgC*SFZQ5hjbYL865`4f&Ep
z8yLN!AccJ?FnExW;neqf+)J3c@&AeZE)HXmL81*Bg%WL2OhVc)pAJCe(U1`VM^1wX
zYiQ}<a!I{JiKi&}JgLu9_$l%`NB-CO@uQF{P};%DiDe0xdqbv*;@+Q2$tX@ZZBTOf
zo5<}lIsPK90r?jq#uc`E&<WEGpU~KoW?Z94Zk1fG=l*NyE(0YU!G|m9{p9uwxuU+>
ze}y+~yN&EFa(WGt058(`)y!FEj<OEK@VBThK<LLUjNiq2$roYdQ(_NV*|xvf=1b$Y
z4LVnlV>3DKCyyt|>jRqboz?v!|1CfpY^HZbWD~0+oD7KqiCbfl?+sae`Yi!Ne4C^O
zVcA3}7by0<Qw|+3v(-EKjF7g}9c>wXvn+LwZu-H*d5<J7c_wk$bNW?Jd!{uxr<9M^
z?O>`cOulKHe5;$H$WLAtmQWMJhZm}4@tAtp0m8tw5TVVB#)(79^*F8zV|jdUiD65W
zZcDfp2ik23F>DS>+#Ha&$xlM^X1}CD5A_ZoT~$DGV^B()e_|`#Jx=l#ds(Z8;ec1_
zJps2q6}99wW!9(3TYs2V_V<kHi)*p6bSKBEo#MU(%&3AqVhJR!EVax7TlSjf>>U<?
zCj>~$kvgylSvbIJ$u2I?E#GNg)`N4uMRX0V>>OIvH4HGrirVBA4U?93+_Gf<txNtj
zed%+PmLJxv`8=`Wf@-%Z`Z2!$$>l36@&R-h@pf}j$co5do0o<b9Bww^_K6UL8XdI3
zp8>StvoScC!$j=(uq*EUACb%bR#tSGrJoy{HR!T~+&jqeZgP5se7@&{H~8`>UNrZL
z^#-4N6UWPVYZZAK0NU_yA?jd!Z-MyZ(9;aW9rrt858Oa4!K!aSgK~Xd@ch!^d)9<&
zoyS8H8AQ&Y1%z*ctpK2qnC$Vbyh<8Y(=hm%XtGW;aq>(-EE3njaaxLvQ63q@04r|5
zf&PYM5)-MzSV`q<9>xkDBxz^?xnClOABnwgWe=EH>ouo4Fq~(e@DbS^=fgrWzWx_O
ze%$fEr+wh`1FJ2T2(uMEt5EEAgHrz@v3&!qy_el#9CwHuJ|<tR^TiR|w)_yeB~nv_
z9juu*L-<CU*ylVCN>1=$RKa0zk(CZiXc3%lBZr5`^$qg=mhXg!fW8kSpI{pzlvX<8
zCWt*-;!q$@t+~a?%Dr>=WQ<*?jbyk^^6##BJ@hHt@i1GwhpB3?+#^JPXLRbrakF01
zt^I6D$zRi(FV5a|J-PX>h{BI2tbBs+lvCHRxKb8f%pyu0V=8=Q<#8$X9?_Y8igaJi
z(vYMT;rbO3+Lhs&6(Q=Cp_(;enoRDU)2$Cy=7(suO1{!yb#Z`Vd!Vuyvqm^81%Mp^
zPVG<~4k4~ZQ`KL~RX8_gyOXlhLAldGyWctKZjZ?idQbbeZ1vHk4QJ<7o7Qww=3WkH
zankNzq;rrB7Vi<TvUL7lGXM(B-_K7&!QqhnC{=_eNDI*A4$xA*B@1g7E!Z)FLm)Bi
zjfuS_?Sm`3EC5SdQIoQwZt~LB=}ULbT5|WSB~MIW`fBpZk0)+FWoR3e-%CNylH2<h
zw?D|c-!`!f2TgzlYth9<yTuH32p7H7cNb~Hcg<kArn6N5Z6Nr#A1BWv<P1Y6oM{^_
zu>9U}rQ~yf+{8bv-;adVn{XLN@u{1B9Axlcv_YOG(FQRAp$7d*J`yMXI1~wf*U0?`
zafqOMfR$f$+mG=9v&SbUzta{}y`T-`FPK3QUbvz?LK~@#BZ}#e)L}{(#!rE7NN<9F
z$E3hM_>n+*6K7D-fi|e_IBhVhjZ=NFEfHoST!^7&^1MXOKhT&%0y}77b=REt5|n7D
z;r4%80?_c`KhXj~F9n+nM3)^3EttDemqVkAI8Pp55PNEv?K;N`l}E|_eSXnLo+Y(e
zyi-7pFc}jB6NVS?k^(GdoFa^Z6Y{-1mMd@-?mVlx${Giq_LK9&<cMzRr#yv=925=L
zqOnz)y+yu&Jq^$pkZ1a1RVpADu~RO{0&r0t1mi;-n3cF>_WSIl!@VDkEqjh7wJ=qW
zqvCGAq<@Fc{Wv=7xA>B)@l7V}0n!{K!-J%M$QXayN%zVR*s2HF<W82@hAj&$q0+@r
z&7#&hY1WOEEo4#CCdAEhPgv-uT@jp=5tfu4mXsH!-w+{gM8G!2aMg}TMRBB}G+I#}
ztKui4gsHcN>NbVxw(@h14J9#HeTtR8hPo(2OPHZOK-cD@ZTHl5dKz|mCGYi2IXFS{
zkX!2WA+rz1ul!t}_v5r3Cl@qaUx|}X+Xmn#Shbthh-x)h0sR_|xW#yWJFUEvR^Lsl
z@8pUc=IU;#@0Vmc?npq&hYAWhIJVbAtiwj^E-M3PCM~a@wybH!(vCSx_Rd*+|IEca
z;7nZqb8^)c^<GQF!{qs@)%_dtxxiBqf27nFtCz_VAaLCb7EBEN^Y_}eRglvLvaEQZ
zr+{+^G3*Jg%=H^`ewD@?=w~Ho*_LyxwAcOs**!@1uad_x3cJMDFX0lGk$IFhTon*_
zw2^4TQ!KHtfnYE?eT^%?=LWf=%l=R=>*{AkIDnd*9yGeWYxMtHh(c>0nIulA76S-m
zG--+x7cOW6X+iY7apoZ4XlX!8eH!Z3&>-?RC1(gqa3kUjFKIY1!3rVKMjLefv62eE
zrni+}zNihn&hc@EvCp8;GP9Da&Rt}G4|%*#j=u{?aZz@|kp$jOhQCe>oA{+5a!BK)
zsy+&Yn)mz-w&!=Y`viM@h&_D+o;~lMNOnWqO(&Vqd4kG+eFS{GvHaJE4E+^NcmwvO
z3#{=nYwvfs+v@a;#o?`Cr=vq2r%j<IR6$nU&Ui?%EjB^N3CkNG(1A02;V}3kQV2=|
zemg1I=HUpPvB*9O`s$+3BR{Z3IFtA`CObGT@sQWTWB$1p{YuT=%@p$>MgN;LZ_(s0
z>6T-Z_#uTqe$sQ-S1kK!w&($7=w#Yf7P*ZXDwuL33tP@)8RHVNCMa?o<vEVBY}fc)
zANhtLIOyOaRTV|TI~TVtE^Y^;Hbp!J@3v#EE?AR?y^Y~GeMr3}TC)wCfMV2T3CXCk
z8lv<KVY-GOU4x&l(MN}UJ?&me``nZ7bW6I!*>Ibq;ZC>I`~7cyDs2AC$s3PND>}Zg
z?%JwOD__~OkMi!I{DU9{FJ>SoEZ@(aK9J@xbqA_QyL`EjfSvR`JnTroQf#z?*rer+
zQ<gVPU*0%#Y3sZtJ#!Y{Hhb}-Q<uJyn*MoG;VE5*G45XSf0jHx<ilY;S4AlX%rT=`
zA)qP+bFeT8%;ER`k!W)>ZID2_pfB+oxng<EW0>(g&We6#Ri{z)vPWp_TjcR0g<+Bu
zlQ3MDL-BwWO-P?C5^W?g4ZsDLMA-PjwL4tOsD+SS;5Lu;yz?)h%>e5hWQEt9TgmBO
zlk1yCpWm!eL%j5ggCj;<Rim|V67qEd;<hS@HllzI$s764FbE@(XHZ203Dg(E5Azv7
z*x{NXFGyhYMkGW=ifL1RtL8i0aKE<@S}`Gm<b8&4a~b=vm9-h!mdkD}<a8JLyhl#I
zP(Z(pHW*&u^9v$#P_RuDELscy)zV;NJwu)!4zv4yW4n*C`;L!&bHMd8b2!>8gEpE;
zRL3j9FD0Iu3L%aw788!q1WcaqzQQ`LvF?8Rdrgin7#-g4b@`>w`^pghMi9#Vxm?bV
zz8e8d+?WmobHE?Y9F%GzZ3f}*=FyCw>cuyo@YdF43jg#vt9+9!ev;|_#gbnhxA;fr
z{A;e|<kU_<&k)*E-fvRm*Q7f|Dd$O#)%BkX;s3r8c*kkG!Xs?W>+F^XJ=0$a%6>Tj
zCZH+pY+^G*2$Jt$F`HRjfg`r<B-i+=OM;apQOXjS4`OB8;$pYP$83&^-yEjd7^KY)
z*5rq(H-xD+MJV~E>?rv*m{4O>rIG4Vcp5^r6@l7v@i56MY`XF?V8#wB>{?y*ZO{^&
zwcSp--EN6@`%Zo|Z1$^hOFq!7IW}p-$pz)uJ0;}bP1y%2^G-^~;>A6D;Wa<`2c^+|
z4&-$!z$GMX7-a>Q_ymLnT_8B1Tc$5>p0TWP*7BygOIzkG*)@CdJ<}IIo3!NJ#Jpb;
zYx|V9Q5Y(m*U0ND^7va!45Y2iwr!f?acjuY|6jC0c>`epM?*VbBKJRN%)7(v;0f0C
z8|%Rh)yxhF<3AV0qCRmuGg;$k80}Q4*}_{l|4AF%t;GzD2RCf-6psz%XYqoQ?BQY7
zVP;!zxHVx0&g}l0#p4GH40U`Q4a0Wg$PlJP^-^^q7D%}LL)VfzPI#0)t_Nd<;doPD
ztho=pU3?lJ84#DMq0NdTeem)m+K6{<vGZuAAegD)6FUKM0-~0Q)eN!RD;{;^c02jM
z&CO#0*ZGgbuY<~JGPDQGnewn>og@sa2YQoHdO;%ZCB<Llc=#N9;wSd+-|Y`ya{uy1
z$e)&kE8PCfhmOpoL=Gt~3Bu7SQibW)OHMzK6U6I#jF7z9t|9yT&CZ96u15xaP7dK%
zZ!u>^_0J)0a^$DqNr!fgat5@4SuTKvf`y?d6GVA_SAyVyB?`O;ZK9aJV88f_Rej9n
zKgXuN$QFFXaxXZRlVcCL{+q_VYIOW~DBw@NBrI-#0#6xTzrGRu9mOKgyiag$zQHn%
zd2Bx$Ti2`Y?31;gjBWfSr2K2o^>2<@{1BVCi)reatdvD=85_UN3r;r8j$p;6(D;0u
zyNcr&{I#3Bb({S(TxY{9cer@ycZ_OBglsE3J5h?l2=#V&=lGbat~f|vitpfSD8rua
z@X3u~Q<{U*TKtlmyb~cLG&w0daD0Qa>VS*#9xuaV;kUdTx8R*yihiA0^VhP@>lu5^
z>u#fU2Y4n1aFaDeISdPl$3kLp4}jE;$DAHBo4KrW=JK{#%Nu7eub;cLaqg0i*^6$U
zzUa}Eg|8>3AJrFMQuhdv_mdw2&b!>tiv|T~BTd*tL6Wvoz-jWov~jZt=y;BYTM-*M
zOyqH$3r+0(e)iHe_WX7B+;#TWAXcb3pQeC5UZp@Ris$MUKW$1YSYQ<3(wP`-c;Uf?
zHNII2k`%^Bk)7krP`nanTsJ)0n8&o4*`{9CdUCykn_`^5H3nn(j<|Xg*BQ^p1fz}*
z^T0R2g=u_{Pe-F>5F3AR#DN>Qqa%@%^g3zUhp&pafv>=ZRuRt~qWnJ1oA(y-=p&EI
z<a(0EeLFnnbt7wsL9UmjU-GFUkG<T@>GX}@f6<D@A{>372WZ2$8h|!ro<c*Z=Dx{R
z(-f0%tA%EoNPmuEelf-$C#R22dVPC2^w0jdizdy001r4`72XpBVT4Dl!h%f!w-3ZI
zm!Cux?g1;?Yhk;F$K7joe97eY*^vJk6Sjrm=HkP5IAVVlXd~?nw(Yf&oEG9#lJv-D
z4P{m{7>?TwVi7ooy93_9_-hpUmC^CxlWhA(Y|-m%;YTd{Z-+9gQ#Va`ir8C2j^7OX
zo#*T65Ca`9^|D{C!i(to5jp*v9PT8x=^Wd3&86HJ+)ak(DCHHJ_9jhxnRNfU5!iTi
z%(`dT%zZ3w2TLeo%A#?KLPyya*Z2)yvJF1UEfeG$?B$zqhNN@+Ca?HS{s~({6@@Tm
zgEKtrD8Zs_k(xrjYfgt5??N025s_LMnN|hhNoaCqaAG+Q0}oAY3QcPZOY06!>haa~
zc&m2$YW8|3-Z?>cyX%yPgBKo>XMPMhZf4~f%w4bLCvb5`&nozOaC`&KBo=!<{3w(?
z!qVLqHfw43tYsY{ZEEK(tDU!`Y3`yuGZ#KMdC{we6<;XVpGc@548M*19wCpH$>E6A
z^(q&aCCMKO5^?}>JX5j}*dzfdNfT)!j3tY`cqp_pY>2(&bBV7A<euM;huGH^_8U1~
zAg_L&e?y^)r(q;QXi$#Q2L2j}Hc+DVVkZ<eF}j3EJbcIk^V@{0WDiN;0W{amn2vU>
zCb!$j`xSEdY#`t=mg(@nI&Pm_i6+nn4bVOX&%w-m&y!@9!8iaalM{X3Cwqg=55!^6
z-@?g*U252IqZL<patNcefdnE12xcy|L)E>(m+6f8ewe*=hCOhKm7nEK>SdRF09!z$
zze>riOWd&Ysmc4aNy|5iSP*Tdh_W6qm8{cfcrIDzk}#XB3y4<W5D}so;%Oq&u?Ljs
zR?aS@bNmsx9M(DDO@PK2DGZ6G3JM<NJs%Hf(Dw^$x0Q99$K8YJ9jp82rhwBHC;^hh
z%n4%^kW50l;Cm6!?=tZ)f?pqkgtS}%As6ow>L|omJV$ya(Ft!L-_3B6f<LveJCC!?
zAF-7mviv_m1DAGke2T`rN#nn<xSkQ*hAeQ|Ian+c1|YlBxT9qE2Dv<ok$5Ak8e}!Y
zE_;P>5A``88jSvwru<Cv{-D|4(xlf(|Hws_`!rjAAG>uAOKN87TBfXF^2)IZ<ztk^
zV-?%S%C?S=+v*sz*)4X9Pi#S8TmjB7#u;9iXb6{Y2~%we)oy`o0NNnaLw^>Bv|>Ot
zGO0W^tulU6W&D)NP(vZMWQE}zu9Px<uyJB}kf8!2-)_l!os#dKkaYKW{XOFo|K&dI
zxsb*0#b<mwtG;i3D||P$WxtEDvIklM4phak!+tB9wW592iq=`n>t`>op1Ztq-qQNH
zi#ld5ymQ9V=aZJct<3yZUUDwFYbfMy!5>Zicgf|X#dDaSsKKj7h&fP!Mub|p@$t9h
zi(v)c%|M1GMjI0jxV0Co6O3l$btq@?$MC^^&1>XwgFN6X86qzX*;}xHvkw9^`j09B
zm4=w4iQ@xMpHPs9nLY9oK1Pb_7T*f}2f*&4fIknnfi;?0-VOIk^4ur*bM>~@?;_ux
z0W#y>DGtEF(`l0PJP_S_3%d1?+W04!x$5fp+c5j|B0Ka4J9NVL(_!^rd{M>}WJgAk
zIpWBGm>}^fYuuGbZ`FZ=<bic?-(6>~{L1e9fffG5GET9@XMKvuqm}#~Cx>^fK7Uy8
z9S|$v6++(@wp3W`W|DOdnU;`s8JQQ7WeL$LqJ>1WOtch+F)O!?%@i9-uz|x{gfE$B
zF{g^-=&_aoG;_H%ZirleCpV~YkC@r*R<@5E9whfS39hKni=uZ<OeVR|Y_^7&Aa?Ve
zJD^SA4GSE7kvKl+Iwf2q#SN18Q^Fw6RH5A^nDIuzMTMJ`1v)+m7Il>({vwB0Z?L9s
zS^g2W^>0t~tXs(WQL=lJ?|gJRgKXSvZ!zL8?qDX&j4<3B-=s0G5bGIY^=J<a+3gk_
zU#9Wz4P!4?)Nhn@mL{F0wBIQCYl?s2ivRu}$8LRtEq;tme}GNAkLmAZ$~|Kz-7zL@
zH%snfhGwR%W9rH=>eBJ*Vl;>llw1-vdM0f2M^dNS5~<x5soNftQWKL@8)vAA$3a6n
zj0_dUXbU3HBTCs8o3s^eqJ+c(c}k&TVoChulISV5!IQcJr`#Sm<xU)5hZ7*NFU~#r
zKF`UIc;50<(1KTD(mzpdIA$pPGqv*E^yXeHMZk>1+S`cDS=%%_y?)M`n%S!==d7%p
zx2&26oW(n5FS&Qpk|#7PUXf*g5nFarw%ek7n4(@Ir!O!2UG9&Ac^*41aJv%E>x9-6
zYc<6R1Ci7pO8p20xS&zh@Qx=Jqj6+3Z%%u{DJ%}CNUlT#B|tpo22~22kr1Hyc#M*R
zmoWhDe9*z9W$jArjxm`+&DeC#P1cBDj%FHjddU7OGrRLTs~cwP`<#l*uC0CHFOvHw
z{my?5!nlPi1~o2HvJ|r5w#~~GF+Ib^YB*Nl3gaj3-~@ydNHzp^7liB^Fk-UCOKjhu
z^K(MX-`sVA;DbAHj0Hj;^fU`JXg=#^qOg7ezBpV}@E@`7PP6BJW=+4b^?$OAlg@=h
z0d2fh@AD=<pGU#pD6~8KCsF?_GJ+?|1X?c88nUF5F`a~T;^)1M0+DAKasGf#(j%}~
z<kcL}T!+`g6@T=6`Y_aFW-kr12Z!0+H11J89*3b1oEZUnBN349aWK*0{Dnx%Fd7ED
z;bY-%-I$}jj-L#;9v%q#jAZ9XYvjKv-fD4k!wnf-i4m^iuoDC*SjGL$kn`t8cJNPD
z^)uUg#xCCy(oW9zayOIxH=H)GARrTQ6Zrjc-1&j=e_I&tXpk%JHL~3T+tKH?o5nwB
z8u#`9X2rwL@LDJqV}IAhRV06q{A&t+YRGpFntPwH8Sk)3Z?i>TdFKA)x&D~P=5IWA
zeC<(u#D4Q@Z0$pA(H?ed154b_G+UigO8pay17w?U?zh5F8j`dlJQc?nmua*m8aW2V
zx2e=SR5$}qvrVnto}}5HtSZzi3$Tq|mQoxyxgv@m^;Hv@R2z|47jCEz(PP7KJ@#m0
zT4{o+dV;>ibNXHW3;!Lo@{PFl-=vnGzP0k|{Q3blH?tm?y|!k~nrhHyE@-o~e%@j(
zMo(LEZ_4tgHJR_q3x1L}URUiW_&dGdyW#osji^hc!h#OSvSMMtbzbXW9NO5{{xD$>
z)?i%LaC3q-;17<P@e+@Ml1_v25SI<$6Qpp#y}hE;gD@2)Mgz1UK^rM&0MJ(FK*WGE
zz+}RV8d47M&`w;AGv?0$r_ZhI&VE*7WH~q7iY(r(1Ch^KJ>I|J{QH0><^sg*5ic^;
zV#1vsw^HdfqfHbj?%l-Di!-b-r_iayK);cd^|MXa+4?K2VsOmuCjTS6d7=@wk$@_o
zjVN94!+fL!FW$)IX^7o!m--xzqU-t_+x#P2^8?#>-n&llhbr(Yh5XEaDB`4wdMDDg
zc{H$$jH@h^Zl!f($s$u0S=SL|5&p@MK?D}9ApxWUz@;@{l$BP2X(Dr=yg^k%gC)qs
zXH>`kFvwo*Wsmf;d(2qoh+%BBfU&e1h6af?&;YQH4nBiOOMjSk81IRLk=O|%`+R8J
zYlC(VT^aw_fa81QdW=*<(#92tmc2^46i-)pp9-B&oDzu^*{{T&KF!+yW;-shJPK$g
z>=1T*%{<{NZsEZ!5dWJ(<s+-})qcmb!|X>R`$%994YK{jiu$o)!Tt%0-QfYZ?}Y%Y
zSU{G6$qP(F^pVT&CZBJ~<&eewL1Lw+*@iQ4FuIqU6bDK1AgP`t?MtM6lT>d}{HuL2
zFP;f~;F~e!kFq&6Ouq@zwq3$zZ{-f8UcSkNIK&}ZQJ|LXNR$;P%1Y%4MUdtcnj)>H
zBu!nKrYcHQ6e`u*aAZ(aT2XvzG58}-E>)zIC8U%CvedHJiDj`<D}z%@{Zgv}rgr#D
z!LqB{Tqi#0JM+b;MeoKhJHi&NYg?Gvx*)S@-rBmkt83=0sF}O0e$L|7nG1JLUvgL4
ziYK*eUzO#37PsS%_{M9Y2Zn=Q!%(3q<X6%T^4tkR6ZXv_?THlz5pdu@L8lovGrS!z
z3*z3x?I&Ve&`q$q!4QQjUi`R3A<zjK9YPJdZy{VF18r>4M&gW(Hljxc)oP5z$Vab5
zPL9DJ{yK_^e(o;04VU{6%e?NrodR2jW1k|AH?O<@GUz|RvlK)dcm-08!<cweiU~bj
zZ4uV~0J`E%*Bo*FE=J8RlO67hJ$-E3dA9Bp%RI^F>RPXkdHk}=Cw)<;`FWtYJHW<@
zYcZa$r^5C&F>r}{s(A20$h}VN3B)~o`FAY$xO4STz+K$e=ZkfX(4)B5t`D@CL;cG{
zGOQIiRdUFjL)Kg(AZG+3k<36QnKDUOCs3w^GOWD*k!S<C0_hWu0}30W2|o?8H?U%R
zfYatNa(bIQzK4GTGetN9>%V9-h|`M^t{sPnof7Q7#Bc<z=~HaiN%n9bd#OM8Z&D11
zwpV`hx)x8@uxAUYI8y6SSl2Glgm<yC;Sz7Ht~a?gkmEtY;Z+l-%_&YB{+C772l;N;
z>~_V%el_u}`HveJgn^O)C~tO8TG%^7PCt_OMLy<@yC+78+y=@1Z=>4}<nX4&1GNym
zlC{^zl=pe<q=*+N>}`rY0?Ptl{-`}khO;#BJT2&@>AzWIhkkbNe351EWz#E}W($kW
zb4)4?(3N5RqFzy#s%S`2G^kZ|@tP{QQR6kmV2)l}qSlwhB^8HEDhZxkgah#8DLWDn
zpeB|nCYP(GRw{3)P~K9eno$uywLE@Gjci(LbZS?is@*T4+h21aAnjiNN%ym5**lkJ
z^(@Zn1bODAH_cgvBBy@VlIH0PI;Staeagy*lGZ#g&wL|x{b%vTzlJrPjrliMheE#~
z%{fRzxHF-lJ{U*+(SqoT$9f8!>tx}_%fMjEosN74*aV<hlulCvcqq3Bo>r^3)q<k}
zMN2%Y1vrb58=(0J@FYaa(<CII8sRMk+&;M?gu;mpSlD^<*e?gX-=#6ThS|10w&IFk
z0fjW1;-4UoS8jOyFdU2p?a~@+xH*T1ry;jN5sVra-%98Wq{|+38E00bE$efc9JmMh
zI@@}ZWgKT~|6&`?vMSi3|7K5|_WEo%_B{V4*b{<OhdabLdBiXaU!1u0m#2b56pjn%
zej_XTlWjikST}_6W}GbI`3*%6aRU)rLlf!792!_oR#Y;eO^!ghf;CS(jyL}w;3MWd
zGU9PPnezo}zLoMUl#PI66_Bk>8lerQ@!-vI`qjt|53(n0v~f60o<IB-ZTQk)tg4E%
zfHpkK#X<JtPm$YK<nYWeYd^sn|7LgfvL^?fKQ%|5<Ea|DWXRLvs29gB_?yKg-B>Y)
z?F&Ef`9ju%ohtnk8p!^jb;8SL(9hwdc=Fc>ZSW&JdXWoQ`ChJnn;6D>OZ)7)5v&CE
zp3xo?-RMb}c|PKV&}R}HE(tEjY21evnDN+wtE?Ml@Ifc^i{B+z9HjRL1!Hg4MT$U1
z2k#GB*n<S^asDt(KS}EM&-mZ}39ERNt?XfmrFNQ1Z)JUGLQ`~1V`OY&n7lq%UF#z&
z3Xm5@%C}>>8=FZ&wcITV#3XIU`QGxR?TXYQ#iSzT)ME85rJCs_s%b^4DaG1pmC4ha
zlBcxkliGELZf)wG_^G=?;UQX`bzo)ozU4W)mt^-W%xs^R-ZFbt^DH!T7Pj8HbkD7;
z?wz>m5#5?+RqI|;Z~7#*?3cJZDEtYEd6VM4CIy%ys<dbVM$j;HVT5w1kf045(`aKO
z0mFH>E_{s9h+#%&@i=u?5=7cq5Y2u0j3gEV@FM}a`pMbC_Y;Uiqv9zbBTdP8T&%({
zeu^lgk&9Viy(V_jH2z}~){(KsK9+ZtExh2r(Hd4~Q9MH4FAn&8YY7#Hx*-#(Of<=8
z1Z_~+BTC5l>`-qoUGb$Wt~7wx<9e1RVAJQ`OKj^OY}HR}`7dKK{$QK_VD0DG-RGTN
z?hnJ(!W&$7LQw?HU^ECx7k3rGR0eGd;{Z@cdP#Mf{PBm!NJjDBcJ(*Ba26_D<X=))
zFE7ybG(3@dXHg%@70_ml7;SO|0xQ-Fw4Q{GWZg)Xjbz$PMqm?}wg{HZR$;S+fDKk!
zV-fJRFwWtfhok`$07&chr^W95VO;d=pb5sD35Yg7P{0*T`0%9*Vv;sWn_*<RScEX{
zBH8`HHR5si_p;JoInQ_Xvj<IfuMPzL%q`*2yK(<QCk-hsH~@D@5EjhicZ8q%LmGDv
z7KU6O(?a78n(Pl*#(ho`aD)+zd3YzSn7!i@K{(87>`8$gBld!Y{R<0*uCs1|!A0@D
zdCX7baG4viC!h)<%Xq65;u|I`*q0`__}F88=qQZ4pRiEH{(E8kSqoHtbUyfTl;Wvj
zcuCw}i>zUg^JkiwruE{6b=CgC&)JIIY-%e@sA17{_VMkm$}W3(Gm9y6(pGtBN`sU;
zBGD>UZi`kIhUhusb%p#K6hnbBsZf=+T{CH$W=f%U>Nfo}pfGuAQQFjsi4!YRldC2s
zRi{m?Q%$Lkn^wg#*4>$&b;ru=1Ix4bF3jqlm(ekMP3!EHEiettTHZ5j#cj8&IGDET
z!KC!3lXDNLw;qY9xe~LFFjEk5$Q<>xS#y<Fkhm^!1{1D+Xk3t_`Ha3eyd}}bnTEg_
zCmP1m31_mn@|=xlZazHD*f@iLAm%)LIe`dK3=m5ish1xkI_F^q!=81Xq?v^=*No%-
zvO2tNb;CT_j*Bew5?grQFV7TIIvjs5`93%7{kb{lEJYc49)-5aM6(I463Nnrz0!z~
z;dITPuEEyLM^Mg^+Yxd&c!O>KldU*5X7P7+%YS5Re{^W+9e?Mb+w+#tuXyE(m5Y3`
zjnLj5f|(z37F>DIc@b*xheXp8pA!Jr^Icc@&|A$phuclA7#}!7!B==h(i7It_s$Ud
zmXLsSYo&!U1s+2HP96b)#k7HjHqy`*G8J%`3&>h12-_?KY=VSfwxYbbh0i|NXcLR5
ze!^=1k%|3Vq|GB_Z=(%wrirvce-pL<DcXQG{Ju8sJO*#c?J&-6xXw2G$nt-OpN-va
z8ZW+Y@^jovA^v6gI3Pj|)>HXGvqpREDusVXj{nA4DKI5i#@szT{w1s35oB4MHc)^-
z8!z5n72JpUPZ^7H2=|dkh*e)_yGf*t1+=lhO!ih9PlOi57_va%na~UI%{2ZS9P2X3
zu-yT}Lb#FrHpnjBa2z)C^d6}`88H%zH71fcj$j)$%mp5l*%6n?_gMe9hrVa!ue160
zvRm$RU-DvP=3(y@&y2bCHm0s%@&Xo}Hz6|HEe4LBP2O65#tBxF;^0(%(y)H3GI^^y
zb*pyLR^8-+#K{FoQ?{i{*`5YW+L1bO`=m)llctt{QJPx|S!U+l8SCy&&$(-5?rn>6
zcFoV~nv>B!Yjw+<RjqSYbj)7XGjqwFTbAE3ZS}p=G9FIJJruv;o0!Ve@jI_a{cAYn
zkbsLR3g^9GK`S4<LKl*IH2nCiFZVd|sKO%*4>cpS;RzcuH}QtVT~he{rO_QMB7-A%
zWG9ZOo{1V7E_4OIuMdUO037?Ghl)H+*!f~|`GXu^vN-HBvAh#3^P=POE56yo-dlSk
z_LJXJX3q~y-p9pD8LOnQGRn%;TWO*W6%=yINa_oseym@D@`F7a5N*948MZ4u&Q^Xq
zZqZTaWk1@b|LRyX2v>E`3l#DR;gkf~fOI4d6e?_jM)WgdJwT|LAJYrkgbY*YWeWNc
z%?cCiI>TyCvfbD1pA+mqvif5J4Bbrl>!|M*p??txTqv{9YJt{TdA61zpqMe`&`>@N
zt``Qkl6fmx3y3%b8#W_AjnD?s02w3tFi<spPg)(1nAx*~+$-=fjem>VTm5l;+q|1L
z+Taa@EXHYLCP%EW`;|O!{@kt`EbAARd3=23@VG9s^OJ(p8$+Q#nzaL5j7B0V_2sy+
zT{<-g6SY_-`Z`*xm?<){y9dU<U>)}<jXzFqMsYk*z!;JR6%Gz*b+}3sPKsh0HqkZS
z7;`|R&3nR_AISb9O)w+WfHtNv{N9RGjo9B-``_^iL+o|@Is>SJ9L^aX`g?I)7p@l3
zR)?+~3O0-HP#^q|qH&%ZxAD$Hi|0p0{EFj`uCeB0IOEl?@()$V1!d3K_|9M0x`)})
z-E2AzO4-CzxlEbMRQXJi@21%7pRgIe1e_rmZ`iCz+M-L@rcW&}q=7R<6K^S<cxy?@
z%;J>UMagpufs{F00k$sV9xw;AS(SI^lDvKMvU_H)>zKW^WkE*A{8eppm$c1U+%bE}
z&N(ag&0c-SE$NRKGG13~_)=YcLbZP&{2`0yvlibY6mgDXjhs2+%wxZJppqHypa|46
zoGN0HhHTA78$a<~zzsv<40I|41dUmu4cus)Gcf#1v_TXSeaG0I4%+Y`VGJR=T@t*1
zBj+bA_U!{~?XUJ(7d_Lj`)BvN<X;NvviLn=bvtZy|Jm$!g%5$E<&uOwQrPP$UJB^{
zK*qpc8|)iE6CG{?L>q{<9*+;OvftS9&+Ha{?Xvu*2^oLb6<_nbgTh{>U~HHEUWmNP
zzg!e{K^xcwU@S5R3=aq5=mh-nP}lWyY1QuoVs~9+4JTR8<*`p$?BBOSy~WK13RMdA
zPc`??Co6KTg;uUIBYFTU$-0J&88o<#hCm)L1lTh2C_ULI(gtB_x@Z~@X~VZ~`<*sB
zeP-g<*}>ueq79fMp1sARjYJz{(MY$jp4;&!@_daZboH`~U)h?!9Ci$jZx}}C9{=<W
zuaEl`7r9u6>k=6n?runpV2QwYiyo$^bL8@YiQRsI-HG@%fct{oCop<*+MsX+ZQOCM
zv<_j~%CR4_;>7T7pwMr-G43wZIm8arxMSpqkq2zjgCfdlGwgs>N1)AFixbuqe`R5x
z3ha$x_Q{a@WsCbjpT7l%wIIUaWMbsTL-@mqJtcFb)eI*D4iw|pL$B5CBA;AxewW7G
zdkyQn?W;}%bX=7`C`^2nRve{S@AW1={6$3jQyzu4kI8ChE9=>U8ou~So<CNV@1ot{
zZ`c@-xH%ztn<{yG(xi$>x71FVSvPTZb>f^7?c74eyp4+4dCBuc+T>;3o1J}cM&3ct
zW@-NZ`FXqM<#f-_>R6i9wRCO!!WAv^7dOsd)H;7j=lqpD^VZxoE903dxrY-AzLHh`
z9@2T)?>>|7OXT@6>4!w4p%~D;Y1j?3C}eaLhA=?@wFI|wBto7D+&n`wBZx~k53ycH
z=_zh7#Fh+hvV@6-dr7#R<e$?5kt!GwYVfzmosW0NVJbOpbb{63_Uo(^jrrrwIcL4r
zo^xG!(tYD)|8}GIBZAZG#tGjHcwgl6+n}Y&JcNza(nJ7Px5b$=I7-CH6TZbS5aZvk
zkqe?k-gj)tS8T~QY~>G*^+u0_6!-%9y+;8*5>_tayK&u&-}A5u1g8{YfwdI}kAgXH
zh+~f#N?W_vd)b|5S?ejb6SNVWKjt3(@EiR8J&DZIgrON`<7|QE2sG0|vx#>$7jUD%
z;6h99a_c|_2^j*V8!6LBIVgKXsQ}tq2ALrVEU@x6D5`d3eYj|l;)h%$hffE`975k_
zfS)GxJh>hv&oj8!NOMs9CLPSH6zl*Q#zltO#`QXR{Y3%qkk<hVTl1?!#%Y#$)~(Uv
zg{_Fsm|zHyVF@_KTE+c|xUE20#=C^{J7hozJ<ER_d;A)^uOILK*e9%ZAJO<<$qB|h
z@o$aH9DikpA=89@ZWkDXQz)@Nv8Q(o{JW0}+*N=T2pBOE^#Ddg6UgE~!nh%`>mYU%
z3+!7X`^w5bwv0Vy4H<yt&l+Yl^W8~;MPcQN5k@d@1Ja=NEEvO$!)WF4PL$E&4wuu9
z<bcy%5B87ixZ==o&85BH=K*Wzt7Q0zW?iC1*Km5Valz@q$v>WpeCPY1XFg-w?;4Z8
zk1cOvleVz9l`MR@V;q+36i%GgHF-wYq?sMbvl=zC$`x}8<qJ2f7H-kb-j+D4n61yb
zHz)UCX8ys|8}3}P;lRT6dluyFT$IzbET?l>M%&VrEsK{mE?nHWaB0)R<*f@>>|D6=
z!P)80PtJK$wdK?BvLF50PP*T5&Ee6$@GmL$Z%XLr4N&~GaDPG?=FSh1<#i5>1@6NZ
z4=jlm?^~c2ItjQ7ipOihfXt(cM3qt6fHNuZafz#R;i-Z3CIntXK3U1r*?{vHqc?iK
zW9C}~3jSi*zqsc7>AUv0+sfa(HeC$vum(KBS1`MMXMx3rTbL2k%vk;;^<KFB3~CHc
zAHbD|*MkcJBbEVwk=N@(tP3Z9e8!f;Lh=R6JciW5{vdh0$R{QJei2}#7Y7{p!Jq(Y
z?orxc1_?bx93t;<S{VQS5PR@4+j)`gx`zHej<7~xL`xn8$z&3y3C3Hj##;q$pF!_m
zG7a2Hra5GsPbNejXw&H4K#T@q@VEv+$wVuxCNy)FS!uqRn{M!}IBi53I^-%jecETE
z4R&iiZFT${hcQ}3gRH0>bA6wO>reqF1yKMtbmKV3pLw)#+cm)N4y*rWYfpLBTfFWh
z_s31aA6Vl~kPNdr=p~>XEH=tShmIv=$Q+7+ysrkitMHKlOr(x^#Da4r?0zE0Yg|wU
z$MAReLhHtCaT=zv*qjPg0J6=V8)NRWqSm*2g&c9{9lVGFk4|H){q{VEx7wS`D2MD$
zSs9#4$1LnKqumdjHW0q~8a?db=H@`#%&$hktOo{W`SGmyL*R;n*9lu0xMS-z8vB58
z?5+X3=E3n@Myy=*e1}4RC!LYM)#%;f-J1$|2FrnaY061q>akws%ip`UKfzX1vy}DB
zu*Ea2Dm<+%VPdyo`tH;@yHggnB`&O1E+~-C*{q*anmD_ftzUn4-iEugHr|!K>GtKD
zZ(Fiy-{K9smgetVncKZ;UFWLwwiPRzm#%1Dw5)OdvWEFf+m|lAcfs;Ur>8$<$bK$*
z!z%&X-*>D0(yr-8=esWXy+rZflkPIXe}xMR!%v=rH{c4x+Ls_4P=WO{+*t^z!0KrN
z0g&fNw86v+Hayu5qs3i>FW(S3lZJCPu?av74k1=#X#5x)UJLedMbh=7KDTH4Smkdl
z{|EcbZ@sg>_e=lIeZ_Y^o6d%HSi&EpkasEY7YY|w^77owf{u`F5dvgrbi~!eeD4{;
z7jB~7y%?tx#=mxv^?b{=yw9@VXZc@or_8?IagSSF-{L!Ru}Pg9QpDGQySS8C<9^J~
zPP7V`xb(JSIFBY=z&3Qt_-7Ca2l(l6F#fz?_W6k-ukg$Yf3~S)0(0;u#2-(ZBf>7L
zDFg?NdAijy`NnY4*}*B7Xz?&7PbSO(R+=;X|JC&t&{5r8_wdDCTS{@axQx5IXX5Ve
z?(XhLfI@*%3iUp9f9me8w55V4#FP8&-wf^B_y4YMt+VdBnM{(I$=&zZKKszRnF%XB
zYn+);P7#>FaiNsw;9n9yw*0N(g^wfC7^NUncyfHfEA)zQry)H@IN<++4ZuV05xB=Y
zN&>5&f{gu5%Kw8U0bf&r->xYhUDGkW`ksjm&rk0DmfkZ=Mm}+;0~kOu#1CQ<J#PG4
zuhGjso}eDT>6okgiyE4M1DW$7y$HGvx4q3c@j0_IP(_^o2R$Dw8%zZCT%S8QLk&$X
zc$i-CI*}1ZVut50oVm4_9!GOx@#Hj8)&;1ieK|#ahE2(n^S_+E`xfa<p(l1b0d{b#
zaE0ujx`kc^Sgq(j<E8+L;F_sPOnp;d&~smd;43}1e}<?sw9KH9H4mkQPiEJTkekRZ
zjQUNDqHm7-JK}#_U07tqAEiZ?COL0k+&%Qd!h``zRJMRyznamsh1u#VXk#f`__}JY
zE}x~zVyR1+nkp(baUeQrAUvr*IHlj7(rZiV^h@aQk82N#ZVQcS2?=ir3T^NUsxkRj
zs{KkaSztEx7|n-O_9McOV_s1w_ryK0Dd|6JQeRzJ@Y$NiOLrfi*!m>B{XKfmMdEoN
zH-TF$@&U4M5K=>vx06&{si+^I_~u|e{|h!yNF)AtnoMGn+7|&2-WV=-+$PBm9LJsh
zz)iBtdfDUG77hH0HD(J!-d`E|#>VhBH~7D?#{BwS(cf>c8sBp-A+GQFhE$Ca0pd68
zpC-K^XWW9Wzwa0^cSs+RsUdm;C_jJEOJ2WB9sP)^{4Z7TI#u#9Rs9up>@xN6@cfs@
z7k@Ci^2f3DH%I{hEm!B5G>(oH?F2SxK%q~C)OO(&u%78<kIgPPG()wGQ3Dgy{Zn^*
zOk-UsbH)jqX&}dm8C?CS;G;=`b9~G9`pAY4f1dOB#|xkMaq~NqENox9MQa`WG&Ma7
z4x4snRz69T72t%CT;aY$B1W*V-@COG>K}(8`VM022qX<wd@+Ta0;-63o!A5C*r|vW
zBcKQnv4@4X=;g@4UZwBoAErXzTN?B^74XrDyua2oPplal+xX1*uCHdXxB+jC6a)Sa
z(l*UZ05;peI{$ovdUBXLH$gT3K^+{WKs<v+&+mj^15B<aDK6=BBe^9Nv#-fgJwkvQ
zoTLtpEqsVx`5L|cM{+&jX^7+z&#xe`nZg_wZaUOAQxsV3A56^odIt1gP#@t`@n6Fm
zamL$lGNAn8zv-l$BJ3o5tI65r=oDe^<wx|KN03{hDl-9+9tGXTS1_7R-iV_rIJRYS
zc+1!*?&kP~xD*i$ZJQ*5uNs8Smu48x{k*sT-8)kr+YosEZu#*o{9YG+Ge=%4)s#u~
zWn4`qOIu6Dq;yB8bVsChho<)iq;>nJ_5>t#1t)Zc#&t%-bVfvVg+;Xogg2W*>q$$;
zUTZM68%^C>ziwqnuO#vyFXp&!-2JYx|2dHQ&gRq)S7m;-q~z<R-9K-9Y}D&(T5^k&
zN-@s}rLuh^L>l+5YqQ`!qUeO=@?Wqal>o3dVgH5=9}-@;zENRwZf|#dK-)*?9ni;k
zZF<p(KjzkaLq)!`GT^n1!7uFyePN5`={xk#+!gxi*1~K1hG^I4XtytE563;8Fcclt
z1?S}s$z0Aj%>{BMZ;x0b;lV#<<E81Pum3?^_<=h0BX#d3>hvY*;oqqj{-oZ$I`69+
z%YPqPKZ<4uc_uDa{0O<1J9OW{;3o7odJ4h%{429dK%0LYi{K%}G(q)Fp^k)^XUKXm
zCWwa-($hkEl258?x2NUs_CfK+*s1FuV@qH8o;vy#)%gx}{HJA4kMH`F=3FB@G$o;R
zjwokl#9%x)>hjwNi*JV9LD7Z$WNH=EKaNZg*c`cq+y9DBr_dRM`YsrQMDyAKn+^2n
zouv9tbcNPH$Ll@%u45xq^t+40KBNNQTAFk5j>eI7L*tvDnb`5g^v>&~LP!cQ4#Nqy
zGH!FgX4w~$gqLs@f+2sQ20iyNdLH&B{6^~32zu9$QYuQus27tpK8wDg=c6is;{L!W
zHE?Uu1p*t$Bp}$vlhKesp=deMo5^WZ9TsA{__tFO%GU3XQD03hyE==|7xy)M>~QQ(
z<TwL+TovTMhNQspI3R(0-&~*$jQEGfsjh2O?M;w+<~&HF(1&g;qEr08FfN5=I_|~b
zxf6nt<GR8>6u;`g)U?;lY2jts_#4f6YQp#6?|ZY}+Ge}3M={_dZ<ea66uJt&zLI08
zref0Dqte?WGCIRDyFxO$gVMS~l6yiEyCdRzqvCpF;`(q5itVyRx1dy@4Qtf;H5shU
zdV8xO7=4~zedJ+v#7Sw`{rrfhe4<|27ys_IjL+^W{(eo<Z|jCe?|x)v->0<eFEneI
zTrp52!zYf;J(4_p1dx(&cpD;+dmA>y)l307L;xBbd$jhkhtRQ}7H=47_o^%7i@zRT
z@Dir&FH#L(P#Nzp34P;k`zt#FpWYkz_)gP9cWWQr;Q!th3`KR$?0=X(@E+~{>ur%V
zGT`|)Y@pN0n;-;d?xeFRbUW@M{M5@cTYs3{@hw^QbT>l5??4-wUH-@P%Inj2jL*QU
z;Qxo`6Q}23)*ukVdjV|Lk>VkN&78|K5R50Vff{<lB-Q2MiM>K^{*Ly!MMeOGM2Ls%
z*n*RRs{sJqrzSinCa^uu=Ni5Ky%FkuRIWdzdVWTN1kTv@FKFJFlL0-VoF$xS#C602
z6Y(^#ONnU4rY~pLA`}5^j)FRQYwnY?%RZdkb&XI5LBPYs0U>$72JHc-c02lvD{m5s
z>UAe?QgLrC3V)jldUbK`cXu}4+<0_++q2_4KcCofjg0i*Y6E8A7LZkLnI^EoTCQbZ
zPgBp_BJKaCtMdk@7CcJNdz%bUkXaIv-oWvWgVPDv%!i=Q_e3XzI&zc1rf+yY<d9ya
z*L+PvcLaWpTI)2C!=40pk$5NSd$4k5srPSDUyLrkJcgx71T#cB0#2UDJfJ*4B!v-R
z-=y#QV;UQTHeaK+ji6jXuOS-tq$)u5-K46AiH!YO5~MA_Z1O0{@pjA%qo{z33Vl*H
z3`AhVu@f_Y-v{@P(cV{R&UZB9Ir>2FA3HN%-y3k&$1ot#HK~lXB2y*bQb|Q+G)HDO
zM`X5!WwwW8bcUpLg~9nG_C&|`#l{cBBn(6)41^~2+mgEViJgY{c901yp-txC7F%$a
zJ!HThe8}d1*kn7RwVhA}otMYGEXw@MCF{qvDZkE5zD(s^qOf{u=;OsNTt(Bv1@V$&
zvpMYFu)*vXTAYxNaja+iKd^D=-;pu{_OxA{Uiis%>Y<;gHgvV#1%rJ}@EhA~&+oH8
z;^OzfKEr)m)EBne-rSz_)7HxCJ5EgRdSP_mSGU-Xi<OOhhVUr~sSKNR`Qy{#QF6qG
zh+gV?`UdXDyB!+vu2?1QEEqV#16&8VXq=Z03V?9ycri$p$fPvx<$w(uo#Y7<D7=!=
z(B)}xv{pVrq<N?s@K7d^Z=eN<?T8*XXsEPdt`o2UBO9>6lmzCBu(!u+Y!a@Ody`0n
z&VK|+(N9$0H45_9=Z06mKI-;6aoS>f3N<uP1mHjM0>IA14s8-ELaXT;n!pCl`@=Yn
z%zJ`f{La|6-^s|3tAkO7y5A-On-N4%ptPW20{!*7KBCv2x<<wScV6V{RN!+9vOmAG
z@y4d1@$Juy@Azzd`&Dwi;;lIc6G<^<a(w5+G{khEZb!W|NkK8E@#_468Kf)ovE1q+
z5t0WI){>86)Cn8FVA+pIy&#h^ObUHH*XKMyFM5d#rQYeNydgY86C5RhB{L*?Ty%|e
zZ>X;Un-S{sktKiJ*f=$Y_eh*76iXa|85o9Gn70qGzc@Ajy}zjEf1;khH0QNjV1nQE
zBfae_dN%}(2d5X84KJt}UvLzKO?njulO0=-p*81D<is#4OM6273u7sxQ+sY)-8%Xg
zDvi!A>ULyPm+0;9(wk4;T3_|ambgb<1CR0jI+gxSGQUbGD%bzOCL*&vEVDB#ts^|S
zGcu_=CZRVru@4rN+!vnG7nssxP3^KKcLpc72gbGrM6~&bb@_+%_y_h`t=&dLhsAF|
zV>=}AKgA0px>Z}lKG_s^X+0LU<h@7Lyfe4wo8@ONuX}!K>!<W?XmE{@`VYJfLUT-X
z;0iz`4cR|Znty^c;%&f)nIw<nP2!&FAjQy|f2HR?{TtQwK9&C(mH5(~Ay4nHJ$S%+
z#?yS#Lwjn!@_~Jp*Z0N#urc?KEq&vgA0ONP;e^jHiE1!_K&Z+jL7KwF%9xyj)F6&-
zq!obc4@)`8otqdKG?1kpq1WCZmkSuE%cdq*Oi!$x#qcmG+Bm%pA*4B0Ta%&!fem63
z6jA{jLZn~uB%wJ{Rl`KU>Coh&CuVVtx?UyIZ+J0m?4_f}ndrJsjJi%wxXnyL{|UE1
zpAmWk){>sRPF4R%bxq8<hn|0qUiISCt{=${BESP|#3(R2?v9uRz@H|83K**pwtab;
z2!bM^8W_jzf5{sYn=Z~cVS@w=6+gU3z-A3SvVv5u=#^tMBtDj(y-6j#Lq)zz*<W6e
z`qf?aH#Q#`-~ROUmQSX(T*esb?LjxVQ?Nh)8!|02Me1PGYmguXXWRjs2Z<uwvL8tQ
z1fSTd*$H$Yz(ORjxk^S~SD*%c?@j9P2-Qg{@(Z4$7eZA23Jn1hkgXigB(uPzLVyvB
zpJs`-d1sXRXn4h?>)R(0RuLWmC>lhK1a~Wxb8s`oEc(T1>e(Nt!8fRuH>sZQsAIpc
zdg#jTC#QWb&^w`FR5UuTVr>2~8kNi?UlHd9ftS6r!}yQm7a~$X03yCct0rbqTf;3I
zG!$qB<23`l=m>B6iN5p6=~eAtuS<DmXW+eTTff4uO&L(92&|;yGTURa+M_c%oEDYd
z8IwjVDyiLR5lL;KNv%N%%>fBb0db8%aV^0yt-&$vfsyUD@K#e$vo4@XZEu#_TE%`H
zd|N-qe$+SMKA(_Byu+UHh<e#I@vS{6A8${;xIX)@rA2?tZMro7?5~So|84syVQmma
zm1!~@k1K|Q70Q#&)c~`&uyBcxS`!|3C}xlk4j1b<?K@6$uFz|q{E4c3jLLjqQ|j3r
z_5l}jKht!KAwBLby3betnoG!+`wB;P4$xa3r`Nto6sIum?<@x3QX{2>*;(hNC;U!l
z0~Ftd0T0MkpmhY_jadMcdYoJbWVcB75#9hBAOzuMlN#FJLtxIN8P6g000J9&=@<0k
zCnpyjf#}&^RN;^F`))0GU}D3^^p;;q6&1e1jUMr$uY1m3#i+JJ#%UD&Tv9B)Mc?@u
zz2u1*s`N)fv}>51*Eh4`iCIEWa1dun*?I;sEty{+uNf4(DbirM>*DO<cP6O&hRK*`
z-Bkz<E_rG4u5TwWbFpV?X6HCTj-xR!wvN7h2cdosldiy5^b%}UfL>nM_mu66RY`wt
zsTtilMDKW(-uy9*?OI6N9GxEwt4*L)iWU&@KjUPSoqBhY!Vvht7^zvEm{{=S4CYKX
z43qR`$&51&oWW`^P-d20q~U%kP_2)EkMrA{q0xm;k;yh}{#k$9Q9D0C(jQ-hVGtD4
zKAWaqo1k8~OuhZZy5Ii9ErV3*CrPl1P!_2v{&6Hy#m2rDzN0FhrJ|mqLf@Yk^EDOm
z#+;~k@6NjH+DLD2BWvx*xcKw*(r;(LriG>#YvKlHY79|2{^__Kt7d59G_7&eV-WXZ
z6ddH8)%5I2&`iPU{_phq6VT{+<1WJ~PsO0#?}*&iF7~UZ60*Bt@mXC?i_7eE8gd?3
zbXprMDz!Btxj8JcF*Lp@JiaYFt|KhIBPg!TKe`pvX-#mwGO(U+t>IZAirdVxv@<Qe
zEb9Q<ewgEbn(O}*E8q?9ps)5v{JJUr_cg_Tt~&hJs%NjP`)*|O*ln>jyj<TYntzK-
z0z1=pzy@KJBYP&pz`&9-w8M=P)pwK&_kVtaI`ImXcXV0Y&}RRGp8ognET`Gx6CBZb
zp8jQbzxTH%{J5=o0;-~Spw;*dy%GM+0SssP3w2v$$+#MD9Ux+c;i~$_z`SxATs9!#
z2+y426;OM@O&4MYq>@HT6pjWIYEu~22d=Q5#OZCIny}Cb8_09!|GcPsbit|d_3zN@
ze<DF0@==TxdDAz&W^W=wgNBnsk(6|DHzEG{f=F^w#lK)J>HKPXaVNdzQF_}qB#FYc
z;(!el5r8p5Bn8M3vAE(M4@&<>V~FUeBhysnb?V^kvS()2e>LNRBD-@}g2U&ab#piU
z`#Sp4UG#N$NbKBR{W3il<BG8mO#Nw1@{P^aV>^!0+n=X5eM)cqi^QLBzrf6%>9L(N
z;2Ds&jMCPMKQWd!Nj*D(Y-;Y{EQCnsJu<!$JB+c9k$ea<q<u-xKuKcRIK3J~9!Nkw
zK1vN;rW$_(4Rqmu=q2ydD=!hsuahc@`4PApfDP0ELNon>Uij+N{1<Leum8CI=NqWs
z$Y*F7Sev6H-Ni+M&k$DyZ2u3ltIuAdazCW3Z&Iq)mIl7J(DwH7fbUkv4=>F}ZV7o7
z3?NP|{bm|VH8CD57`-7FAHxQ6zy|kX1k5r#NJiGlVq<4e>P+X>JA&Ay$9|ui^73-y
z{Vuu_a{m(&TbIPIl}gI(by{LhkJJ7ykYeG89N>veZi`H8i%KF<To?*&@$G(5Eyl1$
zbx@tiQpGpda831WV<Xep$}o1a%sp(=VZPr3T-$$Gfv<Rmzq>d7^KFGcZ5z10{qgBl
zFW%VlGnw%s5gJ)TJ<S`Ng0I9a1Rn}ugGo$a2H?S$cmkm1>@y;xz$1c5dc(UvQjO=S
z$a8nt9y;LvA0~9)CFi)}3j*!4p5~X_l0J8B{_DWG*&Q#^n?9V{_S?+9QAEa=${Ydr
zc^4)f=#f42C<+o_?l`IfxaET^LnZ=Hp@Qr>DS6N09M?23HVqyJz6C-D5+VGQQ`Htj
z1bEtJ=*Y}YV6J=ZGzdg9^Da*kZv);6Drr^YRMzE1?Xz=^j@<bwlzV5O?LtcG)6<yH
zgcuDrwBfRLL{JXxKdejxJp==DWq)A!er`Fvv>j}28vJW;bMS8lY$Ve}p312+NR+8j
z3`9tI9YtXWj|(F=g;%N0nMIGxuK$SMb%nG6aK}MRvSSKSG4OfyZu$@8a=2c1fI&wl
z9k8_PS1R!P6{*+O*NkjFHoN&ndczmsP~WaHfdIOD8ca|E8)vt4%~f34Gjm>?f?Fed
z;IJk8!I35JPTzHzzy=ZE0(xqmLq-B2?z$U}`8JScj$Wo3ew#Nqu?QVc5Cwk4sJ4Ui
zL8zjH019t|#uu1u^S`3!gPi*GP3omT*L*vTQi>Q3XA)JM8R6`d04)l~no6`>L=X^#
zg7yjQS$T&Fe1`IaFx4}Z@h!BWsO0NZEzuRjY&LXSb`R6oH^Cmi#hILdw?V{?D+&$=
zmlF(eFDbNQ8V?}`M$NC%TTo^I!|lEiD({^+!H?b@c*Z^Gs34$MYHy=b^7~=Qd3~^d
zAoTyB2wn)tA<ieYBRaVQHS3tv9+a&^6FdE5F|N}H0-GwZRuRx3u{VmW&3tnU&(g-X
zbo0%966<lH?KI#2es=hyKJhPm<bCSVdEN6Yq;SUfenfj+p?xQaKjC2zWqL&7XkR!n
zLtKqRe%1*aXYJTo@xVC{<$h&)*@aK2^ylaK{pSwTqb}C}_$VJ>NX~Fo=b5I*eB$2l
zt@_sI=rz}eZ|#41(&rmma)Z=xT}OxaPhLAPbHin3WdE$g$Qqn+5D}oy<`~`t@d=j~
zVL)N^C;lkPYs988@H4${W(pJ!=nUcnGjrP}N4HFjZJrw6G&Q+lYVyuWDBMn=__GRu
z7P1TI*%sr999hsv&pkML$BWpEJHGp-qk)T}8A=q8%Ygt0XuOVCg?Jml2LCD)f~u|%
z*@~j6rLFYZ3-p$c=$$vny#ob(;Ved9A$>>`G=Ur<qk)<dv^c(*C56|)Y3$9V>LwSS
znFU>N)6b-CftwLL6<#pp5qHxgn?e1axj~v>!}QLt=ry2M7G9*HzFC-gd3D3hyG~5s
z`69i>F-hi3qj7iGK6MKt>*%%-QU#Kbn~+VU=e;~dotU1BA_Pho7lr|wb$^j!8cGiH
z=n0U`7C07!VR&=R4>ZBpiEC8dAM^UB7eVX+!ShAO$`FUX)cV<R)ZYLb{Or=}^a6-@
zK_2Yk>(ryy;FnylkV1$UAek8#(j$;A;b8~n)C|hA$n<2xwCCsa-H%>hdE{p*^c-b9
zPnln&LO!FC{-jE9A0f@wrJvB-L2g1?I61*0yEF-J+F1@I^#%+Pf}{%xY^+^?+W|e|
z5BdNmte|oCAZ}rwQ(=!R3_7_p;wU$`Pv+l#8#e!c;Bn#y{s&<jK~7p{bZRFsLtvBI
z8J5%$7~f)xsyBz%St6TEVa>YGW_4hT!oO8+Z&TPi74{yHzEhxY6IxmUQ+C+BjD%;I
zB|rOiPP^Vi`@TdozoyxjN&8Ps-{R9ZMD(bLo`9r_WOf=Rp22+uv}M!@2y&d=61-2`
zcftEa%f;gxdRh0EROF+S{`lRtQwI#k+@yU><zb%X%s#{EeNj*EE&Om#>yIucZV6wa
z1Ad?*CeUjq95A#&kXsW>5_y_g_sN<4lT&-9fP|@?7zCJ^A`=0#lbGnjOdFC466ZKq
zsX3%x@Zg@w*<F(}+a{;c{X{DrRX*IM)=gsR@6^gm<7+>lmqF{O6+7f;DrRDS6+M4o
z>dxoE7QMCWDk<Y2vI0%PZIq-XSPTl17@|Q$O~}F4f}i><P1W6?3Lwurwz!#Idyc;Q
zt(mRAPXla7_Ml4$9xApT0A>g(5PcyT+crV32Ak`#X~+ps$ycc2Nl4h-`5akcjgA>^
zO!y3uL^&3F!Kq^K6CFeHF}hu%w~+a~pQ*I3soY<dwBJ~HdVKjy^x{uumq9$<;SI=i
z;MB+#GVYF@JdW*D4(i>4S0<@r<8uxnNE@Xd9$5xyt1ARH%SePa4r=pUdKNV2Rp`cE
zr00R1c?Lz{Kj-w!EO>}s^ajx|!v0Czm=G|dDhT2{C_Dr?H;K2Qp2gtZHR|{^_=;`c
z(Hut!9)J<h*F^M96cusn<Ut%5t*{wdit}4E6O-+|Z&GPbP}YYi|JSJK@2G5OLI8fG
z0S(P$%m|akEsh0B2>1aTe1NM)#;DK!q+b4c;cM5>AII%gKS^tbY4Jsx^8xMh7`?pa
zXDSH4@YuH4lbq0gc|Zr1RyYVtEf{c`6F>it|8XWiPKOhl-VKXM>jG@TQ`$+jAigOe
zwjm&<$v?Uov!bTZCVg;=F0fS>*rpHYkQo{=@g}#F$pfmzL0y7~v)tqlSjCswT{Q1J
z&H69R{$^G-HY=ww=syCJ(GxOy3gNqKnh>{yvlCnziw4LZHN-{mHuz`ngl@wKxm!yo
z5ryBRDsgK$yfE<KMoZUjWs{G#Q(zu)RUX=7JiRI4(K{nvT$}pQ=E_TM{j~Qvn)No#
z_=4vDM2j!cqCaU7KnsPMF@hz8hyW6N9j-7^DsX7nBEo?o>P6;F00X*;V`GPdrt!BF
zwKL3cEQi`KJhS$ji927WmmZ}TR6rt(rb4IY7Sr>4r|*1<Uj6RP?SFwuMQ(4{1d4Vf
z!c^3z$as<?j{^2kz*zbM`lO>&{s@(IYeB>8>eFDvO|Ji8eE-M{L_Og;;4~cVOVW<S
zuo?{H>opf==U|&6ZmtQ}sH`c-#H@LUUIl*o@3+$ca34WP!3RiMAC6BFBqI;#)_)2y
z(O;?3FR9{hu(Eg2{WllCGPU5->BWD{V8+H#lbzf)eG6;0Fuso5oxD^qj(-AdCQ$6A
z2BxT?kvR{IVwn%pAq<l(ot<7lyv+m#Y35>@V1!=t2aW0?b>S9O{TEu0^DfW}UL)@m
z${wWe<d|QAz9GoIfDN8c#s|<Bf_G~B3pBQMuhYJN(Gmox1T8m&^mP$EE_W1-5Un8M
zLU@9I9L9|~Py#vo4VC&d<^L2F@gbFa34OB#7YGXuGaXJ3<H(a|T&Kw9Ot9>hU%N%U
z^9yzFdsNRG)Zx$9p1rvBq3g_-X|Mm#+$U-Ie`ufnndR{>FR&lq7IuUmI-m^dqB4pH
zot9oS;Iy>DeppI=uhWwAdXjQcHt3Ac>L5NRs}~lR(c|zu9VFpNX@P|$G>61B1x7X6
z!y9ZNP1aziwHWP9dcPWDV5upj$`I153^^i=dRCD71(3t-n_)jV%YB<xTp_oA_!%iZ
zjn6^}p*^q_r)I@da5!jjOrtZ1xk!|cTxZAjO^%>q>pFq{;1oVJ?_cSa4}MIQouZ=J
z=3A<_nCpD4oos!NuV&Cyb#$xZ^cv*@i<FPg(L6WD_!<@P4wZ(B<Htn}msa&$xpU|y
zq*73?-2cj~eU!FNkfMqX)fI@9ISjual_(V42on^#Y7VqG4Cy!#2PHi?M?DP{w&mla
zE3b}j{AqUG>(h5WNH1xl=fl;6)0BN|VFAWkrpP+Z*G4w{O6V}#FhGtYu6T#eKat3C
z=$0TcSn(6R=oKvOxJ4BVQyDks*3GUsL9cjvYR%{4@KjvKTqn{hFq*+@MtDPp$w3&K
zoLKqm<UF`3Y_5+Vp^|48)XuKCmtG2o^A+tja$AxH^LcnLct@np4*X9{A}sN|0<mIL
z5q_rHf1(C{o^#>Kg106XeLIc#9?3d+n=@cfV=NDvw`AuR91e1lHNVgcu^GGn2Gxzo
z@FsQt@QOEQHlR2*f&654Aw52yo<w>xACuDK^t$Ub2B9hRqpE)gZ04S)=l_?$20{pU
zJ=o*0lkjwJA-%!xhj7Xg2nnGt1W}Rl-^pIO^R(L+gmi*in*@wLdRR!`!UcoIF*@V8
z%^_}wzyox=P!r=`q}Sj7D^>6&75ffV@Y~}4*(K*dbtOD;GG^y^`()~VavT}v*6|7W
z>g^ZlxexzLRlPtZKSU)yNX0)oH|Nded7tg6za<-_m910mN$+m6-Mio1Ck`G^1@}@}
zrH7oBS#r>6P86jU^!;Q1LQYpgZhw4EUwmfIZReBG7MszADt2U2V?<(808&7$zi4b@
zP*i<Tczs}4LttoAKzOS)u+b7&?-x|%7g}Y-3{PN(D)=;z;~V!5ulf(+@k#0PwDxCO
zi=l8w;6$!h+*BM9JRd7>=_z1_gQ?8vo9GX^Ob_oGxe9qslu+EpZy~-y3GVKXFHwh|
zq>@`HYw2A<ZC)V*?wWQNVaq;2=PuEqZPJq)6z5hc9$X@QaE|iv`2nxb4SRP%^!xMU
z-=CZG?)>ccmsfngy7}v!C$H={`{(Y*t{r%B#OLiP&S$jfJDUG1%^#;xbpbhuq#}+s
zC>%YhPyDUCK|*tEy+z~v&xsY^O>F;w-h$zrA$mpm<f6n0DsY&xURjn$&#fI_aX-D}
zg)vYV$$k+cpMb~&B?l7VIYupRts><((rd?^m~46~yiR2xsf8Fky%g&8A5LxmlQ16O
zcy~Eq17%RK$Wg8UO=<1K@QU9i<~~o)0ZSlugo>M8Trzv-5qjxk^s0C1{eRugdO_s^
z`0@I2({<!*qevYQ?d-w=-DiiX3)iT_Ur>+yw%~)YWk1hC_R-nf-8P8|3fbibe+X<7
z#sD&w-yj_d>d;N9ZG!5)LEU>}*(=i<E|NWmYsj?o<U)FC0fIClT||SCxfton<2R^U
z-2SIAelz#KnB}IojS|>E3d=G3bjzWji&&Rlf!Bi%;S5HvG3|GC^`Qxm*J<fxS~25P
zU7S(SQ~cTM{F$4$)$$0f5kokm=$pHajpG(9{RN9*Afoa*mG|w!?(umiXE3|5>334E
z$J-$@k(j!l81<YS+c$wUi56qZ?UkEU!)sLX!&K^H^K)LLZ1+*NCzb@iyDj=F&*)DX
zfiE&4Rjch*1`ZO~6jUG2t2~-hek8l>FpjwuL(cPm{13qMkI#YI`KNM_nA4S%-I18p
zo{-rFi%V^aO=*lyY>bR=43BOIjjRt21AZDU0S)HBCV<Bl?68nlLuj8W`~)xZVP@JJ
z?1o=?r>1!?(!4Ke$uKFOBYyRsyzVmkr!PH%@XVKNNFR3CE$lyYogH_b!Dr+Yr*(+X
z;@>|^Z~ov4)%z3`)w3kDe{c9bE{48cyoUY!rUT;EJ>r(lqNY29jjKcrOC?Q<<jo6|
zZSz%K3p4`@3`6s+_s+GSqXHkH?2nN;Oz5jr^gC4Ir(`r5E7yy@qXsTg$G)TPzqI=K
zi9MgvEQEG9X$dN!6SM>)><*F+Cd0S>KvpN9n7iW<8hf@^7hk4=KcDCKEw%zus?XMB
zPpoX5UIV$Tm+3WMI%a!_81L|E`gar_u)Oh3dSog6*HR)tHNEVQ*|~2(|8j&(DW_hi
z%BGgzL$7|4tN}&1g|3CmB=Yj%b@VWX&m1xgq-(rp>L!*iEJLdv3sI7%sE`pV{1>YD
z&n4J6|H91HuW65AqAdzOD{>)ckVb5FcprqFW~T^yV}fvhunTqR8#kyoFD?6R60_6j
zs$rl7s~S;##$N$!&|N{0hPvPSOY~}}(txN6mT}untQcDH+|0)B2x$-@D>|4<9iz^u
zs=<X~{|}m;^PmksI860I_F$5FWfnl%JnU%45v~Kubm%rtlBJFqTLL8q+G`I?p&LVm
ze??_{yR7{$-*<0IVR0vBrSv#|`Y-nQMIQZ!h`uFoEPI8#8v;?}s|dV(KEd8GjHZ7^
zRsA&Y#N@)~$pjntkZ_)kZuKN%dIGdSCZW|58C3NZ+U=FGMW6zvpQ9p9Q2K+E<s8Mo
zk7AtJqkWdGc~~Ml6clt)=ijdi?xRZT?=7skC%@_#z=Nan#EBf3(+fExH>dM)L~h-%
z<h-ur+>WH2)})+v91}BI64IL!Qk&zFn`081qvDz)qZ`Acn(d*@fQmJ!!xq?K4Q#dq
zH5fzc)M2ga*puRvr?^F5usg1?@0;PiLMtxPdX!2?YBAwDOEmxY(Zf(qLCF|`gXr`*
zR=VF-fg%hG$BhuMA%u%7^u`x1Qsoa*K?he@2Dj?^U6eg;lI8<~s(pgeUBaTR;`|Mw
zoI3@X>jb&$1o`WPrRzjhcZeI-$~smn4z5t$vsiIvf%2jG>Zj)GUY&1xXTJ5Lh5lcn
z8ndYESE~AlxphA;X!~th|CP1JhVMK%dgtlUO%Kj)dYIn)7>yCAjgQiIJU}l#Ilg>w
zbXm!78}j~G6!bo&d|`?9wYjpF=4;>HnsaS!`Hi)M(|0~WMz1j*10RL3XFYvo<K)dP
z(-1Sbx_I{2h4dfG>Ct6>5^fH41nZ_Jh@5oZ#DacYK=iU#>CHb9*7ku(44;qOMLIAO
zq<cF%zF~H3&Fsh>P?3L+Xqi)KlT`2xD(E{Z^Jl8_+QR!sHoP@?;8&v2<%~J;p^@7S
zv77tp8!q%12}HsDL!JPwAGG;?9^W`}dolt87mig6C@?y->K(8_>2Ez2xxGAtHKSC`
z3}HeHj;wfM8tX>3T_(&aqKrrSFtca@12$;7V_(anx5ww)heT+M6v|&2pZmoG))T??
z;U)+b7pR&bcLI?InI87ot@s4t_c#^x!HV#&=R|!;#eYaueslN36M|oe%%W%%xy=nN
zeO*ak<;`4XO<u>=Jx9xO6q2q`_TgcZG1NQ;G4k{ab?C~xr)HLYLa&EXhXXcHGiJ?z
zgbq$E7__sZF<LuH8?Mk2aDQLEviISisLb<J!lRV&JS9KB-}JOV|FB9nXg3|y+a0i}
zXg*!mcyDptJw>(0VTCov3aXFhRUUC#ZUtP>q3kjQZv(K5;{NoazVzbW^unIBg3i?Z
zj+Fe?l)U!jyq46QHdu02TVh6QVp>~#a$9U-TTEhSU_^^OtlbvWV-4)J1h$)k8_Z#4
z`iKfkVy`;+tSsjZarMu<{+qmW)6&;z`A;OZfWtx7ix6Vtt+7p$Xp%VRq8)vk>6v{q
zGYB)B-A2b2s%h*y0Lg&=JH6??FR9dXi~Nsov!8Ug9`#VP@8_5A<>u`bX73SaZ55_&
zVkhrn=I{0)3s!gdmhWIxZDZAJ;WTgJwXNrMtmF0EAsARKJh)Okv{G_>x%9+h`TdKn
zFH!c_=lEmc#A{UWf2r`-=El6WAmP0gS)VV<{(M=^SIcw1UY`5yqV%t*xR0rjcUFXb
zvLW=d<?3e_3oqQMcodq#D^*YKPW*Cr#nqj?Q`^teTmDOL|B(~}LFL#md3oQ|EUs6K
zSg)b4ETb>4m>gO1%jBXr>A4u2shOtY5vSo6NUy*~!I$Wr-_zc=NY57xP6r{4bbA~E
z!sDyyTiC+6@?&}}R<h;5+gzb6?^DqqQsqA_8@jgT`6<^gX~y;28|g?fz%d!VpB_Fy
z-*hGQcZUo#6c^CSa17;RXw-2_A~}HT1bjZ(s{?Q29P<Q!;%;beJP$>ISyaiX;_FoR
z=!)|*cm9{&g84S5;+JDP7<kb5DKda?6`Or$(G{hRU!l(ZPQ7}Q`s&u&t7P^TcSFad
zEmnz;ZG6ZT;f1gUc-`YrT774Ez(<t*L(29R75q9?_O;imWO~msMH^;kwe+xj_KJA&
zns9Oy4IH4wlZX|Mq3(*o6Ax%mJ&h{gU(}&rsmDeZy*GQu6%uyhjUb%EmK2oZkiIhD
zc_zr5x!IwFs-2*fSlIg9&r1$}KxIBp*)Ocpo;#qqSFRtj+K*tsSsl_()wJJV*>VQR
zDXBjJbK(ccf#p{X{Tn%1rGo@KrTrNtJ(<Nl8O5FHMIEUHZ7_f*J-;n2w>>qdBPF{d
zIkPh{y)z-LD<QQzEV?x)qQgI|*Ba7m3F)+kH`=1gZ82rm*hXv8kS*^8d+j&I&MWdm
z*M+CA`aX7Z_p39|Ulxu!mH`rV0U~+2`O>H-hOS5PADxBX06t}Sc1Oh>66PbD#)pyW
zs1aklKDwsywZ&0qR~ZlPwH$U;wI2{v9N^~d<3qMIV<#_V2Q$UPH{Z*v$jht51Ljle
z#wc@RRUBYf>}6N(=2Y(FRBz)|Z{byK=GDN~^IDg)4ld`PUMW7iTykca?92-F*>%=O
zHibOto%5m7d~-kb&Hc<H?W5f>Z)^#CWnb94uAy)5Fg&qIdTNL4#3s?uX4!pvLtfgS
z_Ql?sKlUA(*o|SncZtdzvNGfqV5zwv#v7hkKR&W<b{w)sSTMW-{4O-<%4UchPwe$2
zjbzU$)+lUzpLV}SIHt&Fu+#~64ZwzU*-4{td@Y1yJ|jyb=NC-Q3B5$=UZ(8N&(Hd7
zZTp|w&yO8=jrRSSB#kIKp~s2~3-Js#W;@ql!IR@7g@MKbBQ=9r+@XjVHx?y$jbp<p
z@lMX%28D8r%G`}5c^Flwhf3`<m33oI^Tdji)2se72?<iL7Lf0u!vn<#6ie`Y!jA-T
z4tzq48=SmI-Ty5Vx99vY3O1*6BN%1@5I*6Vz$C70<UDtNL*Mn(_|k$e7Y4pZ>0YM{
zZ&LdIQs(EW-1j%!^NZyYZ5yF2GZ0&xQO`~*!66u@c|`IOUA0?KSwQh~&t-b;d-U8#
zMyRuwsfYfW|Hjyw-^hM>+z#+g5uu3*4}AhTGj0|z?3!nyi%9Bqv=keMKAYI^z@^o7
zA1w`gc%9)m!+KO1G^7qbqzN6M>N+2&ZaeGrIk3`(lK{{EMGhPeV3SogkW<#5Q{J0h
z){|M%kx|^9UeuOW*qU0<Lhu9Ulh>J?)0vdrm6+9?nAw|<(VLLc6_?nBF@&JlZvU80
zunO$awf5-hfT+6Qgx-jp3$fL2g*Ja-ZvIZ%{0*b?_qDZGmbFgIYa4;a+{y>0y+5Zd
zn5TCLP~aXD?vUlj%og0MFO3y6H%Q0bu_H~1y8aim=egh3=RG|~*|tg7wO85XDyeo8
z6hWbM4>NlwBWI6qj;D8quV)&=E8WLC-Nz?`!N_8;a(&r(p3H1_R<=7khZr~4m6z|v
zFL31-?%|Z}6;>aR)$f(n?vhsSQC7K`n|*^i6{)8*nP<%f4+WP$9Z~aAMAb`y#ZOyu
zp5a71#tMGS%XG?BanN1d>niMalMZ=XF05DHw=v{}Z3SQK==^=(xpD6MwDAU++_Yl^
z4a8^1b&UTd`a+O2So17d@i4!Lrfk1cAy=VLzWy+M$GM4hFO7IyCf>%8osX`YAkrUX
zLXM2AjDbeA40q_K=|!cZ^TU6jRF6}-2dTi9R%L&FcgOEr&kno3O^dFPViYbBlxxvw
zCmgL=blhhhZE~asSdKy#*Ew3|klg}Jo2<5n;ya<SI5$f>s_ht5##F^Rz^0O>5@)HT
zVXAs!*`cYG4^9HxuGdM0-AO6H_AZ<dLIYZgH=xNhOFjM@_2`$>n-{5{ZX&QGJSK;j
zIQVZwkj;@ep)kGf+nKe`-dxu73l;V*C3}WayhO=gqBsvy5|WE<8TyI;&P?zG8GtpQ
zNHBR@l>rz)3mc*~Xr`@4viCJeIz)Avdf=D2uUx||32KaZBWRtG>2z@6XK%9TsoQUP
z45I{7DA?ljAx)uiTJk&1eS_ZC{`IP;hc?&`yV`mc;YW30hpGCm^R*r4s#@=Z&w&9y
z<xQsmo`1r&f4iOB%0u~81Nl{bd6m66<=xq3UH^awo+rJqE3Kg0k^l6k<P9e04yNP`
zWastg<n*Ox_oHTwk%Yj6&VZP<fXF6zAxw{k$M;2~p9;=-+?@NOI{zI>$@ktxSN9Z-
z++FbJih`e(RD8d@>HB5v-z;kTX!AWkZaee+-bXKb{`V^Pt10mnvJD2sH|*%!@dxb!
zfzw}U&3ClzJz8=4PnX;$*Xvq$t7_e3r5?gOsFm*Z$=L0az27Ix%{{@>EgI(K9^>s1
z@8g-^<CEm;o9ykK2=nnx_GP3nS!rxe28Wx;;b-}<Gnj%LmL#7c$?+Cvc*`=_x_rL1
zL>5%7iD<MZ_C#bIi_brkRB&HH?tL-24`|{~$fE8M2OQ$*yV&9;mY{_#>}JRgZ(?+B
zlHa?*|H(V!Ub`dz^DSLhydI{7Z_v7LX~DO&=MS_CMA(Mu{a5JipVM2Pqqkvk0AS-k
zydYw9M-%OOigtUN_WG3exk>;AZ-5AGJyhf;XYa<^^yw)g3<Md2&k1anRNPz~^$n%D
zK&j7A$_q;ZURjax+1ln`w?90_`u6rN1Kd#2Rwk|YY0^R^R|!4NrAJY^BtsTtyFHfA
zjS<0+nJM8U$+J;_aS*-9#l7J=D5tY`J?)@D#L!gq2vsn?q-$#R*{O|h(0hKS4>-u9
za58Wmx3h3Kv2ofPJ)AG5sdp|>AN@l8coj4jQqx0BfxfCEv~$7+h3R$QjjekA`kap6
zsN~Nm!wZ!B8H)D=#dwGkJvPVo0%dz>QNt&0uTD$;pf%H^fq@DecXEO`eamGAdU}p|
zhZ&-egI(D~4eP>>b6>u4$B#62?BUbH-4<*#%-C|t+E$cmN#h&4EpP%_x~C(|L#BY*
zkMLW1-+fn?WIwYw_{g^4gVN|@`tT!EWABCf?gwf*&i-T7?e|r+ovCO!1uJj9x2)-8
zNy9zG^~a0qjuq7nmDC?9tsN+-=`X74&9Caot?0@r?*x9b%6hWO`rwYzOAn@(98ND8
zN-sH@RWwvs&|g^Com146R@4)p*B_ZN7@l}IJmyerbZ=O2L!f_&EuhpG*rW<NC<}W`
z5%UT!^eK1yLtCu(-C;kwGWh(Gh)0&iJ+(UdRVw&tD)e~@V<E|(Pz66xRhOxzVXE^A
z5eIwt`qHQ9tuN3%FVnhL>42wb+tDAl*xU9RN?jG{uEJPXPRxF03{<f8c_$oji*$7j
zcX0`GSOm=7B@*W08sq5}>*F5J@JwX;q;VM;d{&l#o$c$B!DME0IC*St4uhY~5M^@|
z1w2i$+*$+bw=JyOKe8_{0uqgbVG&0{Vh@|*JC)H*+R#==%t@p*B3_-4)5P>`^x<~z
z=MQca9a}EFe~$V|MC$Wmeq4}!jVk<$s``cM!g{&isb_C4|1W*_EA+mn>D^Gl&8HXo
zk1q}$-dHrdt7q2jA)50E?Rt@DFCy7PQ312AC|-jNxdvhv5P%2W_%r%W>~E{Sxgz0f
z%KX4W!>Oh6^UF0)Ee?8Xapvc%dw$#X!Wa=y#sDyBPtBk|rbH-@<1Fq$plE=o0q~5m
zh>dgTQSQ{RaBNgC4nhj4x#L<yRbbs1y$TiHcgEIY_Be;8B55iN8z6@lHjS@2J+tve
zdgG@vAV3iE=-;9Xq`T^nz$aPjFcIsbez{5gb_3dLg!Tb%0~d<8&Y?B|@3fu3W^~z0
z5XS*&`sUo6?-zu=yF~TuBF-ZVd>*3Mk5c>#RM2x&^(V_8zT$q-vC#x~lbe6BrqI?S
zq*!P5V(|#Q<U4xl2h-Gp-z|9Que*MjfgUfs`fb?Ej7v$VHj8_Q0KUb+HsH`B?$ehs
zE{O?tuRm$e&v3(<UET2hoVfe8C7cq+o-jlVQ7wHJntC3n?>q}@=(@kL`y7sS9rwX%
z+t1Xrovv;@Rn>BDW%J33rW55&$0{3#DjJTI*B>seJy=vbSWw;XbUyGxc~uAVst!BS
zp`qN0W7!q=<dq-G&Fjj~Z_g`k2ei^l2a*ei60(mcq#cWn=?Mv{vY0c}x@3taldq~^
ztNZwdV**_-L)qe~Zg)2i>@W_jR&*^E_binkU9CR9QuCjc#+O(5f3Pb0r`4&~R^*H>
zEV@aSMpS-0zwMVhdN1GI`pe#yAH?0?nL0mm3GZ{Ylpm0#?B_+eawFV0Q3n{&`@JGv
z-NRko2xj*0581ab_<+;)hI$-`^mK{xagAnr#Bscmc|NHEM!Jxh!Dr|2xVd~@E|-_X
z=4JZwG8lq%UtxwPH;q6@R>+kULQ{*yD-?+<CE6;XUkTTmi(w{(rA{KN6pL!a{Cb8@
zIfLD>&!=vKXZ>pLuH~E)OC*mk)x5gE^4^N%U*={1NR@s=)qFuUe@3-^v$FHgt=-eU
z9rT_odYR#uInvMPn?70)_TC*upYA$&Y0v97;aVXgN2r;gPK^<A_v+yZsJ#&a;wSXo
zfKC0Cl_*j8J+j<<Zj<W#2HAtlHUC)@^3tOG_wP9Tt?L`tB+&PmrF9_8PY^?th;=$j
z8a)scre+~94;@DM98{UOvt!)JG2EEZiAT0T#=oa%Flbg{;^3u;)u`HKj8b6}*d9m4
z{JE@R7&OrBPtu!S8}-FbMF%AI6DrW`ftgtdvjBTOke>xd39evdcFC1dU>+$G2z5>s
z9(Xe#lQdYz=sUk3UHl5>sIM+MOm8_tyL68;YX9`i`DR1tTa@-GN{2~-SE&4VsJ<_k
z{r5V`>kftoCjW>i6qfBF2Lo)tzg~I~8@y+!hksn~@-^gR`$ujoZ@|g$X2`ok)asmK
z5==}X_!rQ(So8=M?~$RV%e3b&wDcQ#$En}vWIS?L{8>TlDShNIs;mEj&c1UUz4y2G
zoNepAuch;JbH^zdPMn7G#<r97E%($m9jk6UTGe#Cvhi4X{ZMHwkVCAj?g*@~;$U9I
zU~c(9Zu!C6R(7bM{BU03Ku$q#RzX*0K^LO9)IwCQL2~PiPHPE{tFeWbTEZ($q4i4t
z27#fGt8H`_6cQ5_d&sIi)Qukcb`O2GtA4=KbdRg?)PDWxJ%;;t8XnwXeq@{FKO5~&
ztn|OIEciV9(wg84>w@mz6mVvn-?2Tm0k42QZ%d=6s>n@*9YFCetSA@Xhyz|>2V6p2
z_6P6Z8|XBby+N+~g5CCqx*rJha0&Nvi}dk`X8FYPSjj?enuwRi;U;qh$$U{NU!2Af
zCo_Zz-n@8sZmg#u$(^6##!GeOXJE&W3pZyUD|;6s4f=GuIVmpuY$9LJ&G+CI`U*?g
zl4@^3g$uWA7o%j0cgcE>l678{>%8k$Gdq?F&n%E#SfskJP<LUG@gdj}-J?rYkFVFi
zyiWP@TH*7{*-tKJJv^U%ZXW+Er94aJzD0F?x%Tm!`@SHRF$^Oi>%hGe!z9?TvGWUh
z+cVSqPR%aN`kYc+n6JFBTXu3Ue_)&Z#CmigY-g#6^Q$^P^nHCgaGZ{CYy^S6hhlP8
zPfzRV8?vb@ND&pt5+}$oDLjMYmy=DyJIBWfWf+5^SLpSS^Mnj1M$-nzsK`$!`zMt7
zBg*>G%7mYGH{bGpfOdazO7JIG`nPp?FtO`;Ym9MoRx(D5$7uc-&2+4(1M)HF2aklz
z2T*ZFRw$#PDT%j33fw5fqkG#;FCU!Vewt=JPRpLB84pdb?f#J}dW$OmfU5tH>iLkm
z|C4#|{Ji7FEpnmH%mRnbQ7FPX#ZrUc>1B}4dF9r^*QVyZHoo(B^1hHCqTq&d2zz#l
zJ2i!keNM(b>I4W@F+_--GTb8;u^hNg@5c<qxl3!S-`bG)s4(#%easo^=&{F!jz4;2
z=#jz07X}8;_YU0O(|fk7_e@*+$<~e&?H%{Fb)Li#=dEogTG~$3HlL_&zNfO`9w4W@
zj^L-F{&+>hQ5*@DYKDrc1`Emu^UM44O8fIl`U*-9<rNR+6!v8oc4ro%^45}++ZdN!
z7nND(A75b%E7gUR>4IzI_9}^`R&1`37^_A4N};Y&q^}Vg>iPNxV2P=2WT;zx)onhi
zPETc*r>e(8)9<Z0=%GKd-*|Md`Pe?o@dK8lE|w#%=0k3#eos?3)7s^0sP|G7dx$dK
zc?s^UNLSwwH_sq9*Fd)eK`#3O4(#=Jn(Kb@h=*&4r(3AEM>xYfio=W(aFfJ>REap1
zFH8_f5`?lOu_8sNOySFu*pftFalDr>(Or<}!cX4MOWDUs+r`HE$mETlWE<W#pA_s)
z*~QAh5{P}g91n4!kG#}dR)P~$8uqa=_xZxo_cJnfF$;GHySIprY?mI}DZgiz;@+L|
zlRIShZkL|iEI7W2f6oT)Jsa7_H?WQoW8SkuaAsccOH}H63o5=^*7M8C(^pqLI=uS%
zF)-F3JiY&I65Y9;r`h+?TZ?~LV0(Rw-%IYwqh7+c{nFkYx}jBu<COK@#i`G&Z~p?6
z2;nDm&?HHzQDrgFvnG1fguQNXL0~pb(|j@=2oBpA`evZOVRnFwcwn~YPkQI4q!_T^
zB)y>J8Ws84V*i&H8DCtee`$s7jrGZ&?QH&a&zV2P-_wGdBw=uy9Ce!*!F7q+6sFMS
zQ!~=ZX~-d9I2^;32soX6By<!U^Khi+1Hut1X=^XiOR%lrm}4ht@9^S#N3pR3ZE^WU
zGAVEXGk*~MK>zfaOVo4UQ}13}^Xv7k<Je7hTaJVLDo4fdPE7k?#QKfN1#i-GUYXd6
zABPVFArO1}&@*)uNF3^XD0|G3U9nR$@QS!!kb4B&;1zoBXY{rUS2opswlnDkVd7)D
zn6uQep+|;}K6LcR!$*c5bRLHeKXmxWg(HU^Jb37V!v}$-`v>~&@9sI(-Fd3L<0Sk~
zWBa}Jt@qY8AFpnN9mTPx=~z|6vC8@*6?H@9wTH@T4wqIBmR28vl~o-rDIY3==P4V=
zFYV1K>C7xd!MZs<yD=!S!W34l3o6wHR**FZesya927PdoE}%hU$7;kn?BWHM_=Xym
zwwk4_V`%DqHO=0dR&Pz4x3-g^@AfwIdYcEmtOsG<wn1O(ps%IJ*WAT4w=wm2eyNWv
z*GHJ*!;ST3g}HkNxV!tgyZU(?u(|ECxb8K(>@nepbC{Q_-P_II*CUYa6Ut{rO89XK
zagsuoB9bJE<w-JCs$82c(`87t>0)iVK$F3grTIwHJjBSI(hu-5c5~8!mQ5ZBn>^$G
zvF+aRdsy&1neL(-Z)v`lB-cZf<0i;-<7T*W(p=c7`#2f<WtDpr_50OLF4`71ZM&<c
z%|+REK+(E|UA>i4yN%nhliRqH+YH;zZoG@rxkPe`Qa?hO&`)@i3d8a#Fqf}U*<%Y!
z=#35Z&RW{Lj+QsljFjJY7+?3cJ@2DF=&fjWmACIv^>5Y=t<oNyZ#Y5uouR_++ur|W
z;Ma7HLuNX7n&`L(BC(zx6VIYBaDz=>^P#VK($}%d3YQnQ7(tT}Gy;!b=soY!kYb(F
zH$s*DI5+s2m6nIs7#?10d}NK`(N%#@Z!CCk=fR&{p1Z~Sja2QNE(?i=4E_7jza;d8
zgH(n-BPoUByN<y`NNi4E9T+A!_RLJ9WwddaUi$OIT+FLt`#|?is{1eM=ru5mT)(0v
zBQ&~`B1}U#m<<>QSoi1j%3mi|{XVt!FTz-dw;^+~j*2!s80c`zew?M=otXE|EcMFh
z`tQm539K7JrwC<tLI8y}i(~K}uM^%I#S#=45b~knu^;Oh5fXn$yI&aJSM%M0gjXf;
zj~Qa`r%H>OOG{cR%i61}x*BTxni>b&nh&?N9vSRAJ<xypVE>uHfipPo?>p7sb8m0Y
zslLH;-2({f?rZBl-P(m%?p_>W^)1Khnvd2r9;vE7R8c!vUNhjZgB6uW%F2(FmK`iE
zA1JQqE3D|wFYC-MMz^OWDy=Rsrot9kZ4RwA1l8yQ>x?0dhTvLVP^A{KZ1xJ3UnMrb
z3r%G_Lz&QAjWIZ$xq)YC=2=?!)>gi?ooDOd*}8bP9=>0<<D>%*@+}znse^PPLzc}D
zr!n~POm3tPGsw%+>gKBVa51_cFuCqCxa`#**rS8t1Ydf(TD;t>zMgirZ;*f!E)~Wo
zr3p%Tl0=>$Q6$RMX)0a1+L*00W-CnDa$~kwpDWPja@BcERi3ZBz(bsSfR_dEAV}Nl
z6TitbW`ld=7SHJIJ~6vl341w7?xHlOxe1fqc!@5Y_yg>?J)Bq<Y5D<0wwpT7Q&;Gt
zD<P&S@m3cf5at{Z<hk+-TzL5|oc#T)+`WwK?Tq3(8O>{XLwAVJtx-L*T>JV$!^iVX
z-_0{$Txz{^cgWQpp?`Ws{U(k3ndkquxBM}V=00C_hnJ?#L)k>osOsCPI((;Oa6Yf+
z?y!fJhCgse{&T$Ze@0)VGp6X+3EDJ_Az=(|T<6hOS@d7t^ksMYvMYTZJ_oZE5L5U0
zgLeOd-USI<EJ&;Ul}i6$Wza)+8INz$AKz{{wb5{5mHx!qsK@TEe1F~HAACP14Mf7N
zB-jA$@*1DMB61M^Vdyi8NC_C+mI(np#+bc|TCm&ft-VtdP%yje2E72+`dd?EvaRzk
zYTz1mYGmHylNh|)f0^DlOz(9p8i1#A^6SVNR_Ku2BAr$wRC`JD6uTlxb>S||I6znW
z!&`H(s_~WUYrZ9;^^hz_rwC0?B2IwDsB=5zEqd=?(_Vkjg6j@fd4={t0UGs&k7?P%
zG`Hpt-`KYl@sAr~@1yLNNV_d6$Q~OKln@b;92Jok7nK<ola-lTlAT#zm|Ig;(o|d7
z(OBQx+BneGa<IMgSX<{kPV4AC+0}chr~gdf;MwlJ)15tt<&HPC4%Iattgh>?tm&<+
z>Z_?4s;U~QtUOdvF;G$2S610mT-KRi(vervm6q2EK2A((TSP)@AZR$@P5zL@38^py
zml^^}v;oCh|5BA5HLD`AsZeGwliDj}{#6qH8i{|M*uPF>uf<fi)LJjIHQ)%7Sn5Qk
zYO%3GtS@CrGg#sjwjhqpjbyWf8NPnrZU*=LDt8yH+X0QsKGlJ}N~hu6b-&uvP4DGy
z@bxmY8Gd|DphOs^kVYvLaWYw)R1q&zB`b9)N`1P*m?1Z1$c<TYbH3DEATk#54MjX%
zDa0Cm<i(zn0vA#CenG}QZt5;({0`sP9lp`K7*TszF|L9{n1?XIT@dTWk8$Nk9^i)W
z<%YNj!(Aoeo{CssO(IL1!q%m7G-+IQ8cUYKkR<u=6TH~5UJPR1%s3BbstYsUg<tC_
z>-5kJxnQQq^uTV*BYQ0W*=Iuq@@XI26Fl3a3cm-H#*=d8LAkz<p{w;ZSL5g@ukjW)
zF{E9d!v0;H?z@<sYb6Kou%1~R{qVA^r#7~~uX&wLbBKVVQ*LzJadBm%p@Fh2`heU~
zn*As3^ErrH^!Br(tJ{8?ll%F+s28_{p4nmS*rjc8F?R0KbgU9K&BIm<+vzpcpX|6W
z&ikI0BTYigGkr@)|D~Wu@C$(=+{<D3{ZcX~G9ra@oBV?{eZ_NfWZ(GA7Koof_UT)i
z!k)ag->G`Q1~O5z&_ur*Lz*|}U05hYsNblV-P$}ovTb|@-owqQM2QnJ9Y98LL81d}
zz;Ii3Wrq6bI)$O}SFbGpnm`WEcQB05@VzZn2fVHO;THAMx6~tFEq(mg9dFaS-zMus
zxgXHR*J<ko+E6<tkN-#$`?x9g3}w{^nhk;f_h``tSd2k|*6`4v*qE@S_?YyRxU7_f
zyv&TUjI8pU?8@Bynv&wi^2)Z_ny$Kr-nQ05Ep3OK$M&|vZ5>CtI*)dDoosDB(cEyN
zx&ByV{gH;cL$!4Wfh7W-ik|$EuAHKt?1H|uoW8i^&ZzkIsQA|KgvOBAnxN<^|ELOk
zWF=wZ1eK@*3bB+IrVK2F$pgz|0i_aqvDB{^yWcdH3bnahVJ?%KN~MNksV-ll&ErW^
zdE$7EAeze!<*)-;zBX?+orjCk<AB=TMeVvDXi>m$f-gN>HD2yIZx6k%myyk|@;Uwz
zVTeK+p^(SO<S|l3tW238Q^(6SNit26RF^E)rN|6fVtuwymnYB{@U=y3bqPaJ<SEG`
zCeQa#6ksFQ0e<QpcKkk0+yP#^n<&9e6z3_9^%O^Yh@#wu5pIHTS3!si-``yr=q(9n
z$fG#Qc&;jer%2?>6Zz!D$1qrtzCK|L?+7*{n$L+B^3!<yJg%^WBQ4`7s+gJ@mZ6^G
z*TM_x5C`{&0=uOCJs1sCTe|h8Zi~6sVC#a$1IMq5Ypi0)%GkVeKDSQHYZLOidE&!6
zeA*Vd)KNYyl<M$;h==bgeP`>zZ+zb#bNl3$&)2i8pK1OdwD1Zoy-LfkkmeHOUE1|Y
zdhe-;EghFv7k)cG_Pu3+PwojgbU;^iKwa*st9Dn{Y!X#1=GIf9?)foKFRJ)>%Y_jj
z2OxCda8yhj4$)Tuco3l>a)ZN>(Bl%+yC<&$TAovr*z~joL)1jl*0HLf<~mg}Om&Pc
zy?<uq3zK($O=OQ8<7Jp-7y&^E(j~a!J;0_LBU*gOzR12x$0PzK9njsv#=a%jC#g@a
zQlE@c|NV3Em$RS_?SZo*6MoY~5XY(Rf(?q_-l87*gzEV()%G^k{W;bD<C3Avn@>(5
zSJd{?rlK3_=nr+VPnhFQQ-1mYn;`&4KVzWX5ctp0thSjnHjBp3qO)7IcIVNo^|RZ;
z{rn;W?9qXNv0)*JQBi5}u~`WTIhpAtSy^RyIhBQlwdJKvH8pThJuOWGwN?GKl>@aE
z{dLs?^)-X_bpy3EeU()`08eRoPf1ySY1!fYqQQ*pzO;-!a?I{b2CF)~B__2gD!Cyd
zu?{E-h_19omVuRHjjFUnR+z%e45203kRokBfhC~C>R)29mzeyD4c20vxlm=wR~qvq
z%2Xk__+t3{a4shZJ_L~Pa@9J`(?tXP;0P@FdKeg<Mi|q}#P+f9n0_K|piCI5ltifH
zQ9zDd0gII?V<gI0u_{)ij1ef~M4D8dI)$T3C!0?c**@}Y)E-;}8M~ROZo*uKvXr4j
zH6nXIC;0#m5n76;DA`jK57+=Ro}x%kQMl9G1>v582p@3_Qx?yWCUIe+B(5Nl$4TU{
z;+fvDOpj=`M=akvLBdLva&!3nVy?K96uwnCa7{c*k}xnq5|$(hOA!a9$wTwip;an-
zz0Tfg@IR=sc1r`BME;c$Q@KQ4DiIgUxFsrXwN}t7Vl}h59j^SIP5hoEoGyyjOUaL*
z)k7t`LFJ&P|LfAuk#$Gut!L<6=jeUs=>uozod@a7O=DXsZfwc>eO=twYl1(z+wbZ9
z)^2ZOnYXskS6%F*M8&^mqqt$Qu$9srrot{PtbB9(`9GxJ5%qUF(wniH5_*_V-{g~A
zoX`=R2LA&_4>ENSv3HNs8=(<}In9qpsHbMAqM11<^n%LC6+`2z9>09|M_2aVoB$(x
z(-gE2FwIVyzxWLVful5Mc#;L=%wRXcIJ6?R(a?#XS~D}T==vD-<qZlOPG9|f&L`uL
zR)(@1&;s!j!XkB0!hswJFVB1Fck0lGRLgr*!K+l^+f>m<RPpB<ihuAe{7sSYlPctG
zUF>7V_|ud^s8otn8mU&VFqqV4o8HfE^barx`kMk^IB}liK|h1tYVb1~{7puGqrnct
ziP;ojwFLRuLIMLKU_pVAA;Hn%VR4ZW2{F;hvC$bB$>o_TRoQ7(`B`-(`OW1etrewh
zWu>h()x8y!y_My?6%_|dO9u<``wH^=i%SlnTAi8Kou1Q~mfZmZHxz+fY)V69Qhj7{
zT|{zycyeuMLQPOyg<o`;J-j41tTZIFBq+2jAh^sgummVF*^7*}A_dZSb*fMr#}`HN
z_#rH|zpszg%hTlJZiFFVgW-hXX=eM_IKF-?Z!1o?47-3GDCUL8MG-1#v|16XR7T4c
zQ81Z2QY?=U$|7NW994;M3v78ZQ=Y<*ren#L8!ve;D|!9i$X&h}UJ}G?rEdJ3ee9G2
z+!Ry}JOs(^f_M_42_gU$cR{F!Fbw$d62<z65*U&cwj`A;PG^hK*xVF0JDJN$<@hGE
zJ(F1O32e6{fk&#?J43|CW^)VJl460TNMg#D+A<{mNz%X=X-K3bI7%ECtBTCk$5bmr
z>eZoL>fk}lqH9806n+&-V<{G{7&N77QNCJGWRTQo#dQK+?E$Zfjcyg|yzB4ecCQf+
zE&zKq=#6<%Uo1`jX-)PYcNAP*TX1!C{$HzdE^R3M>(12g*TjCZG47-7vG4B)eSKTN
z^ZWcC@D1qT2UhaTWein;rzqb=P_l(nzJggkU(!6se25A;PNm$xZs0x5$8_vXI-Crp
zkICulj@IfF9F9|{2*DLHau_aRI1|j1t@(D6di(0U3zJkbJwJ?Ik~_1mfAX$}FK_+u
zvg^$Wu<4w$4`g--qBE03sZ=pFp_rcJPE7ku&tjYruz}tPCYnH@SafZI`UYZ!6BJPK
z!3aiBp&S6506C~xW0f+h2t=;}YcoH(Mcw-a)%reF^)6NVK2iBgeREmj+k2Bg6~=s~
z2z^Z(_k=D1qc}!CtHEa0S&SN!L2dlUEP5*pkHbt_GroKK30jL$XN8&cHnYLcVzfKW
z&l32L*{wnTw%|Z}Xpnz|pCQbq54Y;V{LILp;v+*-Vk0wRqq2#`<R!!wB_))lrPO4m
zHD+hD=47_z<aTH0_v93G!?N-^)3REV(i-Da>l4$O6Vn@UOh{{pPiu-#Ylus2j8AEd
zNvMsDtBHyw78zF;7E>1-S!)lg(pd6P_(HV96-V<WQ9N-tiyOpX+I>AOOiwG@+mFWx
z5OG3ef(WHJS}l#03nQica4|Pj$PUKiI9G_GfE+j+xjbAZ3zNt~VPbihL=`1b$MO}?
zJY_6LnaEbApsEIEvY!>Z!#i$=SL!zRw5@KbTiudzzL%Ni0fvqw8}$iP-n>OgUZQwF
z#Y+_JC5-eGM!ECiJorgo!W3UgI#Zm*7H4rpIb1=uP>|1MWimZenQo~Jmt@BNB%Vi>
zh*Q88<uR2cEEE}vMdlKTtwio$st77q235&}t7O5o(%^bYXp1DQO&Zp%4C^umH`@a0
zt$wveOSwT;tWo7FW!W-Gj#!+}<>z^`^AGqI?O|2y<<{>Nw(gMjZ84lehv#m;hc*N}
zvLWO@8^fR45c%Y$h$l9LKD0IRq5ZKBxyC)<8FQ~!#E^IBATy+c7hEq40Bq2$F7*|c
zy77wla7s4wN|teQ=ZcFcO%2&$=GV6{;|b+sbj2hca+B2SY>p9H4k%$n0eyTYeFIE>
zXI>6jrVZbXEq(1O_24B^tP7rAV4;^JPHpU(-Tv_C);BNj{p}V?UI3ee#7#OdP?ZqO
zOe^UzHP~g-=y;-bJc>ANE5<9xq6&J+<>|TKO;PWUQqTN0_k&4LY`icf;n+CfF*61}
zD{>*kY(Os?)idZG-hYv5hJNKsRNC`Y((`j-pWB@91~=(LS={To)Mqq_=cr(NNT7en
z|2+au7L%XzNN{DeIgciT6=u|1{%r=GS+6zeG)AZ49FO1$fEKbQlP18Rvg?(8TBV;(
zW!I_vwMx5DA8gWxSxu34Yh0i`IV3PGEGRQPI43;3Fe0KPDyk$lwmc!BDk-TZIkhe&
ztv)TI5tg3W1Ot}Rvs==$+cI)mvvOK9vYM0An-Wr+;*uMq5*mVIYW>5jZDEy$fMOKB
zq`FjzDnTfZWeY-lS^gYGAdeBmX962xQhuaD6r+~LYvqYJkqe{oh?ooTggd{blEkQF
z;R;!hLKY%d1S=F_N_B)%8>7_6tBlEVL$cJ6Ceo+#)tL-wD!N1am<ik5Q?|Qj@9@mp
z<(;*UnS-_6KH>s~G#|+Ek)%5osV4hK5`84`-lAwvVGLZ5hcL-YoZ>4>XUa0!vMjDN
zhbzt&;j-ptvV6fhP4#w7^ga;Jc2DDTvf+_f$~3+vM~p1aSSdHxs;rG_d#fg}OBL9z
z2x^xFwMc^-MWOXDpvYve^Rw1k%~eKIiOyJ{(&s9)nNn?rM4io(6UKulE6t6W<-yAL
z=9Kves=bvh9@-8MLyxCr(8KSrr~fg}z~df4_jrZg!w5glia5d!>*WNuae^ATL3M)Q
zYQDWp;8!lN)Clx7Tt$tqsLX|%w~Ldxk(azin6g5aw^&t8Nt!6x!8MVuy0!ewd+c}3
z7h|Ef2!|DNfga=IkjdOdh8IY74nt#p=g0K=rzaN;U0+afW1j8SLc_$0*wL-6^u7n_
z-LFm__<jcR&F~UL@4*2cT<#1InWt|u$FKWNOhVCS-#F&AFgZI0IU&p}Ectb24mN+h
zF-|@4$NaY^)_y&WL0uwQJUz*nL6a0aKgK}!hQk55MtOtK0=!<F*mM5!hGQ33G=DTN
z<2f+AcE`Wy74xh(=}BecLzGc(G8)Zhqs45t;)ruRqSqTOCMzDn(PXmVaU4ws^S_P!
z;Quy*&Ir?M4bBtCcP$oUfJq-<)cM13Le8~;dbPhoVp2*iDw$O+w`&ysI#r-f8LU%>
zX_aAGb)-%cWz@%7&4~f_w2+{zu#lXHu)N6d{HTb6n5d$-=(2>kvc$N`goKL3<f??E
z%J}4(gtWS(^p>Q|w)l+pxb%+Dga-epDsxb=&M!}6%0Ma)hr{Fq3OJ!6R;ZXACgFt3
zc##T0v`Q4K7RSMq!WcO}O3ICZIlrZpL@6Y}a%liTjxt!K4%6r&HO6R-DNb)o)7a9L
zmJF#WL#Rc1lj<o*IKWQY?vc8~BWJfy?mlLo8#~{VpYJWo_m$-PigOu~ES5BrB}-#U
zlNpi(A5;k7Z6whi(r7PvEJG2`QYCRz$vjn>K$#|1Wr3H#<|g|x<2^m2-P|I)J)@bd
zIIcL7Cy$b<lNH)*wZ2qqsMVQTjn*z?|2n@8t-S@kFRWgb1((W#i%>){nJaDjO0&M)
zU@TUf3l!EoxeeT(Y_TOnpig7T6MT7b_!A690+*4-V`m5iIRbHkKw2tLRtePg0$nRO
zsa!+1#CBM0JtVaB^9&svT?<dwATl&dtc^U>5-p`-YpulGB+@l-WR;$R?ET!tUHr(6
zoUpt3@pnrLm-Ff=_ZG_SC?$AcPVmDkOI|r}`d7(&wDBsb+TCuFp!EqMIPh6NruRNg
zZ#+!T&%Z(id{2o#qC{A0_w`-n!>;$y%(rOipJYr8GwNtr?wTZHr$}|+rT`xJ8K=1r
zDLGOrgP_m$QKC&ny*^Dn57Fu?i(i`t%K<YfXtsg_E~3X!lVD;>2OQAJNl4)svBmxt
z9eSO%|3YiNqP<@jy`%TTWu?!r%YJ5C;$z+kkBXC@pduopBco!XqGNFkkBABh4G#<o
zvHJ(w{QO~7o84mdGnuVMlf_^(!+;8>IkEANS<N;W0UI#p02qOy(WX%v)C!|UY0_&D
z*V$3E`v*k^cp*&yAOsf#N2E~@(<*}0ia?D#P$dsg%L1I|YDKVK6$Ue?!%f;Ki$2C`
zh_xBxZ00ztInHd0xB4adhh&6A6-C5XMkUrpCDx%9XAdv41r?e6bM%%Br4C)oD83+s
z#9)7$m=hx5hDv#1FqJr3DT<N{BAj2sWc+ZsAXF|6P)O}cS)fW8s8NM#wBcHV6FCVg
zW1>u-B-SMfR4Ghpq9;H004sKncRHxnFtk}*SXpk|ECgv@!fdV*NqjC(nZr?J05-sk
zw=mXQ673}o_mqcuD<fFyXpT0XuS*o_QY89RnITi8LNPFr#gFx7M7nweyLpEAFhY^4
zNl+PACTrB$I!&=&Uu`lrn$7J7LyO+ngk6zZYmM4giNd|ozd&OvFz8B6>Qb$`K&i=<
z>9QrpEQvK!?3W=hCgX4MRS8^iESD3-^^N3tN5DkB5kh8!h#f28rHBL>B2k_|RLB>X
ziR85+dA&ed%N13z1r=;T8DCN*Qd9`k`9fWR)L1Sv)rfUfTxBu#lDqR`P%7NT3f#sF
z+bPJz2-7;R*41u@mNQSxlN_Rq#}~&xv7zyExAS8@uh9qIp|`z7yI|<!Q`-A2q8H<O
zoZeA$3(Uhg>bDlipQA*NQu6;$fv?WZ_-JX<4|hLs0|QHP)WjX!P*lU9|As~loDTrw
zEGWB7V*{*XzYA*3H|Uk0IJZxaQ)jL$ePIeaSfNsnLIT=H(&<}RVh1W3W{SPwvOqN&
zrwuphun9VNlnxrh(n8wjt<gOX{=9YY!%Zd6?M-@I09l;u!or-QqTJ$Q7|t{D@=~+2
zlQT1;6BDBn5+dT_!eU~ABO?RD!~H`;{ewfSR=@v?5I7d#1V^Aqr_pQGI*m%}G_A&{
zQ4`Z5cOrR_AMm49nNeE!2SNs|U9a&os{ITq64asaZ8L<J^}%LsAP}XISrihpT4K>i
ztq$|k%Ir>qn^MW`3b|D-v&j_!N_Ci07lj~7u8lVOr5kK%dUJ}-l&CYrYcx?RMVL$+
zBIX5&Ie{W}z`qTTD1>1EiIf*C;gTod2$S*yl@h?lrj+?B6?Py;0aJ&_)S*&!xL6$_
zP$9#K=E@To(l}3kv<o|SKQm>oPud>u)IC1QICtk|_=>YxvK)!AL}Dxy>2kQrR9{If
z5*uG(l&>VpM;7X%2w|uq*y<>rE>>hnkeHI?rZkl`8xf#Tnaq{OGq{oNK0$8Yfj*3I
zP<|A`ER8fzuPD-~ijn1;3>79*4a{h+F_>%grdq9`8k<;TnlibnSRu=kNiv0^G*stU
zqIi}h4jzWBNMK9jc*=OOGEpRp7YJf_%y5o(2*=A`;A<23T0~5{kVQN*pBu&J#PYcD
zEOsJ?lZHLxn5STKvS8r0!Zk@{SyDx&OqDB9=ZaK$xGA9BggXNMiaR$F^w|Tfl)cP?
z?Ovr@z3MPXdxxNGm9loF>EL4fc`E27D(Vv|<M+A6V^se2g(btQORufT{$X|WyUQ$3
z&E=n>+&VX?POOz4S|mC|iH}m!dnm(cD(Pve;XU>Dw92vehlo!Q0j*i+W`Pg7dwdFO
zYIctk!3k*4?Y>6uzeKNj6C|A*)IDR=k*iCen7sS#an1-K1YpaMa{QKddK{V_&dMHI
zN%)<5bbhhEZkX0h62hzeCe8nuc75f_=D`nl<vh>Mc!^5M%T3MCOD`<QC@ORs&eIC=
zbIMAyOG`3~i*TNto0FKG6`z@rn4X@Hn4B1&6d#)q8yy!F5gid085R;A8XRV~`TJSz
zKop7e9Qc7p(&-E;rB12PDiu1FQm<7Rb!wANZ2@dla)Z<0iyWrcDNH)0L9a3xH3rln
zkV-jf6;=a&L}Aj)VHTK9X4c89TB$`ZvEZnYnG{lkRBVuljZ%q4BJ~r?{6!LfxiUnd
z3RS8?)yfc!)8xTw#9Htkq9B<dK+5-*@a#^*`M)0#bNz0Ai*Hki{gebYelUgHPb#-b
zWp*L{ggj6n3*yVdxUxtN5+`XKaup9=+yQ3dK1R|WpM<@Pco$A0Dr_8OHuwZGOQFn?
zFEi%|v}sHko*cmtN3x`m3|Sac9?nuka8%JEZM@Wwq%fxgGg@1&!jL7=068ff_$5|^
zCnJo>ijna%RHAIHEMKq8*QxV#+I*cMUvJL0_?4Ob$_<ust+7m@DH6-_dBPknKLb2e
z7B>$5%H0e83fK$w^a=H5g!{0<xcnHgG)V@!oj4xw;IboltY9H0NXQ9530}kv7V?4x
zoM0X|l*<ld_yn_=;XH0MpBKmHC7_8!XnWj5wJ23BN><2{L8n7ipUwaPA$v(gK~xaR
z<MC0jB(V%p3`>~k!%g&MC1YU8lbN~Co7_sk+}pxQyo;T(np?U`+`B^a;BxEhtHZxr
zpY+$-_)A;T|J;{#aevaM`@&z|ZGCXN^4Mm6`(}2<W_HO2UiDf*(;|KcCGDT%cmMq4
zXV*4<y8YZ=n50l36~Q1dbORmYwD2&b-|x`x#d@E+zo9pLM9;^dcptVw&QOO(S3fqp
z;f)FYFqyqWvlY8gEz^+c18gvR0|qoSab~er4NJVGqhxUy6t+Om6GN~4XL|SJzwN7e
zhmrL<l~PcUT3DD?RFqy^oKaGeSy~FqDl5yatb}Ekm*X6d<GXlHT0voMUO{edeojta
zc6M%NW_CtKR(g77T3SX*N?KA<N<u<XTwFqQbZk^qOk`wqL_`#lBAp)I#-LW|VU81E
zNw1RYoTij%iODr;nN}rLDnufcRHT+lwQ{*mq0lP<G9b<0)jEYrr_dM`TBBTVkP*{M
z4KRbmpp)ozay_c&DhYm%QKPVF5%Q^nU=A1Ir*fEDV$({j2C2;;w`-*V8i~JJY*&f=
zltSA-Cg)pZJhRhq4pRuMFu9ocI)%)llw0A}02?4vAhiQFd}$z862g&%vt?1vYXxzd
zE1H|^I8@d4vtz)#1}(r>n#58hi}c8XvJ~(krVN2Po-K=DO2TktOQK<1SuBcvB2|h^
zm#H%4YAl6%OOeJ{sMO`j)p-IG*|<r*>;yJ1Q6@=LDiTzxB$YNrsY_DmljP=PxiwW~
z&s6$lC~TP^F_4~;ERiXU@!`RzMSF6?;j9k0`MVtOb2|{=?HbB-kAU%g<D~plnK(@>
zN)m_?!B`c`Qu(4}fshzDtUN(1hZn`+gn?$kWcjl>fqY(wfFDYZoM0h01o@$!7iZ+h
z>JZ|}!o`XZzB~l)3xCuXSs5!G!?PT24x69q&5L&B2Doqo*s64pU$^+CZ5LJSH}<&(
zoN@_0cOc?H_m~HL5*}g3Kg5i@pB;3XZ9CzsIpQs9V{(h}d)$0ec6z6*_e@{zkw4d`
zj^g!E;**%aT2}Yn)(3C7d_a5uMl;4RojQ$bG7(+_31E~4{cQ2;lSJb5=p<Etjq04D
zdPdhgNN@b_m<Yr+vY>7XtcW1T#sLiEG3ikT(YPE{V%Z}ZB_KnuMD+y=R1p4Q6ymK>
z@80iya^9wr3JMeR^Aqz667upA^2iAc4?4~HrQ^H#u%z7ll+2ve%$$tO-1O|ctnB>E
z+=86k!mPZ)@`{?`@~YyJioBxI+`J+@j_+pV<d;>|78F;cXXVEyrANmkhJ?oi1xNS?
zgjy^?W{bbk<fqeHG+L7q(2xTuN~J=kkc;IqkxVL-NpT`kN@XgUT&vJHOs9okk~>Uh
zBuAxQBM06z3MqbrUZpUr6jrCH<qngY)nb!cY%oa82INjM8^SyQNhP)_MV7zKB;y<9
z0s~ATG{Tf36HFyG<6JH#)eFEzrLf&bj?5~S5%BQD{#<b|7XT23Gx_0OoG=ew6e4>!
za+O55aKc?Up)TxT7*`P|(I=~IS!!DbTnbMWMersG=ZK<sqG-N2P9#f|DpORtOsz53
zU@0(|i?oJ(wJu+!&6CKp;51Mk;0fUz;*_d5wI)HOOH}9+Wrlc}DM4yUM06)MCW?(o
z5`Cgbo5)wh0*y>jB;r6%L6{pO$i>6Y&DGD_HIV5Z#`TJ1xrgz6q9vR}DL+{(NR)^Z
z6{>WNK2Idc5{feg;xyogCx~}YlR{bS05&Ip$0Nq&`tjH{F3ZYi+Ysn#`H==ev`!kO
zR)$MeAp%7(VB;f*VhhvZXM7mx3^r(iiQa-R7p`eP+vLUy+3y>5fEmA+ld^-Ax{Z_J
zs;%UQcKG_$G5zXU)&{<{Meg6L4;s?g4lA_nMqRZ+l)+>~xqF4X_{5?owvk!3n$^6R
zcZ8Cjp-hipKk<s@4|ZJm(fjQ`8Q+fH{qbKLza8K5)ikuNv8i(9qZ1HuUfMrRrF=~l
z{Xtd!vG(Zrmj6she<$-12*B*)bigDTt5%X!XOc(Xf-Wq^_@q<dG~8B$fhsga?!AX;
z?@wsoW0zeEKB6KsvLiDyqcU@%va+JHa$~adV6i#*cpMMLWEaHd6vpNh!(y_FVlwj+
zGV@?bnfXcC1u5BuDY->yxy7k@#aV^rIi*$kWi@%_H8|#1)Rt5?mN&GP)wh*5biyhc
zyDA$xDjGY;vA(sUzP-4rA-kvozbZN@BRoDSG$tt|Cf*hlWcIUb3`UtsEtD$wVwpf9
zhsl&WiCiO+Y6N1HP@)#gv@)ehp*Be*8o5HNP#7d~9lpgEDJ2qkDU)1oRLV^-nM5xU
zY9vAxaHUZ?T!+Mpqoast!x8Y133L*!R?62Z1bUd9uY)NBT9ruefE+QEgcu5daw##y
zgkq^hEU`<Z0U}ADKorR12ck8^7Dh6K(F{S9uOQrqALh*q_TmP?d<DU`O&G)w1+ydo
zTL@1YE|f+PlSGQ7Q4(-@lyOQ;vRap_HDu~d*)WYhL#6-`J3%Op$5AYfmCF;g+H`zN
zXUtOTGnCpixi(d%Nr4fXwkA!gN)ySG1ybZi2|$jIFw~P1=<e(1?ql=twtIW~`+5g3
zy#sitD57#Gj8TZ<P=dD_vx4l!c3X+rn6FW#!>x$~(L7E#iy7$aV`DMx9HtG-BQ6V^
zJwE~0&gbIX%HswJgrRr~B6$=~fp|5NEd%2r4n16NVWJm5$%CKh!6$uQcVQ%9NbsY*
z1TiF<WyS3AiQMBG<HAYy6lVBJbC}XxwzPmNFA~X1#fmbSqEfCZS7-|5nk=a*L!d}S
zN7!4C;zIc5`CIwb8$@k)NC%e656@R0TVTC!MdV|PGhSKK|NDZ$%S#VlT{|?g@yOU+
z9oMi~k&1YqGQ3P_U!cs$rrx6RKUsD78s;V-P!TXio2Mr=Q)7ymTPk{7Lr=j}h}NcV
zVYvgw*RVVW3qoW!XwJ_x;~kpy;Pj5luPA^7Xn_GF1VS0v5$QRwsLZ^`jNI_F+_2P~
z(B#{O^VqDSgzUn9TVhU;(^B$F(~BxHiYsARrPX<r^+mPK1ywDD)vZM}Z6$RbrS%SL
zXmeUgeQR-DQ&DZB({K)}Xlf~~t<Nv7%qS>I&d!TZ&x%XSgxP{4ErH>BTZq~eAlKQX
z8jD127OPBBjY*``3#4kESjm^Dgffj-p_3}~fEExXlj)o$mm@?pY1KA^E=Z>dK*^4H
z4`4=WmWd324GIrPlhjfp{1VXO0E`Ynpjx0)iSVFFBQfF$TDe)RFav%Pk)D{?ED>8J
zB0q`9E)@C;guyI+FjH{bSi%q%9{C>&WC?E@TO2@OBMIV*LjXDapimSdmV`-V5nv?1
zpJ;VSdP9odkgC-sE0l4lo{7cL5=jhTqfjL2bXg`-j^3E7(PgPL8A^4AT%GPTnKDD7
zOc%;h_!89I<9+xcUfci=w%vni^YFELGHgCfJA)nI16BkpK*S4`2*T0o)ym>*`t)F1
zUZ{U@fUVF1(5h2ZvUq%#$0CeArjM23ZDIO=>S~3z0dja83!80a@%%aBP|zkw|5qNv
zkw*hIEO|W29^T>@^maXjap+AsnXDk@AztMulfbSZC3t?ECnwgE73alD@L?r*GZMXh
z6JWl+iA-iPhnd3Rr0}HaTv;kOM{IEdh`CH&j1M>7gPZ8i1C^;{uc&T^sAHpOV3YLZ
z4#T6Hf?ryj{^g>aZ<l0!wLJ5SmB}BkjDBZ@|Eo(4&#f^&zfAko0_{_C%+FGRFJgh*
zy4EW$k4~dv<~K_eaNv2sGmuS>NoP?Zn1sVYoQ9=Q@^M-{N^6H{(Qh>OL)!C!TbrxC
zqzdYrVEJ{8u)Nx)+?qyMPIW_iS#3&jRYHCloJ>?kVRU9uOcwDuaoNSOSp`mm^MOTY
z<iVoSb8-I9x8eb$ypoi{^7NA0jMBQS@`jwsro8Ir{F)Z071TBt*0q#0wwE<`R<zx=
zs@BeiuAYXD?)vtw+SU$OO-p-iOBbxBxw8sSu5B$Sug@y1OwB1t%qWUa%L|T5Fb9RJ
zjD8Ze(P=;tz{3?OxB@v(sDS^0;e^YTvRD$4NXO@CxNIexE$4BRLcT^M(250WcqAM}
ze3h6hm+%x)zEUn!D}@?@7KsB(5`$8#SBNw+vBM;K)FH$om`NbC3PpYdHljeTFog5B
zaYdmdgmV~26vPn)u!a7An;*X9iv0zWU^pugIR=ZRAyRpmOc4&`s5NmKZM;UCh<Hq)
zjFrlwfE=kT8sJeV67>2^vpL&f&LIv*m#!p$NpqS^l}0A&WJ!EU0#6d>D+u-G1$uM*
zJz0Lh49p9xR)!y7!vvFv6Aa{l5`o69Nu6vrW`+fn1p5^RSn_O!ETcLZkApJ-@{z9x
zcu9Kj_rToaFw8ubmCLj+nP$L-EehewBL0C5sS3#BL9z2inM9lbCLR(zS0O<TV8d0X
z@U$tYi=f#h#xT4Bd?E})B{DdPK#r$(jGIR!np7ShQ9fQVK0c(U<im<(uwxkPC>AH0
z#fgM5Ik60WvbPXCz~Td<%AJCS?UMGL%E9f1d+)M6dPnfn>qDQrE9j{^Ef3$RJGW7D
zZkzi4z1j;q6lXWf&)%hZV2S2_N`H1<&|?d-UhsN$g!%D|`VZQFh7O`>e+PpB^A6Y<
z1KEB7LbcfmQd@)D8K0K?Jmr1<`mXA)sN9-*SYB-dEWfU~puPoG*w9+k*ilg5mRr-D
zUDcRb-T=$0XoRH}RizeH{9j8eu7u(7<bpB)#%W3U<!Qw=Fd!$htUjx}KD(m9X&I%}
znPoLu<+VAL^*L3wxz%+z=2h0_S5_BQ)t1!Mm(?{^HnvnZw*fG)=B@!)Q|Dkqdw*SP
zZ*@~wMSW*!U3*n?7kpH1S#?HXSz3Mx!o}pALIi&H&}e&bq%|PS;vZ`A3)UD>VbgN>
z3NByG<|-L18H~kIuz7MWSI*(c1Uw~=E9G(|FacL47AR4e0CIp6g-GWxoB%K?U`8s@
zXw+u4(ju3er4o}!4A|I2V!KcrAe4p(WMKjsjv;($u)|34fF~g!;fQVjHlElIML&lD
zIe}tn5KJNq28&0bii9cEQ3?ks4qgZ*kwywd09C9`pJKLT8Z8Lq(zOQUBxwpws$87{
z!;FJinJiKyfypY6#UaTdwLWnegCFG0vcuuHd74~3jJ^y%cpE+&!$pKlB;kdDWTcVB
z`I~d?rW}}`G1H<;)5_x|{7@d#A5>Rh#z}i+`54*0CZ@N+$5ThVjW9%jI8zZNkVkQ4
zq`ZMdjw4F|a!4^+oPbGKAO~a!92s~}9tTpA5NJ2&s>}s)U7kppDVCG=E{l`o?HlXq
z73Jj>!}N`3v68&KV_`nN(G1^kl=3*NaFq2BI&!&@9DW>AnCd0UbQ9+97nSUnRP9r=
z>{j+2FrM%UdVm#ro)LJ~+j5V$ZirMDqyrwJUJvo0tMc$J^^x_egKKn$7W$u}a{s%a
z{hMv)huLq^itlO7AGG>9tsEwS8Acvdvji~cu9B&t>*KtyM&(aT`854NWmZ&YRn}xx
z)@4`K!^yw^6&Nr>tghW@`L%6%HLbA1y0)VFc35#kM@eI6X;T-hthu|gt*^YL2j_SY
z--6*ecyex4Gc2#V#c7T&n{p}}(~7H|2n5*DOR4~&?8+J^Y+!{|^{}Gq1{@128}cja
z^UCXREUIoPsci#VU?nx}IIn2#tZeJ4>+G-Z8h|x)54Q9lX&*RR*WOpt++Ee!iBzh*
zt_?maGq)@$Ek7<PJ32liA|^F7GAS@L&Mz>^WDU_7?MkH))jOe3$>U4-JPD60=CFlK
z29M8?0!1ReLLyWEGjf^QSxko^p_0i+k|UCsfgI#HVrhU-76cQ?gJEJt(7!BDBo7eE
z{RA?b({N6{6x*DT`(Gvskjq2=ws0pYPAUzRNW%DnAb<xhh=g$FG@U6`XH0`Bw8><y
zL7fPbsFKi-aag><n1Y_HOq&4WHsML|f)J#^;c$7uFx13&++c)ZFhB^sU?2xj(JK=T
zsw9|N79$me@mT>JUr=2wEN?Ts4PZks<DvC*Q?ogCwlI({33tK<aT-q^2iS1siD=!B
z#;zn0Ogx9B07B^Sg7@tuj$kPh#D)xoH5WZujU`{H%OMF2Mx=O&AVV-&32aUhmz&CB
zrLY;v9FQ2$ko64%D+&!Ilp;`&5O4$df>4eC$z%*krQo$hS-$*Yjueg87NuX8CZJbo
z@09AB#L8Nsv|1pn;0bHEvPK_y6NK-!Nvl^YdzXg4K}CN-#eYU+eYv3O`xTuRS0DX#
z^{HQWK6`8bYm<ymY0;0g<X2jBkrw<!Yrmueo}*=L7paoww&JGN;->bZ#x^(^m=iES
zPF`&b08?1sQPj|hqZ2$%w7>ur;H09p7l47)bPiUv_u~XdoY!?9g4K2%EN|)k$I6<4
ztL}f}D!-;F7lnZeL~*r%4J@s&6n-PKs64B<;%}?UEUL^dsm>`yrJ?bEkW<@%d<pOa
z;PJ9cn(@NgVHK_2KvDa_V@~Tlc)aV-JspF`TKkSPbsq$x+WLoDdJp5gp>v?NrKh~E
zrKqw#6Cq`0ZfsIocuWGyefGdGqt&j}nb4(0uNG}307fEK2nA9;Pt4_sfG%P@1&^oV
z@-;kxo-ZKL8eqT^+xQYcM{g%Utn`<t0;K9doQRcvB85$)usBVuw91rrbb(}Ye;C@e
z4hw?G<bg1$ED-eu9A&Z)m_!mJ5(V%C{z7pO0y#J$Ied{OR;i6s>EflTSeS%NH^gAN
zL86Y6YT|(wl{pQjF{i0bsVY67m4NXfF@ilwBupd>Ck1785SJOi_O-Ko{NQ;&0EhDd
z2b^$=#F2oFvzfzZ*|`j`#jOa^fE>Wa!&Sv%`*8#g*br_wL5@Hc!&k(kOAChs-32%t
z@VkA9r7>j5a5X;C2rqFMMvU0X7@;10s!X*l2h>}&AxEjnM)OoGNfU|F9hNB)<$$ap
z<RuH)u|jq@Fe72xrEH6YWs-2rVxC1Lun7dfPY_oS$|sb<L=ig$+9@bMs`W)`Q;9-f
zC{^Z2q@ej_3Wb?MNj|C$Ol7gRs&tR0Wk=|9tAk!&WPf2^;4=#%UtEy<@`B7)shn4-
zg4gC$zPr5j>vaQ{HXgmU?cPzhhiUnfbnrv8rr{gpHY6%Tcm@MboUSIPx(PUO8ZcAT
z&|Tcv11o9j2BHcZ+MR|#uDGewX{9aQIO36#=B|qNzN*fF>aIZ?E86-i+YkP2<V#q2
zYahM~tL*46Z|lVskLw<pRbF*tCOl6`6)Fm_?6R7i^4i?;2B+m!H0D?SPmQ6drlSBb
z#vetThQHp@jX&))K-%d8f$FLjV%4pEwe17-od-J(9e2VBC*4E$^bVcqIdV_ep<@Hb
zPV^i;+TMS#wYR^utG5y9R&7I3c~x3^c2aVBEK290kycxPR%ehaR6?mtELWkjCy^T@
zGLu+pg2NGtECQifD6zr>Vv9g*;){$<<B4=Iu2AbR4Ns)yi**8tUMMxl6jr&?u2cu8
zG=XYukV*~aTd-0E<oHXZ{=|bx04)L`xFdnk#uNAnL_kgej4QHpg#mnV5E3MzJRFAj
zken!@#HtvnCQhzPP#TiJI@Maz^tKFxHA4po8B^4TWI1r9j6upI5{C-~p(u!<R)*9c
zjs+7(h{^<Pe=sVff)JR59}Ghe6b}MCxYHxCA;>|H=75br5(?Oe<kA0z4O<!y*noBf
zpMzmlmOLHx0In{Ht%+eMBD|%cK9W$DA`<K*ba>I;h3PDLT2r1{mkTP8T#+S{XDgKX
zGI=hvP(*@Q!m{NAO1V}!&#d5^<UAt`odXHqEa6*4JYe1~XNO8yq)RIhM~F4?pj^O&
z`ULdW5Iyo_2|Rg<K$C@eVUfL@7g*~V*y0h^<rOpNlYGJ_^Q=eKh5eb2Y)^Y~Yv#)v
z(%xL3^5LDypRbDhW_jcfYZLz5p831E^SZ3yEecM?X>d5W;BfWdwr1d@3$THQ$*%=|
zx&WKf<~|_jpR4m9%s64=Oosp`oC6GHtpsNHk+K%x1b`W=Y#(qMkb~y{T{YbY>-&y0
z4jgIdKU~*)u(oHgW$<Y0z)(}~A$W}Hww}tCu8OAaaulzddn%g&fIe7tYky7KL0EPB
zq0+X#{HBha`qtdKR^X?!t(TbN#gNz8*oi<9r3e^;Mz}{{3GWou2v~7rOKCI0ymnYs
zD~_#|&8@ZV9ZfyGu)4O6iiW0oBv%d1l~r{GCFNP!xrwQ1F$sy0v2lT+5rH9*0l`uB
zz$l9?(qIBzEfiNgaUcpmk<`Ky8rXcbFI(=-lzRG#y%@L6n<?>OiGA4;CRfJd$@xO9
zK%^H*j1rldP!P~gMN>#>alppU8T(0P@Hqj9<#>EEUu1>x1XecJ<ipl`vvj^36O(6g
z=1yEOlBgi`97U>dsU}9Ijl~h|9knTiAjgvKgpEp{q|_!TH1U|g19HTY2o#aIoIo}k
zJ(=4s1&-zaK@Ly^pF@m`T7n-zj;{%@Ve<nBY~<mnK?61*n<Js(NaDbp0CE5uV1_ME
zWh*i`s#Lxqg{P0_XkvW9#1jYmN<)R(Sh*<$!K4y-nmJo*&eb^Fj!d2@k|Jlzm#MQQ
zvSg7cMkojq^8+M&8=BE-kyR=92RvpO$0Ffa#7v9W$0A|+Nd*29WvJK?DKbZMP0?&a
zBtsPj`Vy!ZYy^;+WT6qnydtrGi8Q!U5niv1Zk9#2i=sNX(f$1R6Wrtn1nJK(6JKB?
zyunKTl$H3MckK7Raldj?epKXsAWHlX^>5dKON}^?Mj(U$=70N{-jb$%2L^ge5o<LQ
zNWh#=hB=<x={$F!r3cT10WjsQ2Z0>?tkeE}ypK4dj(+%@+MYuIPecDu<G|6D!Q)PA
zJ$O&sp%WcP?(G~p-F5U#&+)Up#~<js=Y0Q(3$VU>9_&4Sq5Jr``h&-72ai?v4^{RY
zD(@OBY40m)?I~*R&Z}>SIRnf8F+dC0CQh>bpSZLDRYa6SP!y@DEvP0|SW|~%UPWb2
zS$Td*d0|;4I=MK>&MyFAFDZkNj3Qza!lL6tBIAR@6GFpN!Xi>5qjI96a|42rG9?Cu
zB?X5k;}{T<U=ND51w{J=#=y*eVJe*;U#uaR;Y)cUHJh*UWy=66z=_FKas(O}@CF|R
ze<hTdc|sGLr)P0>T!9gO3`cy)<mi1_S|2trqi6C=EWR0)2A<R|L{6muW<n*JaG5Sz
zVTc9a)M*-Xvfi3XczV`UtvOj^N>UjTq)IefgW)c4L=l|J3qXCD!?d8jEadr%_yI7S
zz|g4q=Mf|o=-#k=Ej~Ub2FuLgSdrfFBq4A(PH%(402mJfRWeAfAarA}f~U!3E7O?r
zB%VH5XiAdUQpDC|fgzr!ixwH<(5{sj5-^AXuDZ^W1BXLW9Yv;4mPVu}mC0yH=`9Gl
zk`Rw^SOyU&9YQ~toNtryY%;E0iaB<6fRq)WVh1SL<dvfs!_|axj1eqLG}9dAtq${5
zN5awZv`Kt@n$VJsxq5X-y&<U27;wlKbl4Pj#2S6v9D9!`?jA$jNp1WoZTuNc{C(Pl
zbGpO}`oxEg$&VS5pU|d0p-Xw}Hf)@LxgB)<4Fd-TNNo0Zgx2pw&i@}~oUn0X=5N@*
z&m45Z=HG@O4fb!?)b$>ABFAZseFQ?y1IJnpo@hICvhDD_9f!|!4Bgi?bhh*8*{-2;
zok!1g9=X5$$l0dDry7r(Y#KV<G<0wCk<-n8AM5&$*YzK5=sVufcl+4be@|ooF+5m5
zaI|6YSi`}grom%yboG6Q>Uswn`UY#ddQl25X=*NQXey~kE2+M)x~8C_x}c&eudE`w
zxHL1bFeNK1J~cHaDJd>F4T%t{3z4xIk<qX$U?wsqH!LzEEGjKLDl;M`BPup0CLt#-
zDK9=bFDb1MmYiOcl2Q0?1ELIm;aX#mOzS6AT6q!!i>LBrN?BYDgQsC~)IJ=Q4^!pE
zQu(sfOpexRKn~#K%%EW4EAVlm2o$S=CF&5dIt(V&M#%M1N(16FLe_S|1`Jn|J>xdd
z)tag?BuW&KBJg755jX-oC_EsK0#inzz)X0AQW>dMMB-d52o#Beq@rMnBt$9-6{69^
z3j}EZAtEkNQaqH0qYTZKhA~B<zM?ReG}>1ZgJEwU(2it@49vjFl6_<e-m)kkc{r%6
zEKL+w8^zN{ay8)`br`q~3QLN{nxVJm!r?e!gO06Oo5n+tL!XRq0p~nXAe(F9^Zmp^
z;!?o1b&zgDKxW4<h?E(uVg@Ub|DXvg3FWK9d4?#qF@|M`#-C%UqF9P(mOPfLN){Qi
zRsLnzimCVSHQ0wtfkW1iqka*`{G*Qf#T@mEIpP;L<QI1sW{*1v3yMD+oNy!{;e;*u
zygB6|1Zn>-V4QBGqzUeYco<l5W8Xhk)Z7azZ0d1ZL1VYmoL@S>TinuD(%N6zHbAWP
z;QunmgJtc5W$pd2iq65xu7lM*hiZBc*Y+K$>mR~p?YP{@v1#CV<H3`yhfcK|Io*2b
zOv~Z>fRolEXWI{-YdvzV?a=+rht4z~xeu6WK6<wG_}TX34|JY*pzGv=-S=MTKJj4p
z$@9G@9_%}Lq5s~8`|o{l;M5}nrylCL_ko_%=lf1y=sSJB|ICB^XD$q!K0k2!+~Ao9
zy6-vNdhi(VQ`gfEYv@LGuBW!G8<AT{T|+@tHNb-=Y(hp>R8k6P4aA~T!lF==OAQW7
z4GK+zg@k7WMWh8qV2U9lJT^BrxiBfCEG?%ZtDq*ku(qJAsk*7Bt@mj6!INEsCveiy
zf4rr4sHyW%Eqrf%cVT4<o|ckR9+y%Sk&qh@mSV6+D|Bd#*cdz=z~kX9aq|?pxD#{r
zmU=K0UQCTQODj|aNDv}w!{oYfsTP<Clj$M>8?`a<AA^U1pV3)TfFFRz5|9P_;6XgW
z`MAaqZ?+~G&2a`poK_zL+Bg}@Q-;fxjyj$Uh$3wz87YN_i^ZX6(V`8-k@)ju!2(67
zP!Yk0`+}B;9GHnBMKfU(Bu5BZ<ODf5x=SKl#X;`kKp#aIM;9e9M)S0&s)Y!((Qr5j
z)l}vz)CMq3BvEGa<tZX<GEWl=#*o^Y2JrKxp)7$N;6V*QAP5riL(q2;aWLr+1(UL(
z)IQ;ApHO5tGE7*YI--r^Yh&>ngsNB)J&K~a{AivqL8{8qT1t!|4a&eqWpJA&tkV$H
zYl`W&#0}bF4*@Oy3CHa5NB!fD2gDqM1;rc>jy(|+dv8F}IeYTC{|_{!Wp;12Z-)Q?
N002ovPDHLkV1m!NmvR6A

literal 0
HcmV?d00001

diff --git a/deploy/tvm/include/crt_config.h b/deploy/tvm/include/crt_config.h
new file mode 100644
index 0000000000..4b9ccca02b
--- /dev/null
+++ b/deploy/tvm/include/crt_config.h
@@ -0,0 +1,26 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#ifndef TVM_RUNTIME_CRT_CONFIG_H_
+#define TVM_RUNTIME_CRT_CONFIG_H_
+
+/*! Log level of the CRT runtime */
+#define TVM_CRT_LOG_LEVEL TVM_CRT_LOG_LEVEL_DEBUG
+
+#endif  // TVM_RUNTIME_CRT_CONFIG_H_
diff --git a/deploy/tvm/include/tvm_runtime.h b/deploy/tvm/include/tvm_runtime.h
new file mode 100644
index 0000000000..2b59d93470
--- /dev/null
+++ b/deploy/tvm/include/tvm_runtime.h
@@ -0,0 +1,55 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <tvm/runtime/c_runtime_api.h>
+#include <tvm/runtime/crt/stack_allocator.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void __attribute__((noreturn)) TVMPlatformAbort(tvm_crt_error_t error_code) {
+  printf("TVMPlatformAbort: %d\n", error_code);
+  printf("EXITTHESIM\n");
+  exit(-1);
+}
+
+tvm_crt_error_t TVMPlatformMemoryAllocate(size_t num_bytes, DLDevice dev, void** out_ptr) {
+  return kTvmErrorFunctionCallNotImplemented;
+}
+
+tvm_crt_error_t TVMPlatformMemoryFree(void* ptr, DLDevice dev) {
+  return kTvmErrorFunctionCallNotImplemented;
+}
+
+void TVMLogf(const char* msg, ...) {
+  va_list args;
+  va_start(args, msg);
+  vfprintf(stdout, msg, args);
+  va_end(args);
+}
+
+TVM_DLL int TVMFuncRegisterGlobal(const char* name, TVMFunctionHandle f, int override) { return 0; }
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/deploy/tvm/requirements.txt b/deploy/tvm/requirements.txt
new file mode 100644
index 0000000000..1bf86ed110
--- /dev/null
+++ b/deploy/tvm/requirements.txt
@@ -0,0 +1,3 @@
+paddlepaddle
+numpy
+opencv-python
\ No newline at end of file
diff --git a/deploy/tvm/run_demo.sh b/deploy/tvm/run_demo.sh
new file mode 100755
index 0000000000..dd8e8f163d
--- /dev/null
+++ b/deploy/tvm/run_demo.sh
@@ -0,0 +1,152 @@
+#!/bin/bash
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+set -e
+set -u
+set -o pipefail
+
+# Show usage
+function show_usage() {
+    cat <<EOF
+Usage: run_demo.sh
+-h, --help
+    Display this help message.
+--cmsis_path CMSIS_PATH
+    Set path to CMSIS.
+--ethosu_platform_path ETHOSU_PLATFORM_PATH
+    Set path to Arm(R) Ethos(TM)-U core platform.
+--fvp_path FVP_PATH
+   Set path to FVP.
+--cmake_path
+   Set path to cmake.
+EOF
+}
+
+# Parse arguments
+while (( $# )); do
+    case "$1" in
+        -h|--help)
+            show_usage
+            exit 0
+            ;;
+
+        --cmsis_path)
+            if [ $# -gt 1 ]
+            then
+                export CMSIS_PATH="$2"
+                shift 2
+            else
+                echo 'ERROR: --cmsis_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --ethosu_platform_path)
+            if [ $# -gt 1 ]
+            then
+                export ETHOSU_PLATFORM_PATH="$2"
+                shift 2
+            else
+                echo 'ERROR: --ethosu_platform_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --fvp_path)
+            if [ $# -gt 1 ]
+            then
+                export PATH="$2/models/Linux64_GCC-6.4:$PATH"
+                shift 2
+            else
+                echo 'ERROR: --fvp_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --cmake_path)
+            if [ $# -gt 1 ]
+            then
+                export CMAKE="$2"
+                shift 2
+            else
+                echo 'ERROR: --cmake_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        -*|--*)
+            echo "Error: Unknown flag: $1" >&2
+            show_usage >&2
+            exit 1
+            ;;
+    esac
+done
+
+
+# Directories
+script_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
+
+
+# Make build directory
+rm -rf build
+make cleanall
+mkdir -p build
+cd build
+
+wget https://paddleocr.bj.bcebos.com/tvm/ocr_en.tar
+tar -xf ocr_en.tar
+
+# Compile model for Arm(R) Cortex(R)-M55 CPU and CMSIS-NN
+# An alternative to using "python3 -m tvm.driver.tvmc" is to call
+# "tvmc" directly once TVM has been pip installed.
+python3 -m tvm.driver.tvmc compile --target=cmsis-nn,c \
+    --target-cmsis-nn-mcpu=cortex-m55 \
+    --target-c-mcpu=cortex-m55 \
+    --runtime=crt \
+    --executor=aot \
+    --executor-aot-interface-api=c \
+    --executor-aot-unpacked-api=1 \
+    --pass-config tir.usmp.enable=1 \
+    --pass-config tir.usmp.algorithm=hill_climb \
+    --pass-config tir.disable_storage_rewrite=1 \
+    --pass-config tir.disable_vectorize=1 ocr_en/inference.pdmodel \
+    --output-format=mlf \
+    --model-format=paddle \
+    --module-name=rec \
+    --input-shapes x:[1,3,32,320] \
+    --output=rec.tar
+tar -xf rec.tar
+
+# Create C header files
+cd ..
+python3 ./convert_image.py imgs_words_en/word_116.png
+
+# Build demo executable
+cd ${script_dir}
+echo ${script_dir}
+make
+
+# Run demo executable on the FVP
+FVP_Corstone_SSE-300_Ethos-U55 -C cpu0.CFGDTCMSZ=15 \
+-C cpu0.CFGITCMSZ=15 -C mps3_board.uart0.out_file=\"-\" -C mps3_board.uart0.shutdown_tag=\"EXITTHESIM\" \
+-C mps3_board.visualisation.disable-visualisation=1 -C mps3_board.telnetterminal0.start_telnet=0 \
+-C mps3_board.telnetterminal1.start_telnet=0 -C mps3_board.telnetterminal2.start_telnet=0 -C mps3_board.telnetterminal5.start_telnet=0 \
+./build/demo
\ No newline at end of file
diff --git a/deploy/tvm/src/demo_bare_metal.c b/deploy/tvm/src/demo_bare_metal.c
new file mode 100644
index 0000000000..3f5f1bc4b0
--- /dev/null
+++ b/deploy/tvm/src/demo_bare_metal.c
@@ -0,0 +1,78 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <stdio.h>
+#include <tvm_runtime.h>
+#include <tvmgen_rec.h>
+
+#include "uart.h"
+
+// Header files generated by convert_image.py
+#include "inputs.h"
+#include "outputs.h"
+
+
+int main(int argc, char** argv) {
+  char dict[]={"#0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuvwxyz{|}~!\"#$%&'()*+,-./  "};
+  int char_dict_nums = 97;
+  uart_init();
+  printf("Starting ocr rec inference\n");
+  struct tvmgen_rec_outputs rec_outputs = {
+      .output = output,
+  };
+  struct tvmgen_rec_inputs rec_inputs = {
+      .x = input,
+  };
+
+  tvmgen_rec_run(&rec_inputs, &rec_outputs);
+
+  // post process
+  int char_nums = output_len / char_dict_nums;
+  
+  int last_index = 0;
+  float score = 0.f;
+  int count = 0;
+  
+  printf("text: ");
+  for (int i = 0; i < char_nums; i++) {
+    int argmax_idx = 0;
+    float max_value = 0.0f;
+    for (int j = 0; j < char_dict_nums; j++){
+      if (output[i * char_dict_nums + j] > max_value){
+        max_value = output[i * char_dict_nums + j];
+        argmax_idx = j;
+      }
+    }
+    if (argmax_idx > 0 && (!(i > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      // printf("%d，%f，%c\n", argmax_idx, max_value, dict[argmax_idx]);
+      printf("%c", dict[argmax_idx]);
+    }
+    last_index = argmax_idx;
+  }
+  score /= count;
+  printf(", score: %f\n", score);
+  
+  // The FVP will shut down when it receives "EXITTHESIM" on the UART
+  printf("EXITTHESIM\n");
+  while (1 == 1)
+    ;
+  return 0;
+}

From cdb0804746a797a4520f5d75319d8838e00d9a51 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Mon, 20 Jun 2022 02:10:22 +0000
Subject: [PATCH 026/134] fix fleet tipc

---
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 .../train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt | 2 +-
 7 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 5fb87fae18..91a6288eb0 100644
--- a/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ infer_model:./inference/ch_PP-OCRv2_det_infer/
 infer_export:null
 infer_quant:False
 inference:tools/infer/predict_det.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 4af814036a..5795bc27e6 100644
--- a/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ infer_model:./inference/ch_PP-OCRv2_rec_infer
 infer_export:null
 infer_quant:False
 inference:tools/infer/predict_rec.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1|6
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 8275254f3f..7e987125a6 100644
--- a/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ infer_model:./inference/ch_PP-OCRv3_det_infer/
 infer_export:null
 infer_quant:False
 inference:tools/infer/predict_det.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index b76048813e..5271f78bb7 100644
--- a/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
 infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
 infer_quant:False
 inference:tools/infer/predict_det.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 3a7c487173..631118c0a9 100644
--- a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ train_model:./inference/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
 infer_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
 infer_quant:False
 inference:tools/infer/predict_rec.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1|6
diff --git a/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index 77b69f6c3a..12388d9677 100644
--- a/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_ppocr_server_v2.0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ train_model:./inference/ch_ppocr_server_v2.0_det_train/best_accuracy
 infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o
 infer_quant:False
 inference:tools/infer/predict_det.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1
diff --git a/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
index d486d37668..9884ab247b 100644
--- a/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
+++ b/test_tipc/configs/ch_ppocr_server_v2.0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -38,7 +38,7 @@ train_model:./inference/ch_ppocr_server_v2.0_rec_train/best_accuracy
 infer_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2.0_rec/rec_icdar15_train.yml -o
 infer_quant:False
 inference:tools/infer/predict_rec.py
---use_gpu:True|False
+--use_gpu:False
 --enable_mkldnn:False
 --cpu_threads:6
 --rec_batch_num:1|6

From 49215a1afeffb6ca1a61d14c60ed87440a865a7a Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 20 Jun 2022 04:00:47 +0000
Subject: [PATCH 027/134] rename to avh

---
 deploy/{tvm => avh}/.gitignore                 |   0
 deploy/{tvm => avh}/Makefile                   |   0
 deploy/{tvm => avh}/README.md                  |   0
 deploy/{tvm => avh}/arm-none-eabi-gcc.cmake    |   0
 deploy/{tvm => avh}/convert_image.py           |   0
 deploy/{tvm => avh}/corstone300.ld             |   0
 deploy/{tvm => avh}/imgs_words_en/word_10.png  | Bin
 deploy/{tvm => avh}/imgs_words_en/word_116.png | Bin
 deploy/{tvm => avh}/include/crt_config.h       |   0
 deploy/{tvm => avh}/include/tvm_runtime.h      |   0
 deploy/{tvm => avh}/requirements.txt           |   0
 deploy/{tvm => avh}/run_demo.sh                |   0
 deploy/{tvm => avh}/src/demo_bare_metal.c      |   0
 13 files changed, 0 insertions(+), 0 deletions(-)
 rename deploy/{tvm => avh}/.gitignore (100%)
 rename deploy/{tvm => avh}/Makefile (100%)
 rename deploy/{tvm => avh}/README.md (100%)
 rename deploy/{tvm => avh}/arm-none-eabi-gcc.cmake (100%)
 rename deploy/{tvm => avh}/convert_image.py (100%)
 rename deploy/{tvm => avh}/corstone300.ld (100%)
 rename deploy/{tvm => avh}/imgs_words_en/word_10.png (100%)
 rename deploy/{tvm => avh}/imgs_words_en/word_116.png (100%)
 rename deploy/{tvm => avh}/include/crt_config.h (100%)
 rename deploy/{tvm => avh}/include/tvm_runtime.h (100%)
 rename deploy/{tvm => avh}/requirements.txt (100%)
 rename deploy/{tvm => avh}/run_demo.sh (100%)
 rename deploy/{tvm => avh}/src/demo_bare_metal.c (100%)

diff --git a/deploy/tvm/.gitignore b/deploy/avh/.gitignore
similarity index 100%
rename from deploy/tvm/.gitignore
rename to deploy/avh/.gitignore
diff --git a/deploy/tvm/Makefile b/deploy/avh/Makefile
similarity index 100%
rename from deploy/tvm/Makefile
rename to deploy/avh/Makefile
diff --git a/deploy/tvm/README.md b/deploy/avh/README.md
similarity index 100%
rename from deploy/tvm/README.md
rename to deploy/avh/README.md
diff --git a/deploy/tvm/arm-none-eabi-gcc.cmake b/deploy/avh/arm-none-eabi-gcc.cmake
similarity index 100%
rename from deploy/tvm/arm-none-eabi-gcc.cmake
rename to deploy/avh/arm-none-eabi-gcc.cmake
diff --git a/deploy/tvm/convert_image.py b/deploy/avh/convert_image.py
similarity index 100%
rename from deploy/tvm/convert_image.py
rename to deploy/avh/convert_image.py
diff --git a/deploy/tvm/corstone300.ld b/deploy/avh/corstone300.ld
similarity index 100%
rename from deploy/tvm/corstone300.ld
rename to deploy/avh/corstone300.ld
diff --git a/deploy/tvm/imgs_words_en/word_10.png b/deploy/avh/imgs_words_en/word_10.png
similarity index 100%
rename from deploy/tvm/imgs_words_en/word_10.png
rename to deploy/avh/imgs_words_en/word_10.png
diff --git a/deploy/tvm/imgs_words_en/word_116.png b/deploy/avh/imgs_words_en/word_116.png
similarity index 100%
rename from deploy/tvm/imgs_words_en/word_116.png
rename to deploy/avh/imgs_words_en/word_116.png
diff --git a/deploy/tvm/include/crt_config.h b/deploy/avh/include/crt_config.h
similarity index 100%
rename from deploy/tvm/include/crt_config.h
rename to deploy/avh/include/crt_config.h
diff --git a/deploy/tvm/include/tvm_runtime.h b/deploy/avh/include/tvm_runtime.h
similarity index 100%
rename from deploy/tvm/include/tvm_runtime.h
rename to deploy/avh/include/tvm_runtime.h
diff --git a/deploy/tvm/requirements.txt b/deploy/avh/requirements.txt
similarity index 100%
rename from deploy/tvm/requirements.txt
rename to deploy/avh/requirements.txt
diff --git a/deploy/tvm/run_demo.sh b/deploy/avh/run_demo.sh
similarity index 100%
rename from deploy/tvm/run_demo.sh
rename to deploy/avh/run_demo.sh
diff --git a/deploy/tvm/src/demo_bare_metal.c b/deploy/avh/src/demo_bare_metal.c
similarity index 100%
rename from deploy/tvm/src/demo_bare_metal.c
rename to deploy/avh/src/demo_bare_metal.c

From bdc43081771e0a63803026d177f56ab2ef60f50c Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Mon, 20 Jun 2022 11:40:35 +0000
Subject: [PATCH 028/134] fix timeout

---
 .../det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt       | 4 ++--
 test_tipc/prepare.sh                                          | 3 ++-
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt
index de9e336332..b70ef46b4a 100644
--- a/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt
+++ b/test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/train_infer_python.txt
@@ -7,13 +7,13 @@ Global.auto_cast:null
 Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
 Global.save_model_dir:./output/
 Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
+Global.pretrained_model:./pretrain_models/det_r50_vd_sast_icdar15_v2.0_train/best_accuracy
 train_model_name:latest
 train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
 null:null
 ##
 trainer:norm_train
-norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/det_r50_vd_sast_icdar2015.yml -o Global.pretrained_model=./pretrain_models/ResNet50_vd_ssld_pretrained
+norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_sast_icdar15_v2.0/det_r50_vd_sast_icdar2015.yml -o
 pact_train:null
 fpgm_train:null
 distill_train:null
diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 8c9e12511d..cac373dedc 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -98,8 +98,9 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
     fi
     if [ ${model_name} == "det_r50_vd_sast_icdar15_v2.0" ] || [ ${model_name} == "det_r50_vd_sast_totaltext_v2.0" ]; then
         wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar --no-check-certificate
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar --no-check-certificate
-        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && cd ../
+        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && cd ../
     fi
     if [ ${model_name} == "det_mv3_db_v2_0" ]; then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar  --no-check-certificate

From 89567ae300e6cd827d225a1dd1f7888904fdb51f Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 21 Jun 2022 15:21:58 +0800
Subject: [PATCH 029/134] fix

---
 tools/infer/predict_det.py | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py
index 5f2675d667..73387aef59 100755
--- a/tools/infer/predict_det.py
+++ b/tools/infer/predict_det.py
@@ -24,6 +24,7 @@ import cv2
 import numpy as np
 import time
 import sys
+from scipy.spatial import distance as dist
 
 import tools.infer.utility as utility
 from ppocr.utils.logging import get_logger
@@ -150,14 +151,22 @@ class TextDetector(object):
                 logger=logger)
 
     def order_points_clockwise(self, pts):
-        rect = np.zeros((4, 2), dtype="float32")
-        s = pts.sum(axis=1)
-        rect[0] = pts[np.argmin(s)]
-        rect[2] = pts[np.argmax(s)]
-        diff = np.diff(pts, axis=1)
-        rect[1] = pts[np.argmin(diff)]
-        rect[3] = pts[np.argmax(diff)]
-        return rect
+        """
+        refer to :https://github.com/PyImageSearch/imutils/blob/9f740a53bcc2ed7eba2558afed8b4c17fd8a1d4c/imutils/perspective.py#L9
+        """
+        # sort the points based on their x-coordinates
+        xSorted = pts[np.argsort(pts[:, 0]), :]
+
+        leftMost = xSorted[:2, :]
+        rightMost = xSorted[2:, :]
+
+        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
+        (tl, bl) = leftMost
+
+        D = dist.cdist(tl[np.newaxis], rightMost, "euclidean")[0]
+        (br, tr) = rightMost[np.argsort(D)[::-1], :]
+
+        return np.array([tl, tr, br, bl], dtype="float32")
 
     def clip_det_res(self, points, img_height, img_width):
         for pno in range(points.shape[0]):

From 4b85f7e3e4b0c10b7d0f76e181e30ab56d80c6ce Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 21 Jun 2022 15:24:47 +0800
Subject: [PATCH 030/134] fix

---
 applications/轻量级车牌识别.md | 2831 --------------------------------
 1 file changed, 2831 deletions(-)
 delete mode 100644 applications/轻量级车牌识别.md

diff --git a/applications/轻量级车牌识别.md b/applications/轻量级车牌识别.md
deleted file mode 100644
index b6c3847d14..0000000000
--- a/applications/轻量级车牌识别.md
+++ /dev/null
@@ -1,2831 +0,0 @@
-
-
-
-
-
-
-
-<!DOCTYPE html>
-<html lang="en" data-color-mode="auto" data-light-theme="light" data-dark-theme="dark" data-a11y-animated-images="system">
-  <head>
-    <meta charset="utf-8">
-  <link rel="dns-prefetch" href="https://github.githubassets.com">
-  <link rel="dns-prefetch" href="https://avatars.githubusercontent.com">
-  <link rel="dns-prefetch" href="https://github-cloud.s3.amazonaws.com">
-  <link rel="dns-prefetch" href="https://user-images.githubusercontent.com/">
-  <link rel="preconnect" href="https://github.githubassets.com" crossorigin>
-  <link rel="preconnect" href="https://avatars.githubusercontent.com">
-
-
-
-  <link crossorigin="anonymous" media="all" integrity="sha512-ksfTgQOOnE+FFXf+yNfVjKSlEckJAdufFIYGK7ZjRhWcZgzAGcmZqqArTgMLpu90FwthqcCX4ldDgKXbmVMeuQ==" rel="stylesheet" href="https://github.githubassets.com/assets/light-92c7d381038e.css" /><link crossorigin="anonymous" media="all" integrity="sha512-1KkMNn8M/al/dtzBLupRwkIOgnA9MWkm8oxS+solP87jByEvY/g4BmoxLihRogKcX1obPnf4Yp7dI0ZTWO+ljg==" rel="stylesheet" href="https://github.githubassets.com/assets/dark-d4a90c367f0c.css" /><link data-color-theme="dark_dimmed" crossorigin="anonymous" media="all" integrity="sha512-cZa7DZqvMBwD236uzEunO/G1dvw8/QftyT2UtLWKQFEy0z0eq0R5WPwqVME+3NSZG1YaLJAaIqtU+m0zWf/6SQ==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_dimmed-7196bb0d9aaf.css" /><link data-color-theme="dark_high_contrast" crossorigin="anonymous" media="all" integrity="sha512-WVoKqJ4y1nLsdNH4RkRT5qrM9+n9RFe1RHSiTnQkBf5TSZkJEc9GpLpTIS7T15EQaUQBJ8BwmKvwFPVqfpTEIQ==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_high_contrast-595a0aa89e32.css" /><link data-color-theme="dark_colorblind" crossorigin="anonymous" media="all" integrity="sha512-XpAMBMSRZ6RTXgepS8LjKiOeNK3BilRbv8qEiA/M3m+Q4GoqxtHedOI5BAZRikCzfBL4KWYvVzYZSZ8Gp/UnUg==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_colorblind-5e900c04c491.css" /><link data-color-theme="light_colorblind" crossorigin="anonymous" media="all" integrity="sha512-3HF2HZ4LgEIQm77yOzoeR20CX1n2cUQlcywscqF4s+5iplolajiHV7E5ranBwkX65jN9TNciHEVSYebQ+8xxEw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_colorblind-dc71761d9e0b.css" /><link data-color-theme="light_high_contrast" crossorigin="anonymous" media="all" integrity="sha512-+J8j3T0kbK9/sL3zbkCfPtgYcRD4qQfRbT6xnfOrOTjvz4zhr0M7AXPuE642PpaxGhHs1t77cTtieW9hI2K6Gw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_high_contrast-f89f23dd3d24.css" /><link data-color-theme="light_tritanopia" crossorigin="anonymous" media="all" integrity="sha512-AQeAx5wHQAXNf0DmkvVlHYwA3f6BkxunWTI0GGaRN57GqD+H9tW8RKIKlopLS0qGaC54seFsPc601GDlqIuuHg==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_tritanopia-010780c79c07.css" /><link data-color-theme="dark_tritanopia" crossorigin="anonymous" media="all" integrity="sha512-+u5pmgAE0T03d/yI6Ha0NWwz6Pk0W6S6WEfIt8veDVdK8NTjcMbZmQB9XUCkDlrBoAKkABva8HuGJ+SzEpV1Uw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_tritanopia-faee699a0004.css" />
-
-    <link crossorigin="anonymous" media="all" integrity="sha512-EAhBCLIJ/pXHG3Y6yQhs9s53SHV80sjJ+yCwlQtfv7LaVkD+VoEuZBZ5betQJFUNj/5qBSfZk5GFtazEDzWLAg==" rel="stylesheet" href="https://github.githubassets.com/assets/primer-10084108b209.css" />
-    <link crossorigin="anonymous" media="all" integrity="sha512-4cUFwwYgpnyRoJKr15lavpQFa6HaoGIW6sap6JXLP9cQ33dOLfadeyjfJypUJ8GfE8MnAfdfvcNZlgn7oZ70PA==" rel="stylesheet" href="https://github.githubassets.com/assets/global-e1c505c30620.css" />
-    <link crossorigin="anonymous" media="all" integrity="sha512-ZkqzH/tvOPQgSkHnN9TlzL/abuPzgfV2KLooWN2Xq+jrP7VFzNKd0dib2+zSde1277kCzr5gp1piayo6AthIsQ==" rel="stylesheet" href="https://github.githubassets.com/assets/github-664ab31ffb6f.css" />
-  <link crossorigin="anonymous" media="all" integrity="sha512-PXtwH8brj9qH7LskWA0iAMLw4Wx3RXN7txSxjlqdzQEalZKu3UcsQ+ajfHJMDkYDA2iWL1FIrf2zyLflA5Xu7A==" rel="stylesheet" href="https://github.githubassets.com/assets/code-3d7b701fc6eb.css" />
-
-    <meta name="optimizely-datafile" content="{&quot;groups&quot;: [], &quot;environmentKey&quot;: &quot;production&quot;, &quot;rollouts&quot;: [], &quot;typedAudiences&quot;: [], &quot;projectId&quot;: &quot;16737760170&quot;, &quot;variables&quot;: [], &quot;featureFlags&quot;: [], &quot;experiments&quot;: [{&quot;status&quot;: &quot;Running&quot;, &quot;audienceIds&quot;: [], &quot;variations&quot;: [{&quot;variables&quot;: [], &quot;id&quot;: &quot;20667381018&quot;, &quot;key&quot;: &quot;control&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;20680930759&quot;, &quot;key&quot;: &quot;treatment&quot;}], &quot;id&quot;: &quot;20652570897&quot;, &quot;key&quot;: &quot;project_genesis&quot;, &quot;layerId&quot;: &quot;20672300363&quot;, &quot;trafficAllocation&quot;: [{&quot;entityId&quot;: &quot;20667381018&quot;, &quot;endOfRange&quot;: 5000}, {&quot;entityId&quot;: &quot;20680930759&quot;, &quot;endOfRange&quot;: 10000}], &quot;forcedVariations&quot;: {&quot;83356e17066d336d1803024138ecb683&quot;: &quot;treatment&quot;, &quot;18e31c8a9b2271332466133162a4aa0d&quot;: &quot;treatment&quot;, &quot;10f8ab3fbc5ebe989a36a05f79d48f32&quot;: &quot;treatment&quot;, &quot;1686089f6d540cd2deeaec60ee43ecf7&quot;: &quot;treatment&quot;}}, {&quot;status&quot;: &quot;Running&quot;, &quot;audienceIds&quot;: [], &quot;variations&quot;: [{&quot;variables&quot;: [], &quot;id&quot;: &quot;21427950901&quot;, &quot;key&quot;: &quot;control&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;21429710665&quot;, &quot;key&quot;: &quot;beginner&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;21437291543&quot;, &quot;key&quot;: &quot;upstart&quot;}], &quot;id&quot;: &quot;21445030708&quot;, &quot;key&quot;: &quot;_259_zero_user_dashboard&quot;, &quot;layerId&quot;: &quot;21434011841&quot;, &quot;trafficAllocation&quot;: [{&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 3334}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 5000}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 8333}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 10000}], &quot;forcedVariations&quot;: {&quot;3c64268131793aa297119a343c19e345&quot;: &quot;beginner&quot;, &quot;95b24126db31ea8693c0fe5ea9f53b65&quot;: &quot;beginner&quot;, &quot;086e2abe64e9101112af53b95d2d90b9&quot;: &quot;upstart&quot;, &quot;bae688df9d297afac98e2d254e912ada&quot;: &quot;control&quot;, &quot;6c2cfda7c41396fcc31a4db759a42b94&quot;: &quot;beginner&quot;, &quot;16ed2b4ff7de02663b7c606309695916&quot;: &quot;control&quot;, &quot;1971768911.1635962195&quot;: &quot;beginner&quot;, &quot;830bf802470ec6c9c5800c99d8e57445&quot;: &quot;beginner&quot;}}], &quot;version&quot;: &quot;4&quot;, &quot;audiences&quot;: [{&quot;conditions&quot;: &quot;[\&quot;or\&quot;, {\&quot;match\&quot;: \&quot;exact\&quot;, \&quot;name\&quot;: \&quot;$opt_dummy_attribute\&quot;, \&quot;type\&quot;: \&quot;custom_attribute\&quot;, \&quot;value\&quot;: \&quot;$opt_dummy_value\&quot;}]&quot;, &quot;id&quot;: &quot;$opt_dummy_audience&quot;, &quot;name&quot;: &quot;Optimizely-Generated Audience for Backwards Compatibility&quot;}], &quot;anonymizeIP&quot;: true, &quot;sdkKey&quot;: &quot;WTc6awnGuYDdG98CYRban&quot;, &quot;attributes&quot;: [{&quot;id&quot;: &quot;16822470375&quot;, &quot;key&quot;: &quot;user_id&quot;}, {&quot;id&quot;: &quot;17143601254&quot;, &quot;key&quot;: &quot;spammy&quot;}, {&quot;id&quot;: &quot;18175660309&quot;, &quot;key&quot;: &quot;organization_plan&quot;}, {&quot;id&quot;: &quot;18813001570&quot;, &quot;key&quot;: &quot;is_logged_in&quot;}, {&quot;id&quot;: &quot;19073851829&quot;, &quot;key&quot;: &quot;geo&quot;}, {&quot;id&quot;: &quot;20175462351&quot;, &quot;key&quot;: &quot;requestedCurrency&quot;}, {&quot;id&quot;: &quot;20785470195&quot;, &quot;key&quot;: &quot;country_code&quot;}, {&quot;id&quot;: &quot;21656311196&quot;, &quot;key&quot;: &quot;opened_downgrade_dialog&quot;}], &quot;botFiltering&quot;: false, &quot;accountId&quot;: &quot;16737760170&quot;, &quot;events&quot;: [{&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;17911811441&quot;, &quot;key&quot;: &quot;hydro_click.dashboard.teacher_toolbox_cta&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18124116703&quot;, &quot;key&quot;: &quot;submit.organizations.complete_sign_up&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18145892387&quot;, &quot;key&quot;: &quot;no_metric.tracked_outside_of_optimizely&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18178755568&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.add_repo&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18180553241&quot;, &quot;key&quot;: &quot;submit.repository_imports.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18186103728&quot;, &quot;key&quot;: &quot;click.help.learn_more_about_repository_creation&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18188530140&quot;, &quot;key&quot;: &quot;test_event&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18191963644&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.transfer_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18195612788&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.import_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18210945499&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.invite_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18211063248&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.create_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18215721889&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.update_profile&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18224360785&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.dismiss&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18234832286&quot;, &quot;key&quot;: &quot;submit.organization_activation.complete&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18252392383&quot;, &quot;key&quot;: &quot;submit.org_repository.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18257551537&quot;, &quot;key&quot;: &quot;submit.org_member_invitation.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18259522260&quot;, &quot;key&quot;: &quot;submit.organization_profile.update&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18564603625&quot;, &quot;key&quot;: &quot;view.classroom_select_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18568612016&quot;, &quot;key&quot;: &quot;click.classroom_sign_in_click&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18572592540&quot;, &quot;key&quot;: &quot;view.classroom_name&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18574203855&quot;, &quot;key&quot;: &quot;click.classroom_create_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18582053415&quot;, &quot;key&quot;: &quot;click.classroom_select_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18589463420&quot;, &quot;key&quot;: &quot;click.classroom_create_classroom&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18591323364&quot;, &quot;key&quot;: &quot;click.classroom_create_first_classroom&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18591652321&quot;, &quot;key&quot;: &quot;click.classroom_grant_access&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18607131425&quot;, &quot;key&quot;: &quot;view.classroom_creation&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18831680583&quot;, &quot;key&quot;: &quot;upgrade_account_plan&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19064064515&quot;, &quot;key&quot;: &quot;click.signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19075373687&quot;, &quot;key&quot;: &quot;click.view_account_billing_page&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19077355841&quot;, &quot;key&quot;: &quot;click.dismiss_signup_prompt&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19079713938&quot;, &quot;key&quot;: &quot;click.contact_sales&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19120963070&quot;, &quot;key&quot;: &quot;click.compare_account_plans&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19151690317&quot;, &quot;key&quot;: &quot;click.upgrade_account_cta&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19424193129&quot;, &quot;key&quot;: &quot;click.open_account_switcher&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19520330825&quot;, &quot;key&quot;: &quot;click.visit_account_profile&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19540970635&quot;, &quot;key&quot;: &quot;click.switch_account_context&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19730198868&quot;, &quot;key&quot;: &quot;submit.homepage_signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19820830627&quot;, &quot;key&quot;: &quot;click.homepage_signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19988571001&quot;, &quot;key&quot;: &quot;click.create_enterprise_trial&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20036538294&quot;, &quot;key&quot;: &quot;click.create_organization_team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20040653299&quot;, &quot;key&quot;: &quot;click.input_enterprise_trial_form&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20062030003&quot;, &quot;key&quot;: &quot;click.continue_with_team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20068947153&quot;, &quot;key&quot;: &quot;click.create_organization_free&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20086636658&quot;, &quot;key&quot;: &quot;click.signup_continue.username&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20091648988&quot;, &quot;key&quot;: &quot;click.signup_continue.create_account&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20103637615&quot;, &quot;key&quot;: &quot;click.signup_continue.email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20111574253&quot;, &quot;key&quot;: &quot;click.signup_continue.password&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20120044111&quot;, &quot;key&quot;: &quot;view.pricing_page&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20152062109&quot;, &quot;key&quot;: &quot;submit.create_account&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20165800992&quot;, &quot;key&quot;: &quot;submit.upgrade_payment_form&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20171520319&quot;, &quot;key&quot;: &quot;submit.create_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20222645674&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.discuss_your_needs&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20227443657&quot;, &quot;key&quot;: &quot;submit.verify_primary_user_email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20234607160&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.try_enterprise&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20238175784&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20239847212&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.continue_free&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20251097193&quot;, &quot;key&quot;: &quot;recommended_plan&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20438619534&quot;, &quot;key&quot;: &quot;click.pricing_calculator.1_member&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20456699683&quot;, &quot;key&quot;: &quot;click.pricing_calculator.15_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20467868331&quot;, &quot;key&quot;: &quot;click.pricing_calculator.10_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20476267432&quot;, &quot;key&quot;: &quot;click.trial_days_remaining&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20476357660&quot;, &quot;key&quot;: &quot;click.discover_feature&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20479287901&quot;, &quot;key&quot;: &quot;click.pricing_calculator.custom_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20481107083&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.apply_teacher_benefits&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20483089392&quot;, &quot;key&quot;: &quot;click.pricing_calculator.5_members&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20484283944&quot;, &quot;key&quot;: &quot;click.onboarding_task&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20484996281&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.apply_student_benefits&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20486713726&quot;, &quot;key&quot;: &quot;click.onboarding_task_breadcrumb&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20490791319&quot;, &quot;key&quot;: &quot;click.upgrade_to_enterprise&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20491786766&quot;, &quot;key&quot;: &quot;click.talk_to_us&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20494144087&quot;, &quot;key&quot;: &quot;click.dismiss_enterprise_trial&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20499722759&quot;, &quot;key&quot;: &quot;completed_all_tasks&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20500710104&quot;, &quot;key&quot;: &quot;completed_onboarding_tasks&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20513160672&quot;, &quot;key&quot;: &quot;click.read_doc&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20516196762&quot;, &quot;key&quot;: &quot;actions_enabled&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20518980986&quot;, &quot;key&quot;: &quot;click.dismiss_trial_banner&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20535446721&quot;, &quot;key&quot;: &quot;click.issue_actions_prompt.dismiss_prompt&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20557002247&quot;, &quot;key&quot;: &quot;click.issue_actions_prompt.setup_workflow&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20595070227&quot;, &quot;key&quot;: &quot;click.pull_request_setup_workflow&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20626600314&quot;, &quot;key&quot;: &quot;click.seats_input&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20642310305&quot;, &quot;key&quot;: &quot;click.decrease_seats_number&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20662990045&quot;, &quot;key&quot;: &quot;click.increase_seats_number&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20679620969&quot;, &quot;key&quot;: &quot;click.public_product_roadmap&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20761240940&quot;, &quot;key&quot;: &quot;click.dismiss_survey_banner&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20767210721&quot;, &quot;key&quot;: &quot;click.take_survey&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20795281201&quot;, &quot;key&quot;: &quot;click.archive_list&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20966790249&quot;, &quot;key&quot;: &quot;contact_sales.submit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20996500333&quot;, &quot;key&quot;: &quot;contact_sales.existing_customer&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20996890162&quot;, &quot;key&quot;: &quot;contact_sales.blank_message_field&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21000470317&quot;, &quot;key&quot;: &quot;contact_sales.personal_email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21002790172&quot;, &quot;key&quot;: &quot;contact_sales.blank_phone_field&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21354412592&quot;, &quot;key&quot;: &quot;click.dismiss_create_readme&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21366102546&quot;, &quot;key&quot;: &quot;click.dismiss_zero_user_content&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21370252505&quot;, &quot;key&quot;: &quot;account_did_downgrade&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21370840408&quot;, &quot;key&quot;: &quot;click.cta_create_readme&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21375451068&quot;, &quot;key&quot;: &quot;click.cta_create_new_repository&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21385390948&quot;, &quot;key&quot;: &quot;click.zero_user_content&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21467712175&quot;, &quot;key&quot;: &quot;click.downgrade_keep&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21484112202&quot;, &quot;key&quot;: &quot;click.downgrade&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21495292213&quot;, &quot;key&quot;: &quot;click.downgrade_survey_exit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21508241468&quot;, &quot;key&quot;: &quot;click.downgrade_survey_submit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21512030356&quot;, &quot;key&quot;: &quot;click.downgrade_support&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21539090022&quot;, &quot;key&quot;: &quot;click.downgrade_exit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21543640644&quot;, &quot;key&quot;: &quot;click_fetch_upstream&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21646510300&quot;, &quot;key&quot;: &quot;click.move_your_work&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21656151116&quot;, &quot;key&quot;: &quot;click.add_branch_protection_rule&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21663860599&quot;, &quot;key&quot;: &quot;click.downgrade_dialog_open&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21687860483&quot;, &quot;key&quot;: &quot;click.learn_about_protected_branches&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21689050333&quot;, &quot;key&quot;: &quot;click.dismiss_protect_this_branch&quot;}], &quot;revision&quot;: &quot;1288&quot;}" />
-
-
-  <script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-IlxrV3fPKJ3H3zXBv8ZNFZA1UGUCcfs482E0tD9JDa8vDF+oIfVC8m5/T6o5u/ccu/oNvjJNPZhYkX02Ao349g==" src="https://github.githubassets.com/assets/runtime-225c6b5777cf.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-XKV1seh7UvAY3d4/PUvi5uHqcCJ9ThVg73aqJM9r8QFid4KRpTHbebXZdhznn39F6fvGFcgvAhTdTrUp/oWL1A==" src="https://github.githubassets.com/assets/environment-5ca575b1e87b.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-io+1MvgXPXTw8Kp4eOdNMJl8uGASuw8VfTY5VeIFETaAknimWi8GoxggMEeQ6mq0de4Dest4iIJ/9gUbCo0hgw==" src="https://github.githubassets.com/assets/vendors-node_modules_selector-observer_dist_index_esm_js-8a8fb532f817.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-Es25N4GyPa8Yfp5wpahoe5b2fyPtkRMyR6mKIXyCJC0ocqQazeWvxhGZhx3StRxOfqDfHDR5SS35u/R3Wux6Cg==" src="https://github.githubassets.com/assets/vendors-node_modules_delegated-events_dist_index_js-node_modules_github_details-dialog-elemen-63debe-12cdb93781b2.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-lmiecOIgf+hakg5oKMNM7grVhEDyPoIrT39Px448JJH5PSAaK21PH0Twgyz5O5oi8+dnlLr3Jt8bBCtAcpNdRw==" src="https://github.githubassets.com/assets/vendors-node_modules_github_filter-input-element_dist_index_js-node_modules_github_remote-inp-c7e9ed-96689e70e220.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-y67eNkVaNK4RguUGcHOvIbHFlgf1Qje+LDdjVw2eFuuvBOqta2GePz/CwoLIR/PJhhRAj5RPGxCWoomnimSw6w==" src="https://github.githubassets.com/assets/vendors-node_modules_github_catalyst_lib_index_js-node_modules_github_time-elements_dist_index_js-cbaede36455a.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-DE9GK/TS3Hf/dJfP1GMiyjvsGRnpR4A1eGH64s0Ot7Vy7Tj7/aD1d5EOMWHFWQXlXy/ZSwu5MHInG6RZN31bMA==" src="https://github.githubassets.com/assets/vendors-node_modules_github_file-attachment-element_dist_index_js-node_modules_primer_view-co-52e104-0c4f462bf4d2.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-l//8hwOiwPAmBg0NwsWUYovdQ7/r9kPHiD9/LL4fD3M7El8gbgaOYU5+o6cYLB6puSfOTxyN9M6fE38HSDr2Bw==" src="https://github.githubassets.com/assets/github-elements-97fffc8703a2.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-lIm60Y77qCoue/4Srnoe0QDZN/ReE8mrMZE2uyNTZteQP0e5kPJGSn4NtlWfYkVvI1DuyDC76KsOz04K76v+iA==" src="https://github.githubassets.com/assets/element-registry-9489bad18efb.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-uo73yUZcm4EicwjSbfxFZcKfjWniOxhLBp+q1n7IFRfutFM6/lzbQMgD0Xrxp7QD1HzqdvrV8UclPhi3mEOyzQ==" src="https://github.githubassets.com/assets/vendors-node_modules_lit-html_lit-html_js-ba8ef7c9465c.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-5XdknqL9iTuaAhbOnKJDHf+5bJReOYDwlz5p2mW+SAVfab4RN6AfIliqtPwheKld40DrhPs47We/mT5nlZq2Sg==" src="https://github.githubassets.com/assets/vendors-node_modules_github_remote-form_dist_index_js-node_modules_github_catalyst_lib_index_-87b1b3-e577649ea2fd.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-0r1nf/rfPz54kyePp4f63bcPxkFo7wyaUZJD/SwIVDK3q0WzurAK9ydOm88tzKtPJm8xWI0Vo25NyCfecwxJ9g==" src="https://github.githubassets.com/assets/vendors-node_modules_github_mini-throttle_dist_index_js-node_modules_github_hotkey_dist_index-9f48bd-d2bd677ffadf.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-VK56d0N1hPZ20mOzPoy84zlTGCjGbKKmVvfjoyDqSF+VxTD4f6X8QDs2RgG1R1cdBmsCiea+ZxP6ukV3tHlD+Q==" src="https://github.githubassets.com/assets/vendors-node_modules_github_paste-markdown_dist_index_esm_js-node_modules_github_quote-select-df2537-54ae7a774375.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-vneJnOQi1L9PJESlqC6veCutbVwyQDQCzMoPlkEhfWJ592tNYva9p1SV5uTde09VMWh17HFMDbfuVoy0I5tqHA==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_pjax_ts-be77899ce422.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-2hsYB2KHyayxUJJxJGhXeTeTZZG2c7lzRtO0uB1txmpc7rfvIt4mf0iossT0MHIHknYlaslgi98jmmlVXcXaZQ==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_keyboard-shortcuts-helper_ts-app_assets_modules_github_be-af52ef-da1b18076287.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-8eyyuvsdeUUBrkTsu7OePie4HdGA6yQww2w70dhs28WU0cMh5qhDM1vHrLOwJdbFQodch4KNQ1rv6KNyw1XMpA==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_details_ts-app_assets_modules_github_behaviors_include-fr-a5a4c7-f1ecb2bafb1d.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-i0EzX4ojOyIXe3MB4Kc/20cXNNTwDWHPd9i/bwOEDOd3bRNP7xJCQkX4f7DIELn+WFHtJSpEEfPRzi64gEENFw==" src="https://github.githubassets.com/assets/behaviors-8b41335f8a23.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-rLnbCtiKIpRtm1jgAz913tDImW3g8G+FLRMBIMwrkZ33pO8PMzWDG4m8Zl0uIIs10hCyFr6Dc8s6VDC200S9lg==" src="https://github.githubassets.com/assets/vendors-node_modules_delegated-events_dist_index_js-node_modules_github_catalyst_lib_index_js-01c6c51-acb9db0ad88a.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-zJONQW0ZfaXs3mk/bws2nP0XemlIUbA4+vWsLqTzjTTz7X2yCAI0VXa+i8y/vgLF3Ooeyus5ZnyhiKYj3Jwz6w==" src="https://github.githubassets.com/assets/notifications-global-cc938d416d19.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-1b+YhvQsW1BdKi3eWSc7oNSGJcrQalc3oxxMoXI8foh3weo2AOLx86EMZ5J1a541yu2y7Axj7Vntn1T66EW8dA==" src="https://github.githubassets.com/assets/vendors-node_modules_optimizely_optimizely-sdk_dist_optimizely_browser_es_min_js-node_modules-77839b-d5bf9886f42c.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-NH4Gv218MnzYs/oudrFw7n14lPMa75qmHypJ9MbJn1RW03KwjJKCegU17usZk1uQD8cLWseA4vhCL638+tKM+A==" src="https://github.githubassets.com/assets/optimizely-347e06bf6d7c.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-yG3t7Lpt5yKkdMNXFpl75Xemm+g4ghZ5e8xRBwl/HMbmgc3htVvxrjbohbK9MMfnnci1SnogZNH9DhJHYdQf3A==" src="https://github.githubassets.com/assets/vendors-node_modules_virtualized-list_es_index_js-node_modules_github_template-parts_lib_index_js-c86dedecba6d.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-eE/fdOH+uhuRT992TtBa/oI33EN3x7EBNkJnTdiozp5RMDDiteODOWGCazeuulOPF5ZSFPR7r0PBZFn8enpsRQ==" src="https://github.githubassets.com/assets/vendors-node_modules_github_mini-throttle_dist_decorators_js-node_modules_github_remote-form_-2c55be-784fdf74e1fe.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-FNWxKHkUqs6+xuyyZpMx9BmvcxQqIgVJ5A+PmdU82lZO41cFlo6y6ddHkRfWh2Zd6j9ozvndvrTaOuEH58VvEw==" src="https://github.githubassets.com/assets/vendors-node_modules_github_file-attachment-element_dist_index_js-node_modules_github_filter--fd67a9-14d5b1287914.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-dAabCMULUfibZDcNGDpWbINNJawdJeCfhl8ZTGFEYvgpblXXQIBlutl5zv0oOqc0p1ginRnWy4mAZES/swKmZA==" src="https://github.githubassets.com/assets/app_assets_modules_github_fetch_ts-app_assets_modules_github_input-demux_ts-app_assets_module-e5b55c-74069b08c50b.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-SXRchiPOYprh1D0HcC85QDp+f15IP1oshRaSxymGKZ5jw+se3OVL6zw5QYPnxl9cOf8WxcyBuW+t+raq61yg5g==" src="https://github.githubassets.com/assets/repositories-49745c8623ce.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-2IO4DDKs14Xk+vMRzfe3NVcyY5Bsb071hi/LJEZcTTo74dZ9sgelYQaCPu0nh3qsny1AvzutHZyuvBmt0fbguA==" src="https://github.githubassets.com/assets/app_assets_modules_github_diffs_blob-lines_ts-app_assets_modules_github_diffs_linkable-line-n-f314c3-d883b80c32ac.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-9b6aISppMLr5F8Imzz33ChOeks/lElt5UEQ1fHaKruZEmJ/Jf2bZ8A96OL0fPGqhReDoWK5CZec/Btp1ny7+oQ==" src="https://github.githubassets.com/assets/diffs-f5be9a212a69.js"></script>
-
-
-  <title>PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR · GitHub</title>
-
-
-
-
-
-  <meta name="request-id" content="B29E:3EF7:A6757:C396D:62B1711D" data-pjax-transient="true"/><meta name="html-safe-nonce" content="15daa2bd7ccba656c9766df0cb28166a7316ec9ab50a1896f789379774d4eaae" data-pjax-transient="true"/><meta name="visitor-payload" content="eyJyZWZlcnJlciI6IiIsInJlcXVlc3RfaWQiOiJCMjlFOjNFRjc6QTY3NTc6QzM5NkQ6NjJCMTcxMUQiLCJ2aXNpdG9yX2lkIjoiNTYzMDgzMDUyMDQ2NDE0MjYyMiIsInJlZ2lvbl9lZGdlIjoic2VhIiwicmVnaW9uX3JlbmRlciI6InNlYSJ9" data-pjax-transient="true"/><meta name="visitor-hmac" content="24e8ead041aed97e2f99f905d69cd7a4bca272e545058aabf0af164142b62c03" data-pjax-transient="true"/>
-
-    <meta name="hovercard-subject-tag" content="repository:262296122" data-pjax-transient>
-
-
-  <meta name="github-keyboard-shortcuts" content="repository,source-code" data-pjax-transient="true" />
-
-
-  <meta name="selected-link" value="repo_source" data-pjax-transient>
-
-    <meta name="google-site-verification" content="c1kuD-K2HIVF635lypcsWPoD4kilo5-jA_wBFyT4uMY">
-  <meta name="google-site-verification" content="KT5gs8h0wvaagLKAVWq8bbeNwnZZK1r1XQysX3xurLU">
-  <meta name="google-site-verification" content="ZzhVyEFwb7w3e0-uOTltm8Jsck2F5StVihD0exw2fsA">
-  <meta name="google-site-verification" content="GXs5KoUUkNCoaAZn7wPN-t01Pywp9M3sEjnt_3_ZWPc">
-
-<meta name="octolytics-url" content="https://collector.github.com/github/collect" />
-
-  <meta name="analytics-location" content="/&lt;user-name&gt;/&lt;repo-name&gt;/blob/show" data-pjax-transient="true" />
-
-
-
-
-
-
-
-
-    <meta name="user-login" content="">
-
-    <meta name="viewport" content="width=device-width">
-
-      <meta name="description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices) - PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR">
-      <link rel="search" type="application/opensearchdescription+xml" href="/opensearch.xml" title="GitHub">
-    <link rel="fluid-icon" href="https://github.com/fluidicon.png" title="GitHub">
-    <meta property="fb:app_id" content="1401488693436528">
-    <meta name="apple-itunes-app" content="app-id=1477376905" />
-      <meta name="twitter:image:src" content="https://opengraph.githubassets.com/1eca697559d2aabccdeb858570f3f508b575ece90f37d8d4e2156599d7d26d4f/PaddlePaddle/PaddleOCR" /><meta name="twitter:site" content="@github" /><meta name="twitter:card" content="summary_large_image" /><meta name="twitter:title" content="PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR" /><meta name="twitter:description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." />
-      <meta property="og:image" content="https://opengraph.githubassets.com/1eca697559d2aabccdeb858570f3f508b575ece90f37d8d4e2156599d7d26d4f/PaddlePaddle/PaddleOCR" /><meta property="og:image:alt" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." /><meta property="og:image:width" content="1200" /><meta property="og:image:height" content="600" /><meta property="og:site_name" content="GitHub" /><meta property="og:type" content="object" /><meta property="og:title" content="PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR" /><meta property="og:url" content="https://github.com/PaddlePaddle/PaddleOCR" /><meta property="og:description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." />
-
-    <link rel="assets" href="https://github.githubassets.com/">
-
-
-
-        <meta name="hostname" content="github.com">
-
-
-
-        <meta name="expected-hostname" content="github.com">
-
-
-    <meta name="enabled-features" content="IMAGE_METRIC_TRACKING">
-
-
-  <meta http-equiv="x-pjax-version" content="6c58939d21332326d08c0f7a7a642997da526ff183c0d52243bccca116e52b17" data-turbo-track="reload">
-  <meta http-equiv="x-pjax-csp-version" content="485d6a5ccbb1eeae9c86b616b4870b531f6f458e8bd5c309c40280dc4f51defb" data-turbo-track="reload">
-  <meta http-equiv="x-pjax-css-version" content="4219d8fbdfa0920ce6d03f0d9d444f678ff205e0a832d33151d083532f704774" data-turbo-track="reload">
-  <meta http-equiv="x-pjax-js-version" content="29ca8c5c927eacd73f2fe0e83209b72749f0b338ac4cfd01387c31f0f4001844" data-turbo-track="reload">
-
-  <meta name="turbo-cache-control" content="no-preview"></meta>
-
-
-  <meta name="go-import" content="github.com/PaddlePaddle/PaddleOCR git https://github.com/PaddlePaddle/PaddleOCR.git">
-
-  <meta name="octolytics-dimension-user_id" content="23534030" /><meta name="octolytics-dimension-user_login" content="PaddlePaddle" /><meta name="octolytics-dimension-repository_id" content="262296122" /><meta name="octolytics-dimension-repository_nwo" content="PaddlePaddle/PaddleOCR" /><meta name="octolytics-dimension-repository_public" content="true" /><meta name="octolytics-dimension-repository_is_fork" content="false" /><meta name="octolytics-dimension-repository_network_root_id" content="262296122" /><meta name="octolytics-dimension-repository_network_root_nwo" content="PaddlePaddle/PaddleOCR" />
-
-
-
-    <link rel="canonical" href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-pjax-transient>
-
-
-  <meta name="browser-stats-url" content="https://api.github.com/_private/browser/stats">
-
-  <meta name="browser-errors-url" content="https://api.github.com/_private/browser/errors">
-
-  <meta name="browser-optimizely-client-errors-url" content="https://api.github.com/_private/browser/optimizely_client/errors">
-
-  <link rel="mask-icon" href="https://github.githubassets.com/pinned-octocat.svg" color="#000000">
-  <link rel="alternate icon" class="js-site-favicon" type="image/png" href="https://github.githubassets.com/favicons/favicon.png">
-  <link rel="icon" class="js-site-favicon" type="image/svg+xml" href="https://github.githubassets.com/favicons/favicon.svg">
-
-<meta name="theme-color" content="#1e2327">
-<meta name="color-scheme" content="light dark" />
-
-
-  <link rel="manifest" href="/manifest.json" crossOrigin="use-credentials">
-
-  </head>
-
-  <body class="logged-out env-production page-responsive page-blob" style="word-wrap: break-word;">
-
-
-    <div class="position-relative js-header-wrapper ">
-      <a href="#start-of-content" class="px-2 py-4 color-bg-accent-emphasis color-fg-on-emphasis show-on-focus js-skip-to-content">Skip to content</a>
-      <span data-view-component="true" class="progress-pjax-loader js-pjax-loader-bar Progress position-fixed width-full">
-    <span style="width: 0%;" data-view-component="true" class="Progress-item progress-pjax-loader-bar left-0 top-0 color-bg-accent-emphasis"></span>
-</span>  
-
-
-
-
-
-            <script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-Z1hiPkiRLA+tojAnxZLHuRa1IrpZhnqK66AX75QbEn6PgZIpDJnWcJhQ0pOU2cI6Pe9yO0i3s4JIc9G//vkosQ==" src="https://github.githubassets.com/assets/sessions-6758623e4891.js"></script>
-<header class="Header-old header-logged-out js-details-container Details position-relative f4 py-2" role="banner">
-  <div class="container-xl d-lg-flex flex-items-center p-responsive">
-    <div class="d-flex flex-justify-between flex-items-center">
-      <a class="mr-4 color-fg-inherit" href="https://github.com/" aria-label="Homepage" data-ga-click="(Logged out) Header, go to homepage, icon:logo-wordmark">
-        <svg height="32" aria-hidden="true" viewBox="0 0 16 16" version="1.1" width="32" data-view-component="true" class="octicon octicon-mark-github">
-    <path fill-rule="evenodd" d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"></path>
-</svg>
-      </a>
-
-        <div class="d-lg-none css-truncate css-truncate-target width-fit p-2">
-
-
-        </div>
-
-      <div class="d-flex flex-items-center">
-            <a href="/signup?ref_cta=Sign+up&amp;ref_loc=header+logged+out&amp;ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E%2Fblob%2Fshow&amp;source=header-repo"
-              class="d-inline-block d-lg-none f5 no-underline border color-border-default rounded-2 px-2 py-1 mr-3 mr-sm-5 color-fg-inherit"
-              data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="e697b45b9e440c22e4e3985c3ab81484e1433b0710fb937a52cb9aad23cfc3ca"
-            >
-              Sign&nbsp;up
-            </a>
-
-        <button aria-label="Toggle navigation" aria-expanded="false" type="button" data-view-component="true" class="js-details-target btn-link d-lg-none mt-1 color-fg-inherit">  <svg aria-hidden="true" height="24" viewBox="0 0 16 16" version="1.1" width="24" data-view-component="true" class="octicon octicon-three-bars">
-    <path fill-rule="evenodd" d="M1 2.75A.75.75 0 011.75 2h12.5a.75.75 0 110 1.5H1.75A.75.75 0 011 2.75zm0 5A.75.75 0 011.75 7h12.5a.75.75 0 110 1.5H1.75A.75.75 0 011 7.75zM1.75 12a.75.75 0 100 1.5h12.5a.75.75 0 100-1.5H1.75z"></path>
-</svg>
-
-</button>      </div>
-    </div>
-
-    <div class="HeaderMenu HeaderMenu--logged-out position-fixed top-0 right-0 bottom-0 height-fit position-lg-relative d-lg-flex flex-justify-between flex-items-center flex-auto">
-      <div class="d-flex d-lg-none flex-justify-end border-bottom color-bg-subtle p-3">
-        <button aria-label="Toggle navigation" aria-expanded="false" type="button" data-view-component="true" class="js-details-target btn-link">  <svg aria-hidden="true" height="24" viewBox="0 0 24 24" version="1.1" width="24" data-view-component="true" class="octicon octicon-x color-fg-muted">
-    <path fill-rule="evenodd" d="M5.72 5.72a.75.75 0 011.06 0L12 10.94l5.22-5.22a.75.75 0 111.06 1.06L13.06 12l5.22 5.22a.75.75 0 11-1.06 1.06L12 13.06l-5.22 5.22a.75.75 0 01-1.06-1.06L10.94 12 5.72 6.78a.75.75 0 010-1.06z"></path>
-</svg>
-
-</button>      </div>
-
-        <nav class="mt-0 px-3 px-lg-0 mb-5 mb-lg-0" aria-label="Global">
-          <ul class="d-lg-flex list-style-none">
-              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <details class="HeaderMenu-details details-overlay details-reset width-full">
-      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
-        Product
-        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
-      </summary>
-      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
-        <ul class="list-style-none f5 pb-1">
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Features&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Features;&quot;}" href="/features">
-      Features
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Mobile&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Mobile;&quot;}" href="/mobile">
-      Mobile
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Actions&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Actions;&quot;}" href="/features/actions">
-      Actions
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Codespaces&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Codespaces;&quot;}" href="/features/codespaces">
-      Codespaces
-</a>  </li>
-
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Packages&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Packages;&quot;}" href="/features/packages">
-      Packages
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Security&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Security;&quot;}" href="/features/security">
-      Security
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Code review&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Code review;&quot;}" href="/features/code-review">
-      Code review
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Issues&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Issues;&quot;}" href="/features/issues">
-      Issues
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Integrations&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Integrations;&quot;}" href="/features/integrations">
-      Integrations
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold border-top pt-4 pb-2 mt-3" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to GitHub Sponsors&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Sponsors;&quot;}" href="/sponsors">
-      GitHub Sponsors
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Customer stories&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Customer stories;&quot;}" href="/customer-stories">
-      Customer stories
-</a>  </li>
-
-        </ul>
-      </div>
-    </details>
-</li>
-
-
-              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Team&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Team;&quot;}" href="/team">Team</a>
-</li>
-
-              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Enterprise&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Enterprise;&quot;}" href="/enterprise">Enterprise</a>
-</li>
-
-
-            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <details class="HeaderMenu-details details-overlay details-reset width-full">
-      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
-        Explore
-        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
-      </summary>
-      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
-        <ul class="list-style-none f5 pb-1">
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Explore GitHub&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Explore GitHub;&quot;}" href="/explore">
-      Explore GitHub
-</a>  </li>
-
-              <li class="color-fg-muted text-normal f6 text-mono mb-1 border-top pt-3 mt-3 mb-1">Learn and contribute</li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Topics&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Topics;&quot;}" href="/topics">
-      Topics
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Collections&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Collections;&quot;}" href="/collections">
-      Collections
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Trending&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Trending;&quot;}" href="/trending">
-      Trending
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Skills&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Skills;&quot;}" href="https://skills.github.com/">
-      Skills
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Sponsors&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Sponsors;&quot;}" href="/sponsors/explore">
-      GitHub Sponsors
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Open source guides&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Open source guides;&quot;}" href="https://opensource.guide">
-      Open source guides
-</a>  </li>
-
-              <li class="color-fg-muted text-normal f6 text-mono mb-1 border-top pt-3 mt-3 mb-1">Connect with others</li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to The ReadME Project&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:The ReadME Project;&quot;}" href="/readme">
-      The ReadME Project
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Events&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Events;&quot;}" href="/events">
-      Events
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Community forum&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Community forum;&quot;}" href="https://github.community">
-      Community forum
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Education&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Education;&quot;}" href="https://education.github.com">
-      GitHub Education
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Stars program&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Stars program;&quot;}" href="https://stars.github.com">
-      GitHub Stars program
-</a>  </li>
-
-        </ul>
-      </div>
-    </details>
-</li>
-
-
-            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Marketplace&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Marketplace;&quot;}" href="/marketplace">Marketplace</a>
-</li>
-
-
-            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <details class="HeaderMenu-details details-overlay details-reset width-full">
-      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
-        Pricing
-        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
-      </summary>
-      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
-        <ul class="list-style-none f5 pb-1">
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Plans&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Plans;&quot;}" href="/pricing">
-      Plans
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Compare plans&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Compare plans;&quot;}" href="/pricing#compare-features">
-      Compare plans
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Contact Sales&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Contact Sales;&quot;}" href="https://github.com/enterprise/contact">
-      Contact Sales
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold border-top pt-4 pb-2 mt-3" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Education&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Education;&quot;}" href="https://education.github.com">
-      Education
-</a>  </li>
-
-        </ul>
-      </div>
-    </details>
-</li>
-
-          </ul>
-        </nav>
-
-      <div class="d-lg-flex flex-items-center px-3 px-lg-0 text-center text-lg-left">
-          <div class="d-lg-flex min-width-0 mb-3 mb-lg-0">
-
-
-
-
-<div class="header-search flex-auto js-site-search position-relative flex-self-stretch flex-md-self-auto mb-3 mb-md-0 mr-0 mr-md-3 scoped-search site-scoped-search js-jump-to"
->
-  <div class="position-relative">
-    <!-- '"` --><!-- </textarea></xmp> --></option></form><form class="js-site-search-form" role="search" aria-label="Site" data-scope-type="Repository" data-scope-id="262296122" data-scoped-search-url="/PaddlePaddle/PaddleOCR/search" data-owner-scoped-search-url="/orgs/PaddlePaddle/search" data-unscoped-search-url="/search" data-turbo="false" action="/PaddlePaddle/PaddleOCR/search" accept-charset="UTF-8" method="get">
-      <label class="form-control input-sm header-search-wrapper p-0 js-chromeless-input-container header-search-wrapper-jump-to position-relative d-flex flex-justify-between flex-items-center">
-        <input type="text"
-          class="form-control input-sm header-search-input jump-to-field js-jump-to-field js-site-search-focus js-site-search-field is-clearable"
-          data-hotkey=s,/
-          name="q"
-          data-test-selector="nav-search-input"
-          placeholder="Search"
-          data-unscoped-placeholder="Search GitHub"
-          data-scoped-placeholder="Search"
-          autocapitalize="off"
-          role="combobox"
-          aria-haspopup="listbox"
-          aria-expanded="false"
-          aria-autocomplete="list"
-          aria-controls="jump-to-results"
-          aria-label="Search"
-          data-jump-to-suggestions-path="/_graphql/GetSuggestedNavigationDestinations"
-          spellcheck="false"
-          autocomplete="off"
-        >
-        <input type="hidden" data-csrf="true" class="js-data-jump-to-suggestions-path-csrf" value="cz2/lNe1YAUCKqnoH+OVnKVZRsUzANIx9wkvFLF9HpiPdeWlILE+JGF5aQ6h3DuBmCy+fi/xtF/lq64Lvjposw==" />
-        <input type="hidden" class="js-site-search-type-field" name="type" >
-            <svg xmlns="http://www.w3.org/2000/svg" width="22" height="20" aria-hidden="true" class="mr-1 header-search-key-slash"><path fill="none" stroke="#979A9C" opacity=".4" d="M3.5.5h12c1.7 0 3 1.3 3 3v13c0 1.7-1.3 3-3 3h-12c-1.7 0-3-1.3-3-3v-13c0-1.7 1.3-3 3-3z"></path><path fill="#979A9C" d="M11.8 6L8 15.1h-.9L10.8 6h1z"></path></svg>
-
-
-          <div class="Box position-absolute overflow-hidden d-none jump-to-suggestions js-jump-to-suggestions-container">
-
-<ul class="d-none js-jump-to-suggestions-template-container">
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-suggestion" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="suggestion">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
-        In this repository
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-</ul>
-
-<ul class="d-none js-jump-to-no-results-template-container">
-  <li class="d-flex flex-justify-center flex-items-center f5 d-none js-jump-to-suggestion p-2">
-    <span class="color-fg-muted">No suggested jump to results</span>
-  </li>
-</ul>
-
-<ul id="jump-to-results" role="listbox" class="p-0 m-0 js-navigation-container jump-to-suggestions-results-container js-jump-to-suggestions-results-container">
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-scoped-search d-none" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="scoped_search">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
-        In this repository
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-owner-scoped-search d-none" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="owner_scoped_search">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this organization">
-        In this organization
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-global-search d-none" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="global_search">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
-        In this repository
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-
-</ul>
-
-          </div>
-      </label>
-</form>  </div>
-</div>
-
-          </div>
-
-        <div class="position-relative mr-3 mb-4 mb-lg-0 d-inline-block">
-          <a href="/login?return_to=https%3A%2F%2Fgithub.com%2FPaddlePaddle%2FPaddleOCR%2Fblob%2Fdygraph%2Fapplications%2F%25E8%25BD%25BB%25E9%2587%258F%25E7%25BA%25A7%25E8%25BD%25A6%25E7%2589%258C%25E8%25AF%2586%25E5%2588%25AB.md"
-            class="HeaderMenu-link flex-shrink-0 no-underline"
-            data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header menu&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="0de5d29e004c06e7301cb66a02d668f7cd98725fd6fa97ab9b512671e1a9905b"
-            data-ga-click="(Logged out) Header, clicked Sign in, text:sign-in">
-            Sign in
-          </a>
-        </div>
-
-          <a href="/signup?ref_cta=Sign+up&amp;ref_loc=header+logged+out&amp;ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E%2Fblob%2Fshow&amp;source=header-repo&amp;source_repo=PaddlePaddle%2FPaddleOCR"
-            class="HeaderMenu-link flex-shrink-0 d-inline-block no-underline border color-border-default rounded px-2 py-1"
-            data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header menu&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="0de5d29e004c06e7301cb66a02d668f7cd98725fd6fa97ab9b512671e1a9905b"
-            data-analytics-event="{&quot;category&quot;:&quot;Sign up&quot;,&quot;action&quot;:&quot;click to sign up for account&quot;,&quot;label&quot;:&quot;ref_page:/&lt;user-name&gt;/&lt;repo-name&gt;/blob/show;ref_cta:Sign up;ref_loc:header logged out&quot;}"
-          >
-            Sign up
-          </a>
-      </div>
-    </div>
-  </div>
-</header>
-
-    </div>
-
-  <div id="start-of-content" class="show-on-focus"></div>
-
-
-
-
-
-
-
-    <div data-pjax-replace id="js-flash-container" data-turbo-replace>
-
-
-  <template class="js-flash-template">
-    <div class="flash flash-full  {{ className }}">
-  <div class="px-2" >
-    <button class="flash-close js-flash-close" type="button" aria-label="Dismiss this message">
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-    </button>
-
-      <div>{{ message }}</div>
-
-  </div>
-</div>
-  </template>
-</div>
-
-
-
-  <include-fragment class="js-notification-shelf-include-fragment" data-base-src="https://github.com/notifications/beta/shelf"></include-fragment>
-
-
-
-
-
-  <div
-    class="application-main "
-    data-commit-hovercards-enabled
-    data-discussion-hovercards-enabled
-    data-issue-and-pr-hovercards-enabled
-  >
-        <div itemscope itemtype="http://schema.org/SoftwareSourceCode" class="">
-    <main id="js-repo-pjax-container" data-pjax-container >
-
-
-
-
-
-
-
-
-
-
-
-
-
-  <div id="repository-container-header" class="pt-3 hide-full-screen" style="background-color: var(--color-page-header-bg);" data-pjax-replace data-turbo-replace>
-
-      <div class="d-flex mb-3 px-3 px-md-4 px-lg-5">
-
-        <div class="flex-auto min-width-0 width-fit mr-3">
-            <div class=" d-flex flex-wrap flex-items-center wb-break-word f3 text-normal">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo color-fg-muted mr-2">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-  <span class="author flex-self-stretch" itemprop="author">
-    <a class="url fn" rel="author" data-hovercard-type="organization" data-hovercard-url="/orgs/PaddlePaddle/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle">PaddlePaddle</a>
-  </span>
-  <span class="mx-1 flex-self-stretch color-fg-muted">/</span>
-  <strong itemprop="name" class="mr-2 flex-self-stretch">
-    <a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR">PaddleOCR</a>
-  </strong>
-
-  <span></span><span class="Label Label--secondary v-align-middle mr-1">Public</span>
-</div>
-
-        </div>
-
-          <ul class="pagehead-actions flex-shrink-0 d-none d-md-inline" style="padding: 2px 0;">
-
-
-
-  <li>
-      <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;notification subscription menu watch&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="2c00da16a0d8d066b49f1f6e26149784faffba39c427495160ebf3bac257b4a9" aria-label="You must be signed in to change notification settings" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-bell mr-2">
-    <path d="M8 16a2 2 0 001.985-1.75c.017-.137-.097-.25-.235-.25h-3.5c-.138 0-.252.113-.235.25A2 2 0 008 16z"></path><path fill-rule="evenodd" d="M8 1.5A3.5 3.5 0 004.5 5v2.947c0 .346-.102.683-.294.97l-1.703 2.556a.018.018 0 00-.003.01l.001.006c0 .002.002.004.004.006a.017.017 0 00.006.004l.007.001h10.964l.007-.001a.016.016 0 00.006-.004.016.016 0 00.004-.006l.001-.007a.017.017 0 00-.003-.01l-1.703-2.554a1.75 1.75 0 01-.294-.97V5A3.5 3.5 0 008 1.5zM3 5a5 5 0 0110 0v2.947c0 .05.015.098.042.139l1.703 2.555A1.518 1.518 0 0113.482 13H2.518a1.518 1.518 0 01-1.263-2.36l1.703-2.554A.25.25 0 003 7.947V5z"></path>
-</svg>Notifications
-
-</a>
-  </li>
-
-  <li>
-        <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;repo details fork button&quot;,&quot;repository_id&quot;:262296122,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f55097530c92c084f3a79350400121a7c267c872da2b69858747a12db51b5c71" aria-label="You must be signed in to fork a repository" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo-forked mr-2">
-    <path fill-rule="evenodd" d="M5 3.25a.75.75 0 11-1.5 0 .75.75 0 011.5 0zm0 2.122a2.25 2.25 0 10-1.5 0v.878A2.25 2.25 0 005.75 8.5h1.5v2.128a2.251 2.251 0 101.5 0V8.5h1.5a2.25 2.25 0 002.25-2.25v-.878a2.25 2.25 0 10-1.5 0v.878a.75.75 0 01-.75.75h-4.5A.75.75 0 015 6.25v-.878zm3.75 7.378a.75.75 0 11-1.5 0 .75.75 0 011.5 0zm3-8.75a.75.75 0 100-1.5.75.75 0 000 1.5z"></path>
-</svg>Fork
-    <span id="repo-network-counter" data-pjax-replace="true" title="4,668" data-view-component="true" class="Counter">4.7k</span>
-
-</a>
-  </li>
-
-  <li>
-        <div data-view-component="true" class="BtnGroup d-flex">
-      <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;star button&quot;,&quot;repository_id&quot;:262296122,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="93ad0192b3e15b4610691426f776e3dab114d4c7666db7f44c0d706f50c74c23" aria-label="You must be signed in to star a repository" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-star v-align-text-bottom d-inline-block mr-2">
-    <path fill-rule="evenodd" d="M8 .25a.75.75 0 01.673.418l1.882 3.815 4.21.612a.75.75 0 01.416 1.279l-3.046 2.97.719 4.192a.75.75 0 01-1.088.791L8 12.347l-3.766 1.98a.75.75 0 01-1.088-.79l.72-4.194L.818 6.374a.75.75 0 01.416-1.28l4.21-.611L7.327.668A.75.75 0 018 .25zm0 2.445L6.615 5.5a.75.75 0 01-.564.41l-3.097.45 2.24 2.184a.75.75 0 01.216.664l-.528 3.084 2.769-1.456a.75.75 0 01.698 0l2.77 1.456-.53-3.084a.75.75 0 01.216-.664l2.24-2.183-3.096-.45a.75.75 0 01-.564-.41L8 2.694v.001z"></path>
-</svg><span data-view-component="true" class="d-inline">
-          Star
-</span>          <span id="repo-stars-counter-star" aria-label="22861 users starred this repository" data-singular-suffix="user starred this repository" data-plural-suffix="users starred this repository" data-pjax-replace="true" data-turbo-replace="true" title="22,861" data-view-component="true" class="Counter js-social-count">22.9k</span>
-
-</a>      <button disabled="disabled" aria-label="You must be signed in to add this repository to a list" type="button" data-view-component="true" class="btn-sm btn BtnGroup-item px-2">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-triangle-down">
-    <path d="M4.427 7.427l3.396 3.396a.25.25 0 00.354 0l3.396-3.396A.25.25 0 0011.396 7H4.604a.25.25 0 00-.177.427z"></path>
-</svg>
-
-</button></div>
-  </li>
-
-
-
-</ul>
-
-      </div>
-
-      <div id="responsive-meta-container" data-pjax-replace data-turbo-replace>
-</div>
-
-
-
-<nav data-pjax="#js-repo-pjax-container" aria-label="Repository" data-view-component="true" class="js-repo-nav js-sidenav-container-pjax js-responsive-underlinenav overflow-hidden UnderlineNav px-3 px-md-4 px-lg-5">
-
-  <ul data-view-component="true" class="UnderlineNav-body list-style-none">
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="code-tab" href="/PaddlePaddle/PaddleOCR/tree/dygraph" data-tab-item="i0code-tab" data-selected-links="repo_source repo_downloads repo_commits repo_releases repo_tags repo_branches repo_packages repo_deployments /PaddlePaddle/PaddleOCR/tree/dygraph" data-pjax="#repo-content-pjax-container" data-hotkey="g c" data-ga-click="Repository, Navigation click, Code tab" aria-current="page" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item selected">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-code UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M4.72 3.22a.75.75 0 011.06 1.06L2.06 8l3.72 3.72a.75.75 0 11-1.06 1.06L.47 8.53a.75.75 0 010-1.06l4.25-4.25zm6.56 0a.75.75 0 10-1.06 1.06L13.94 8l-3.72 3.72a.75.75 0 101.06 1.06l4.25-4.25a.75.75 0 000-1.06l-4.25-4.25z"></path>
-</svg>
-          <span data-content="Code">Code</span>
-            <span id="code-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="issues-tab" href="/PaddlePaddle/PaddleOCR/issues" data-tab-item="i1issues-tab" data-selected-links="repo_issues repo_labels repo_milestones /PaddlePaddle/PaddleOCR/issues" data-pjax="#repo-content-pjax-container" data-hotkey="g i" data-ga-click="Repository, Navigation click, Issues tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-issue-opened UnderlineNav-octicon d-none d-sm-inline">
-    <path d="M8 9.5a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path><path fill-rule="evenodd" d="M8 0a8 8 0 100 16A8 8 0 008 0zM1.5 8a6.5 6.5 0 1113 0 6.5 6.5 0 01-13 0z"></path>
-</svg>
-          <span data-content="Issues">Issues</span>
-            <span id="issues-repo-tab-count" data-pjax-replace="" title="1,078" data-view-component="true" class="Counter">1.1k</span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="pull-requests-tab" href="/PaddlePaddle/PaddleOCR/pulls" data-tab-item="i2pull-requests-tab" data-selected-links="repo_pulls checks /PaddlePaddle/PaddleOCR/pulls" data-pjax="#repo-content-pjax-container" data-hotkey="g p" data-ga-click="Repository, Navigation click, Pull requests tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-git-pull-request UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M7.177 3.073L9.573.677A.25.25 0 0110 .854v4.792a.25.25 0 01-.427.177L7.177 3.427a.25.25 0 010-.354zM3.75 2.5a.75.75 0 100 1.5.75.75 0 000-1.5zm-2.25.75a2.25 2.25 0 113 2.122v5.256a2.251 2.251 0 11-1.5 0V5.372A2.25 2.25 0 011.5 3.25zM11 2.5h-1V4h1a1 1 0 011 1v5.628a2.251 2.251 0 101.5 0V5A2.5 2.5 0 0011 2.5zm1 10.25a.75.75 0 111.5 0 .75.75 0 01-1.5 0zM3.75 12a.75.75 0 100 1.5.75.75 0 000-1.5z"></path>
-</svg>
-          <span data-content="Pull requests">Pull requests</span>
-            <span id="pull-requests-repo-tab-count" data-pjax-replace="" title="97" data-view-component="true" class="Counter">97</span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="discussions-tab" href="/PaddlePaddle/PaddleOCR/discussions" data-tab-item="i3discussions-tab" data-selected-links="repo_discussions /PaddlePaddle/PaddleOCR/discussions" data-pjax="#repo-content-pjax-container" data-hotkey="g g" data-ga-click="Repository, Navigation click, Discussions tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-comment-discussion UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M1.5 2.75a.25.25 0 01.25-.25h8.5a.25.25 0 01.25.25v5.5a.25.25 0 01-.25.25h-3.5a.75.75 0 00-.53.22L3.5 11.44V9.25a.75.75 0 00-.75-.75h-1a.25.25 0 01-.25-.25v-5.5zM1.75 1A1.75 1.75 0 000 2.75v5.5C0 9.216.784 10 1.75 10H2v1.543a1.457 1.457 0 002.487 1.03L7.061 10h3.189A1.75 1.75 0 0012 8.25v-5.5A1.75 1.75 0 0010.25 1h-8.5zM14.5 4.75a.25.25 0 00-.25-.25h-.5a.75.75 0 110-1.5h.5c.966 0 1.75.784 1.75 1.75v5.5A1.75 1.75 0 0114.25 12H14v1.543a1.457 1.457 0 01-2.487 1.03L9.22 12.28a.75.75 0 111.06-1.06l2.22 2.22v-2.19a.75.75 0 01.75-.75h1a.25.25 0 00.25-.25v-5.5z"></path>
-</svg>
-          <span data-content="Discussions">Discussions</span>
-            <span id="discussions-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="actions-tab" href="/PaddlePaddle/PaddleOCR/actions" data-tab-item="i4actions-tab" data-selected-links="repo_actions /PaddlePaddle/PaddleOCR/actions" data-pjax="#repo-content-pjax-container" data-hotkey="g a" data-analytics-event="{&quot;category&quot;:&quot;Actions&quot;,&quot;action&quot;:&quot;clicked&quot;,&quot;label&quot;:&quot;ref_cta:Actions;ref_loc:navigation_helper&quot;}" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-play UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M1.5 8a6.5 6.5 0 1113 0 6.5 6.5 0 01-13 0zM8 0a8 8 0 100 16A8 8 0 008 0zM6.379 5.227A.25.25 0 006 5.442v5.117a.25.25 0 00.379.214l4.264-2.559a.25.25 0 000-.428L6.379 5.227z"></path>
-</svg>
-          <span data-content="Actions">Actions</span>
-            <span id="actions-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="projects-tab" href="/PaddlePaddle/PaddleOCR/projects?type=beta" data-tab-item="i5projects-tab" data-selected-links="repo_projects new_repo_project repo_project /PaddlePaddle/PaddleOCR/projects?type=beta" data-pjax="#repo-content-pjax-container" data-hotkey="g b" data-ga-click="Repository, Navigation click, Projects tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-table UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M0 1.75C0 .784.784 0 1.75 0h12.5C15.216 0 16 .784 16 1.75v3.585a.746.746 0 010 .83v8.085A1.75 1.75 0 0114.25 16H6.309a.748.748 0 01-1.118 0H1.75A1.75 1.75 0 010 14.25V6.165a.746.746 0 010-.83V1.75zM1.5 6.5v7.75c0 .138.112.25.25.25H5v-8H1.5zM5 5H1.5V1.75a.25.25 0 01.25-.25H5V5zm1.5 1.5v8h7.75a.25.25 0 00.25-.25V6.5h-8zm8-1.5h-8V1.5h7.75a.25.25 0 01.25.25V5z"></path>
-</svg>
-          <span data-content="Projects">Projects</span>
-            <span id="projects-repo-tab-count" data-pjax-replace="" title="0" hidden="hidden" data-view-component="true" class="Counter">0</span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="wiki-tab" href="/PaddlePaddle/PaddleOCR/wiki" data-tab-item="i6wiki-tab" data-selected-links="repo_wiki /PaddlePaddle/PaddleOCR/wiki" data-pjax="#repo-content-pjax-container" data-hotkey="g w" data-ga-click="Repository, Navigation click, Wikis tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-book UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M0 1.75A.75.75 0 01.75 1h4.253c1.227 0 2.317.59 3 1.501A3.744 3.744 0 0111.006 1h4.245a.75.75 0 01.75.75v10.5a.75.75 0 01-.75.75h-4.507a2.25 2.25 0 00-1.591.659l-.622.621a.75.75 0 01-1.06 0l-.622-.621A2.25 2.25 0 005.258 13H.75a.75.75 0 01-.75-.75V1.75zm8.755 3a2.25 2.25 0 012.25-2.25H14.5v9h-3.757c-.71 0-1.4.201-1.992.572l.004-7.322zm-1.504 7.324l.004-5.073-.002-2.253A2.25 2.25 0 005.003 2.5H1.5v9h3.757a3.75 3.75 0 011.994.574z"></path>
-</svg>
-          <span data-content="Wiki">Wiki</span>
-            <span id="wiki-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="security-tab" href="/PaddlePaddle/PaddleOCR/security" data-tab-item="i7security-tab" data-selected-links="security overview alerts policy token_scanning code_scanning /PaddlePaddle/PaddleOCR/security" data-pjax="#repo-content-pjax-container" data-hotkey="g s" data-ga-click="Repository, Navigation click, Security tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-shield UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M7.467.133a1.75 1.75 0 011.066 0l5.25 1.68A1.75 1.75 0 0115 3.48V7c0 1.566-.32 3.182-1.303 4.682-.983 1.498-2.585 2.813-5.032 3.855a1.7 1.7 0 01-1.33 0c-2.447-1.042-4.049-2.357-5.032-3.855C1.32 10.182 1 8.566 1 7V3.48a1.75 1.75 0 011.217-1.667l5.25-1.68zm.61 1.429a.25.25 0 00-.153 0l-5.25 1.68a.25.25 0 00-.174.238V7c0 1.358.275 2.666 1.057 3.86.784 1.194 2.121 2.34 4.366 3.297a.2.2 0 00.154 0c2.245-.956 3.582-2.104 4.366-3.298C13.225 9.666 13.5 8.36 13.5 7V3.48a.25.25 0 00-.174-.237l-5.25-1.68zM9 10.5a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.75a.75.75 0 10-1.5 0v3a.75.75 0 001.5 0v-3z"></path>
-</svg>
-          <span data-content="Security">Security</span>
-            <include-fragment src="/PaddlePaddle/PaddleOCR/security/overall-count" accept="text/fragment+html"></include-fragment>
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="insights-tab" href="/PaddlePaddle/PaddleOCR/pulse" data-tab-item="i8insights-tab" data-selected-links="repo_graphs repo_contributors dependency_graph dependabot_updates pulse people community /PaddlePaddle/PaddleOCR/pulse" data-pjax="#repo-content-pjax-container" data-ga-click="Repository, Navigation click, Insights tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-graph UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M1.5 1.75a.75.75 0 00-1.5 0v12.5c0 .414.336.75.75.75h14.5a.75.75 0 000-1.5H1.5V1.75zm14.28 2.53a.75.75 0 00-1.06-1.06L10 7.94 7.53 5.47a.75.75 0 00-1.06 0L3.22 8.72a.75.75 0 001.06 1.06L7 7.06l2.47 2.47a.75.75 0 001.06 0l5.25-5.25z"></path>
-</svg>
-          <span data-content="Insights">Insights</span>
-            <span id="insights-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-</ul>
-    <div style="visibility:hidden;" data-view-component="true" class="UnderlineNav-actions js-responsive-underlinenav-overflow position-absolute pr-3 pr-md-4 pr-lg-5 right-0">      <details data-view-component="true" class="details-overlay details-reset position-relative">
-  <summary role="button" data-view-component="true">          <div class="UnderlineNav-item mr-0 border-0">
-            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
-    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
-</svg>
-            <span class="sr-only">More</span>
-          </div>
-</summary>
-  <div data-view-component="true">          <details-menu role="menu" data-view-component="true" class="dropdown-menu dropdown-menu-sw">
-
-            <ul>
-                <li data-menu-item="i0code-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item selected dropdown-item" aria-current="page" data-selected-links="repo_source repo_downloads repo_commits repo_releases repo_tags repo_branches repo_packages repo_deployments /PaddlePaddle/PaddleOCR/tree/dygraph" href="/PaddlePaddle/PaddleOCR/tree/dygraph">
-                    Code
-</a>                </li>
-                <li data-menu-item="i1issues-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_issues repo_labels repo_milestones /PaddlePaddle/PaddleOCR/issues" href="/PaddlePaddle/PaddleOCR/issues">
-                    Issues
-</a>                </li>
-                <li data-menu-item="i2pull-requests-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_pulls checks /PaddlePaddle/PaddleOCR/pulls" href="/PaddlePaddle/PaddleOCR/pulls">
-                    Pull requests
-</a>                </li>
-                <li data-menu-item="i3discussions-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_discussions /PaddlePaddle/PaddleOCR/discussions" href="/PaddlePaddle/PaddleOCR/discussions">
-                    Discussions
-</a>                </li>
-                <li data-menu-item="i4actions-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_actions /PaddlePaddle/PaddleOCR/actions" href="/PaddlePaddle/PaddleOCR/actions">
-                    Actions
-</a>                </li>
-                <li data-menu-item="i5projects-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_projects new_repo_project repo_project /PaddlePaddle/PaddleOCR/projects?type=beta" href="/PaddlePaddle/PaddleOCR/projects?type=beta">
-                    Projects
-</a>                </li>
-                <li data-menu-item="i6wiki-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_wiki /PaddlePaddle/PaddleOCR/wiki" href="/PaddlePaddle/PaddleOCR/wiki">
-                    Wiki
-</a>                </li>
-                <li data-menu-item="i7security-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="security overview alerts policy token_scanning code_scanning /PaddlePaddle/PaddleOCR/security" href="/PaddlePaddle/PaddleOCR/security">
-                    Security
-</a>                </li>
-                <li data-menu-item="i8insights-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_graphs repo_contributors dependency_graph dependabot_updates pulse people community /PaddlePaddle/PaddleOCR/pulse" href="/PaddlePaddle/PaddleOCR/pulse">
-                    Insights
-</a>                </li>
-            </ul>
-
-</details-menu></div>
-</details></div>
-</nav>
-  </div>
-
-
-
-<div id="repo-content-pjax-container" class="repository-content " >
-
-
-
-
-
-
-  <div class="clearfix container-xl px-3 px-md-4 px-lg-5 mt-4">
-
-
-<div>
-
-
-
-
-
-
-
-
-<a class="d-none js-permalink-shortcut" data-hotkey="y" href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">Permalink</a>
-
-<div class="d-flex flex-items-start flex-shrink-0 pb-3 flex-wrap flex-md-nowrap flex-justify-between flex-md-justify-start">
-
-<div class="position-relative">
-  <details class="details-reset details-overlay mr-0 mb-0 " id="branch-select-menu">
-    <summary class="btn css-truncate"
-            data-hotkey="w"
-            title="Switch branches or tags">
-      <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-git-branch">
-    <path fill-rule="evenodd" d="M11.75 2.5a.75.75 0 100 1.5.75.75 0 000-1.5zm-2.25.75a2.25 2.25 0 113 2.122V6A2.5 2.5 0 0110 8.5H6a1 1 0 00-1 1v1.128a2.251 2.251 0 11-1.5 0V5.372a2.25 2.25 0 111.5 0v1.836A2.492 2.492 0 016 7h4a1 1 0 001-1v-.628A2.25 2.25 0 019.5 3.25zM4.25 12a.75.75 0 100 1.5.75.75 0 000-1.5zM3.5 3.25a.75.75 0 111.5 0 .75.75 0 01-1.5 0z"></path>
-</svg>
-      <span class="css-truncate-target" data-menu-button>dygraph</span>
-      <span class="dropdown-caret"></span>
-    </summary>
-
-
-<div class="SelectMenu">
-  <div class="SelectMenu-modal">
-    <header class="SelectMenu-header">
-      <span class="SelectMenu-title">Switch branches/tags</span>
-      <button class="SelectMenu-closeButton" type="button" data-toggle-for="branch-select-menu"><svg aria-label="Close menu" aria-hidden="false" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg></button>
-    </header>
-
-    <input-demux data-action="tab-container-change:input-demux#storeInput tab-container-changed:input-demux#updateInput">
-      <tab-container class="d-flex flex-column js-branches-tags-tabs" style="min-height: 0;">
-        <div class="SelectMenu-filter">
-          <input data-target="input-demux.source"
-                 id="context-commitish-filter-field"
-                 class="SelectMenu-input form-control"
-                 aria-owns="ref-list-branches"
-                 data-controls-ref-menu-id="ref-list-branches"
-                 autofocus
-                 autocomplete="off"
-                 aria-label="Filter branches/tags"
-                 placeholder="Filter branches/tags"
-                 type="text"
-          >
-        </div>
-
-        <div class="SelectMenu-tabs" role="tablist" data-target="input-demux.control" >
-          <button class="SelectMenu-tab" type="button" role="tab" aria-selected="true">Branches</button>
-          <button class="SelectMenu-tab" type="button" role="tab">Tags</button>
-        </div>
-
-        <div role="tabpanel" id="ref-list-branches" data-filter-placeholder="Filter branches/tags" tabindex="" class="d-flex flex-column flex-auto overflow-auto">
-          <ref-selector
-            type="branch"
-            data-targets="input-demux.sinks"
-            data-action="
-              input-entered:ref-selector#inputEntered
-              tab-selected:ref-selector#tabSelected
-              focus-list:ref-selector#focusFirstListMember
-            "
-            query-endpoint="/PaddlePaddle/PaddleOCR/refs"
-
-            cache-key="v0:1655177037.9190001"
-            current-committish="ZHlncmFwaA=="
-            default-branch="cmVsZWFzZS8yLjU="
-            name-with-owner="UGFkZGxlUGFkZGxlL1BhZGRsZU9DUg=="
-            prefetch-on-mouseover
-          >
-
-            <template data-target="ref-selector.fetchFailedTemplate">
-              <div class="SelectMenu-message" data-index="{{ index }}">Could not load branches</div>
-            </template>
-
-              <template data-target="ref-selector.noMatchTemplate">
-    <div class="SelectMenu-message">Nothing to show</div>
-</template>
-
-
-            <div data-target="ref-selector.listContainer" role="menu" class="SelectMenu-list " data-pjax="#repo-content-pjax-container">
-              <div class="SelectMenu-loading pt-3 pb-0 overflow-hidden" aria-label="Menu is loading">
-                <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="anim-rotate">
-  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
-  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
-</svg>
-              </div>
-            </div>
-
-              <template data-target="ref-selector.itemTemplate">
-  <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{ urlEncodedRefName }}/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="SelectMenu-item" role="menuitemradio" rel="nofollow" aria-checked="{{ isCurrent }}" data-index="{{ index }}">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check SelectMenu-icon SelectMenu-icon--check">
-    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
-</svg>
-    <span class="flex-1 css-truncate css-truncate-overflow {{ isFilteringClass }}">{{ refName }}</span>
-    <span hidden="{{ isNotDefault }}" class="Label Label--secondary flex-self-start">default</span>
-  </a>
-</template>
-
-
-              <footer class="SelectMenu-footer"><a href="/PaddlePaddle/PaddleOCR/branches">View all branches</a></footer>
-          </ref-selector>
-
-        </div>
-
-        <div role="tabpanel" id="tags-menu" data-filter-placeholder="Find a tag" tabindex="" hidden class="d-flex flex-column flex-auto overflow-auto">
-          <ref-selector
-            type="tag"
-            data-action="
-              input-entered:ref-selector#inputEntered
-              tab-selected:ref-selector#tabSelected
-              focus-list:ref-selector#focusFirstListMember
-            "
-            data-targets="input-demux.sinks"
-            query-endpoint="/PaddlePaddle/PaddleOCR/refs"
-            cache-key="v0:1655177037.9190001"
-            current-committish="ZHlncmFwaA=="
-            default-branch="cmVsZWFzZS8yLjU="
-            name-with-owner="UGFkZGxlUGFkZGxlL1BhZGRsZU9DUg=="
-          >
-
-            <template data-target="ref-selector.fetchFailedTemplate">
-              <div class="SelectMenu-message" data-index="{{ index }}">Could not load tags</div>
-            </template>
-
-            <template data-target="ref-selector.noMatchTemplate">
-              <div class="SelectMenu-message" data-index="{{ index }}">Nothing to show</div>
-            </template>
-
-              <template data-target="ref-selector.itemTemplate">
-  <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{ urlEncodedRefName }}/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="SelectMenu-item" role="menuitemradio" rel="nofollow" aria-checked="{{ isCurrent }}" data-index="{{ index }}">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check SelectMenu-icon SelectMenu-icon--check">
-    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
-</svg>
-    <span class="flex-1 css-truncate css-truncate-overflow {{ isFilteringClass }}">{{ refName }}</span>
-    <span hidden="{{ isNotDefault }}" class="Label Label--secondary flex-self-start">default</span>
-  </a>
-</template>
-
-
-            <div data-target="ref-selector.listContainer" role="menu" class="SelectMenu-list" data-pjax="#repo-content-pjax-container">
-              <div class="SelectMenu-loading pt-3 pb-0 overflow-hidden" aria-label="Menu is loading">
-                <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="anim-rotate">
-  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
-  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
-</svg>
-              </div>
-            </div>
-              <footer class="SelectMenu-footer"><a href="/PaddlePaddle/PaddleOCR/tags">View all tags</a></footer>
-          </ref-selector>
-        </div>
-      </tab-container>
-    </input-demux>
-  </div>
-</div>
-
-  </details>
-
-</div>
-
-  <h2 id="blob-path" class="breadcrumb flex-auto flex-self-center min-width-0 text-normal mx-2 width-full width-md-auto flex-order-1 flex-md-order-none mt-3 mt-md-0">
-    <span class="js-repo-root text-bold"><span class="js-path-segment d-inline-block wb-break-all"><a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR/tree/dygraph"><span>PaddleOCR</span></a></span></span><span class="separator">/</span><span class="js-path-segment d-inline-block wb-break-all"><a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR/tree/dygraph/applications"><span>applications</span></a></span><span class="separator">/</span><strong class="final-path">轻量级车牌识别.md</strong>
-  </h2>
-  <a href="/PaddlePaddle/PaddleOCR/find/dygraph" data-pjax="" data-hotkey="t" data-view-component="true" class="btn mr-2 d-none d-md-block">  Go to file
-
-</a>
-  <details id="blob-more-options-details" data-view-component="true" class="details-overlay details-reset position-relative">
-  <summary role="button" data-view-component="true" class="btn">  <svg aria-label="More options" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
-    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
-</svg>
-
-</summary>
-  <div data-view-component="true">      <ul class="dropdown-menu dropdown-menu-sw">
-        <li class="d-block d-md-none">
-          <a class="dropdown-item d-flex flex-items-baseline" data-hydro-click="{&quot;event_type&quot;:&quot;repository.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;FIND_FILE_BUTTON&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="719caca13290ec1595dbf25afc93e9874844c0bcf4645d121411be9e6fbaa205" data-ga-click="Repository, find file, location:repo overview" data-hotkey="t" data-pjax="true" href="/PaddlePaddle/PaddleOCR/find/dygraph">
-            <span class="flex-auto">Go to file</span>
-            <span class="text-small color-fg-muted" aria-hidden="true">T</span>
-</a>        </li>
-        <li data-toggle-for="blob-more-options-details">
-          <button data-toggle-for="jumpto-line-details-dialog" type="button" data-view-component="true" class="dropdown-item btn-link">  <span class="d-flex flex-items-baseline">
-              <span class="flex-auto">Go to line</span>
-              <span class="text-small color-fg-muted" aria-hidden="true">L</span>
-            </span>
-
-</button>        </li>
-        <li class="dropdown-divider" role="none"></li>
-        <li>
-          <clipboard-copy data-toggle-for="blob-more-options-details" aria-label="Copy path" value="applications/轻量级车牌识别.md" data-view-component="true" class="dropdown-item cursor-pointer">
-
-            Copy path
-
-</clipboard-copy>        </li>
-        <li>
-          <clipboard-copy data-toggle-for="blob-more-options-details" aria-label="Copy permalink" value="https://github.com/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-view-component="true" class="dropdown-item cursor-pointer">
-
-            <span class="d-flex flex-items-baseline">
-              <span class="flex-auto">Copy permalink</span>
-            </span>
-
-</clipboard-copy>        </li>
-      </ul>
-</div>
-</details></div>
-
-
-
-
-
-    <div id="spoof-warning" class="mt-0 pb-3" hidden aria-hidden>
-  <div data-view-component="true" class="flash flash-warn mt-0 clearfix">
-
-
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert float-left mt-1">
-    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
-</svg>
-
-      <div class="overflow-hidden">This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.</div>
-
-
-
-</div></div>
-
-    <include-fragment src="/PaddlePaddle/PaddleOCR/spoofed_commit_check/a3dfc10b7824743939976c42db6d50d118e2b3b9" data-test-selector="spoofed-commit-check"></include-fragment>
-
-    <div class="Box d-flex flex-column flex-shrink-0 mb-3">
-  <include-fragment src="/PaddlePaddle/PaddleOCR/contributors/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="commit-loader">
-    <div class="Box-header d-flex flex-items-center">
-      <div class="Skeleton avatar avatar-user flex-shrink-0 ml-n1 mr-n1 mt-n1 mb-n1" style="width:24px;height:24px;"></div>
-      <div class="Skeleton Skeleton--text col-5 ml-2">&nbsp;</div>
-    </div>
-
-    <div class="Box-body d-flex flex-items-center" >
-      <div class="Skeleton Skeleton--text col-1">&nbsp;</div>
-      <span class="color-fg-danger h6 loader-error">Cannot retrieve contributors at this time</span>
-    </div>
-</include-fragment></div>
-
-
-
-
-
-
-
-    <readme-toc>
-
-    <div data-target="readme-toc.content" class="Box mt-3 position-relative">
-
-  <div
-    class="Box-header js-blob-header blob-header js-sticky js-position-sticky top-0 p-2 d-flex flex-shrink-0 flex-md-row flex-items-center"
-    style="position: sticky; z-index: 1;"
-  >
-
-      <details
-  data-target="readme-toc.trigger"
-  data-menu-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;trigger&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}"
-  data-menu-hydro-click-hmac="532f79aab7b37e9668b558163fc4a6d01f017a94245e04831a2e20c0a43fff35"
-  class="dropdown details-reset details-overlay"
->
-  <summary
-    class="btn btn-octicon m-0 mr-2 p-2"
-    aria-haspopup="true"
-    aria-label="Table of Contents">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-list-unordered">
-    <path fill-rule="evenodd" d="M2 4a1 1 0 100-2 1 1 0 000 2zm3.75-1.5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zm0 5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zm0 5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zM3 8a1 1 0 11-2 0 1 1 0 012 0zm-1 6a1 1 0 100-2 1 1 0 000 2z"></path>
-</svg>
-  </summary>
-
-
-  <details-menu class="SelectMenu" role="menu">
-    <div class="SelectMenu-modal rounded-3 mt-1" style="max-height:340px;">
-
-        <div class="SelectMenu-filter">
-          <input
-            class="SelectMenu-input form-control js-filterable-field"
-            id="toc-filter-field"
-            type="text"
-            autocomplete="off"
-            spellcheck="false"
-            autofocus
-            placeholder="Filter headings"
-            aria-label="Filter headings">
-        </div>
-
-      <div class="SelectMenu-list SelectMenu-list--borderless p-2" style="overscroll-behavior: contain;" data-filterable-for="toc-filter-field" data-filterable-type="substring">
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 text-emphasized" style="-webkit-box-orient: vertical; padding-left: 12px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#一种基于paddleocr的轻量级车牌识别模型">一种基于PaddleOCR的轻量级车牌识别模型</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#1-项目介绍">1. 项目介绍</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#2-环境搭建">2. 环境搭建</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#3-数据集准备">3. 数据集准备</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#31-数据集标注规则">3.1 数据集标注规则</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#32-制作符合pp-ocr训练格式的标注文件">3.2 制作符合PP-OCR训练格式的标注文件</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#4-实验">4. 实验</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#41-检测">4.1 检测</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#411-预训练模型直接预测">4.1.1 预训练模型直接预测</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#412-ccpd车牌数据集fine-tune">4.1.2 CCPD车牌数据集fine-tune</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#413-ccpd车牌数据集fine-tune量化训练">4.1.3 CCPD车牌数据集fine-tune+量化训练</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#414-模型导出">4.1.4 模型导出</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#42-识别">4.2 识别</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#421-预训练模型直接预测">4.2.1 预训练模型直接预测</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#422-预训练模型直接预测改动后处理">4.2.2 预训练模型直接预测+改动后处理</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#423-ccpd车牌数据集fine-tune">4.2.3 CCPD车牌数据集fine-tune</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#424-ccpd车牌数据集fine-tune量化训练">4.2.4 CCPD车牌数据集fine-tune+量化训练</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#425-模型导出">4.2.5 模型导出</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#43-计算end2end指标">4.3 计算End2End指标</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#44-部署">4.4 部署</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#45-实验总结">4.5 实验总结</a>
-      </div>
-    </div>
-  </details-menu>
-</details>
-
-
-  <div class="text-mono f6 flex-auto pr-3 flex-order-2 flex-md-order-1">
-
-      831 lines (628 sloc)
-      <span class="file-info-divider"></span>
-    41.2 KB
-  </div>
-
-  <div class="d-flex py-1 py-md-0 flex-auto flex-order-1 flex-md-order-2 flex-sm-grow-0 flex-justify-between hide-sm hide-md">
-        <div class="BtnGroup">
-    <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?plain=1" data-permalink-href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?plain=1" aria-label="Display the source blob" data-view-component="true" class="source tooltipped tooltipped tooltipped-n  js-permalink-replaceable-link btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-code">
-    <path fill-rule="evenodd" d="M4.72 3.22a.75.75 0 011.06 1.06L2.06 8l3.72 3.72a.75.75 0 11-1.06 1.06L.47 8.53a.75.75 0 010-1.06l4.25-4.25zm6.56 0a.75.75 0 10-1.06 1.06L13.94 8l-3.72 3.72a.75.75 0 101.06 1.06l4.25-4.25a.75.75 0 000-1.06l-4.25-4.25z"></path>
-</svg>
-
-</a>    <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-permalink-href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" aria-label="Display the rendered blob" data-view-component="true" class="rendered tooltipped tooltipped tooltipped-n selected js-permalink-replaceable-link btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-file">
-    <path fill-rule="evenodd" d="M3.75 1.5a.25.25 0 00-.25.25v12.5c0 .138.112.25.25.25h9.5a.25.25 0 00.25-.25V6h-2.75A1.75 1.75 0 019 4.25V1.5H3.75zm6.75.062V4.25c0 .138.112.25.25.25h2.688a.252.252 0 00-.011-.013l-2.914-2.914a.272.272 0 00-.013-.011zM2 1.75C2 .784 2.784 0 3.75 0h6.586c.464 0 .909.184 1.237.513l2.914 2.914c.329.328.513.773.513 1.237v9.586A1.75 1.75 0 0113.25 16h-9.5A1.75 1.75 0 012 14.25V1.75z"></path>
-</svg>
-
-</a>  </div>
-
-
-    <div class="BtnGroup">
-      <a data-permalink-href="/PaddlePaddle/PaddleOCR/raw/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" href="/PaddlePaddle/PaddleOCR/raw/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" id="raw-url" data-view-component="true" class="js-permalink-replaceable-link btn-sm btn BtnGroup-item">  Raw
-
-</a>        <a data-permalink-href="/PaddlePaddle/PaddleOCR/blame/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" href="/PaddlePaddle/PaddleOCR/blame/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-hotkey="b" data-view-component="true" class="js-update-url-with-hash js-permalink-replaceable-link btn-sm btn BtnGroup-item">  Blame
-
-</a>    </div>
-
-    <div class="d-flex">
-
-
-
-<div class="ml-1" data-test-selector="edit-dropdown-menu-component">
-  <form class="BtnGroup-parent js-update-url-with-hash " data-turbo="false" action="/PaddlePaddle/PaddleOCR/edit/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" accept-charset="UTF-8" method="post"><input type="hidden" name="authenticity_token" value="1Z4UlqXTCelIqAiU_KBcJn79l74PWWcGleoasGb6THCV1oxFbwiUxk0rAA2agNGcRF_HxL8kuNLZFJTzIo-u9w" autocomplete="off" />
-    <button disabled="disabled" title="You must be signed in to make or propose changes" data-hotkey="e" data-disable-with="" data-test-selector="edit-icon-button" type="submit" data-view-component="true" class="btn-sm BtnGroup-item btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-pencil">
-    <path fill-rule="evenodd" d="M11.013 1.427a1.75 1.75 0 012.474 0l1.086 1.086a1.75 1.75 0 010 2.474l-8.61 8.61c-.21.21-.47.364-.756.445l-3.251.93a.75.75 0 01-.927-.928l.929-3.25a1.75 1.75 0 01.445-.758l8.61-8.61zm1.414 1.06a.25.25 0 00-.354 0L10.811 3.75l1.439 1.44 1.263-1.263a.25.25 0 000-.354l-1.086-1.086zM11.189 6.25L9.75 4.81l-6.286 6.287a.25.25 0 00-.064.108l-.558 1.953 1.953-.558a.249.249 0 00.108-.064l6.286-6.286z"></path>
-</svg>
-
-</button></form>
-  <details class="details-reset details-overlay select-menu BtnGroup-parent d-inline-block position-relative">
-    <summary data-disable-invalid="" data-disable-with="" data-hydro-click="{&quot;event_type&quot;:&quot;blob_edit_dropdown.more_options_click&quot;,&quot;payload&quot;:{&quot;repository_id&quot;:262296122,&quot;actor_id&quot;:null,&quot;github_dev_enabled&quot;:false,&quot;edit_enabled&quot;:false,&quot;small_screen&quot;:false,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="8e1c91138fac18e1604a705b864e87574bdf4b05c6f3b559e5a5a9051b4f9f4c" aria-label="Select additional options" data-view-component="true" class="select-menu-button btn-sm btn BtnGroup-item float-none px-2">
-</summary>    <div class="SelectMenu right-0">
-      <div class="SelectMenu-modal width-full">
-        <div class="SelectMenu-list SelectMenu-list--borderless py-2">
-          <form class="SelectMenu-item js-update-url-with-hash " data-turbo="false" action="/PaddlePaddle/PaddleOCR/edit/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" accept-charset="UTF-8" method="post"><input type="hidden" name="authenticity_token" value="srFxcI2OKUzpJ_365ftK_v24mPkLiynPIhsZBVXEjdzy-emjR1W0Y-yk9WOD28dExxrIg7v29htu5ZdGEbFvWw" autocomplete="off" />
-            <button disabled="disabled" title="You must be signed in to make or propose changes" data-test-selector="edit-text-button" type="submit" data-view-component="true" class="btn-invisible btn width-full d-flex flex-justify-between color-fg-muted text-normal p-0">  <div>Edit this file</div>
-              <div class="color-fg-muted">E</div>
-
-</button></form>
-
-            <a data-platforms="windows,mac" data-test-selector="github-desktop-link" aria-label="Open this file in GitHub Desktop" href="https://desktop.github.com" data-view-component="true" class="SelectMenu-item no-wrap js-remove-unless-platform width-full text-normal color-fg-default f5">
-              Open in GitHub Desktop
-</a>        </div>
-      </div>
-    </div>
-  </details>
-</div>
-
-
-
-<div data-test-selector="remote-clipboard-copy">
-  </div>
-
-
-          <button class="btn-octicon btn-octicon-danger disabled tooltipped tooltipped-nw" disabled
-            aria-label="You must be signed in to make or propose changes" type="button">
-            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-trash">
-    <path fill-rule="evenodd" d="M6.5 1.75a.25.25 0 01.25-.25h2.5a.25.25 0 01.25.25V3h-3V1.75zm4.5 0V3h2.25a.75.75 0 010 1.5H2.75a.75.75 0 010-1.5H5V1.75C5 .784 5.784 0 6.75 0h2.5C10.216 0 11 .784 11 1.75zM4.496 6.675a.75.75 0 10-1.492.15l.66 6.6A1.75 1.75 0 005.405 15h5.19c.9 0 1.652-.681 1.741-1.576l.66-6.6a.75.75 0 00-1.492-.149l-.66 6.6a.25.25 0 01-.249.225h-5.19a.25.25 0 01-.249-.225l-.66-6.6z"></path>
-</svg>
-          </button>
-    </div>
-  </div>
-
-    <div class="d-flex hide-lg hide-xl flex-order-2 flex-grow-0">
-      <details class="dropdown details-reset details-overlay d-inline-block">
-        <summary
-          class="btn-octicon p-2"
-          aria-haspopup="true"
-          aria-label="possible actions"
-          data-test-selector="small-screen-more-options"
-          data-hydro-click="{&quot;event_type&quot;:&quot;blob_edit_dropdown.more_options_click&quot;,&quot;payload&quot;:{&quot;repository_id&quot;:262296122,&quot;actor_id&quot;:null,&quot;github_dev_enabled&quot;:false,&quot;edit_enabled&quot;:false,&quot;small_screen&quot;:true,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}"
-          data-hydro-click-hmac="3711aba9bcc87015fd0b3b29260720d15e89cbf158ae938adc99aa4a84a101e5"
-        >
-          <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
-    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
-</svg>
-        </summary>
-
-        <ul class="dropdown-menu dropdown-menu-sw" style="width: 175px">
-            <li>
-                <a class="dropdown-item tooltipped tooltipped-nw js-remove-unless-platform"
-                   data-platforms="windows,mac"
-                   href="https://desktop.github.com"
-                   data-ga-click="Repository, open with desktop">
-                  Open with Desktop
-                </a>
-            </li>
-          <li>
-            <a class="dropdown-item" href="/PaddlePaddle/PaddleOCR/raw/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">
-              View raw
-            </a>
-          </li>
-            <li>
-                          </li>
-            <li>
-              <a class="dropdown-item" href="/PaddlePaddle/PaddleOCR/blame/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">
-                View blame
-              </a>
-            </li>
-
-        </ul>
-      </details>
-    </div>
-</div>
-
-
-        <div id="readme" class="Box-body readme blob js-code-block-container p-5 p-xl-6 gist-border-0">
-    <article class="markdown-body entry-content container-lg" itemprop="text"><h1 dir="auto"><a id="user-content-一种基于paddleocr的轻量级车牌识别模型" class="anchor" aria-hidden="true" href="#一种基于paddleocr的轻量级车牌识别模型"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>一种基于PaddleOCR的轻量级车牌识别模型</h1>
-<ul dir="auto">
-<li><a href="#1-%E9%A1%B9%E7%9B%AE%E4%BB%8B%E7%BB%8D">1. 项目介绍</a></li>
-<li><a href="#2-%E7%8E%AF%E5%A2%83%E6%90%AD%E5%BB%BA">2. 环境搭建</a></li>
-<li><a href="#3-%E6%95%B0%E6%8D%AE%E9%9B%86%E5%87%86%E5%A4%87">3. 数据集准备</a>
-<ul dir="auto">
-<li><a href="#31-%E6%95%B0%E6%8D%AE%E9%9B%86%E6%A0%87%E6%B3%A8%E8%A7%84%E5%88%99">3.1 数据集标注规则</a></li>
-<li><a href="#32-%E5%88%B6%E4%BD%9C%E7%AC%A6%E5%90%88pp-ocr%E8%AE%AD%E7%BB%83%E6%A0%BC%E5%BC%8F%E7%9A%84%E6%A0%87%E6%B3%A8%E6%96%87%E4%BB%B6">3.2 制作符合PP-OCR训练格式的标注文件</a></li>
-</ul>
-</li>
-<li><a href="#4-%E5%AE%9E%E9%AA%8C">4. 实验</a>
-<ul dir="auto">
-<li><a href="#41-%E6%A3%80%E6%B5%8B">4.1 检测</a>
-<ul dir="auto">
-<li><a href="#411-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B">4.1.1 预训练模型直接预测</a></li>
-<li><a href="#412-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune">4.1.2 CCPD车牌数据集fine-tune</a></li>
-<li><a href="#413-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune%E9%87%8F%E5%8C%96%E8%AE%AD%E7%BB%83">4.1.3 CCPD车牌数据集fine-tune+量化训练</a></li>
-<li><a href="#414-%E6%A8%A1%E5%9E%8B%E5%AF%BC%E5%87%BA">4.1.4 模型导出</a></li>
-</ul>
-</li>
-<li><a href="#42-%E8%AF%86%E5%88%AB">4.2 识别</a>
-<ul dir="auto">
-<li><a href="#421-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B">4.2.1 预训练模型直接预测</a></li>
-<li><a href="#422-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B%E6%94%B9%E5%8A%A8%E5%90%8E%E5%A4%84%E7%90%86">4.2.2 预训练模型直接预测+改动后处理</a></li>
-<li><a href="#423-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune">4.2.3 CCPD车牌数据集fine-tune</a></li>
-<li><a href="#424-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune%E9%87%8F%E5%8C%96%E8%AE%AD%E7%BB%83">4.2.4 CCPD车牌数据集fine-tune+量化训练</a></li>
-<li><a href="#425-%E6%A8%A1%E5%9E%8B%E5%AF%BC%E5%87%BA">4.2.5 模型导出</a></li>
-</ul>
-</li>
-<li><a href="#43-%E8%AE%A1%E7%AE%97End2End%E6%8C%87%E6%A0%87">4.3 计算End2End指标</a></li>
-<li><a href="#44-%E9%83%A8%E7%BD%B2">4.4 部署</a></li>
-<li><a href="#45-%E5%AE%9E%E9%AA%8C%E6%80%BB%E7%BB%93">4.5 实验总结</a></li>
-</ul>
-</li>
-</ul>
-<h2 dir="auto"><a id="user-content-1-项目介绍" class="anchor" aria-hidden="true" href="#1-项目介绍"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>1. 项目介绍</h2>
-<p dir="auto">车牌识别(Vehicle License Plate Recognition，VLPR) 是计算机视频图像识别技术在车辆牌照识别中的一种应用。车牌识别技术要求能够将运动中的汽车牌照从复杂背景中提取并识别出来，在高速公路车辆管理，停车场管理和城市交通中得到广泛应用。</p>
-<p dir="auto">本项目难点如下：</p>
-<ol dir="auto">
-<li>车牌在图像中的尺度差异大、在车辆上的悬挂位置不固定</li>
-<li>车牌图像质量层次不齐: 角度倾斜、图片模糊、光照不足、过曝等问题严重</li>
-<li>边缘和端测场景应用对模型大小有限制，推理速度有要求</li>
-</ol>
-<p dir="auto">针对以上问题， 本例选用 PP-OCRv3 这一开源超轻量OCR系统进行车牌识别系统的开发。基于PP-OCRv3模型，在CCPD数据集达到99%的检测和94%的识别精度，模型大小12.8M(2.5M+10.3M)。基于量化对模型体积进行进一步压缩到5.8M(1M+4.8M), 同时推理速度提升25%。</p>
-<p dir="auto">aistudio项目链接: <a href="https://aistudio.baidu.com/aistudio/projectdetail/3919091?contributionType=1" rel="nofollow">基于PaddleOCR的轻量级车牌识别范例</a></p>
-<h2 dir="auto"><a id="user-content-2-环境搭建" class="anchor" aria-hidden="true" href="#2-环境搭建"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>2. 环境搭建</h2>
-<p dir="auto">本任务基于Aistudio完成, 具体环境如下：</p>
-<ul dir="auto">
-<li>操作系统: Linux</li>
-<li>PaddlePaddle: 2.3</li>
-<li>paddleslim: 2.2.2</li>
-<li>PaddleOCR: Release/2.5</li>
-</ul>
-<p dir="auto">下载 PaddleOCR代码</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR"><pre>git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR</pre></div>
-<p dir="auto">安装依赖库</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="pip install -r PaddleOCR/requirements.txt"><pre>pip install -r PaddleOCR/requirements.txt</pre></div>
-<h2 dir="auto"><a id="user-content-3-数据集准备" class="anchor" aria-hidden="true" href="#3-数据集准备"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3. 数据集准备</h2>
-<p dir="auto">所使用的数据集为 CCPD2020 新能源车牌数据集，该数据集为</p>
-<p dir="auto">该数据集分布如下：</p>
-<table>
-<thead>
-<tr>
-<th>数据集类型</th>
-<th>数量</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>训练集</td>
-<td>5769</td>
-</tr>
-<tr>
-<td>验证集</td>
-<td>1001</td>
-</tr>
-<tr>
-<td>测试集</td>
-<td>5006</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">数据集图片示例如下:
-<a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/422ba572727331bfacf4a748100bcdf775d5ed4b2409ae78456849b2d94b1a8d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f33626365303537613865306334306130616362643236623265323965346532353930613331626334313237363462653762396534393739396336396362393163"><img src="https://camo.githubusercontent.com/422ba572727331bfacf4a748100bcdf775d5ed4b2409ae78456849b2d94b1a8d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f33626365303537613865306334306130616362643236623265323965346532353930613331626334313237363462653762396534393739396336396362393163" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/3bce057a8e0c40a0acbd26b2e29e4e2590a31bc412764be7b9e49799c69cb91c" style="max-width: 100%;"></a></p>
-<p dir="auto">数据集可以从这里下载 <a href="https://aistudio.baidu.com/aistudio/datasetdetail/101595" rel="nofollow">https://aistudio.baidu.com/aistudio/datasetdetail/101595</a></p>
-<p dir="auto">下载好数据集后对数据集进行解压</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip"><pre>unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip</pre></div>
-<h3 dir="auto"><a id="user-content-31-数据集标注规则" class="anchor" aria-hidden="true" href="#31-数据集标注规则"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3.1 数据集标注规则</h3>
-<p dir="auto">CPPD数据集的图片文件名具有特殊规则，详细可查看：<a href="https://github.com/detectRecog/CCPD">https://github.com/detectRecog/CCPD</a></p>
-<p dir="auto">具体规则如下：</p>
-<p dir="auto">例如: 025-95_113-154&amp;383_386&amp;473-386&amp;473_177&amp;454_154&amp;383_363&amp;402-0_0_22_27_27_33_16-37-15.jpg</p>
-<p dir="auto">每个名称可以分为七个字段，以-符号作为分割。这些字段解释如下。</p>
-<ul dir="auto">
-<li>
-<p dir="auto">025：车牌面积与整个图片区域的面积比。025 (25%)</p>
-</li>
-<li>
-<p dir="auto">95_113：水平倾斜程度和垂直倾斜度。水平 95度 垂直 113度</p>
-</li>
-<li>
-<p dir="auto">154&amp;383_386&amp;473：左上和右下顶点的坐标。左上(154,383) 右下(386,473)</p>
-</li>
-<li>
-<p dir="auto">386&amp;473_177&amp;454_154&amp;383_363&amp;402：整个图像中车牌的四个顶点的精确（x，y）坐标。这些坐标从右下角顶点开始。(386,473) (177,454) (154,383) (363,402)</p>
-</li>
-<li>
-<p dir="auto">0_0_22_27_27_33_16：CCPD中的每个图像只有一个车牌。每个车牌号码由一个汉字，一个字母和五个字母或数字组成。有效的中文车牌由七个字符组成：省（1个字符），字母（1个字符），字母+数字（5个字符）。“ 0_0_22_27_27_33_16”是每个字符的索引。这三个数组定义如下。每个数组的最后一个字符是字母O，而不是数字0。我们将O用作“无字符”的符号，因为中文车牌字符中没有O。因此以上车牌拼起来即为 皖AY339S</p>
-</li>
-<li>
-<p dir="auto">37：牌照区域的亮度。 37 (37%)</p>
-</li>
-<li>
-<p dir="auto">15：车牌区域的模糊度。15 (15%)</p>
-</li>
-</ul>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="provinces = [&quot;皖&quot;, &quot;沪&quot;, &quot;津&quot;, &quot;渝&quot;, &quot;冀&quot;, &quot;晋&quot;, &quot;蒙&quot;, &quot;辽&quot;, &quot;吉&quot;, &quot;黑&quot;, &quot;苏&quot;, &quot;浙&quot;, &quot;京&quot;, &quot;闽&quot;, &quot;赣&quot;, &quot;鲁&quot;, &quot;豫&quot;, &quot;鄂&quot;, &quot;湘&quot;, &quot;粤&quot;, &quot;桂&quot;, &quot;琼&quot;, &quot;川&quot;, &quot;贵&quot;, &quot;云&quot;, &quot;藏&quot;, &quot;陕&quot;, &quot;甘&quot;, &quot;青&quot;, &quot;宁&quot;, &quot;新&quot;, &quot;警&quot;, &quot;学&quot;, &quot;O&quot;]
-alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W','X', 'Y', 'Z', 'O']
-ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X','Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']"><pre><span class="pl-s1">provinces</span> <span class="pl-c1">=</span> [<span class="pl-s">"皖"</span>, <span class="pl-s">"沪"</span>, <span class="pl-s">"津"</span>, <span class="pl-s">"渝"</span>, <span class="pl-s">"冀"</span>, <span class="pl-s">"晋"</span>, <span class="pl-s">"蒙"</span>, <span class="pl-s">"辽"</span>, <span class="pl-s">"吉"</span>, <span class="pl-s">"黑"</span>, <span class="pl-s">"苏"</span>, <span class="pl-s">"浙"</span>, <span class="pl-s">"京"</span>, <span class="pl-s">"闽"</span>, <span class="pl-s">"赣"</span>, <span class="pl-s">"鲁"</span>, <span class="pl-s">"豫"</span>, <span class="pl-s">"鄂"</span>, <span class="pl-s">"湘"</span>, <span class="pl-s">"粤"</span>, <span class="pl-s">"桂"</span>, <span class="pl-s">"琼"</span>, <span class="pl-s">"川"</span>, <span class="pl-s">"贵"</span>, <span class="pl-s">"云"</span>, <span class="pl-s">"藏"</span>, <span class="pl-s">"陕"</span>, <span class="pl-s">"甘"</span>, <span class="pl-s">"青"</span>, <span class="pl-s">"宁"</span>, <span class="pl-s">"新"</span>, <span class="pl-s">"警"</span>, <span class="pl-s">"学"</span>, <span class="pl-s">"O"</span>]
-<span class="pl-s1">alphabets</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>,<span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'O'</span>]
-<span class="pl-s1">ads</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>,<span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'0'</span>, <span class="pl-s">'1'</span>, <span class="pl-s">'2'</span>, <span class="pl-s">'3'</span>, <span class="pl-s">'4'</span>, <span class="pl-s">'5'</span>, <span class="pl-s">'6'</span>, <span class="pl-s">'7'</span>, <span class="pl-s">'8'</span>, <span class="pl-s">'9'</span>, <span class="pl-s">'O'</span>]</pre></div>
-<h3 dir="auto"><a id="user-content-32-制作符合pp-ocr训练格式的标注文件" class="anchor" aria-hidden="true" href="#32-制作符合pp-ocr训练格式的标注文件"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3.2 制作符合PP-OCR训练格式的标注文件</h3>
-<p dir="auto">在开始训练之前，可使用如下代码制作符合PP-OCR训练格式的标注文件。</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="import cv2
-import os
-import json
-from tqdm import tqdm
-import numpy as np
-
-provinces = [&quot;皖&quot;, &quot;沪&quot;, &quot;津&quot;, &quot;渝&quot;, &quot;冀&quot;, &quot;晋&quot;, &quot;蒙&quot;, &quot;辽&quot;, &quot;吉&quot;, &quot;黑&quot;, &quot;苏&quot;, &quot;浙&quot;, &quot;京&quot;, &quot;闽&quot;, &quot;赣&quot;, &quot;鲁&quot;, &quot;豫&quot;, &quot;鄂&quot;, &quot;湘&quot;, &quot;粤&quot;, &quot;桂&quot;, &quot;琼&quot;, &quot;川&quot;, &quot;贵&quot;, &quot;云&quot;, &quot;藏&quot;, &quot;陕&quot;, &quot;甘&quot;, &quot;青&quot;, &quot;宁&quot;, &quot;新&quot;, &quot;警&quot;, &quot;学&quot;, &quot;O&quot;]
-alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'O']
-ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']
-
-def make_label(img_dir, save_gt_folder, phase):
-    crop_img_save_dir = os.path.join(save_gt_folder, phase, 'crop_imgs')
-    os.makedirs(crop_img_save_dir, exist_ok=True)
-
-    f_det = open(os.path.join(save_gt_folder, phase, 'det.txt'), 'w', encoding='utf-8')
-    f_rec = open(os.path.join(save_gt_folder, phase, 'rec.txt'), 'w', encoding='utf-8')
-
-    i = 0
-    for filename in tqdm(os.listdir(os.path.join(img_dir, phase))):
-        str_list = filename.split('-')
-        if len(str_list) &lt; 5:
-            continue
-        coord_list = str_list[3].split('_')
-        txt_list = str_list[4].split('_')
-        boxes = []
-        for coord in coord_list:
-            boxes.append([int(x) for x in coord.split(&quot;&amp;&quot;)])
-        boxes = [boxes[2], boxes[3], boxes[0], boxes[1]]
-        lp_number = provinces[int(txt_list[0])] + alphabets[int(txt_list[1])] + ''.join([ads[int(x)] for x in txt_list[2:]])
-
-        # det
-        det_info = [{'points':boxes, 'transcription':lp_number}]
-        f_det.write('{}\t{}\n'.format(os.path.join(phase, filename), json.dumps(det_info, ensure_ascii=False)))
-
-        # rec
-        boxes = np.float32(boxes)
-        img = cv2.imread(os.path.join(img_dir, phase, filename))
-        # crop_img = img[int(boxes[:,1].min()):int(boxes[:,1].max()),int(boxes[:,0].min()):int(boxes[:,0].max())]
-        crop_img = get_rotate_crop_image(img, boxes)
-        crop_img_save_filename = '{}_{}.jpg'.format(i,'_'.join(txt_list))
-        crop_img_save_path = os.path.join(crop_img_save_dir, crop_img_save_filename)
-        cv2.imwrite(crop_img_save_path, crop_img)
-        f_rec.write('{}/crop_imgs/{}\t{}\n'.format(phase, crop_img_save_filename, lp_number))
-        i+=1
-    f_det.close()
-    f_rec.close()
-
-def get_rotate_crop_image(img, points):
-    '''
-    img_height, img_width = img.shape[0:2]
-    left = int(np.min(points[:, 0]))
-    right = int(np.max(points[:, 0]))
-    top = int(np.min(points[:, 1]))
-    bottom = int(np.max(points[:, 1]))
-    img_crop = img[top:bottom, left:right, :].copy()
-    points[:, 0] = points[:, 0] - left
-    points[:, 1] = points[:, 1] - top
-    '''
-    assert len(points) == 4, &quot;shape of points must be 4*2&quot;
-    img_crop_width = int(
-        max(
-            np.linalg.norm(points[0] - points[1]),
-            np.linalg.norm(points[2] - points[3])))
-    img_crop_height = int(
-        max(
-            np.linalg.norm(points[0] - points[3]),
-            np.linalg.norm(points[1] - points[2])))
-    pts_std = np.float32([[0, 0], [img_crop_width, 0],
-                          [img_crop_width, img_crop_height],
-                          [0, img_crop_height]])
-    M = cv2.getPerspectiveTransform(points, pts_std)
-    dst_img = cv2.warpPerspective(
-        img,
-        M, (img_crop_width, img_crop_height),
-        borderMode=cv2.BORDER_REPLICATE,
-        flags=cv2.INTER_CUBIC)
-    dst_img_height, dst_img_width = dst_img.shape[0:2]
-    if dst_img_height * 1.0 / dst_img_width &gt;= 1.5:
-        dst_img = np.rot90(dst_img)
-    return dst_img
-
-img_dir = '/home/aistudio/data/CCPD2020/ccpd_green'
-save_gt_folder = '/home/aistudio/data/CCPD2020/PPOCR'
-# phase = 'train' # change to val and test to make val dataset and test dataset
-for phase in ['train','val','test']:
-    make_label(img_dir, save_gt_folder, phase)"><pre><span class="pl-k">import</span> <span class="pl-s1">cv2</span>
-<span class="pl-k">import</span> <span class="pl-s1">os</span>
-<span class="pl-k">import</span> <span class="pl-s1">json</span>
-<span class="pl-k">from</span> <span class="pl-s1">tqdm</span> <span class="pl-k">import</span> <span class="pl-s1">tqdm</span>
-<span class="pl-k">import</span> <span class="pl-s1">numpy</span> <span class="pl-k">as</span> <span class="pl-s1">np</span>
-
-<span class="pl-s1">provinces</span> <span class="pl-c1">=</span> [<span class="pl-s">"皖"</span>, <span class="pl-s">"沪"</span>, <span class="pl-s">"津"</span>, <span class="pl-s">"渝"</span>, <span class="pl-s">"冀"</span>, <span class="pl-s">"晋"</span>, <span class="pl-s">"蒙"</span>, <span class="pl-s">"辽"</span>, <span class="pl-s">"吉"</span>, <span class="pl-s">"黑"</span>, <span class="pl-s">"苏"</span>, <span class="pl-s">"浙"</span>, <span class="pl-s">"京"</span>, <span class="pl-s">"闽"</span>, <span class="pl-s">"赣"</span>, <span class="pl-s">"鲁"</span>, <span class="pl-s">"豫"</span>, <span class="pl-s">"鄂"</span>, <span class="pl-s">"湘"</span>, <span class="pl-s">"粤"</span>, <span class="pl-s">"桂"</span>, <span class="pl-s">"琼"</span>, <span class="pl-s">"川"</span>, <span class="pl-s">"贵"</span>, <span class="pl-s">"云"</span>, <span class="pl-s">"藏"</span>, <span class="pl-s">"陕"</span>, <span class="pl-s">"甘"</span>, <span class="pl-s">"青"</span>, <span class="pl-s">"宁"</span>, <span class="pl-s">"新"</span>, <span class="pl-s">"警"</span>, <span class="pl-s">"学"</span>, <span class="pl-s">"O"</span>]
-<span class="pl-s1">alphabets</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'O'</span>]
-<span class="pl-s1">ads</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'0'</span>, <span class="pl-s">'1'</span>, <span class="pl-s">'2'</span>, <span class="pl-s">'3'</span>, <span class="pl-s">'4'</span>, <span class="pl-s">'5'</span>, <span class="pl-s">'6'</span>, <span class="pl-s">'7'</span>, <span class="pl-s">'8'</span>, <span class="pl-s">'9'</span>, <span class="pl-s">'O'</span>]
-
-<span class="pl-k">def</span> <span class="pl-en">make_label</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>):
-    <span class="pl-s1">crop_img_save_dir</span> <span class="pl-c1">=</span> <span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'crop_imgs'</span>)
-    <span class="pl-s1">os</span>.<span class="pl-en">makedirs</span>(<span class="pl-s1">crop_img_save_dir</span>, <span class="pl-s1">exist_ok</span><span class="pl-c1">=</span><span class="pl-c1">True</span>)
-
-    <span class="pl-s1">f_det</span> <span class="pl-c1">=</span> <span class="pl-en">open</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'det.txt'</span>), <span class="pl-s">'w'</span>, <span class="pl-s1">encoding</span><span class="pl-c1">=</span><span class="pl-s">'utf-8'</span>)
-    <span class="pl-s1">f_rec</span> <span class="pl-c1">=</span> <span class="pl-en">open</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'rec.txt'</span>), <span class="pl-s">'w'</span>, <span class="pl-s1">encoding</span><span class="pl-c1">=</span><span class="pl-s">'utf-8'</span>)
-
-    <span class="pl-s1">i</span> <span class="pl-c1">=</span> <span class="pl-c1">0</span>
-    <span class="pl-k">for</span> <span class="pl-s1">filename</span> <span class="pl-c1">in</span> <span class="pl-en">tqdm</span>(<span class="pl-s1">os</span>.<span class="pl-en">listdir</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">phase</span>))):
-        <span class="pl-s1">str_list</span> <span class="pl-c1">=</span> <span class="pl-s1">filename</span>.<span class="pl-en">split</span>(<span class="pl-s">'-'</span>)
-        <span class="pl-k">if</span> <span class="pl-en">len</span>(<span class="pl-s1">str_list</span>) <span class="pl-c1">&lt;</span> <span class="pl-c1">5</span>:
-            <span class="pl-k">continue</span>
-        <span class="pl-s1">coord_list</span> <span class="pl-c1">=</span> <span class="pl-s1">str_list</span>[<span class="pl-c1">3</span>].<span class="pl-en">split</span>(<span class="pl-s">'_'</span>)
-        <span class="pl-s1">txt_list</span> <span class="pl-c1">=</span> <span class="pl-s1">str_list</span>[<span class="pl-c1">4</span>].<span class="pl-en">split</span>(<span class="pl-s">'_'</span>)
-        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> []
-        <span class="pl-k">for</span> <span class="pl-s1">coord</span> <span class="pl-c1">in</span> <span class="pl-s1">coord_list</span>:
-            <span class="pl-s1">boxes</span>.<span class="pl-en">append</span>([<span class="pl-en">int</span>(<span class="pl-s1">x</span>) <span class="pl-k">for</span> <span class="pl-s1">x</span> <span class="pl-c1">in</span> <span class="pl-s1">coord</span>.<span class="pl-en">split</span>(<span class="pl-s">"&amp;"</span>)])
-        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> [<span class="pl-s1">boxes</span>[<span class="pl-c1">2</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">3</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">0</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">1</span>]]
-        <span class="pl-s1">lp_number</span> <span class="pl-c1">=</span> <span class="pl-s1">provinces</span>[<span class="pl-en">int</span>(<span class="pl-s1">txt_list</span>[<span class="pl-c1">0</span>])] <span class="pl-c1">+</span> <span class="pl-s1">alphabets</span>[<span class="pl-en">int</span>(<span class="pl-s1">txt_list</span>[<span class="pl-c1">1</span>])] <span class="pl-c1">+</span> <span class="pl-s">''</span>.<span class="pl-en">join</span>([<span class="pl-s1">ads</span>[<span class="pl-en">int</span>(<span class="pl-s1">x</span>)] <span class="pl-k">for</span> <span class="pl-s1">x</span> <span class="pl-c1">in</span> <span class="pl-s1">txt_list</span>[<span class="pl-c1">2</span>:]])
-
-        <span class="pl-c"># det</span>
-        <span class="pl-s1">det_info</span> <span class="pl-c1">=</span> [{<span class="pl-s">'points'</span>:<span class="pl-s1">boxes</span>, <span class="pl-s">'transcription'</span>:<span class="pl-s1">lp_number</span>}]
-        <span class="pl-s1">f_det</span>.<span class="pl-en">write</span>(<span class="pl-s">'{}<span class="pl-cce">\t</span>{}<span class="pl-cce">\n</span>'</span>.<span class="pl-en">format</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">phase</span>, <span class="pl-s1">filename</span>), <span class="pl-s1">json</span>.<span class="pl-en">dumps</span>(<span class="pl-s1">det_info</span>, <span class="pl-s1">ensure_ascii</span><span class="pl-c1">=</span><span class="pl-c1">False</span>)))
-
-        <span class="pl-c"># rec</span>
-        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">float32</span>(<span class="pl-s1">boxes</span>)
-        <span class="pl-s1">img</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">imread</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">phase</span>, <span class="pl-s1">filename</span>))
-        <span class="pl-c"># crop_img = img[int(boxes[:,1].min()):int(boxes[:,1].max()),int(boxes[:,0].min()):int(boxes[:,0].max())]</span>
-        <span class="pl-s1">crop_img</span> <span class="pl-c1">=</span> <span class="pl-en">get_rotate_crop_image</span>(<span class="pl-s1">img</span>, <span class="pl-s1">boxes</span>)
-        <span class="pl-s1">crop_img_save_filename</span> <span class="pl-c1">=</span> <span class="pl-s">'{}_{}.jpg'</span>.<span class="pl-en">format</span>(<span class="pl-s1">i</span>,<span class="pl-s">'_'</span>.<span class="pl-en">join</span>(<span class="pl-s1">txt_list</span>))
-        <span class="pl-s1">crop_img_save_path</span> <span class="pl-c1">=</span> <span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">crop_img_save_dir</span>, <span class="pl-s1">crop_img_save_filename</span>)
-        <span class="pl-s1">cv2</span>.<span class="pl-en">imwrite</span>(<span class="pl-s1">crop_img_save_path</span>, <span class="pl-s1">crop_img</span>)
-        <span class="pl-s1">f_rec</span>.<span class="pl-en">write</span>(<span class="pl-s">'{}/crop_imgs/{}<span class="pl-cce">\t</span>{}<span class="pl-cce">\n</span>'</span>.<span class="pl-en">format</span>(<span class="pl-s1">phase</span>, <span class="pl-s1">crop_img_save_filename</span>, <span class="pl-s1">lp_number</span>))
-        <span class="pl-s1">i</span><span class="pl-c1">+=</span><span class="pl-c1">1</span>
-    <span class="pl-s1">f_det</span>.<span class="pl-en">close</span>()
-    <span class="pl-s1">f_rec</span>.<span class="pl-en">close</span>()
-
-<span class="pl-k">def</span> <span class="pl-en">get_rotate_crop_image</span>(<span class="pl-s1">img</span>, <span class="pl-s1">points</span>):
-    <span class="pl-s">'''</span>
-<span class="pl-s">    img_height, img_width = img.shape[0:2]</span>
-<span class="pl-s">    left = int(np.min(points[:, 0]))</span>
-<span class="pl-s">    right = int(np.max(points[:, 0]))</span>
-<span class="pl-s">    top = int(np.min(points[:, 1]))</span>
-<span class="pl-s">    bottom = int(np.max(points[:, 1]))</span>
-<span class="pl-s">    img_crop = img[top:bottom, left:right, :].copy()</span>
-<span class="pl-s">    points[:, 0] = points[:, 0] - left</span>
-<span class="pl-s">    points[:, 1] = points[:, 1] - top</span>
-<span class="pl-s">    '''</span>
-    <span class="pl-k">assert</span> <span class="pl-en">len</span>(<span class="pl-s1">points</span>) <span class="pl-c1">==</span> <span class="pl-c1">4</span>, <span class="pl-s">"shape of points must be 4*2"</span>
-    <span class="pl-s1">img_crop_width</span> <span class="pl-c1">=</span> <span class="pl-en">int</span>(
-        <span class="pl-en">max</span>(
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">0</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">1</span>]),
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">2</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">3</span>])))
-    <span class="pl-s1">img_crop_height</span> <span class="pl-c1">=</span> <span class="pl-en">int</span>(
-        <span class="pl-en">max</span>(
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">0</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">3</span>]),
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">1</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">2</span>])))
-    <span class="pl-s1">pts_std</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">float32</span>([[<span class="pl-c1">0</span>, <span class="pl-c1">0</span>], [<span class="pl-s1">img_crop_width</span>, <span class="pl-c1">0</span>],
-                          [<span class="pl-s1">img_crop_width</span>, <span class="pl-s1">img_crop_height</span>],
-                          [<span class="pl-c1">0</span>, <span class="pl-s1">img_crop_height</span>]])
-    <span class="pl-v">M</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">getPerspectiveTransform</span>(<span class="pl-s1">points</span>, <span class="pl-s1">pts_std</span>)
-    <span class="pl-s1">dst_img</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">warpPerspective</span>(
-        <span class="pl-s1">img</span>,
-        <span class="pl-v">M</span>, (<span class="pl-s1">img_crop_width</span>, <span class="pl-s1">img_crop_height</span>),
-        <span class="pl-s1">borderMode</span><span class="pl-c1">=</span><span class="pl-s1">cv2</span>.<span class="pl-v">BORDER_REPLICATE</span>,
-        <span class="pl-s1">flags</span><span class="pl-c1">=</span><span class="pl-s1">cv2</span>.<span class="pl-v">INTER_CUBIC</span>)
-    <span class="pl-s1">dst_img_height</span>, <span class="pl-s1">dst_img_width</span> <span class="pl-c1">=</span> <span class="pl-s1">dst_img</span>.<span class="pl-s1">shape</span>[<span class="pl-c1">0</span>:<span class="pl-c1">2</span>]
-    <span class="pl-k">if</span> <span class="pl-s1">dst_img_height</span> <span class="pl-c1">*</span> <span class="pl-c1">1.0</span> <span class="pl-c1">/</span> <span class="pl-s1">dst_img_width</span> <span class="pl-c1">&gt;=</span> <span class="pl-c1">1.5</span>:
-        <span class="pl-s1">dst_img</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">rot90</span>(<span class="pl-s1">dst_img</span>)
-    <span class="pl-k">return</span> <span class="pl-s1">dst_img</span>
-
-<span class="pl-s1">img_dir</span> <span class="pl-c1">=</span> <span class="pl-s">'/home/aistudio/data/CCPD2020/ccpd_green'</span>
-<span class="pl-s1">save_gt_folder</span> <span class="pl-c1">=</span> <span class="pl-s">'/home/aistudio/data/CCPD2020/PPOCR'</span>
-<span class="pl-c"># phase = 'train' # change to val and test to make val dataset and test dataset</span>
-<span class="pl-k">for</span> <span class="pl-s1">phase</span> <span class="pl-c1">in</span> [<span class="pl-s">'train'</span>,<span class="pl-s">'val'</span>,<span class="pl-s">'test'</span>]:
-    <span class="pl-en">make_label</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>)</pre></div>
-<p dir="auto">通过上述命令可以完成了<code>训练集</code>，<code>验证集</code>和<code>测试集</code>的制作，制作完成的数据集信息如下：</p>
-<table>
-<thead>
-<tr>
-<th>类型</th>
-<th>数据集</th>
-<th>图片地址</th>
-<th>标签地址</th>
-<th>图片数量</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>检测</td>
-<td>训练集</td>
-<td>/home/aistudio/data/CCPD2020/ccpd_green/train</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/train/det.txt</td>
-<td>5769</td>
-</tr>
-<tr>
-<td>检测</td>
-<td>验证集</td>
-<td>/home/aistudio/data/CCPD2020/ccpd_green/val</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/val/det.txt</td>
-<td>1001</td>
-</tr>
-<tr>
-<td>检测</td>
-<td>测试集</td>
-<td>/home/aistudio/data/CCPD2020/ccpd_green/test</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/test/det.txt</td>
-<td>5006</td>
-</tr>
-<tr>
-<td>识别</td>
-<td>训练集</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/train/crop_imgs</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt</td>
-<td>5769</td>
-</tr>
-<tr>
-<td>识别</td>
-<td>验证集</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/val/crop_imgs</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/val/rec.txt</td>
-<td>1001</td>
-</tr>
-<tr>
-<td>识别</td>
-<td>测试集</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt</td>
-<td>5006</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">在普遍的深度学习流程中，都是在训练集训练，在验证集选择最优模型后在测试集上进行测试。在本例中，我们省略中间步骤，直接在训练集训练，在测试集选择最优模型，因此我们只使用训练集和测试集。</p>
-<h2 dir="auto"><a id="user-content-4-实验" class="anchor" aria-hidden="true" href="#4-实验"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4. 实验</h2>
-<p dir="auto">由于数据集比较少，为了模型更好和更快的收敛，这里选用 PaddleOCR 中的 PP-OCRv3 模型进行文本检测和识别，并且使用 PP-OCRv3 模型参数作为预训练模型。PP-OCRv3在PP-OCRv2的基础上，中文场景端到端Hmean指标相比于PP-OCRv2提升5%, 英文数字模型端到端效果提升11%。详细优化细节请参考<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md">PP-OCRv3</a>技术报告。</p>
-<p dir="auto">由于车牌场景均为端侧设备部署，因此对速度和模型大小有比较高的要求，因此还需要采用量化训练的方式进行模型大小的压缩和模型推理速度的加速。模型量化可以在基本不损失模型的精度的情况下，将FP32精度的模型参数转换为Int8精度，减小模型参数大小并加速计算，使用量化后的模型在移动端等部署时更具备速度优势。</p>
-<p dir="auto">因此，本实验中对于车牌检测和识别有如下3种方案：</p>
-<ol dir="auto">
-<li>PP-OCRv3中英文超轻量预训练模型直接预测</li>
-<li>CCPD车牌数据集在PP-OCRv3模型上fine-tune</li>
-<li>CCPD车牌数据集在PP-OCRv3模型上fine-tune后量化</li>
-</ol>
-<h3 dir="auto"><a id="user-content-41-检测" class="anchor" aria-hidden="true" href="#41-检测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1 检测</h3>
-<h4 dir="auto"><a id="user-content-411-预训练模型直接预测" class="anchor" aria-hidden="true" href="#411-预训练模型直接预测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.1 预训练模型直接预测</h4>
-<p dir="auto">从下表中下载PP-OCRv3文本检测预训练模型</p>
-<table>
-<thead>
-<tr>
-<th>模型名称</th>
-<th>模型简介</th>
-<th>配置文件</th>
-<th>推理模型大小</th>
-<th>下载地址</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>ch_PP-OCRv3_det</td>
-<td>【最新】原始超轻量模型，支持中英文、多语种文本检测</td>
-<td><a href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml">ch_PP-OCRv3_det_cml.yml</a></td>
-<td>3.8M</td>
-<td><a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar" rel="nofollow">推理模型</a> / <a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar" rel="nofollow">训练模型</a></td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">使用如下命令下载预训练模型</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="mkdir models
-cd models
-wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
-tar -xf ch_PP-OCRv3_det_distill_train.tar
-cd /home/aistudio/PaddleOCR"><pre>mkdir models
-<span class="pl-c1">cd</span> models
-wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
-tar -xf ch_PP-OCRv3_det_distill_train.tar
-<span class="pl-c1">cd</span> /home/aistudio/PaddleOCR</pre></div>
-<p dir="auto">预训练模型下载完成后，我们使用<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml">ch_PP-OCRv3_det_student.yml</a> 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址</li>
-</ol>
-</li>
-<li>数据集相关
-<ol dir="auto">
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">上述字段均为必须修改的字段，可以通过修改配置文件的方式改动，也可在不需要修改配置文件的情况下，改变训练的参数。这里使用不改变配置文件的方式 。使用如下命令进行PP-OCRv3文本检测预训练模型的评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">上述指令中，通过-c 选择训练使用配置文件，通过-o参数在不需要修改配置文件的情况下，改变训练的参数。</p>
-<p dir="auto">使用预训练模型进行评估，指标如下所示：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
-<td>76.12%</td>
-</tr>
-</tbody>
-</table>
-<h4 dir="auto"><a id="user-content-412-ccpd车牌数据集fine-tune" class="anchor" aria-hidden="true" href="#412-ccpd车牌数据集fine-tune"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.2 CCPD车牌数据集fine-tune</h4>
-<p dir="auto"><strong>训练</strong></p>
-<p dir="auto">为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址</li>
-<li>Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔772个step评估一次，772为一个epoch总的step数。</li>
-</ol>
-</li>
-<li>优化器相关:
-<ol dir="auto">
-<li>Optimizer.lr.name: 学习率衰减器设为常量 Const</li>
-<li>Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍</li>
-<li>Optimizer.lr.warmup_epoch: warmup_epoch设为0</li>
-</ol>
-</li>
-<li>数据集相关:
-<ol dir="auto">
-<li>Train.dataset.data_dir：指向训练集图片存放目录</li>
-<li>Train.dataset.label_file_list：指向训练集标注文件</li>
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">使用如下代码即可启动在CCPD车牌数据集上的fine-tune。</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Global.save_model_dir=output/CCPD/det \
-    Global.eval_batch_step=&quot;[0, 772]&quot; \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Global.save_model_dir=output/CCPD/det \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 772]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">在上述命令中，通过<code>-o</code>的方式修改了配置文件中的参数。</p>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_finetune.tar" rel="nofollow">det_ppocr_v3_finetune.tar</a></p>
-<p dir="auto"><strong>评估</strong></p>
-<p dir="auto">训练完成后使用如下命令进行评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
-<td>76.12%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
-<td>99%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到进行fine-tune能显著提升车牌检测的效果。</p>
-<h4 dir="auto"><a id="user-content-413-ccpd车牌数据集fine-tune量化训练" class="anchor" aria-hidden="true" href="#413-ccpd车牌数据集fine-tune量化训练"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.3 CCPD车牌数据集fine-tune+量化训练</h4>
-<p dir="auto">此处采用 PaddleOCR 中提供好的<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/slim/quantization/README.md">量化教程</a>对模型进行量化训练。</p>
-<p dir="auto">量化训练可通过如下命令启动:</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_model_dir=output/CCPD/det_quant \
-    Global.eval_batch_step=&quot;[0, 772]&quot; \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_model_dir=output/CCPD/det_quant \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 772]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_quant.tar" rel="nofollow">det_ppocr_v3_quant.tar</a></p>
-<p dir="auto">量化后指标对比如下</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
-<td>99%</td>
-<td>2.5M</td>
-<td>223ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化</td>
-<td>98.91%</td>
-<td>1M</td>
-<td>189ms</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到通过量化训练在精度几乎无损的情况下，降低模型体积60%并且推理速度提升15%。</p>
-<p dir="auto">速度测试基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>完成。</p>
-<h4 dir="auto"><a id="user-content-414-模型导出" class="anchor" aria-hidden="true" href="#414-模型导出"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.4 模型导出</h4>
-<p dir="auto">使用如下命令可以将训练好的模型进行导出</p>
-<ul dir="auto">
-<li>非量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer"><pre>python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer</pre></div>
-<ul dir="auto">
-<li>量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer"><pre>python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer</pre></div>
-<h3 dir="auto"><a id="user-content-42-识别" class="anchor" aria-hidden="true" href="#42-识别"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2 识别</h3>
-<h4 dir="auto"><a id="user-content-421-预训练模型直接预测" class="anchor" aria-hidden="true" href="#421-预训练模型直接预测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.1 预训练模型直接预测</h4>
-<p dir="auto">从下表中下载PP-OCRv3文本识别预训练模型</p>
-<table>
-<thead>
-<tr>
-<th>模型名称</th>
-<th>模型简介</th>
-<th>配置文件</th>
-<th>推理模型大小</th>
-<th>下载地址</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>ch_PP-OCRv3_rec</td>
-<td>【最新】原始超轻量模型，支持中英文、数字识别</td>
-<td><a href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml">ch_PP-OCRv3_rec_distillation.yml</a></td>
-<td>12.4M</td>
-<td><a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar" rel="nofollow">推理模型</a> / <a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar" rel="nofollow">训练模型</a></td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">使用如下命令下载预训练模型</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="mkdir models
-cd models
-wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
-tar -xf ch_PP-OCRv3_rec_train.tar
-cd /home/aistudio/PaddleOCR"><pre>mkdir models
-<span class="pl-c1">cd</span> models
-wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
-tar -xf ch_PP-OCRv3_rec_train.tar
-<span class="pl-c1">cd</span> /home/aistudio/PaddleOCR</pre></div>
-<p dir="auto">PaddleOCR提供的PP-OCRv3识别模型采用蒸馏训练策略，因此提供的预训练模型中会包含<code>Teacher</code>和<code>Student</code>模型的参数，详细信息可参考<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/knowledge_distillation.md">knowledge_distillation.md</a>。 因此，模型下载完成后需要使用如下代码提取<code>Student</code>模型的参数：</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="import paddle
-# 加载预训练模型
-all_params = paddle.load(&quot;models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams&quot;)
-# 查看权重参数的keys
-print(all_params.keys())
-# 学生模型的权重提取
-s_params = {key[len(&quot;Student.&quot;):]: all_params[key] for key in all_params if &quot;Student.&quot; in key}
-# 查看学生模型权重参数的keys
-print(s_params.keys())
-# 保存
-paddle.save(s_params, &quot;models/ch_PP-OCRv3_rec_train/student.pdparams&quot;)"><pre><span class="pl-k">import</span> <span class="pl-s1">paddle</span>
-<span class="pl-c"># 加载预训练模型</span>
-<span class="pl-s1">all_params</span> <span class="pl-c1">=</span> <span class="pl-s1">paddle</span>.<span class="pl-en">load</span>(<span class="pl-s">"models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams"</span>)
-<span class="pl-c"># 查看权重参数的keys</span>
-<span class="pl-en">print</span>(<span class="pl-s1">all_params</span>.<span class="pl-en">keys</span>())
-<span class="pl-c"># 学生模型的权重提取</span>
-<span class="pl-s1">s_params</span> <span class="pl-c1">=</span> {<span class="pl-s1">key</span>[<span class="pl-en">len</span>(<span class="pl-s">"Student."</span>):]: <span class="pl-s1">all_params</span>[<span class="pl-s1">key</span>] <span class="pl-k">for</span> <span class="pl-s1">key</span> <span class="pl-c1">in</span> <span class="pl-s1">all_params</span> <span class="pl-k">if</span> <span class="pl-s">"Student."</span> <span class="pl-c1">in</span> <span class="pl-s1">key</span>}
-<span class="pl-c"># 查看学生模型权重参数的keys</span>
-<span class="pl-en">print</span>(<span class="pl-s1">s_params</span>.<span class="pl-en">keys</span>())
-<span class="pl-c"># 保存</span>
-<span class="pl-s1">paddle</span>.<span class="pl-en">save</span>(<span class="pl-s1">s_params</span>, <span class="pl-s">"models/ch_PP-OCRv3_rec_train/student.pdparams"</span>)</pre></div>
-<p dir="auto">预训练模型下载完成后，我们使用<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml">ch_PP-OCRv3_rec.yml</a> 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址</li>
-</ol>
-</li>
-<li>数据集相关
-<ol dir="auto">
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">使用如下命令进行PP-OCRv3文本识别预训练模型的评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">评估部分日志如下：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
-eval model:: 100%|██████████████████████████████| 40/40 [00:15&lt;00:00,  2.57it/s]
-[2022/05/12 19:52:17] ppocr INFO: metric eval ***************
-[2022/05/12 19:52:17] ppocr INFO: acc:0.0
-[2022/05/12 19:52:17] ppocr INFO: norm_edit_dis:0.8656084923002452
-[2022/05/12 19:52:17] ppocr INFO: Teacher_acc:0.000399520574511545
-[2022/05/12 19:52:17] ppocr INFO: Teacher_norm_edit_dis:0.8657902943394548
-[2022/05/12 19:52:17] ppocr INFO: fps:1443.1801978719905
-"><pre>[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
-<span class="pl-c1">eval</span> model:: 100%<span class="pl-k">|</span>██████████████████████████████<span class="pl-k">|</span> 40/40 [00:<span class="pl-k">15&lt;</span>00:00,  2.57it/s]
-[2022/05/12 19:52:17] ppocr INFO: metric <span class="pl-c1">eval</span> <span class="pl-k">***************</span>
-[2022/05/12 19:52:17] ppocr INFO: acc:0.0
-[2022/05/12 19:52:17] ppocr INFO: norm_edit_dis:0.8656084923002452
-[2022/05/12 19:52:17] ppocr INFO: Teacher_acc:0.000399520574511545
-[2022/05/12 19:52:17] ppocr INFO: Teacher_norm_edit_dis:0.8657902943394548
-[2022/05/12 19:52:17] ppocr INFO: fps:1443.1801978719905
-</pre></div>
-<p dir="auto">使用预训练模型进行评估，指标如下所示：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">从评估日志中可以看到，直接使用PP-OCRv3预训练模型进行评估，acc非常低，但是norm_edit_dis很高。因此，我们猜测是模型大部分文字识别是对的，只有少部分文字识别错误。使用如下命令进行infer查看模型的推理结果进行验证：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg"><pre>python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg</pre></div>
-<p dir="auto">输出部分日志如下：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
-W0501 08:51:57.127391 11326 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
-W0501 08:51:57.132315 11326 device_context.cc:465] device: 0, cuDNN Version: 7.6.
-[2022/05/01 08:52:00] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/student
-[2022/05/01 08:52:00] ppocr INFO: infer_img: /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_3_32_30_31_30_30.jpg
-[2022/05/01 08:52:00] ppocr INFO:      result: {&quot;Student&quot;: {&quot;label&quot;: &quot;皖A·D86766&quot;, &quot;score&quot;: 0.9552637934684753}, &quot;Teacher&quot;: {&quot;label&quot;: &quot;皖A·D86766&quot;, &quot;score&quot;: 0.9917094707489014}}
-[2022/05/01 08:52:00] ppocr INFO: success!"><pre>[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
-W0501 08:51:57.127391 11326 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
-W0501 08:51:57.132315 11326 device_context.cc:465] device: 0, cuDNN Version: 7.6.
-[2022/05/01 08:52:00] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/student
-[2022/05/01 08:52:00] ppocr INFO: infer_img: /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_3_32_30_31_30_30.jpg
-[2022/05/01 08:52:00] ppocr INFO:      result: {<span class="pl-s"><span class="pl-pds">"</span>Student<span class="pl-pds">"</span></span>: {<span class="pl-s"><span class="pl-pds">"</span>label<span class="pl-pds">"</span></span>: <span class="pl-s"><span class="pl-pds">"</span>皖A·D86766<span class="pl-pds">"</span></span>, <span class="pl-s"><span class="pl-pds">"</span>score<span class="pl-pds">"</span></span>: 0.9552637934684753}, <span class="pl-s"><span class="pl-pds">"</span>Teacher<span class="pl-pds">"</span></span>: {<span class="pl-s"><span class="pl-pds">"</span>label<span class="pl-pds">"</span></span>: <span class="pl-s"><span class="pl-pds">"</span>皖A·D86766<span class="pl-pds">"</span></span>, <span class="pl-s"><span class="pl-pds">"</span>score<span class="pl-pds">"</span></span>: 0.9917094707489014}}
-[2022/05/01 08:52:00] ppocr INFO: success<span class="pl-k">!</span></pre></div>
-<p dir="auto">从infer结果可以看到，车牌中的文字大部分都识别正确，只是多识别出了一个<code>·</code>。针对这种情况，有如下两种方案：</p>
-<ol dir="auto">
-<li>直接通过后处理去掉多识别的<code>·</code>。</li>
-<li>进行 fine-tune。</li>
-</ol>
-<h4 dir="auto"><a id="user-content-422-预训练模型直接预测改动后处理" class="anchor" aria-hidden="true" href="#422-预训练模型直接预测改动后处理"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.2 预训练模型直接预测+改动后处理</h4>
-<p dir="auto">直接通过后处理去掉多识别的<code>·</code>，在后处理的改动比较简单，只需在 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/postprocess/rec_postprocess.py">ppocr/postprocess/rec_postprocess.py</a> 文件的76行添加如下代码:</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="text = text.replace('·','')"><pre><span class="pl-s1">text</span> <span class="pl-c1">=</span> <span class="pl-s1">text</span>.<span class="pl-en">replace</span>(<span class="pl-s">'·'</span>,<span class="pl-s">''</span>)</pre></div>
-<p dir="auto">改动前后指标对比:</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0.2%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
-<td>90.97%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到，去掉多余的<code>·</code>能大幅提高精度。</p>
-<h4 dir="auto"><a id="user-content-423-ccpd车牌数据集fine-tune" class="anchor" aria-hidden="true" href="#423-ccpd车牌数据集fine-tune"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.3 CCPD车牌数据集fine-tune</h4>
-<p dir="auto"><strong>训练</strong></p>
-<p dir="auto">为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址</li>
-<li>Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔45个step评估一次，45为一个epoch总的step数。</li>
-</ol>
-</li>
-<li>优化器相关
-<ol dir="auto">
-<li>Optimizer.lr.name: 学习率衰减器设为常量 Const</li>
-<li>Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍</li>
-<li>Optimizer.lr.warmup_epoch: warmup_epoch设为0</li>
-</ol>
-</li>
-<li>数据集相关
-<ol dir="auto">
-<li>Train.dataset.data_dir：指向训练集图片存放目录</li>
-<li>Train.dataset.label_file_list：指向训练集标注文件</li>
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">使用如下命令启动 fine-tune</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.save_model_dir=output/CCPD/rec/ \
-    Global.eval_batch_step=&quot;[0, 90]&quot; \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.save_model_dir=output/CCPD/rec/ \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 90]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_finetune.tar" rel="nofollow">rec_ppocr_v3_finetune.tar</a></p>
-<p dir="auto"><strong>评估</strong></p>
-<p dir="auto">训练完成后使用如下命令进行评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
-<td>90.97%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
-<td>94.54%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到进行fine-tune能显著提升车牌识别的效果。</p>
-<h4 dir="auto"><a id="user-content-424-ccpd车牌数据集fine-tune量化训练" class="anchor" aria-hidden="true" href="#424-ccpd车牌数据集fine-tune量化训练"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.4 CCPD车牌数据集fine-tune+量化训练</h4>
-<p dir="auto">此处采用 PaddleOCR 中提供好的<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/slim/quantization/README.md">量化教程</a>对模型进行量化训练。</p>
-<p dir="auto">量化训练可通过如下命令启动:</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_model_dir=output/CCPD/rec_quant/ \
-    Global.eval_batch_step=&quot;[0, 90]&quot; \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_model_dir=output/CCPD/rec_quant/ \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 90]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_quant.tar" rel="nofollow">rec_ppocr_v3_quant.tar</a></p>
-<p dir="auto">量化后指标对比如下</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
-<td>94.54%</td>
-<td>10.3M</td>
-<td>4.2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化</td>
-<td>93.4%</td>
-<td>4.8M</td>
-<td>1.8ms</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到量化后能降低模型体积53%并且推理速度提升57%，但是由于识别数据过少，量化带来了1%的精度下降。</p>
-<p dir="auto">速度测试基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>完成。</p>
-<h4 dir="auto"><a id="user-content-425-模型导出" class="anchor" aria-hidden="true" href="#425-模型导出"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.5 模型导出</h4>
-<p dir="auto">使用如下命令可以将训练好的模型进行导出。</p>
-<ul dir="auto">
-<li>非量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer"><pre>python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer</pre></div>
-<ul dir="auto">
-<li>量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer"><pre>python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer</pre></div>
-<h3 dir="auto"><a id="user-content-43-计算end2end指标" class="anchor" aria-hidden="true" href="#43-计算end2end指标"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.3 计算End2End指标</h3>
-<p dir="auto">端到端指标可通过 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/tools/end2end/readme.md">PaddleOCR内置脚本</a> 进行计算，具体步骤如下：</p>
-<ol dir="auto">
-<li>导出模型</li>
-</ol>
-<p dir="auto">通过如下命令进行模型的导出。注意，量化模型导出时，需要配置eval数据集</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="# 检测模型
-
-# 预训练模型
-python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Global.save_inference_dir=output/ch_PP-OCRv3_det_distill_train/infer
-
-# 非量化模型
-python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/det/infer
-
-# 量化模型
-python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/det_quant/infer \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt] \
-    Eval.loader.num_workers=0
-
-# 识别模型
-
-# 预训练模型
-python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.save_inference_dir=output/ch_PP-OCRv3_rec_train/infer
-
-# 非量化模型
-python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer
-
-# 量化模型
-python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre><span class="pl-c"><span class="pl-c">#</span> 检测模型</span>
-
-<span class="pl-c"><span class="pl-c">#</span> 预训练模型</span>
-python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Global.save_inference_dir=output/ch_PP-OCRv3_det_distill_train/infer
-
-<span class="pl-c"><span class="pl-c">#</span> 非量化模型</span>
-python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/det/infer
-
-<span class="pl-c"><span class="pl-c">#</span> 量化模型</span>
-python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/det_quant/infer \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt] \
-    Eval.loader.num_workers=0
-
-<span class="pl-c"><span class="pl-c">#</span> 识别模型</span>
-
-<span class="pl-c"><span class="pl-c">#</span> 预训练模型</span>
-python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.save_inference_dir=output/ch_PP-OCRv3_rec_train/infer
-
-<span class="pl-c"><span class="pl-c">#</span> 非量化模型</span>
-python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer
-
-<span class="pl-c"><span class="pl-c">#</span> 量化模型</span>
-python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<ol start="2" dir="auto">
-<li>用导出的模型对测试集进行预测</li>
-</ol>
-<p dir="auto">此处，分别使用PP-OCRv3预训练模型，fintune模型和量化模型对测试集的所有图像进行预测，命令如下：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型
-python3 tools/infer/predict_system.py --det_model_dir=models/ch_PP-OCRv3_det_distill_train/infer --rec_model_dir=models/ch_PP-OCRv3_rec_train/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/pretrain --use_dilation=true
-
-# PP-OCRv3中英文超轻量检测预训练模型+fine-tune，PP-OCRv3中英文超轻量识别预训练模型+fine-tune
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det/infer --rec_model_dir=output/CCPD/rec/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/fine-tune --use_dilation=true
-
-# PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true"><pre><span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型</span>
-python3 tools/infer/predict_system.py --det_model_dir=models/ch_PP-OCRv3_det_distill_train/infer --rec_model_dir=models/ch_PP-OCRv3_rec_train/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/pretrain --use_dilation=true
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型+fine-tune，PP-OCRv3中英文超轻量识别预训练模型+fine-tune</span>
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det/infer --rec_model_dir=output/CCPD/rec/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/fine-tune --use_dilation=true
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估</span>
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true</pre></div>
-<ol start="3" dir="auto">
-<li>转换label并计算指标</li>
-</ol>
-<p dir="auto">将gt和上一步保存的预测结果转换为端对端评测需要的数据格式，并根据转换后的数据进行端到端指标计算</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
-
-# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型 结果转换和评估
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/pretrain/system_results.txt --save_folder=end2end/pretrain
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/pretrain
-
-# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型+后处理去掉多识别的`·` 结果转换和评估
-# 需手动修改后处理函数
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/post/system_results.txt --save_folder=end2end/post
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/post
-
-# PP-OCRv3中英文超轻量检测预训练模型 fine-tune，PP-OCRv3中英文超轻量识别预训练模型 fine-tune 结果转换和评估
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/fine-tune/system_results.txt --save_folder=end2end/fine-tune
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/fine-tune
-
-# PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/quant/system_results.txt --save_folder=end2end/quant
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant"><pre>python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型 结果转换和评估</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/pretrain/system_results.txt --save_folder=end2end/pretrain
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/pretrain
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型+后处理去掉多识别的`·` 结果转换和评估</span>
-<span class="pl-c"><span class="pl-c">#</span> 需手动修改后处理函数</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/post/system_results.txt --save_folder=end2end/post
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/post
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune，PP-OCRv3中英文超轻量识别预训练模型 fine-tune 结果转换和评估</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/fine-tune/system_results.txt --save_folder=end2end/fine-tune
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/fine-tune
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/quant/system_results.txt --save_folder=end2end/quant
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant</pre></div>
-<p dir="auto">日志如下:</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="The convert label saved in end2end/gt
-The convert label saved in end2end/pretrain
-start testing...
-hit, dt_count, gt_count 2 5988 5006
-character_acc: 70.42%
-avg_edit_dist_field: 2.37
-avg_edit_dist_img: 2.37
-precision: 0.03%
-recall: 0.04%
-fmeasure: 0.04%
-The convert label saved in end2end/post
-start testing...
-hit, dt_count, gt_count 4224 5988 5006
-character_acc: 81.59%
-avg_edit_dist_field: 1.47
-avg_edit_dist_img: 1.47
-precision: 70.54%
-recall: 84.38%
-fmeasure: 76.84%
-The convert label saved in end2end/fine-tune
-start testing...
-hit, dt_count, gt_count 4286 4898 5006
-character_acc: 94.16%
-avg_edit_dist_field: 0.47
-avg_edit_dist_img: 0.47
-precision: 87.51%
-recall: 85.62%
-fmeasure: 86.55%
-The convert label saved in end2end/quant
-start testing...
-hit, dt_count, gt_count 4349 4951 5006
-character_acc: 94.13%
-avg_edit_dist_field: 0.47
-avg_edit_dist_img: 0.47
-precision: 87.84%
-recall: 86.88%
-fmeasure: 87.36%"><pre>The convert label saved <span class="pl-k">in</span> end2end/gt
-The convert label saved <span class="pl-k">in</span> end2end/pretrain
-start testing...
-hit, dt_count, gt_count 2 5988 5006
-character_acc: 70.42%
-avg_edit_dist_field: 2.37
-avg_edit_dist_img: 2.37
-precision: 0.03%
-recall: 0.04%
-fmeasure: 0.04%
-The convert label saved <span class="pl-k">in</span> end2end/post
-start testing...
-hit, dt_count, gt_count 4224 5988 5006
-character_acc: 81.59%
-avg_edit_dist_field: 1.47
-avg_edit_dist_img: 1.47
-precision: 70.54%
-recall: 84.38%
-fmeasure: 76.84%
-The convert label saved <span class="pl-k">in</span> end2end/fine-tune
-start testing...
-hit, dt_count, gt_count 4286 4898 5006
-character_acc: 94.16%
-avg_edit_dist_field: 0.47
-avg_edit_dist_img: 0.47
-precision: 87.51%
-recall: 85.62%
-fmeasure: 86.55%
-The convert label saved <span class="pl-k">in</span> end2end/quant
-start testing...
-hit, dt_count, gt_count 4349 4951 5006
-character_acc: 94.13%
-avg_edit_dist_field: 0.47
-avg_edit_dist_img: 0.47
-precision: 87.84%
-recall: 86.88%
-fmeasure: 87.36%</pre></div>
-<p dir="auto">各个方案端到端指标如下：</p>
-<table>
-<thead>
-<tr>
-<th>模型</th>
-<th>指标</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
-<td>0.04%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
-<td>78.27%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
-<td>87.14%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
-<td>88%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到78.27%，在CCPD数据集上进行 fine-tune 后指标进一步提升到87.14%, 在经过量化训练之后，由于检测模型的recall变高，指标进一步提升到88%。但是这个结果仍旧不符合检测模型+识别模型的真实性能(99%*94%=93%)，因此我们需要对 base case 进行具体分析。</p>
-<p dir="auto">在之前的端到端预测结果中，可以看到很多不符合车牌标注的文字被识别出来, 因此可以进行简单的过滤来提升precision</p>
-<p dir="auto">为了快速评估，我们在 <code> tools/end2end/convert_ppocr_label.py</code> 脚本的 58 行加入如下代码，对非8个字符的结果进行过滤</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="if len(txt) != 8: # 车牌字符串长度为8
-    continue"><pre><span class="pl-k">if</span> <span class="pl-en">len</span>(<span class="pl-s1">txt</span>) <span class="pl-c1">!=</span> <span class="pl-c1">8</span>: <span class="pl-c"># 车牌字符串长度为8</span>
-    <span class="pl-k">continue</span></pre></div>
-<p dir="auto">此外，通过可视化box可以发现有很多框都是竖直翻转之后的框，并且没有完全框住车牌边界，因此需要进行框的竖直翻转以及轻微扩大，示意图如下：</p>
-<p dir="auto"><a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/199752de33937424916ca3fd6ef50f83461dca6034c35d251e32a4eecd4c3228/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f35396162303431316338656234646664393137666232623665356236396131376565376361343833353134343461656339616336313034623739666631303238"><img src="https://camo.githubusercontent.com/199752de33937424916ca3fd6ef50f83461dca6034c35d251e32a4eecd4c3228/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f35396162303431316338656234646664393137666232623665356236396131376565376361343833353134343461656339616336313034623739666631303238" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/59ab0411c8eb4dfd917fb2b6e5b69a17ee7ca48351444aec9ac6104b79ff1028" style="max-width: 100%;"></a></p>
-<p dir="auto">修改前后个方案指标对比如下：</p>
-<p dir="auto">各个方案端到端指标如下：</p>
-<table>
-<thead>
-<tr>
-<th>模型</th>
-<th>base</th>
-<th>A:识别结果过滤</th>
-<th>B:use_dilation</th>
-<th>C:flip_box</th>
-<th>best</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
-<td>0.04%</td>
-<td>0.08%</td>
-<td>0.02%</td>
-<td>0.05%</td>
-<td>0.00%(A)</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
-<td>78.27%</td>
-<td>90.84%</td>
-<td>78.61%</td>
-<td>79.43%</td>
-<td>91.66%(A+B+C)</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
-<td>87.14%</td>
-<td>90.40%</td>
-<td>87.66%</td>
-<td>89.98</td>
-<td>92.5%(A+B+C)</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
-<td>88%</td>
-<td>90.54%</td>
-<td>88.5%</td>
-<td>89.46%</td>
-<td>92.02%(A+B+C)</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标变为92.02%。</p>
-<h3 dir="auto"><a id="user-content-44-部署" class="anchor" aria-hidden="true" href="#44-部署"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.4 部署</h3>
-<ul dir="auto">
-<li>基于 Paddle Inference 的python推理</li>
-</ul>
-<p dir="auto">检测模型和识别模型分别 fine-tune 并导出为inference模型之后，可以使用如下命令基于 Paddle Inference 进行端到端推理并对结果进行可视化。</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/infer/predict_system.py \
-    --det_model_dir=output/CCPD/det/infer/ \
-    --rec_model_dir=output/CCPD/rec/infer/ \
-    --image_dir=&quot;/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&amp;509_530&amp;611-527&amp;611_172&amp;599_159&amp;509_530&amp;525-0_0_3_32_30_31_30_30-109-106.jpg&quot; \
-    --rec_image_shape=3,48,320"><pre>python tools/infer/predict_system.py \
-    --det_model_dir=output/CCPD/det/infer/ \
-    --rec_model_dir=output/CCPD/rec/infer/ \
-    --image_dir=<span class="pl-s"><span class="pl-pds">"</span>/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&amp;509_530&amp;611-527&amp;611_172&amp;599_159&amp;509_530&amp;525-0_0_3_32_30_31_30_30-109-106.jpg<span class="pl-pds">"</span></span> \
-    --rec_image_shape=3,48,320</pre></div>
-<p dir="auto">推理结果如下</p>
-<p dir="auto"><a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/b5c04d7affeca84aebcff3bf829edff2d142d26d700fd208c24b983144df766d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f37366236613039333963326334636634393033396236353633633462323865323431653131323835643734363465373939653831633538633066373730376137"><img src="https://camo.githubusercontent.com/b5c04d7affeca84aebcff3bf829edff2d142d26d700fd208c24b983144df766d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f37366236613039333963326334636634393033396236353633633462323865323431653131323835643734363465373939653831633538633066373730376137" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/76b6a0939c2c4cf49039b6563c4b28e241e11285d7464e799e81c58c0f7707a7" style="max-width: 100%;"></a></p>
-<ul dir="auto">
-<li>端侧部署</li>
-</ul>
-<p dir="auto">端侧部署我们采用基于 PaddleLite 的 cpp 推理。Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。具体可参考 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a></p>
-<h3 dir="auto"><a id="user-content-45-实验总结" class="anchor" aria-hidden="true" href="#45-实验总结"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.5 实验总结</h3>
-<p dir="auto">我们分别使用PP-OCRv3中英文超轻量预训练模型在车牌数据集上进行了直接评估和 fine-tune 和 fine-tune +量化3种方案的实验，并基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>进行了速度测试，指标对比如下：</p>
-<ul dir="auto">
-<li>检测</li>
-</ul>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
-<td>76.12%</td>
-<td>2.5M</td>
-<td>233ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
-<td>99%</td>
-<td>2.5M</td>
-<td>233ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune + 量化</td>
-<td>98.91%</td>
-<td>1M</td>
-<td>189ms</td>
-</tr>
-</tbody>
-</table>
-<ul dir="auto">
-<li>识别</li>
-</ul>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0%</td>
-<td>10.3M</td>
-<td>4.2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
-<td>90.97%</td>
-<td>10.3M</td>
-<td>4.2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
-<td>94.54%</td>
-<td>10.3M</td>
-<td>4,2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化</td>
-<td>93.4%</td>
-<td>4.8M</td>
-<td>1.8ms</td>
-</tr>
-</tbody>
-</table>
-<ul dir="auto">
-<li>端到端指标如下：</li>
-</ul>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>fmeasure</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
-<td>0.08%</td>
-<td>12.8M</td>
-<td>298ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
-<td>91.66%</td>
-<td>12.8M</td>
-<td>298ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
-<td>92.5%</td>
-<td>12.8M</td>
-<td>298ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
-<td>92.02%</td>
-<td>5.8M</td>
-<td>224ms</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto"><strong>结论</strong></p>
-<p dir="auto">PP-OCRv3的检测模型在未经过fine-tune的情况下，在车牌数据集上也有一定的精度，经过 fine-tune 后能够极大的提升检测效果，精度达到99%。在使用量化训练后检测模型的精度几乎无损，并且模型大小压缩60%。</p>
-<p dir="auto">PP-OCRv3的识别模型在未经过fine-tune的情况下，在车牌数据集上精度为0，但是经过分析可以知道，模型大部分字符都预测正确，但是会多预测一个特殊字符，去掉这个特殊字符后，精度达到90%。PP-OCRv3识别模型在经过 fine-tune 后识别精度进一步提升，达到94.4%。在使用量化训练后识别模型大小压缩53%，但是由于数据量多少，带来了1%的精度损失。</p>
-<p dir="auto">从端到端结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标轻微下降到92.02%但模型大小降低54%。</p>
-</article>
-  </div>
-
-    </div>
-
-  </readme-toc>
-
-
-
-  <details class="details-reset details-overlay details-overlay-dark" id="jumpto-line-details-dialog">
-    <summary data-hotkey="l" aria-label="Jump to line"></summary>
-    <details-dialog class="Box Box--overlay d-flex flex-column anim-fade-in fast linejump" aria-label="Jump to line">
-      <!-- '"` --><!-- </textarea></xmp> --></option></form><form class="js-jump-to-line-form Box-body d-flex" data-turbo="false" action="" accept-charset="UTF-8" method="get">
-        <input class="form-control flex-auto mr-3 linejump-input js-jump-to-line-field" type="text" placeholder="Jump to line&hellip;" aria-label="Jump to line" autofocus>
-        <button data-close-dialog="" type="submit" data-view-component="true" class="btn">  Go
-
-</button>
-</form>    </details-dialog>
-  </details>
-
-    <div class="Popover anim-scale-in js-tagsearch-popover"
-     hidden
-     data-tagsearch-url="/PaddlePaddle/PaddleOCR/find-definition"
-     data-tagsearch-ref="dygraph"
-     data-tagsearch-code-nav-context="BLOB_VIEW">
-  <div class="Popover-message Popover-message--large Popover-message--top-left TagsearchPopover mt-1 mb-4 mx-auto Box color-shadow-large">
-    <div class="TagsearchPopover-content js-tagsearch-popover-content overflow-auto" style="will-change:transform;">
-    </div>
-
-  </div>
-</div>
-
-
-</div>
-
-  </div>
-
-
-</div>
-
-
-    </main>
-  </div>
-
-  </div>
-
-          <footer class="footer width-full container-xl p-responsive" role="contentinfo">
-
-
-  <div class="position-relative d-flex flex-items-center pb-2 f6 color-fg-muted border-top color-border-muted flex-column-reverse flex-lg-row flex-wrap flex-lg-nowrap mt-6 pt-6">
-    <ul class="list-style-none d-flex flex-wrap col-0 col-lg-2 flex-justify-start flex-lg-justify-between mb-2 mb-lg-0">
-      <li class="mt-2 mt-lg-0 d-flex flex-items-center">
-        <a aria-label="Homepage" title="GitHub" class="footer-octicon mr-2" href="https://github.com">
-          <svg aria-hidden="true" height="24" viewBox="0 0 16 16" version="1.1" width="24" data-view-component="true" class="octicon octicon-mark-github">
-    <path fill-rule="evenodd" d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"></path>
-</svg>
-</a>        <span>
-        &copy; 2022 GitHub, Inc.
-        </span>
-      </li>
-    </ul>
-    <ul class="list-style-none d-flex flex-wrap col-12 col-lg-8 flex-justify-center flex-lg-justify-between mb-2 mb-lg-0">
-        <li class="mr-3 mr-lg-0"><a href="https://docs.github.com/en/github/site-policy/github-terms-of-service" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to terms&quot;,&quot;label&quot;:&quot;text:terms&quot;}">Terms</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://docs.github.com/en/github/site-policy/github-privacy-statement" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to privacy&quot;,&quot;label&quot;:&quot;text:privacy&quot;}">Privacy</a></li>
-        <li class="mr-3 mr-lg-0"><a data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to security&quot;,&quot;label&quot;:&quot;text:security&quot;}" href="https://github.com/security">Security</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://www.githubstatus.com/" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to status&quot;,&quot;label&quot;:&quot;text:status&quot;}">Status</a></li>
-        <li class="mr-3 mr-lg-0"><a data-ga-click="Footer, go to help, text:Docs" href="https://docs.github.com">Docs</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://support.github.com?tags=dotcom-footer" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to contact&quot;,&quot;label&quot;:&quot;text:contact&quot;}">Contact GitHub</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://github.com/pricing" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to Pricing&quot;,&quot;label&quot;:&quot;text:Pricing&quot;}">Pricing</a></li>
-      <li class="mr-3 mr-lg-0"><a href="https://docs.github.com" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to api&quot;,&quot;label&quot;:&quot;text:api&quot;}">API</a></li>
-      <li class="mr-3 mr-lg-0"><a href="https://services.github.com" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to training&quot;,&quot;label&quot;:&quot;text:training&quot;}">Training</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://github.blog" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to blog&quot;,&quot;label&quot;:&quot;text:blog&quot;}">Blog</a></li>
-        <li><a data-ga-click="Footer, go to about, text:about" href="https://github.com/about">About</a></li>
-    </ul>
-  </div>
-  <div class="d-flex flex-justify-center pb-6">
-    <span class="f6 color-fg-muted"></span>
-  </div>
-</footer>
-
-
-
-
-  <div id="ajax-error-message" class="ajax-error-message flash flash-error" hidden>
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert">
-    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
-</svg>
-    <button type="button" class="flash-close js-ajax-error-dismiss" aria-label="Dismiss error">
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-    </button>
-    You can’t perform that action at this time.
-  </div>
-
-  <div class="js-stale-session-flash flash flash-warn flash-banner" hidden
-    >
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert">
-    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
-</svg>
-    <span class="js-stale-session-flash-signed-in" hidden>You signed in with another tab or window. <a href="">Reload</a> to refresh your session.</span>
-    <span class="js-stale-session-flash-signed-out" hidden>You signed out in another tab or window. <a href="">Reload</a> to refresh your session.</span>
-  </div>
-    <template id="site-details-dialog">
-  <details class="details-reset details-overlay details-overlay-dark lh-default color-fg-default hx_rsm" open>
-    <summary role="button" aria-label="Close dialog"></summary>
-    <details-dialog class="Box Box--overlay d-flex flex-column anim-fade-in fast hx_rsm-dialog hx_rsm-modal">
-      <button class="Box-btn-octicon m-0 btn-octicon position-absolute right-0 top-0" type="button" aria-label="Close dialog" data-close-dialog>
-        <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-      </button>
-      <div class="octocat-spinner my-6 js-details-dialog-spinner"></div>
-    </details-dialog>
-  </details>
-</template>
-
-    <div class="Popover js-hovercard-content position-absolute" style="display: none; outline: none;" tabindex="0">
-  <div class="Popover-message Popover-message--bottom-left Popover-message--large Box color-shadow-large" style="width:360px;">
-  </div>
-</div>
-
-    <template id="snippet-clipboard-copy-button">
-  <div class="zeroclipboard-container position-absolute right-0 top-0">
-    <clipboard-copy aria-label="Copy" class="ClipboardButton btn js-clipboard-copy m-2 p-0 tooltipped-no-delay" data-copy-feedback="Copied!" data-tooltip-direction="w">
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-copy js-clipboard-copy-icon m-2">
-    <path fill-rule="evenodd" d="M0 6.75C0 5.784.784 5 1.75 5h1.5a.75.75 0 010 1.5h-1.5a.25.25 0 00-.25.25v7.5c0 .138.112.25.25.25h7.5a.25.25 0 00.25-.25v-1.5a.75.75 0 011.5 0v1.5A1.75 1.75 0 019.25 16h-7.5A1.75 1.75 0 010 14.25v-7.5z"></path><path fill-rule="evenodd" d="M5 1.75C5 .784 5.784 0 6.75 0h7.5C15.216 0 16 .784 16 1.75v7.5A1.75 1.75 0 0114.25 11h-7.5A1.75 1.75 0 015 9.25v-7.5zm1.75-.25a.25.25 0 00-.25.25v7.5c0 .138.112.25.25.25h7.5a.25.25 0 00.25-.25v-7.5a.25.25 0 00-.25-.25h-7.5z"></path>
-</svg>
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check js-clipboard-check-icon color-fg-success d-none m-2">
-    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
-</svg>
-    </clipboard-copy>
-  </div>
-</template>
-
-
-
-
-  </body>
-</html>

From 1ad3ba3ee6ba7debb4b4970faf3ce9fc3245f56c Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Tue, 21 Jun 2022 07:27:24 +0000
Subject: [PATCH 031/134] update kl pact config

---
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 .../train_ptq_infer_python.txt}               |   0
 ...python.txt => train_pact_infer_python.txt} |   0
 test_tipc/prepare.sh                          |   5 +-
 test_tipc/test_ptq_inference_python.sh        | 157 ++++++++++++++++++
 16 files changed, 160 insertions(+), 2 deletions(-)
 rename test_tipc/configs/{ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => ch_PP-OCRv2_det/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/ch_PP-OCRv2_det_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 rename test_tipc/configs/{ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => ch_PP-OCRv2_rec/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/ch_PP-OCRv2_rec_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 rename test_tipc/configs/{ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => ch_PP-OCRv3_det/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/ch_PP-OCRv3_det_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 rename test_tipc/configs/{ch_PP-OCRv3_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => ch_PP-OCRv3_rec/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/ch_PP-OCRv3_rec_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 rename test_tipc/configs/{ch_ppocr_mobile_v2.0_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => ch_ppocr_mobile_v2.0_det/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 rename test_tipc/configs/{ch_ppocr_mobile_v2.0_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => ch_ppocr_mobile_v2.0_rec/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 rename test_tipc/configs/{en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt => en_table_structure/train_ptq_infer_python.txt} (100%)
 rename test_tipc/configs/en_table_structure_PACT/{train_infer_python.txt => train_pact_infer_python.txt} (100%)
 create mode 100644 test_tipc/test_ptq_inference_python.sh

diff --git a/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_PACT/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_det_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv2_det_PACT/train_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv2_det_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/ch_PP-OCRv2_rec/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/ch_PP-OCRv3_det/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_PACT/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_det_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv3_det_PACT/train_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv3_det_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv3_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_rec/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv3_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/ch_PP-OCRv3_rec/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv3_rec_PACT/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_rec_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv3_rec_PACT/train_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv3_rec_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_ppocr_mobile_v2.0_det_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_infer_python.txt
rename to test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_ppocr_mobile_v2.0_rec_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_infer_python.txt
rename to test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/configs/en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure/train_ptq_infer_python.txt
similarity index 100%
rename from test_tipc/configs/en_table_structure_KL/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
rename to test_tipc/configs/en_table_structure/train_ptq_infer_python.txt
diff --git a/test_tipc/configs/en_table_structure_PACT/train_infer_python.txt b/test_tipc/configs/en_table_structure_PACT/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/en_table_structure_PACT/train_infer_python.txt
rename to test_tipc/configs/en_table_structure_PACT/train_pact_infer_python.txt
diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index cac373dedc..2c9bd2901b 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -100,7 +100,8 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
         wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar --no-check-certificate
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar --no-check-certificate
-        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && cd ../
+        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && cd ../
+        cd ./pretrain_models && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && cd ../
     fi
     if [ ${model_name} == "det_mv3_db_v2_0" ]; then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar  --no-check-certificate
@@ -353,7 +354,7 @@ elif [ ${MODE} = "whole_infer" ];then
     fi
 fi
 
-if [ ${MODE} = "klquant_whole_infer" ]; then
+if [[ ${model_name} =~ "KL" ]]; then
     wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar --no-check-certificate
     cd ./train_data/ && tar xf icdar2015_lite.tar && rm -rf ./icdar2015 && ln -s ./icdar2015_lite ./icdar2015 && cd ../
     if [ ${model_name} = "ch_ppocr_mobile_v2.0_det_KL" ]; then
diff --git a/test_tipc/test_ptq_inference_python.sh b/test_tipc/test_ptq_inference_python.sh
new file mode 100644
index 0000000000..2454c0b50a
--- /dev/null
+++ b/test_tipc/test_ptq_inference_python.sh
@@ -0,0 +1,157 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+# MODE be one of [''whole_infer']
+MODE=$2
+
+IFS=$'\n'
+# parser klquant_infer params
+
+dataline=$(awk 'NR==1, NR==17{print}'  $FILENAME)
+lines=(${dataline})
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+export_weight=$(func_parser_key "${lines[3]}")
+save_infer_key=$(func_parser_key "${lines[4]}")
+# parser inference model 
+infer_model_dir_list=$(func_parser_value "${lines[5]}")
+infer_export_list=$(func_parser_value "${lines[6]}")
+infer_is_quant=$(func_parser_value "${lines[7]}")
+# parser inference 
+inference_py=$(func_parser_value "${lines[8]}")
+use_gpu_key=$(func_parser_key "${lines[9]}")
+use_gpu_list=$(func_parser_value "${lines[9]}")
+use_mkldnn_key=$(func_parser_key "${lines[10]}")
+use_mkldnn_list=$(func_parser_value "${lines[10]}")
+cpu_threads_key=$(func_parser_key "${lines[11]}")
+cpu_threads_list=$(func_parser_value "${lines[11]}")
+batch_size_key=$(func_parser_key "${lines[12]}")
+batch_size_list=$(func_parser_value "${lines[12]}")
+use_trt_key=$(func_parser_key "${lines[13]}")
+use_trt_list=$(func_parser_value "${lines[13]}")
+precision_key=$(func_parser_key "${lines[14]}")
+precision_list=$(func_parser_value "${lines[14]}")
+infer_model_key=$(func_parser_key "${lines[15]}")
+image_dir_key=$(func_parser_key "${lines[16]}")
+infer_img_dir=$(func_parser_value "${lines[16]}")
+save_log_key=$(func_parser_key "${lines[17]}")
+save_log_value=$(func_parser_value "${lines[17]}")
+benchmark_key=$(func_parser_key "${lines[18]}")
+benchmark_value=$(func_parser_value "${lines[18]}")
+infer_key1=$(func_parser_key "${lines[19]}")
+infer_value1=$(func_parser_value "${lines[19]}")
+
+
+LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+
+function func_inference(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    _log_path=$4
+    _img_dir=$5
+    _flag_quant=$6
+    # inference 
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${use_mkldnn_list[*]}; do
+                for threads in ${cpu_threads_list[*]}; do
+                    for batch_size in ${batch_size_list[*]}; do
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
+                            _save_log_path="${_log_path}/python_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                            set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                            set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                            set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                            set_mkldnn=$(func_set_params "${use_mkldnn_key}" "${use_mkldnn}")
+                            set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
+                            set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                            set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                            set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_params0} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            eval $command
+                            last_status=${PIPESTATUS[0]}
+                            eval "cat ${_save_log_path}"
+                            status_check $last_status "${command}" "${status_log}" "${model_name}"
+                        done
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${use_trt_list[*]}; do
+                for precision in ${precision_list[*]}; do
+                    if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                        continue
+                    fi # skip when quant model inference but precision is not int8
+                    for batch_size in ${batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/python_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                        set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                        set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                        set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} ${set_infer_params0} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+if [ ${MODE} = "whole_infer" ]; then
+    GPUID=$3
+    if [ ${#GPUID} -le 0 ];then
+        env=" "
+    else
+        env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+    fi
+    # set CUDA_VISIBLE_DEVICES
+    eval $env
+    export Count=0
+    IFS="|"
+    infer_run_exports=(${infer_export_list})
+    infer_quant_flag=(${infer_is_quant})
+    for infer_model in ${infer_model_dir_list[*]}; do
+        # run export
+        if [ ${infer_run_exports[Count]} != "null" ];then
+            save_infer_dir="${infer_model}_klquant"
+            set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
+            set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key}"
+            echo ${infer_run_exports[Count]} 
+            echo $export_cmd
+            eval $export_cmd
+            status_export=$?
+            status_check $status_export "${export_cmd}" "${status_log}" "${model_name}"
+        else
+            save_infer_dir=${infer_model}
+        fi
+        #run inference
+        is_quant="True"
+        func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant}
+        Count=$(($Count + 1))
+    done
+fi
+

From 4f5b2b9759982822c5a076fc0d023915e7714ab7 Mon Sep 17 00:00:00 2001
From: huangjun12 <12272008@bjtu.edu.cn>
Date: Tue, 21 Jun 2022 07:27:35 +0000
Subject: [PATCH 032/134] add package producation data applications

---
 applications/包装生产日期识别.md | 827 +++++++++++++++++++++++++++++++
 1 file changed, 827 insertions(+)
 create mode 100644 applications/包装生产日期识别.md

diff --git a/applications/包装生产日期识别.md b/applications/包装生产日期识别.md
new file mode 100644
index 0000000000..4f74a4a148
--- /dev/null
+++ b/applications/包装生产日期识别.md
@@ -0,0 +1,827 @@
+# 一种基于PaddleOCR的产品包装生产日期识别模型
+
+- [1. 项目介绍](#1-项目介绍)
+- [2. 环境搭建](#2-环境搭建)
+- [3. 数据准备](#3-数据准备)
+- [4. 直接使用PP-OCRv3模型评估](#4-直接使用PPOCRv3模型评估)
+- [5. 基于合成数据finetune](#5-基于合成数据finetune)
+  - [5.1 Text Renderer数据合成方法](#51-TextRenderer数据合成方法)
+    - [5.1.1 下载Text Renderer代码](#511-下载TextRenderer代码)
+    - [5.1.2 准备背景图片](#512-准备背景图片)
+    - [5.1.3 准备语料](#513-准备语料)
+    - [5.1.4 下载字体](#514-下载字体)
+    - [5.1.5 运行数据合成命令](#515-运行数据合成命令)
+  - [5.2 模型训练](#52-模型训练)
+- [6. 基于真实数据finetune](#6-基于真实数据finetune)
+  - [6.1 python爬虫获取数据](#61-python爬虫获取数据)
+  - [6.2 数据挖掘](#62-数据挖掘)
+  - [6.3 模型训练](#63-模型训练)
+- [7. 基于合成+真实数据finetune](#7-基于合成+真实数据finetune)
+
+
+## 1. 项目介绍
+
+产品包装生产日期是计算机视觉图像识别技术在工业场景中的一种应用。产品包装生产日期识别技术要求能够将产品生产日期从复杂背景中提取并识别出来，在物流管理、物资管理中得到广泛应用。
+
+![](https://ai-studio-static-online.cdn.bcebos.com/d9e0533cc1df47ffa3bbe99de9e42639a3ebfa5bce834bafb1ca4574bf9db684)
+
+
+- 项目难点
+
+1. 没有训练数据
+2. 图像质量层次不齐: 角度倾斜、图片模糊、光照不足、过曝等问题严重
+
+针对以上问题， 本例选用PP-OCRv3这一开源超轻量OCR系统进行包装产品生产日期识别系统的开发。直接使用PP-OCRv3进行评估的精度为62.99%。为提升识别精度，我们首先使用数据合成工具合成了3k数据，基于这部分数据进行finetune，识别精度提升至73.66%。由于合成数据与真实数据之间的分布存在差异，为进一步提升精度，我们使用网络爬虫配合数据挖掘策略得到了1k带标签的真实数据，基于真实数据finetune的精度为71.33%。最后，我们综合使用合成数据和真实数据进行finetune，将识别精度提升至86.99%。各策略的精度提升效果如下：
+
+| 策略 | 精度|
+| :--------------- | :-------- |
+| PP-OCRv3评估 | 62.99|
+| 合成数据finetune | 73.66|
+| 真实数据finetune | 71.33|
+| 真实+合成数据finetune | 86.99|
+
+## 2. 环境搭建
+
+本任务基于Aistudio完成, 具体环境如下：
+
+- 操作系统: Linux
+- PaddlePaddle: 2.3
+- PaddleOCR: Release/2.5
+- text_renderer: master
+
+下载PaddlleOCR代码并安装依赖库:
+```bash
+git clone -b dygraph https://gitee.com/paddlepaddle/PaddleOCR
+
+# 安装依赖库
+cd PaddleOCR
+pip install -r PaddleOCR/requirements.txt
+```
+
+## 3. 数据准备
+
+本项目使用人工预标注的300张图像作为测试集。
+
+部分数据示例如下：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/39ff30e0ab0442579712255e6a9ea6b5271169c98e624e6eb2b8781f003bfea0)
+
+
+标签文件格式如下：
+```txt
+数据路径 标签（中间以制表符分隔）
+```
+
+|数据集类型|数量|
+|---|---|
+|测试集| 300|
+
+数据集[下载链接](https://aistudio.baidu.com/aistudio/datasetdetail/149770)，下载后可以通过下方命令解压:
+
+```bash
+tar -xvf data.tar
+mv data ${PaddleOCR_root}
+```
+
+数据解压后的文件结构如下：
+
+```shell
+PaddleOCR
+├── data
+│   ├── mining_images            # 挖掘的真实数据示例
+│   ├── mining_train.list        # 挖掘的真实数据文件列表
+│   ├── render_images            # 合成数据示例
+│   ├── render_train.list        # 合成数据文件列表
+│   ├── val                      # 测试集数据
+│   └── val.list                 # 测试集数据文件列表
+|   ├── bg                       # 合成数据所需背景图像
+│   └── corpus                   # 合成数据所需语料
+│   └── rec_vit_sub_64_363_all/  # svtr_tiny高精度识别模型
+```
+
+## 4. 直接使用PP-OCRv3模型评估
+
+准备好测试数据后，可以使用PaddleOCR的PP-OCRv3模型进行识别。
+
+- 下载预训练模型
+
+首先需要下载PP-OCR v3中英文识别模型文件，下载链接可以在https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/ppocr_introduction.md#6 获取，下载命令:
+
+```bash
+cd ${PaddleOCR_root}
+mkdir ckpt
+wget -nc -P ckpt https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+pushd ckpt/
+tar -xvf ch_PP-OCRv3_rec_train.tar
+popd
+```
+
+- 模型评估
+
+使用以下命令进行PP-OCRv3评估:
+
+```bash
+python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
+                         -o Global.checkpoints=ckpt/ch_PP-OCRv3_rec_train/best_accuracy \
+                         Eval.dataset.data_dir=./data \
+                         Eval.dataset.label_file_list=["./data/val.list"]
+
+```
+
+其中各参数含义如下：
+
+```bash
+-c: 指定使用的配置文件，ch_PP-OCRv3_rec_distillation.yml对应于OCRv3识别模型。
+-o: 覆盖配置文件中参数
+Global.checkpoints: 指定评估使用的模型文件路径
+Eval.dataset.data_dir: 指定评估数据集路径
+Eval.dataset.label_file_list: 指定评估数据集文件列表
+```
+
+## 5. 基于合成数据finetune
+
+### 5.1 Text Renderer数据合成方法
+
+#### 5.1.1 下载Text Renderer代码
+
+首先从github或gitee下载Text Renderer代码，并安装相关依赖。
+
+```bash
+git clone https://gitee.com/wowowoll/text_renderer.git
+
+# 安装依赖库
+cd text_renderer
+pip install -r requirements.txt
+```
+
+使用text renderer合成数据之前需要准备好背景图片、语料以及字体库，下面将逐一介绍各个步骤。
+
+#### 5.1.2 准备背景图片
+
+观察日常生活中常见的包装生产日期图片，我们可以发现其背景相对简单。为此我们可以从网上找一下图片，截取部分图像块作为背景图像。
+
+本项目已准备了部分图像作为背景图片，在第3部分完成数据准备后,可以得到我们准备好的背景图像，示例如下：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/456ae2acb27d4a94896c478812aee0bc3551c703d7bd40c9be4dc983c7b3fc8a)
+
+
+
+背景图像存放于如下位置：
+
+```shell
+PaddleOCR
+├── data
+｜   ├── bg     # 合成数据所需背景图像
+```
+
+#### 5.1.3 准备语料
+
+观察测试集生产日期图像，我们可以知道如下数据有如下特点：
+1. 由年月日组成，中间可能以“/”、“-”、“:”、“.”或者空格间隔，也可能以汉字年月日分隔
+2. 有些生产日期包含在产品批号中，此时可能包含具体时间、英文字母或数字标识
+
+基于以上两点，我们编写语料生成脚本：
+
+```python
+import random
+from random import choice
+import os
+
+cropus_num = 2000 #设置语料数量
+
+def get_cropus(f):
+    # 随机生成年份
+    year = random.randint(0, 22)
+    # 随机生成月份
+    month = random.randint(1, 12)
+    # 随机生成日期
+    day_dict = {31: [1,3,5,7,8,10,12], 30: [4,6,9,11], 28: [2]}
+    for item in day_dict:
+        if month in day_dict[item]:
+            day = random.randint(0, item)
+    # 随机生成小时
+    hours = random.randint(0, 24)
+    # 随机生成分钟
+    minute = random.randint(0, 60)
+     # 随机生成秒数
+    second = random.randint(0, 60)
+
+    # 随机生成产品标识字符
+    length = random.randint(0, 6)
+    file_id = []
+    flag = 0
+    my_dict = [i for i in range(48,58)] + [j for j in range(40, 42)] + [k for k in range(65,90)]  # 大小写字母 + 括号
+
+    for i in range(1, length):
+        if flag:
+            if i == flag+2:  #括号匹配
+                file_id.append(')')
+                flag = 0
+                continue
+        sel = choice(my_dict)
+        if sel == 41:
+            continue
+        if sel == 40:
+            if i == 1 or i > length-3:
+                continue
+            flag = i
+        my_ascii = chr(sel)
+        file_id.append(my_ascii)
+    file_id_str = ''.join(file_id)
+
+    #随机生成产品标识字符
+    file_id2 = random.randint(0, 9)
+
+    rad = random.random()
+    if rad < 0.3:
+        f.write('20{:02d}{:02d}{:02d} {}'.format(year, month, day, file_id_str))
+    elif 0.3 < rad < 0.5:
+        f.write('20{:02d}年{:02d}月{:02d}日'.format(year, month, day))
+    elif 0.5 < rad < 0.7:
+        f.write('20{:02d}/{:02d}/{:02d}'.format(year, month, day))
+    elif 0.7 < rad < 0.8:
+        f.write('20{:02d}-{:02d}-{:02d}'.format(year, month, day))
+    elif 0.8 < rad < 0.9:
+        f.write('20{:02d}.{:02d}.{:02d}'.format(year, month, day))  
+    else:
+        f.write('{:02d}:{:02d}:{:02d} {:02d}'.format(hours, minute, second, file_id2))
+
+if __name__ == "__main__":
+    file_path = '/home/aistudio/text_renderer/my_data/cropus'
+    if not os.path.exists(file_path):
+        os.makedirs(file_path)
+    file_name = os.path.join(file_path, 'books.txt')
+    f = open(file_name, 'w')
+    for i in range(cropus_num):
+        get_cropus(f)
+        if i < cropus_num-1:
+            f.write('\n')
+
+    f.close()
+```
+
+本项目已准备了部分语料，在第3部分完成数据准备后,可以得到我们准备好的语料库，默认位置如下：
+
+```shell
+PaddleOCR
+├── data
+│   └── corpus              #合成数据所需语料
+```
+
+#### 5.1.4 下载字体
+
+观察包装生产日期，我们可以发现其使用的字体为点阵体。字体可以在如下网址下载：
+https://www.fonts.net.cn/fonts-en/tag-dianzhen-1.html
+
+本项目已准备了部分字体，在第3部分完成数据准备后,可以得到我们准备好的字体，默认位置如下：
+
+```shell
+PaddleOCR
+├── data
+│   └── fonts                #合成数据所需字体
+```
+
+下载好字体后，还需要在list文件中指定字体文件存放路径，脚本如下:
+
+```bash
+cd text_renderer/my_data/
+touch fonts.list
+ls /home/aistudio/PaddleOCR/data/fonts/* > fonts.list
+```
+
+#### 5.1.5 运行数据合成命令
+
+完成数据准备后，my_data文件结构如下：
+
+```shell
+my_data/
+├── cropus
+│   └── books.txt #语料库
+├── eng.txt    #字符列表
+└── fonts.list #字体列表
+```
+
+在运行合成数据命令之前，还有两处细节需要手动修改：
+1. 将默认配置文件`text_renderer/configs/default.yaml`中第9行enable的值设为`true`，即允许合成彩色图像。否则合成的都是灰度图。
+
+```yaml
+ # color boundary is in R,G,B format
+ font_color:
++  enable: true #false
+```
+
+2. 将`text_renderer/textrenderer/renderer.py`第184行作如下修改，取消padding。否则图片两端会有一些空白。
+
+```python
+padding = random.randint(s_bbox_width // 10, s_bbox_width // 8) #修改前
+padding = 0 #修改后
+```
+
+运行数据合成命令:
+
+```bash
+cd /home/aistudio/text_renderer/
+python main.py --num_img=3000 \
+                  --fonts_list='./my_data/fonts.list' \
+                  --corpus_dir "./my_data/cropus" \
+                  --corpus_mode "list" \
+                  --bg_dir "/home/aistudio/PaddleOCR/data/bg/" \
+                  --img_width 0
+```
+
+合成好的数据默认保存在`text_renderer/output`目录下，可进入该目录查看合成的数据。
+
+
+合成数据示例如下
+![](https://ai-studio-static-online.cdn.bcebos.com/d686a48d465a43d09fbee51924fdca42ee21c50e676646da8559fb9967b94185)
+
+数据合成好后，还需要生成如下格式的训练所需的标注文件，
+```
+图像路径 标签
+```
+
+使用如下脚本即可生成标注文件：
+
+```python
+import random
+
+abspath = '/home/aistudio/text_renderer/output/default/'
+
+#标注文件生成路径
+fout = open('./render_train.list', 'w', encoding='utf-8')
+
+with open('./output/default/tmp_labels.txt','r') as f:
+    lines = f.readlines()
+    for item in lines:
+        label = item[9:]
+        filename = item[:8] + '.jpg'
+        fout.write(abspath + filename + '\t' + label)
+
+    fout.close()
+```
+
+经过以上步骤，我们便完成了包装生产日期数据合成。
+数据位于`text_renderer/output`，标注文件位于`text_renderer/render_train.list`。
+
+本项目提供了生成好的数据供大家体验,完成步骤3的数据准备后，可得数据路径位于:
+
+```shell
+PaddleOCR
+├── data
+│   ├── render_images     # 合成数据示例
+│   ├── render_train.list   #合成数据文件列表
+```
+
+###  5.2 模型训练
+
+准备好合成数据后，我们可以使用以下命令，利用合成数据进行finetune:
+```bash
+cd ${PaddleOCR_root}
+python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
+                       -o Global.pretrained_model=./ckpt/ch_PP-OCRv3_rec_train/best_accuracy \
+                       Global.epoch_num=20 \
+                       Global.eval_batch_step='[0, 20]' \
+                       Train.dataset.data_dir=./data \
+                       Train.dataset.label_file_list=['./data/render_train.list'] \
+                       Train.loader.batch_size_per_card=64 \
+                       Eval.dataset.data_dir=./data \
+                       Eval.dataset.label_file_list=["./data/val.list"] \
+                       Eval.loader.batch_size_per_card=64
+
+```
+
+其中各参数含义如下：
+
+```txt
+-c: 指定使用的配置文件，ch_PP-OCRv3_rec_distillation.yml对应于OCRv3识别模型。
+-o: 覆盖配置文件中参数
+Global.pretrained_model: 指定finetune使用的预训练模型
+Global.epoch_num: 指定训练的epoch数
+Global.eval_batch_step: 间隔多少step做一次评估
+Train.dataset.data_dir: 训练数据集路径
+Train.dataset.label_file_list: 训练集文件列表
+Train.loader.batch_size_per_card: 训练单卡batch size
+Eval.dataset.data_dir: 评估数据集路径
+Eval.dataset.label_file_list: 评估数据集文件列表
+Eval.loader.batch_size_per_card: 评估单卡batch size
+```
+
+## 6. 基于真实数据finetune
+
+
+使用合成数据finetune能提升我们模型的识别精度，但由于合成数据和真实数据之间的分布可能有一定差异，因此作用有限。为进一步提高识别精度，本节介绍如何挖掘真实数据进行模型finetune。
+
+数据挖掘的整体思路如下：
+1. 使用python爬虫从网上获取大量无标签数据
+2. 使用模型从大量无标签数据中构建出有效训练集
+
+### 6.1 python爬虫获取数据
+
+本节介绍如何使用python脚本爬取网络图片，这里以爬取百度图片为例。首先将图片链接都爬取下来，然后用多进程下载图片。使用到的工具有requests库和谷歌浏览器（抓包工具）。
+
+准备工作：
+1. 获取请求头需要的的设备信息：
+(1) 使用谷歌浏览器，打开百度图片，搜索关键词【包装生产日期】
+(2) 右键，选择【检查】，打开抓包工具
+(3) Network >> XHR >> 选定网址 >> Headers >> RequestHeaders >> UserAgents
+![](https://ai-studio-static-online.cdn.bcebos.com/715fa3f1255544789ddf5699084783b6b88055835a7049e3b588f4f564e14972)
+
+2. 获取网页参数
+用类似的方式，在XHR中选择Payload获取网页参数
+![](https://ai-studio-static-online.cdn.bcebos.com/2a8b91ed854a40e589cc9a9596873192cc730c60d5754373a403f395a81ec44f)
+
+
+接下来编写代码进行图片抓取：
+
+```python
+import requests
+import json
+import tqdm
+import os
+
+img_num = 0        # 记录保存下来的图片数量
+num_per_page = 50  # 每一页展示的图片数量
+page_num = 40      # 页数
+page_id = 0        # 每一页开始索引
+
+# 图片链接保存文件路径
+list_path = '/home/aistudio/sprider'
+if not os.path.exists(list_path):
+    os.mkdir(list_path)
+f_list = open(os.path.join(list_path, 'file.list'), 'w')
+
+
+img_dict = {}   # 图片链接字典，去重
+
+# 搜索关键词
+key_word = ['包装产品生产日期', '包装日期', '生产日期', '保质期', '纸箱生产日期', '商品日期', '黑色生产日期', '黑底生产日期']
+
+for key in key_word:
+    bar = tqdm.tqdm(total=page_num*num_per_page)
+    for page_id in range(page_num):
+        # 使用准备好的请求头设备信息
+        header = {
+            'User-Agent':'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/101.0.4951.64 Safari/537.36'
+        }
+
+        # 使用准备好的网页参数
+        url = 'https://image.baidu.com/search/acjson?'
+        param = {
+        'tn': 'resultjson_com',
+        'logid': '8931113288182398301',
+        'ipn': 'rj',
+        'ct': '201326592',
+        'is': '',
+        'fp': 'result',
+        'fr':'',
+        'word': key,       # 搜索关键词
+        'queryWord': key,  # 搜索关键词
+        'cl': '2',
+        'lm': '-1',
+        'ie': 'utf-8',
+        'oe': 'utf-8',
+        'adpicid': '',
+        'st': '-1',
+        'z': '',
+        'ic': '0',
+        'hd': '',
+        'latest': '',
+        'copyright': '',
+        's': '',
+        'se': '',
+        'tab': '',
+        'width': '',
+        'height': '',
+        'face': '0',
+        'istype': '2',
+        'qc': '',
+        'nc': '1',
+        'expermode': '',
+        'nojc': '',
+        'isAsync': '',
+        'pn': str(page_id*num_per_page),  # 每一页开始索引
+        'rn': str(num_per_page),          # 每一页爬取的图片数量
+        'gsm': '1e',
+        '1652689395185':''
+        }
+
+        # 获取网页信息  
+        page_text = requests.get(url=url, headers=header, params=param, timeout=20)
+
+        try:
+            page_text = page_text.json()  # 网页信息转成json字典
+        except:
+            continue
+
+        info_list = page_text['data']
+
+        # 获取不到有效信息后，退出搜索
+        if info_list == [{}]:
+            print('===break===')
+            break
+
+        del info_list[-1]
+
+        img_path_list = []
+        for info in info_list:
+            img_url = info['thumbURL']   # 图片源地址
+            #bar.update(1)
+            if img_url not in img_dict:
+                img_dict[img_url] = 1
+                f_list.write(img_url + '\n')
+
+f_list.close()
+print('saved image number:', img_num+1)
+```
+
+```python
+# 使用多进程下载网页图片
+
+import requests
+from concurrent.futures import ProcessPoolExecutor
+
+file_path = '/home/aistudio/sprider/file.list'
+image_save_path = '/home/aistudio/sprider/data'
+
+if not os.path.exists(image_save_path):
+    os.mkdir(image_save_path)
+
+f = open(file_path, 'r')
+lines = f.readlines()
+
+header = {
+    'User-Agent':'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/101.0.4951.64 Safari/537.36'
+}
+
+def process(img_url, img_num):
+    try:
+        img_data = requests.get(url=img_url, headers=header, timeout=20).content
+        img_path = os.path.join(image_save_path, '{:05d}.jpg'.format(img_num))
+        with open(img_path, 'wb') as fp:
+            fp.write(img_data)
+        img_num += 1
+    except Exception as e:
+        print('connect Failed, just continue')
+
+with ProcessPoolExecutor(max_workers=10) as executer:
+    for idx, line in enumerate(lines):
+        executer.submit(process, line.strip(), idx)
+
+print('done')
+```
+
+完成以上步骤，便可得到获取的图片，目录格式如下：
+
+```txt
+sprider
+├── file.list
+├── data
+│   ├── 00000.jpg
+│   ├── 00001.jpg
+...
+```
+
+### 6.2 数据挖掘
+
+我们使用PaddleOCR对获取到的图片进行挖掘，具体步骤如下：
+1. 使用 PP-OCRv3检测模型+svtr-tiny识别模型，对每张图片进行预测。
+2. 使用数据挖掘策略，得到有效图片。
+3. 将有效图片对应的图像区域和标签提取出来，构建训练集。
+
+
+首先下载预训练模型，PP-OCRv3检测模型下载链接：https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar
+
+svtr-tiny识别模型，我们使用的是在内部数据上训练的高精度模型，本项目提供下载链接，完成步骤3的数据准备后，可得数据路径位于:
+
+```shell
+PaddleOCR
+├── data
+│   ├── rec_vit_sub_64_363_all/  # svtr_tiny高精度识别模型
+```
+
+```bash
+# 下载解压PP-OCRv3检测模型
+cd ${PaddleOCR_root}
+wget -nc -P ckpt https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar
+pushd ckpt
+tar -xvf ch_PP-OCRv3_det_infer.tar
+popd ckpt
+```
+
+在使用PPOCRv3检测模型+svtr-tiny识别模型进行预测之前，有如下两处细节需要手动修改：
+1. 将`tools/infer/predict_rec.py`中第110行`imgW`修改为`320`
+
+```python
+#imgW = int((imgH * max_wh_ratio))
+imgW = 320
+```
+
+2. 将`tools/infer/predict_system.py`第169行添加如下一行，将预测分数也写入结果文件中。
+
+```python
+"scores": rec_res[idx][1],
+```
+
+模型预测命令:
+```bash
+python tools/infer/predict_system.py \
+        --image_dir="/home/aistudio/sprider/data" \
+        --det_model_dir="./ckpt/ch_PP-OCRv3_det_infer/"  \
+        --rec_model_dir="/home/aistudio/PaddleOCR/data/rec_vit_sub_64_363_all/" \
+        --rec_image_shape="3,32,320"
+```
+
+获得预测结果后，我们使用数据挖掘策略得到有效图片。具体挖掘策略如下：
+1. 预测置信度高于95%
+2. 识别结果包含字符‘20’，即年份
+3. 没有中文，或者有中文并且‘日’和'月'同时在识别结果中
+
+```python
+# 获取有效预测
+
+import json
+import re
+
+zh_pattern = re.compile(u'[\u4e00-\u9fa5]+')  #正则表达式，筛选字符是否包含中文
+
+file_path = '/home/aistudio/PaddleOCR/inference_results/system_results.txt'
+out_path = '/home/aistudio/PaddleOCR/selected_results.txt'
+f_out = open(out_path, 'w')
+
+with open(file_path, "r", encoding='utf-8') as fin:
+    lines = fin.readlines()
+
+
+for line in lines:
+    flag = False
+    # 读取文件内容
+    file_name, json_file = line.strip().split('\t')
+    preds = json.loads(json_file)
+    res = []
+    for item in preds:
+        transcription = item['transcription'] #获取识别结果
+        scores = item['scores']               #获取识别得分
+        # 挖掘策略
+        if scores > 0.95:
+            if '20' in transcription and len(transcription) > 4 and len(transcription) < 12:
+                word = transcription
+                if not(zh_pattern.search(word) and ('日' not in word or '月' not in word)):
+                    flag = True
+                    res.append(item)
+    save_pred = file_name + "\t" + json.dumps(
+        res, ensure_ascii=False) + "\n"
+    if flag ==True:
+        f_out.write(save_pred)
+
+f_out.close()
+```
+
+然后将有效预测对应的图像区域和标签提取出来，构建训练集。具体实现脚本如下：
+
+```python
+import cv2
+import json
+import numpy as np
+
+PATH = '/home/aistudio/PaddleOCR/inference_results/'  #数据原始路径
+SAVE_PATH = '/home/aistudio/mining_images/'             #裁剪后数据保存路径
+file_list = '/home/aistudio/PaddleOCR/selected_results.txt' #数据预测结果
+label_file = '/home/aistudio/mining_images/mining_train.list'  #输出真实数据训练集标签list
+
+if not os.path.exists(SAVE_PATH):
+    os.mkdir(SAVE_PATH)
+
+f_label = open(label_file, 'w')
+
+
+def get_rotate_crop_image(img, points):
+    """
+    根据检测结果points，从输入图像img中裁剪出相应的区域
+    """
+    assert len(points) == 4, "shape of points must be 4*2"
+    img_crop_width = int(
+        max(
+            np.linalg.norm(points[0] - points[1]),
+            np.linalg.norm(points[2] - points[3])))
+    img_crop_height = int(
+        max(
+            np.linalg.norm(points[0] - points[3]),
+            np.linalg.norm(points[1] - points[2])))
+    pts_std = np.float32([[0, 0], [img_crop_width, 0],
+                          [img_crop_width, img_crop_height],
+                          [0, img_crop_height]])
+    M = cv2.getPerspectiveTransform(points, pts_std)
+    # 形变或倾斜，会做透视变换，reshape成矩形
+    dst_img = cv2.warpPerspective(
+        img,
+        M, (img_crop_width, img_crop_height),
+        borderMode=cv2.BORDER_REPLICATE,
+        flags=cv2.INTER_CUBIC)
+    dst_img_height, dst_img_width = dst_img.shape[0:2]
+    if dst_img_height * 1.0 / dst_img_width >= 1.5:
+        dst_img = np.rot90(dst_img)
+    return dst_img
+
+def crop_and_get_filelist(file_list):
+    with open(file_list, "r", encoding='utf-8') as fin:
+        lines = fin.readlines()
+
+    img_num = 0
+    for line in lines:
+        img_name, json_file = line.strip().split('\t')
+        preds = json.loads(json_file)
+        for item in preds:
+            transcription = item['transcription']
+            points = item['points']
+            points = np.array(points).astype('float32')
+            #print('processing {}...'.format(img_name))
+
+            img = cv2.imread(PATH+img_name)
+            dst_img = get_rotate_crop_image(img, points)
+            h, w, c = dst_img.shape
+            newWidth = int((32. / h) * w)
+            newImg = cv2.resize(dst_img, (newWidth, 32))
+            new_img_name = '{:05d}.jpg'.format(img_num)
+            cv2.imwrite(SAVE_PATH+new_img_name, dst_img)
+            f_label.write(SAVE_PATH+new_img_name+'\t'+transcription+'\n')
+            img_num += 1
+
+
+crop_and_get_filelist(file_list)
+f_label.close()
+```
+
+### 6.3 模型训练
+
+通过数据挖掘，我们得到了真实场景数据和对应的标签。接下来使用真实数据finetune，观察精度提升效果。
+
+
+利用真实数据进行finetune:
+
+```bash
+cd ${PaddleOCR_root}
+python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
+                       -o Global.pretrained_model=./ckpt/ch_PP-OCRv3_rec_train/best_accuracy \
+                       Global.epoch_num=20 \
+                       Global.eval_batch_step='[0, 20]' \
+                       Train.dataset.data_dir=./data \
+                       Train.dataset.label_file_list=['./data/mining_train.list'] \
+                       Train.loader.batch_size_per_card=64 \
+                       Eval.dataset.data_dir=./data \
+                       Eval.dataset.label_file_list=["./data/val.list"] \
+                       Eval.loader.batch_size_per_card=64
+```
+
+各参数含义参考第6部分合成数据finetune，只需要对训练数据路径做相应的修改：
+
+```txt
+Train.dataset.data_dir: 训练数据集路径
+Train.dataset.label_file_list: 训练集文件列表
+```
+
+示例使用我们提供的真实数据进行finetune，如想换成自己的数据，只需要相应的修改`Train.dataset.data_dir`和`Train.dataset.label_file_list`参数即可。
+
+由于数据量不大，这里仅训练20个epoch即可。训练完成后，可以得到合成数据finetune后的精度为best acc=**71.33%**。
+
+由于数量比较少，精度会比合成数据finetue的略低。
+
+
+## 7. 基于合成+真实数据finetune
+
+为进一步提升模型精度，我们结合使用合成数据和挖掘到的真实数据进行finetune。
+
+利用合成+真实数据进行finetune，各参数含义参考第6部分合成数据finetune，只需要对训练数据路径做相应的修改：
+
+```txt
+Train.dataset.data_dir: 训练数据集路径
+Train.dataset.label_file_list: 训练集文件列表
+```
+
+生成训练list文件:
+```bash
+# 生成训练集文件list
+cat /home/aistudio/PaddleOCR/data/render_train.list /home/aistudio/PaddleOCR/data/mining_train.list > /home/aistudio/PaddleOCR/data/render_mining_train.list
+```
+
+启动训练:
+```bash
+cd ${PaddleOCR_root}
+python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
+                       -o Global.pretrained_model=./ckpt/ch_PP-OCRv3_rec_train/best_accuracy \
+                       Global.epoch_num=40 \
+                       Global.eval_batch_step='[0, 20]' \
+                       Train.dataset.data_dir=./data \
+                       Train.dataset.label_file_list=['./data/render_mining_train.list'] \
+                       Train.loader.batch_size_per_card=64 \
+                       Eval.dataset.data_dir=./data \
+                       Eval.dataset.label_file_list=["./data/val.list"] \
+                       Eval.loader.batch_size_per_card=64
+```
+
+示例使用我们提供的真实+合成数据进行finetune，如想换成自己的数据，只需要相应的修改Train.dataset.data_dir和Train.dataset.label_file_list参数即可。
+
+由于数据量不大，这里仅训练40个epoch即可。训练完成后，可以得到合成数据finetune后的精度为best acc=**86.99%**。
+
+可以看到，相较于原始PP-OCRv3的识别精度62.99%，使用合成数据+真实数据finetune后，识别精度能提升24%。
+
+训练完成后，PP-OCRv3模型的推理部署方法可以参考repo文档： https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/deploy/README_ch.md

From 8b0b5d1ee56bdef51bb0e1d9897d9051df291d4f Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 21 Jun 2022 15:36:36 +0800
Subject: [PATCH 033/134] pre-commit

---
 applications/轻量级车牌识别.md | 2902 ++++++++++++++++++++++++++++++++
 1 file changed, 2902 insertions(+)
 create mode 100644 applications/轻量级车牌识别.md

diff --git a/applications/轻量级车牌识别.md b/applications/轻量级车牌识别.md
new file mode 100644
index 0000000000..e0f4706f14
--- /dev/null
+++ b/applications/轻量级车牌识别.md
@@ -0,0 +1,2902 @@
+
+
+
+
+
+
+
+<!DOCTYPE html>
+<html lang="en" data-color-mode="auto" data-light-theme="light" data-dark-theme="dark" data-a11y-animated-images="system">
+  <head>
+    <meta charset="utf-8">
+  <link rel="dns-prefetch" href="https://github.githubassets.com">
+  <link rel="dns-prefetch" href="https://avatars.githubusercontent.com">
+  <link rel="dns-prefetch" href="https://github-cloud.s3.amazonaws.com">
+  <link rel="dns-prefetch" href="https://user-images.githubusercontent.com/">
+  <link rel="preconnect" href="https://github.githubassets.com" crossorigin>
+  <link rel="preconnect" href="https://avatars.githubusercontent.com">
+
+
+
+  <link crossorigin="anonymous" media="all" integrity="sha512-ksfTgQOOnE+FFXf+yNfVjKSlEckJAdufFIYGK7ZjRhWcZgzAGcmZqqArTgMLpu90FwthqcCX4ldDgKXbmVMeuQ==" rel="stylesheet" href="https://github.githubassets.com/assets/light-92c7d381038e.css" /><link crossorigin="anonymous" media="all" integrity="sha512-1KkMNn8M/al/dtzBLupRwkIOgnA9MWkm8oxS+solP87jByEvY/g4BmoxLihRogKcX1obPnf4Yp7dI0ZTWO+ljg==" rel="stylesheet" href="https://github.githubassets.com/assets/dark-d4a90c367f0c.css" /><link data-color-theme="dark_dimmed" crossorigin="anonymous" media="all" integrity="sha512-cZa7DZqvMBwD236uzEunO/G1dvw8/QftyT2UtLWKQFEy0z0eq0R5WPwqVME+3NSZG1YaLJAaIqtU+m0zWf/6SQ==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_dimmed-7196bb0d9aaf.css" /><link data-color-theme="dark_high_contrast" crossorigin="anonymous" media="all" integrity="sha512-WVoKqJ4y1nLsdNH4RkRT5qrM9+n9RFe1RHSiTnQkBf5TSZkJEc9GpLpTIS7T15EQaUQBJ8BwmKvwFPVqfpTEIQ==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_high_contrast-595a0aa89e32.css" /><link data-color-theme="dark_colorblind" crossorigin="anonymous" media="all" integrity="sha512-XpAMBMSRZ6RTXgepS8LjKiOeNK3BilRbv8qEiA/M3m+Q4GoqxtHedOI5BAZRikCzfBL4KWYvVzYZSZ8Gp/UnUg==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_colorblind-5e900c04c491.css" /><link data-color-theme="light_colorblind" crossorigin="anonymous" media="all" integrity="sha512-3HF2HZ4LgEIQm77yOzoeR20CX1n2cUQlcywscqF4s+5iplolajiHV7E5ranBwkX65jN9TNciHEVSYebQ+8xxEw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_colorblind-dc71761d9e0b.css" /><link data-color-theme="light_high_contrast" crossorigin="anonymous" media="all" integrity="sha512-+J8j3T0kbK9/sL3zbkCfPtgYcRD4qQfRbT6xnfOrOTjvz4zhr0M7AXPuE642PpaxGhHs1t77cTtieW9hI2K6Gw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_high_contrast-f89f23dd3d24.css" /><link data-color-theme="light_tritanopia" crossorigin="anonymous" media="all" integrity="sha512-AQeAx5wHQAXNf0DmkvVlHYwA3f6BkxunWTI0GGaRN57GqD+H9tW8RKIKlopLS0qGaC54seFsPc601GDlqIuuHg==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_tritanopia-010780c79c07.css" /><link data-color-theme="dark_tritanopia" crossorigin="anonymous" media="all" integrity="sha512-+u5pmgAE0T03d/yI6Ha0NWwz6Pk0W6S6WEfIt8veDVdK8NTjcMbZmQB9XUCkDlrBoAKkABva8HuGJ+SzEpV1Uw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_tritanopia-faee699a0004.css" />
+
+    <link crossorigin="anonymous" media="all" integrity="sha512-EAhBCLIJ/pXHG3Y6yQhs9s53SHV80sjJ+yCwlQtfv7LaVkD+VoEuZBZ5betQJFUNj/5qBSfZk5GFtazEDzWLAg==" rel="stylesheet" href="https://github.githubassets.com/assets/primer-10084108b209.css" />
+    <link crossorigin="anonymous" media="all" integrity="sha512-4cUFwwYgpnyRoJKr15lavpQFa6HaoGIW6sap6JXLP9cQ33dOLfadeyjfJypUJ8GfE8MnAfdfvcNZlgn7oZ70PA==" rel="stylesheet" href="https://github.githubassets.com/assets/global-e1c505c30620.css" />
+    <link crossorigin="anonymous" media="all" integrity="sha512-ZkqzH/tvOPQgSkHnN9TlzL/abuPzgfV2KLooWN2Xq+jrP7VFzNKd0dib2+zSde1277kCzr5gp1piayo6AthIsQ==" rel="stylesheet" href="https://github.githubassets.com/assets/github-664ab31ffb6f.css" />
+  <link crossorigin="anonymous" media="all" integrity="sha512-PXtwH8brj9qH7LskWA0iAMLw4Wx3RXN7txSxjlqdzQEalZKu3UcsQ+ajfHJMDkYDA2iWL1FIrf2zyLflA5Xu7A==" rel="stylesheet" href="https://github.githubassets.com/assets/code-3d7b701fc6eb.css" />
+
+    <meta name="optimizely-datafile" content="{&quot;groups&quot;: [], &quot;environmentKey&quot;: &quot;production&quot;, &quot;rollouts&quot;: [], &quot;typedAudiences&quot;: [], &quot;projectId&quot;: &quot;16737760170&quot;, &quot;variables&quot;: [], &quot;featureFlags&quot;: [], &quot;experiments&quot;: [{&quot;status&quot;: &quot;Running&quot;, &quot;audienceIds&quot;: [], &quot;variations&quot;: [{&quot;variables&quot;: [], &quot;id&quot;: &quot;20667381018&quot;, &quot;key&quot;: &quot;control&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;20680930759&quot;, &quot;key&quot;: &quot;treatment&quot;}], &quot;id&quot;: &quot;20652570897&quot;, &quot;key&quot;: &quot;project_genesis&quot;, &quot;layerId&quot;: &quot;20672300363&quot;, &quot;trafficAllocation&quot;: [{&quot;entityId&quot;: &quot;20667381018&quot;, &quot;endOfRange&quot;: 5000}, {&quot;entityId&quot;: &quot;20680930759&quot;, &quot;endOfRange&quot;: 10000}], &quot;forcedVariations&quot;: {&quot;83356e17066d336d1803024138ecb683&quot;: &quot;treatment&quot;, &quot;18e31c8a9b2271332466133162a4aa0d&quot;: &quot;treatment&quot;, &quot;10f8ab3fbc5ebe989a36a05f79d48f32&quot;: &quot;treatment&quot;, &quot;1686089f6d540cd2deeaec60ee43ecf7&quot;: &quot;treatment&quot;}}, {&quot;status&quot;: &quot;Running&quot;, &quot;audienceIds&quot;: [], &quot;variations&quot;: [{&quot;variables&quot;: [], &quot;id&quot;: &quot;21427950901&quot;, &quot;key&quot;: &quot;control&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;21429710665&quot;, &quot;key&quot;: &quot;beginner&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;21437291543&quot;, &quot;key&quot;: &quot;upstart&quot;}], &quot;id&quot;: &quot;21445030708&quot;, &quot;key&quot;: &quot;_259_zero_user_dashboard&quot;, &quot;layerId&quot;: &quot;21434011841&quot;, &quot;trafficAllocation&quot;: [{&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 3334}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 5000}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 8333}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 10000}], &quot;forcedVariations&quot;: {&quot;3c64268131793aa297119a343c19e345&quot;: &quot;beginner&quot;, &quot;95b24126db31ea8693c0fe5ea9f53b65&quot;: &quot;beginner&quot;, &quot;086e2abe64e9101112af53b95d2d90b9&quot;: &quot;upstart&quot;, &quot;bae688df9d297afac98e2d254e912ada&quot;: &quot;control&quot;, &quot;6c2cfda7c41396fcc31a4db759a42b94&quot;: &quot;beginner&quot;, &quot;16ed2b4ff7de02663b7c606309695916&quot;: &quot;control&quot;, &quot;1971768911.1635962195&quot;: &quot;beginner&quot;, &quot;830bf802470ec6c9c5800c99d8e57445&quot;: &quot;beginner&quot;}}], &quot;version&quot;: &quot;4&quot;, &quot;audiences&quot;: [{&quot;conditions&quot;: &quot;[\&quot;or\&quot;, {\&quot;match\&quot;: \&quot;exact\&quot;, \&quot;name\&quot;: \&quot;$opt_dummy_attribute\&quot;, \&quot;type\&quot;: \&quot;custom_attribute\&quot;, \&quot;value\&quot;: \&quot;$opt_dummy_value\&quot;}]&quot;, &quot;id&quot;: &quot;$opt_dummy_audience&quot;, &quot;name&quot;: &quot;Optimizely-Generated Audience for Backwards Compatibility&quot;}], &quot;anonymizeIP&quot;: true, &quot;sdkKey&quot;: &quot;WTc6awnGuYDdG98CYRban&quot;, &quot;attributes&quot;: [{&quot;id&quot;: &quot;16822470375&quot;, &quot;key&quot;: &quot;user_id&quot;}, {&quot;id&quot;: &quot;17143601254&quot;, &quot;key&quot;: &quot;spammy&quot;}, {&quot;id&quot;: &quot;18175660309&quot;, &quot;key&quot;: &quot;organization_plan&quot;}, {&quot;id&quot;: &quot;18813001570&quot;, &quot;key&quot;: &quot;is_logged_in&quot;}, {&quot;id&quot;: &quot;19073851829&quot;, &quot;key&quot;: &quot;geo&quot;}, {&quot;id&quot;: &quot;20175462351&quot;, &quot;key&quot;: &quot;requestedCurrency&quot;}, {&quot;id&quot;: &quot;20785470195&quot;, &quot;key&quot;: &quot;country_code&quot;}, {&quot;id&quot;: &quot;21656311196&quot;, &quot;key&quot;: &quot;opened_downgrade_dialog&quot;}], &quot;botFiltering&quot;: false, &quot;accountId&quot;: &quot;16737760170&quot;, &quot;events&quot;: [{&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;17911811441&quot;, &quot;key&quot;: &quot;hydro_click.dashboard.teacher_toolbox_cta&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18124116703&quot;, &quot;key&quot;: &quot;submit.organizations.complete_sign_up&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18145892387&quot;, &quot;key&quot;: &quot;no_metric.tracked_outside_of_optimizely&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18178755568&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.add_repo&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18180553241&quot;, &quot;key&quot;: &quot;submit.repository_imports.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18186103728&quot;, &quot;key&quot;: &quot;click.help.learn_more_about_repository_creation&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18188530140&quot;, &quot;key&quot;: &quot;test_event&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18191963644&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.transfer_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18195612788&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.import_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18210945499&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.invite_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18211063248&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.create_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18215721889&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.update_profile&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18224360785&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.dismiss&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18234832286&quot;, &quot;key&quot;: &quot;submit.organization_activation.complete&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18252392383&quot;, &quot;key&quot;: &quot;submit.org_repository.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18257551537&quot;, &quot;key&quot;: &quot;submit.org_member_invitation.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18259522260&quot;, &quot;key&quot;: &quot;submit.organization_profile.update&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18564603625&quot;, &quot;key&quot;: &quot;view.classroom_select_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18568612016&quot;, &quot;key&quot;: &quot;click.classroom_sign_in_click&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18572592540&quot;, &quot;key&quot;: &quot;view.classroom_name&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18574203855&quot;, &quot;key&quot;: &quot;click.classroom_create_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18582053415&quot;, &quot;key&quot;: &quot;click.classroom_select_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18589463420&quot;, &quot;key&quot;: &quot;click.classroom_create_classroom&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18591323364&quot;, &quot;key&quot;: &quot;click.classroom_create_first_classroom&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18591652321&quot;, &quot;key&quot;: &quot;click.classroom_grant_access&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18607131425&quot;, &quot;key&quot;: &quot;view.classroom_creation&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18831680583&quot;, &quot;key&quot;: &quot;upgrade_account_plan&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19064064515&quot;, &quot;key&quot;: &quot;click.signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19075373687&quot;, &quot;key&quot;: &quot;click.view_account_billing_page&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19077355841&quot;, &quot;key&quot;: &quot;click.dismiss_signup_prompt&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19079713938&quot;, &quot;key&quot;: &quot;click.contact_sales&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19120963070&quot;, &quot;key&quot;: &quot;click.compare_account_plans&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19151690317&quot;, &quot;key&quot;: &quot;click.upgrade_account_cta&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19424193129&quot;, &quot;key&quot;: &quot;click.open_account_switcher&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19520330825&quot;, &quot;key&quot;: &quot;click.visit_account_profile&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19540970635&quot;, &quot;key&quot;: &quot;click.switch_account_context&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19730198868&quot;, &quot;key&quot;: &quot;submit.homepage_signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19820830627&quot;, &quot;key&quot;: &quot;click.homepage_signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19988571001&quot;, &quot;key&quot;: &quot;click.create_enterprise_trial&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20036538294&quot;, &quot;key&quot;: &quot;click.create_organization_team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20040653299&quot;, &quot;key&quot;: &quot;click.input_enterprise_trial_form&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20062030003&quot;, &quot;key&quot;: &quot;click.continue_with_team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20068947153&quot;, &quot;key&quot;: &quot;click.create_organization_free&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20086636658&quot;, &quot;key&quot;: &quot;click.signup_continue.username&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20091648988&quot;, &quot;key&quot;: &quot;click.signup_continue.create_account&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20103637615&quot;, &quot;key&quot;: &quot;click.signup_continue.email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20111574253&quot;, &quot;key&quot;: &quot;click.signup_continue.password&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20120044111&quot;, &quot;key&quot;: &quot;view.pricing_page&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20152062109&quot;, &quot;key&quot;: &quot;submit.create_account&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20165800992&quot;, &quot;key&quot;: &quot;submit.upgrade_payment_form&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20171520319&quot;, &quot;key&quot;: &quot;submit.create_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20222645674&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.discuss_your_needs&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20227443657&quot;, &quot;key&quot;: &quot;submit.verify_primary_user_email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20234607160&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.try_enterprise&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20238175784&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20239847212&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.continue_free&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20251097193&quot;, &quot;key&quot;: &quot;recommended_plan&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20438619534&quot;, &quot;key&quot;: &quot;click.pricing_calculator.1_member&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20456699683&quot;, &quot;key&quot;: &quot;click.pricing_calculator.15_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20467868331&quot;, &quot;key&quot;: &quot;click.pricing_calculator.10_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20476267432&quot;, &quot;key&quot;: &quot;click.trial_days_remaining&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20476357660&quot;, &quot;key&quot;: &quot;click.discover_feature&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20479287901&quot;, &quot;key&quot;: &quot;click.pricing_calculator.custom_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20481107083&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.apply_teacher_benefits&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20483089392&quot;, &quot;key&quot;: &quot;click.pricing_calculator.5_members&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20484283944&quot;, &quot;key&quot;: &quot;click.onboarding_task&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20484996281&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.apply_student_benefits&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20486713726&quot;, &quot;key&quot;: &quot;click.onboarding_task_breadcrumb&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20490791319&quot;, &quot;key&quot;: &quot;click.upgrade_to_enterprise&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20491786766&quot;, &quot;key&quot;: &quot;click.talk_to_us&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20494144087&quot;, &quot;key&quot;: &quot;click.dismiss_enterprise_trial&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20499722759&quot;, &quot;key&quot;: &quot;completed_all_tasks&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20500710104&quot;, &quot;key&quot;: &quot;completed_onboarding_tasks&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20513160672&quot;, &quot;key&quot;: &quot;click.read_doc&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20516196762&quot;, &quot;key&quot;: &quot;actions_enabled&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20518980986&quot;, &quot;key&quot;: &quot;click.dismiss_trial_banner&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20535446721&quot;, &quot;key&quot;: &quot;click.issue_actions_prompt.dismiss_prompt&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20557002247&quot;, &quot;key&quot;: &quot;click.issue_actions_prompt.setup_workflow&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20595070227&quot;, &quot;key&quot;: &quot;click.pull_request_setup_workflow&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20626600314&quot;, &quot;key&quot;: &quot;click.seats_input&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20642310305&quot;, &quot;key&quot;: &quot;click.decrease_seats_number&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20662990045&quot;, &quot;key&quot;: &quot;click.increase_seats_number&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20679620969&quot;, &quot;key&quot;: &quot;click.public_product_roadmap&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20761240940&quot;, &quot;key&quot;: &quot;click.dismiss_survey_banner&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20767210721&quot;, &quot;key&quot;: &quot;click.take_survey&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20795281201&quot;, &quot;key&quot;: &quot;click.archive_list&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20966790249&quot;, &quot;key&quot;: &quot;contact_sales.submit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20996500333&quot;, &quot;key&quot;: &quot;contact_sales.existing_customer&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20996890162&quot;, &quot;key&quot;: &quot;contact_sales.blank_message_field&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21000470317&quot;, &quot;key&quot;: &quot;contact_sales.personal_email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21002790172&quot;, &quot;key&quot;: &quot;contact_sales.blank_phone_field&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21354412592&quot;, &quot;key&quot;: &quot;click.dismiss_create_readme&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21366102546&quot;, &quot;key&quot;: &quot;click.dismiss_zero_user_content&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21370252505&quot;, &quot;key&quot;: &quot;account_did_downgrade&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21370840408&quot;, &quot;key&quot;: &quot;click.cta_create_readme&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21375451068&quot;, &quot;key&quot;: &quot;click.cta_create_new_repository&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21385390948&quot;, &quot;key&quot;: &quot;click.zero_user_content&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21467712175&quot;, &quot;key&quot;: &quot;click.downgrade_keep&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21484112202&quot;, &quot;key&quot;: &quot;click.downgrade&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21495292213&quot;, &quot;key&quot;: &quot;click.downgrade_survey_exit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21508241468&quot;, &quot;key&quot;: &quot;click.downgrade_survey_submit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21512030356&quot;, &quot;key&quot;: &quot;click.downgrade_support&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21539090022&quot;, &quot;key&quot;: &quot;click.downgrade_exit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21543640644&quot;, &quot;key&quot;: &quot;click_fetch_upstream&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21646510300&quot;, &quot;key&quot;: &quot;click.move_your_work&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21656151116&quot;, &quot;key&quot;: &quot;click.add_branch_protection_rule&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21663860599&quot;, &quot;key&quot;: &quot;click.downgrade_dialog_open&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21687860483&quot;, &quot;key&quot;: &quot;click.learn_about_protected_branches&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21689050333&quot;, &quot;key&quot;: &quot;click.dismiss_protect_this_branch&quot;}], &quot;revision&quot;: &quot;1288&quot;}" />
+
+
+  <script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-IlxrV3fPKJ3H3zXBv8ZNFZA1UGUCcfs482E0tD9JDa8vDF+oIfVC8m5/T6o5u/ccu/oNvjJNPZhYkX02Ao349g==" src="https://github.githubassets.com/assets/runtime-225c6b5777cf.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-XKV1seh7UvAY3d4/PUvi5uHqcCJ9ThVg73aqJM9r8QFid4KRpTHbebXZdhznn39F6fvGFcgvAhTdTrUp/oWL1A==" src="https://github.githubassets.com/assets/environment-5ca575b1e87b.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-io+1MvgXPXTw8Kp4eOdNMJl8uGASuw8VfTY5VeIFETaAknimWi8GoxggMEeQ6mq0de4Dest4iIJ/9gUbCo0hgw==" src="https://github.githubassets.com/assets/vendors-node_modules_selector-observer_dist_index_esm_js-8a8fb532f817.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-Es25N4GyPa8Yfp5wpahoe5b2fyPtkRMyR6mKIXyCJC0ocqQazeWvxhGZhx3StRxOfqDfHDR5SS35u/R3Wux6Cg==" src="https://github.githubassets.com/assets/vendors-node_modules_delegated-events_dist_index_js-node_modules_github_details-dialog-elemen-63debe-12cdb93781b2.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-lmiecOIgf+hakg5oKMNM7grVhEDyPoIrT39Px448JJH5PSAaK21PH0Twgyz5O5oi8+dnlLr3Jt8bBCtAcpNdRw==" src="https://github.githubassets.com/assets/vendors-node_modules_github_filter-input-element_dist_index_js-node_modules_github_remote-inp-c7e9ed-96689e70e220.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-y67eNkVaNK4RguUGcHOvIbHFlgf1Qje+LDdjVw2eFuuvBOqta2GePz/CwoLIR/PJhhRAj5RPGxCWoomnimSw6w==" src="https://github.githubassets.com/assets/vendors-node_modules_github_catalyst_lib_index_js-node_modules_github_time-elements_dist_index_js-cbaede36455a.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-DE9GK/TS3Hf/dJfP1GMiyjvsGRnpR4A1eGH64s0Ot7Vy7Tj7/aD1d5EOMWHFWQXlXy/ZSwu5MHInG6RZN31bMA==" src="https://github.githubassets.com/assets/vendors-node_modules_github_file-attachment-element_dist_index_js-node_modules_primer_view-co-52e104-0c4f462bf4d2.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-l//8hwOiwPAmBg0NwsWUYovdQ7/r9kPHiD9/LL4fD3M7El8gbgaOYU5+o6cYLB6puSfOTxyN9M6fE38HSDr2Bw==" src="https://github.githubassets.com/assets/github-elements-97fffc8703a2.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-lIm60Y77qCoue/4Srnoe0QDZN/ReE8mrMZE2uyNTZteQP0e5kPJGSn4NtlWfYkVvI1DuyDC76KsOz04K76v+iA==" src="https://github.githubassets.com/assets/element-registry-9489bad18efb.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-uo73yUZcm4EicwjSbfxFZcKfjWniOxhLBp+q1n7IFRfutFM6/lzbQMgD0Xrxp7QD1HzqdvrV8UclPhi3mEOyzQ==" src="https://github.githubassets.com/assets/vendors-node_modules_lit-html_lit-html_js-ba8ef7c9465c.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-5XdknqL9iTuaAhbOnKJDHf+5bJReOYDwlz5p2mW+SAVfab4RN6AfIliqtPwheKld40DrhPs47We/mT5nlZq2Sg==" src="https://github.githubassets.com/assets/vendors-node_modules_github_remote-form_dist_index_js-node_modules_github_catalyst_lib_index_-87b1b3-e577649ea2fd.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-0r1nf/rfPz54kyePp4f63bcPxkFo7wyaUZJD/SwIVDK3q0WzurAK9ydOm88tzKtPJm8xWI0Vo25NyCfecwxJ9g==" src="https://github.githubassets.com/assets/vendors-node_modules_github_mini-throttle_dist_index_js-node_modules_github_hotkey_dist_index-9f48bd-d2bd677ffadf.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-VK56d0N1hPZ20mOzPoy84zlTGCjGbKKmVvfjoyDqSF+VxTD4f6X8QDs2RgG1R1cdBmsCiea+ZxP6ukV3tHlD+Q==" src="https://github.githubassets.com/assets/vendors-node_modules_github_paste-markdown_dist_index_esm_js-node_modules_github_quote-select-df2537-54ae7a774375.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-vneJnOQi1L9PJESlqC6veCutbVwyQDQCzMoPlkEhfWJ592tNYva9p1SV5uTde09VMWh17HFMDbfuVoy0I5tqHA==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_pjax_ts-be77899ce422.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-2hsYB2KHyayxUJJxJGhXeTeTZZG2c7lzRtO0uB1txmpc7rfvIt4mf0iossT0MHIHknYlaslgi98jmmlVXcXaZQ==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_keyboard-shortcuts-helper_ts-app_assets_modules_github_be-af52ef-da1b18076287.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-8eyyuvsdeUUBrkTsu7OePie4HdGA6yQww2w70dhs28WU0cMh5qhDM1vHrLOwJdbFQodch4KNQ1rv6KNyw1XMpA==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_details_ts-app_assets_modules_github_behaviors_include-fr-a5a4c7-f1ecb2bafb1d.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-i0EzX4ojOyIXe3MB4Kc/20cXNNTwDWHPd9i/bwOEDOd3bRNP7xJCQkX4f7DIELn+WFHtJSpEEfPRzi64gEENFw==" src="https://github.githubassets.com/assets/behaviors-8b41335f8a23.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-rLnbCtiKIpRtm1jgAz913tDImW3g8G+FLRMBIMwrkZ33pO8PMzWDG4m8Zl0uIIs10hCyFr6Dc8s6VDC200S9lg==" src="https://github.githubassets.com/assets/vendors-node_modules_delegated-events_dist_index_js-node_modules_github_catalyst_lib_index_js-01c6c51-acb9db0ad88a.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-zJONQW0ZfaXs3mk/bws2nP0XemlIUbA4+vWsLqTzjTTz7X2yCAI0VXa+i8y/vgLF3Ooeyus5ZnyhiKYj3Jwz6w==" src="https://github.githubassets.com/assets/notifications-global-cc938d416d19.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-1b+YhvQsW1BdKi3eWSc7oNSGJcrQalc3oxxMoXI8foh3weo2AOLx86EMZ5J1a541yu2y7Axj7Vntn1T66EW8dA==" src="https://github.githubassets.com/assets/vendors-node_modules_optimizely_optimizely-sdk_dist_optimizely_browser_es_min_js-node_modules-77839b-d5bf9886f42c.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-NH4Gv218MnzYs/oudrFw7n14lPMa75qmHypJ9MbJn1RW03KwjJKCegU17usZk1uQD8cLWseA4vhCL638+tKM+A==" src="https://github.githubassets.com/assets/optimizely-347e06bf6d7c.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-yG3t7Lpt5yKkdMNXFpl75Xemm+g4ghZ5e8xRBwl/HMbmgc3htVvxrjbohbK9MMfnnci1SnogZNH9DhJHYdQf3A==" src="https://github.githubassets.com/assets/vendors-node_modules_virtualized-list_es_index_js-node_modules_github_template-parts_lib_index_js-c86dedecba6d.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-eE/fdOH+uhuRT992TtBa/oI33EN3x7EBNkJnTdiozp5RMDDiteODOWGCazeuulOPF5ZSFPR7r0PBZFn8enpsRQ==" src="https://github.githubassets.com/assets/vendors-node_modules_github_mini-throttle_dist_decorators_js-node_modules_github_remote-form_-2c55be-784fdf74e1fe.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-FNWxKHkUqs6+xuyyZpMx9BmvcxQqIgVJ5A+PmdU82lZO41cFlo6y6ddHkRfWh2Zd6j9ozvndvrTaOuEH58VvEw==" src="https://github.githubassets.com/assets/vendors-node_modules_github_file-attachment-element_dist_index_js-node_modules_github_filter--fd67a9-14d5b1287914.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-dAabCMULUfibZDcNGDpWbINNJawdJeCfhl8ZTGFEYvgpblXXQIBlutl5zv0oOqc0p1ginRnWy4mAZES/swKmZA==" src="https://github.githubassets.com/assets/app_assets_modules_github_fetch_ts-app_assets_modules_github_input-demux_ts-app_assets_module-e5b55c-74069b08c50b.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-SXRchiPOYprh1D0HcC85QDp+f15IP1oshRaSxymGKZ5jw+se3OVL6zw5QYPnxl9cOf8WxcyBuW+t+raq61yg5g==" src="https://github.githubassets.com/assets/repositories-49745c8623ce.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-2IO4DDKs14Xk+vMRzfe3NVcyY5Bsb071hi/LJEZcTTo74dZ9sgelYQaCPu0nh3qsny1AvzutHZyuvBmt0fbguA==" src="https://github.githubassets.com/assets/app_assets_modules_github_diffs_blob-lines_ts-app_assets_modules_github_diffs_linkable-line-n-f314c3-d883b80c32ac.js"></script>
+<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-9b6aISppMLr5F8Imzz33ChOeks/lElt5UEQ1fHaKruZEmJ/Jf2bZ8A96OL0fPGqhReDoWK5CZec/Btp1ny7+oQ==" src="https://github.githubassets.com/assets/diffs-f5be9a212a69.js"></script>
+
+
+  <title>PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR · GitHub</title>
+
+
+
+
+
+  <meta name="request-id" content="B6FC:313F:2FB16:4AF11:62B1726D" data-pjax-transient="true"/><meta name="html-safe-nonce" content="95c166e7500cd689d260f3b8707ad62a8e0f0dec50b5a900780ad09a4cb58361" data-pjax-transient="true"/><meta name="visitor-payload" content="eyJyZWZlcnJlciI6IiIsInJlcXVlc3RfaWQiOiJCNkZDOjMxM0Y6MkZCMTY6NEFGMTE6NjJCMTcyNkQiLCJ2aXNpdG9yX2lkIjoiNDc4ODQxMzQ5MTgzODAyMjI1NCIsInJlZ2lvbl9lZGdlIjoic2VhIiwicmVnaW9uX3JlbmRlciI6InNlYSJ9" data-pjax-transient="true"/><meta name="visitor-hmac" content="60769063a92f53a010f7783fa7370a5d4b255dce3cfb2ae919babf28c57cc46d" data-pjax-transient="true"/>
+
+    <meta name="hovercard-subject-tag" content="repository:262296122" data-pjax-transient>
+
+
+  <meta name="github-keyboard-shortcuts" content="repository,source-code" data-pjax-transient="true" />
+
+
+  <meta name="selected-link" value="repo_source" data-pjax-transient>
+
+    <meta name="google-site-verification" content="c1kuD-K2HIVF635lypcsWPoD4kilo5-jA_wBFyT4uMY">
+  <meta name="google-site-verification" content="KT5gs8h0wvaagLKAVWq8bbeNwnZZK1r1XQysX3xurLU">
+  <meta name="google-site-verification" content="ZzhVyEFwb7w3e0-uOTltm8Jsck2F5StVihD0exw2fsA">
+  <meta name="google-site-verification" content="GXs5KoUUkNCoaAZn7wPN-t01Pywp9M3sEjnt_3_ZWPc">
+
+<meta name="octolytics-url" content="https://collector.github.com/github/collect" />
+
+  <meta name="analytics-location" content="/&lt;user-name&gt;/&lt;repo-name&gt;/blob/show" data-pjax-transient="true" />
+
+
+
+
+
+
+
+
+    <meta name="user-login" content="">
+
+    <meta name="viewport" content="width=device-width">
+
+      <meta name="description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices) - PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR">
+      <link rel="search" type="application/opensearchdescription+xml" href="/opensearch.xml" title="GitHub">
+    <link rel="fluid-icon" href="https://github.com/fluidicon.png" title="GitHub">
+    <meta property="fb:app_id" content="1401488693436528">
+    <meta name="apple-itunes-app" content="app-id=1477376905" />
+      <meta name="twitter:image:src" content="https://opengraph.githubassets.com/1eca697559d2aabccdeb858570f3f508b575ece90f37d8d4e2156599d7d26d4f/PaddlePaddle/PaddleOCR" /><meta name="twitter:site" content="@github" /><meta name="twitter:card" content="summary_large_image" /><meta name="twitter:title" content="PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR" /><meta name="twitter:description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." />
+      <meta property="og:image" content="https://opengraph.githubassets.com/1eca697559d2aabccdeb858570f3f508b575ece90f37d8d4e2156599d7d26d4f/PaddlePaddle/PaddleOCR" /><meta property="og:image:alt" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." /><meta property="og:image:width" content="1200" /><meta property="og:image:height" content="600" /><meta property="og:site_name" content="GitHub" /><meta property="og:type" content="object" /><meta property="og:title" content="PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR" /><meta property="og:url" content="https://github.com/PaddlePaddle/PaddleOCR" /><meta property="og:description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." />
+
+    <link rel="assets" href="https://github.githubassets.com/">
+
+
+
+        <meta name="hostname" content="github.com">
+
+
+
+        <meta name="expected-hostname" content="github.com">
+
+
+    <meta name="enabled-features" content="IMAGE_METRIC_TRACKING">
+
+
+  <meta http-equiv="x-pjax-version" content="6c58939d21332326d08c0f7a7a642997da526ff183c0d52243bccca116e52b17" data-turbo-track="reload">
+  <meta http-equiv="x-pjax-csp-version" content="485d6a5ccbb1eeae9c86b616b4870b531f6f458e8bd5c309c40280dc4f51defb" data-turbo-track="reload">
+  <meta http-equiv="x-pjax-css-version" content="4219d8fbdfa0920ce6d03f0d9d444f678ff205e0a832d33151d083532f704774" data-turbo-track="reload">
+  <meta http-equiv="x-pjax-js-version" content="29ca8c5c927eacd73f2fe0e83209b72749f0b338ac4cfd01387c31f0f4001844" data-turbo-track="reload">
+
+  <meta name="turbo-cache-control" content="no-preview"></meta>
+
+
+  <meta name="go-import" content="github.com/PaddlePaddle/PaddleOCR git https://github.com/PaddlePaddle/PaddleOCR.git">
+
+  <meta name="octolytics-dimension-user_id" content="23534030" /><meta name="octolytics-dimension-user_login" content="PaddlePaddle" /><meta name="octolytics-dimension-repository_id" content="262296122" /><meta name="octolytics-dimension-repository_nwo" content="PaddlePaddle/PaddleOCR" /><meta name="octolytics-dimension-repository_public" content="true" /><meta name="octolytics-dimension-repository_is_fork" content="false" /><meta name="octolytics-dimension-repository_network_root_id" content="262296122" /><meta name="octolytics-dimension-repository_network_root_nwo" content="PaddlePaddle/PaddleOCR" />
+
+
+
+    <link rel="canonical" href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-pjax-transient>
+
+
+  <meta name="browser-stats-url" content="https://api.github.com/_private/browser/stats">
+
+  <meta name="browser-errors-url" content="https://api.github.com/_private/browser/errors">
+
+  <meta name="browser-optimizely-client-errors-url" content="https://api.github.com/_private/browser/optimizely_client/errors">
+
+  <link rel="mask-icon" href="https://github.githubassets.com/pinned-octocat.svg" color="#000000">
+  <link rel="alternate icon" class="js-site-favicon" type="image/png" href="https://github.githubassets.com/favicons/favicon.png">
+  <link rel="icon" class="js-site-favicon" type="image/svg+xml" href="https://github.githubassets.com/favicons/favicon.svg">
+
+<meta name="theme-color" content="#1e2327">
+<meta name="color-scheme" content="light dark" />
+
+
+  <link rel="manifest" href="/manifest.json" crossOrigin="use-credentials">
+
+  </head>
+
+  <body class="logged-out env-production page-responsive page-blob" style="word-wrap: break-word;">
+
+
+    <div class="position-relative js-header-wrapper ">
+      <a href="#start-of-content" class="px-2 py-4 color-bg-accent-emphasis color-fg-on-emphasis show-on-focus js-skip-to-content">Skip to content</a>
+      <span data-view-component="true" class="progress-pjax-loader js-pjax-loader-bar Progress position-fixed width-full">
+    <span style="width: 0%;" data-view-component="true" class="Progress-item progress-pjax-loader-bar left-0 top-0 color-bg-accent-emphasis"></span>
+</span>  
+
+
+
+
+
+            <script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-Z1hiPkiRLA+tojAnxZLHuRa1IrpZhnqK66AX75QbEn6PgZIpDJnWcJhQ0pOU2cI6Pe9yO0i3s4JIc9G//vkosQ==" src="https://github.githubassets.com/assets/sessions-6758623e4891.js"></script>
+<header class="Header-old header-logged-out js-details-container Details position-relative f4 py-2" role="banner">
+  <div class="container-xl d-lg-flex flex-items-center p-responsive">
+    <div class="d-flex flex-justify-between flex-items-center">
+      <a class="mr-4 color-fg-inherit" href="https://github.com/" aria-label="Homepage" data-ga-click="(Logged out) Header, go to homepage, icon:logo-wordmark">
+        <svg height="32" aria-hidden="true" viewBox="0 0 16 16" version="1.1" width="32" data-view-component="true" class="octicon octicon-mark-github">
+    <path fill-rule="evenodd" d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"></path>
+</svg>
+      </a>
+
+        <div class="d-lg-none css-truncate css-truncate-target width-fit p-2">
+
+
+        </div>
+
+      <div class="d-flex flex-items-center">
+            <a href="/signup?ref_cta=Sign+up&amp;ref_loc=header+logged+out&amp;ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E%2Fblob%2Fshow&amp;source=header-repo"
+              class="d-inline-block d-lg-none f5 no-underline border color-border-default rounded-2 px-2 py-1 mr-3 mr-sm-5 color-fg-inherit"
+              data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="e697b45b9e440c22e4e3985c3ab81484e1433b0710fb937a52cb9aad23cfc3ca"
+            >
+              Sign&nbsp;up
+            </a>
+
+        <button aria-label="Toggle navigation" aria-expanded="false" type="button" data-view-component="true" class="js-details-target btn-link d-lg-none mt-1 color-fg-inherit">  <svg aria-hidden="true" height="24" viewBox="0 0 16 16" version="1.1" width="24" data-view-component="true" class="octicon octicon-three-bars">
+    <path fill-rule="evenodd" d="M1 2.75A.75.75 0 011.75 2h12.5a.75.75 0 110 1.5H1.75A.75.75 0 011 2.75zm0 5A.75.75 0 011.75 7h12.5a.75.75 0 110 1.5H1.75A.75.75 0 011 7.75zM1.75 12a.75.75 0 100 1.5h12.5a.75.75 0 100-1.5H1.75z"></path>
+</svg>
+
+</button>      </div>
+    </div>
+
+    <div class="HeaderMenu HeaderMenu--logged-out position-fixed top-0 right-0 bottom-0 height-fit position-lg-relative d-lg-flex flex-justify-between flex-items-center flex-auto">
+      <div class="d-flex d-lg-none flex-justify-end border-bottom color-bg-subtle p-3">
+        <button aria-label="Toggle navigation" aria-expanded="false" type="button" data-view-component="true" class="js-details-target btn-link">  <svg aria-hidden="true" height="24" viewBox="0 0 24 24" version="1.1" width="24" data-view-component="true" class="octicon octicon-x color-fg-muted">
+    <path fill-rule="evenodd" d="M5.72 5.72a.75.75 0 011.06 0L12 10.94l5.22-5.22a.75.75 0 111.06 1.06L13.06 12l5.22 5.22a.75.75 0 11-1.06 1.06L12 13.06l-5.22 5.22a.75.75 0 01-1.06-1.06L10.94 12 5.72 6.78a.75.75 0 010-1.06z"></path>
+</svg>
+
+</button>      </div>
+
+        <nav class="mt-0 px-3 px-lg-0 mb-5 mb-lg-0" aria-label="Global">
+          <ul class="d-lg-flex list-style-none">
+              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
+    <details class="HeaderMenu-details details-overlay details-reset width-full">
+      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
+        Product
+        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
+      </summary>
+      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
+        <ul class="list-style-none f5 pb-1">
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Features&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Features;&quot;}" href="/features">
+      Features
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Mobile&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Mobile;&quot;}" href="/mobile">
+      Mobile
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Actions&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Actions;&quot;}" href="/features/actions">
+      Actions
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Codespaces&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Codespaces;&quot;}" href="/features/codespaces">
+      Codespaces
+</a>  </li>
+
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Packages&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Packages;&quot;}" href="/features/packages">
+      Packages
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Security&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Security;&quot;}" href="/features/security">
+      Security
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Code review&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Code review;&quot;}" href="/features/code-review">
+      Code review
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Issues&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Issues;&quot;}" href="/features/issues">
+      Issues
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Integrations&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Integrations;&quot;}" href="/features/integrations">
+      Integrations
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold border-top pt-4 pb-2 mt-3" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to GitHub Sponsors&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Sponsors;&quot;}" href="/sponsors">
+      GitHub Sponsors
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Customer stories&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Customer stories;&quot;}" href="/customer-stories">
+      Customer stories
+</a>  </li>
+
+        </ul>
+      </div>
+    </details>
+</li>
+
+
+              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
+    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Team&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Team;&quot;}" href="/team">Team</a>
+</li>
+
+              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
+    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Enterprise&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Enterprise;&quot;}" href="/enterprise">Enterprise</a>
+</li>
+
+
+            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
+    <details class="HeaderMenu-details details-overlay details-reset width-full">
+      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
+        Explore
+        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
+      </summary>
+      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
+        <ul class="list-style-none f5 pb-1">
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Explore GitHub&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Explore GitHub;&quot;}" href="/explore">
+      Explore GitHub
+</a>  </li>
+
+              <li class="color-fg-muted text-normal f6 text-mono mb-1 border-top pt-3 mt-3 mb-1">Learn and contribute</li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Topics&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Topics;&quot;}" href="/topics">
+      Topics
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Collections&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Collections;&quot;}" href="/collections">
+      Collections
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Trending&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Trending;&quot;}" href="/trending">
+      Trending
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Skills&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Skills;&quot;}" href="https://skills.github.com/">
+      Skills
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Sponsors&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Sponsors;&quot;}" href="/sponsors/explore">
+      GitHub Sponsors
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Open source guides&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Open source guides;&quot;}" href="https://opensource.guide">
+      Open source guides
+</a>  </li>
+
+              <li class="color-fg-muted text-normal f6 text-mono mb-1 border-top pt-3 mt-3 mb-1">Connect with others</li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to The ReadME Project&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:The ReadME Project;&quot;}" href="/readme">
+      The ReadME Project
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Events&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Events;&quot;}" href="/events">
+      Events
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Community forum&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Community forum;&quot;}" href="https://github.community">
+      Community forum
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Education&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Education;&quot;}" href="https://education.github.com">
+      GitHub Education
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Stars program&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Stars program;&quot;}" href="https://stars.github.com">
+      GitHub Stars program
+</a>  </li>
+
+        </ul>
+      </div>
+    </details>
+</li>
+
+
+            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
+    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Marketplace&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Marketplace;&quot;}" href="/marketplace">Marketplace</a>
+</li>
+
+
+            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
+    <details class="HeaderMenu-details details-overlay details-reset width-full">
+      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
+        Pricing
+        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
+      </summary>
+      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
+        <ul class="list-style-none f5 pb-1">
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Plans&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Plans;&quot;}" href="/pricing">
+      Plans
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Compare plans&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Compare plans;&quot;}" href="/pricing#compare-features">
+      Compare plans
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Contact Sales&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Contact Sales;&quot;}" href="https://github.com/enterprise/contact">
+      Contact Sales
+</a>  </li>
+
+              <li>
+    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold border-top pt-4 pb-2 mt-3" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Education&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Education;&quot;}" href="https://education.github.com">
+      Education
+</a>  </li>
+
+        </ul>
+      </div>
+    </details>
+</li>
+
+          </ul>
+        </nav>
+
+      <div class="d-lg-flex flex-items-center px-3 px-lg-0 text-center text-lg-left">
+          <div class="d-lg-flex min-width-0 mb-3 mb-lg-0">
+
+
+
+
+<div class="header-search flex-auto js-site-search position-relative flex-self-stretch flex-md-self-auto mb-3 mb-md-0 mr-0 mr-md-3 scoped-search site-scoped-search js-jump-to"
+>
+  <div class="position-relative">
+    <!-- '"` --><!-- </textarea></xmp> --></option></form><form class="js-site-search-form" role="search" aria-label="Site" data-scope-type="Repository" data-scope-id="262296122" data-scoped-search-url="/PaddlePaddle/PaddleOCR/search" data-owner-scoped-search-url="/orgs/PaddlePaddle/search" data-unscoped-search-url="/search" data-turbo="false" action="/PaddlePaddle/PaddleOCR/search" accept-charset="UTF-8" method="get">
+      <label class="form-control input-sm header-search-wrapper p-0 js-chromeless-input-container header-search-wrapper-jump-to position-relative d-flex flex-justify-between flex-items-center">
+        <input type="text"
+          class="form-control input-sm header-search-input jump-to-field js-jump-to-field js-site-search-focus js-site-search-field is-clearable"
+          data-hotkey=s,/
+          name="q"
+          data-test-selector="nav-search-input"
+          placeholder="Search"
+          data-unscoped-placeholder="Search GitHub"
+          data-scoped-placeholder="Search"
+          autocapitalize="off"
+          role="combobox"
+          aria-haspopup="listbox"
+          aria-expanded="false"
+          aria-autocomplete="list"
+          aria-controls="jump-to-results"
+          aria-label="Search"
+          data-jump-to-suggestions-path="/_graphql/GetSuggestedNavigationDestinations"
+          spellcheck="false"
+          autocomplete="off"
+        >
+        <input type="hidden" data-csrf="true" class="js-data-jump-to-suggestions-path-csrf" value="3sIzmZY19XiVvvmqOda8K/tw/otB4voFktXu/QVkH/wlFcIW1DALJozQM+9OJnW6NJEvC+nr88icYnU0YgDkJA==" />
+        <input type="hidden" class="js-site-search-type-field" name="type" >
+            <svg xmlns="http://www.w3.org/2000/svg" width="22" height="20" aria-hidden="true" class="mr-1 header-search-key-slash"><path fill="none" stroke="#979A9C" opacity=".4" d="M3.5.5h12c1.7 0 3 1.3 3 3v13c0 1.7-1.3 3-3 3h-12c-1.7 0-3-1.3-3-3v-13c0-1.7 1.3-3 3-3z"></path><path fill="#979A9C" d="M11.8 6L8 15.1h-.9L10.8 6h1z"></path></svg>
+
+
+          <div class="Box position-absolute overflow-hidden d-none jump-to-suggestions js-jump-to-suggestions-container">
+
+<ul class="d-none js-jump-to-suggestions-template-container">
+
+
+<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-suggestion" role="option">
+  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="suggestion">
+    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
+      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
+</svg>
+      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
+</svg>
+      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
+</svg>
+    </div>
+
+    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
+
+    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
+    </div>
+
+    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
+      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
+        In this repository
+      </span>
+      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
+        All GitHub
+      </span>
+      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+
+    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
+      Jump to
+      <span class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+  </a>
+</li>
+
+</ul>
+
+<ul class="d-none js-jump-to-no-results-template-container">
+  <li class="d-flex flex-justify-center flex-items-center f5 d-none js-jump-to-suggestion p-2">
+    <span class="color-fg-muted">No suggested jump to results</span>
+  </li>
+</ul>
+
+<ul id="jump-to-results" role="listbox" class="p-0 m-0 js-navigation-container jump-to-suggestions-results-container js-jump-to-suggestions-results-container">
+
+
+<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-scoped-search d-none" role="option">
+  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="scoped_search">
+    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
+      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
+</svg>
+      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
+</svg>
+      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
+</svg>
+    </div>
+
+    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
+
+    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
+    </div>
+
+    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
+      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
+        In this repository
+      </span>
+      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
+        All GitHub
+      </span>
+      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+
+    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
+      Jump to
+      <span class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+  </a>
+</li>
+
+
+
+<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-owner-scoped-search d-none" role="option">
+  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="owner_scoped_search">
+    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
+      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
+</svg>
+      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
+</svg>
+      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
+</svg>
+    </div>
+
+    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
+
+    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
+    </div>
+
+    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
+      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this organization">
+        In this organization
+      </span>
+      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
+        All GitHub
+      </span>
+      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+
+    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
+      Jump to
+      <span class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+  </a>
+</li>
+
+
+
+<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-global-search d-none" role="option">
+  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="global_search">
+    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
+      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
+</svg>
+      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
+</svg>
+      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
+    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
+</svg>
+    </div>
+
+    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
+
+    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
+    </div>
+
+    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
+      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
+        In this repository
+      </span>
+      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
+        All GitHub
+      </span>
+      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+
+    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
+      Jump to
+      <span class="d-inline-block ml-1 v-align-middle">↵</span>
+    </div>
+  </a>
+</li>
+
+
+</ul>
+
+          </div>
+      </label>
+</form>  </div>
+</div>
+
+          </div>
+
+        <div class="position-relative mr-3 mb-4 mb-lg-0 d-inline-block">
+          <a href="/login?return_to=https%3A%2F%2Fgithub.com%2FPaddlePaddle%2FPaddleOCR%2Fblob%2Fdygraph%2Fapplications%2F%25E8%25BD%25BB%25E9%2587%258F%25E7%25BA%25A7%25E8%25BD%25A6%25E7%2589%258C%25E8%25AF%2586%25E5%2588%25AB.md"
+            class="HeaderMenu-link flex-shrink-0 no-underline"
+            data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header menu&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="0de5d29e004c06e7301cb66a02d668f7cd98725fd6fa97ab9b512671e1a9905b"
+            data-ga-click="(Logged out) Header, clicked Sign in, text:sign-in">
+            Sign in
+          </a>
+        </div>
+
+          <a href="/signup?ref_cta=Sign+up&amp;ref_loc=header+logged+out&amp;ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E%2Fblob%2Fshow&amp;source=header-repo&amp;source_repo=PaddlePaddle%2FPaddleOCR"
+            class="HeaderMenu-link flex-shrink-0 d-inline-block no-underline border color-border-default rounded px-2 py-1"
+            data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header menu&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="0de5d29e004c06e7301cb66a02d668f7cd98725fd6fa97ab9b512671e1a9905b"
+            data-analytics-event="{&quot;category&quot;:&quot;Sign up&quot;,&quot;action&quot;:&quot;click to sign up for account&quot;,&quot;label&quot;:&quot;ref_page:/&lt;user-name&gt;/&lt;repo-name&gt;/blob/show;ref_cta:Sign up;ref_loc:header logged out&quot;}"
+          >
+            Sign up
+          </a>
+      </div>
+    </div>
+  </div>
+</header>
+
+    </div>
+
+  <div id="start-of-content" class="show-on-focus"></div>
+
+
+
+
+
+
+
+    <div data-pjax-replace id="js-flash-container" data-turbo-replace>
+
+
+  <template class="js-flash-template">
+    <div class="flash flash-full  {{ className }}">
+  <div class="px-2" >
+    <button class="flash-close js-flash-close" type="button" aria-label="Dismiss this message">
+      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
+    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
+</svg>
+    </button>
+
+      <div>{{ message }}</div>
+
+  </div>
+</div>
+  </template>
+</div>
+
+
+
+  <include-fragment class="js-notification-shelf-include-fragment" data-base-src="https://github.com/notifications/beta/shelf"></include-fragment>
+
+
+
+
+
+  <div
+    class="application-main "
+    data-commit-hovercards-enabled
+    data-discussion-hovercards-enabled
+    data-issue-and-pr-hovercards-enabled
+  >
+        <div itemscope itemtype="http://schema.org/SoftwareSourceCode" class="">
+    <main id="js-repo-pjax-container" data-pjax-container >
+
+
+
+
+
+
+
+
+
+
+
+
+
+  <div id="repository-container-header" class="pt-3 hide-full-screen" style="background-color: var(--color-page-header-bg);" data-pjax-replace data-turbo-replace>
+
+      <div class="d-flex mb-3 px-3 px-md-4 px-lg-5">
+
+        <div class="flex-auto min-width-0 width-fit mr-3">
+            <div class=" d-flex flex-wrap flex-items-center wb-break-word f3 text-normal">
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo color-fg-muted mr-2">
+    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
+</svg>
+  <span class="author flex-self-stretch" itemprop="author">
+    <a class="url fn" rel="author" data-hovercard-type="organization" data-hovercard-url="/orgs/PaddlePaddle/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle">PaddlePaddle</a>
+  </span>
+  <span class="mx-1 flex-self-stretch color-fg-muted">/</span>
+  <strong itemprop="name" class="mr-2 flex-self-stretch">
+    <a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR">PaddleOCR</a>
+  </strong>
+
+  <span></span><span class="Label Label--secondary v-align-middle mr-1">Public</span>
+</div>
+
+        </div>
+
+          <ul class="pagehead-actions flex-shrink-0 d-none d-md-inline" style="padding: 2px 0;">
+
+
+
+  <li>
+      <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;notification subscription menu watch&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="2c00da16a0d8d066b49f1f6e26149784faffba39c427495160ebf3bac257b4a9" aria-label="You must be signed in to change notification settings" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-bell mr-2">
+    <path d="M8 16a2 2 0 001.985-1.75c.017-.137-.097-.25-.235-.25h-3.5c-.138 0-.252.113-.235.25A2 2 0 008 16z"></path><path fill-rule="evenodd" d="M8 1.5A3.5 3.5 0 004.5 5v2.947c0 .346-.102.683-.294.97l-1.703 2.556a.018.018 0 00-.003.01l.001.006c0 .002.002.004.004.006a.017.017 0 00.006.004l.007.001h10.964l.007-.001a.016.016 0 00.006-.004.016.016 0 00.004-.006l.001-.007a.017.017 0 00-.003-.01l-1.703-2.554a1.75 1.75 0 01-.294-.97V5A3.5 3.5 0 008 1.5zM3 5a5 5 0 0110 0v2.947c0 .05.015.098.042.139l1.703 2.555A1.518 1.518 0 0113.482 13H2.518a1.518 1.518 0 01-1.263-2.36l1.703-2.554A.25.25 0 003 7.947V5z"></path>
+</svg>Notifications
+
+</a>
+  </li>
+
+  <li>
+        <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;repo details fork button&quot;,&quot;repository_id&quot;:262296122,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f55097530c92c084f3a79350400121a7c267c872da2b69858747a12db51b5c71" aria-label="You must be signed in to fork a repository" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo-forked mr-2">
+    <path fill-rule="evenodd" d="M5 3.25a.75.75 0 11-1.5 0 .75.75 0 011.5 0zm0 2.122a2.25 2.25 0 10-1.5 0v.878A2.25 2.25 0 005.75 8.5h1.5v2.128a2.251 2.251 0 101.5 0V8.5h1.5a2.25 2.25 0 002.25-2.25v-.878a2.25 2.25 0 10-1.5 0v.878a.75.75 0 01-.75.75h-4.5A.75.75 0 015 6.25v-.878zm3.75 7.378a.75.75 0 11-1.5 0 .75.75 0 011.5 0zm3-8.75a.75.75 0 100-1.5.75.75 0 000 1.5z"></path>
+</svg>Fork
+    <span id="repo-network-counter" data-pjax-replace="true" title="4,668" data-view-component="true" class="Counter">4.7k</span>
+
+</a>
+  </li>
+
+  <li>
+        <div data-view-component="true" class="BtnGroup d-flex">
+      <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;star button&quot;,&quot;repository_id&quot;:262296122,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="93ad0192b3e15b4610691426f776e3dab114d4c7666db7f44c0d706f50c74c23" aria-label="You must be signed in to star a repository" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-star v-align-text-bottom d-inline-block mr-2">
+    <path fill-rule="evenodd" d="M8 .25a.75.75 0 01.673.418l1.882 3.815 4.21.612a.75.75 0 01.416 1.279l-3.046 2.97.719 4.192a.75.75 0 01-1.088.791L8 12.347l-3.766 1.98a.75.75 0 01-1.088-.79l.72-4.194L.818 6.374a.75.75 0 01.416-1.28l4.21-.611L7.327.668A.75.75 0 018 .25zm0 2.445L6.615 5.5a.75.75 0 01-.564.41l-3.097.45 2.24 2.184a.75.75 0 01.216.664l-.528 3.084 2.769-1.456a.75.75 0 01.698 0l2.77 1.456-.53-3.084a.75.75 0 01.216-.664l2.24-2.183-3.096-.45a.75.75 0 01-.564-.41L8 2.694v.001z"></path>
+</svg><span data-view-component="true" class="d-inline">
+          Star
+</span>          <span id="repo-stars-counter-star" aria-label="22861 users starred this repository" data-singular-suffix="user starred this repository" data-plural-suffix="users starred this repository" data-pjax-replace="true" data-turbo-replace="true" title="22,861" data-view-component="true" class="Counter js-social-count">22.9k</span>
+
+</a>      <button disabled="disabled" aria-label="You must be signed in to add this repository to a list" type="button" data-view-component="true" class="btn-sm btn BtnGroup-item px-2">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-triangle-down">
+    <path d="M4.427 7.427l3.396 3.396a.25.25 0 00.354 0l3.396-3.396A.25.25 0 0011.396 7H4.604a.25.25 0 00-.177.427z"></path>
+</svg>
+
+</button></div>
+  </li>
+
+
+
+</ul>
+
+      </div>
+
+      <div id="responsive-meta-container" data-pjax-replace data-turbo-replace>
+</div>
+
+
+
+<nav data-pjax="#js-repo-pjax-container" aria-label="Repository" data-view-component="true" class="js-repo-nav js-sidenav-container-pjax js-responsive-underlinenav overflow-hidden UnderlineNav px-3 px-md-4 px-lg-5">
+
+  <ul data-view-component="true" class="UnderlineNav-body list-style-none">
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="code-tab" href="/PaddlePaddle/PaddleOCR/tree/dygraph" data-tab-item="i0code-tab" data-selected-links="repo_source repo_downloads repo_commits repo_releases repo_tags repo_branches repo_packages repo_deployments /PaddlePaddle/PaddleOCR/tree/dygraph" data-pjax="#repo-content-pjax-container" data-hotkey="g c" data-ga-click="Repository, Navigation click, Code tab" aria-current="page" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item selected">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-code UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M4.72 3.22a.75.75 0 011.06 1.06L2.06 8l3.72 3.72a.75.75 0 11-1.06 1.06L.47 8.53a.75.75 0 010-1.06l4.25-4.25zm6.56 0a.75.75 0 10-1.06 1.06L13.94 8l-3.72 3.72a.75.75 0 101.06 1.06l4.25-4.25a.75.75 0 000-1.06l-4.25-4.25z"></path>
+</svg>
+          <span data-content="Code">Code</span>
+            <span id="code-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="issues-tab" href="/PaddlePaddle/PaddleOCR/issues" data-tab-item="i1issues-tab" data-selected-links="repo_issues repo_labels repo_milestones /PaddlePaddle/PaddleOCR/issues" data-pjax="#repo-content-pjax-container" data-hotkey="g i" data-ga-click="Repository, Navigation click, Issues tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-issue-opened UnderlineNav-octicon d-none d-sm-inline">
+    <path d="M8 9.5a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path><path fill-rule="evenodd" d="M8 0a8 8 0 100 16A8 8 0 008 0zM1.5 8a6.5 6.5 0 1113 0 6.5 6.5 0 01-13 0z"></path>
+</svg>
+          <span data-content="Issues">Issues</span>
+            <span id="issues-repo-tab-count" data-pjax-replace="" title="1,078" data-view-component="true" class="Counter">1.1k</span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="pull-requests-tab" href="/PaddlePaddle/PaddleOCR/pulls" data-tab-item="i2pull-requests-tab" data-selected-links="repo_pulls checks /PaddlePaddle/PaddleOCR/pulls" data-pjax="#repo-content-pjax-container" data-hotkey="g p" data-ga-click="Repository, Navigation click, Pull requests tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-git-pull-request UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M7.177 3.073L9.573.677A.25.25 0 0110 .854v4.792a.25.25 0 01-.427.177L7.177 3.427a.25.25 0 010-.354zM3.75 2.5a.75.75 0 100 1.5.75.75 0 000-1.5zm-2.25.75a2.25 2.25 0 113 2.122v5.256a2.251 2.251 0 11-1.5 0V5.372A2.25 2.25 0 011.5 3.25zM11 2.5h-1V4h1a1 1 0 011 1v5.628a2.251 2.251 0 101.5 0V5A2.5 2.5 0 0011 2.5zm1 10.25a.75.75 0 111.5 0 .75.75 0 01-1.5 0zM3.75 12a.75.75 0 100 1.5.75.75 0 000-1.5z"></path>
+</svg>
+          <span data-content="Pull requests">Pull requests</span>
+            <span id="pull-requests-repo-tab-count" data-pjax-replace="" title="98" data-view-component="true" class="Counter">98</span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="discussions-tab" href="/PaddlePaddle/PaddleOCR/discussions" data-tab-item="i3discussions-tab" data-selected-links="repo_discussions /PaddlePaddle/PaddleOCR/discussions" data-pjax="#repo-content-pjax-container" data-hotkey="g g" data-ga-click="Repository, Navigation click, Discussions tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-comment-discussion UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M1.5 2.75a.25.25 0 01.25-.25h8.5a.25.25 0 01.25.25v5.5a.25.25 0 01-.25.25h-3.5a.75.75 0 00-.53.22L3.5 11.44V9.25a.75.75 0 00-.75-.75h-1a.25.25 0 01-.25-.25v-5.5zM1.75 1A1.75 1.75 0 000 2.75v5.5C0 9.216.784 10 1.75 10H2v1.543a1.457 1.457 0 002.487 1.03L7.061 10h3.189A1.75 1.75 0 0012 8.25v-5.5A1.75 1.75 0 0010.25 1h-8.5zM14.5 4.75a.25.25 0 00-.25-.25h-.5a.75.75 0 110-1.5h.5c.966 0 1.75.784 1.75 1.75v5.5A1.75 1.75 0 0114.25 12H14v1.543a1.457 1.457 0 01-2.487 1.03L9.22 12.28a.75.75 0 111.06-1.06l2.22 2.22v-2.19a.75.75 0 01.75-.75h1a.25.25 0 00.25-.25v-5.5z"></path>
+</svg>
+          <span data-content="Discussions">Discussions</span>
+            <span id="discussions-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="actions-tab" href="/PaddlePaddle/PaddleOCR/actions" data-tab-item="i4actions-tab" data-selected-links="repo_actions /PaddlePaddle/PaddleOCR/actions" data-pjax="#repo-content-pjax-container" data-hotkey="g a" data-analytics-event="{&quot;category&quot;:&quot;Actions&quot;,&quot;action&quot;:&quot;clicked&quot;,&quot;label&quot;:&quot;ref_cta:Actions;ref_loc:navigation_helper&quot;}" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-play UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M1.5 8a6.5 6.5 0 1113 0 6.5 6.5 0 01-13 0zM8 0a8 8 0 100 16A8 8 0 008 0zM6.379 5.227A.25.25 0 006 5.442v5.117a.25.25 0 00.379.214l4.264-2.559a.25.25 0 000-.428L6.379 5.227z"></path>
+</svg>
+          <span data-content="Actions">Actions</span>
+            <span id="actions-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="projects-tab" href="/PaddlePaddle/PaddleOCR/projects?type=beta" data-tab-item="i5projects-tab" data-selected-links="repo_projects new_repo_project repo_project /PaddlePaddle/PaddleOCR/projects?type=beta" data-pjax="#repo-content-pjax-container" data-hotkey="g b" data-ga-click="Repository, Navigation click, Projects tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-table UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M0 1.75C0 .784.784 0 1.75 0h12.5C15.216 0 16 .784 16 1.75v3.585a.746.746 0 010 .83v8.085A1.75 1.75 0 0114.25 16H6.309a.748.748 0 01-1.118 0H1.75A1.75 1.75 0 010 14.25V6.165a.746.746 0 010-.83V1.75zM1.5 6.5v7.75c0 .138.112.25.25.25H5v-8H1.5zM5 5H1.5V1.75a.25.25 0 01.25-.25H5V5zm1.5 1.5v8h7.75a.25.25 0 00.25-.25V6.5h-8zm8-1.5h-8V1.5h7.75a.25.25 0 01.25.25V5z"></path>
+</svg>
+          <span data-content="Projects">Projects</span>
+            <span id="projects-repo-tab-count" data-pjax-replace="" title="0" hidden="hidden" data-view-component="true" class="Counter">0</span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="wiki-tab" href="/PaddlePaddle/PaddleOCR/wiki" data-tab-item="i6wiki-tab" data-selected-links="repo_wiki /PaddlePaddle/PaddleOCR/wiki" data-pjax="#repo-content-pjax-container" data-hotkey="g w" data-ga-click="Repository, Navigation click, Wikis tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-book UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M0 1.75A.75.75 0 01.75 1h4.253c1.227 0 2.317.59 3 1.501A3.744 3.744 0 0111.006 1h4.245a.75.75 0 01.75.75v10.5a.75.75 0 01-.75.75h-4.507a2.25 2.25 0 00-1.591.659l-.622.621a.75.75 0 01-1.06 0l-.622-.621A2.25 2.25 0 005.258 13H.75a.75.75 0 01-.75-.75V1.75zm8.755 3a2.25 2.25 0 012.25-2.25H14.5v9h-3.757c-.71 0-1.4.201-1.992.572l.004-7.322zm-1.504 7.324l.004-5.073-.002-2.253A2.25 2.25 0 005.003 2.5H1.5v9h3.757a3.75 3.75 0 011.994.574z"></path>
+</svg>
+          <span data-content="Wiki">Wiki</span>
+            <span id="wiki-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
+
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="security-tab" href="/PaddlePaddle/PaddleOCR/security" data-tab-item="i7security-tab" data-selected-links="security overview alerts policy token_scanning code_scanning /PaddlePaddle/PaddleOCR/security" data-pjax="#repo-content-pjax-container" data-hotkey="g s" data-ga-click="Repository, Navigation click, Security tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-shield UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M7.467.133a1.75 1.75 0 011.066 0l5.25 1.68A1.75 1.75 0 0115 3.48V7c0 1.566-.32 3.182-1.303 4.682-.983 1.498-2.585 2.813-5.032 3.855a1.7 1.7 0 01-1.33 0c-2.447-1.042-4.049-2.357-5.032-3.855C1.32 10.182 1 8.566 1 7V3.48a1.75 1.75 0 011.217-1.667l5.25-1.68zm.61 1.429a.25.25 0 00-.153 0l-5.25 1.68a.25.25 0 00-.174.238V7c0 1.358.275 2.666 1.057 3.86.784 1.194 2.121 2.34 4.366 3.297a.2.2 0 00.154 0c2.245-.956 3.582-2.104 4.366-3.298C13.225 9.666 13.5 8.36 13.5 7V3.48a.25.25 0 00-.174-.237l-5.25-1.68zM9 10.5a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.75a.75.75 0 10-1.5 0v3a.75.75 0 001.5 0v-3z"></path>
+</svg>
+          <span data-content="Security">Security</span>
+            <include-fragment src="/PaddlePaddle/PaddleOCR/security/overall-count" accept="text/fragment+html"></include-fragment>
+
+
+</a></li>
+      <li data-view-component="true" class="d-inline-flex">
+  <a id="insights-tab" href="/PaddlePaddle/PaddleOCR/pulse" data-tab-item="i8insights-tab" data-selected-links="repo_graphs repo_contributors dependency_graph dependabot_updates pulse people community /PaddlePaddle/PaddleOCR/pulse" data-pjax="#repo-content-pjax-container" data-ga-click="Repository, Navigation click, Insights tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
+
+                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-graph UnderlineNav-octicon d-none d-sm-inline">
+    <path fill-rule="evenodd" d="M1.5 1.75a.75.75 0 00-1.5 0v12.5c0 .414.336.75.75.75h14.5a.75.75 0 000-1.5H1.5V1.75zm14.28 2.53a.75.75 0 00-1.06-1.06L10 7.94 7.53 5.47a.75.75 0 00-1.06 0L3.22 8.72a.75.75 0 001.06 1.06L7 7.06l2.47 2.47a.75.75 0 001.06 0l5.25-5.25z"></path>
+</svg>
+          <span data-content="Insights">Insights</span>
+            <span id="insights-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
+
+
+
+</a></li>
+</ul>
+    <div style="visibility:hidden;" data-view-component="true" class="UnderlineNav-actions js-responsive-underlinenav-overflow position-absolute pr-3 pr-md-4 pr-lg-5 right-0">      <details data-view-component="true" class="details-overlay details-reset position-relative">
+  <summary role="button" data-view-component="true">          <div class="UnderlineNav-item mr-0 border-0">
+            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
+    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
+</svg>
+            <span class="sr-only">More</span>
+          </div>
+</summary>
+  <div data-view-component="true">          <details-menu role="menu" data-view-component="true" class="dropdown-menu dropdown-menu-sw">
+
+            <ul>
+                <li data-menu-item="i0code-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item selected dropdown-item" aria-current="page" data-selected-links="repo_source repo_downloads repo_commits repo_releases repo_tags repo_branches repo_packages repo_deployments /PaddlePaddle/PaddleOCR/tree/dygraph" href="/PaddlePaddle/PaddleOCR/tree/dygraph">
+                    Code
+</a>                </li>
+                <li data-menu-item="i1issues-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_issues repo_labels repo_milestones /PaddlePaddle/PaddleOCR/issues" href="/PaddlePaddle/PaddleOCR/issues">
+                    Issues
+</a>                </li>
+                <li data-menu-item="i2pull-requests-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_pulls checks /PaddlePaddle/PaddleOCR/pulls" href="/PaddlePaddle/PaddleOCR/pulls">
+                    Pull requests
+</a>                </li>
+                <li data-menu-item="i3discussions-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_discussions /PaddlePaddle/PaddleOCR/discussions" href="/PaddlePaddle/PaddleOCR/discussions">
+                    Discussions
+</a>                </li>
+                <li data-menu-item="i4actions-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_actions /PaddlePaddle/PaddleOCR/actions" href="/PaddlePaddle/PaddleOCR/actions">
+                    Actions
+</a>                </li>
+                <li data-menu-item="i5projects-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_projects new_repo_project repo_project /PaddlePaddle/PaddleOCR/projects?type=beta" href="/PaddlePaddle/PaddleOCR/projects?type=beta">
+                    Projects
+</a>                </li>
+                <li data-menu-item="i6wiki-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_wiki /PaddlePaddle/PaddleOCR/wiki" href="/PaddlePaddle/PaddleOCR/wiki">
+                    Wiki
+</a>                </li>
+                <li data-menu-item="i7security-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="security overview alerts policy token_scanning code_scanning /PaddlePaddle/PaddleOCR/security" href="/PaddlePaddle/PaddleOCR/security">
+                    Security
+</a>                </li>
+                <li data-menu-item="i8insights-tab" hidden>
+                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_graphs repo_contributors dependency_graph dependabot_updates pulse people community /PaddlePaddle/PaddleOCR/pulse" href="/PaddlePaddle/PaddleOCR/pulse">
+                    Insights
+</a>                </li>
+            </ul>
+
+</details-menu></div>
+</details></div>
+</nav>
+  </div>
+
+
+
+<div id="repo-content-pjax-container" class="repository-content " >
+
+
+
+
+
+
+  <div class="clearfix container-xl px-3 px-md-4 px-lg-5 mt-4">
+
+
+<div>
+
+
+
+
+
+
+
+
+<a class="d-none js-permalink-shortcut" data-hotkey="y" href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">Permalink</a>
+
+<div class="d-flex flex-items-start flex-shrink-0 pb-3 flex-wrap flex-md-nowrap flex-justify-between flex-md-justify-start">
+
+<div class="position-relative">
+  <details class="details-reset details-overlay mr-0 mb-0 " id="branch-select-menu">
+    <summary class="btn css-truncate"
+            data-hotkey="w"
+            title="Switch branches or tags">
+      <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-git-branch">
+    <path fill-rule="evenodd" d="M11.75 2.5a.75.75 0 100 1.5.75.75 0 000-1.5zm-2.25.75a2.25 2.25 0 113 2.122V6A2.5 2.5 0 0110 8.5H6a1 1 0 00-1 1v1.128a2.251 2.251 0 11-1.5 0V5.372a2.25 2.25 0 111.5 0v1.836A2.492 2.492 0 016 7h4a1 1 0 001-1v-.628A2.25 2.25 0 019.5 3.25zM4.25 12a.75.75 0 100 1.5.75.75 0 000-1.5zM3.5 3.25a.75.75 0 111.5 0 .75.75 0 01-1.5 0z"></path>
+</svg>
+      <span class="css-truncate-target" data-menu-button>dygraph</span>
+      <span class="dropdown-caret"></span>
+    </summary>
+
+
+<div class="SelectMenu">
+  <div class="SelectMenu-modal">
+    <header class="SelectMenu-header">
+      <span class="SelectMenu-title">Switch branches/tags</span>
+      <button class="SelectMenu-closeButton" type="button" data-toggle-for="branch-select-menu"><svg aria-label="Close menu" aria-hidden="false" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
+    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
+</svg></button>
+    </header>
+
+    <input-demux data-action="tab-container-change:input-demux#storeInput tab-container-changed:input-demux#updateInput">
+      <tab-container class="d-flex flex-column js-branches-tags-tabs" style="min-height: 0;">
+        <div class="SelectMenu-filter">
+          <input data-target="input-demux.source"
+                 id="context-commitish-filter-field"
+                 class="SelectMenu-input form-control"
+                 aria-owns="ref-list-branches"
+                 data-controls-ref-menu-id="ref-list-branches"
+                 autofocus
+                 autocomplete="off"
+                 aria-label="Filter branches/tags"
+                 placeholder="Filter branches/tags"
+                 type="text"
+          >
+        </div>
+
+        <div class="SelectMenu-tabs" role="tablist" data-target="input-demux.control" >
+          <button class="SelectMenu-tab" type="button" role="tab" aria-selected="true">Branches</button>
+          <button class="SelectMenu-tab" type="button" role="tab">Tags</button>
+        </div>
+
+        <div role="tabpanel" id="ref-list-branches" data-filter-placeholder="Filter branches/tags" tabindex="" class="d-flex flex-column flex-auto overflow-auto">
+          <ref-selector
+            type="branch"
+            data-targets="input-demux.sinks"
+            data-action="
+              input-entered:ref-selector#inputEntered
+              tab-selected:ref-selector#tabSelected
+              focus-list:ref-selector#focusFirstListMember
+            "
+            query-endpoint="/PaddlePaddle/PaddleOCR/refs"
+
+            cache-key="v0:1655177037.9190001"
+            current-committish="ZHlncmFwaA=="
+            default-branch="cmVsZWFzZS8yLjU="
+            name-with-owner="UGFkZGxlUGFkZGxlL1BhZGRsZU9DUg=="
+            prefetch-on-mouseover
+          >
+
+            <template data-target="ref-selector.fetchFailedTemplate">
+              <div class="SelectMenu-message" data-index="{{ index }}">Could not load branches</div>
+            </template>
+
+              <template data-target="ref-selector.noMatchTemplate">
+    <div class="SelectMenu-message">Nothing to show</div>
+</template>
+
+
+            <div data-target="ref-selector.listContainer" role="menu" class="SelectMenu-list " data-pjax="#repo-content-pjax-container">
+              <div class="SelectMenu-loading pt-3 pb-0 overflow-hidden" aria-label="Menu is loading">
+                <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="anim-rotate">
+  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
+  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
+</svg>
+              </div>
+            </div>
+
+              <template data-target="ref-selector.itemTemplate">
+  <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{ urlEncodedRefName }}/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="SelectMenu-item" role="menuitemradio" rel="nofollow" aria-checked="{{ isCurrent }}" data-index="{{ index }}">
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check SelectMenu-icon SelectMenu-icon--check">
+    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
+</svg>
+    <span class="flex-1 css-truncate css-truncate-overflow {{ isFilteringClass }}">{{ refName }}</span>
+    <span hidden="{{ isNotDefault }}" class="Label Label--secondary flex-self-start">default</span>
+  </a>
+</template>
+
+
+              <footer class="SelectMenu-footer"><a href="/PaddlePaddle/PaddleOCR/branches">View all branches</a></footer>
+          </ref-selector>
+
+        </div>
+
+        <div role="tabpanel" id="tags-menu" data-filter-placeholder="Find a tag" tabindex="" hidden class="d-flex flex-column flex-auto overflow-auto">
+          <ref-selector
+            type="tag"
+            data-action="
+              input-entered:ref-selector#inputEntered
+              tab-selected:ref-selector#tabSelected
+              focus-list:ref-selector#focusFirstListMember
+            "
+            data-targets="input-demux.sinks"
+            query-endpoint="/PaddlePaddle/PaddleOCR/refs"
+            cache-key="v0:1655177037.9190001"
+            current-committish="ZHlncmFwaA=="
+            default-branch="cmVsZWFzZS8yLjU="
+            name-with-owner="UGFkZGxlUGFkZGxlL1BhZGRsZU9DUg=="
+          >
+
+            <template data-target="ref-selector.fetchFailedTemplate">
+              <div class="SelectMenu-message" data-index="{{ index }}">Could not load tags</div>
+            </template>
+
+            <template data-target="ref-selector.noMatchTemplate">
+              <div class="SelectMenu-message" data-index="{{ index }}">Nothing to show</div>
+            </template>
+
+              <template data-target="ref-selector.itemTemplate">
+  <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{ urlEncodedRefName }}/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="SelectMenu-item" role="menuitemradio" rel="nofollow" aria-checked="{{ isCurrent }}" data-index="{{ index }}">
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check SelectMenu-icon SelectMenu-icon--check">
+    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
+</svg>
+    <span class="flex-1 css-truncate css-truncate-overflow {{ isFilteringClass }}">{{ refName }}</span>
+    <span hidden="{{ isNotDefault }}" class="Label Label--secondary flex-self-start">default</span>
+  </a>
+</template>
+
+
+            <div data-target="ref-selector.listContainer" role="menu" class="SelectMenu-list" data-pjax="#repo-content-pjax-container">
+              <div class="SelectMenu-loading pt-3 pb-0 overflow-hidden" aria-label="Menu is loading">
+                <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="anim-rotate">
+  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
+  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
+</svg>
+              </div>
+            </div>
+              <footer class="SelectMenu-footer"><a href="/PaddlePaddle/PaddleOCR/tags">View all tags</a></footer>
+          </ref-selector>
+        </div>
+      </tab-container>
+    </input-demux>
+  </div>
+</div>
+
+  </details>
+
+</div>
+
+  <h2 id="blob-path" class="breadcrumb flex-auto flex-self-center min-width-0 text-normal mx-2 width-full width-md-auto flex-order-1 flex-md-order-none mt-3 mt-md-0">
+    <span class="js-repo-root text-bold"><span class="js-path-segment d-inline-block wb-break-all"><a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR/tree/dygraph"><span>PaddleOCR</span></a></span></span><span class="separator">/</span><span class="js-path-segment d-inline-block wb-break-all"><a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR/tree/dygraph/applications"><span>applications</span></a></span><span class="separator">/</span><strong class="final-path">轻量级车牌识别.md</strong>
+  </h2>
+  <a href="/PaddlePaddle/PaddleOCR/find/dygraph" data-pjax="" data-hotkey="t" data-view-component="true" class="btn mr-2 d-none d-md-block">  Go to file
+
+</a>
+  <details id="blob-more-options-details" data-view-component="true" class="details-overlay details-reset position-relative">
+  <summary role="button" data-view-component="true" class="btn">  <svg aria-label="More options" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
+    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
+</svg>
+
+</summary>
+  <div data-view-component="true">      <ul class="dropdown-menu dropdown-menu-sw">
+        <li class="d-block d-md-none">
+          <a class="dropdown-item d-flex flex-items-baseline" data-hydro-click="{&quot;event_type&quot;:&quot;repository.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;FIND_FILE_BUTTON&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="719caca13290ec1595dbf25afc93e9874844c0bcf4645d121411be9e6fbaa205" data-ga-click="Repository, find file, location:repo overview" data-hotkey="t" data-pjax="true" href="/PaddlePaddle/PaddleOCR/find/dygraph">
+            <span class="flex-auto">Go to file</span>
+            <span class="text-small color-fg-muted" aria-hidden="true">T</span>
+</a>        </li>
+        <li data-toggle-for="blob-more-options-details">
+          <button data-toggle-for="jumpto-line-details-dialog" type="button" data-view-component="true" class="dropdown-item btn-link">  <span class="d-flex flex-items-baseline">
+              <span class="flex-auto">Go to line</span>
+              <span class="text-small color-fg-muted" aria-hidden="true">L</span>
+            </span>
+
+</button>        </li>
+        <li class="dropdown-divider" role="none"></li>
+        <li>
+          <clipboard-copy data-toggle-for="blob-more-options-details" aria-label="Copy path" value="applications/轻量级车牌识别.md" data-view-component="true" class="dropdown-item cursor-pointer">
+
+            Copy path
+
+</clipboard-copy>        </li>
+        <li>
+          <clipboard-copy data-toggle-for="blob-more-options-details" aria-label="Copy permalink" value="https://github.com/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-view-component="true" class="dropdown-item cursor-pointer">
+
+            <span class="d-flex flex-items-baseline">
+              <span class="flex-auto">Copy permalink</span>
+            </span>
+
+</clipboard-copy>        </li>
+      </ul>
+</div>
+</details></div>
+
+
+
+
+
+    <div id="spoof-warning" class="mt-0 pb-3" hidden aria-hidden>
+  <div data-view-component="true" class="flash flash-warn mt-0 clearfix">
+
+
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert float-left mt-1">
+    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
+</svg>
+
+      <div class="overflow-hidden">This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.</div>
+
+
+
+</div></div>
+
+    <include-fragment src="/PaddlePaddle/PaddleOCR/spoofed_commit_check/a3dfc10b7824743939976c42db6d50d118e2b3b9" data-test-selector="spoofed-commit-check"></include-fragment>
+
+    <div class="Box d-flex flex-column flex-shrink-0 mb-3">
+
+  <div class="Box-header Details js-details-container">
+      <div class="d-flex flex-items-center">
+        <span class="flex-shrink-0 ml-n1 mr-n1 mt-n1 mb-n1">
+          <a rel="contributor" data-skip-pjax="true" data-hovercard-type="user" data-hovercard-url="/users/MissPenguin/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/MissPenguin"><img class="avatar avatar-user" src="https://avatars.githubusercontent.com/u/25809855?s=48&amp;v=4" width="24" height="24" alt="@MissPenguin" /></a>
+        </span>
+        <div class="flex-1 d-flex flex-items-center ml-3 min-width-0">
+          <div class="css-truncate css-truncate-overflow">
+            <a class="text-bold Link--primary" rel="contributor" data-hovercard-type="user" data-hovercard-url="/users/MissPenguin/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/MissPenguin">MissPenguin</a>
+
+              <span class="markdown-title">
+                <a data-pjax="true" title="Update 轻量级车牌识别.md" class="Link--secondary" href="/PaddlePaddle/PaddleOCR/commit/10a3f2573d8e875c69683a71dc067d9060fb6875">Update 轻量级车牌识别.md</a>
+              </span>
+          </div>
+
+
+          <span class="ml-2">
+            <include-fragment accept="text/fragment+html" src="/PaddlePaddle/PaddleOCR/commit/10a3f2573d8e875c69683a71dc067d9060fb6875/rollup?direction=e" class="d-inline"></include-fragment>
+          </span>
+        </div>
+        <div class="ml-3 d-flex flex-shrink-0 flex-items-center flex-justify-end color-fg-muted no-wrap">
+          <span class="d-none d-md-inline">
+            <span>Latest commit</span>
+            <a class="text-small text-mono Link--secondary" href="/PaddlePaddle/PaddleOCR/commit/10a3f2573d8e875c69683a71dc067d9060fb6875" data-pjax>10a3f25</a>
+            <span itemprop="dateModified"><relative-time datetime="2022-06-15T10:13:01Z" class="no-wrap">Jun 15, 2022</relative-time></span>
+          </span>
+
+          <a data-pjax href="/PaddlePaddle/PaddleOCR/commits/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="ml-3 no-wrap Link--primary no-underline">
+            <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-history">
+    <path fill-rule="evenodd" d="M1.643 3.143L.427 1.927A.25.25 0 000 2.104V5.75c0 .138.112.25.25.25h3.646a.25.25 0 00.177-.427L2.715 4.215a6.5 6.5 0 11-1.18 4.458.75.75 0 10-1.493.154 8.001 8.001 0 101.6-5.684zM7.75 4a.75.75 0 01.75.75v2.992l2.028.812a.75.75 0 01-.557 1.392l-2.5-1A.75.75 0 017 8.25v-3.5A.75.75 0 017.75 4z"></path>
+</svg>
+            <span class="d-none d-sm-inline">
+              <strong>History</strong>
+            </span>
+          </a>
+        </div>
+      </div>
+
+  </div>
+
+  <div class="Box-body d-flex flex-items-center flex-auto border-bottom-0 flex-wrap" >
+    <details class="details-reset details-overlay details-overlay-dark lh-default color-fg-default float-left mr-3" id="blob_contributors_box">
+      <summary class="Link--primary">
+        <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-people">
+    <path fill-rule="evenodd" d="M5.5 3.5a2 2 0 100 4 2 2 0 000-4zM2 5.5a3.5 3.5 0 115.898 2.549 5.507 5.507 0 013.034 4.084.75.75 0 11-1.482.235 4.001 4.001 0 00-7.9 0 .75.75 0 01-1.482-.236A5.507 5.507 0 013.102 8.05 3.49 3.49 0 012 5.5zM11 4a.75.75 0 100 1.5 1.5 1.5 0 01.666 2.844.75.75 0 00-.416.672v.352a.75.75 0 00.574.73c1.2.289 2.162 1.2 2.522 2.372a.75.75 0 101.434-.44 5.01 5.01 0 00-2.56-3.012A3 3 0 0011 4z"></path>
+</svg>
+        <strong>2</strong>
+
+        contributors
+      </summary>
+      <details-dialog
+        class="Box Box--overlay d-flex flex-column anim-fade-in fast"
+        aria-label="Users who have contributed to this file"
+        src="/PaddlePaddle/PaddleOCR/contributors-list/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" preload>
+        <div class="Box-header">
+          <button class="Box-btn-octicon btn-octicon float-right" type="button" aria-label="Close dialog" data-close-dialog>
+            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
+    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
+</svg>
+          </button>
+          <h3 class="Box-title">
+            Users who have contributed to this file
+          </h3>
+        </div>
+        <include-fragment>
+          <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="my-3 mx-auto d-block anim-rotate">
+  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
+  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
+</svg>
+        </include-fragment>
+      </details-dialog>
+    </details>
+      <span class="">
+    <a class="avatar-link" data-hovercard-type="user" data-hovercard-url="/users/WenmuZhou/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle/PaddleOCR/commits/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?author=WenmuZhou">
+      <img class="avatar mr-2 avatar-user" src="https://avatars.githubusercontent.com/u/12406017?s=48&amp;v=4" width="24" height="24" alt="@WenmuZhou" />
+</a>    <a class="avatar-link" data-hovercard-type="user" data-hovercard-url="/users/MissPenguin/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle/PaddleOCR/commits/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?author=MissPenguin">
+      <img class="avatar mr-2 avatar-user" src="https://avatars.githubusercontent.com/u/25809855?s=48&amp;v=4" width="24" height="24" alt="@MissPenguin" />
+</a>
+</span>
+
+  </div>
+</div>
+
+
+
+
+
+
+
+    <readme-toc>
+
+    <div data-target="readme-toc.content" class="Box mt-3 position-relative">
+
+  <div
+    class="Box-header js-blob-header blob-header js-sticky js-position-sticky top-0 p-2 d-flex flex-shrink-0 flex-md-row flex-items-center"
+    style="position: sticky; z-index: 1;"
+  >
+
+      <details
+  data-target="readme-toc.trigger"
+  data-menu-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;trigger&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}"
+  data-menu-hydro-click-hmac="532f79aab7b37e9668b558163fc4a6d01f017a94245e04831a2e20c0a43fff35"
+  class="dropdown details-reset details-overlay"
+>
+  <summary
+    class="btn btn-octicon m-0 mr-2 p-2"
+    aria-haspopup="true"
+    aria-label="Table of Contents">
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-list-unordered">
+    <path fill-rule="evenodd" d="M2 4a1 1 0 100-2 1 1 0 000 2zm3.75-1.5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zm0 5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zm0 5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zM3 8a1 1 0 11-2 0 1 1 0 012 0zm-1 6a1 1 0 100-2 1 1 0 000 2z"></path>
+</svg>
+  </summary>
+
+
+  <details-menu class="SelectMenu" role="menu">
+    <div class="SelectMenu-modal rounded-3 mt-1" style="max-height:340px;">
+
+        <div class="SelectMenu-filter">
+          <input
+            class="SelectMenu-input form-control js-filterable-field"
+            id="toc-filter-field"
+            type="text"
+            autocomplete="off"
+            spellcheck="false"
+            autofocus
+            placeholder="Filter headings"
+            aria-label="Filter headings">
+        </div>
+
+      <div class="SelectMenu-list SelectMenu-list--borderless p-2" style="overscroll-behavior: contain;" data-filterable-for="toc-filter-field" data-filterable-type="substring">
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 text-emphasized" style="-webkit-box-orient: vertical; padding-left: 12px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#一种基于paddleocr的轻量级车牌识别模型">一种基于PaddleOCR的轻量级车牌识别模型</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#1-项目介绍">1. 项目介绍</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#2-环境搭建">2. 环境搭建</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#3-数据集准备">3. 数据集准备</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#31-数据集标注规则">3.1 数据集标注规则</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#32-制作符合pp-ocr训练格式的标注文件">3.2 制作符合PP-OCR训练格式的标注文件</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#4-实验">4. 实验</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#41-检测">4.1 检测</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#411-预训练模型直接预测">4.1.1 预训练模型直接预测</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#412-ccpd车牌数据集fine-tune">4.1.2 CCPD车牌数据集fine-tune</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#413-ccpd车牌数据集fine-tune量化训练">4.1.3 CCPD车牌数据集fine-tune+量化训练</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#414-模型导出">4.1.4 模型导出</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#42-识别">4.2 识别</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#421-预训练模型直接预测">4.2.1 预训练模型直接预测</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#422-预训练模型直接预测改动后处理">4.2.2 预训练模型直接预测+改动后处理</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#423-ccpd车牌数据集fine-tune">4.2.3 CCPD车牌数据集fine-tune</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#424-ccpd车牌数据集fine-tune量化训练">4.2.4 CCPD车牌数据集fine-tune+量化训练</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#425-模型导出">4.2.5 模型导出</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#43-计算end2end指标">4.3 计算End2End指标</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#44-部署">4.4 部署</a>
+          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#45-实验总结">4.5 实验总结</a>
+      </div>
+    </div>
+  </details-menu>
+</details>
+
+
+  <div class="text-mono f6 flex-auto pr-3 flex-order-2 flex-md-order-1">
+
+      831 lines (628 sloc)
+      <span class="file-info-divider"></span>
+    41.2 KB
+  </div>
+
+  <div class="d-flex py-1 py-md-0 flex-auto flex-order-1 flex-md-order-2 flex-sm-grow-0 flex-justify-between hide-sm hide-md">
+        <div class="BtnGroup">
+    <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?plain=1" data-permalink-href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?plain=1" aria-label="Display the source blob" data-view-component="true" class="source tooltipped tooltipped tooltipped-n  js-permalink-replaceable-link btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-code">
+    <path fill-rule="evenodd" d="M4.72 3.22a.75.75 0 011.06 1.06L2.06 8l3.72 3.72a.75.75 0 11-1.06 1.06L.47 8.53a.75.75 0 010-1.06l4.25-4.25zm6.56 0a.75.75 0 10-1.06 1.06L13.94 8l-3.72 3.72a.75.75 0 101.06 1.06l4.25-4.25a.75.75 0 000-1.06l-4.25-4.25z"></path>
+</svg>
+
+</a>    <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-permalink-href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" aria-label="Display the rendered blob" data-view-component="true" class="rendered tooltipped tooltipped tooltipped-n selected js-permalink-replaceable-link btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-file">
+    <path fill-rule="evenodd" d="M3.75 1.5a.25.25 0 00-.25.25v12.5c0 .138.112.25.25.25h9.5a.25.25 0 00.25-.25V6h-2.75A1.75 1.75 0 019 4.25V1.5H3.75zm6.75.062V4.25c0 .138.112.25.25.25h2.688a.252.252 0 00-.011-.013l-2.914-2.914a.272.272 0 00-.013-.011zM2 1.75C2 .784 2.784 0 3.75 0h6.586c.464 0 .909.184 1.237.513l2.914 2.914c.329.328.513.773.513 1.237v9.586A1.75 1.75 0 0113.25 16h-9.5A1.75 1.75 0 012 14.25V1.75z"></path>
+</svg>
+
+</a>  </div>
+
+
+    <div class="BtnGroup">
+      <a data-permalink-href="/PaddlePaddle/PaddleOCR/raw/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" href="/PaddlePaddle/PaddleOCR/raw/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" id="raw-url" data-view-component="true" class="js-permalink-replaceable-link btn-sm btn BtnGroup-item">  Raw
+
+</a>        <a data-permalink-href="/PaddlePaddle/PaddleOCR/blame/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" href="/PaddlePaddle/PaddleOCR/blame/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-hotkey="b" data-view-component="true" class="js-update-url-with-hash js-permalink-replaceable-link btn-sm btn BtnGroup-item">  Blame
+
+</a>    </div>
+
+    <div class="d-flex">
+
+
+
+<div class="ml-1" data-test-selector="edit-dropdown-menu-component">
+  <form class="BtnGroup-parent js-update-url-with-hash " data-turbo="false" action="/PaddlePaddle/PaddleOCR/edit/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" accept-charset="UTF-8" method="post"><input type="hidden" name="authenticity_token" value="llz78Eh-h4wCyHYKnMSomzLitBJmBsXnGWrYyE5RqU8R0h9nBeLKEJMLkCaprphwyYcNklhkrLFDqGzHP14SMA" autocomplete="off" />
+    <button disabled="disabled" title="You must be signed in to make or propose changes" data-hotkey="e" data-disable-with="" data-test-selector="edit-icon-button" type="submit" data-view-component="true" class="btn-sm BtnGroup-item btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-pencil">
+    <path fill-rule="evenodd" d="M11.013 1.427a1.75 1.75 0 012.474 0l1.086 1.086a1.75 1.75 0 010 2.474l-8.61 8.61c-.21.21-.47.364-.756.445l-3.251.93a.75.75 0 01-.927-.928l.929-3.25a1.75 1.75 0 01.445-.758l8.61-8.61zm1.414 1.06a.25.25 0 00-.354 0L10.811 3.75l1.439 1.44 1.263-1.263a.25.25 0 000-.354l-1.086-1.086zM11.189 6.25L9.75 4.81l-6.286 6.287a.25.25 0 00-.064.108l-.558 1.953 1.953-.558a.249.249 0 00.108-.064l6.286-6.286z"></path>
+</svg>
+
+</button></form>
+  <details class="details-reset details-overlay select-menu BtnGroup-parent d-inline-block position-relative">
+    <summary data-disable-invalid="" data-disable-with="" data-hydro-click="{&quot;event_type&quot;:&quot;blob_edit_dropdown.more_options_click&quot;,&quot;payload&quot;:{&quot;repository_id&quot;:262296122,&quot;actor_id&quot;:null,&quot;github_dev_enabled&quot;:false,&quot;edit_enabled&quot;:false,&quot;small_screen&quot;:false,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="8e1c91138fac18e1604a705b864e87574bdf4b05c6f3b559e5a5a9051b4f9f4c" aria-label="Select additional options" data-view-component="true" class="select-menu-button btn-sm btn BtnGroup-item float-none px-2">
+</summary>    <div class="SelectMenu right-0">
+      <div class="SelectMenu-modal width-full">
+        <div class="SelectMenu-list SelectMenu-list--borderless py-2">
+          <form class="SelectMenu-item js-update-url-with-hash " data-turbo="false" action="/PaddlePaddle/PaddleOCR/edit/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" accept-charset="UTF-8" method="post"><input type="hidden" name="authenticity_token" value="Ovbx0rgGgsdbao-R_-A0l2RC6K1eBosPF4oakk00YT69eBVF9ZrPW8qpab3KigR8nydRLWBk4llNSK6dPDvaQQ" autocomplete="off" />
+            <button disabled="disabled" title="You must be signed in to make or propose changes" data-test-selector="edit-text-button" type="submit" data-view-component="true" class="btn-invisible btn width-full d-flex flex-justify-between color-fg-muted text-normal p-0">  <div>Edit this file</div>
+              <div class="color-fg-muted">E</div>
+
+</button></form>
+
+            <a data-platforms="windows,mac" data-test-selector="github-desktop-link" aria-label="Open this file in GitHub Desktop" href="https://desktop.github.com" data-view-component="true" class="SelectMenu-item no-wrap js-remove-unless-platform width-full text-normal color-fg-default f5">
+              Open in GitHub Desktop
+</a>        </div>
+      </div>
+    </div>
+  </details>
+</div>
+
+
+
+<div data-test-selector="remote-clipboard-copy">
+  </div>
+
+
+          <button class="btn-octicon btn-octicon-danger disabled tooltipped tooltipped-nw" disabled
+            aria-label="You must be signed in to make or propose changes" type="button">
+            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-trash">
+    <path fill-rule="evenodd" d="M6.5 1.75a.25.25 0 01.25-.25h2.5a.25.25 0 01.25.25V3h-3V1.75zm4.5 0V3h2.25a.75.75 0 010 1.5H2.75a.75.75 0 010-1.5H5V1.75C5 .784 5.784 0 6.75 0h2.5C10.216 0 11 .784 11 1.75zM4.496 6.675a.75.75 0 10-1.492.15l.66 6.6A1.75 1.75 0 005.405 15h5.19c.9 0 1.652-.681 1.741-1.576l.66-6.6a.75.75 0 00-1.492-.149l-.66 6.6a.25.25 0 01-.249.225h-5.19a.25.25 0 01-.249-.225l-.66-6.6z"></path>
+</svg>
+          </button>
+    </div>
+  </div>
+
+    <div class="d-flex hide-lg hide-xl flex-order-2 flex-grow-0">
+      <details class="dropdown details-reset details-overlay d-inline-block">
+        <summary
+          class="btn-octicon p-2"
+          aria-haspopup="true"
+          aria-label="possible actions"
+          data-test-selector="small-screen-more-options"
+          data-hydro-click="{&quot;event_type&quot;:&quot;blob_edit_dropdown.more_options_click&quot;,&quot;payload&quot;:{&quot;repository_id&quot;:262296122,&quot;actor_id&quot;:null,&quot;github_dev_enabled&quot;:false,&quot;edit_enabled&quot;:false,&quot;small_screen&quot;:true,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}"
+          data-hydro-click-hmac="3711aba9bcc87015fd0b3b29260720d15e89cbf158ae938adc99aa4a84a101e5"
+        >
+          <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
+    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
+</svg>
+        </summary>
+
+        <ul class="dropdown-menu dropdown-menu-sw" style="width: 175px">
+            <li>
+                <a class="dropdown-item tooltipped tooltipped-nw js-remove-unless-platform"
+                   data-platforms="windows,mac"
+                   href="https://desktop.github.com"
+                   data-ga-click="Repository, open with desktop">
+                  Open with Desktop
+                </a>
+            </li>
+          <li>
+            <a class="dropdown-item" href="/PaddlePaddle/PaddleOCR/raw/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">
+              View raw
+            </a>
+          </li>
+            <li>
+                          </li>
+            <li>
+              <a class="dropdown-item" href="/PaddlePaddle/PaddleOCR/blame/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">
+                View blame
+              </a>
+            </li>
+
+        </ul>
+      </details>
+    </div>
+</div>
+
+
+        <div id="readme" class="Box-body readme blob js-code-block-container p-5 p-xl-6 gist-border-0">
+    <article class="markdown-body entry-content container-lg" itemprop="text"><h1 dir="auto"><a id="user-content-一种基于paddleocr的轻量级车牌识别模型" class="anchor" aria-hidden="true" href="#一种基于paddleocr的轻量级车牌识别模型"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>一种基于PaddleOCR的轻量级车牌识别模型</h1>
+<ul dir="auto">
+<li><a href="#1-%E9%A1%B9%E7%9B%AE%E4%BB%8B%E7%BB%8D">1. 项目介绍</a></li>
+<li><a href="#2-%E7%8E%AF%E5%A2%83%E6%90%AD%E5%BB%BA">2. 环境搭建</a></li>
+<li><a href="#3-%E6%95%B0%E6%8D%AE%E9%9B%86%E5%87%86%E5%A4%87">3. 数据集准备</a>
+<ul dir="auto">
+<li><a href="#31-%E6%95%B0%E6%8D%AE%E9%9B%86%E6%A0%87%E6%B3%A8%E8%A7%84%E5%88%99">3.1 数据集标注规则</a></li>
+<li><a href="#32-%E5%88%B6%E4%BD%9C%E7%AC%A6%E5%90%88pp-ocr%E8%AE%AD%E7%BB%83%E6%A0%BC%E5%BC%8F%E7%9A%84%E6%A0%87%E6%B3%A8%E6%96%87%E4%BB%B6">3.2 制作符合PP-OCR训练格式的标注文件</a></li>
+</ul>
+</li>
+<li><a href="#4-%E5%AE%9E%E9%AA%8C">4. 实验</a>
+<ul dir="auto">
+<li><a href="#41-%E6%A3%80%E6%B5%8B">4.1 检测</a>
+<ul dir="auto">
+<li><a href="#411-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B">4.1.1 预训练模型直接预测</a></li>
+<li><a href="#412-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune">4.1.2 CCPD车牌数据集fine-tune</a></li>
+<li><a href="#413-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune%E9%87%8F%E5%8C%96%E8%AE%AD%E7%BB%83">4.1.3 CCPD车牌数据集fine-tune+量化训练</a></li>
+<li><a href="#414-%E6%A8%A1%E5%9E%8B%E5%AF%BC%E5%87%BA">4.1.4 模型导出</a></li>
+</ul>
+</li>
+<li><a href="#42-%E8%AF%86%E5%88%AB">4.2 识别</a>
+<ul dir="auto">
+<li><a href="#421-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B">4.2.1 预训练模型直接预测</a></li>
+<li><a href="#422-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B%E6%94%B9%E5%8A%A8%E5%90%8E%E5%A4%84%E7%90%86">4.2.2 预训练模型直接预测+改动后处理</a></li>
+<li><a href="#423-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune">4.2.3 CCPD车牌数据集fine-tune</a></li>
+<li><a href="#424-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune%E9%87%8F%E5%8C%96%E8%AE%AD%E7%BB%83">4.2.4 CCPD车牌数据集fine-tune+量化训练</a></li>
+<li><a href="#425-%E6%A8%A1%E5%9E%8B%E5%AF%BC%E5%87%BA">4.2.5 模型导出</a></li>
+</ul>
+</li>
+<li><a href="#43-%E8%AE%A1%E7%AE%97End2End%E6%8C%87%E6%A0%87">4.3 计算End2End指标</a></li>
+<li><a href="#44-%E9%83%A8%E7%BD%B2">4.4 部署</a></li>
+<li><a href="#45-%E5%AE%9E%E9%AA%8C%E6%80%BB%E7%BB%93">4.5 实验总结</a></li>
+</ul>
+</li>
+</ul>
+<h2 dir="auto"><a id="user-content-1-项目介绍" class="anchor" aria-hidden="true" href="#1-项目介绍"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>1. 项目介绍</h2>
+<p dir="auto">车牌识别(Vehicle License Plate Recognition，VLPR) 是计算机视频图像识别技术在车辆牌照识别中的一种应用。车牌识别技术要求能够将运动中的汽车牌照从复杂背景中提取并识别出来，在高速公路车辆管理，停车场管理和城市交通中得到广泛应用。</p>
+<p dir="auto">本项目难点如下：</p>
+<ol dir="auto">
+<li>车牌在图像中的尺度差异大、在车辆上的悬挂位置不固定</li>
+<li>车牌图像质量层次不齐: 角度倾斜、图片模糊、光照不足、过曝等问题严重</li>
+<li>边缘和端测场景应用对模型大小有限制，推理速度有要求</li>
+</ol>
+<p dir="auto">针对以上问题， 本例选用 PP-OCRv3 这一开源超轻量OCR系统进行车牌识别系统的开发。基于PP-OCRv3模型，在CCPD数据集达到99%的检测和94%的识别精度，模型大小12.8M(2.5M+10.3M)。基于量化对模型体积进行进一步压缩到5.8M(1M+4.8M), 同时推理速度提升25%。</p>
+<p dir="auto">aistudio项目链接: <a href="https://aistudio.baidu.com/aistudio/projectdetail/3919091?contributionType=1" rel="nofollow">基于PaddleOCR的轻量级车牌识别范例</a></p>
+<h2 dir="auto"><a id="user-content-2-环境搭建" class="anchor" aria-hidden="true" href="#2-环境搭建"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>2. 环境搭建</h2>
+<p dir="auto">本任务基于Aistudio完成, 具体环境如下：</p>
+<ul dir="auto">
+<li>操作系统: Linux</li>
+<li>PaddlePaddle: 2.3</li>
+<li>paddleslim: 2.2.2</li>
+<li>PaddleOCR: Release/2.5</li>
+</ul>
+<p dir="auto">下载 PaddleOCR代码</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR"><pre>git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR</pre></div>
+<p dir="auto">安装依赖库</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="pip install -r PaddleOCR/requirements.txt"><pre>pip install -r PaddleOCR/requirements.txt</pre></div>
+<h2 dir="auto"><a id="user-content-3-数据集准备" class="anchor" aria-hidden="true" href="#3-数据集准备"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3. 数据集准备</h2>
+<p dir="auto">所使用的数据集为 CCPD2020 新能源车牌数据集，该数据集为</p>
+<p dir="auto">该数据集分布如下：</p>
+<table>
+<thead>
+<tr>
+<th>数据集类型</th>
+<th>数量</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>训练集</td>
+<td>5769</td>
+</tr>
+<tr>
+<td>验证集</td>
+<td>1001</td>
+</tr>
+<tr>
+<td>测试集</td>
+<td>5006</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">数据集图片示例如下:
+<a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/422ba572727331bfacf4a748100bcdf775d5ed4b2409ae78456849b2d94b1a8d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f33626365303537613865306334306130616362643236623265323965346532353930613331626334313237363462653762396534393739396336396362393163"><img src="https://camo.githubusercontent.com/422ba572727331bfacf4a748100bcdf775d5ed4b2409ae78456849b2d94b1a8d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f33626365303537613865306334306130616362643236623265323965346532353930613331626334313237363462653762396534393739396336396362393163" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/3bce057a8e0c40a0acbd26b2e29e4e2590a31bc412764be7b9e49799c69cb91c" style="max-width: 100%;"></a></p>
+<p dir="auto">数据集可以从这里下载 <a href="https://aistudio.baidu.com/aistudio/datasetdetail/101595" rel="nofollow">https://aistudio.baidu.com/aistudio/datasetdetail/101595</a></p>
+<p dir="auto">下载好数据集后对数据集进行解压</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip"><pre>unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip</pre></div>
+<h3 dir="auto"><a id="user-content-31-数据集标注规则" class="anchor" aria-hidden="true" href="#31-数据集标注规则"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3.1 数据集标注规则</h3>
+<p dir="auto">CPPD数据集的图片文件名具有特殊规则，详细可查看：<a href="https://github.com/detectRecog/CCPD">https://github.com/detectRecog/CCPD</a></p>
+<p dir="auto">具体规则如下：</p>
+<p dir="auto">例如: 025-95_113-154&amp;383_386&amp;473-386&amp;473_177&amp;454_154&amp;383_363&amp;402-0_0_22_27_27_33_16-37-15.jpg</p>
+<p dir="auto">每个名称可以分为七个字段，以-符号作为分割。这些字段解释如下。</p>
+<ul dir="auto">
+<li>
+<p dir="auto">025：车牌面积与整个图片区域的面积比。025 (25%)</p>
+</li>
+<li>
+<p dir="auto">95_113：水平倾斜程度和垂直倾斜度。水平 95度 垂直 113度</p>
+</li>
+<li>
+<p dir="auto">154&amp;383_386&amp;473：左上和右下顶点的坐标。左上(154,383) 右下(386,473)</p>
+</li>
+<li>
+<p dir="auto">386&amp;473_177&amp;454_154&amp;383_363&amp;402：整个图像中车牌的四个顶点的精确（x，y）坐标。这些坐标从右下角顶点开始。(386,473) (177,454) (154,383) (363,402)</p>
+</li>
+<li>
+<p dir="auto">0_0_22_27_27_33_16：CCPD中的每个图像只有一个车牌。每个车牌号码由一个汉字，一个字母和五个字母或数字组成。有效的中文车牌由七个字符组成：省（1个字符），字母（1个字符），字母+数字（5个字符）。“ 0_0_22_27_27_33_16”是每个字符的索引。这三个数组定义如下。每个数组的最后一个字符是字母O，而不是数字0。我们将O用作“无字符”的符号，因为中文车牌字符中没有O。因此以上车牌拼起来即为 皖AY339S</p>
+</li>
+<li>
+<p dir="auto">37：牌照区域的亮度。 37 (37%)</p>
+</li>
+<li>
+<p dir="auto">15：车牌区域的模糊度。15 (15%)</p>
+</li>
+</ul>
+<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="provinces = [&quot;皖&quot;, &quot;沪&quot;, &quot;津&quot;, &quot;渝&quot;, &quot;冀&quot;, &quot;晋&quot;, &quot;蒙&quot;, &quot;辽&quot;, &quot;吉&quot;, &quot;黑&quot;, &quot;苏&quot;, &quot;浙&quot;, &quot;京&quot;, &quot;闽&quot;, &quot;赣&quot;, &quot;鲁&quot;, &quot;豫&quot;, &quot;鄂&quot;, &quot;湘&quot;, &quot;粤&quot;, &quot;桂&quot;, &quot;琼&quot;, &quot;川&quot;, &quot;贵&quot;, &quot;云&quot;, &quot;藏&quot;, &quot;陕&quot;, &quot;甘&quot;, &quot;青&quot;, &quot;宁&quot;, &quot;新&quot;, &quot;警&quot;, &quot;学&quot;, &quot;O&quot;]
+alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W','X', 'Y', 'Z', 'O']
+ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X','Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']"><pre><span class="pl-s1">provinces</span> <span class="pl-c1">=</span> [<span class="pl-s">"皖"</span>, <span class="pl-s">"沪"</span>, <span class="pl-s">"津"</span>, <span class="pl-s">"渝"</span>, <span class="pl-s">"冀"</span>, <span class="pl-s">"晋"</span>, <span class="pl-s">"蒙"</span>, <span class="pl-s">"辽"</span>, <span class="pl-s">"吉"</span>, <span class="pl-s">"黑"</span>, <span class="pl-s">"苏"</span>, <span class="pl-s">"浙"</span>, <span class="pl-s">"京"</span>, <span class="pl-s">"闽"</span>, <span class="pl-s">"赣"</span>, <span class="pl-s">"鲁"</span>, <span class="pl-s">"豫"</span>, <span class="pl-s">"鄂"</span>, <span class="pl-s">"湘"</span>, <span class="pl-s">"粤"</span>, <span class="pl-s">"桂"</span>, <span class="pl-s">"琼"</span>, <span class="pl-s">"川"</span>, <span class="pl-s">"贵"</span>, <span class="pl-s">"云"</span>, <span class="pl-s">"藏"</span>, <span class="pl-s">"陕"</span>, <span class="pl-s">"甘"</span>, <span class="pl-s">"青"</span>, <span class="pl-s">"宁"</span>, <span class="pl-s">"新"</span>, <span class="pl-s">"警"</span>, <span class="pl-s">"学"</span>, <span class="pl-s">"O"</span>]
+<span class="pl-s1">alphabets</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>,<span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'O'</span>]
+<span class="pl-s1">ads</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>,<span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'0'</span>, <span class="pl-s">'1'</span>, <span class="pl-s">'2'</span>, <span class="pl-s">'3'</span>, <span class="pl-s">'4'</span>, <span class="pl-s">'5'</span>, <span class="pl-s">'6'</span>, <span class="pl-s">'7'</span>, <span class="pl-s">'8'</span>, <span class="pl-s">'9'</span>, <span class="pl-s">'O'</span>]</pre></div>
+<h3 dir="auto"><a id="user-content-32-制作符合pp-ocr训练格式的标注文件" class="anchor" aria-hidden="true" href="#32-制作符合pp-ocr训练格式的标注文件"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3.2 制作符合PP-OCR训练格式的标注文件</h3>
+<p dir="auto">在开始训练之前，可使用如下代码制作符合PP-OCR训练格式的标注文件。</p>
+<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="import cv2
+import os
+import json
+from tqdm import tqdm
+import numpy as np
+
+provinces = [&quot;皖&quot;, &quot;沪&quot;, &quot;津&quot;, &quot;渝&quot;, &quot;冀&quot;, &quot;晋&quot;, &quot;蒙&quot;, &quot;辽&quot;, &quot;吉&quot;, &quot;黑&quot;, &quot;苏&quot;, &quot;浙&quot;, &quot;京&quot;, &quot;闽&quot;, &quot;赣&quot;, &quot;鲁&quot;, &quot;豫&quot;, &quot;鄂&quot;, &quot;湘&quot;, &quot;粤&quot;, &quot;桂&quot;, &quot;琼&quot;, &quot;川&quot;, &quot;贵&quot;, &quot;云&quot;, &quot;藏&quot;, &quot;陕&quot;, &quot;甘&quot;, &quot;青&quot;, &quot;宁&quot;, &quot;新&quot;, &quot;警&quot;, &quot;学&quot;, &quot;O&quot;]
+alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'O']
+ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']
+
+def make_label(img_dir, save_gt_folder, phase):
+    crop_img_save_dir = os.path.join(save_gt_folder, phase, 'crop_imgs')
+    os.makedirs(crop_img_save_dir, exist_ok=True)
+
+    f_det = open(os.path.join(save_gt_folder, phase, 'det.txt'), 'w', encoding='utf-8')
+    f_rec = open(os.path.join(save_gt_folder, phase, 'rec.txt'), 'w', encoding='utf-8')
+
+    i = 0
+    for filename in tqdm(os.listdir(os.path.join(img_dir, phase))):
+        str_list = filename.split('-')
+        if len(str_list) &lt; 5:
+            continue
+        coord_list = str_list[3].split('_')
+        txt_list = str_list[4].split('_')
+        boxes = []
+        for coord in coord_list:
+            boxes.append([int(x) for x in coord.split(&quot;&amp;&quot;)])
+        boxes = [boxes[2], boxes[3], boxes[0], boxes[1]]
+        lp_number = provinces[int(txt_list[0])] + alphabets[int(txt_list[1])] + ''.join([ads[int(x)] for x in txt_list[2:]])
+
+        # det
+        det_info = [{'points':boxes, 'transcription':lp_number}]
+        f_det.write('{}\t{}\n'.format(os.path.join(phase, filename), json.dumps(det_info, ensure_ascii=False)))
+
+        # rec
+        boxes = np.float32(boxes)
+        img = cv2.imread(os.path.join(img_dir, phase, filename))
+        # crop_img = img[int(boxes[:,1].min()):int(boxes[:,1].max()),int(boxes[:,0].min()):int(boxes[:,0].max())]
+        crop_img = get_rotate_crop_image(img, boxes)
+        crop_img_save_filename = '{}_{}.jpg'.format(i,'_'.join(txt_list))
+        crop_img_save_path = os.path.join(crop_img_save_dir, crop_img_save_filename)
+        cv2.imwrite(crop_img_save_path, crop_img)
+        f_rec.write('{}/crop_imgs/{}\t{}\n'.format(phase, crop_img_save_filename, lp_number))
+        i+=1
+    f_det.close()
+    f_rec.close()
+
+def get_rotate_crop_image(img, points):
+    '''
+    img_height, img_width = img.shape[0:2]
+    left = int(np.min(points[:, 0]))
+    right = int(np.max(points[:, 0]))
+    top = int(np.min(points[:, 1]))
+    bottom = int(np.max(points[:, 1]))
+    img_crop = img[top:bottom, left:right, :].copy()
+    points[:, 0] = points[:, 0] - left
+    points[:, 1] = points[:, 1] - top
+    '''
+    assert len(points) == 4, &quot;shape of points must be 4*2&quot;
+    img_crop_width = int(
+        max(
+            np.linalg.norm(points[0] - points[1]),
+            np.linalg.norm(points[2] - points[3])))
+    img_crop_height = int(
+        max(
+            np.linalg.norm(points[0] - points[3]),
+            np.linalg.norm(points[1] - points[2])))
+    pts_std = np.float32([[0, 0], [img_crop_width, 0],
+                          [img_crop_width, img_crop_height],
+                          [0, img_crop_height]])
+    M = cv2.getPerspectiveTransform(points, pts_std)
+    dst_img = cv2.warpPerspective(
+        img,
+        M, (img_crop_width, img_crop_height),
+        borderMode=cv2.BORDER_REPLICATE,
+        flags=cv2.INTER_CUBIC)
+    dst_img_height, dst_img_width = dst_img.shape[0:2]
+    if dst_img_height * 1.0 / dst_img_width &gt;= 1.5:
+        dst_img = np.rot90(dst_img)
+    return dst_img
+
+img_dir = '/home/aistudio/data/CCPD2020/ccpd_green'
+save_gt_folder = '/home/aistudio/data/CCPD2020/PPOCR'
+# phase = 'train' # change to val and test to make val dataset and test dataset
+for phase in ['train','val','test']:
+    make_label(img_dir, save_gt_folder, phase)"><pre><span class="pl-k">import</span> <span class="pl-s1">cv2</span>
+<span class="pl-k">import</span> <span class="pl-s1">os</span>
+<span class="pl-k">import</span> <span class="pl-s1">json</span>
+<span class="pl-k">from</span> <span class="pl-s1">tqdm</span> <span class="pl-k">import</span> <span class="pl-s1">tqdm</span>
+<span class="pl-k">import</span> <span class="pl-s1">numpy</span> <span class="pl-k">as</span> <span class="pl-s1">np</span>
+
+<span class="pl-s1">provinces</span> <span class="pl-c1">=</span> [<span class="pl-s">"皖"</span>, <span class="pl-s">"沪"</span>, <span class="pl-s">"津"</span>, <span class="pl-s">"渝"</span>, <span class="pl-s">"冀"</span>, <span class="pl-s">"晋"</span>, <span class="pl-s">"蒙"</span>, <span class="pl-s">"辽"</span>, <span class="pl-s">"吉"</span>, <span class="pl-s">"黑"</span>, <span class="pl-s">"苏"</span>, <span class="pl-s">"浙"</span>, <span class="pl-s">"京"</span>, <span class="pl-s">"闽"</span>, <span class="pl-s">"赣"</span>, <span class="pl-s">"鲁"</span>, <span class="pl-s">"豫"</span>, <span class="pl-s">"鄂"</span>, <span class="pl-s">"湘"</span>, <span class="pl-s">"粤"</span>, <span class="pl-s">"桂"</span>, <span class="pl-s">"琼"</span>, <span class="pl-s">"川"</span>, <span class="pl-s">"贵"</span>, <span class="pl-s">"云"</span>, <span class="pl-s">"藏"</span>, <span class="pl-s">"陕"</span>, <span class="pl-s">"甘"</span>, <span class="pl-s">"青"</span>, <span class="pl-s">"宁"</span>, <span class="pl-s">"新"</span>, <span class="pl-s">"警"</span>, <span class="pl-s">"学"</span>, <span class="pl-s">"O"</span>]
+<span class="pl-s1">alphabets</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'O'</span>]
+<span class="pl-s1">ads</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'0'</span>, <span class="pl-s">'1'</span>, <span class="pl-s">'2'</span>, <span class="pl-s">'3'</span>, <span class="pl-s">'4'</span>, <span class="pl-s">'5'</span>, <span class="pl-s">'6'</span>, <span class="pl-s">'7'</span>, <span class="pl-s">'8'</span>, <span class="pl-s">'9'</span>, <span class="pl-s">'O'</span>]
+
+<span class="pl-k">def</span> <span class="pl-en">make_label</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>):
+    <span class="pl-s1">crop_img_save_dir</span> <span class="pl-c1">=</span> <span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'crop_imgs'</span>)
+    <span class="pl-s1">os</span>.<span class="pl-en">makedirs</span>(<span class="pl-s1">crop_img_save_dir</span>, <span class="pl-s1">exist_ok</span><span class="pl-c1">=</span><span class="pl-c1">True</span>)
+
+    <span class="pl-s1">f_det</span> <span class="pl-c1">=</span> <span class="pl-en">open</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'det.txt'</span>), <span class="pl-s">'w'</span>, <span class="pl-s1">encoding</span><span class="pl-c1">=</span><span class="pl-s">'utf-8'</span>)
+    <span class="pl-s1">f_rec</span> <span class="pl-c1">=</span> <span class="pl-en">open</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'rec.txt'</span>), <span class="pl-s">'w'</span>, <span class="pl-s1">encoding</span><span class="pl-c1">=</span><span class="pl-s">'utf-8'</span>)
+
+    <span class="pl-s1">i</span> <span class="pl-c1">=</span> <span class="pl-c1">0</span>
+    <span class="pl-k">for</span> <span class="pl-s1">filename</span> <span class="pl-c1">in</span> <span class="pl-en">tqdm</span>(<span class="pl-s1">os</span>.<span class="pl-en">listdir</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">phase</span>))):
+        <span class="pl-s1">str_list</span> <span class="pl-c1">=</span> <span class="pl-s1">filename</span>.<span class="pl-en">split</span>(<span class="pl-s">'-'</span>)
+        <span class="pl-k">if</span> <span class="pl-en">len</span>(<span class="pl-s1">str_list</span>) <span class="pl-c1">&lt;</span> <span class="pl-c1">5</span>:
+            <span class="pl-k">continue</span>
+        <span class="pl-s1">coord_list</span> <span class="pl-c1">=</span> <span class="pl-s1">str_list</span>[<span class="pl-c1">3</span>].<span class="pl-en">split</span>(<span class="pl-s">'_'</span>)
+        <span class="pl-s1">txt_list</span> <span class="pl-c1">=</span> <span class="pl-s1">str_list</span>[<span class="pl-c1">4</span>].<span class="pl-en">split</span>(<span class="pl-s">'_'</span>)
+        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> []
+        <span class="pl-k">for</span> <span class="pl-s1">coord</span> <span class="pl-c1">in</span> <span class="pl-s1">coord_list</span>:
+            <span class="pl-s1">boxes</span>.<span class="pl-en">append</span>([<span class="pl-en">int</span>(<span class="pl-s1">x</span>) <span class="pl-k">for</span> <span class="pl-s1">x</span> <span class="pl-c1">in</span> <span class="pl-s1">coord</span>.<span class="pl-en">split</span>(<span class="pl-s">"&amp;"</span>)])
+        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> [<span class="pl-s1">boxes</span>[<span class="pl-c1">2</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">3</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">0</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">1</span>]]
+        <span class="pl-s1">lp_number</span> <span class="pl-c1">=</span> <span class="pl-s1">provinces</span>[<span class="pl-en">int</span>(<span class="pl-s1">txt_list</span>[<span class="pl-c1">0</span>])] <span class="pl-c1">+</span> <span class="pl-s1">alphabets</span>[<span class="pl-en">int</span>(<span class="pl-s1">txt_list</span>[<span class="pl-c1">1</span>])] <span class="pl-c1">+</span> <span class="pl-s">''</span>.<span class="pl-en">join</span>([<span class="pl-s1">ads</span>[<span class="pl-en">int</span>(<span class="pl-s1">x</span>)] <span class="pl-k">for</span> <span class="pl-s1">x</span> <span class="pl-c1">in</span> <span class="pl-s1">txt_list</span>[<span class="pl-c1">2</span>:]])
+
+        <span class="pl-c"># det</span>
+        <span class="pl-s1">det_info</span> <span class="pl-c1">=</span> [{<span class="pl-s">'points'</span>:<span class="pl-s1">boxes</span>, <span class="pl-s">'transcription'</span>:<span class="pl-s1">lp_number</span>}]
+        <span class="pl-s1">f_det</span>.<span class="pl-en">write</span>(<span class="pl-s">'{}<span class="pl-cce">\t</span>{}<span class="pl-cce">\n</span>'</span>.<span class="pl-en">format</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">phase</span>, <span class="pl-s1">filename</span>), <span class="pl-s1">json</span>.<span class="pl-en">dumps</span>(<span class="pl-s1">det_info</span>, <span class="pl-s1">ensure_ascii</span><span class="pl-c1">=</span><span class="pl-c1">False</span>)))
+
+        <span class="pl-c"># rec</span>
+        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">float32</span>(<span class="pl-s1">boxes</span>)
+        <span class="pl-s1">img</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">imread</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">phase</span>, <span class="pl-s1">filename</span>))
+        <span class="pl-c"># crop_img = img[int(boxes[:,1].min()):int(boxes[:,1].max()),int(boxes[:,0].min()):int(boxes[:,0].max())]</span>
+        <span class="pl-s1">crop_img</span> <span class="pl-c1">=</span> <span class="pl-en">get_rotate_crop_image</span>(<span class="pl-s1">img</span>, <span class="pl-s1">boxes</span>)
+        <span class="pl-s1">crop_img_save_filename</span> <span class="pl-c1">=</span> <span class="pl-s">'{}_{}.jpg'</span>.<span class="pl-en">format</span>(<span class="pl-s1">i</span>,<span class="pl-s">'_'</span>.<span class="pl-en">join</span>(<span class="pl-s1">txt_list</span>))
+        <span class="pl-s1">crop_img_save_path</span> <span class="pl-c1">=</span> <span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">crop_img_save_dir</span>, <span class="pl-s1">crop_img_save_filename</span>)
+        <span class="pl-s1">cv2</span>.<span class="pl-en">imwrite</span>(<span class="pl-s1">crop_img_save_path</span>, <span class="pl-s1">crop_img</span>)
+        <span class="pl-s1">f_rec</span>.<span class="pl-en">write</span>(<span class="pl-s">'{}/crop_imgs/{}<span class="pl-cce">\t</span>{}<span class="pl-cce">\n</span>'</span>.<span class="pl-en">format</span>(<span class="pl-s1">phase</span>, <span class="pl-s1">crop_img_save_filename</span>, <span class="pl-s1">lp_number</span>))
+        <span class="pl-s1">i</span><span class="pl-c1">+=</span><span class="pl-c1">1</span>
+    <span class="pl-s1">f_det</span>.<span class="pl-en">close</span>()
+    <span class="pl-s1">f_rec</span>.<span class="pl-en">close</span>()
+
+<span class="pl-k">def</span> <span class="pl-en">get_rotate_crop_image</span>(<span class="pl-s1">img</span>, <span class="pl-s1">points</span>):
+    <span class="pl-s">'''</span>
+<span class="pl-s">    img_height, img_width = img.shape[0:2]</span>
+<span class="pl-s">    left = int(np.min(points[:, 0]))</span>
+<span class="pl-s">    right = int(np.max(points[:, 0]))</span>
+<span class="pl-s">    top = int(np.min(points[:, 1]))</span>
+<span class="pl-s">    bottom = int(np.max(points[:, 1]))</span>
+<span class="pl-s">    img_crop = img[top:bottom, left:right, :].copy()</span>
+<span class="pl-s">    points[:, 0] = points[:, 0] - left</span>
+<span class="pl-s">    points[:, 1] = points[:, 1] - top</span>
+<span class="pl-s">    '''</span>
+    <span class="pl-k">assert</span> <span class="pl-en">len</span>(<span class="pl-s1">points</span>) <span class="pl-c1">==</span> <span class="pl-c1">4</span>, <span class="pl-s">"shape of points must be 4*2"</span>
+    <span class="pl-s1">img_crop_width</span> <span class="pl-c1">=</span> <span class="pl-en">int</span>(
+        <span class="pl-en">max</span>(
+            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">0</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">1</span>]),
+            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">2</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">3</span>])))
+    <span class="pl-s1">img_crop_height</span> <span class="pl-c1">=</span> <span class="pl-en">int</span>(
+        <span class="pl-en">max</span>(
+            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">0</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">3</span>]),
+            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">1</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">2</span>])))
+    <span class="pl-s1">pts_std</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">float32</span>([[<span class="pl-c1">0</span>, <span class="pl-c1">0</span>], [<span class="pl-s1">img_crop_width</span>, <span class="pl-c1">0</span>],
+                          [<span class="pl-s1">img_crop_width</span>, <span class="pl-s1">img_crop_height</span>],
+                          [<span class="pl-c1">0</span>, <span class="pl-s1">img_crop_height</span>]])
+    <span class="pl-v">M</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">getPerspectiveTransform</span>(<span class="pl-s1">points</span>, <span class="pl-s1">pts_std</span>)
+    <span class="pl-s1">dst_img</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">warpPerspective</span>(
+        <span class="pl-s1">img</span>,
+        <span class="pl-v">M</span>, (<span class="pl-s1">img_crop_width</span>, <span class="pl-s1">img_crop_height</span>),
+        <span class="pl-s1">borderMode</span><span class="pl-c1">=</span><span class="pl-s1">cv2</span>.<span class="pl-v">BORDER_REPLICATE</span>,
+        <span class="pl-s1">flags</span><span class="pl-c1">=</span><span class="pl-s1">cv2</span>.<span class="pl-v">INTER_CUBIC</span>)
+    <span class="pl-s1">dst_img_height</span>, <span class="pl-s1">dst_img_width</span> <span class="pl-c1">=</span> <span class="pl-s1">dst_img</span>.<span class="pl-s1">shape</span>[<span class="pl-c1">0</span>:<span class="pl-c1">2</span>]
+    <span class="pl-k">if</span> <span class="pl-s1">dst_img_height</span> <span class="pl-c1">*</span> <span class="pl-c1">1.0</span> <span class="pl-c1">/</span> <span class="pl-s1">dst_img_width</span> <span class="pl-c1">&gt;=</span> <span class="pl-c1">1.5</span>:
+        <span class="pl-s1">dst_img</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">rot90</span>(<span class="pl-s1">dst_img</span>)
+    <span class="pl-k">return</span> <span class="pl-s1">dst_img</span>
+
+<span class="pl-s1">img_dir</span> <span class="pl-c1">=</span> <span class="pl-s">'/home/aistudio/data/CCPD2020/ccpd_green'</span>
+<span class="pl-s1">save_gt_folder</span> <span class="pl-c1">=</span> <span class="pl-s">'/home/aistudio/data/CCPD2020/PPOCR'</span>
+<span class="pl-c"># phase = 'train' # change to val and test to make val dataset and test dataset</span>
+<span class="pl-k">for</span> <span class="pl-s1">phase</span> <span class="pl-c1">in</span> [<span class="pl-s">'train'</span>,<span class="pl-s">'val'</span>,<span class="pl-s">'test'</span>]:
+    <span class="pl-en">make_label</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>)</pre></div>
+<p dir="auto">通过上述命令可以完成了<code>训练集</code>，<code>验证集</code>和<code>测试集</code>的制作，制作完成的数据集信息如下：</p>
+<table>
+<thead>
+<tr>
+<th>类型</th>
+<th>数据集</th>
+<th>图片地址</th>
+<th>标签地址</th>
+<th>图片数量</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>检测</td>
+<td>训练集</td>
+<td>/home/aistudio/data/CCPD2020/ccpd_green/train</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/train/det.txt</td>
+<td>5769</td>
+</tr>
+<tr>
+<td>检测</td>
+<td>验证集</td>
+<td>/home/aistudio/data/CCPD2020/ccpd_green/val</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/val/det.txt</td>
+<td>1001</td>
+</tr>
+<tr>
+<td>检测</td>
+<td>测试集</td>
+<td>/home/aistudio/data/CCPD2020/ccpd_green/test</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/test/det.txt</td>
+<td>5006</td>
+</tr>
+<tr>
+<td>识别</td>
+<td>训练集</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/train/crop_imgs</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt</td>
+<td>5769</td>
+</tr>
+<tr>
+<td>识别</td>
+<td>验证集</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/val/crop_imgs</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/val/rec.txt</td>
+<td>1001</td>
+</tr>
+<tr>
+<td>识别</td>
+<td>测试集</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs</td>
+<td>/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt</td>
+<td>5006</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">在普遍的深度学习流程中，都是在训练集训练，在验证集选择最优模型后在测试集上进行测试。在本例中，我们省略中间步骤，直接在训练集训练，在测试集选择最优模型，因此我们只使用训练集和测试集。</p>
+<h2 dir="auto"><a id="user-content-4-实验" class="anchor" aria-hidden="true" href="#4-实验"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4. 实验</h2>
+<p dir="auto">由于数据集比较少，为了模型更好和更快的收敛，这里选用 PaddleOCR 中的 PP-OCRv3 模型进行文本检测和识别，并且使用 PP-OCRv3 模型参数作为预训练模型。PP-OCRv3在PP-OCRv2的基础上，中文场景端到端Hmean指标相比于PP-OCRv2提升5%, 英文数字模型端到端效果提升11%。详细优化细节请参考<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md">PP-OCRv3</a>技术报告。</p>
+<p dir="auto">由于车牌场景均为端侧设备部署，因此对速度和模型大小有比较高的要求，因此还需要采用量化训练的方式进行模型大小的压缩和模型推理速度的加速。模型量化可以在基本不损失模型的精度的情况下，将FP32精度的模型参数转换为Int8精度，减小模型参数大小并加速计算，使用量化后的模型在移动端等部署时更具备速度优势。</p>
+<p dir="auto">因此，本实验中对于车牌检测和识别有如下3种方案：</p>
+<ol dir="auto">
+<li>PP-OCRv3中英文超轻量预训练模型直接预测</li>
+<li>CCPD车牌数据集在PP-OCRv3模型上fine-tune</li>
+<li>CCPD车牌数据集在PP-OCRv3模型上fine-tune后量化</li>
+</ol>
+<h3 dir="auto"><a id="user-content-41-检测" class="anchor" aria-hidden="true" href="#41-检测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1 检测</h3>
+<h4 dir="auto"><a id="user-content-411-预训练模型直接预测" class="anchor" aria-hidden="true" href="#411-预训练模型直接预测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.1 预训练模型直接预测</h4>
+<p dir="auto">从下表中下载PP-OCRv3文本检测预训练模型</p>
+<table>
+<thead>
+<tr>
+<th>模型名称</th>
+<th>模型简介</th>
+<th>配置文件</th>
+<th>推理模型大小</th>
+<th>下载地址</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>ch_PP-OCRv3_det</td>
+<td>【最新】原始超轻量模型，支持中英文、多语种文本检测</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml">ch_PP-OCRv3_det_cml.yml</a></td>
+<td>3.8M</td>
+<td><a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar" rel="nofollow">推理模型</a> / <a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar" rel="nofollow">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">使用如下命令下载预训练模型</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="mkdir models
+cd models
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar -xf ch_PP-OCRv3_det_distill_train.tar
+cd /home/aistudio/PaddleOCR"><pre>mkdir models
+<span class="pl-c1">cd</span> models
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar -xf ch_PP-OCRv3_det_distill_train.tar
+<span class="pl-c1">cd</span> /home/aistudio/PaddleOCR</pre></div>
+<p dir="auto">预训练模型下载完成后，我们使用<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml">ch_PP-OCRv3_det_student.yml</a> 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：</p>
+<ol dir="auto">
+<li>模型存储和训练相关:
+<ol dir="auto">
+<li>Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址</li>
+</ol>
+</li>
+<li>数据集相关
+<ol dir="auto">
+<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
+<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
+</ol>
+</li>
+</ol>
+<p dir="auto">上述字段均为必须修改的字段，可以通过修改配置文件的方式改动，也可在不需要修改配置文件的情况下，改变训练的参数。这里使用不改变配置文件的方式 。使用如下命令进行PP-OCRv3文本检测预训练模型的评估</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
+<p dir="auto">上述指令中，通过-c 选择训练使用配置文件，通过-o参数在不需要修改配置文件的情况下，改变训练的参数。</p>
+<p dir="auto">使用预训练模型进行评估，指标如下所示：</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>hmeans</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
+<td>76.12%</td>
+</tr>
+</tbody>
+</table>
+<h4 dir="auto"><a id="user-content-412-ccpd车牌数据集fine-tune" class="anchor" aria-hidden="true" href="#412-ccpd车牌数据集fine-tune"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.2 CCPD车牌数据集fine-tune</h4>
+<p dir="auto"><strong>训练</strong></p>
+<p dir="auto">为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:</p>
+<ol dir="auto">
+<li>模型存储和训练相关:
+<ol dir="auto">
+<li>Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址</li>
+<li>Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔772个step评估一次，772为一个epoch总的step数。</li>
+</ol>
+</li>
+<li>优化器相关:
+<ol dir="auto">
+<li>Optimizer.lr.name: 学习率衰减器设为常量 Const</li>
+<li>Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍</li>
+<li>Optimizer.lr.warmup_epoch: warmup_epoch设为0</li>
+</ol>
+</li>
+<li>数据集相关:
+<ol dir="auto">
+<li>Train.dataset.data_dir：指向训练集图片存放目录</li>
+<li>Train.dataset.label_file_list：指向训练集标注文件</li>
+<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
+<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
+</ol>
+</li>
+</ol>
+<p dir="auto">使用如下代码即可启动在CCPD车牌数据集上的fine-tune。</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
+    Global.save_model_dir=output/CCPD/det \
+    Global.eval_batch_step=&quot;[0, 772]&quot; \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
+    Global.save_model_dir=output/CCPD/det \
+    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 772]<span class="pl-pds">"</span></span> \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
+<p dir="auto">在上述命令中，通过<code>-o</code>的方式修改了配置文件中的参数。</p>
+<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_finetune.tar" rel="nofollow">det_ppocr_v3_finetune.tar</a></p>
+<p dir="auto"><strong>评估</strong></p>
+<p dir="auto">训练完成后使用如下命令进行评估</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
+<p dir="auto">使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>hmeans</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
+<td>76.12%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
+<td>99%</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">可以看到进行fine-tune能显著提升车牌检测的效果。</p>
+<h4 dir="auto"><a id="user-content-413-ccpd车牌数据集fine-tune量化训练" class="anchor" aria-hidden="true" href="#413-ccpd车牌数据集fine-tune量化训练"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.3 CCPD车牌数据集fine-tune+量化训练</h4>
+<p dir="auto">此处采用 PaddleOCR 中提供好的<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/slim/quantization/README.md">量化教程</a>对模型进行量化训练。</p>
+<p dir="auto">量化训练可通过如下命令启动:</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Global.save_model_dir=output/CCPD/det_quant \
+    Global.eval_batch_step=&quot;[0, 772]&quot; \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Global.save_model_dir=output/CCPD/det_quant \
+    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 772]<span class="pl-pds">"</span></span> \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
+<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_quant.tar" rel="nofollow">det_ppocr_v3_quant.tar</a></p>
+<p dir="auto">量化后指标对比如下</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>hmeans</th>
+<th>模型大小</th>
+<th>预测速度(lite)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
+<td>99%</td>
+<td>2.5M</td>
+<td>223ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化</td>
+<td>98.91%</td>
+<td>1M</td>
+<td>189ms</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">可以看到通过量化训练在精度几乎无损的情况下，降低模型体积60%并且推理速度提升15%。</p>
+<p dir="auto">速度测试基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>完成。</p>
+<h4 dir="auto"><a id="user-content-414-模型导出" class="anchor" aria-hidden="true" href="#414-模型导出"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.4 模型导出</h4>
+<p dir="auto">使用如下命令可以将训练好的模型进行导出</p>
+<ul dir="auto">
+<li>非量化模型</li>
+</ul>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Global.save_inference_dir=output/det/infer"><pre>python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Global.save_inference_dir=output/det/infer</pre></div>
+<ul dir="auto">
+<li>量化模型</li>
+</ul>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/det/infer"><pre>python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/det/infer</pre></div>
+<h3 dir="auto"><a id="user-content-42-识别" class="anchor" aria-hidden="true" href="#42-识别"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2 识别</h3>
+<h4 dir="auto"><a id="user-content-421-预训练模型直接预测" class="anchor" aria-hidden="true" href="#421-预训练模型直接预测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.1 预训练模型直接预测</h4>
+<p dir="auto">从下表中下载PP-OCRv3文本识别预训练模型</p>
+<table>
+<thead>
+<tr>
+<th>模型名称</th>
+<th>模型简介</th>
+<th>配置文件</th>
+<th>推理模型大小</th>
+<th>下载地址</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>ch_PP-OCRv3_rec</td>
+<td>【最新】原始超轻量模型，支持中英文、数字识别</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml">ch_PP-OCRv3_rec_distillation.yml</a></td>
+<td>12.4M</td>
+<td><a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar" rel="nofollow">推理模型</a> / <a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar" rel="nofollow">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">使用如下命令下载预训练模型</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="mkdir models
+cd models
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+tar -xf ch_PP-OCRv3_rec_train.tar
+cd /home/aistudio/PaddleOCR"><pre>mkdir models
+<span class="pl-c1">cd</span> models
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+tar -xf ch_PP-OCRv3_rec_train.tar
+<span class="pl-c1">cd</span> /home/aistudio/PaddleOCR</pre></div>
+<p dir="auto">PaddleOCR提供的PP-OCRv3识别模型采用蒸馏训练策略，因此提供的预训练模型中会包含<code>Teacher</code>和<code>Student</code>模型的参数，详细信息可参考<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/knowledge_distillation.md">knowledge_distillation.md</a>。 因此，模型下载完成后需要使用如下代码提取<code>Student</code>模型的参数：</p>
+<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="import paddle
+# 加载预训练模型
+all_params = paddle.load(&quot;models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams&quot;)
+# 查看权重参数的keys
+print(all_params.keys())
+# 学生模型的权重提取
+s_params = {key[len(&quot;Student.&quot;):]: all_params[key] for key in all_params if &quot;Student.&quot; in key}
+# 查看学生模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, &quot;models/ch_PP-OCRv3_rec_train/student.pdparams&quot;)"><pre><span class="pl-k">import</span> <span class="pl-s1">paddle</span>
+<span class="pl-c"># 加载预训练模型</span>
+<span class="pl-s1">all_params</span> <span class="pl-c1">=</span> <span class="pl-s1">paddle</span>.<span class="pl-en">load</span>(<span class="pl-s">"models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams"</span>)
+<span class="pl-c"># 查看权重参数的keys</span>
+<span class="pl-en">print</span>(<span class="pl-s1">all_params</span>.<span class="pl-en">keys</span>())
+<span class="pl-c"># 学生模型的权重提取</span>
+<span class="pl-s1">s_params</span> <span class="pl-c1">=</span> {<span class="pl-s1">key</span>[<span class="pl-en">len</span>(<span class="pl-s">"Student."</span>):]: <span class="pl-s1">all_params</span>[<span class="pl-s1">key</span>] <span class="pl-k">for</span> <span class="pl-s1">key</span> <span class="pl-c1">in</span> <span class="pl-s1">all_params</span> <span class="pl-k">if</span> <span class="pl-s">"Student."</span> <span class="pl-c1">in</span> <span class="pl-s1">key</span>}
+<span class="pl-c"># 查看学生模型权重参数的keys</span>
+<span class="pl-en">print</span>(<span class="pl-s1">s_params</span>.<span class="pl-en">keys</span>())
+<span class="pl-c"># 保存</span>
+<span class="pl-s1">paddle</span>.<span class="pl-en">save</span>(<span class="pl-s1">s_params</span>, <span class="pl-s">"models/ch_PP-OCRv3_rec_train/student.pdparams"</span>)</pre></div>
+<p dir="auto">预训练模型下载完成后，我们使用<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml">ch_PP-OCRv3_rec.yml</a> 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：</p>
+<ol dir="auto">
+<li>模型存储和训练相关:
+<ol dir="auto">
+<li>Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址</li>
+</ol>
+</li>
+<li>数据集相关
+<ol dir="auto">
+<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
+<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
+</ol>
+</li>
+</ol>
+<p dir="auto">使用如下命令进行PP-OCRv3文本识别预训练模型的评估</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
+<p dir="auto">评估部分日志如下：</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
+eval model:: 100%|██████████████████████████████| 40/40 [00:15&lt;00:00,  2.57it/s]
+[2022/05/12 19:52:17] ppocr INFO: metric eval ***************
+[2022/05/12 19:52:17] ppocr INFO: acc:0.0
+[2022/05/12 19:52:17] ppocr INFO: norm_edit_dis:0.8656084923002452
+[2022/05/12 19:52:17] ppocr INFO: Teacher_acc:0.000399520574511545
+[2022/05/12 19:52:17] ppocr INFO: Teacher_norm_edit_dis:0.8657902943394548
+[2022/05/12 19:52:17] ppocr INFO: fps:1443.1801978719905
+"><pre>[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
+<span class="pl-c1">eval</span> model:: 100%<span class="pl-k">|</span>██████████████████████████████<span class="pl-k">|</span> 40/40 [00:<span class="pl-k">15&lt;</span>00:00,  2.57it/s]
+[2022/05/12 19:52:17] ppocr INFO: metric <span class="pl-c1">eval</span> <span class="pl-k">***************</span>
+[2022/05/12 19:52:17] ppocr INFO: acc:0.0
+[2022/05/12 19:52:17] ppocr INFO: norm_edit_dis:0.8656084923002452
+[2022/05/12 19:52:17] ppocr INFO: Teacher_acc:0.000399520574511545
+[2022/05/12 19:52:17] ppocr INFO: Teacher_norm_edit_dis:0.8657902943394548
+[2022/05/12 19:52:17] ppocr INFO: fps:1443.1801978719905
+</pre></div>
+<p dir="auto">使用预训练模型进行评估，指标如下所示：</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>acc</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
+<td>0%</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">从评估日志中可以看到，直接使用PP-OCRv3预训练模型进行评估，acc非常低，但是norm_edit_dis很高。因此，我们猜测是模型大部分文字识别是对的，只有少部分文字识别错误。使用如下命令进行infer查看模型的推理结果进行验证：</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg"><pre>python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg</pre></div>
+<p dir="auto">输出部分日志如下：</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
+W0501 08:51:57.127391 11326 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
+W0501 08:51:57.132315 11326 device_context.cc:465] device: 0, cuDNN Version: 7.6.
+[2022/05/01 08:52:00] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/student
+[2022/05/01 08:52:00] ppocr INFO: infer_img: /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_3_32_30_31_30_30.jpg
+[2022/05/01 08:52:00] ppocr INFO:      result: {&quot;Student&quot;: {&quot;label&quot;: &quot;皖A·D86766&quot;, &quot;score&quot;: 0.9552637934684753}, &quot;Teacher&quot;: {&quot;label&quot;: &quot;皖A·D86766&quot;, &quot;score&quot;: 0.9917094707489014}}
+[2022/05/01 08:52:00] ppocr INFO: success!"><pre>[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
+W0501 08:51:57.127391 11326 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
+W0501 08:51:57.132315 11326 device_context.cc:465] device: 0, cuDNN Version: 7.6.
+[2022/05/01 08:52:00] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/student
+[2022/05/01 08:52:00] ppocr INFO: infer_img: /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_3_32_30_31_30_30.jpg
+[2022/05/01 08:52:00] ppocr INFO:      result: {<span class="pl-s"><span class="pl-pds">"</span>Student<span class="pl-pds">"</span></span>: {<span class="pl-s"><span class="pl-pds">"</span>label<span class="pl-pds">"</span></span>: <span class="pl-s"><span class="pl-pds">"</span>皖A·D86766<span class="pl-pds">"</span></span>, <span class="pl-s"><span class="pl-pds">"</span>score<span class="pl-pds">"</span></span>: 0.9552637934684753}, <span class="pl-s"><span class="pl-pds">"</span>Teacher<span class="pl-pds">"</span></span>: {<span class="pl-s"><span class="pl-pds">"</span>label<span class="pl-pds">"</span></span>: <span class="pl-s"><span class="pl-pds">"</span>皖A·D86766<span class="pl-pds">"</span></span>, <span class="pl-s"><span class="pl-pds">"</span>score<span class="pl-pds">"</span></span>: 0.9917094707489014}}
+[2022/05/01 08:52:00] ppocr INFO: success<span class="pl-k">!</span></pre></div>
+<p dir="auto">从infer结果可以看到，车牌中的文字大部分都识别正确，只是多识别出了一个<code>·</code>。针对这种情况，有如下两种方案：</p>
+<ol dir="auto">
+<li>直接通过后处理去掉多识别的<code>·</code>。</li>
+<li>进行 fine-tune。</li>
+</ol>
+<h4 dir="auto"><a id="user-content-422-预训练模型直接预测改动后处理" class="anchor" aria-hidden="true" href="#422-预训练模型直接预测改动后处理"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.2 预训练模型直接预测+改动后处理</h4>
+<p dir="auto">直接通过后处理去掉多识别的<code>·</code>，在后处理的改动比较简单，只需在 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/postprocess/rec_postprocess.py">ppocr/postprocess/rec_postprocess.py</a> 文件的76行添加如下代码:</p>
+<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="text = text.replace('·','')"><pre><span class="pl-s1">text</span> <span class="pl-c1">=</span> <span class="pl-s1">text</span>.<span class="pl-en">replace</span>(<span class="pl-s">'·'</span>,<span class="pl-s">''</span>)</pre></div>
+<p dir="auto">改动前后指标对比:</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>acc</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
+<td>0.2%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
+<td>90.97%</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">可以看到，去掉多余的<code>·</code>能大幅提高精度。</p>
+<h4 dir="auto"><a id="user-content-423-ccpd车牌数据集fine-tune" class="anchor" aria-hidden="true" href="#423-ccpd车牌数据集fine-tune"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.3 CCPD车牌数据集fine-tune</h4>
+<p dir="auto"><strong>训练</strong></p>
+<p dir="auto">为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:</p>
+<ol dir="auto">
+<li>模型存储和训练相关:
+<ol dir="auto">
+<li>Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址</li>
+<li>Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔45个step评估一次，45为一个epoch总的step数。</li>
+</ol>
+</li>
+<li>优化器相关
+<ol dir="auto">
+<li>Optimizer.lr.name: 学习率衰减器设为常量 Const</li>
+<li>Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍</li>
+<li>Optimizer.lr.warmup_epoch: warmup_epoch设为0</li>
+</ol>
+</li>
+<li>数据集相关
+<ol dir="auto">
+<li>Train.dataset.data_dir：指向训练集图片存放目录</li>
+<li>Train.dataset.label_file_list：指向训练集标注文件</li>
+<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
+<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
+</ol>
+</li>
+</ol>
+<p dir="auto">使用如下命令启动 fine-tune</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Global.save_model_dir=output/CCPD/rec/ \
+    Global.eval_batch_step=&quot;[0, 90]&quot; \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Global.save_model_dir=output/CCPD/rec/ \
+    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 90]<span class="pl-pds">"</span></span> \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
+<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_finetune.tar" rel="nofollow">rec_ppocr_v3_finetune.tar</a></p>
+<p dir="auto"><strong>评估</strong></p>
+<p dir="auto">训练完成后使用如下命令进行评估</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
+<p dir="auto">使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>acc</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
+<td>0%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
+<td>90.97%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
+<td>94.54%</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">可以看到进行fine-tune能显著提升车牌识别的效果。</p>
+<h4 dir="auto"><a id="user-content-424-ccpd车牌数据集fine-tune量化训练" class="anchor" aria-hidden="true" href="#424-ccpd车牌数据集fine-tune量化训练"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.4 CCPD车牌数据集fine-tune+量化训练</h4>
+<p dir="auto">此处采用 PaddleOCR 中提供好的<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/slim/quantization/README.md">量化教程</a>对模型进行量化训练。</p>
+<p dir="auto">量化训练可通过如下命令启动:</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Global.save_model_dir=output/CCPD/rec_quant/ \
+    Global.eval_batch_step=&quot;[0, 90]&quot; \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Global.save_model_dir=output/CCPD/rec_quant/ \
+    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 90]<span class="pl-pds">"</span></span> \
+    Optimizer.lr.name=Const \
+    Optimizer.lr.learning_rate=0.0005 \
+    Optimizer.lr.warmup_epoch=0 \
+    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
+<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_quant.tar" rel="nofollow">rec_ppocr_v3_quant.tar</a></p>
+<p dir="auto">量化后指标对比如下</p>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>acc</th>
+<th>模型大小</th>
+<th>预测速度(lite)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
+<td>94.54%</td>
+<td>10.3M</td>
+<td>4.2ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化</td>
+<td>93.4%</td>
+<td>4.8M</td>
+<td>1.8ms</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">可以看到量化后能降低模型体积53%并且推理速度提升57%，但是由于识别数据过少，量化带来了1%的精度下降。</p>
+<p dir="auto">速度测试基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>完成。</p>
+<h4 dir="auto"><a id="user-content-425-模型导出" class="anchor" aria-hidden="true" href="#425-模型导出"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.5 模型导出</h4>
+<p dir="auto">使用如下命令可以将训练好的模型进行导出。</p>
+<ul dir="auto">
+<li>非量化模型</li>
+</ul>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec/infer"><pre>python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec/infer</pre></div>
+<ul dir="auto">
+<li>量化模型</li>
+</ul>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec_quant/infer"><pre>python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec_quant/infer</pre></div>
+<h3 dir="auto"><a id="user-content-43-计算end2end指标" class="anchor" aria-hidden="true" href="#43-计算end2end指标"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.3 计算End2End指标</h3>
+<p dir="auto">端到端指标可通过 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/tools/end2end/readme.md">PaddleOCR内置脚本</a> 进行计算，具体步骤如下：</p>
+<ol dir="auto">
+<li>导出模型</li>
+</ol>
+<p dir="auto">通过如下命令进行模型的导出。注意，量化模型导出时，需要配置eval数据集</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="# 检测模型
+
+# 预训练模型
+python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
+    Global.save_inference_dir=output/ch_PP-OCRv3_det_distill_train/infer
+
+# 非量化模型
+python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/det/infer
+
+# 量化模型
+python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/det_quant/infer \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt] \
+    Eval.loader.num_workers=0
+
+# 识别模型
+
+# 预训练模型
+python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Global.save_inference_dir=output/ch_PP-OCRv3_rec_train/infer
+
+# 非量化模型
+python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec/infer
+
+# 量化模型
+python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec_quant/infer \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre><span class="pl-c"><span class="pl-c">#</span> 检测模型</span>
+
+<span class="pl-c"><span class="pl-c">#</span> 预训练模型</span>
+python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
+    Global.save_inference_dir=output/ch_PP-OCRv3_det_distill_train/infer
+
+<span class="pl-c"><span class="pl-c">#</span> 非量化模型</span>
+python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/det/infer
+
+<span class="pl-c"><span class="pl-c">#</span> 量化模型</span>
+python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/det_quant/infer \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt] \
+    Eval.loader.num_workers=0
+
+<span class="pl-c"><span class="pl-c">#</span> 识别模型</span>
+
+<span class="pl-c"><span class="pl-c">#</span> 预训练模型</span>
+python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
+    Global.save_inference_dir=output/ch_PP-OCRv3_rec_train/infer
+
+<span class="pl-c"><span class="pl-c">#</span> 非量化模型</span>
+python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec/infer
+
+<span class="pl-c"><span class="pl-c">#</span> 量化模型</span>
+python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
+    Global.save_inference_dir=output/CCPD/rec_quant/infer \
+    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
+<ol start="2" dir="auto">
+<li>用导出的模型对测试集进行预测</li>
+</ol>
+<p dir="auto">此处，分别使用PP-OCRv3预训练模型，fintune模型和量化模型对测试集的所有图像进行预测，命令如下：</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型
+python3 tools/infer/predict_system.py --det_model_dir=models/ch_PP-OCRv3_det_distill_train/infer --rec_model_dir=models/ch_PP-OCRv3_rec_train/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/pretrain --use_dilation=true
+
+# PP-OCRv3中英文超轻量检测预训练模型+fine-tune，PP-OCRv3中英文超轻量识别预训练模型+fine-tune
+python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det/infer --rec_model_dir=output/CCPD/rec/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/fine-tune --use_dilation=true
+
+# PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估
+python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true"><pre><span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型</span>
+python3 tools/infer/predict_system.py --det_model_dir=models/ch_PP-OCRv3_det_distill_train/infer --rec_model_dir=models/ch_PP-OCRv3_rec_train/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/pretrain --use_dilation=true
+
+<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型+fine-tune，PP-OCRv3中英文超轻量识别预训练模型+fine-tune</span>
+python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det/infer --rec_model_dir=output/CCPD/rec/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/fine-tune --use_dilation=true
+
+<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估</span>
+python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true</pre></div>
+<ol start="3" dir="auto">
+<li>转换label并计算指标</li>
+</ol>
+<p dir="auto">将gt和上一步保存的预测结果转换为端对端评测需要的数据格式，并根据转换后的数据进行端到端指标计算</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
+
+# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型 结果转换和评估
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/pretrain/system_results.txt --save_folder=end2end/pretrain
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/pretrain
+
+# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型+后处理去掉多识别的`·` 结果转换和评估
+# 需手动修改后处理函数
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/post/system_results.txt --save_folder=end2end/post
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/post
+
+# PP-OCRv3中英文超轻量检测预训练模型 fine-tune，PP-OCRv3中英文超轻量识别预训练模型 fine-tune 结果转换和评估
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/fine-tune/system_results.txt --save_folder=end2end/fine-tune
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/fine-tune
+
+# PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/quant/system_results.txt --save_folder=end2end/quant
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant"><pre>python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
+
+<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型 结果转换和评估</span>
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/pretrain/system_results.txt --save_folder=end2end/pretrain
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/pretrain
+
+<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型+后处理去掉多识别的`·` 结果转换和评估</span>
+<span class="pl-c"><span class="pl-c">#</span> 需手动修改后处理函数</span>
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/post/system_results.txt --save_folder=end2end/post
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/post
+
+<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune，PP-OCRv3中英文超轻量识别预训练模型 fine-tune 结果转换和评估</span>
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/fine-tune/system_results.txt --save_folder=end2end/fine-tune
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/fine-tune
+
+<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估</span>
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/quant/system_results.txt --save_folder=end2end/quant
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant</pre></div>
+<p dir="auto">日志如下:</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="The convert label saved in end2end/gt
+The convert label saved in end2end/pretrain
+start testing...
+hit, dt_count, gt_count 2 5988 5006
+character_acc: 70.42%
+avg_edit_dist_field: 2.37
+avg_edit_dist_img: 2.37
+precision: 0.03%
+recall: 0.04%
+fmeasure: 0.04%
+The convert label saved in end2end/post
+start testing...
+hit, dt_count, gt_count 4224 5988 5006
+character_acc: 81.59%
+avg_edit_dist_field: 1.47
+avg_edit_dist_img: 1.47
+precision: 70.54%
+recall: 84.38%
+fmeasure: 76.84%
+The convert label saved in end2end/fine-tune
+start testing...
+hit, dt_count, gt_count 4286 4898 5006
+character_acc: 94.16%
+avg_edit_dist_field: 0.47
+avg_edit_dist_img: 0.47
+precision: 87.51%
+recall: 85.62%
+fmeasure: 86.55%
+The convert label saved in end2end/quant
+start testing...
+hit, dt_count, gt_count 4349 4951 5006
+character_acc: 94.13%
+avg_edit_dist_field: 0.47
+avg_edit_dist_img: 0.47
+precision: 87.84%
+recall: 86.88%
+fmeasure: 87.36%"><pre>The convert label saved <span class="pl-k">in</span> end2end/gt
+The convert label saved <span class="pl-k">in</span> end2end/pretrain
+start testing...
+hit, dt_count, gt_count 2 5988 5006
+character_acc: 70.42%
+avg_edit_dist_field: 2.37
+avg_edit_dist_img: 2.37
+precision: 0.03%
+recall: 0.04%
+fmeasure: 0.04%
+The convert label saved <span class="pl-k">in</span> end2end/post
+start testing...
+hit, dt_count, gt_count 4224 5988 5006
+character_acc: 81.59%
+avg_edit_dist_field: 1.47
+avg_edit_dist_img: 1.47
+precision: 70.54%
+recall: 84.38%
+fmeasure: 76.84%
+The convert label saved <span class="pl-k">in</span> end2end/fine-tune
+start testing...
+hit, dt_count, gt_count 4286 4898 5006
+character_acc: 94.16%
+avg_edit_dist_field: 0.47
+avg_edit_dist_img: 0.47
+precision: 87.51%
+recall: 85.62%
+fmeasure: 86.55%
+The convert label saved <span class="pl-k">in</span> end2end/quant
+start testing...
+hit, dt_count, gt_count 4349 4951 5006
+character_acc: 94.13%
+avg_edit_dist_field: 0.47
+avg_edit_dist_img: 0.47
+precision: 87.84%
+recall: 86.88%
+fmeasure: 87.36%</pre></div>
+<p dir="auto">各个方案端到端指标如下：</p>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>指标</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
+<td>0.04%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
+<td>78.27%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
+<td>87.14%</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
+<td>88%</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到78.27%，在CCPD数据集上进行 fine-tune 后指标进一步提升到87.14%, 在经过量化训练之后，由于检测模型的recall变高，指标进一步提升到88%。但是这个结果仍旧不符合检测模型+识别模型的真实性能(99%*94%=93%)，因此我们需要对 base case 进行具体分析。</p>
+<p dir="auto">在之前的端到端预测结果中，可以看到很多不符合车牌标注的文字被识别出来, 因此可以进行简单的过滤来提升precision</p>
+<p dir="auto">为了快速评估，我们在 <code> tools/end2end/convert_ppocr_label.py</code> 脚本的 58 行加入如下代码，对非8个字符的结果进行过滤</p>
+<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="if len(txt) != 8: # 车牌字符串长度为8
+    continue"><pre><span class="pl-k">if</span> <span class="pl-en">len</span>(<span class="pl-s1">txt</span>) <span class="pl-c1">!=</span> <span class="pl-c1">8</span>: <span class="pl-c"># 车牌字符串长度为8</span>
+    <span class="pl-k">continue</span></pre></div>
+<p dir="auto">此外，通过可视化box可以发现有很多框都是竖直翻转之后的框，并且没有完全框住车牌边界，因此需要进行框的竖直翻转以及轻微扩大，示意图如下：</p>
+<p dir="auto"><a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/199752de33937424916ca3fd6ef50f83461dca6034c35d251e32a4eecd4c3228/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f35396162303431316338656234646664393137666232623665356236396131376565376361343833353134343461656339616336313034623739666631303238"><img src="https://camo.githubusercontent.com/199752de33937424916ca3fd6ef50f83461dca6034c35d251e32a4eecd4c3228/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f35396162303431316338656234646664393137666232623665356236396131376565376361343833353134343461656339616336313034623739666631303238" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/59ab0411c8eb4dfd917fb2b6e5b69a17ee7ca48351444aec9ac6104b79ff1028" style="max-width: 100%;"></a></p>
+<p dir="auto">修改前后个方案指标对比如下：</p>
+<p dir="auto">各个方案端到端指标如下：</p>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>base</th>
+<th>A:识别结果过滤</th>
+<th>B:use_dilation</th>
+<th>C:flip_box</th>
+<th>best</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
+<td>0.04%</td>
+<td>0.08%</td>
+<td>0.02%</td>
+<td>0.05%</td>
+<td>0.00%(A)</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
+<td>78.27%</td>
+<td>90.84%</td>
+<td>78.61%</td>
+<td>79.43%</td>
+<td>91.66%(A+B+C)</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
+<td>87.14%</td>
+<td>90.40%</td>
+<td>87.66%</td>
+<td>89.98</td>
+<td>92.5%(A+B+C)</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
+<td>88%</td>
+<td>90.54%</td>
+<td>88.5%</td>
+<td>89.46%</td>
+<td>92.02%(A+B+C)</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto">从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标变为92.02%。</p>
+<h3 dir="auto"><a id="user-content-44-部署" class="anchor" aria-hidden="true" href="#44-部署"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.4 部署</h3>
+<ul dir="auto">
+<li>基于 Paddle Inference 的python推理</li>
+</ul>
+<p dir="auto">检测模型和识别模型分别 fine-tune 并导出为inference模型之后，可以使用如下命令基于 Paddle Inference 进行端到端推理并对结果进行可视化。</p>
+<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/infer/predict_system.py \
+    --det_model_dir=output/CCPD/det/infer/ \
+    --rec_model_dir=output/CCPD/rec/infer/ \
+    --image_dir=&quot;/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&amp;509_530&amp;611-527&amp;611_172&amp;599_159&amp;509_530&amp;525-0_0_3_32_30_31_30_30-109-106.jpg&quot; \
+    --rec_image_shape=3,48,320"><pre>python tools/infer/predict_system.py \
+    --det_model_dir=output/CCPD/det/infer/ \
+    --rec_model_dir=output/CCPD/rec/infer/ \
+    --image_dir=<span class="pl-s"><span class="pl-pds">"</span>/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&amp;509_530&amp;611-527&amp;611_172&amp;599_159&amp;509_530&amp;525-0_0_3_32_30_31_30_30-109-106.jpg<span class="pl-pds">"</span></span> \
+    --rec_image_shape=3,48,320</pre></div>
+<p dir="auto">推理结果如下</p>
+<p dir="auto"><a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/b5c04d7affeca84aebcff3bf829edff2d142d26d700fd208c24b983144df766d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f37366236613039333963326334636634393033396236353633633462323865323431653131323835643734363465373939653831633538633066373730376137"><img src="https://camo.githubusercontent.com/b5c04d7affeca84aebcff3bf829edff2d142d26d700fd208c24b983144df766d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f37366236613039333963326334636634393033396236353633633462323865323431653131323835643734363465373939653831633538633066373730376137" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/76b6a0939c2c4cf49039b6563c4b28e241e11285d7464e799e81c58c0f7707a7" style="max-width: 100%;"></a></p>
+<ul dir="auto">
+<li>端侧部署</li>
+</ul>
+<p dir="auto">端侧部署我们采用基于 PaddleLite 的 cpp 推理。Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。具体可参考 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a></p>
+<h3 dir="auto"><a id="user-content-45-实验总结" class="anchor" aria-hidden="true" href="#45-实验总结"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.5 实验总结</h3>
+<p dir="auto">我们分别使用PP-OCRv3中英文超轻量预训练模型在车牌数据集上进行了直接评估和 fine-tune 和 fine-tune +量化3种方案的实验，并基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>进行了速度测试，指标对比如下：</p>
+<ul dir="auto">
+<li>检测</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>hmeans</th>
+<th>模型大小</th>
+<th>预测速度(lite)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
+<td>76.12%</td>
+<td>2.5M</td>
+<td>233ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
+<td>99%</td>
+<td>2.5M</td>
+<td>233ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune + 量化</td>
+<td>98.91%</td>
+<td>1M</td>
+<td>189ms</td>
+</tr>
+</tbody>
+</table>
+<ul dir="auto">
+<li>识别</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>acc</th>
+<th>模型大小</th>
+<th>预测速度(lite)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
+<td>0%</td>
+<td>10.3M</td>
+<td>4.2ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
+<td>90.97%</td>
+<td>10.3M</td>
+<td>4.2ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
+<td>94.54%</td>
+<td>10.3M</td>
+<td>4,2ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化</td>
+<td>93.4%</td>
+<td>4.8M</td>
+<td>1.8ms</td>
+</tr>
+</tbody>
+</table>
+<ul dir="auto">
+<li>端到端指标如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>方案</th>
+<th>fmeasure</th>
+<th>模型大小</th>
+<th>预测速度(lite)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
+<td>0.08%</td>
+<td>12.8M</td>
+<td>298ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
+<td>91.66%</td>
+<td>12.8M</td>
+<td>298ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
+<td>92.5%</td>
+<td>12.8M</td>
+<td>298ms</td>
+</tr>
+<tr>
+<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
+<td>92.02%</td>
+<td>5.8M</td>
+<td>224ms</td>
+</tr>
+</tbody>
+</table>
+<p dir="auto"><strong>结论</strong></p>
+<p dir="auto">PP-OCRv3的检测模型在未经过fine-tune的情况下，在车牌数据集上也有一定的精度，经过 fine-tune 后能够极大的提升检测效果，精度达到99%。在使用量化训练后检测模型的精度几乎无损，并且模型大小压缩60%。</p>
+<p dir="auto">PP-OCRv3的识别模型在未经过fine-tune的情况下，在车牌数据集上精度为0，但是经过分析可以知道，模型大部分字符都预测正确，但是会多预测一个特殊字符，去掉这个特殊字符后，精度达到90%。PP-OCRv3识别模型在经过 fine-tune 后识别精度进一步提升，达到94.4%。在使用量化训练后识别模型大小压缩53%，但是由于数据量多少，带来了1%的精度损失。</p>
+<p dir="auto">从端到端结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标轻微下降到92.02%但模型大小降低54%。</p>
+</article>
+  </div>
+
+    </div>
+
+  </readme-toc>
+
+
+
+  <details class="details-reset details-overlay details-overlay-dark" id="jumpto-line-details-dialog">
+    <summary data-hotkey="l" aria-label="Jump to line"></summary>
+    <details-dialog class="Box Box--overlay d-flex flex-column anim-fade-in fast linejump" aria-label="Jump to line">
+      <!-- '"` --><!-- </textarea></xmp> --></option></form><form class="js-jump-to-line-form Box-body d-flex" data-turbo="false" action="" accept-charset="UTF-8" method="get">
+        <input class="form-control flex-auto mr-3 linejump-input js-jump-to-line-field" type="text" placeholder="Jump to line&hellip;" aria-label="Jump to line" autofocus>
+        <button data-close-dialog="" type="submit" data-view-component="true" class="btn">  Go
+
+</button>
+</form>    </details-dialog>
+  </details>
+
+    <div class="Popover anim-scale-in js-tagsearch-popover"
+     hidden
+     data-tagsearch-url="/PaddlePaddle/PaddleOCR/find-definition"
+     data-tagsearch-ref="dygraph"
+     data-tagsearch-code-nav-context="BLOB_VIEW">
+  <div class="Popover-message Popover-message--large Popover-message--top-left TagsearchPopover mt-1 mb-4 mx-auto Box color-shadow-large">
+    <div class="TagsearchPopover-content js-tagsearch-popover-content overflow-auto" style="will-change:transform;">
+    </div>
+
+  </div>
+</div>
+
+
+</div>
+
+  </div>
+
+
+</div>
+
+
+    </main>
+  </div>
+
+  </div>
+
+          <footer class="footer width-full container-xl p-responsive" role="contentinfo">
+
+
+  <div class="position-relative d-flex flex-items-center pb-2 f6 color-fg-muted border-top color-border-muted flex-column-reverse flex-lg-row flex-wrap flex-lg-nowrap mt-6 pt-6">
+    <ul class="list-style-none d-flex flex-wrap col-0 col-lg-2 flex-justify-start flex-lg-justify-between mb-2 mb-lg-0">
+      <li class="mt-2 mt-lg-0 d-flex flex-items-center">
+        <a aria-label="Homepage" title="GitHub" class="footer-octicon mr-2" href="https://github.com">
+          <svg aria-hidden="true" height="24" viewBox="0 0 16 16" version="1.1" width="24" data-view-component="true" class="octicon octicon-mark-github">
+    <path fill-rule="evenodd" d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"></path>
+</svg>
+</a>        <span>
+        &copy; 2022 GitHub, Inc.
+        </span>
+      </li>
+    </ul>
+    <ul class="list-style-none d-flex flex-wrap col-12 col-lg-8 flex-justify-center flex-lg-justify-between mb-2 mb-lg-0">
+        <li class="mr-3 mr-lg-0"><a href="https://docs.github.com/en/github/site-policy/github-terms-of-service" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to terms&quot;,&quot;label&quot;:&quot;text:terms&quot;}">Terms</a></li>
+        <li class="mr-3 mr-lg-0"><a href="https://docs.github.com/en/github/site-policy/github-privacy-statement" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to privacy&quot;,&quot;label&quot;:&quot;text:privacy&quot;}">Privacy</a></li>
+        <li class="mr-3 mr-lg-0"><a data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to security&quot;,&quot;label&quot;:&quot;text:security&quot;}" href="https://github.com/security">Security</a></li>
+        <li class="mr-3 mr-lg-0"><a href="https://www.githubstatus.com/" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to status&quot;,&quot;label&quot;:&quot;text:status&quot;}">Status</a></li>
+        <li class="mr-3 mr-lg-0"><a data-ga-click="Footer, go to help, text:Docs" href="https://docs.github.com">Docs</a></li>
+        <li class="mr-3 mr-lg-0"><a href="https://support.github.com?tags=dotcom-footer" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to contact&quot;,&quot;label&quot;:&quot;text:contact&quot;}">Contact GitHub</a></li>
+        <li class="mr-3 mr-lg-0"><a href="https://github.com/pricing" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to Pricing&quot;,&quot;label&quot;:&quot;text:Pricing&quot;}">Pricing</a></li>
+      <li class="mr-3 mr-lg-0"><a href="https://docs.github.com" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to api&quot;,&quot;label&quot;:&quot;text:api&quot;}">API</a></li>
+      <li class="mr-3 mr-lg-0"><a href="https://services.github.com" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to training&quot;,&quot;label&quot;:&quot;text:training&quot;}">Training</a></li>
+        <li class="mr-3 mr-lg-0"><a href="https://github.blog" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to blog&quot;,&quot;label&quot;:&quot;text:blog&quot;}">Blog</a></li>
+        <li><a data-ga-click="Footer, go to about, text:about" href="https://github.com/about">About</a></li>
+    </ul>
+  </div>
+  <div class="d-flex flex-justify-center pb-6">
+    <span class="f6 color-fg-muted"></span>
+  </div>
+</footer>
+
+
+
+
+  <div id="ajax-error-message" class="ajax-error-message flash flash-error" hidden>
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert">
+    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
+</svg>
+    <button type="button" class="flash-close js-ajax-error-dismiss" aria-label="Dismiss error">
+      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
+    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
+</svg>
+    </button>
+    You can’t perform that action at this time.
+  </div>
+
+  <div class="js-stale-session-flash flash flash-warn flash-banner" hidden
+    >
+    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert">
+    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
+</svg>
+    <span class="js-stale-session-flash-signed-in" hidden>You signed in with another tab or window. <a href="">Reload</a> to refresh your session.</span>
+    <span class="js-stale-session-flash-signed-out" hidden>You signed out in another tab or window. <a href="">Reload</a> to refresh your session.</span>
+  </div>
+    <template id="site-details-dialog">
+  <details class="details-reset details-overlay details-overlay-dark lh-default color-fg-default hx_rsm" open>
+    <summary role="button" aria-label="Close dialog"></summary>
+    <details-dialog class="Box Box--overlay d-flex flex-column anim-fade-in fast hx_rsm-dialog hx_rsm-modal">
+      <button class="Box-btn-octicon m-0 btn-octicon position-absolute right-0 top-0" type="button" aria-label="Close dialog" data-close-dialog>
+        <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
+    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
+</svg>
+      </button>
+      <div class="octocat-spinner my-6 js-details-dialog-spinner"></div>
+    </details-dialog>
+  </details>
+</template>
+
+    <div class="Popover js-hovercard-content position-absolute" style="display: none; outline: none;" tabindex="0">
+  <div class="Popover-message Popover-message--bottom-left Popover-message--large Box color-shadow-large" style="width:360px;">
+  </div>
+</div>
+
+    <template id="snippet-clipboard-copy-button">
+  <div class="zeroclipboard-container position-absolute right-0 top-0">
+    <clipboard-copy aria-label="Copy" class="ClipboardButton btn js-clipboard-copy m-2 p-0 tooltipped-no-delay" data-copy-feedback="Copied!" data-tooltip-direction="w">
+      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-copy js-clipboard-copy-icon m-2">
+    <path fill-rule="evenodd" d="M0 6.75C0 5.784.784 5 1.75 5h1.5a.75.75 0 010 1.5h-1.5a.25.25 0 00-.25.25v7.5c0 .138.112.25.25.25h7.5a.25.25 0 00.25-.25v-1.5a.75.75 0 011.5 0v1.5A1.75 1.75 0 019.25 16h-7.5A1.75 1.75 0 010 14.25v-7.5z"></path><path fill-rule="evenodd" d="M5 1.75C5 .784 5.784 0 6.75 0h7.5C15.216 0 16 .784 16 1.75v7.5A1.75 1.75 0 0114.25 11h-7.5A1.75 1.75 0 015 9.25v-7.5zm1.75-.25a.25.25 0 00-.25.25v7.5c0 .138.112.25.25.25h7.5a.25.25 0 00.25-.25v-7.5a.25.25 0 00-.25-.25h-7.5z"></path>
+</svg>
+      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check js-clipboard-check-icon color-fg-success d-none m-2">
+    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
+</svg>
+    </clipboard-copy>
+  </div>
+</template>
+
+
+
+
+  </body>
+</html>

From aba127fe595773f7f2f3554c6bee5300e3c27a03 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 21 Jun 2022 15:50:39 +0800
Subject: [PATCH 034/134] fix

---
 applications/轻量级车牌识别.md | 3113 ++++++--------------------------
 1 file changed, 521 insertions(+), 2592 deletions(-)

diff --git a/applications/轻量级车牌识别.md b/applications/轻量级车牌识别.md
index e0f4706f14..7012c7f4bb 100644
--- a/applications/轻量级车牌识别.md
+++ b/applications/轻量级车牌识别.md
@@ -1,1571 +1,129 @@
+# 一种基于PaddleOCR的轻量级车牌识别模型
 
+- [1. 项目介绍](#1-项目介绍)
+- [2. 环境搭建](#2-环境搭建)
+- [3. 数据集准备](#3-数据集准备)
+  - [3.1 数据集标注规则](#31-数据集标注规则)
+  - [3.2 制作符合PP-OCR训练格式的标注文件](#32-制作符合pp-ocr训练格式的标注文件)
+- [4. 实验](#4-实验)
+  - [4.1 检测](#41-检测)
+    - [4.1.1 预训练模型直接预测](#411-预训练模型直接预测)
+    - [4.1.2 CCPD车牌数据集fine-tune](#412-ccpd车牌数据集fine-tune)
+    - [4.1.3 CCPD车牌数据集fine-tune+量化训练](#413-ccpd车牌数据集fine-tune量化训练)
+    - [4.1.4 模型导出](#414-模型导出)
+  - [4.2 识别](#42-识别)
+    - [4.2.1 预训练模型直接预测](#421-预训练模型直接预测)
+    - [4.2.2 预训练模型直接预测+改动后处理](#422-预训练模型直接预测改动后处理)
+    - [4.2.3 CCPD车牌数据集fine-tune](#423-ccpd车牌数据集fine-tune)
+    - [4.2.4 CCPD车牌数据集fine-tune+量化训练](#424-ccpd车牌数据集fine-tune量化训练)
+    - [4.2.5 模型导出](#425-模型导出)
+  - [4.3 计算End2End指标](#43-计算End2End指标)
+  - [4.4 部署](#44-部署)
+  - [4.5 实验总结](#45-实验总结)
 
+## 1. 项目介绍
 
+车牌识别(Vehicle License Plate Recognition，VLPR) 是计算机视频图像识别技术在车辆牌照识别中的一种应用。车牌识别技术要求能够将运动中的汽车牌照从复杂背景中提取并识别出来，在高速公路车辆管理，停车场管理和城市交通中得到广泛应用。
 
+本项目难点如下：
 
+1. 车牌在图像中的尺度差异大、在车辆上的悬挂位置不固定
+2. 车牌图像质量层次不齐: 角度倾斜、图片模糊、光照不足、过曝等问题严重
+3. 边缘和端测场景应用对模型大小有限制，推理速度有要求
 
+针对以上问题， 本例选用 PP-OCRv3 这一开源超轻量OCR系统进行车牌识别系统的开发。基于PP-OCRv3模型，在CCPD数据集达到99%的检测和94%的识别精度，模型大小12.8M(2.5M+10.3M)。基于量化对模型体积进行进一步压缩到5.8M(1M+4.8M), 同时推理速度提升25%。
 
-<!DOCTYPE html>
-<html lang="en" data-color-mode="auto" data-light-theme="light" data-dark-theme="dark" data-a11y-animated-images="system">
-  <head>
-    <meta charset="utf-8">
-  <link rel="dns-prefetch" href="https://github.githubassets.com">
-  <link rel="dns-prefetch" href="https://avatars.githubusercontent.com">
-  <link rel="dns-prefetch" href="https://github-cloud.s3.amazonaws.com">
-  <link rel="dns-prefetch" href="https://user-images.githubusercontent.com/">
-  <link rel="preconnect" href="https://github.githubassets.com" crossorigin>
-  <link rel="preconnect" href="https://avatars.githubusercontent.com">
 
 
+aistudio项目链接: [基于PaddleOCR的轻量级车牌识别范例](https://aistudio.baidu.com/aistudio/projectdetail/3919091?contributionType=1)
 
-  <link crossorigin="anonymous" media="all" integrity="sha512-ksfTgQOOnE+FFXf+yNfVjKSlEckJAdufFIYGK7ZjRhWcZgzAGcmZqqArTgMLpu90FwthqcCX4ldDgKXbmVMeuQ==" rel="stylesheet" href="https://github.githubassets.com/assets/light-92c7d381038e.css" /><link crossorigin="anonymous" media="all" integrity="sha512-1KkMNn8M/al/dtzBLupRwkIOgnA9MWkm8oxS+solP87jByEvY/g4BmoxLihRogKcX1obPnf4Yp7dI0ZTWO+ljg==" rel="stylesheet" href="https://github.githubassets.com/assets/dark-d4a90c367f0c.css" /><link data-color-theme="dark_dimmed" crossorigin="anonymous" media="all" integrity="sha512-cZa7DZqvMBwD236uzEunO/G1dvw8/QftyT2UtLWKQFEy0z0eq0R5WPwqVME+3NSZG1YaLJAaIqtU+m0zWf/6SQ==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_dimmed-7196bb0d9aaf.css" /><link data-color-theme="dark_high_contrast" crossorigin="anonymous" media="all" integrity="sha512-WVoKqJ4y1nLsdNH4RkRT5qrM9+n9RFe1RHSiTnQkBf5TSZkJEc9GpLpTIS7T15EQaUQBJ8BwmKvwFPVqfpTEIQ==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_high_contrast-595a0aa89e32.css" /><link data-color-theme="dark_colorblind" crossorigin="anonymous" media="all" integrity="sha512-XpAMBMSRZ6RTXgepS8LjKiOeNK3BilRbv8qEiA/M3m+Q4GoqxtHedOI5BAZRikCzfBL4KWYvVzYZSZ8Gp/UnUg==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_colorblind-5e900c04c491.css" /><link data-color-theme="light_colorblind" crossorigin="anonymous" media="all" integrity="sha512-3HF2HZ4LgEIQm77yOzoeR20CX1n2cUQlcywscqF4s+5iplolajiHV7E5ranBwkX65jN9TNciHEVSYebQ+8xxEw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_colorblind-dc71761d9e0b.css" /><link data-color-theme="light_high_contrast" crossorigin="anonymous" media="all" integrity="sha512-+J8j3T0kbK9/sL3zbkCfPtgYcRD4qQfRbT6xnfOrOTjvz4zhr0M7AXPuE642PpaxGhHs1t77cTtieW9hI2K6Gw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_high_contrast-f89f23dd3d24.css" /><link data-color-theme="light_tritanopia" crossorigin="anonymous" media="all" integrity="sha512-AQeAx5wHQAXNf0DmkvVlHYwA3f6BkxunWTI0GGaRN57GqD+H9tW8RKIKlopLS0qGaC54seFsPc601GDlqIuuHg==" rel="stylesheet" data-href="https://github.githubassets.com/assets/light_tritanopia-010780c79c07.css" /><link data-color-theme="dark_tritanopia" crossorigin="anonymous" media="all" integrity="sha512-+u5pmgAE0T03d/yI6Ha0NWwz6Pk0W6S6WEfIt8veDVdK8NTjcMbZmQB9XUCkDlrBoAKkABva8HuGJ+SzEpV1Uw==" rel="stylesheet" data-href="https://github.githubassets.com/assets/dark_tritanopia-faee699a0004.css" />
+## 2. 环境搭建
 
-    <link crossorigin="anonymous" media="all" integrity="sha512-EAhBCLIJ/pXHG3Y6yQhs9s53SHV80sjJ+yCwlQtfv7LaVkD+VoEuZBZ5betQJFUNj/5qBSfZk5GFtazEDzWLAg==" rel="stylesheet" href="https://github.githubassets.com/assets/primer-10084108b209.css" />
-    <link crossorigin="anonymous" media="all" integrity="sha512-4cUFwwYgpnyRoJKr15lavpQFa6HaoGIW6sap6JXLP9cQ33dOLfadeyjfJypUJ8GfE8MnAfdfvcNZlgn7oZ70PA==" rel="stylesheet" href="https://github.githubassets.com/assets/global-e1c505c30620.css" />
-    <link crossorigin="anonymous" media="all" integrity="sha512-ZkqzH/tvOPQgSkHnN9TlzL/abuPzgfV2KLooWN2Xq+jrP7VFzNKd0dib2+zSde1277kCzr5gp1piayo6AthIsQ==" rel="stylesheet" href="https://github.githubassets.com/assets/github-664ab31ffb6f.css" />
-  <link crossorigin="anonymous" media="all" integrity="sha512-PXtwH8brj9qH7LskWA0iAMLw4Wx3RXN7txSxjlqdzQEalZKu3UcsQ+ajfHJMDkYDA2iWL1FIrf2zyLflA5Xu7A==" rel="stylesheet" href="https://github.githubassets.com/assets/code-3d7b701fc6eb.css" />
+本任务基于Aistudio完成, 具体环境如下：
 
-    <meta name="optimizely-datafile" content="{&quot;groups&quot;: [], &quot;environmentKey&quot;: &quot;production&quot;, &quot;rollouts&quot;: [], &quot;typedAudiences&quot;: [], &quot;projectId&quot;: &quot;16737760170&quot;, &quot;variables&quot;: [], &quot;featureFlags&quot;: [], &quot;experiments&quot;: [{&quot;status&quot;: &quot;Running&quot;, &quot;audienceIds&quot;: [], &quot;variations&quot;: [{&quot;variables&quot;: [], &quot;id&quot;: &quot;20667381018&quot;, &quot;key&quot;: &quot;control&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;20680930759&quot;, &quot;key&quot;: &quot;treatment&quot;}], &quot;id&quot;: &quot;20652570897&quot;, &quot;key&quot;: &quot;project_genesis&quot;, &quot;layerId&quot;: &quot;20672300363&quot;, &quot;trafficAllocation&quot;: [{&quot;entityId&quot;: &quot;20667381018&quot;, &quot;endOfRange&quot;: 5000}, {&quot;entityId&quot;: &quot;20680930759&quot;, &quot;endOfRange&quot;: 10000}], &quot;forcedVariations&quot;: {&quot;83356e17066d336d1803024138ecb683&quot;: &quot;treatment&quot;, &quot;18e31c8a9b2271332466133162a4aa0d&quot;: &quot;treatment&quot;, &quot;10f8ab3fbc5ebe989a36a05f79d48f32&quot;: &quot;treatment&quot;, &quot;1686089f6d540cd2deeaec60ee43ecf7&quot;: &quot;treatment&quot;}}, {&quot;status&quot;: &quot;Running&quot;, &quot;audienceIds&quot;: [], &quot;variations&quot;: [{&quot;variables&quot;: [], &quot;id&quot;: &quot;21427950901&quot;, &quot;key&quot;: &quot;control&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;21429710665&quot;, &quot;key&quot;: &quot;beginner&quot;}, {&quot;variables&quot;: [], &quot;id&quot;: &quot;21437291543&quot;, &quot;key&quot;: &quot;upstart&quot;}], &quot;id&quot;: &quot;21445030708&quot;, &quot;key&quot;: &quot;_259_zero_user_dashboard&quot;, &quot;layerId&quot;: &quot;21434011841&quot;, &quot;trafficAllocation&quot;: [{&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 3334}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 5000}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 8333}, {&quot;entityId&quot;: &quot;21427950901&quot;, &quot;endOfRange&quot;: 10000}], &quot;forcedVariations&quot;: {&quot;3c64268131793aa297119a343c19e345&quot;: &quot;beginner&quot;, &quot;95b24126db31ea8693c0fe5ea9f53b65&quot;: &quot;beginner&quot;, &quot;086e2abe64e9101112af53b95d2d90b9&quot;: &quot;upstart&quot;, &quot;bae688df9d297afac98e2d254e912ada&quot;: &quot;control&quot;, &quot;6c2cfda7c41396fcc31a4db759a42b94&quot;: &quot;beginner&quot;, &quot;16ed2b4ff7de02663b7c606309695916&quot;: &quot;control&quot;, &quot;1971768911.1635962195&quot;: &quot;beginner&quot;, &quot;830bf802470ec6c9c5800c99d8e57445&quot;: &quot;beginner&quot;}}], &quot;version&quot;: &quot;4&quot;, &quot;audiences&quot;: [{&quot;conditions&quot;: &quot;[\&quot;or\&quot;, {\&quot;match\&quot;: \&quot;exact\&quot;, \&quot;name\&quot;: \&quot;$opt_dummy_attribute\&quot;, \&quot;type\&quot;: \&quot;custom_attribute\&quot;, \&quot;value\&quot;: \&quot;$opt_dummy_value\&quot;}]&quot;, &quot;id&quot;: &quot;$opt_dummy_audience&quot;, &quot;name&quot;: &quot;Optimizely-Generated Audience for Backwards Compatibility&quot;}], &quot;anonymizeIP&quot;: true, &quot;sdkKey&quot;: &quot;WTc6awnGuYDdG98CYRban&quot;, &quot;attributes&quot;: [{&quot;id&quot;: &quot;16822470375&quot;, &quot;key&quot;: &quot;user_id&quot;}, {&quot;id&quot;: &quot;17143601254&quot;, &quot;key&quot;: &quot;spammy&quot;}, {&quot;id&quot;: &quot;18175660309&quot;, &quot;key&quot;: &quot;organization_plan&quot;}, {&quot;id&quot;: &quot;18813001570&quot;, &quot;key&quot;: &quot;is_logged_in&quot;}, {&quot;id&quot;: &quot;19073851829&quot;, &quot;key&quot;: &quot;geo&quot;}, {&quot;id&quot;: &quot;20175462351&quot;, &quot;key&quot;: &quot;requestedCurrency&quot;}, {&quot;id&quot;: &quot;20785470195&quot;, &quot;key&quot;: &quot;country_code&quot;}, {&quot;id&quot;: &quot;21656311196&quot;, &quot;key&quot;: &quot;opened_downgrade_dialog&quot;}], &quot;botFiltering&quot;: false, &quot;accountId&quot;: &quot;16737760170&quot;, &quot;events&quot;: [{&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;17911811441&quot;, &quot;key&quot;: &quot;hydro_click.dashboard.teacher_toolbox_cta&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18124116703&quot;, &quot;key&quot;: &quot;submit.organizations.complete_sign_up&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18145892387&quot;, &quot;key&quot;: &quot;no_metric.tracked_outside_of_optimizely&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18178755568&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.add_repo&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18180553241&quot;, &quot;key&quot;: &quot;submit.repository_imports.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18186103728&quot;, &quot;key&quot;: &quot;click.help.learn_more_about_repository_creation&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18188530140&quot;, &quot;key&quot;: &quot;test_event&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18191963644&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.transfer_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18195612788&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.import_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18210945499&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.invite_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18211063248&quot;, &quot;key&quot;: &quot;click.empty_org_repo_cta.create_repository&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18215721889&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.update_profile&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18224360785&quot;, &quot;key&quot;: &quot;click.org_onboarding_checklist.dismiss&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18234832286&quot;, &quot;key&quot;: &quot;submit.organization_activation.complete&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18252392383&quot;, &quot;key&quot;: &quot;submit.org_repository.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18257551537&quot;, &quot;key&quot;: &quot;submit.org_member_invitation.create&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18259522260&quot;, &quot;key&quot;: &quot;submit.organization_profile.update&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18564603625&quot;, &quot;key&quot;: &quot;view.classroom_select_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18568612016&quot;, &quot;key&quot;: &quot;click.classroom_sign_in_click&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18572592540&quot;, &quot;key&quot;: &quot;view.classroom_name&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18574203855&quot;, &quot;key&quot;: &quot;click.classroom_create_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18582053415&quot;, &quot;key&quot;: &quot;click.classroom_select_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18589463420&quot;, &quot;key&quot;: &quot;click.classroom_create_classroom&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18591323364&quot;, &quot;key&quot;: &quot;click.classroom_create_first_classroom&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18591652321&quot;, &quot;key&quot;: &quot;click.classroom_grant_access&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18607131425&quot;, &quot;key&quot;: &quot;view.classroom_creation&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;18831680583&quot;, &quot;key&quot;: &quot;upgrade_account_plan&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19064064515&quot;, &quot;key&quot;: &quot;click.signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19075373687&quot;, &quot;key&quot;: &quot;click.view_account_billing_page&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19077355841&quot;, &quot;key&quot;: &quot;click.dismiss_signup_prompt&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19079713938&quot;, &quot;key&quot;: &quot;click.contact_sales&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19120963070&quot;, &quot;key&quot;: &quot;click.compare_account_plans&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19151690317&quot;, &quot;key&quot;: &quot;click.upgrade_account_cta&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19424193129&quot;, &quot;key&quot;: &quot;click.open_account_switcher&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19520330825&quot;, &quot;key&quot;: &quot;click.visit_account_profile&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19540970635&quot;, &quot;key&quot;: &quot;click.switch_account_context&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19730198868&quot;, &quot;key&quot;: &quot;submit.homepage_signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19820830627&quot;, &quot;key&quot;: &quot;click.homepage_signup&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;19988571001&quot;, &quot;key&quot;: &quot;click.create_enterprise_trial&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20036538294&quot;, &quot;key&quot;: &quot;click.create_organization_team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20040653299&quot;, &quot;key&quot;: &quot;click.input_enterprise_trial_form&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20062030003&quot;, &quot;key&quot;: &quot;click.continue_with_team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20068947153&quot;, &quot;key&quot;: &quot;click.create_organization_free&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20086636658&quot;, &quot;key&quot;: &quot;click.signup_continue.username&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20091648988&quot;, &quot;key&quot;: &quot;click.signup_continue.create_account&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20103637615&quot;, &quot;key&quot;: &quot;click.signup_continue.email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20111574253&quot;, &quot;key&quot;: &quot;click.signup_continue.password&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20120044111&quot;, &quot;key&quot;: &quot;view.pricing_page&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20152062109&quot;, &quot;key&quot;: &quot;submit.create_account&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20165800992&quot;, &quot;key&quot;: &quot;submit.upgrade_payment_form&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20171520319&quot;, &quot;key&quot;: &quot;submit.create_organization&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20222645674&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.discuss_your_needs&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20227443657&quot;, &quot;key&quot;: &quot;submit.verify_primary_user_email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20234607160&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.try_enterprise&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20238175784&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.team&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20239847212&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.continue_free&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20251097193&quot;, &quot;key&quot;: &quot;recommended_plan&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20438619534&quot;, &quot;key&quot;: &quot;click.pricing_calculator.1_member&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20456699683&quot;, &quot;key&quot;: &quot;click.pricing_calculator.15_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20467868331&quot;, &quot;key&quot;: &quot;click.pricing_calculator.10_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20476267432&quot;, &quot;key&quot;: &quot;click.trial_days_remaining&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20476357660&quot;, &quot;key&quot;: &quot;click.discover_feature&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20479287901&quot;, &quot;key&quot;: &quot;click.pricing_calculator.custom_members&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20481107083&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.apply_teacher_benefits&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20483089392&quot;, &quot;key&quot;: &quot;click.pricing_calculator.5_members&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20484283944&quot;, &quot;key&quot;: &quot;click.onboarding_task&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20484996281&quot;, &quot;key&quot;: &quot;click.recommended_plan_in_signup.apply_student_benefits&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20486713726&quot;, &quot;key&quot;: &quot;click.onboarding_task_breadcrumb&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20490791319&quot;, &quot;key&quot;: &quot;click.upgrade_to_enterprise&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20491786766&quot;, &quot;key&quot;: &quot;click.talk_to_us&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20494144087&quot;, &quot;key&quot;: &quot;click.dismiss_enterprise_trial&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20499722759&quot;, &quot;key&quot;: &quot;completed_all_tasks&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20500710104&quot;, &quot;key&quot;: &quot;completed_onboarding_tasks&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20513160672&quot;, &quot;key&quot;: &quot;click.read_doc&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20516196762&quot;, &quot;key&quot;: &quot;actions_enabled&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20518980986&quot;, &quot;key&quot;: &quot;click.dismiss_trial_banner&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20535446721&quot;, &quot;key&quot;: &quot;click.issue_actions_prompt.dismiss_prompt&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20557002247&quot;, &quot;key&quot;: &quot;click.issue_actions_prompt.setup_workflow&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20595070227&quot;, &quot;key&quot;: &quot;click.pull_request_setup_workflow&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20626600314&quot;, &quot;key&quot;: &quot;click.seats_input&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20642310305&quot;, &quot;key&quot;: &quot;click.decrease_seats_number&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20662990045&quot;, &quot;key&quot;: &quot;click.increase_seats_number&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20679620969&quot;, &quot;key&quot;: &quot;click.public_product_roadmap&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20761240940&quot;, &quot;key&quot;: &quot;click.dismiss_survey_banner&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20767210721&quot;, &quot;key&quot;: &quot;click.take_survey&quot;}, {&quot;experimentIds&quot;: [&quot;20652570897&quot;], &quot;id&quot;: &quot;20795281201&quot;, &quot;key&quot;: &quot;click.archive_list&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20966790249&quot;, &quot;key&quot;: &quot;contact_sales.submit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20996500333&quot;, &quot;key&quot;: &quot;contact_sales.existing_customer&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;20996890162&quot;, &quot;key&quot;: &quot;contact_sales.blank_message_field&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21000470317&quot;, &quot;key&quot;: &quot;contact_sales.personal_email&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21002790172&quot;, &quot;key&quot;: &quot;contact_sales.blank_phone_field&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21354412592&quot;, &quot;key&quot;: &quot;click.dismiss_create_readme&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21366102546&quot;, &quot;key&quot;: &quot;click.dismiss_zero_user_content&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21370252505&quot;, &quot;key&quot;: &quot;account_did_downgrade&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21370840408&quot;, &quot;key&quot;: &quot;click.cta_create_readme&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21375451068&quot;, &quot;key&quot;: &quot;click.cta_create_new_repository&quot;}, {&quot;experimentIds&quot;: [&quot;21445030708&quot;], &quot;id&quot;: &quot;21385390948&quot;, &quot;key&quot;: &quot;click.zero_user_content&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21467712175&quot;, &quot;key&quot;: &quot;click.downgrade_keep&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21484112202&quot;, &quot;key&quot;: &quot;click.downgrade&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21495292213&quot;, &quot;key&quot;: &quot;click.downgrade_survey_exit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21508241468&quot;, &quot;key&quot;: &quot;click.downgrade_survey_submit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21512030356&quot;, &quot;key&quot;: &quot;click.downgrade_support&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21539090022&quot;, &quot;key&quot;: &quot;click.downgrade_exit&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21543640644&quot;, &quot;key&quot;: &quot;click_fetch_upstream&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21646510300&quot;, &quot;key&quot;: &quot;click.move_your_work&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21656151116&quot;, &quot;key&quot;: &quot;click.add_branch_protection_rule&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21663860599&quot;, &quot;key&quot;: &quot;click.downgrade_dialog_open&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21687860483&quot;, &quot;key&quot;: &quot;click.learn_about_protected_branches&quot;}, {&quot;experimentIds&quot;: [], &quot;id&quot;: &quot;21689050333&quot;, &quot;key&quot;: &quot;click.dismiss_protect_this_branch&quot;}], &quot;revision&quot;: &quot;1288&quot;}" />
+- 操作系统: Linux
+- PaddlePaddle: 2.3
+- paddleslim: 2.2.2
+- PaddleOCR: Release/2.5
 
+下载 PaddleOCR代码
 
-  <script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-IlxrV3fPKJ3H3zXBv8ZNFZA1UGUCcfs482E0tD9JDa8vDF+oIfVC8m5/T6o5u/ccu/oNvjJNPZhYkX02Ao349g==" src="https://github.githubassets.com/assets/runtime-225c6b5777cf.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-XKV1seh7UvAY3d4/PUvi5uHqcCJ9ThVg73aqJM9r8QFid4KRpTHbebXZdhznn39F6fvGFcgvAhTdTrUp/oWL1A==" src="https://github.githubassets.com/assets/environment-5ca575b1e87b.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-io+1MvgXPXTw8Kp4eOdNMJl8uGASuw8VfTY5VeIFETaAknimWi8GoxggMEeQ6mq0de4Dest4iIJ/9gUbCo0hgw==" src="https://github.githubassets.com/assets/vendors-node_modules_selector-observer_dist_index_esm_js-8a8fb532f817.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-Es25N4GyPa8Yfp5wpahoe5b2fyPtkRMyR6mKIXyCJC0ocqQazeWvxhGZhx3StRxOfqDfHDR5SS35u/R3Wux6Cg==" src="https://github.githubassets.com/assets/vendors-node_modules_delegated-events_dist_index_js-node_modules_github_details-dialog-elemen-63debe-12cdb93781b2.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-lmiecOIgf+hakg5oKMNM7grVhEDyPoIrT39Px448JJH5PSAaK21PH0Twgyz5O5oi8+dnlLr3Jt8bBCtAcpNdRw==" src="https://github.githubassets.com/assets/vendors-node_modules_github_filter-input-element_dist_index_js-node_modules_github_remote-inp-c7e9ed-96689e70e220.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-y67eNkVaNK4RguUGcHOvIbHFlgf1Qje+LDdjVw2eFuuvBOqta2GePz/CwoLIR/PJhhRAj5RPGxCWoomnimSw6w==" src="https://github.githubassets.com/assets/vendors-node_modules_github_catalyst_lib_index_js-node_modules_github_time-elements_dist_index_js-cbaede36455a.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-DE9GK/TS3Hf/dJfP1GMiyjvsGRnpR4A1eGH64s0Ot7Vy7Tj7/aD1d5EOMWHFWQXlXy/ZSwu5MHInG6RZN31bMA==" src="https://github.githubassets.com/assets/vendors-node_modules_github_file-attachment-element_dist_index_js-node_modules_primer_view-co-52e104-0c4f462bf4d2.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-l//8hwOiwPAmBg0NwsWUYovdQ7/r9kPHiD9/LL4fD3M7El8gbgaOYU5+o6cYLB6puSfOTxyN9M6fE38HSDr2Bw==" src="https://github.githubassets.com/assets/github-elements-97fffc8703a2.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-lIm60Y77qCoue/4Srnoe0QDZN/ReE8mrMZE2uyNTZteQP0e5kPJGSn4NtlWfYkVvI1DuyDC76KsOz04K76v+iA==" src="https://github.githubassets.com/assets/element-registry-9489bad18efb.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-uo73yUZcm4EicwjSbfxFZcKfjWniOxhLBp+q1n7IFRfutFM6/lzbQMgD0Xrxp7QD1HzqdvrV8UclPhi3mEOyzQ==" src="https://github.githubassets.com/assets/vendors-node_modules_lit-html_lit-html_js-ba8ef7c9465c.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-5XdknqL9iTuaAhbOnKJDHf+5bJReOYDwlz5p2mW+SAVfab4RN6AfIliqtPwheKld40DrhPs47We/mT5nlZq2Sg==" src="https://github.githubassets.com/assets/vendors-node_modules_github_remote-form_dist_index_js-node_modules_github_catalyst_lib_index_-87b1b3-e577649ea2fd.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-0r1nf/rfPz54kyePp4f63bcPxkFo7wyaUZJD/SwIVDK3q0WzurAK9ydOm88tzKtPJm8xWI0Vo25NyCfecwxJ9g==" src="https://github.githubassets.com/assets/vendors-node_modules_github_mini-throttle_dist_index_js-node_modules_github_hotkey_dist_index-9f48bd-d2bd677ffadf.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-VK56d0N1hPZ20mOzPoy84zlTGCjGbKKmVvfjoyDqSF+VxTD4f6X8QDs2RgG1R1cdBmsCiea+ZxP6ukV3tHlD+Q==" src="https://github.githubassets.com/assets/vendors-node_modules_github_paste-markdown_dist_index_esm_js-node_modules_github_quote-select-df2537-54ae7a774375.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-vneJnOQi1L9PJESlqC6veCutbVwyQDQCzMoPlkEhfWJ592tNYva9p1SV5uTde09VMWh17HFMDbfuVoy0I5tqHA==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_pjax_ts-be77899ce422.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-2hsYB2KHyayxUJJxJGhXeTeTZZG2c7lzRtO0uB1txmpc7rfvIt4mf0iossT0MHIHknYlaslgi98jmmlVXcXaZQ==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_keyboard-shortcuts-helper_ts-app_assets_modules_github_be-af52ef-da1b18076287.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-8eyyuvsdeUUBrkTsu7OePie4HdGA6yQww2w70dhs28WU0cMh5qhDM1vHrLOwJdbFQodch4KNQ1rv6KNyw1XMpA==" src="https://github.githubassets.com/assets/app_assets_modules_github_behaviors_details_ts-app_assets_modules_github_behaviors_include-fr-a5a4c7-f1ecb2bafb1d.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-i0EzX4ojOyIXe3MB4Kc/20cXNNTwDWHPd9i/bwOEDOd3bRNP7xJCQkX4f7DIELn+WFHtJSpEEfPRzi64gEENFw==" src="https://github.githubassets.com/assets/behaviors-8b41335f8a23.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-rLnbCtiKIpRtm1jgAz913tDImW3g8G+FLRMBIMwrkZ33pO8PMzWDG4m8Zl0uIIs10hCyFr6Dc8s6VDC200S9lg==" src="https://github.githubassets.com/assets/vendors-node_modules_delegated-events_dist_index_js-node_modules_github_catalyst_lib_index_js-01c6c51-acb9db0ad88a.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-zJONQW0ZfaXs3mk/bws2nP0XemlIUbA4+vWsLqTzjTTz7X2yCAI0VXa+i8y/vgLF3Ooeyus5ZnyhiKYj3Jwz6w==" src="https://github.githubassets.com/assets/notifications-global-cc938d416d19.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-1b+YhvQsW1BdKi3eWSc7oNSGJcrQalc3oxxMoXI8foh3weo2AOLx86EMZ5J1a541yu2y7Axj7Vntn1T66EW8dA==" src="https://github.githubassets.com/assets/vendors-node_modules_optimizely_optimizely-sdk_dist_optimizely_browser_es_min_js-node_modules-77839b-d5bf9886f42c.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-NH4Gv218MnzYs/oudrFw7n14lPMa75qmHypJ9MbJn1RW03KwjJKCegU17usZk1uQD8cLWseA4vhCL638+tKM+A==" src="https://github.githubassets.com/assets/optimizely-347e06bf6d7c.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-yG3t7Lpt5yKkdMNXFpl75Xemm+g4ghZ5e8xRBwl/HMbmgc3htVvxrjbohbK9MMfnnci1SnogZNH9DhJHYdQf3A==" src="https://github.githubassets.com/assets/vendors-node_modules_virtualized-list_es_index_js-node_modules_github_template-parts_lib_index_js-c86dedecba6d.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-eE/fdOH+uhuRT992TtBa/oI33EN3x7EBNkJnTdiozp5RMDDiteODOWGCazeuulOPF5ZSFPR7r0PBZFn8enpsRQ==" src="https://github.githubassets.com/assets/vendors-node_modules_github_mini-throttle_dist_decorators_js-node_modules_github_remote-form_-2c55be-784fdf74e1fe.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-FNWxKHkUqs6+xuyyZpMx9BmvcxQqIgVJ5A+PmdU82lZO41cFlo6y6ddHkRfWh2Zd6j9ozvndvrTaOuEH58VvEw==" src="https://github.githubassets.com/assets/vendors-node_modules_github_file-attachment-element_dist_index_js-node_modules_github_filter--fd67a9-14d5b1287914.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-dAabCMULUfibZDcNGDpWbINNJawdJeCfhl8ZTGFEYvgpblXXQIBlutl5zv0oOqc0p1ginRnWy4mAZES/swKmZA==" src="https://github.githubassets.com/assets/app_assets_modules_github_fetch_ts-app_assets_modules_github_input-demux_ts-app_assets_module-e5b55c-74069b08c50b.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-SXRchiPOYprh1D0HcC85QDp+f15IP1oshRaSxymGKZ5jw+se3OVL6zw5QYPnxl9cOf8WxcyBuW+t+raq61yg5g==" src="https://github.githubassets.com/assets/repositories-49745c8623ce.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-2IO4DDKs14Xk+vMRzfe3NVcyY5Bsb071hi/LJEZcTTo74dZ9sgelYQaCPu0nh3qsny1AvzutHZyuvBmt0fbguA==" src="https://github.githubassets.com/assets/app_assets_modules_github_diffs_blob-lines_ts-app_assets_modules_github_diffs_linkable-line-n-f314c3-d883b80c32ac.js"></script>
-<script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-9b6aISppMLr5F8Imzz33ChOeks/lElt5UEQ1fHaKruZEmJ/Jf2bZ8A96OL0fPGqhReDoWK5CZec/Btp1ny7+oQ==" src="https://github.githubassets.com/assets/diffs-f5be9a212a69.js"></script>
+```bash
+git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR
+```
 
+安装依赖库
 
-  <title>PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR · GitHub</title>
+```bash
+pip install -r PaddleOCR/requirements.txt
+```
 
+## 3. 数据集准备
 
+所使用的数据集为 CCPD2020 新能源车牌数据集，该数据集为
 
+该数据集分布如下：
 
+|数据集类型|数量|
+|---|---|
+|训练集| 5769|
+|验证集| 1001|
+|测试集| 5006|
 
-  <meta name="request-id" content="B6FC:313F:2FB16:4AF11:62B1726D" data-pjax-transient="true"/><meta name="html-safe-nonce" content="95c166e7500cd689d260f3b8707ad62a8e0f0dec50b5a900780ad09a4cb58361" data-pjax-transient="true"/><meta name="visitor-payload" content="eyJyZWZlcnJlciI6IiIsInJlcXVlc3RfaWQiOiJCNkZDOjMxM0Y6MkZCMTY6NEFGMTE6NjJCMTcyNkQiLCJ2aXNpdG9yX2lkIjoiNDc4ODQxMzQ5MTgzODAyMjI1NCIsInJlZ2lvbl9lZGdlIjoic2VhIiwicmVnaW9uX3JlbmRlciI6InNlYSJ9" data-pjax-transient="true"/><meta name="visitor-hmac" content="60769063a92f53a010f7783fa7370a5d4b255dce3cfb2ae919babf28c57cc46d" data-pjax-transient="true"/>
+数据集图片示例如下:
+![](https://ai-studio-static-online.cdn.bcebos.com/3bce057a8e0c40a0acbd26b2e29e4e2590a31bc412764be7b9e49799c69cb91c)
 
-    <meta name="hovercard-subject-tag" content="repository:262296122" data-pjax-transient>
+数据集可以从这里下载 https://aistudio.baidu.com/aistudio/datasetdetail/101595
 
+下载好数据集后对数据集进行解压
 
-  <meta name="github-keyboard-shortcuts" content="repository,source-code" data-pjax-transient="true" />
+```bash
+unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip
+```
 
+### 3.1 数据集标注规则
 
-  <meta name="selected-link" value="repo_source" data-pjax-transient>
+CPPD数据集的图片文件名具有特殊规则，详细可查看：https://github.com/detectRecog/CCPD
 
-    <meta name="google-site-verification" content="c1kuD-K2HIVF635lypcsWPoD4kilo5-jA_wBFyT4uMY">
-  <meta name="google-site-verification" content="KT5gs8h0wvaagLKAVWq8bbeNwnZZK1r1XQysX3xurLU">
-  <meta name="google-site-verification" content="ZzhVyEFwb7w3e0-uOTltm8Jsck2F5StVihD0exw2fsA">
-  <meta name="google-site-verification" content="GXs5KoUUkNCoaAZn7wPN-t01Pywp9M3sEjnt_3_ZWPc">
+具体规则如下：
 
-<meta name="octolytics-url" content="https://collector.github.com/github/collect" />
+例如: 025-95_113-154&383_386&473-386&473_177&454_154&383_363&402-0_0_22_27_27_33_16-37-15.jpg
 
-  <meta name="analytics-location" content="/&lt;user-name&gt;/&lt;repo-name&gt;/blob/show" data-pjax-transient="true" />
+每个名称可以分为七个字段，以-符号作为分割。这些字段解释如下。
 
+- 025：车牌面积与整个图片区域的面积比。025 (25%)
 
+- 95_113：水平倾斜程度和垂直倾斜度。水平 95度 垂直 113度
 
+- 154&383_386&473：左上和右下顶点的坐标。左上(154,383) 右下(386,473)
 
+- 386&473_177&454_154&383_363&402：整个图像中车牌的四个顶点的精确（x，y）坐标。这些坐标从右下角顶点开始。(386,473) (177,454) (154,383) (363,402)
 
+- 0_0_22_27_27_33_16：CCPD中的每个图像只有一个车牌。每个车牌号码由一个汉字，一个字母和五个字母或数字组成。有效的中文车牌由七个字符组成：省（1个字符），字母（1个字符），字母+数字（5个字符）。“ 0_0_22_27_27_33_16”是每个字符的索引。这三个数组定义如下。每个数组的最后一个字符是字母O，而不是数字0。我们将O用作“无字符”的符号，因为中文车牌字符中没有O。因此以上车牌拼起来即为 皖AY339S
 
+- 37：牌照区域的亮度。 37 (37%)
 
+- 15：车牌区域的模糊度。15 (15%)
 
-    <meta name="user-login" content="">
-
-    <meta name="viewport" content="width=device-width">
-
-      <meta name="description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices) - PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR">
-      <link rel="search" type="application/opensearchdescription+xml" href="/opensearch.xml" title="GitHub">
-    <link rel="fluid-icon" href="https://github.com/fluidicon.png" title="GitHub">
-    <meta property="fb:app_id" content="1401488693436528">
-    <meta name="apple-itunes-app" content="app-id=1477376905" />
-      <meta name="twitter:image:src" content="https://opengraph.githubassets.com/1eca697559d2aabccdeb858570f3f508b575ece90f37d8d4e2156599d7d26d4f/PaddlePaddle/PaddleOCR" /><meta name="twitter:site" content="@github" /><meta name="twitter:card" content="summary_large_image" /><meta name="twitter:title" content="PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR" /><meta name="twitter:description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." />
-      <meta property="og:image" content="https://opengraph.githubassets.com/1eca697559d2aabccdeb858570f3f508b575ece90f37d8d4e2156599d7d26d4f/PaddlePaddle/PaddleOCR" /><meta property="og:image:alt" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." /><meta property="og:image:width" content="1200" /><meta property="og:image:height" content="600" /><meta property="og:site_name" content="GitHub" /><meta property="og:type" content="object" /><meta property="og:title" content="PaddleOCR/轻量级车牌识别.md at dygraph · PaddlePaddle/PaddleOCR" /><meta property="og:url" content="https://github.com/PaddlePaddle/PaddleOCR" /><meta property="og:description" content="Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and..." />
-
-    <link rel="assets" href="https://github.githubassets.com/">
-
-
-
-        <meta name="hostname" content="github.com">
-
-
-
-        <meta name="expected-hostname" content="github.com">
-
-
-    <meta name="enabled-features" content="IMAGE_METRIC_TRACKING">
-
-
-  <meta http-equiv="x-pjax-version" content="6c58939d21332326d08c0f7a7a642997da526ff183c0d52243bccca116e52b17" data-turbo-track="reload">
-  <meta http-equiv="x-pjax-csp-version" content="485d6a5ccbb1eeae9c86b616b4870b531f6f458e8bd5c309c40280dc4f51defb" data-turbo-track="reload">
-  <meta http-equiv="x-pjax-css-version" content="4219d8fbdfa0920ce6d03f0d9d444f678ff205e0a832d33151d083532f704774" data-turbo-track="reload">
-  <meta http-equiv="x-pjax-js-version" content="29ca8c5c927eacd73f2fe0e83209b72749f0b338ac4cfd01387c31f0f4001844" data-turbo-track="reload">
-
-  <meta name="turbo-cache-control" content="no-preview"></meta>
-
-
-  <meta name="go-import" content="github.com/PaddlePaddle/PaddleOCR git https://github.com/PaddlePaddle/PaddleOCR.git">
-
-  <meta name="octolytics-dimension-user_id" content="23534030" /><meta name="octolytics-dimension-user_login" content="PaddlePaddle" /><meta name="octolytics-dimension-repository_id" content="262296122" /><meta name="octolytics-dimension-repository_nwo" content="PaddlePaddle/PaddleOCR" /><meta name="octolytics-dimension-repository_public" content="true" /><meta name="octolytics-dimension-repository_is_fork" content="false" /><meta name="octolytics-dimension-repository_network_root_id" content="262296122" /><meta name="octolytics-dimension-repository_network_root_nwo" content="PaddlePaddle/PaddleOCR" />
-
-
-
-    <link rel="canonical" href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-pjax-transient>
-
-
-  <meta name="browser-stats-url" content="https://api.github.com/_private/browser/stats">
-
-  <meta name="browser-errors-url" content="https://api.github.com/_private/browser/errors">
-
-  <meta name="browser-optimizely-client-errors-url" content="https://api.github.com/_private/browser/optimizely_client/errors">
-
-  <link rel="mask-icon" href="https://github.githubassets.com/pinned-octocat.svg" color="#000000">
-  <link rel="alternate icon" class="js-site-favicon" type="image/png" href="https://github.githubassets.com/favicons/favicon.png">
-  <link rel="icon" class="js-site-favicon" type="image/svg+xml" href="https://github.githubassets.com/favicons/favicon.svg">
-
-<meta name="theme-color" content="#1e2327">
-<meta name="color-scheme" content="light dark" />
-
-
-  <link rel="manifest" href="/manifest.json" crossOrigin="use-credentials">
-
-  </head>
-
-  <body class="logged-out env-production page-responsive page-blob" style="word-wrap: break-word;">
-
-
-    <div class="position-relative js-header-wrapper ">
-      <a href="#start-of-content" class="px-2 py-4 color-bg-accent-emphasis color-fg-on-emphasis show-on-focus js-skip-to-content">Skip to content</a>
-      <span data-view-component="true" class="progress-pjax-loader js-pjax-loader-bar Progress position-fixed width-full">
-    <span style="width: 0%;" data-view-component="true" class="Progress-item progress-pjax-loader-bar left-0 top-0 color-bg-accent-emphasis"></span>
-</span>  
-
-
-
-
-
-            <script crossorigin="anonymous" defer="defer" type="application/javascript" integrity="sha512-Z1hiPkiRLA+tojAnxZLHuRa1IrpZhnqK66AX75QbEn6PgZIpDJnWcJhQ0pOU2cI6Pe9yO0i3s4JIc9G//vkosQ==" src="https://github.githubassets.com/assets/sessions-6758623e4891.js"></script>
-<header class="Header-old header-logged-out js-details-container Details position-relative f4 py-2" role="banner">
-  <div class="container-xl d-lg-flex flex-items-center p-responsive">
-    <div class="d-flex flex-justify-between flex-items-center">
-      <a class="mr-4 color-fg-inherit" href="https://github.com/" aria-label="Homepage" data-ga-click="(Logged out) Header, go to homepage, icon:logo-wordmark">
-        <svg height="32" aria-hidden="true" viewBox="0 0 16 16" version="1.1" width="32" data-view-component="true" class="octicon octicon-mark-github">
-    <path fill-rule="evenodd" d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"></path>
-</svg>
-      </a>
-
-        <div class="d-lg-none css-truncate css-truncate-target width-fit p-2">
-
-
-        </div>
-
-      <div class="d-flex flex-items-center">
-            <a href="/signup?ref_cta=Sign+up&amp;ref_loc=header+logged+out&amp;ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E%2Fblob%2Fshow&amp;source=header-repo"
-              class="d-inline-block d-lg-none f5 no-underline border color-border-default rounded-2 px-2 py-1 mr-3 mr-sm-5 color-fg-inherit"
-              data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="e697b45b9e440c22e4e3985c3ab81484e1433b0710fb937a52cb9aad23cfc3ca"
-            >
-              Sign&nbsp;up
-            </a>
-
-        <button aria-label="Toggle navigation" aria-expanded="false" type="button" data-view-component="true" class="js-details-target btn-link d-lg-none mt-1 color-fg-inherit">  <svg aria-hidden="true" height="24" viewBox="0 0 16 16" version="1.1" width="24" data-view-component="true" class="octicon octicon-three-bars">
-    <path fill-rule="evenodd" d="M1 2.75A.75.75 0 011.75 2h12.5a.75.75 0 110 1.5H1.75A.75.75 0 011 2.75zm0 5A.75.75 0 011.75 7h12.5a.75.75 0 110 1.5H1.75A.75.75 0 011 7.75zM1.75 12a.75.75 0 100 1.5h12.5a.75.75 0 100-1.5H1.75z"></path>
-</svg>
-
-</button>      </div>
-    </div>
-
-    <div class="HeaderMenu HeaderMenu--logged-out position-fixed top-0 right-0 bottom-0 height-fit position-lg-relative d-lg-flex flex-justify-between flex-items-center flex-auto">
-      <div class="d-flex d-lg-none flex-justify-end border-bottom color-bg-subtle p-3">
-        <button aria-label="Toggle navigation" aria-expanded="false" type="button" data-view-component="true" class="js-details-target btn-link">  <svg aria-hidden="true" height="24" viewBox="0 0 24 24" version="1.1" width="24" data-view-component="true" class="octicon octicon-x color-fg-muted">
-    <path fill-rule="evenodd" d="M5.72 5.72a.75.75 0 011.06 0L12 10.94l5.22-5.22a.75.75 0 111.06 1.06L13.06 12l5.22 5.22a.75.75 0 11-1.06 1.06L12 13.06l-5.22 5.22a.75.75 0 01-1.06-1.06L10.94 12 5.72 6.78a.75.75 0 010-1.06z"></path>
-</svg>
-
-</button>      </div>
-
-        <nav class="mt-0 px-3 px-lg-0 mb-5 mb-lg-0" aria-label="Global">
-          <ul class="d-lg-flex list-style-none">
-              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <details class="HeaderMenu-details details-overlay details-reset width-full">
-      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
-        Product
-        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
-      </summary>
-      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
-        <ul class="list-style-none f5 pb-1">
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Features&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Features;&quot;}" href="/features">
-      Features
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Mobile&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Mobile;&quot;}" href="/mobile">
-      Mobile
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Actions&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Actions;&quot;}" href="/features/actions">
-      Actions
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Codespaces&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Codespaces;&quot;}" href="/features/codespaces">
-      Codespaces
-</a>  </li>
-
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Packages&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Packages;&quot;}" href="/features/packages">
-      Packages
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Security&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Security;&quot;}" href="/features/security">
-      Security
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Code review&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Code review;&quot;}" href="/features/code-review">
-      Code review
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Issues&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Issues;&quot;}" href="/features/issues">
-      Issues
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Integrations&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Integrations;&quot;}" href="/features/integrations">
-      Integrations
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold border-top pt-4 pb-2 mt-3" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to GitHub Sponsors&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Sponsors;&quot;}" href="/sponsors">
-      GitHub Sponsors
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Product&quot;,&quot;action&quot;:&quot;click to go to Customer stories&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Customer stories;&quot;}" href="/customer-stories">
-      Customer stories
-</a>  </li>
-
-        </ul>
-      </div>
-    </details>
-</li>
-
-
-              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Team&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Team;&quot;}" href="/team">Team</a>
-</li>
-
-              <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Enterprise&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Enterprise;&quot;}" href="/enterprise">Enterprise</a>
-</li>
-
-
-            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <details class="HeaderMenu-details details-overlay details-reset width-full">
-      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
-        Explore
-        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
-      </summary>
-      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
-        <ul class="list-style-none f5 pb-1">
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Explore GitHub&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Explore GitHub;&quot;}" href="/explore">
-      Explore GitHub
-</a>  </li>
-
-              <li class="color-fg-muted text-normal f6 text-mono mb-1 border-top pt-3 mt-3 mb-1">Learn and contribute</li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Topics&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Topics;&quot;}" href="/topics">
-      Topics
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Collections&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Collections;&quot;}" href="/collections">
-      Collections
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Trending&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Trending;&quot;}" href="/trending">
-      Trending
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Skills&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Skills;&quot;}" href="https://skills.github.com/">
-      Skills
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Sponsors&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Sponsors;&quot;}" href="/sponsors/explore">
-      GitHub Sponsors
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Open source guides&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Open source guides;&quot;}" href="https://opensource.guide">
-      Open source guides
-</a>  </li>
-
-              <li class="color-fg-muted text-normal f6 text-mono mb-1 border-top pt-3 mt-3 mb-1">Connect with others</li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to The ReadME Project&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:The ReadME Project;&quot;}" href="/readme">
-      The ReadME Project
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Events&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Events;&quot;}" href="/events">
-      Events
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to Community forum&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Community forum;&quot;}" href="https://github.community">
-      Community forum
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Education&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Education;&quot;}" href="https://education.github.com">
-      GitHub Education
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Explore&quot;,&quot;action&quot;:&quot;click to go to GitHub Stars program&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:GitHub Stars program;&quot;}" href="https://stars.github.com">
-      GitHub Stars program
-</a>  </li>
-
-        </ul>
-      </div>
-    </details>
-</li>
-
-
-            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <a class="HeaderMenu-link no-underline py-3 d-block d-lg-inline-block" data-analytics-event="{&quot;category&quot;:&quot;Header menu top item (logged out)&quot;,&quot;action&quot;:&quot;click to go to Marketplace&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Marketplace;&quot;}" href="/marketplace">Marketplace</a>
-</li>
-
-
-            <li class="mr-0 mr-lg-3 position-relative flex-wrap flex-justify-between flex-items-center border-bottom border-lg-bottom-0 d-block d-lg-flex flex-lg-nowrap flex-lg-items-center">
-    <details class="HeaderMenu-details details-overlay details-reset width-full">
-      <summary class="HeaderMenu-summary HeaderMenu-link px-0 py-3 border-0 no-wrap d-block d-lg-inline-block">
-        Pricing
-        <svg x="0" y="0" viewBox="0 0 14 8" xml:space="preserve" fill="none" class="icon-chevon-down-mktg position-absolute position-lg-relative"><path d="M1,1l6.2,6L13,1"></path></svg>
-      </summary>
-      <div class="dropdown-menu flex-auto rounded px-0 mt-0 pb-4 p-lg-4 position-relative position-lg-absolute left-0 left-lg-n4">
-        <ul class="list-style-none f5 pb-1">
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Plans&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Plans;&quot;}" href="/pricing">
-      Plans
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Compare plans&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Compare plans;&quot;}" href="/pricing#compare-features">
-      Compare plans
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--secondary py-2" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Contact Sales&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Contact Sales;&quot;}" href="https://github.com/enterprise/contact">
-      Contact Sales
-</a>  </li>
-
-              <li>
-    <a class="lh-condensed-ultra d-block no-underline position-relative Link--primary text-bold border-top pt-4 pb-2 mt-3" data-analytics-event="{&quot;category&quot;:&quot;Header dropdown (logged out), Pricing&quot;,&quot;action&quot;:&quot;click to go to Education&quot;,&quot;label&quot;:&quot;ref_page:/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md;ref_cta:Education;&quot;}" href="https://education.github.com">
-      Education
-</a>  </li>
-
-        </ul>
-      </div>
-    </details>
-</li>
-
-          </ul>
-        </nav>
-
-      <div class="d-lg-flex flex-items-center px-3 px-lg-0 text-center text-lg-left">
-          <div class="d-lg-flex min-width-0 mb-3 mb-lg-0">
-
-
-
-
-<div class="header-search flex-auto js-site-search position-relative flex-self-stretch flex-md-self-auto mb-3 mb-md-0 mr-0 mr-md-3 scoped-search site-scoped-search js-jump-to"
->
-  <div class="position-relative">
-    <!-- '"` --><!-- </textarea></xmp> --></option></form><form class="js-site-search-form" role="search" aria-label="Site" data-scope-type="Repository" data-scope-id="262296122" data-scoped-search-url="/PaddlePaddle/PaddleOCR/search" data-owner-scoped-search-url="/orgs/PaddlePaddle/search" data-unscoped-search-url="/search" data-turbo="false" action="/PaddlePaddle/PaddleOCR/search" accept-charset="UTF-8" method="get">
-      <label class="form-control input-sm header-search-wrapper p-0 js-chromeless-input-container header-search-wrapper-jump-to position-relative d-flex flex-justify-between flex-items-center">
-        <input type="text"
-          class="form-control input-sm header-search-input jump-to-field js-jump-to-field js-site-search-focus js-site-search-field is-clearable"
-          data-hotkey=s,/
-          name="q"
-          data-test-selector="nav-search-input"
-          placeholder="Search"
-          data-unscoped-placeholder="Search GitHub"
-          data-scoped-placeholder="Search"
-          autocapitalize="off"
-          role="combobox"
-          aria-haspopup="listbox"
-          aria-expanded="false"
-          aria-autocomplete="list"
-          aria-controls="jump-to-results"
-          aria-label="Search"
-          data-jump-to-suggestions-path="/_graphql/GetSuggestedNavigationDestinations"
-          spellcheck="false"
-          autocomplete="off"
-        >
-        <input type="hidden" data-csrf="true" class="js-data-jump-to-suggestions-path-csrf" value="3sIzmZY19XiVvvmqOda8K/tw/otB4voFktXu/QVkH/wlFcIW1DALJozQM+9OJnW6NJEvC+nr88icYnU0YgDkJA==" />
-        <input type="hidden" class="js-site-search-type-field" name="type" >
-            <svg xmlns="http://www.w3.org/2000/svg" width="22" height="20" aria-hidden="true" class="mr-1 header-search-key-slash"><path fill="none" stroke="#979A9C" opacity=".4" d="M3.5.5h12c1.7 0 3 1.3 3 3v13c0 1.7-1.3 3-3 3h-12c-1.7 0-3-1.3-3-3v-13c0-1.7 1.3-3 3-3z"></path><path fill="#979A9C" d="M11.8 6L8 15.1h-.9L10.8 6h1z"></path></svg>
-
-
-          <div class="Box position-absolute overflow-hidden d-none jump-to-suggestions js-jump-to-suggestions-container">
-
-<ul class="d-none js-jump-to-suggestions-template-container">
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-suggestion" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="suggestion">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
-        In this repository
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-</ul>
-
-<ul class="d-none js-jump-to-no-results-template-container">
-  <li class="d-flex flex-justify-center flex-items-center f5 d-none js-jump-to-suggestion p-2">
-    <span class="color-fg-muted">No suggested jump to results</span>
-  </li>
-</ul>
-
-<ul id="jump-to-results" role="listbox" class="p-0 m-0 js-navigation-container jump-to-suggestions-results-container js-jump-to-suggestions-results-container">
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-scoped-search d-none" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="scoped_search">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
-        In this repository
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-owner-scoped-search d-none" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="owner_scoped_search">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this organization">
-        In this organization
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-
-
-<li class="d-flex flex-justify-start flex-items-center p-0 f5 navigation-item js-navigation-item js-jump-to-global-search d-none" role="option">
-  <a tabindex="-1" class="no-underline d-flex flex-auto flex-items-center jump-to-suggestions-path js-jump-to-suggestion-path js-navigation-open p-2" href="" data-item-type="global_search">
-    <div class="jump-to-octicon js-jump-to-octicon flex-shrink-0 mr-2 text-center d-none">
-      <svg title="Repository" aria-label="Repository" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo js-jump-to-octicon-repo d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-      <svg title="Project" aria-label="Project" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-project js-jump-to-octicon-project d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M1.75 0A1.75 1.75 0 000 1.75v12.5C0 15.216.784 16 1.75 16h12.5A1.75 1.75 0 0016 14.25V1.75A1.75 1.75 0 0014.25 0H1.75zM1.5 1.75a.25.25 0 01.25-.25h12.5a.25.25 0 01.25.25v12.5a.25.25 0 01-.25.25H1.75a.25.25 0 01-.25-.25V1.75zM11.75 3a.75.75 0 00-.75.75v7.5a.75.75 0 001.5 0v-7.5a.75.75 0 00-.75-.75zm-8.25.75a.75.75 0 011.5 0v5.5a.75.75 0 01-1.5 0v-5.5zM8 3a.75.75 0 00-.75.75v3.5a.75.75 0 001.5 0v-3.5A.75.75 0 008 3z"></path>
-</svg>
-      <svg title="Search" aria-label="Search" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-search js-jump-to-octicon-search d-none flex-shrink-0">
-    <path fill-rule="evenodd" d="M11.5 7a4.499 4.499 0 11-8.998 0A4.499 4.499 0 0111.5 7zm-.82 4.74a6 6 0 111.06-1.06l3.04 3.04a.75.75 0 11-1.06 1.06l-3.04-3.04z"></path>
-</svg>
-    </div>
-
-    <img class="avatar mr-2 flex-shrink-0 js-jump-to-suggestion-avatar d-none" alt="" aria-label="Team" src="" width="28" height="28">
-
-    <div class="jump-to-suggestion-name js-jump-to-suggestion-name flex-auto overflow-hidden text-left no-wrap css-truncate css-truncate-target">
-    </div>
-
-    <div class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none js-jump-to-badge-search">
-      <span class="js-jump-to-badge-search-text-default d-none" aria-label="in this repository">
-        In this repository
-      </span>
-      <span class="js-jump-to-badge-search-text-global d-none" aria-label="in all of GitHub">
-        All GitHub
-      </span>
-      <span aria-hidden="true" class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-
-    <div aria-hidden="true" class="border rounded-2 flex-shrink-0 color-bg-subtle px-1 color-fg-muted ml-1 f6 d-none d-on-nav-focus js-jump-to-badge-jump">
-      Jump to
-      <span class="d-inline-block ml-1 v-align-middle">↵</span>
-    </div>
-  </a>
-</li>
-
-
-</ul>
-
-          </div>
-      </label>
-</form>  </div>
-</div>
-
-          </div>
-
-        <div class="position-relative mr-3 mb-4 mb-lg-0 d-inline-block">
-          <a href="/login?return_to=https%3A%2F%2Fgithub.com%2FPaddlePaddle%2FPaddleOCR%2Fblob%2Fdygraph%2Fapplications%2F%25E8%25BD%25BB%25E9%2587%258F%25E7%25BA%25A7%25E8%25BD%25A6%25E7%2589%258C%25E8%25AF%2586%25E5%2588%25AB.md"
-            class="HeaderMenu-link flex-shrink-0 no-underline"
-            data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header menu&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="0de5d29e004c06e7301cb66a02d668f7cd98725fd6fa97ab9b512671e1a9905b"
-            data-ga-click="(Logged out) Header, clicked Sign in, text:sign-in">
-            Sign in
-          </a>
-        </div>
-
-          <a href="/signup?ref_cta=Sign+up&amp;ref_loc=header+logged+out&amp;ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E%2Fblob%2Fshow&amp;source=header-repo&amp;source_repo=PaddlePaddle%2FPaddleOCR"
-            class="HeaderMenu-link flex-shrink-0 d-inline-block no-underline border color-border-default rounded px-2 py-1"
-            data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;site header menu&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;SIGN_UP&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="0de5d29e004c06e7301cb66a02d668f7cd98725fd6fa97ab9b512671e1a9905b"
-            data-analytics-event="{&quot;category&quot;:&quot;Sign up&quot;,&quot;action&quot;:&quot;click to sign up for account&quot;,&quot;label&quot;:&quot;ref_page:/&lt;user-name&gt;/&lt;repo-name&gt;/blob/show;ref_cta:Sign up;ref_loc:header logged out&quot;}"
-          >
-            Sign up
-          </a>
-      </div>
-    </div>
-  </div>
-</header>
-
-    </div>
-
-  <div id="start-of-content" class="show-on-focus"></div>
-
-
-
-
-
-
-
-    <div data-pjax-replace id="js-flash-container" data-turbo-replace>
-
-
-  <template class="js-flash-template">
-    <div class="flash flash-full  {{ className }}">
-  <div class="px-2" >
-    <button class="flash-close js-flash-close" type="button" aria-label="Dismiss this message">
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-    </button>
-
-      <div>{{ message }}</div>
-
-  </div>
-</div>
-  </template>
-</div>
-
-
-
-  <include-fragment class="js-notification-shelf-include-fragment" data-base-src="https://github.com/notifications/beta/shelf"></include-fragment>
-
-
-
-
-
-  <div
-    class="application-main "
-    data-commit-hovercards-enabled
-    data-discussion-hovercards-enabled
-    data-issue-and-pr-hovercards-enabled
-  >
-        <div itemscope itemtype="http://schema.org/SoftwareSourceCode" class="">
-    <main id="js-repo-pjax-container" data-pjax-container >
-
-
-
-
-
-
-
-
-
-
-
-
-
-  <div id="repository-container-header" class="pt-3 hide-full-screen" style="background-color: var(--color-page-header-bg);" data-pjax-replace data-turbo-replace>
-
-      <div class="d-flex mb-3 px-3 px-md-4 px-lg-5">
-
-        <div class="flex-auto min-width-0 width-fit mr-3">
-            <div class=" d-flex flex-wrap flex-items-center wb-break-word f3 text-normal">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo color-fg-muted mr-2">
-    <path fill-rule="evenodd" d="M2 2.5A2.5 2.5 0 014.5 0h8.75a.75.75 0 01.75.75v12.5a.75.75 0 01-.75.75h-2.5a.75.75 0 110-1.5h1.75v-2h-8a1 1 0 00-.714 1.7.75.75 0 01-1.072 1.05A2.495 2.495 0 012 11.5v-9zm10.5-1V9h-8c-.356 0-.694.074-1 .208V2.5a1 1 0 011-1h8zM5 12.25v3.25a.25.25 0 00.4.2l1.45-1.087a.25.25 0 01.3 0L8.6 15.7a.25.25 0 00.4-.2v-3.25a.25.25 0 00-.25-.25h-3.5a.25.25 0 00-.25.25z"></path>
-</svg>
-  <span class="author flex-self-stretch" itemprop="author">
-    <a class="url fn" rel="author" data-hovercard-type="organization" data-hovercard-url="/orgs/PaddlePaddle/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle">PaddlePaddle</a>
-  </span>
-  <span class="mx-1 flex-self-stretch color-fg-muted">/</span>
-  <strong itemprop="name" class="mr-2 flex-self-stretch">
-    <a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR">PaddleOCR</a>
-  </strong>
-
-  <span></span><span class="Label Label--secondary v-align-middle mr-1">Public</span>
-</div>
-
-        </div>
-
-          <ul class="pagehead-actions flex-shrink-0 d-none d-md-inline" style="padding: 2px 0;">
-
-
-
-  <li>
-      <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;notification subscription menu watch&quot;,&quot;repository_id&quot;:null,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="2c00da16a0d8d066b49f1f6e26149784faffba39c427495160ebf3bac257b4a9" aria-label="You must be signed in to change notification settings" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-bell mr-2">
-    <path d="M8 16a2 2 0 001.985-1.75c.017-.137-.097-.25-.235-.25h-3.5c-.138 0-.252.113-.235.25A2 2 0 008 16z"></path><path fill-rule="evenodd" d="M8 1.5A3.5 3.5 0 004.5 5v2.947c0 .346-.102.683-.294.97l-1.703 2.556a.018.018 0 00-.003.01l.001.006c0 .002.002.004.004.006a.017.017 0 00.006.004l.007.001h10.964l.007-.001a.016.016 0 00.006-.004.016.016 0 00.004-.006l.001-.007a.017.017 0 00-.003-.01l-1.703-2.554a1.75 1.75 0 01-.294-.97V5A3.5 3.5 0 008 1.5zM3 5a5 5 0 0110 0v2.947c0 .05.015.098.042.139l1.703 2.555A1.518 1.518 0 0113.482 13H2.518a1.518 1.518 0 01-1.263-2.36l1.703-2.554A.25.25 0 003 7.947V5z"></path>
-</svg>Notifications
-
-</a>
-  </li>
-
-  <li>
-        <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;repo details fork button&quot;,&quot;repository_id&quot;:262296122,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f55097530c92c084f3a79350400121a7c267c872da2b69858747a12db51b5c71" aria-label="You must be signed in to fork a repository" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-repo-forked mr-2">
-    <path fill-rule="evenodd" d="M5 3.25a.75.75 0 11-1.5 0 .75.75 0 011.5 0zm0 2.122a2.25 2.25 0 10-1.5 0v.878A2.25 2.25 0 005.75 8.5h1.5v2.128a2.251 2.251 0 101.5 0V8.5h1.5a2.25 2.25 0 002.25-2.25v-.878a2.25 2.25 0 10-1.5 0v.878a.75.75 0 01-.75.75h-4.5A.75.75 0 015 6.25v-.878zm3.75 7.378a.75.75 0 11-1.5 0 .75.75 0 011.5 0zm3-8.75a.75.75 0 100-1.5.75.75 0 000 1.5z"></path>
-</svg>Fork
-    <span id="repo-network-counter" data-pjax-replace="true" title="4,668" data-view-component="true" class="Counter">4.7k</span>
-
-</a>
-  </li>
-
-  <li>
-        <div data-view-component="true" class="BtnGroup d-flex">
-      <a href="/login?return_to=%2FPaddlePaddle%2FPaddleOCR" rel="nofollow" data-hydro-click="{&quot;event_type&quot;:&quot;authentication.click&quot;,&quot;payload&quot;:{&quot;location_in_page&quot;:&quot;star button&quot;,&quot;repository_id&quot;:262296122,&quot;auth_type&quot;:&quot;LOG_IN&quot;,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="93ad0192b3e15b4610691426f776e3dab114d4c7666db7f44c0d706f50c74c23" aria-label="You must be signed in to star a repository" data-view-component="true" class="tooltipped tooltipped-s btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-star v-align-text-bottom d-inline-block mr-2">
-    <path fill-rule="evenodd" d="M8 .25a.75.75 0 01.673.418l1.882 3.815 4.21.612a.75.75 0 01.416 1.279l-3.046 2.97.719 4.192a.75.75 0 01-1.088.791L8 12.347l-3.766 1.98a.75.75 0 01-1.088-.79l.72-4.194L.818 6.374a.75.75 0 01.416-1.28l4.21-.611L7.327.668A.75.75 0 018 .25zm0 2.445L6.615 5.5a.75.75 0 01-.564.41l-3.097.45 2.24 2.184a.75.75 0 01.216.664l-.528 3.084 2.769-1.456a.75.75 0 01.698 0l2.77 1.456-.53-3.084a.75.75 0 01.216-.664l2.24-2.183-3.096-.45a.75.75 0 01-.564-.41L8 2.694v.001z"></path>
-</svg><span data-view-component="true" class="d-inline">
-          Star
-</span>          <span id="repo-stars-counter-star" aria-label="22861 users starred this repository" data-singular-suffix="user starred this repository" data-plural-suffix="users starred this repository" data-pjax-replace="true" data-turbo-replace="true" title="22,861" data-view-component="true" class="Counter js-social-count">22.9k</span>
-
-</a>      <button disabled="disabled" aria-label="You must be signed in to add this repository to a list" type="button" data-view-component="true" class="btn-sm btn BtnGroup-item px-2">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-triangle-down">
-    <path d="M4.427 7.427l3.396 3.396a.25.25 0 00.354 0l3.396-3.396A.25.25 0 0011.396 7H4.604a.25.25 0 00-.177.427z"></path>
-</svg>
-
-</button></div>
-  </li>
-
-
-
-</ul>
-
-      </div>
-
-      <div id="responsive-meta-container" data-pjax-replace data-turbo-replace>
-</div>
-
-
-
-<nav data-pjax="#js-repo-pjax-container" aria-label="Repository" data-view-component="true" class="js-repo-nav js-sidenav-container-pjax js-responsive-underlinenav overflow-hidden UnderlineNav px-3 px-md-4 px-lg-5">
-
-  <ul data-view-component="true" class="UnderlineNav-body list-style-none">
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="code-tab" href="/PaddlePaddle/PaddleOCR/tree/dygraph" data-tab-item="i0code-tab" data-selected-links="repo_source repo_downloads repo_commits repo_releases repo_tags repo_branches repo_packages repo_deployments /PaddlePaddle/PaddleOCR/tree/dygraph" data-pjax="#repo-content-pjax-container" data-hotkey="g c" data-ga-click="Repository, Navigation click, Code tab" aria-current="page" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item selected">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-code UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M4.72 3.22a.75.75 0 011.06 1.06L2.06 8l3.72 3.72a.75.75 0 11-1.06 1.06L.47 8.53a.75.75 0 010-1.06l4.25-4.25zm6.56 0a.75.75 0 10-1.06 1.06L13.94 8l-3.72 3.72a.75.75 0 101.06 1.06l4.25-4.25a.75.75 0 000-1.06l-4.25-4.25z"></path>
-</svg>
-          <span data-content="Code">Code</span>
-            <span id="code-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="issues-tab" href="/PaddlePaddle/PaddleOCR/issues" data-tab-item="i1issues-tab" data-selected-links="repo_issues repo_labels repo_milestones /PaddlePaddle/PaddleOCR/issues" data-pjax="#repo-content-pjax-container" data-hotkey="g i" data-ga-click="Repository, Navigation click, Issues tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-issue-opened UnderlineNav-octicon d-none d-sm-inline">
-    <path d="M8 9.5a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path><path fill-rule="evenodd" d="M8 0a8 8 0 100 16A8 8 0 008 0zM1.5 8a6.5 6.5 0 1113 0 6.5 6.5 0 01-13 0z"></path>
-</svg>
-          <span data-content="Issues">Issues</span>
-            <span id="issues-repo-tab-count" data-pjax-replace="" title="1,078" data-view-component="true" class="Counter">1.1k</span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="pull-requests-tab" href="/PaddlePaddle/PaddleOCR/pulls" data-tab-item="i2pull-requests-tab" data-selected-links="repo_pulls checks /PaddlePaddle/PaddleOCR/pulls" data-pjax="#repo-content-pjax-container" data-hotkey="g p" data-ga-click="Repository, Navigation click, Pull requests tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-git-pull-request UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M7.177 3.073L9.573.677A.25.25 0 0110 .854v4.792a.25.25 0 01-.427.177L7.177 3.427a.25.25 0 010-.354zM3.75 2.5a.75.75 0 100 1.5.75.75 0 000-1.5zm-2.25.75a2.25 2.25 0 113 2.122v5.256a2.251 2.251 0 11-1.5 0V5.372A2.25 2.25 0 011.5 3.25zM11 2.5h-1V4h1a1 1 0 011 1v5.628a2.251 2.251 0 101.5 0V5A2.5 2.5 0 0011 2.5zm1 10.25a.75.75 0 111.5 0 .75.75 0 01-1.5 0zM3.75 12a.75.75 0 100 1.5.75.75 0 000-1.5z"></path>
-</svg>
-          <span data-content="Pull requests">Pull requests</span>
-            <span id="pull-requests-repo-tab-count" data-pjax-replace="" title="98" data-view-component="true" class="Counter">98</span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="discussions-tab" href="/PaddlePaddle/PaddleOCR/discussions" data-tab-item="i3discussions-tab" data-selected-links="repo_discussions /PaddlePaddle/PaddleOCR/discussions" data-pjax="#repo-content-pjax-container" data-hotkey="g g" data-ga-click="Repository, Navigation click, Discussions tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-comment-discussion UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M1.5 2.75a.25.25 0 01.25-.25h8.5a.25.25 0 01.25.25v5.5a.25.25 0 01-.25.25h-3.5a.75.75 0 00-.53.22L3.5 11.44V9.25a.75.75 0 00-.75-.75h-1a.25.25 0 01-.25-.25v-5.5zM1.75 1A1.75 1.75 0 000 2.75v5.5C0 9.216.784 10 1.75 10H2v1.543a1.457 1.457 0 002.487 1.03L7.061 10h3.189A1.75 1.75 0 0012 8.25v-5.5A1.75 1.75 0 0010.25 1h-8.5zM14.5 4.75a.25.25 0 00-.25-.25h-.5a.75.75 0 110-1.5h.5c.966 0 1.75.784 1.75 1.75v5.5A1.75 1.75 0 0114.25 12H14v1.543a1.457 1.457 0 01-2.487 1.03L9.22 12.28a.75.75 0 111.06-1.06l2.22 2.22v-2.19a.75.75 0 01.75-.75h1a.25.25 0 00.25-.25v-5.5z"></path>
-</svg>
-          <span data-content="Discussions">Discussions</span>
-            <span id="discussions-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="actions-tab" href="/PaddlePaddle/PaddleOCR/actions" data-tab-item="i4actions-tab" data-selected-links="repo_actions /PaddlePaddle/PaddleOCR/actions" data-pjax="#repo-content-pjax-container" data-hotkey="g a" data-analytics-event="{&quot;category&quot;:&quot;Actions&quot;,&quot;action&quot;:&quot;clicked&quot;,&quot;label&quot;:&quot;ref_cta:Actions;ref_loc:navigation_helper&quot;}" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-play UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M1.5 8a6.5 6.5 0 1113 0 6.5 6.5 0 01-13 0zM8 0a8 8 0 100 16A8 8 0 008 0zM6.379 5.227A.25.25 0 006 5.442v5.117a.25.25 0 00.379.214l4.264-2.559a.25.25 0 000-.428L6.379 5.227z"></path>
-</svg>
-          <span data-content="Actions">Actions</span>
-            <span id="actions-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="projects-tab" href="/PaddlePaddle/PaddleOCR/projects?type=beta" data-tab-item="i5projects-tab" data-selected-links="repo_projects new_repo_project repo_project /PaddlePaddle/PaddleOCR/projects?type=beta" data-pjax="#repo-content-pjax-container" data-hotkey="g b" data-ga-click="Repository, Navigation click, Projects tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-table UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M0 1.75C0 .784.784 0 1.75 0h12.5C15.216 0 16 .784 16 1.75v3.585a.746.746 0 010 .83v8.085A1.75 1.75 0 0114.25 16H6.309a.748.748 0 01-1.118 0H1.75A1.75 1.75 0 010 14.25V6.165a.746.746 0 010-.83V1.75zM1.5 6.5v7.75c0 .138.112.25.25.25H5v-8H1.5zM5 5H1.5V1.75a.25.25 0 01.25-.25H5V5zm1.5 1.5v8h7.75a.25.25 0 00.25-.25V6.5h-8zm8-1.5h-8V1.5h7.75a.25.25 0 01.25.25V5z"></path>
-</svg>
-          <span data-content="Projects">Projects</span>
-            <span id="projects-repo-tab-count" data-pjax-replace="" title="0" hidden="hidden" data-view-component="true" class="Counter">0</span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="wiki-tab" href="/PaddlePaddle/PaddleOCR/wiki" data-tab-item="i6wiki-tab" data-selected-links="repo_wiki /PaddlePaddle/PaddleOCR/wiki" data-pjax="#repo-content-pjax-container" data-hotkey="g w" data-ga-click="Repository, Navigation click, Wikis tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-book UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M0 1.75A.75.75 0 01.75 1h4.253c1.227 0 2.317.59 3 1.501A3.744 3.744 0 0111.006 1h4.245a.75.75 0 01.75.75v10.5a.75.75 0 01-.75.75h-4.507a2.25 2.25 0 00-1.591.659l-.622.621a.75.75 0 01-1.06 0l-.622-.621A2.25 2.25 0 005.258 13H.75a.75.75 0 01-.75-.75V1.75zm8.755 3a2.25 2.25 0 012.25-2.25H14.5v9h-3.757c-.71 0-1.4.201-1.992.572l.004-7.322zm-1.504 7.324l.004-5.073-.002-2.253A2.25 2.25 0 005.003 2.5H1.5v9h3.757a3.75 3.75 0 011.994.574z"></path>
-</svg>
-          <span data-content="Wiki">Wiki</span>
-            <span id="wiki-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="security-tab" href="/PaddlePaddle/PaddleOCR/security" data-tab-item="i7security-tab" data-selected-links="security overview alerts policy token_scanning code_scanning /PaddlePaddle/PaddleOCR/security" data-pjax="#repo-content-pjax-container" data-hotkey="g s" data-ga-click="Repository, Navigation click, Security tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-shield UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M7.467.133a1.75 1.75 0 011.066 0l5.25 1.68A1.75 1.75 0 0115 3.48V7c0 1.566-.32 3.182-1.303 4.682-.983 1.498-2.585 2.813-5.032 3.855a1.7 1.7 0 01-1.33 0c-2.447-1.042-4.049-2.357-5.032-3.855C1.32 10.182 1 8.566 1 7V3.48a1.75 1.75 0 011.217-1.667l5.25-1.68zm.61 1.429a.25.25 0 00-.153 0l-5.25 1.68a.25.25 0 00-.174.238V7c0 1.358.275 2.666 1.057 3.86.784 1.194 2.121 2.34 4.366 3.297a.2.2 0 00.154 0c2.245-.956 3.582-2.104 4.366-3.298C13.225 9.666 13.5 8.36 13.5 7V3.48a.25.25 0 00-.174-.237l-5.25-1.68zM9 10.5a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.75a.75.75 0 10-1.5 0v3a.75.75 0 001.5 0v-3z"></path>
-</svg>
-          <span data-content="Security">Security</span>
-            <include-fragment src="/PaddlePaddle/PaddleOCR/security/overall-count" accept="text/fragment+html"></include-fragment>
-
-
-</a></li>
-      <li data-view-component="true" class="d-inline-flex">
-  <a id="insights-tab" href="/PaddlePaddle/PaddleOCR/pulse" data-tab-item="i8insights-tab" data-selected-links="repo_graphs repo_contributors dependency_graph dependabot_updates pulse people community /PaddlePaddle/PaddleOCR/pulse" data-pjax="#repo-content-pjax-container" data-ga-click="Repository, Navigation click, Insights tab" data-view-component="true" class="UnderlineNav-item no-wrap js-responsive-underlinenav-item js-selected-navigation-item">
-
-                  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-graph UnderlineNav-octicon d-none d-sm-inline">
-    <path fill-rule="evenodd" d="M1.5 1.75a.75.75 0 00-1.5 0v12.5c0 .414.336.75.75.75h14.5a.75.75 0 000-1.5H1.5V1.75zm14.28 2.53a.75.75 0 00-1.06-1.06L10 7.94 7.53 5.47a.75.75 0 00-1.06 0L3.22 8.72a.75.75 0 001.06 1.06L7 7.06l2.47 2.47a.75.75 0 001.06 0l5.25-5.25z"></path>
-</svg>
-          <span data-content="Insights">Insights</span>
-            <span id="insights-repo-tab-count" data-pjax-replace="" title="Not available" data-view-component="true" class="Counter"></span>
-
-
-
-</a></li>
-</ul>
-    <div style="visibility:hidden;" data-view-component="true" class="UnderlineNav-actions js-responsive-underlinenav-overflow position-absolute pr-3 pr-md-4 pr-lg-5 right-0">      <details data-view-component="true" class="details-overlay details-reset position-relative">
-  <summary role="button" data-view-component="true">          <div class="UnderlineNav-item mr-0 border-0">
-            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
-    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
-</svg>
-            <span class="sr-only">More</span>
-          </div>
-</summary>
-  <div data-view-component="true">          <details-menu role="menu" data-view-component="true" class="dropdown-menu dropdown-menu-sw">
-
-            <ul>
-                <li data-menu-item="i0code-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item selected dropdown-item" aria-current="page" data-selected-links="repo_source repo_downloads repo_commits repo_releases repo_tags repo_branches repo_packages repo_deployments /PaddlePaddle/PaddleOCR/tree/dygraph" href="/PaddlePaddle/PaddleOCR/tree/dygraph">
-                    Code
-</a>                </li>
-                <li data-menu-item="i1issues-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_issues repo_labels repo_milestones /PaddlePaddle/PaddleOCR/issues" href="/PaddlePaddle/PaddleOCR/issues">
-                    Issues
-</a>                </li>
-                <li data-menu-item="i2pull-requests-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_pulls checks /PaddlePaddle/PaddleOCR/pulls" href="/PaddlePaddle/PaddleOCR/pulls">
-                    Pull requests
-</a>                </li>
-                <li data-menu-item="i3discussions-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_discussions /PaddlePaddle/PaddleOCR/discussions" href="/PaddlePaddle/PaddleOCR/discussions">
-                    Discussions
-</a>                </li>
-                <li data-menu-item="i4actions-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_actions /PaddlePaddle/PaddleOCR/actions" href="/PaddlePaddle/PaddleOCR/actions">
-                    Actions
-</a>                </li>
-                <li data-menu-item="i5projects-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_projects new_repo_project repo_project /PaddlePaddle/PaddleOCR/projects?type=beta" href="/PaddlePaddle/PaddleOCR/projects?type=beta">
-                    Projects
-</a>                </li>
-                <li data-menu-item="i6wiki-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_wiki /PaddlePaddle/PaddleOCR/wiki" href="/PaddlePaddle/PaddleOCR/wiki">
-                    Wiki
-</a>                </li>
-                <li data-menu-item="i7security-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="security overview alerts policy token_scanning code_scanning /PaddlePaddle/PaddleOCR/security" href="/PaddlePaddle/PaddleOCR/security">
-                    Security
-</a>                </li>
-                <li data-menu-item="i8insights-tab" hidden>
-                  <a role="menuitem" class="js-selected-navigation-item dropdown-item" data-selected-links="repo_graphs repo_contributors dependency_graph dependabot_updates pulse people community /PaddlePaddle/PaddleOCR/pulse" href="/PaddlePaddle/PaddleOCR/pulse">
-                    Insights
-</a>                </li>
-            </ul>
-
-</details-menu></div>
-</details></div>
-</nav>
-  </div>
-
-
-
-<div id="repo-content-pjax-container" class="repository-content " >
-
-
-
-
-
-
-  <div class="clearfix container-xl px-3 px-md-4 px-lg-5 mt-4">
-
-
-<div>
-
-
-
-
-
-
-
-
-<a class="d-none js-permalink-shortcut" data-hotkey="y" href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">Permalink</a>
-
-<div class="d-flex flex-items-start flex-shrink-0 pb-3 flex-wrap flex-md-nowrap flex-justify-between flex-md-justify-start">
-
-<div class="position-relative">
-  <details class="details-reset details-overlay mr-0 mb-0 " id="branch-select-menu">
-    <summary class="btn css-truncate"
-            data-hotkey="w"
-            title="Switch branches or tags">
-      <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-git-branch">
-    <path fill-rule="evenodd" d="M11.75 2.5a.75.75 0 100 1.5.75.75 0 000-1.5zm-2.25.75a2.25 2.25 0 113 2.122V6A2.5 2.5 0 0110 8.5H6a1 1 0 00-1 1v1.128a2.251 2.251 0 11-1.5 0V5.372a2.25 2.25 0 111.5 0v1.836A2.492 2.492 0 016 7h4a1 1 0 001-1v-.628A2.25 2.25 0 019.5 3.25zM4.25 12a.75.75 0 100 1.5.75.75 0 000-1.5zM3.5 3.25a.75.75 0 111.5 0 .75.75 0 01-1.5 0z"></path>
-</svg>
-      <span class="css-truncate-target" data-menu-button>dygraph</span>
-      <span class="dropdown-caret"></span>
-    </summary>
-
-
-<div class="SelectMenu">
-  <div class="SelectMenu-modal">
-    <header class="SelectMenu-header">
-      <span class="SelectMenu-title">Switch branches/tags</span>
-      <button class="SelectMenu-closeButton" type="button" data-toggle-for="branch-select-menu"><svg aria-label="Close menu" aria-hidden="false" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg></button>
-    </header>
-
-    <input-demux data-action="tab-container-change:input-demux#storeInput tab-container-changed:input-demux#updateInput">
-      <tab-container class="d-flex flex-column js-branches-tags-tabs" style="min-height: 0;">
-        <div class="SelectMenu-filter">
-          <input data-target="input-demux.source"
-                 id="context-commitish-filter-field"
-                 class="SelectMenu-input form-control"
-                 aria-owns="ref-list-branches"
-                 data-controls-ref-menu-id="ref-list-branches"
-                 autofocus
-                 autocomplete="off"
-                 aria-label="Filter branches/tags"
-                 placeholder="Filter branches/tags"
-                 type="text"
-          >
-        </div>
-
-        <div class="SelectMenu-tabs" role="tablist" data-target="input-demux.control" >
-          <button class="SelectMenu-tab" type="button" role="tab" aria-selected="true">Branches</button>
-          <button class="SelectMenu-tab" type="button" role="tab">Tags</button>
-        </div>
-
-        <div role="tabpanel" id="ref-list-branches" data-filter-placeholder="Filter branches/tags" tabindex="" class="d-flex flex-column flex-auto overflow-auto">
-          <ref-selector
-            type="branch"
-            data-targets="input-demux.sinks"
-            data-action="
-              input-entered:ref-selector#inputEntered
-              tab-selected:ref-selector#tabSelected
-              focus-list:ref-selector#focusFirstListMember
-            "
-            query-endpoint="/PaddlePaddle/PaddleOCR/refs"
-
-            cache-key="v0:1655177037.9190001"
-            current-committish="ZHlncmFwaA=="
-            default-branch="cmVsZWFzZS8yLjU="
-            name-with-owner="UGFkZGxlUGFkZGxlL1BhZGRsZU9DUg=="
-            prefetch-on-mouseover
-          >
-
-            <template data-target="ref-selector.fetchFailedTemplate">
-              <div class="SelectMenu-message" data-index="{{ index }}">Could not load branches</div>
-            </template>
-
-              <template data-target="ref-selector.noMatchTemplate">
-    <div class="SelectMenu-message">Nothing to show</div>
-</template>
-
-
-            <div data-target="ref-selector.listContainer" role="menu" class="SelectMenu-list " data-pjax="#repo-content-pjax-container">
-              <div class="SelectMenu-loading pt-3 pb-0 overflow-hidden" aria-label="Menu is loading">
-                <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="anim-rotate">
-  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
-  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
-</svg>
-              </div>
-            </div>
-
-              <template data-target="ref-selector.itemTemplate">
-  <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{ urlEncodedRefName }}/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="SelectMenu-item" role="menuitemradio" rel="nofollow" aria-checked="{{ isCurrent }}" data-index="{{ index }}">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check SelectMenu-icon SelectMenu-icon--check">
-    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
-</svg>
-    <span class="flex-1 css-truncate css-truncate-overflow {{ isFilteringClass }}">{{ refName }}</span>
-    <span hidden="{{ isNotDefault }}" class="Label Label--secondary flex-self-start">default</span>
-  </a>
-</template>
-
-
-              <footer class="SelectMenu-footer"><a href="/PaddlePaddle/PaddleOCR/branches">View all branches</a></footer>
-          </ref-selector>
-
-        </div>
-
-        <div role="tabpanel" id="tags-menu" data-filter-placeholder="Find a tag" tabindex="" hidden class="d-flex flex-column flex-auto overflow-auto">
-          <ref-selector
-            type="tag"
-            data-action="
-              input-entered:ref-selector#inputEntered
-              tab-selected:ref-selector#tabSelected
-              focus-list:ref-selector#focusFirstListMember
-            "
-            data-targets="input-demux.sinks"
-            query-endpoint="/PaddlePaddle/PaddleOCR/refs"
-            cache-key="v0:1655177037.9190001"
-            current-committish="ZHlncmFwaA=="
-            default-branch="cmVsZWFzZS8yLjU="
-            name-with-owner="UGFkZGxlUGFkZGxlL1BhZGRsZU9DUg=="
-          >
-
-            <template data-target="ref-selector.fetchFailedTemplate">
-              <div class="SelectMenu-message" data-index="{{ index }}">Could not load tags</div>
-            </template>
-
-            <template data-target="ref-selector.noMatchTemplate">
-              <div class="SelectMenu-message" data-index="{{ index }}">Nothing to show</div>
-            </template>
-
-              <template data-target="ref-selector.itemTemplate">
-  <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{ urlEncodedRefName }}/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="SelectMenu-item" role="menuitemradio" rel="nofollow" aria-checked="{{ isCurrent }}" data-index="{{ index }}">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check SelectMenu-icon SelectMenu-icon--check">
-    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
-</svg>
-    <span class="flex-1 css-truncate css-truncate-overflow {{ isFilteringClass }}">{{ refName }}</span>
-    <span hidden="{{ isNotDefault }}" class="Label Label--secondary flex-self-start">default</span>
-  </a>
-</template>
-
-
-            <div data-target="ref-selector.listContainer" role="menu" class="SelectMenu-list" data-pjax="#repo-content-pjax-container">
-              <div class="SelectMenu-loading pt-3 pb-0 overflow-hidden" aria-label="Menu is loading">
-                <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="anim-rotate">
-  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
-  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
-</svg>
-              </div>
-            </div>
-              <footer class="SelectMenu-footer"><a href="/PaddlePaddle/PaddleOCR/tags">View all tags</a></footer>
-          </ref-selector>
-        </div>
-      </tab-container>
-    </input-demux>
-  </div>
-</div>
-
-  </details>
-
-</div>
-
-  <h2 id="blob-path" class="breadcrumb flex-auto flex-self-center min-width-0 text-normal mx-2 width-full width-md-auto flex-order-1 flex-md-order-none mt-3 mt-md-0">
-    <span class="js-repo-root text-bold"><span class="js-path-segment d-inline-block wb-break-all"><a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR/tree/dygraph"><span>PaddleOCR</span></a></span></span><span class="separator">/</span><span class="js-path-segment d-inline-block wb-break-all"><a data-pjax="#repo-content-pjax-container" href="/PaddlePaddle/PaddleOCR/tree/dygraph/applications"><span>applications</span></a></span><span class="separator">/</span><strong class="final-path">轻量级车牌识别.md</strong>
-  </h2>
-  <a href="/PaddlePaddle/PaddleOCR/find/dygraph" data-pjax="" data-hotkey="t" data-view-component="true" class="btn mr-2 d-none d-md-block">  Go to file
-
-</a>
-  <details id="blob-more-options-details" data-view-component="true" class="details-overlay details-reset position-relative">
-  <summary role="button" data-view-component="true" class="btn">  <svg aria-label="More options" role="img" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
-    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
-</svg>
-
-</summary>
-  <div data-view-component="true">      <ul class="dropdown-menu dropdown-menu-sw">
-        <li class="d-block d-md-none">
-          <a class="dropdown-item d-flex flex-items-baseline" data-hydro-click="{&quot;event_type&quot;:&quot;repository.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;FIND_FILE_BUTTON&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="719caca13290ec1595dbf25afc93e9874844c0bcf4645d121411be9e6fbaa205" data-ga-click="Repository, find file, location:repo overview" data-hotkey="t" data-pjax="true" href="/PaddlePaddle/PaddleOCR/find/dygraph">
-            <span class="flex-auto">Go to file</span>
-            <span class="text-small color-fg-muted" aria-hidden="true">T</span>
-</a>        </li>
-        <li data-toggle-for="blob-more-options-details">
-          <button data-toggle-for="jumpto-line-details-dialog" type="button" data-view-component="true" class="dropdown-item btn-link">  <span class="d-flex flex-items-baseline">
-              <span class="flex-auto">Go to line</span>
-              <span class="text-small color-fg-muted" aria-hidden="true">L</span>
-            </span>
-
-</button>        </li>
-        <li class="dropdown-divider" role="none"></li>
-        <li>
-          <clipboard-copy data-toggle-for="blob-more-options-details" aria-label="Copy path" value="applications/轻量级车牌识别.md" data-view-component="true" class="dropdown-item cursor-pointer">
-
-            Copy path
-
-</clipboard-copy>        </li>
-        <li>
-          <clipboard-copy data-toggle-for="blob-more-options-details" aria-label="Copy permalink" value="https://github.com/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-view-component="true" class="dropdown-item cursor-pointer">
-
-            <span class="d-flex flex-items-baseline">
-              <span class="flex-auto">Copy permalink</span>
-            </span>
-
-</clipboard-copy>        </li>
-      </ul>
-</div>
-</details></div>
-
-
-
-
-
-    <div id="spoof-warning" class="mt-0 pb-3" hidden aria-hidden>
-  <div data-view-component="true" class="flash flash-warn mt-0 clearfix">
-
-
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert float-left mt-1">
-    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
-</svg>
-
-      <div class="overflow-hidden">This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.</div>
-
-
-
-</div></div>
-
-    <include-fragment src="/PaddlePaddle/PaddleOCR/spoofed_commit_check/a3dfc10b7824743939976c42db6d50d118e2b3b9" data-test-selector="spoofed-commit-check"></include-fragment>
-
-    <div class="Box d-flex flex-column flex-shrink-0 mb-3">
-
-  <div class="Box-header Details js-details-container">
-      <div class="d-flex flex-items-center">
-        <span class="flex-shrink-0 ml-n1 mr-n1 mt-n1 mb-n1">
-          <a rel="contributor" data-skip-pjax="true" data-hovercard-type="user" data-hovercard-url="/users/MissPenguin/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/MissPenguin"><img class="avatar avatar-user" src="https://avatars.githubusercontent.com/u/25809855?s=48&amp;v=4" width="24" height="24" alt="@MissPenguin" /></a>
-        </span>
-        <div class="flex-1 d-flex flex-items-center ml-3 min-width-0">
-          <div class="css-truncate css-truncate-overflow">
-            <a class="text-bold Link--primary" rel="contributor" data-hovercard-type="user" data-hovercard-url="/users/MissPenguin/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/MissPenguin">MissPenguin</a>
-
-              <span class="markdown-title">
-                <a data-pjax="true" title="Update 轻量级车牌识别.md" class="Link--secondary" href="/PaddlePaddle/PaddleOCR/commit/10a3f2573d8e875c69683a71dc067d9060fb6875">Update 轻量级车牌识别.md</a>
-              </span>
-          </div>
-
-
-          <span class="ml-2">
-            <include-fragment accept="text/fragment+html" src="/PaddlePaddle/PaddleOCR/commit/10a3f2573d8e875c69683a71dc067d9060fb6875/rollup?direction=e" class="d-inline"></include-fragment>
-          </span>
-        </div>
-        <div class="ml-3 d-flex flex-shrink-0 flex-items-center flex-justify-end color-fg-muted no-wrap">
-          <span class="d-none d-md-inline">
-            <span>Latest commit</span>
-            <a class="text-small text-mono Link--secondary" href="/PaddlePaddle/PaddleOCR/commit/10a3f2573d8e875c69683a71dc067d9060fb6875" data-pjax>10a3f25</a>
-            <span itemprop="dateModified"><relative-time datetime="2022-06-15T10:13:01Z" class="no-wrap">Jun 15, 2022</relative-time></span>
-          </span>
-
-          <a data-pjax href="/PaddlePaddle/PaddleOCR/commits/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" class="ml-3 no-wrap Link--primary no-underline">
-            <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-history">
-    <path fill-rule="evenodd" d="M1.643 3.143L.427 1.927A.25.25 0 000 2.104V5.75c0 .138.112.25.25.25h3.646a.25.25 0 00.177-.427L2.715 4.215a6.5 6.5 0 11-1.18 4.458.75.75 0 10-1.493.154 8.001 8.001 0 101.6-5.684zM7.75 4a.75.75 0 01.75.75v2.992l2.028.812a.75.75 0 01-.557 1.392l-2.5-1A.75.75 0 017 8.25v-3.5A.75.75 0 017.75 4z"></path>
-</svg>
-            <span class="d-none d-sm-inline">
-              <strong>History</strong>
-            </span>
-          </a>
-        </div>
-      </div>
-
-  </div>
-
-  <div class="Box-body d-flex flex-items-center flex-auto border-bottom-0 flex-wrap" >
-    <details class="details-reset details-overlay details-overlay-dark lh-default color-fg-default float-left mr-3" id="blob_contributors_box">
-      <summary class="Link--primary">
-        <svg text="gray" aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-people">
-    <path fill-rule="evenodd" d="M5.5 3.5a2 2 0 100 4 2 2 0 000-4zM2 5.5a3.5 3.5 0 115.898 2.549 5.507 5.507 0 013.034 4.084.75.75 0 11-1.482.235 4.001 4.001 0 00-7.9 0 .75.75 0 01-1.482-.236A5.507 5.507 0 013.102 8.05 3.49 3.49 0 012 5.5zM11 4a.75.75 0 100 1.5 1.5 1.5 0 01.666 2.844.75.75 0 00-.416.672v.352a.75.75 0 00.574.73c1.2.289 2.162 1.2 2.522 2.372a.75.75 0 101.434-.44 5.01 5.01 0 00-2.56-3.012A3 3 0 0011 4z"></path>
-</svg>
-        <strong>2</strong>
-
-        contributors
-      </summary>
-      <details-dialog
-        class="Box Box--overlay d-flex flex-column anim-fade-in fast"
-        aria-label="Users who have contributed to this file"
-        src="/PaddlePaddle/PaddleOCR/contributors-list/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" preload>
-        <div class="Box-header">
-          <button class="Box-btn-octicon btn-octicon float-right" type="button" aria-label="Close dialog" data-close-dialog>
-            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-          </button>
-          <h3 class="Box-title">
-            Users who have contributed to this file
-          </h3>
-        </div>
-        <include-fragment>
-          <svg style="box-sizing: content-box; color: var(--color-icon-primary);" width="32" height="32" viewBox="0 0 16 16" fill="none" data-view-component="true" class="my-3 mx-auto d-block anim-rotate">
-  <circle cx="8" cy="8" r="7" stroke="currentColor" stroke-opacity="0.25" stroke-width="2" vector-effect="non-scaling-stroke" />
-  <path d="M15 8a7.002 7.002 0 00-7-7" stroke="currentColor" stroke-width="2" stroke-linecap="round" vector-effect="non-scaling-stroke" />
-</svg>
-        </include-fragment>
-      </details-dialog>
-    </details>
-      <span class="">
-    <a class="avatar-link" data-hovercard-type="user" data-hovercard-url="/users/WenmuZhou/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle/PaddleOCR/commits/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?author=WenmuZhou">
-      <img class="avatar mr-2 avatar-user" src="https://avatars.githubusercontent.com/u/12406017?s=48&amp;v=4" width="24" height="24" alt="@WenmuZhou" />
-</a>    <a class="avatar-link" data-hovercard-type="user" data-hovercard-url="/users/MissPenguin/hovercard" data-octo-click="hovercard-link-click" data-octo-dimensions="link_type:self" href="/PaddlePaddle/PaddleOCR/commits/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?author=MissPenguin">
-      <img class="avatar mr-2 avatar-user" src="https://avatars.githubusercontent.com/u/25809855?s=48&amp;v=4" width="24" height="24" alt="@MissPenguin" />
-</a>
-</span>
-
-  </div>
-</div>
-
-
-
-
-
-
-
-    <readme-toc>
-
-    <div data-target="readme-toc.content" class="Box mt-3 position-relative">
-
-  <div
-    class="Box-header js-blob-header blob-header js-sticky js-position-sticky top-0 p-2 d-flex flex-shrink-0 flex-md-row flex-items-center"
-    style="position: sticky; z-index: 1;"
-  >
-
-      <details
-  data-target="readme-toc.trigger"
-  data-menu-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;trigger&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}"
-  data-menu-hydro-click-hmac="532f79aab7b37e9668b558163fc4a6d01f017a94245e04831a2e20c0a43fff35"
-  class="dropdown details-reset details-overlay"
->
-  <summary
-    class="btn btn-octicon m-0 mr-2 p-2"
-    aria-haspopup="true"
-    aria-label="Table of Contents">
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-list-unordered">
-    <path fill-rule="evenodd" d="M2 4a1 1 0 100-2 1 1 0 000 2zm3.75-1.5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zm0 5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zm0 5a.75.75 0 000 1.5h8.5a.75.75 0 000-1.5h-8.5zM3 8a1 1 0 11-2 0 1 1 0 012 0zm-1 6a1 1 0 100-2 1 1 0 000 2z"></path>
-</svg>
-  </summary>
-
-
-  <details-menu class="SelectMenu" role="menu">
-    <div class="SelectMenu-modal rounded-3 mt-1" style="max-height:340px;">
-
-        <div class="SelectMenu-filter">
-          <input
-            class="SelectMenu-input form-control js-filterable-field"
-            id="toc-filter-field"
-            type="text"
-            autocomplete="off"
-            spellcheck="false"
-            autofocus
-            placeholder="Filter headings"
-            aria-label="Filter headings">
-        </div>
-
-      <div class="SelectMenu-list SelectMenu-list--borderless p-2" style="overscroll-behavior: contain;" data-filterable-for="toc-filter-field" data-filterable-type="substring">
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 text-emphasized" style="-webkit-box-orient: vertical; padding-left: 12px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#一种基于paddleocr的轻量级车牌识别模型">一种基于PaddleOCR的轻量级车牌识别模型</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#1-项目介绍">1. 项目介绍</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#2-环境搭建">2. 环境搭建</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#3-数据集准备">3. 数据集准备</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#31-数据集标注规则">3.1 数据集标注规则</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#32-制作符合pp-ocr训练格式的标注文件">3.2 制作符合PP-OCR训练格式的标注文件</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 24px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#4-实验">4. 实验</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#41-检测">4.1 检测</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#411-预训练模型直接预测">4.1.1 预训练模型直接预测</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#412-ccpd车牌数据集fine-tune">4.1.2 CCPD车牌数据集fine-tune</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#413-ccpd车牌数据集fine-tune量化训练">4.1.3 CCPD车牌数据集fine-tune+量化训练</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#414-模型导出">4.1.4 模型导出</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#42-识别">4.2 识别</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#421-预训练模型直接预测">4.2.1 预训练模型直接预测</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#422-预训练模型直接预测改动后处理">4.2.2 预训练模型直接预测+改动后处理</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#423-ccpd车牌数据集fine-tune">4.2.3 CCPD车牌数据集fine-tune</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#424-ccpd车牌数据集fine-tune量化训练">4.2.4 CCPD车牌数据集fine-tune+量化训练</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 48px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#425-模型导出">4.2.5 模型导出</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#43-计算end2end指标">4.3 计算End2End指标</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#44-部署">4.4 部署</a>
-          <a role="menuitem" class="filter-item SelectMenu-item ws-normal wb-break-word line-clamp-2 py-1 " style="-webkit-box-orient: vertical; padding-left: 36px;" data-action="click:readme-toc#blur" data-targets="readme-toc.entries" data-hydro-click="{&quot;event_type&quot;:&quot;repository_toc_menu.click&quot;,&quot;payload&quot;:{&quot;target&quot;:&quot;entry&quot;,&quot;repository_id&quot;:262296122,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="f16ce9e8c1e9aa2401c6c9022492741b158c7acb5e15ff56a78b4eb7f8928d6e" href="#45-实验总结">4.5 实验总结</a>
-      </div>
-    </div>
-  </details-menu>
-</details>
-
-
-  <div class="text-mono f6 flex-auto pr-3 flex-order-2 flex-md-order-1">
-
-      831 lines (628 sloc)
-      <span class="file-info-divider"></span>
-    41.2 KB
-  </div>
-
-  <div class="d-flex py-1 py-md-0 flex-auto flex-order-1 flex-md-order-2 flex-sm-grow-0 flex-justify-between hide-sm hide-md">
-        <div class="BtnGroup">
-    <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?plain=1" data-permalink-href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md?plain=1" aria-label="Display the source blob" data-view-component="true" class="source tooltipped tooltipped tooltipped-n  js-permalink-replaceable-link btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-code">
-    <path fill-rule="evenodd" d="M4.72 3.22a.75.75 0 011.06 1.06L2.06 8l3.72 3.72a.75.75 0 11-1.06 1.06L.47 8.53a.75.75 0 010-1.06l4.25-4.25zm6.56 0a.75.75 0 10-1.06 1.06L13.94 8l-3.72 3.72a.75.75 0 101.06 1.06l4.25-4.25a.75.75 0 000-1.06l-4.25-4.25z"></path>
-</svg>
-
-</a>    <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-permalink-href="/PaddlePaddle/PaddleOCR/blob/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" aria-label="Display the rendered blob" data-view-component="true" class="rendered tooltipped tooltipped tooltipped-n selected js-permalink-replaceable-link btn-sm btn BtnGroup-item">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-file">
-    <path fill-rule="evenodd" d="M3.75 1.5a.25.25 0 00-.25.25v12.5c0 .138.112.25.25.25h9.5a.25.25 0 00.25-.25V6h-2.75A1.75 1.75 0 019 4.25V1.5H3.75zm6.75.062V4.25c0 .138.112.25.25.25h2.688a.252.252 0 00-.011-.013l-2.914-2.914a.272.272 0 00-.013-.011zM2 1.75C2 .784 2.784 0 3.75 0h6.586c.464 0 .909.184 1.237.513l2.914 2.914c.329.328.513.773.513 1.237v9.586A1.75 1.75 0 0113.25 16h-9.5A1.75 1.75 0 012 14.25V1.75z"></path>
-</svg>
-
-</a>  </div>
-
-
-    <div class="BtnGroup">
-      <a data-permalink-href="/PaddlePaddle/PaddleOCR/raw/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" href="/PaddlePaddle/PaddleOCR/raw/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" id="raw-url" data-view-component="true" class="js-permalink-replaceable-link btn-sm btn BtnGroup-item">  Raw
-
-</a>        <a data-permalink-href="/PaddlePaddle/PaddleOCR/blame/a3dfc10b7824743939976c42db6d50d118e2b3b9/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" href="/PaddlePaddle/PaddleOCR/blame/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" data-hotkey="b" data-view-component="true" class="js-update-url-with-hash js-permalink-replaceable-link btn-sm btn BtnGroup-item">  Blame
-
-</a>    </div>
-
-    <div class="d-flex">
-
-
-
-<div class="ml-1" data-test-selector="edit-dropdown-menu-component">
-  <form class="BtnGroup-parent js-update-url-with-hash " data-turbo="false" action="/PaddlePaddle/PaddleOCR/edit/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" accept-charset="UTF-8" method="post"><input type="hidden" name="authenticity_token" value="llz78Eh-h4wCyHYKnMSomzLitBJmBsXnGWrYyE5RqU8R0h9nBeLKEJMLkCaprphwyYcNklhkrLFDqGzHP14SMA" autocomplete="off" />
-    <button disabled="disabled" title="You must be signed in to make or propose changes" data-hotkey="e" data-disable-with="" data-test-selector="edit-icon-button" type="submit" data-view-component="true" class="btn-sm BtnGroup-item btn">  <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-pencil">
-    <path fill-rule="evenodd" d="M11.013 1.427a1.75 1.75 0 012.474 0l1.086 1.086a1.75 1.75 0 010 2.474l-8.61 8.61c-.21.21-.47.364-.756.445l-3.251.93a.75.75 0 01-.927-.928l.929-3.25a1.75 1.75 0 01.445-.758l8.61-8.61zm1.414 1.06a.25.25 0 00-.354 0L10.811 3.75l1.439 1.44 1.263-1.263a.25.25 0 000-.354l-1.086-1.086zM11.189 6.25L9.75 4.81l-6.286 6.287a.25.25 0 00-.064.108l-.558 1.953 1.953-.558a.249.249 0 00.108-.064l6.286-6.286z"></path>
-</svg>
-
-</button></form>
-  <details class="details-reset details-overlay select-menu BtnGroup-parent d-inline-block position-relative">
-    <summary data-disable-invalid="" data-disable-with="" data-hydro-click="{&quot;event_type&quot;:&quot;blob_edit_dropdown.more_options_click&quot;,&quot;payload&quot;:{&quot;repository_id&quot;:262296122,&quot;actor_id&quot;:null,&quot;github_dev_enabled&quot;:false,&quot;edit_enabled&quot;:false,&quot;small_screen&quot;:false,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}" data-hydro-click-hmac="8e1c91138fac18e1604a705b864e87574bdf4b05c6f3b559e5a5a9051b4f9f4c" aria-label="Select additional options" data-view-component="true" class="select-menu-button btn-sm btn BtnGroup-item float-none px-2">
-</summary>    <div class="SelectMenu right-0">
-      <div class="SelectMenu-modal width-full">
-        <div class="SelectMenu-list SelectMenu-list--borderless py-2">
-          <form class="SelectMenu-item js-update-url-with-hash " data-turbo="false" action="/PaddlePaddle/PaddleOCR/edit/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md" accept-charset="UTF-8" method="post"><input type="hidden" name="authenticity_token" value="Ovbx0rgGgsdbao-R_-A0l2RC6K1eBosPF4oakk00YT69eBVF9ZrPW8qpab3KigR8nydRLWBk4llNSK6dPDvaQQ" autocomplete="off" />
-            <button disabled="disabled" title="You must be signed in to make or propose changes" data-test-selector="edit-text-button" type="submit" data-view-component="true" class="btn-invisible btn width-full d-flex flex-justify-between color-fg-muted text-normal p-0">  <div>Edit this file</div>
-              <div class="color-fg-muted">E</div>
-
-</button></form>
-
-            <a data-platforms="windows,mac" data-test-selector="github-desktop-link" aria-label="Open this file in GitHub Desktop" href="https://desktop.github.com" data-view-component="true" class="SelectMenu-item no-wrap js-remove-unless-platform width-full text-normal color-fg-default f5">
-              Open in GitHub Desktop
-</a>        </div>
-      </div>
-    </div>
-  </details>
-</div>
-
-
-
-<div data-test-selector="remote-clipboard-copy">
-  </div>
-
-
-          <button class="btn-octicon btn-octicon-danger disabled tooltipped tooltipped-nw" disabled
-            aria-label="You must be signed in to make or propose changes" type="button">
-            <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-trash">
-    <path fill-rule="evenodd" d="M6.5 1.75a.25.25 0 01.25-.25h2.5a.25.25 0 01.25.25V3h-3V1.75zm4.5 0V3h2.25a.75.75 0 010 1.5H2.75a.75.75 0 010-1.5H5V1.75C5 .784 5.784 0 6.75 0h2.5C10.216 0 11 .784 11 1.75zM4.496 6.675a.75.75 0 10-1.492.15l.66 6.6A1.75 1.75 0 005.405 15h5.19c.9 0 1.652-.681 1.741-1.576l.66-6.6a.75.75 0 00-1.492-.149l-.66 6.6a.25.25 0 01-.249.225h-5.19a.25.25 0 01-.249-.225l-.66-6.6z"></path>
-</svg>
-          </button>
-    </div>
-  </div>
-
-    <div class="d-flex hide-lg hide-xl flex-order-2 flex-grow-0">
-      <details class="dropdown details-reset details-overlay d-inline-block">
-        <summary
-          class="btn-octicon p-2"
-          aria-haspopup="true"
-          aria-label="possible actions"
-          data-test-selector="small-screen-more-options"
-          data-hydro-click="{&quot;event_type&quot;:&quot;blob_edit_dropdown.more_options_click&quot;,&quot;payload&quot;:{&quot;repository_id&quot;:262296122,&quot;actor_id&quot;:null,&quot;github_dev_enabled&quot;:false,&quot;edit_enabled&quot;:false,&quot;small_screen&quot;:true,&quot;originating_url&quot;:&quot;https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md&quot;,&quot;user_id&quot;:null}}"
-          data-hydro-click-hmac="3711aba9bcc87015fd0b3b29260720d15e89cbf158ae938adc99aa4a84a101e5"
-        >
-          <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-kebab-horizontal">
-    <path d="M8 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zM1.5 9a1.5 1.5 0 100-3 1.5 1.5 0 000 3zm13 0a1.5 1.5 0 100-3 1.5 1.5 0 000 3z"></path>
-</svg>
-        </summary>
-
-        <ul class="dropdown-menu dropdown-menu-sw" style="width: 175px">
-            <li>
-                <a class="dropdown-item tooltipped tooltipped-nw js-remove-unless-platform"
-                   data-platforms="windows,mac"
-                   href="https://desktop.github.com"
-                   data-ga-click="Repository, open with desktop">
-                  Open with Desktop
-                </a>
-            </li>
-          <li>
-            <a class="dropdown-item" href="/PaddlePaddle/PaddleOCR/raw/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">
-              View raw
-            </a>
-          </li>
-            <li>
-                          </li>
-            <li>
-              <a class="dropdown-item" href="/PaddlePaddle/PaddleOCR/blame/dygraph/applications/%E8%BD%BB%E9%87%8F%E7%BA%A7%E8%BD%A6%E7%89%8C%E8%AF%86%E5%88%AB.md">
-                View blame
-              </a>
-            </li>
-
-        </ul>
-      </details>
-    </div>
-</div>
-
-
-        <div id="readme" class="Box-body readme blob js-code-block-container p-5 p-xl-6 gist-border-0">
-    <article class="markdown-body entry-content container-lg" itemprop="text"><h1 dir="auto"><a id="user-content-一种基于paddleocr的轻量级车牌识别模型" class="anchor" aria-hidden="true" href="#一种基于paddleocr的轻量级车牌识别模型"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>一种基于PaddleOCR的轻量级车牌识别模型</h1>
-<ul dir="auto">
-<li><a href="#1-%E9%A1%B9%E7%9B%AE%E4%BB%8B%E7%BB%8D">1. 项目介绍</a></li>
-<li><a href="#2-%E7%8E%AF%E5%A2%83%E6%90%AD%E5%BB%BA">2. 环境搭建</a></li>
-<li><a href="#3-%E6%95%B0%E6%8D%AE%E9%9B%86%E5%87%86%E5%A4%87">3. 数据集准备</a>
-<ul dir="auto">
-<li><a href="#31-%E6%95%B0%E6%8D%AE%E9%9B%86%E6%A0%87%E6%B3%A8%E8%A7%84%E5%88%99">3.1 数据集标注规则</a></li>
-<li><a href="#32-%E5%88%B6%E4%BD%9C%E7%AC%A6%E5%90%88pp-ocr%E8%AE%AD%E7%BB%83%E6%A0%BC%E5%BC%8F%E7%9A%84%E6%A0%87%E6%B3%A8%E6%96%87%E4%BB%B6">3.2 制作符合PP-OCR训练格式的标注文件</a></li>
-</ul>
-</li>
-<li><a href="#4-%E5%AE%9E%E9%AA%8C">4. 实验</a>
-<ul dir="auto">
-<li><a href="#41-%E6%A3%80%E6%B5%8B">4.1 检测</a>
-<ul dir="auto">
-<li><a href="#411-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B">4.1.1 预训练模型直接预测</a></li>
-<li><a href="#412-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune">4.1.2 CCPD车牌数据集fine-tune</a></li>
-<li><a href="#413-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune%E9%87%8F%E5%8C%96%E8%AE%AD%E7%BB%83">4.1.3 CCPD车牌数据集fine-tune+量化训练</a></li>
-<li><a href="#414-%E6%A8%A1%E5%9E%8B%E5%AF%BC%E5%87%BA">4.1.4 模型导出</a></li>
-</ul>
-</li>
-<li><a href="#42-%E8%AF%86%E5%88%AB">4.2 识别</a>
-<ul dir="auto">
-<li><a href="#421-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B">4.2.1 预训练模型直接预测</a></li>
-<li><a href="#422-%E9%A2%84%E8%AE%AD%E7%BB%83%E6%A8%A1%E5%9E%8B%E7%9B%B4%E6%8E%A5%E9%A2%84%E6%B5%8B%E6%94%B9%E5%8A%A8%E5%90%8E%E5%A4%84%E7%90%86">4.2.2 预训练模型直接预测+改动后处理</a></li>
-<li><a href="#423-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune">4.2.3 CCPD车牌数据集fine-tune</a></li>
-<li><a href="#424-ccpd%E8%BD%A6%E7%89%8C%E6%95%B0%E6%8D%AE%E9%9B%86fine-tune%E9%87%8F%E5%8C%96%E8%AE%AD%E7%BB%83">4.2.4 CCPD车牌数据集fine-tune+量化训练</a></li>
-<li><a href="#425-%E6%A8%A1%E5%9E%8B%E5%AF%BC%E5%87%BA">4.2.5 模型导出</a></li>
-</ul>
-</li>
-<li><a href="#43-%E8%AE%A1%E7%AE%97End2End%E6%8C%87%E6%A0%87">4.3 计算End2End指标</a></li>
-<li><a href="#44-%E9%83%A8%E7%BD%B2">4.4 部署</a></li>
-<li><a href="#45-%E5%AE%9E%E9%AA%8C%E6%80%BB%E7%BB%93">4.5 实验总结</a></li>
-</ul>
-</li>
-</ul>
-<h2 dir="auto"><a id="user-content-1-项目介绍" class="anchor" aria-hidden="true" href="#1-项目介绍"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>1. 项目介绍</h2>
-<p dir="auto">车牌识别(Vehicle License Plate Recognition，VLPR) 是计算机视频图像识别技术在车辆牌照识别中的一种应用。车牌识别技术要求能够将运动中的汽车牌照从复杂背景中提取并识别出来，在高速公路车辆管理，停车场管理和城市交通中得到广泛应用。</p>
-<p dir="auto">本项目难点如下：</p>
-<ol dir="auto">
-<li>车牌在图像中的尺度差异大、在车辆上的悬挂位置不固定</li>
-<li>车牌图像质量层次不齐: 角度倾斜、图片模糊、光照不足、过曝等问题严重</li>
-<li>边缘和端测场景应用对模型大小有限制，推理速度有要求</li>
-</ol>
-<p dir="auto">针对以上问题， 本例选用 PP-OCRv3 这一开源超轻量OCR系统进行车牌识别系统的开发。基于PP-OCRv3模型，在CCPD数据集达到99%的检测和94%的识别精度，模型大小12.8M(2.5M+10.3M)。基于量化对模型体积进行进一步压缩到5.8M(1M+4.8M), 同时推理速度提升25%。</p>
-<p dir="auto">aistudio项目链接: <a href="https://aistudio.baidu.com/aistudio/projectdetail/3919091?contributionType=1" rel="nofollow">基于PaddleOCR的轻量级车牌识别范例</a></p>
-<h2 dir="auto"><a id="user-content-2-环境搭建" class="anchor" aria-hidden="true" href="#2-环境搭建"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>2. 环境搭建</h2>
-<p dir="auto">本任务基于Aistudio完成, 具体环境如下：</p>
-<ul dir="auto">
-<li>操作系统: Linux</li>
-<li>PaddlePaddle: 2.3</li>
-<li>paddleslim: 2.2.2</li>
-<li>PaddleOCR: Release/2.5</li>
-</ul>
-<p dir="auto">下载 PaddleOCR代码</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR"><pre>git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR</pre></div>
-<p dir="auto">安装依赖库</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="pip install -r PaddleOCR/requirements.txt"><pre>pip install -r PaddleOCR/requirements.txt</pre></div>
-<h2 dir="auto"><a id="user-content-3-数据集准备" class="anchor" aria-hidden="true" href="#3-数据集准备"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3. 数据集准备</h2>
-<p dir="auto">所使用的数据集为 CCPD2020 新能源车牌数据集，该数据集为</p>
-<p dir="auto">该数据集分布如下：</p>
-<table>
-<thead>
-<tr>
-<th>数据集类型</th>
-<th>数量</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>训练集</td>
-<td>5769</td>
-</tr>
-<tr>
-<td>验证集</td>
-<td>1001</td>
-</tr>
-<tr>
-<td>测试集</td>
-<td>5006</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">数据集图片示例如下:
-<a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/422ba572727331bfacf4a748100bcdf775d5ed4b2409ae78456849b2d94b1a8d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f33626365303537613865306334306130616362643236623265323965346532353930613331626334313237363462653762396534393739396336396362393163"><img src="https://camo.githubusercontent.com/422ba572727331bfacf4a748100bcdf775d5ed4b2409ae78456849b2d94b1a8d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f33626365303537613865306334306130616362643236623265323965346532353930613331626334313237363462653762396534393739396336396362393163" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/3bce057a8e0c40a0acbd26b2e29e4e2590a31bc412764be7b9e49799c69cb91c" style="max-width: 100%;"></a></p>
-<p dir="auto">数据集可以从这里下载 <a href="https://aistudio.baidu.com/aistudio/datasetdetail/101595" rel="nofollow">https://aistudio.baidu.com/aistudio/datasetdetail/101595</a></p>
-<p dir="auto">下载好数据集后对数据集进行解压</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip"><pre>unzip -d /home/aistudio/data /home/aistudio/data/data101595/CCPD2020.zip</pre></div>
-<h3 dir="auto"><a id="user-content-31-数据集标注规则" class="anchor" aria-hidden="true" href="#31-数据集标注规则"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3.1 数据集标注规则</h3>
-<p dir="auto">CPPD数据集的图片文件名具有特殊规则，详细可查看：<a href="https://github.com/detectRecog/CCPD">https://github.com/detectRecog/CCPD</a></p>
-<p dir="auto">具体规则如下：</p>
-<p dir="auto">例如: 025-95_113-154&amp;383_386&amp;473-386&amp;473_177&amp;454_154&amp;383_363&amp;402-0_0_22_27_27_33_16-37-15.jpg</p>
-<p dir="auto">每个名称可以分为七个字段，以-符号作为分割。这些字段解释如下。</p>
-<ul dir="auto">
-<li>
-<p dir="auto">025：车牌面积与整个图片区域的面积比。025 (25%)</p>
-</li>
-<li>
-<p dir="auto">95_113：水平倾斜程度和垂直倾斜度。水平 95度 垂直 113度</p>
-</li>
-<li>
-<p dir="auto">154&amp;383_386&amp;473：左上和右下顶点的坐标。左上(154,383) 右下(386,473)</p>
-</li>
-<li>
-<p dir="auto">386&amp;473_177&amp;454_154&amp;383_363&amp;402：整个图像中车牌的四个顶点的精确（x，y）坐标。这些坐标从右下角顶点开始。(386,473) (177,454) (154,383) (363,402)</p>
-</li>
-<li>
-<p dir="auto">0_0_22_27_27_33_16：CCPD中的每个图像只有一个车牌。每个车牌号码由一个汉字，一个字母和五个字母或数字组成。有效的中文车牌由七个字符组成：省（1个字符），字母（1个字符），字母+数字（5个字符）。“ 0_0_22_27_27_33_16”是每个字符的索引。这三个数组定义如下。每个数组的最后一个字符是字母O，而不是数字0。我们将O用作“无字符”的符号，因为中文车牌字符中没有O。因此以上车牌拼起来即为 皖AY339S</p>
-</li>
-<li>
-<p dir="auto">37：牌照区域的亮度。 37 (37%)</p>
-</li>
-<li>
-<p dir="auto">15：车牌区域的模糊度。15 (15%)</p>
-</li>
-</ul>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="provinces = [&quot;皖&quot;, &quot;沪&quot;, &quot;津&quot;, &quot;渝&quot;, &quot;冀&quot;, &quot;晋&quot;, &quot;蒙&quot;, &quot;辽&quot;, &quot;吉&quot;, &quot;黑&quot;, &quot;苏&quot;, &quot;浙&quot;, &quot;京&quot;, &quot;闽&quot;, &quot;赣&quot;, &quot;鲁&quot;, &quot;豫&quot;, &quot;鄂&quot;, &quot;湘&quot;, &quot;粤&quot;, &quot;桂&quot;, &quot;琼&quot;, &quot;川&quot;, &quot;贵&quot;, &quot;云&quot;, &quot;藏&quot;, &quot;陕&quot;, &quot;甘&quot;, &quot;青&quot;, &quot;宁&quot;, &quot;新&quot;, &quot;警&quot;, &quot;学&quot;, &quot;O&quot;]
+```python
+provinces = ["皖", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑", "苏", "浙", "京", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤", "桂", "琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁", "新", "警", "学", "O"]
 alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W','X', 'Y', 'Z', 'O']
-ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X','Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']"><pre><span class="pl-s1">provinces</span> <span class="pl-c1">=</span> [<span class="pl-s">"皖"</span>, <span class="pl-s">"沪"</span>, <span class="pl-s">"津"</span>, <span class="pl-s">"渝"</span>, <span class="pl-s">"冀"</span>, <span class="pl-s">"晋"</span>, <span class="pl-s">"蒙"</span>, <span class="pl-s">"辽"</span>, <span class="pl-s">"吉"</span>, <span class="pl-s">"黑"</span>, <span class="pl-s">"苏"</span>, <span class="pl-s">"浙"</span>, <span class="pl-s">"京"</span>, <span class="pl-s">"闽"</span>, <span class="pl-s">"赣"</span>, <span class="pl-s">"鲁"</span>, <span class="pl-s">"豫"</span>, <span class="pl-s">"鄂"</span>, <span class="pl-s">"湘"</span>, <span class="pl-s">"粤"</span>, <span class="pl-s">"桂"</span>, <span class="pl-s">"琼"</span>, <span class="pl-s">"川"</span>, <span class="pl-s">"贵"</span>, <span class="pl-s">"云"</span>, <span class="pl-s">"藏"</span>, <span class="pl-s">"陕"</span>, <span class="pl-s">"甘"</span>, <span class="pl-s">"青"</span>, <span class="pl-s">"宁"</span>, <span class="pl-s">"新"</span>, <span class="pl-s">"警"</span>, <span class="pl-s">"学"</span>, <span class="pl-s">"O"</span>]
-<span class="pl-s1">alphabets</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>,<span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'O'</span>]
-<span class="pl-s1">ads</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>,<span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'0'</span>, <span class="pl-s">'1'</span>, <span class="pl-s">'2'</span>, <span class="pl-s">'3'</span>, <span class="pl-s">'4'</span>, <span class="pl-s">'5'</span>, <span class="pl-s">'6'</span>, <span class="pl-s">'7'</span>, <span class="pl-s">'8'</span>, <span class="pl-s">'9'</span>, <span class="pl-s">'O'</span>]</pre></div>
-<h3 dir="auto"><a id="user-content-32-制作符合pp-ocr训练格式的标注文件" class="anchor" aria-hidden="true" href="#32-制作符合pp-ocr训练格式的标注文件"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>3.2 制作符合PP-OCR训练格式的标注文件</h3>
-<p dir="auto">在开始训练之前，可使用如下代码制作符合PP-OCR训练格式的标注文件。</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="import cv2
+ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X','Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']
+```
+
+### 3.2 制作符合PP-OCR训练格式的标注文件
+
+在开始训练之前，可使用如下代码制作符合PP-OCR训练格式的标注文件。
+
+
+```python
+import cv2
 import os
 import json
 from tqdm import tqdm
 import numpy as np
 
-provinces = [&quot;皖&quot;, &quot;沪&quot;, &quot;津&quot;, &quot;渝&quot;, &quot;冀&quot;, &quot;晋&quot;, &quot;蒙&quot;, &quot;辽&quot;, &quot;吉&quot;, &quot;黑&quot;, &quot;苏&quot;, &quot;浙&quot;, &quot;京&quot;, &quot;闽&quot;, &quot;赣&quot;, &quot;鲁&quot;, &quot;豫&quot;, &quot;鄂&quot;, &quot;湘&quot;, &quot;粤&quot;, &quot;桂&quot;, &quot;琼&quot;, &quot;川&quot;, &quot;贵&quot;, &quot;云&quot;, &quot;藏&quot;, &quot;陕&quot;, &quot;甘&quot;, &quot;青&quot;, &quot;宁&quot;, &quot;新&quot;, &quot;警&quot;, &quot;学&quot;, &quot;O&quot;]
+provinces = ["皖", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑", "苏", "浙", "京", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤", "桂", "琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁", "新", "警", "学", "O"]
 alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'O']
 ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']
 
@@ -1579,13 +137,13 @@ def make_label(img_dir, save_gt_folder, phase):
     i = 0
     for filename in tqdm(os.listdir(os.path.join(img_dir, phase))):
         str_list = filename.split('-')
-        if len(str_list) &lt; 5:
+        if len(str_list) < 5:
             continue
         coord_list = str_list[3].split('_')
         txt_list = str_list[4].split('_')
         boxes = []
         for coord in coord_list:
-            boxes.append([int(x) for x in coord.split(&quot;&amp;&quot;)])
+            boxes.append([int(x) for x in coord.split("&")])
         boxes = [boxes[2], boxes[3], boxes[0], boxes[1]]
         lp_number = provinces[int(txt_list[0])] + alphabets[int(txt_list[1])] + ''.join([ads[int(x)] for x in txt_list[2:]])
 
@@ -1617,7 +175,7 @@ def get_rotate_crop_image(img, points):
     points[:, 0] = points[:, 0] - left
     points[:, 1] = points[:, 1] - top
     '''
-    assert len(points) == 4, &quot;shape of points must be 4*2&quot;
+    assert len(points) == 4, "shape of points must be 4*2"
     img_crop_width = int(
         max(
             np.linalg.norm(points[0] - points[1]),
@@ -1636,7 +194,7 @@ def get_rotate_crop_image(img, points):
         borderMode=cv2.BORDER_REPLICATE,
         flags=cv2.INTER_CUBIC)
     dst_img_height, dst_img_width = dst_img.shape[0:2]
-    if dst_img_height * 1.0 / dst_img_width &gt;= 1.5:
+    if dst_img_height * 1.0 / dst_img_width >= 1.5:
         dst_img = np.rot90(dst_img)
     return dst_img
 
@@ -1644,695 +202,414 @@ img_dir = '/home/aistudio/data/CCPD2020/ccpd_green'
 save_gt_folder = '/home/aistudio/data/CCPD2020/PPOCR'
 # phase = 'train' # change to val and test to make val dataset and test dataset
 for phase in ['train','val','test']:
-    make_label(img_dir, save_gt_folder, phase)"><pre><span class="pl-k">import</span> <span class="pl-s1">cv2</span>
-<span class="pl-k">import</span> <span class="pl-s1">os</span>
-<span class="pl-k">import</span> <span class="pl-s1">json</span>
-<span class="pl-k">from</span> <span class="pl-s1">tqdm</span> <span class="pl-k">import</span> <span class="pl-s1">tqdm</span>
-<span class="pl-k">import</span> <span class="pl-s1">numpy</span> <span class="pl-k">as</span> <span class="pl-s1">np</span>
+    make_label(img_dir, save_gt_folder, phase)
+```
 
-<span class="pl-s1">provinces</span> <span class="pl-c1">=</span> [<span class="pl-s">"皖"</span>, <span class="pl-s">"沪"</span>, <span class="pl-s">"津"</span>, <span class="pl-s">"渝"</span>, <span class="pl-s">"冀"</span>, <span class="pl-s">"晋"</span>, <span class="pl-s">"蒙"</span>, <span class="pl-s">"辽"</span>, <span class="pl-s">"吉"</span>, <span class="pl-s">"黑"</span>, <span class="pl-s">"苏"</span>, <span class="pl-s">"浙"</span>, <span class="pl-s">"京"</span>, <span class="pl-s">"闽"</span>, <span class="pl-s">"赣"</span>, <span class="pl-s">"鲁"</span>, <span class="pl-s">"豫"</span>, <span class="pl-s">"鄂"</span>, <span class="pl-s">"湘"</span>, <span class="pl-s">"粤"</span>, <span class="pl-s">"桂"</span>, <span class="pl-s">"琼"</span>, <span class="pl-s">"川"</span>, <span class="pl-s">"贵"</span>, <span class="pl-s">"云"</span>, <span class="pl-s">"藏"</span>, <span class="pl-s">"陕"</span>, <span class="pl-s">"甘"</span>, <span class="pl-s">"青"</span>, <span class="pl-s">"宁"</span>, <span class="pl-s">"新"</span>, <span class="pl-s">"警"</span>, <span class="pl-s">"学"</span>, <span class="pl-s">"O"</span>]
-<span class="pl-s1">alphabets</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'O'</span>]
-<span class="pl-s1">ads</span> <span class="pl-c1">=</span> [<span class="pl-s">'A'</span>, <span class="pl-s">'B'</span>, <span class="pl-s">'C'</span>, <span class="pl-s">'D'</span>, <span class="pl-s">'E'</span>, <span class="pl-s">'F'</span>, <span class="pl-s">'G'</span>, <span class="pl-s">'H'</span>, <span class="pl-s">'J'</span>, <span class="pl-s">'K'</span>, <span class="pl-s">'L'</span>, <span class="pl-s">'M'</span>, <span class="pl-s">'N'</span>, <span class="pl-s">'P'</span>, <span class="pl-s">'Q'</span>, <span class="pl-s">'R'</span>, <span class="pl-s">'S'</span>, <span class="pl-s">'T'</span>, <span class="pl-s">'U'</span>, <span class="pl-s">'V'</span>, <span class="pl-s">'W'</span>, <span class="pl-s">'X'</span>, <span class="pl-s">'Y'</span>, <span class="pl-s">'Z'</span>, <span class="pl-s">'0'</span>, <span class="pl-s">'1'</span>, <span class="pl-s">'2'</span>, <span class="pl-s">'3'</span>, <span class="pl-s">'4'</span>, <span class="pl-s">'5'</span>, <span class="pl-s">'6'</span>, <span class="pl-s">'7'</span>, <span class="pl-s">'8'</span>, <span class="pl-s">'9'</span>, <span class="pl-s">'O'</span>]
+通过上述命令可以完成了`训练集`，`验证集`和`测试集`的制作，制作完成的数据集信息如下：
 
-<span class="pl-k">def</span> <span class="pl-en">make_label</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>):
-    <span class="pl-s1">crop_img_save_dir</span> <span class="pl-c1">=</span> <span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'crop_imgs'</span>)
-    <span class="pl-s1">os</span>.<span class="pl-en">makedirs</span>(<span class="pl-s1">crop_img_save_dir</span>, <span class="pl-s1">exist_ok</span><span class="pl-c1">=</span><span class="pl-c1">True</span>)
+| 类型 | 数据集 | 图片地址 | 标签地址 | 图片数量 |
+| --- | --- | --- | --- | --- |
+| 检测 | 训练集 | /home/aistudio/data/CCPD2020/ccpd_green/train | /home/aistudio/data/CCPD2020/PPOCR/train/det.txt | 5769 |
+| 检测 | 验证集 | /home/aistudio/data/CCPD2020/ccpd_green/val | /home/aistudio/data/CCPD2020/PPOCR/val/det.txt | 1001 |
+| 检测 | 测试集 | /home/aistudio/data/CCPD2020/ccpd_green/test | /home/aistudio/data/CCPD2020/PPOCR/test/det.txt | 5006 |
+| 识别 | 训练集 | /home/aistudio/data/CCPD2020/PPOCR/train/crop_imgs | /home/aistudio/data/CCPD2020/PPOCR/train/rec.txt | 5769 |
+| 识别 | 验证集 | /home/aistudio/data/CCPD2020/PPOCR/val/crop_imgs | /home/aistudio/data/CCPD2020/PPOCR/val/rec.txt | 1001 |
+| 识别 | 测试集 | /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs | /home/aistudio/data/CCPD2020/PPOCR/test/rec.txt | 5006 |
 
-    <span class="pl-s1">f_det</span> <span class="pl-c1">=</span> <span class="pl-en">open</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'det.txt'</span>), <span class="pl-s">'w'</span>, <span class="pl-s1">encoding</span><span class="pl-c1">=</span><span class="pl-s">'utf-8'</span>)
-    <span class="pl-s1">f_rec</span> <span class="pl-c1">=</span> <span class="pl-en">open</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>, <span class="pl-s">'rec.txt'</span>), <span class="pl-s">'w'</span>, <span class="pl-s1">encoding</span><span class="pl-c1">=</span><span class="pl-s">'utf-8'</span>)
+在普遍的深度学习流程中，都是在训练集训练，在验证集选择最优模型后在测试集上进行测试。在本例中，我们省略中间步骤，直接在训练集训练，在测试集选择最优模型，因此我们只使用训练集和测试集。
 
-    <span class="pl-s1">i</span> <span class="pl-c1">=</span> <span class="pl-c1">0</span>
-    <span class="pl-k">for</span> <span class="pl-s1">filename</span> <span class="pl-c1">in</span> <span class="pl-en">tqdm</span>(<span class="pl-s1">os</span>.<span class="pl-en">listdir</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">phase</span>))):
-        <span class="pl-s1">str_list</span> <span class="pl-c1">=</span> <span class="pl-s1">filename</span>.<span class="pl-en">split</span>(<span class="pl-s">'-'</span>)
-        <span class="pl-k">if</span> <span class="pl-en">len</span>(<span class="pl-s1">str_list</span>) <span class="pl-c1">&lt;</span> <span class="pl-c1">5</span>:
-            <span class="pl-k">continue</span>
-        <span class="pl-s1">coord_list</span> <span class="pl-c1">=</span> <span class="pl-s1">str_list</span>[<span class="pl-c1">3</span>].<span class="pl-en">split</span>(<span class="pl-s">'_'</span>)
-        <span class="pl-s1">txt_list</span> <span class="pl-c1">=</span> <span class="pl-s1">str_list</span>[<span class="pl-c1">4</span>].<span class="pl-en">split</span>(<span class="pl-s">'_'</span>)
-        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> []
-        <span class="pl-k">for</span> <span class="pl-s1">coord</span> <span class="pl-c1">in</span> <span class="pl-s1">coord_list</span>:
-            <span class="pl-s1">boxes</span>.<span class="pl-en">append</span>([<span class="pl-en">int</span>(<span class="pl-s1">x</span>) <span class="pl-k">for</span> <span class="pl-s1">x</span> <span class="pl-c1">in</span> <span class="pl-s1">coord</span>.<span class="pl-en">split</span>(<span class="pl-s">"&amp;"</span>)])
-        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> [<span class="pl-s1">boxes</span>[<span class="pl-c1">2</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">3</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">0</span>], <span class="pl-s1">boxes</span>[<span class="pl-c1">1</span>]]
-        <span class="pl-s1">lp_number</span> <span class="pl-c1">=</span> <span class="pl-s1">provinces</span>[<span class="pl-en">int</span>(<span class="pl-s1">txt_list</span>[<span class="pl-c1">0</span>])] <span class="pl-c1">+</span> <span class="pl-s1">alphabets</span>[<span class="pl-en">int</span>(<span class="pl-s1">txt_list</span>[<span class="pl-c1">1</span>])] <span class="pl-c1">+</span> <span class="pl-s">''</span>.<span class="pl-en">join</span>([<span class="pl-s1">ads</span>[<span class="pl-en">int</span>(<span class="pl-s1">x</span>)] <span class="pl-k">for</span> <span class="pl-s1">x</span> <span class="pl-c1">in</span> <span class="pl-s1">txt_list</span>[<span class="pl-c1">2</span>:]])
+## 4. 实验
 
-        <span class="pl-c"># det</span>
-        <span class="pl-s1">det_info</span> <span class="pl-c1">=</span> [{<span class="pl-s">'points'</span>:<span class="pl-s1">boxes</span>, <span class="pl-s">'transcription'</span>:<span class="pl-s1">lp_number</span>}]
-        <span class="pl-s1">f_det</span>.<span class="pl-en">write</span>(<span class="pl-s">'{}<span class="pl-cce">\t</span>{}<span class="pl-cce">\n</span>'</span>.<span class="pl-en">format</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">phase</span>, <span class="pl-s1">filename</span>), <span class="pl-s1">json</span>.<span class="pl-en">dumps</span>(<span class="pl-s1">det_info</span>, <span class="pl-s1">ensure_ascii</span><span class="pl-c1">=</span><span class="pl-c1">False</span>)))
+由于数据集比较少，为了模型更好和更快的收敛，这里选用 PaddleOCR 中的 PP-OCRv3 模型进行文本检测和识别，并且使用 PP-OCRv3 模型参数作为预训练模型。PP-OCRv3在PP-OCRv2的基础上，中文场景端到端Hmean指标相比于PP-OCRv2提升5%, 英文数字模型端到端效果提升11%。详细优化细节请参考[PP-OCRv3](../doc/doc_ch/PP-OCRv3_introduction.md)技术报告。
 
-        <span class="pl-c"># rec</span>
-        <span class="pl-s1">boxes</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">float32</span>(<span class="pl-s1">boxes</span>)
-        <span class="pl-s1">img</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">imread</span>(<span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">phase</span>, <span class="pl-s1">filename</span>))
-        <span class="pl-c"># crop_img = img[int(boxes[:,1].min()):int(boxes[:,1].max()),int(boxes[:,0].min()):int(boxes[:,0].max())]</span>
-        <span class="pl-s1">crop_img</span> <span class="pl-c1">=</span> <span class="pl-en">get_rotate_crop_image</span>(<span class="pl-s1">img</span>, <span class="pl-s1">boxes</span>)
-        <span class="pl-s1">crop_img_save_filename</span> <span class="pl-c1">=</span> <span class="pl-s">'{}_{}.jpg'</span>.<span class="pl-en">format</span>(<span class="pl-s1">i</span>,<span class="pl-s">'_'</span>.<span class="pl-en">join</span>(<span class="pl-s1">txt_list</span>))
-        <span class="pl-s1">crop_img_save_path</span> <span class="pl-c1">=</span> <span class="pl-s1">os</span>.<span class="pl-s1">path</span>.<span class="pl-en">join</span>(<span class="pl-s1">crop_img_save_dir</span>, <span class="pl-s1">crop_img_save_filename</span>)
-        <span class="pl-s1">cv2</span>.<span class="pl-en">imwrite</span>(<span class="pl-s1">crop_img_save_path</span>, <span class="pl-s1">crop_img</span>)
-        <span class="pl-s1">f_rec</span>.<span class="pl-en">write</span>(<span class="pl-s">'{}/crop_imgs/{}<span class="pl-cce">\t</span>{}<span class="pl-cce">\n</span>'</span>.<span class="pl-en">format</span>(<span class="pl-s1">phase</span>, <span class="pl-s1">crop_img_save_filename</span>, <span class="pl-s1">lp_number</span>))
-        <span class="pl-s1">i</span><span class="pl-c1">+=</span><span class="pl-c1">1</span>
-    <span class="pl-s1">f_det</span>.<span class="pl-en">close</span>()
-    <span class="pl-s1">f_rec</span>.<span class="pl-en">close</span>()
+由于车牌场景均为端侧设备部署，因此对速度和模型大小有比较高的要求，因此还需要采用量化训练的方式进行模型大小的压缩和模型推理速度的加速。模型量化可以在基本不损失模型的精度的情况下，将FP32精度的模型参数转换为Int8精度，减小模型参数大小并加速计算，使用量化后的模型在移动端等部署时更具备速度优势。
 
-<span class="pl-k">def</span> <span class="pl-en">get_rotate_crop_image</span>(<span class="pl-s1">img</span>, <span class="pl-s1">points</span>):
-    <span class="pl-s">'''</span>
-<span class="pl-s">    img_height, img_width = img.shape[0:2]</span>
-<span class="pl-s">    left = int(np.min(points[:, 0]))</span>
-<span class="pl-s">    right = int(np.max(points[:, 0]))</span>
-<span class="pl-s">    top = int(np.min(points[:, 1]))</span>
-<span class="pl-s">    bottom = int(np.max(points[:, 1]))</span>
-<span class="pl-s">    img_crop = img[top:bottom, left:right, :].copy()</span>
-<span class="pl-s">    points[:, 0] = points[:, 0] - left</span>
-<span class="pl-s">    points[:, 1] = points[:, 1] - top</span>
-<span class="pl-s">    '''</span>
-    <span class="pl-k">assert</span> <span class="pl-en">len</span>(<span class="pl-s1">points</span>) <span class="pl-c1">==</span> <span class="pl-c1">4</span>, <span class="pl-s">"shape of points must be 4*2"</span>
-    <span class="pl-s1">img_crop_width</span> <span class="pl-c1">=</span> <span class="pl-en">int</span>(
-        <span class="pl-en">max</span>(
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">0</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">1</span>]),
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">2</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">3</span>])))
-    <span class="pl-s1">img_crop_height</span> <span class="pl-c1">=</span> <span class="pl-en">int</span>(
-        <span class="pl-en">max</span>(
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">0</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">3</span>]),
-            <span class="pl-s1">np</span>.<span class="pl-s1">linalg</span>.<span class="pl-en">norm</span>(<span class="pl-s1">points</span>[<span class="pl-c1">1</span>] <span class="pl-c1">-</span> <span class="pl-s1">points</span>[<span class="pl-c1">2</span>])))
-    <span class="pl-s1">pts_std</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">float32</span>([[<span class="pl-c1">0</span>, <span class="pl-c1">0</span>], [<span class="pl-s1">img_crop_width</span>, <span class="pl-c1">0</span>],
-                          [<span class="pl-s1">img_crop_width</span>, <span class="pl-s1">img_crop_height</span>],
-                          [<span class="pl-c1">0</span>, <span class="pl-s1">img_crop_height</span>]])
-    <span class="pl-v">M</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">getPerspectiveTransform</span>(<span class="pl-s1">points</span>, <span class="pl-s1">pts_std</span>)
-    <span class="pl-s1">dst_img</span> <span class="pl-c1">=</span> <span class="pl-s1">cv2</span>.<span class="pl-en">warpPerspective</span>(
-        <span class="pl-s1">img</span>,
-        <span class="pl-v">M</span>, (<span class="pl-s1">img_crop_width</span>, <span class="pl-s1">img_crop_height</span>),
-        <span class="pl-s1">borderMode</span><span class="pl-c1">=</span><span class="pl-s1">cv2</span>.<span class="pl-v">BORDER_REPLICATE</span>,
-        <span class="pl-s1">flags</span><span class="pl-c1">=</span><span class="pl-s1">cv2</span>.<span class="pl-v">INTER_CUBIC</span>)
-    <span class="pl-s1">dst_img_height</span>, <span class="pl-s1">dst_img_width</span> <span class="pl-c1">=</span> <span class="pl-s1">dst_img</span>.<span class="pl-s1">shape</span>[<span class="pl-c1">0</span>:<span class="pl-c1">2</span>]
-    <span class="pl-k">if</span> <span class="pl-s1">dst_img_height</span> <span class="pl-c1">*</span> <span class="pl-c1">1.0</span> <span class="pl-c1">/</span> <span class="pl-s1">dst_img_width</span> <span class="pl-c1">&gt;=</span> <span class="pl-c1">1.5</span>:
-        <span class="pl-s1">dst_img</span> <span class="pl-c1">=</span> <span class="pl-s1">np</span>.<span class="pl-en">rot90</span>(<span class="pl-s1">dst_img</span>)
-    <span class="pl-k">return</span> <span class="pl-s1">dst_img</span>
+因此，本实验中对于车牌检测和识别有如下3种方案：
 
-<span class="pl-s1">img_dir</span> <span class="pl-c1">=</span> <span class="pl-s">'/home/aistudio/data/CCPD2020/ccpd_green'</span>
-<span class="pl-s1">save_gt_folder</span> <span class="pl-c1">=</span> <span class="pl-s">'/home/aistudio/data/CCPD2020/PPOCR'</span>
-<span class="pl-c"># phase = 'train' # change to val and test to make val dataset and test dataset</span>
-<span class="pl-k">for</span> <span class="pl-s1">phase</span> <span class="pl-c1">in</span> [<span class="pl-s">'train'</span>,<span class="pl-s">'val'</span>,<span class="pl-s">'test'</span>]:
-    <span class="pl-en">make_label</span>(<span class="pl-s1">img_dir</span>, <span class="pl-s1">save_gt_folder</span>, <span class="pl-s1">phase</span>)</pre></div>
-<p dir="auto">通过上述命令可以完成了<code>训练集</code>，<code>验证集</code>和<code>测试集</code>的制作，制作完成的数据集信息如下：</p>
-<table>
-<thead>
-<tr>
-<th>类型</th>
-<th>数据集</th>
-<th>图片地址</th>
-<th>标签地址</th>
-<th>图片数量</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>检测</td>
-<td>训练集</td>
-<td>/home/aistudio/data/CCPD2020/ccpd_green/train</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/train/det.txt</td>
-<td>5769</td>
-</tr>
-<tr>
-<td>检测</td>
-<td>验证集</td>
-<td>/home/aistudio/data/CCPD2020/ccpd_green/val</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/val/det.txt</td>
-<td>1001</td>
-</tr>
-<tr>
-<td>检测</td>
-<td>测试集</td>
-<td>/home/aistudio/data/CCPD2020/ccpd_green/test</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/test/det.txt</td>
-<td>5006</td>
-</tr>
-<tr>
-<td>识别</td>
-<td>训练集</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/train/crop_imgs</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt</td>
-<td>5769</td>
-</tr>
-<tr>
-<td>识别</td>
-<td>验证集</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/val/crop_imgs</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/val/rec.txt</td>
-<td>1001</td>
-</tr>
-<tr>
-<td>识别</td>
-<td>测试集</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs</td>
-<td>/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt</td>
-<td>5006</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">在普遍的深度学习流程中，都是在训练集训练，在验证集选择最优模型后在测试集上进行测试。在本例中，我们省略中间步骤，直接在训练集训练，在测试集选择最优模型，因此我们只使用训练集和测试集。</p>
-<h2 dir="auto"><a id="user-content-4-实验" class="anchor" aria-hidden="true" href="#4-实验"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4. 实验</h2>
-<p dir="auto">由于数据集比较少，为了模型更好和更快的收敛，这里选用 PaddleOCR 中的 PP-OCRv3 模型进行文本检测和识别，并且使用 PP-OCRv3 模型参数作为预训练模型。PP-OCRv3在PP-OCRv2的基础上，中文场景端到端Hmean指标相比于PP-OCRv2提升5%, 英文数字模型端到端效果提升11%。详细优化细节请参考<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md">PP-OCRv3</a>技术报告。</p>
-<p dir="auto">由于车牌场景均为端侧设备部署，因此对速度和模型大小有比较高的要求，因此还需要采用量化训练的方式进行模型大小的压缩和模型推理速度的加速。模型量化可以在基本不损失模型的精度的情况下，将FP32精度的模型参数转换为Int8精度，减小模型参数大小并加速计算，使用量化后的模型在移动端等部署时更具备速度优势。</p>
-<p dir="auto">因此，本实验中对于车牌检测和识别有如下3种方案：</p>
-<ol dir="auto">
-<li>PP-OCRv3中英文超轻量预训练模型直接预测</li>
-<li>CCPD车牌数据集在PP-OCRv3模型上fine-tune</li>
-<li>CCPD车牌数据集在PP-OCRv3模型上fine-tune后量化</li>
-</ol>
-<h3 dir="auto"><a id="user-content-41-检测" class="anchor" aria-hidden="true" href="#41-检测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1 检测</h3>
-<h4 dir="auto"><a id="user-content-411-预训练模型直接预测" class="anchor" aria-hidden="true" href="#411-预训练模型直接预测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.1 预训练模型直接预测</h4>
-<p dir="auto">从下表中下载PP-OCRv3文本检测预训练模型</p>
-<table>
-<thead>
-<tr>
-<th>模型名称</th>
-<th>模型简介</th>
-<th>配置文件</th>
-<th>推理模型大小</th>
-<th>下载地址</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>ch_PP-OCRv3_det</td>
-<td>【最新】原始超轻量模型，支持中英文、多语种文本检测</td>
-<td><a href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml">ch_PP-OCRv3_det_cml.yml</a></td>
-<td>3.8M</td>
-<td><a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar" rel="nofollow">推理模型</a> / <a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar" rel="nofollow">训练模型</a></td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">使用如下命令下载预训练模型</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="mkdir models
+1. PP-OCRv3中英文超轻量预训练模型直接预测
+2. CCPD车牌数据集在PP-OCRv3模型上fine-tune
+3. CCPD车牌数据集在PP-OCRv3模型上fine-tune后量化
+
+### 4.1 检测
+#### 4.1.1 预训练模型直接预测
+
+从下表中下载PP-OCRv3文本检测预训练模型
+
+|模型名称|模型简介|配置文件|推理模型大小|下载地址|
+| --- | --- | --- | --- | --- |
+|ch_PP-OCRv3_det| 【最新】原始超轻量模型，支持中英文、多语种文本检测 |[ch_PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml)| 3.8M |[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar)|
+
+使用如下命令下载预训练模型
+
+```bash
+mkdir models
 cd models
 wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
 tar -xf ch_PP-OCRv3_det_distill_train.tar
-cd /home/aistudio/PaddleOCR"><pre>mkdir models
-<span class="pl-c1">cd</span> models
-wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
-tar -xf ch_PP-OCRv3_det_distill_train.tar
-<span class="pl-c1">cd</span> /home/aistudio/PaddleOCR</pre></div>
-<p dir="auto">预训练模型下载完成后，我们使用<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml">ch_PP-OCRv3_det_student.yml</a> 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址</li>
-</ol>
-</li>
-<li>数据集相关
-<ol dir="auto">
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">上述字段均为必须修改的字段，可以通过修改配置文件的方式改动，也可在不需要修改配置文件的情况下，改变训练的参数。这里使用不改变配置文件的方式 。使用如下命令进行PP-OCRv3文本检测预训练模型的评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+cd /home/aistudio/PaddleOCR
+```
+
+预训练模型下载完成后，我们使用[ch_PP-OCRv3_det_student.yml](../configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml) 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：
+
+1. 模型存储和训练相关:
+   1. Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址
+2. 数据集相关
+   1. Eval.dataset.data_dir：指向测试集图片存放目录
+   2. Eval.dataset.label_file_list：指向测试集标注文件
+
+上述字段均为必须修改的字段，可以通过修改配置文件的方式改动，也可在不需要修改配置文件的情况下，改变训练的参数。这里使用不改变配置文件的方式 。使用如下命令进行PP-OCRv3文本检测预训练模型的评估
+
+
+```bash
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
     Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">上述指令中，通过-c 选择训练使用配置文件，通过-o参数在不需要修改配置文件的情况下，改变训练的参数。</p>
-<p dir="auto">使用预训练模型进行评估，指标如下所示：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
-<td>76.12%</td>
-</tr>
-</tbody>
-</table>
-<h4 dir="auto"><a id="user-content-412-ccpd车牌数据集fine-tune" class="anchor" aria-hidden="true" href="#412-ccpd车牌数据集fine-tune"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.2 CCPD车牌数据集fine-tune</h4>
-<p dir="auto"><strong>训练</strong></p>
-<p dir="auto">为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址</li>
-<li>Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔772个step评估一次，772为一个epoch总的step数。</li>
-</ol>
-</li>
-<li>优化器相关:
-<ol dir="auto">
-<li>Optimizer.lr.name: 学习率衰减器设为常量 Const</li>
-<li>Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍</li>
-<li>Optimizer.lr.warmup_epoch: warmup_epoch设为0</li>
-</ol>
-</li>
-<li>数据集相关:
-<ol dir="auto">
-<li>Train.dataset.data_dir：指向训练集图片存放目录</li>
-<li>Train.dataset.label_file_list：指向训练集标注文件</li>
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">使用如下代码即可启动在CCPD车牌数据集上的fine-tune。</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]
+```
+上述指令中，通过-c 选择训练使用配置文件，通过-o参数在不需要修改配置文件的情况下，改变训练的参数。
+
+使用预训练模型进行评估，指标如下所示：
+
+| 方案                        |hmeans|
+|---------------------------|---|
+| PP-OCRv3中英文超轻量检测预训练模型直接预测 |76.12%|
+
+#### 4.1.2 CCPD车牌数据集fine-tune
+
+**训练**
+
+为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:
+
+1. 模型存储和训练相关:
+   1. Global.pretrained_model: 指向PP-OCRv3文本检测预训练模型地址
+   2. Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔772个step评估一次，772为一个epoch总的step数。
+2. 优化器相关:
+   1. Optimizer.lr.name: 学习率衰减器设为常量 Const
+   2. Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍
+   3. Optimizer.lr.warmup_epoch: warmup_epoch设为0
+3. 数据集相关:
+   1. Train.dataset.data_dir：指向训练集图片存放目录
+   2. Train.dataset.label_file_list：指向训练集标注文件
+   3. Eval.dataset.data_dir：指向测试集图片存放目录
+   4. Eval.dataset.label_file_list：指向测试集标注文件
+
+使用如下代码即可启动在CCPD车牌数据集上的fine-tune。
+
+```bash
+python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
     Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
     Global.save_model_dir=output/CCPD/det \
-    Global.eval_batch_step=&quot;[0, 772]&quot; \
+    Global.eval_batch_step="[0, 772]" \
     Optimizer.lr.name=Const \
     Optimizer.lr.learning_rate=0.0005 \
     Optimizer.lr.warmup_epoch=0 \
     Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
     Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Global.save_model_dir=output/CCPD/det \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 772]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">在上述命令中，通过<code>-o</code>的方式修改了配置文件中的参数。</p>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_finetune.tar" rel="nofollow">det_ppocr_v3_finetune.tar</a></p>
-<p dir="auto"><strong>评估</strong></p>
-<p dir="auto">训练完成后使用如下命令进行评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]
+```
+
+在上述命令中，通过`-o`的方式修改了配置文件中的参数。
+
+训练好的模型地址为: [det_ppocr_v3_finetune.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_finetune.tar)
+
+**评估**
+
+训练完成后使用如下命令进行评估
+
+
+```bash
+python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
     Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
-<td>76.12%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
-<td>99%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到进行fine-tune能显著提升车牌检测的效果。</p>
-<h4 dir="auto"><a id="user-content-413-ccpd车牌数据集fine-tune量化训练" class="anchor" aria-hidden="true" href="#413-ccpd车牌数据集fine-tune量化训练"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.3 CCPD车牌数据集fine-tune+量化训练</h4>
-<p dir="auto">此处采用 PaddleOCR 中提供好的<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/slim/quantization/README.md">量化教程</a>对模型进行量化训练。</p>
-<p dir="auto">量化训练可通过如下命令启动:</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]
+```
+
+使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：
+
+|方案|hmeans|
+|---|---|
+|PP-OCRv3中英文超轻量检测预训练模型直接预测|76.12%|
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%|
+
+可以看到进行fine-tune能显著提升车牌检测的效果。
+
+#### 4.1.3 CCPD车牌数据集fine-tune+量化训练
+
+此处采用 PaddleOCR 中提供好的[量化教程](../deploy/slim/quantization/README.md)对模型进行量化训练。
+
+量化训练可通过如下命令启动:
+
+```bash
+python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
     Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
     Global.save_model_dir=output/CCPD/det_quant \
-    Global.eval_batch_step=&quot;[0, 772]&quot; \
+    Global.eval_batch_step="[0, 772]" \
     Optimizer.lr.name=Const \
     Optimizer.lr.learning_rate=0.0005 \
     Optimizer.lr.warmup_epoch=0 \
     Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
     Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]"><pre>python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]
+```
+
+训练好的模型地址为: [det_ppocr_v3_quant.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_quant.tar)
+
+量化后指标对比如下
+
+|方案|hmeans| 模型大小 | 预测速度(lite) |
+|---|---|------|------------|
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%| 2.5M | 223ms      |
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化|98.91%| 1M   | 189ms      |
+
+可以看到通过量化训练在精度几乎无损的情况下，降低模型体积60%并且推理速度提升15%。
+
+速度测试基于[PaddleOCR lite教程](../deploy/lite/readme_ch.md)完成。
+
+#### 4.1.4 模型导出
+
+使用如下命令可以将训练好的模型进行导出
+
+* 非量化模型
+```bash
+python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
     Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_model_dir=output/CCPD/det_quant \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 772]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/det.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]</pre></div>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_quant.tar" rel="nofollow">det_ppocr_v3_quant.tar</a></p>
-<p dir="auto">量化后指标对比如下</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
-<td>99%</td>
-<td>2.5M</td>
-<td>223ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune+量化</td>
-<td>98.91%</td>
-<td>1M</td>
-<td>189ms</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到通过量化训练在精度几乎无损的情况下，降低模型体积60%并且推理速度提升15%。</p>
-<p dir="auto">速度测试基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>完成。</p>
-<h4 dir="auto"><a id="user-content-414-模型导出" class="anchor" aria-hidden="true" href="#414-模型导出"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.1.4 模型导出</h4>
-<p dir="auto">使用如下命令可以将训练好的模型进行导出</p>
-<ul dir="auto">
-<li>非量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer"><pre>python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer</pre></div>
-<ul dir="auto">
-<li>量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
+    Global.save_inference_dir=output/det/infer
+```
+* 量化模型
+```bash
+python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
     Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer"><pre>python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/det/infer</pre></div>
-<h3 dir="auto"><a id="user-content-42-识别" class="anchor" aria-hidden="true" href="#42-识别"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2 识别</h3>
-<h4 dir="auto"><a id="user-content-421-预训练模型直接预测" class="anchor" aria-hidden="true" href="#421-预训练模型直接预测"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.1 预训练模型直接预测</h4>
-<p dir="auto">从下表中下载PP-OCRv3文本识别预训练模型</p>
-<table>
-<thead>
-<tr>
-<th>模型名称</th>
-<th>模型简介</th>
-<th>配置文件</th>
-<th>推理模型大小</th>
-<th>下载地址</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>ch_PP-OCRv3_rec</td>
-<td>【最新】原始超轻量模型，支持中英文、数字识别</td>
-<td><a href="https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml">ch_PP-OCRv3_rec_distillation.yml</a></td>
-<td>12.4M</td>
-<td><a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar" rel="nofollow">推理模型</a> / <a href="https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar" rel="nofollow">训练模型</a></td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">使用如下命令下载预训练模型</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="mkdir models
+    Global.save_inference_dir=output/det/infer
+```
+
+### 4.2 识别
+#### 4.2.1 预训练模型直接预测
+
+从下表中下载PP-OCRv3文本识别预训练模型
+
+|模型名称|模型简介|配置文件|推理模型大小|下载地址|
+| --- | --- | --- | --- | --- |
+|ch_PP-OCRv3_rec|【最新】原始超轻量模型，支持中英文、数字识别|[ch_PP-OCRv3_rec_distillation.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml)| 12.4M |[推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+
+使用如下命令下载预训练模型
+
+```bash
+mkdir models
 cd models
 wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
 tar -xf ch_PP-OCRv3_rec_train.tar
-cd /home/aistudio/PaddleOCR"><pre>mkdir models
-<span class="pl-c1">cd</span> models
-wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
-tar -xf ch_PP-OCRv3_rec_train.tar
-<span class="pl-c1">cd</span> /home/aistudio/PaddleOCR</pre></div>
-<p dir="auto">PaddleOCR提供的PP-OCRv3识别模型采用蒸馏训练策略，因此提供的预训练模型中会包含<code>Teacher</code>和<code>Student</code>模型的参数，详细信息可参考<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/knowledge_distillation.md">knowledge_distillation.md</a>。 因此，模型下载完成后需要使用如下代码提取<code>Student</code>模型的参数：</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="import paddle
+cd /home/aistudio/PaddleOCR
+```
+
+PaddleOCR提供的PP-OCRv3识别模型采用蒸馏训练策略，因此提供的预训练模型中会包含`Teacher`和`Student`模型的参数，详细信息可参考[knowledge_distillation.md](../doc/doc_ch/knowledge_distillation.md)。 因此，模型下载完成后需要使用如下代码提取`Student`模型的参数：
+
+```python
+import paddle
 # 加载预训练模型
-all_params = paddle.load(&quot;models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams&quot;)
+all_params = paddle.load("models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams")
 # 查看权重参数的keys
 print(all_params.keys())
 # 学生模型的权重提取
-s_params = {key[len(&quot;Student.&quot;):]: all_params[key] for key in all_params if &quot;Student.&quot; in key}
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
 # 查看学生模型权重参数的keys
 print(s_params.keys())
 # 保存
-paddle.save(s_params, &quot;models/ch_PP-OCRv3_rec_train/student.pdparams&quot;)"><pre><span class="pl-k">import</span> <span class="pl-s1">paddle</span>
-<span class="pl-c"># 加载预训练模型</span>
-<span class="pl-s1">all_params</span> <span class="pl-c1">=</span> <span class="pl-s1">paddle</span>.<span class="pl-en">load</span>(<span class="pl-s">"models/ch_PP-OCRv3_rec_train/best_accuracy.pdparams"</span>)
-<span class="pl-c"># 查看权重参数的keys</span>
-<span class="pl-en">print</span>(<span class="pl-s1">all_params</span>.<span class="pl-en">keys</span>())
-<span class="pl-c"># 学生模型的权重提取</span>
-<span class="pl-s1">s_params</span> <span class="pl-c1">=</span> {<span class="pl-s1">key</span>[<span class="pl-en">len</span>(<span class="pl-s">"Student."</span>):]: <span class="pl-s1">all_params</span>[<span class="pl-s1">key</span>] <span class="pl-k">for</span> <span class="pl-s1">key</span> <span class="pl-c1">in</span> <span class="pl-s1">all_params</span> <span class="pl-k">if</span> <span class="pl-s">"Student."</span> <span class="pl-c1">in</span> <span class="pl-s1">key</span>}
-<span class="pl-c"># 查看学生模型权重参数的keys</span>
-<span class="pl-en">print</span>(<span class="pl-s1">s_params</span>.<span class="pl-en">keys</span>())
-<span class="pl-c"># 保存</span>
-<span class="pl-s1">paddle</span>.<span class="pl-en">save</span>(<span class="pl-s1">s_params</span>, <span class="pl-s">"models/ch_PP-OCRv3_rec_train/student.pdparams"</span>)</pre></div>
-<p dir="auto">预训练模型下载完成后，我们使用<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml">ch_PP-OCRv3_rec.yml</a> 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址</li>
-</ol>
-</li>
-<li>数据集相关
-<ol dir="auto">
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">使用如下命令进行PP-OCRv3文本识别预训练模型的评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+paddle.save(s_params, "models/ch_PP-OCRv3_rec_train/student.pdparams")
+```
+
+预训练模型下载完成后，我们使用[ch_PP-OCRv3_rec.yml](../configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml) 配置文件进行后续实验，在开始评估之前需要对配置文件中部分字段进行设置，具体如下：
+
+1. 模型存储和训练相关:
+   1. Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址
+2. 数据集相关
+   1. Eval.dataset.data_dir：指向测试集图片存放目录
+   2. Eval.dataset.label_file_list：指向测试集标注文件
+
+使用如下命令进行PP-OCRv3文本识别预训练模型的评估
+
+```bash
+python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">评估部分日志如下：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
-eval model:: 100%|██████████████████████████████| 40/40 [00:15&lt;00:00,  2.57it/s]
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
+```
+
+评估部分日志如下：
+```bash
+[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
+eval model:: 100%|██████████████████████████████| 40/40 [00:15<00:00,  2.57it/s]
 [2022/05/12 19:52:17] ppocr INFO: metric eval ***************
 [2022/05/12 19:52:17] ppocr INFO: acc:0.0
 [2022/05/12 19:52:17] ppocr INFO: norm_edit_dis:0.8656084923002452
 [2022/05/12 19:52:17] ppocr INFO: Teacher_acc:0.000399520574511545
 [2022/05/12 19:52:17] ppocr INFO: Teacher_norm_edit_dis:0.8657902943394548
 [2022/05/12 19:52:17] ppocr INFO: fps:1443.1801978719905
-"><pre>[2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
-<span class="pl-c1">eval</span> model:: 100%<span class="pl-k">|</span>██████████████████████████████<span class="pl-k">|</span> 40/40 [00:<span class="pl-k">15&lt;</span>00:00,  2.57it/s]
-[2022/05/12 19:52:17] ppocr INFO: metric <span class="pl-c1">eval</span> <span class="pl-k">***************</span>
-[2022/05/12 19:52:17] ppocr INFO: acc:0.0
-[2022/05/12 19:52:17] ppocr INFO: norm_edit_dis:0.8656084923002452
-[2022/05/12 19:52:17] ppocr INFO: Teacher_acc:0.000399520574511545
-[2022/05/12 19:52:17] ppocr INFO: Teacher_norm_edit_dis:0.8657902943394548
-[2022/05/12 19:52:17] ppocr INFO: fps:1443.1801978719905
-</pre></div>
-<p dir="auto">使用预训练模型进行评估，指标如下所示：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">从评估日志中可以看到，直接使用PP-OCRv3预训练模型进行评估，acc非常低，但是norm_edit_dis很高。因此，我们猜测是模型大部分文字识别是对的，只有少部分文字识别错误。使用如下命令进行infer查看模型的推理结果进行验证：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+
+```
+使用预训练模型进行评估，指标如下所示：
+
+|方案|acc|
+|---|---|
+|PP-OCRv3中英文超轻量识别预训练模型直接预测|0%|
+
+从评估日志中可以看到，直接使用PP-OCRv3预训练模型进行评估，acc非常低，但是norm_edit_dis很高。因此，我们猜测是模型大部分文字识别是对的，只有少部分文字识别错误。使用如下命令进行infer查看模型的推理结果进行验证：
+
+
+```bash
+python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg"><pre>python tools/infer_rec.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg</pre></div>
-<p dir="auto">输出部分日志如下：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
+    Global.infer_img=/home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_0_3_32_30_31_30_30.jpg
+```
+
+输出部分日志如下：
+```bash
+[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
 W0501 08:51:57.127391 11326 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
 W0501 08:51:57.132315 11326 device_context.cc:465] device: 0, cuDNN Version: 7.6.
 [2022/05/01 08:52:00] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/student
 [2022/05/01 08:52:00] ppocr INFO: infer_img: /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_3_32_30_31_30_30.jpg
-[2022/05/01 08:52:00] ppocr INFO:      result: {&quot;Student&quot;: {&quot;label&quot;: &quot;皖A·D86766&quot;, &quot;score&quot;: 0.9552637934684753}, &quot;Teacher&quot;: {&quot;label&quot;: &quot;皖A·D86766&quot;, &quot;score&quot;: 0.9917094707489014}}
-[2022/05/01 08:52:00] ppocr INFO: success!"><pre>[2022/05/01 08:51:57] ppocr INFO: train with paddle 2.2.2 and device CUDAPlace(0)
-W0501 08:51:57.127391 11326 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.0, Runtime API Version: 10.1
-W0501 08:51:57.132315 11326 device_context.cc:465] device: 0, cuDNN Version: 7.6.
-[2022/05/01 08:52:00] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/student
-[2022/05/01 08:52:00] ppocr INFO: infer_img: /home/aistudio/data/CCPD2020/PPOCR/test/crop_imgs/0_0_3_32_30_31_30_30.jpg
-[2022/05/01 08:52:00] ppocr INFO:      result: {<span class="pl-s"><span class="pl-pds">"</span>Student<span class="pl-pds">"</span></span>: {<span class="pl-s"><span class="pl-pds">"</span>label<span class="pl-pds">"</span></span>: <span class="pl-s"><span class="pl-pds">"</span>皖A·D86766<span class="pl-pds">"</span></span>, <span class="pl-s"><span class="pl-pds">"</span>score<span class="pl-pds">"</span></span>: 0.9552637934684753}, <span class="pl-s"><span class="pl-pds">"</span>Teacher<span class="pl-pds">"</span></span>: {<span class="pl-s"><span class="pl-pds">"</span>label<span class="pl-pds">"</span></span>: <span class="pl-s"><span class="pl-pds">"</span>皖A·D86766<span class="pl-pds">"</span></span>, <span class="pl-s"><span class="pl-pds">"</span>score<span class="pl-pds">"</span></span>: 0.9917094707489014}}
-[2022/05/01 08:52:00] ppocr INFO: success<span class="pl-k">!</span></pre></div>
-<p dir="auto">从infer结果可以看到，车牌中的文字大部分都识别正确，只是多识别出了一个<code>·</code>。针对这种情况，有如下两种方案：</p>
-<ol dir="auto">
-<li>直接通过后处理去掉多识别的<code>·</code>。</li>
-<li>进行 fine-tune。</li>
-</ol>
-<h4 dir="auto"><a id="user-content-422-预训练模型直接预测改动后处理" class="anchor" aria-hidden="true" href="#422-预训练模型直接预测改动后处理"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.2 预训练模型直接预测+改动后处理</h4>
-<p dir="auto">直接通过后处理去掉多识别的<code>·</code>，在后处理的改动比较简单，只需在 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/postprocess/rec_postprocess.py">ppocr/postprocess/rec_postprocess.py</a> 文件的76行添加如下代码:</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="text = text.replace('·','')"><pre><span class="pl-s1">text</span> <span class="pl-c1">=</span> <span class="pl-s1">text</span>.<span class="pl-en">replace</span>(<span class="pl-s">'·'</span>,<span class="pl-s">''</span>)</pre></div>
-<p dir="auto">改动前后指标对比:</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0.2%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
-<td>90.97%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到，去掉多余的<code>·</code>能大幅提高精度。</p>
-<h4 dir="auto"><a id="user-content-423-ccpd车牌数据集fine-tune" class="anchor" aria-hidden="true" href="#423-ccpd车牌数据集fine-tune"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.3 CCPD车牌数据集fine-tune</h4>
-<p dir="auto"><strong>训练</strong></p>
-<p dir="auto">为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:</p>
-<ol dir="auto">
-<li>模型存储和训练相关:
-<ol dir="auto">
-<li>Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址</li>
-<li>Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔45个step评估一次，45为一个epoch总的step数。</li>
-</ol>
-</li>
-<li>优化器相关
-<ol dir="auto">
-<li>Optimizer.lr.name: 学习率衰减器设为常量 Const</li>
-<li>Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍</li>
-<li>Optimizer.lr.warmup_epoch: warmup_epoch设为0</li>
-</ol>
-</li>
-<li>数据集相关
-<ol dir="auto">
-<li>Train.dataset.data_dir：指向训练集图片存放目录</li>
-<li>Train.dataset.label_file_list：指向训练集标注文件</li>
-<li>Eval.dataset.data_dir：指向测试集图片存放目录</li>
-<li>Eval.dataset.label_file_list：指向测试集标注文件</li>
-</ol>
-</li>
-</ol>
-<p dir="auto">使用如下命令启动 fine-tune</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+[2022/05/01 08:52:00] ppocr INFO:      result: {"Student": {"label": "皖A·D86766", "score": 0.9552637934684753}, "Teacher": {"label": "皖A·D86766", "score": 0.9917094707489014}}
+[2022/05/01 08:52:00] ppocr INFO: success!
+```
+
+从infer结果可以看到，车牌中的文字大部分都识别正确，只是多识别出了一个`·`。针对这种情况，有如下两种方案：
+1. 直接通过后处理去掉多识别的`·`。
+2. 进行 fine-tune。
+
+#### 4.2.2 预训练模型直接预测+改动后处理
+
+直接通过后处理去掉多识别的`·`，在后处理的改动比较简单，只需在 [ppocr/postprocess/rec_postprocess.py](../ppocr/postprocess/rec_postprocess.py) 文件的76行添加如下代码:
+```python
+text = text.replace('·','')
+```
+
+改动前后指标对比:
+
+|方案|acc|
+|---|---|
+|PP-OCRv3中英文超轻量识别预训练模型直接预测|0.2%|
+|PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的`·`|90.97%|
+
+可以看到，去掉多余的`·`能大幅提高精度。
+
+#### 4.2.3 CCPD车牌数据集fine-tune
+
+**训练**
+
+为了进行fine-tune训练，我们需要在配置文件中设置需要使用的预训练模型地址，学习率和数据集等参数。 具体如下:
+
+1. 模型存储和训练相关:
+   1. Global.pretrained_model: 指向PP-OCRv3文本识别预训练模型地址
+   2. Global.eval_batch_step: 模型多少step评估一次，这里设为从第0个step开始没隔45个step评估一次，45为一个epoch总的step数。
+2. 优化器相关
+   1. Optimizer.lr.name: 学习率衰减器设为常量 Const
+   2. Optimizer.lr.learning_rate: 做 fine-tune 实验，学习率需要设置的比较小，此处学习率设为配置文件中的0.05倍
+   3. Optimizer.lr.warmup_epoch: warmup_epoch设为0
+3. 数据集相关
+   1. Train.dataset.data_dir：指向训练集图片存放目录
+   2. Train.dataset.label_file_list：指向训练集标注文件
+   3. Eval.dataset.data_dir：指向测试集图片存放目录
+   4. Eval.dataset.label_file_list：指向测试集标注文件
+
+使用如下命令启动 fine-tune
+
+```bash
+python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
     Global.save_model_dir=output/CCPD/rec/ \
-    Global.eval_batch_step=&quot;[0, 90]&quot; \
+    Global.eval_batch_step="[0, 90]" \
     Optimizer.lr.name=Const \
     Optimizer.lr.learning_rate=0.0005 \
     Optimizer.lr.warmup_epoch=0 \
     Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
     Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.save_model_dir=output/CCPD/rec/ \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 90]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_finetune.tar" rel="nofollow">rec_ppocr_v3_finetune.tar</a></p>
-<p dir="auto"><strong>评估</strong></p>
-<p dir="auto">训练完成后使用如下命令进行评估</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
+```
+训练好的模型地址为: [rec_ppocr_v3_finetune.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_finetune.tar)
+
+**评估**
+
+训练完成后使用如下命令进行评估
+
+```bash
+python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
-<td>90.97%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
-<td>94.54%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到进行fine-tune能显著提升车牌识别的效果。</p>
-<h4 dir="auto"><a id="user-content-424-ccpd车牌数据集fine-tune量化训练" class="anchor" aria-hidden="true" href="#424-ccpd车牌数据集fine-tune量化训练"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.4 CCPD车牌数据集fine-tune+量化训练</h4>
-<p dir="auto">此处采用 PaddleOCR 中提供好的<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/slim/quantization/README.md">量化教程</a>对模型进行量化训练。</p>
-<p dir="auto">量化训练可通过如下命令启动:</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
+```
+
+使用预训练模型和CCPD车牌数据集fine-tune，指标分别如下：
+
+|方案| acc    |
+|---|--------|
+|PP-OCRv3中英文超轻量识别预训练模型直接预测| 0%     |
+|PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的`·`| 90.97% |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% |
+
+可以看到进行fine-tune能显著提升车牌识别的效果。
+
+#### 4.2.4 CCPD车牌数据集fine-tune+量化训练
+
+此处采用 PaddleOCR 中提供好的[量化教程](../deploy/slim/quantization/README.md)对模型进行量化训练。
+
+量化训练可通过如下命令启动:
+
+```bash
+python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
     Global.save_model_dir=output/CCPD/rec_quant/ \
-    Global.eval_batch_step=&quot;[0, 90]&quot; \
+    Global.eval_batch_step="[0, 90]" \
     Optimizer.lr.name=Const \
     Optimizer.lr.learning_rate=0.0005 \
     Optimizer.lr.warmup_epoch=0 \
     Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
     Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre>python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
+```
+训练好的模型地址为: [rec_ppocr_v3_quant.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_quant.tar)
+
+量化后指标对比如下
+
+|方案| acc    | 模型大小  | 预测速度(lite) |
+|---|--------|-------|------------|
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% | 10.3M | 4.2ms      |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化| 93.4%  | 4.8M  | 1.8ms      |
+
+可以看到量化后能降低模型体积53%并且推理速度提升57%，但是由于识别数据过少，量化带来了1%的精度下降。
+
+速度测试基于[PaddleOCR lite教程](../deploy/lite/readme_ch.md)完成。
+
+#### 4.2.5 模型导出
+
+使用如下命令可以将训练好的模型进行导出。
+
+* 非量化模型
+```bash
+python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_model_dir=output/CCPD/rec_quant/ \
-    Global.eval_batch_step=<span class="pl-s"><span class="pl-pds">"</span>[0, 90]<span class="pl-pds">"</span></span> \
-    Optimizer.lr.name=Const \
-    Optimizer.lr.learning_rate=0.0005 \
-    Optimizer.lr.warmup_epoch=0 \
-    Train.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Train.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/train/rec.txt] \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<p dir="auto">训练好的模型地址为: <a href="https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_quant.tar" rel="nofollow">rec_ppocr_v3_quant.tar</a></p>
-<p dir="auto">量化后指标对比如下</p>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
-<td>94.54%</td>
-<td>10.3M</td>
-<td>4.2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化</td>
-<td>93.4%</td>
-<td>4.8M</td>
-<td>1.8ms</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">可以看到量化后能降低模型体积53%并且推理速度提升57%，但是由于识别数据过少，量化带来了1%的精度下降。</p>
-<p dir="auto">速度测试基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>完成。</p>
-<h4 dir="auto"><a id="user-content-425-模型导出" class="anchor" aria-hidden="true" href="#425-模型导出"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.2.5 模型导出</h4>
-<p dir="auto">使用如下命令可以将训练好的模型进行导出。</p>
-<ul dir="auto">
-<li>非量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer"><pre>python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer</pre></div>
-<ul dir="auto">
-<li>量化模型</li>
-</ul>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
+    Global.save_inference_dir=output/CCPD/rec/infer
+```
+* 量化模型
+```bash
+python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer"><pre>python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer</pre></div>
-<h3 dir="auto"><a id="user-content-43-计算end2end指标" class="anchor" aria-hidden="true" href="#43-计算end2end指标"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.3 计算End2End指标</h3>
-<p dir="auto">端到端指标可通过 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/tools/end2end/readme.md">PaddleOCR内置脚本</a> 进行计算，具体步骤如下：</p>
-<ol dir="auto">
-<li>导出模型</li>
-</ol>
-<p dir="auto">通过如下命令进行模型的导出。注意，量化模型导出时，需要配置eval数据集</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="# 检测模型
+    Global.save_inference_dir=output/CCPD/rec_quant/infer
+```
+
+### 4.3 计算End2End指标
+
+端到端指标可通过 [PaddleOCR内置脚本](../tools/end2end/readme.md) 进行计算，具体步骤如下：
+
+1. 导出模型
+
+通过如下命令进行模型的导出。注意，量化模型导出时，需要配置eval数据集
+
+```bash
+# 检测模型
 
 # 预训练模型
 python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
@@ -2369,68 +646,30 @@ python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OC
     Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
     Global.save_inference_dir=output/CCPD/rec_quant/infer \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]"><pre><span class="pl-c"><span class="pl-c">#</span> 检测模型</span>
+    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
+```
 
-<span class="pl-c"><span class="pl-c">#</span> 预训练模型</span>
-python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_det_distill_train/student.pdparams \
-    Global.save_inference_dir=output/ch_PP-OCRv3_det_distill_train/infer
+2. 用导出的模型对测试集进行预测
 
-<span class="pl-c"><span class="pl-c">#</span> 非量化模型</span>
-python tools/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/det/infer
+此处，分别使用PP-OCRv3预训练模型，fintune模型和量化模型对测试集的所有图像进行预测，命令如下：
 
-<span class="pl-c"><span class="pl-c">#</span> 量化模型</span>
-python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o \
-    Global.pretrained_model=output/CCPD/det_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/det_quant/infer \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/ccpd_green \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt] \
-    Eval.loader.num_workers=0
-
-<span class="pl-c"><span class="pl-c">#</span> 识别模型</span>
-
-<span class="pl-c"><span class="pl-c">#</span> 预训练模型</span>
-python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=models/ch_PP-OCRv3_rec_train/student.pdparams \
-    Global.save_inference_dir=output/ch_PP-OCRv3_rec_train/infer
-
-<span class="pl-c"><span class="pl-c">#</span> 非量化模型</span>
-python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec/infer
-
-<span class="pl-c"><span class="pl-c">#</span> 量化模型</span>
-python deploy/slim/quantization/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
-    Global.pretrained_model=output/CCPD/rec_quant/best_accuracy.pdparams \
-    Global.save_inference_dir=output/CCPD/rec_quant/infer \
-    Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
-    Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]</pre></div>
-<ol start="2" dir="auto">
-<li>用导出的模型对测试集进行预测</li>
-</ol>
-<p dir="auto">此处，分别使用PP-OCRv3预训练模型，fintune模型和量化模型对测试集的所有图像进行预测，命令如下：</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型
+```bash
+# PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型
 python3 tools/infer/predict_system.py --det_model_dir=models/ch_PP-OCRv3_det_distill_train/infer --rec_model_dir=models/ch_PP-OCRv3_rec_train/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/pretrain --use_dilation=true
 
 # PP-OCRv3中英文超轻量检测预训练模型+fine-tune，PP-OCRv3中英文超轻量识别预训练模型+fine-tune
 python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det/infer --rec_model_dir=output/CCPD/rec/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/fine-tune --use_dilation=true
 
 # PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true"><pre><span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型</span>
-python3 tools/infer/predict_system.py --det_model_dir=models/ch_PP-OCRv3_det_distill_train/infer --rec_model_dir=models/ch_PP-OCRv3_rec_train/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/pretrain --use_dilation=true
+python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true
+```
 
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型+fine-tune，PP-OCRv3中英文超轻量识别预训练模型+fine-tune</span>
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det/infer --rec_model_dir=output/CCPD/rec/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/fine-tune --use_dilation=true
+3. 转换label并计算指标
 
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估</span>
-python3 tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer --rec_model_dir=output/CCPD/rec_quant/infer --det_limit_side_len=736 --det_limit_type=min --image_dir=/home/aistudio/data/CCPD2020/ccpd_green/test/ --draw_img_save_dir=infer/quant --use_dilation=true</pre></div>
-<ol start="3" dir="auto">
-<li>转换label并计算指标</li>
-</ol>
-<p dir="auto">将gt和上一步保存的预测结果转换为端对端评测需要的数据格式，并根据转换后的数据进行端到端指标计算</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
+将gt和上一步保存的预测结果转换为端对端评测需要的数据格式，并根据转换后的数据进行端到端指标计算
+
+```bash
+python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
 
 # PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型 结果转换和评估
 python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/pretrain/system_results.txt --save_folder=end2end/pretrain
@@ -2447,26 +686,12 @@ python3 tools/end2end/eval_end2end.py end2end/gt end2end/fine-tune
 
 # PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估
 python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/quant/system_results.txt --save_folder=end2end/quant
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant"><pre>python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=/home/aistudio/data/CCPD2020/PPOCR/test/det.txt --save_folder=end2end/gt
+python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant
+```
 
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型 结果转换和评估</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/pretrain/system_results.txt --save_folder=end2end/pretrain
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/pretrain
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型，PP-OCRv3中英文超轻量识别预训练模型+后处理去掉多识别的`·` 结果转换和评估</span>
-<span class="pl-c"><span class="pl-c">#</span> 需手动修改后处理函数</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/post/system_results.txt --save_folder=end2end/post
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/post
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune，PP-OCRv3中英文超轻量识别预训练模型 fine-tune 结果转换和评估</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/fine-tune/system_results.txt --save_folder=end2end/fine-tune
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/fine-tune
-
-<span class="pl-c"><span class="pl-c">#</span> PP-OCRv3中英文超轻量检测预训练模型 fine-tune +量化，PP-OCRv3中英文超轻量识别预训练模型 fine-tune +量化 结果转换和评估</span>
-python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=infer/quant/system_results.txt --save_folder=end2end/quant
-python3 tools/end2end/eval_end2end.py end2end/gt end2end/quant</pre></div>
-<p dir="auto">日志如下:</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="The convert label saved in end2end/gt
+日志如下:
+```bash
+The convert label saved in end2end/gt
 The convert label saved in end2end/pretrain
 start testing...
 hit, dt_count, gt_count 2 5988 5006
@@ -2502,401 +727,105 @@ avg_edit_dist_field: 0.47
 avg_edit_dist_img: 0.47
 precision: 87.84%
 recall: 86.88%
-fmeasure: 87.36%"><pre>The convert label saved <span class="pl-k">in</span> end2end/gt
-The convert label saved <span class="pl-k">in</span> end2end/pretrain
-start testing...
-hit, dt_count, gt_count 2 5988 5006
-character_acc: 70.42%
-avg_edit_dist_field: 2.37
-avg_edit_dist_img: 2.37
-precision: 0.03%
-recall: 0.04%
-fmeasure: 0.04%
-The convert label saved <span class="pl-k">in</span> end2end/post
-start testing...
-hit, dt_count, gt_count 4224 5988 5006
-character_acc: 81.59%
-avg_edit_dist_field: 1.47
-avg_edit_dist_img: 1.47
-precision: 70.54%
-recall: 84.38%
-fmeasure: 76.84%
-The convert label saved <span class="pl-k">in</span> end2end/fine-tune
-start testing...
-hit, dt_count, gt_count 4286 4898 5006
-character_acc: 94.16%
-avg_edit_dist_field: 0.47
-avg_edit_dist_img: 0.47
-precision: 87.51%
-recall: 85.62%
-fmeasure: 86.55%
-The convert label saved <span class="pl-k">in</span> end2end/quant
-start testing...
-hit, dt_count, gt_count 4349 4951 5006
-character_acc: 94.13%
-avg_edit_dist_field: 0.47
-avg_edit_dist_img: 0.47
-precision: 87.84%
-recall: 86.88%
-fmeasure: 87.36%</pre></div>
-<p dir="auto">各个方案端到端指标如下：</p>
-<table>
-<thead>
-<tr>
-<th>模型</th>
-<th>指标</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
-<td>0.04%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
-<td>78.27%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
-<td>87.14%</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
-<td>88%</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到78.27%，在CCPD数据集上进行 fine-tune 后指标进一步提升到87.14%, 在经过量化训练之后，由于检测模型的recall变高，指标进一步提升到88%。但是这个结果仍旧不符合检测模型+识别模型的真实性能(99%*94%=93%)，因此我们需要对 base case 进行具体分析。</p>
-<p dir="auto">在之前的端到端预测结果中，可以看到很多不符合车牌标注的文字被识别出来, 因此可以进行简单的过滤来提升precision</p>
-<p dir="auto">为了快速评估，我们在 <code> tools/end2end/convert_ppocr_label.py</code> 脚本的 58 行加入如下代码，对非8个字符的结果进行过滤</p>
-<div class="highlight highlight-source-python notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="if len(txt) != 8: # 车牌字符串长度为8
-    continue"><pre><span class="pl-k">if</span> <span class="pl-en">len</span>(<span class="pl-s1">txt</span>) <span class="pl-c1">!=</span> <span class="pl-c1">8</span>: <span class="pl-c"># 车牌字符串长度为8</span>
-    <span class="pl-k">continue</span></pre></div>
-<p dir="auto">此外，通过可视化box可以发现有很多框都是竖直翻转之后的框，并且没有完全框住车牌边界，因此需要进行框的竖直翻转以及轻微扩大，示意图如下：</p>
-<p dir="auto"><a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/199752de33937424916ca3fd6ef50f83461dca6034c35d251e32a4eecd4c3228/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f35396162303431316338656234646664393137666232623665356236396131376565376361343833353134343461656339616336313034623739666631303238"><img src="https://camo.githubusercontent.com/199752de33937424916ca3fd6ef50f83461dca6034c35d251e32a4eecd4c3228/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f35396162303431316338656234646664393137666232623665356236396131376565376361343833353134343461656339616336313034623739666631303238" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/59ab0411c8eb4dfd917fb2b6e5b69a17ee7ca48351444aec9ac6104b79ff1028" style="max-width: 100%;"></a></p>
-<p dir="auto">修改前后个方案指标对比如下：</p>
-<p dir="auto">各个方案端到端指标如下：</p>
-<table>
-<thead>
-<tr>
-<th>模型</th>
-<th>base</th>
-<th>A:识别结果过滤</th>
-<th>B:use_dilation</th>
-<th>C:flip_box</th>
-<th>best</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
-<td>0.04%</td>
-<td>0.08%</td>
-<td>0.02%</td>
-<td>0.05%</td>
-<td>0.00%(A)</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
-<td>78.27%</td>
-<td>90.84%</td>
-<td>78.61%</td>
-<td>79.43%</td>
-<td>91.66%(A+B+C)</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
-<td>87.14%</td>
-<td>90.40%</td>
-<td>87.66%</td>
-<td>89.98</td>
-<td>92.5%(A+B+C)</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
-<td>88%</td>
-<td>90.54%</td>
-<td>88.5%</td>
-<td>89.46%</td>
-<td>92.02%(A+B+C)</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto">从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标变为92.02%。</p>
-<h3 dir="auto"><a id="user-content-44-部署" class="anchor" aria-hidden="true" href="#44-部署"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.4 部署</h3>
-<ul dir="auto">
-<li>基于 Paddle Inference 的python推理</li>
-</ul>
-<p dir="auto">检测模型和识别模型分别 fine-tune 并导出为inference模型之后，可以使用如下命令基于 Paddle Inference 进行端到端推理并对结果进行可视化。</p>
-<div class="highlight highlight-source-shell notranslate position-relative overflow-auto" data-snippet-clipboard-copy-content="python tools/infer/predict_system.py \
+fmeasure: 87.36%
+```
+
+各个方案端到端指标如下：
+
+|模型| 指标     |
+|---|--------|
+|PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型| 0.04%  |
+|PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的`·`| 78.27% |
+|PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune| 87.14% |
+|PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化| 88%    |
+
+从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到78.27%，在CCPD数据集上进行 fine-tune 后指标进一步提升到87.14%, 在经过量化训练之后，由于检测模型的recall变高，指标进一步提升到88%。但是这个结果仍旧不符合检测模型+识别模型的真实性能(99%*94%=93%)，因此我们需要对 base case 进行具体分析。
+
+在之前的端到端预测结果中，可以看到很多不符合车牌标注的文字被识别出来, 因此可以进行简单的过滤来提升precision
+
+为了快速评估，我们在 ` tools/end2end/convert_ppocr_label.py` 脚本的 58 行加入如下代码，对非8个字符的结果进行过滤
+```python
+if len(txt) != 8: # 车牌字符串长度为8
+    continue
+```
+
+此外，通过可视化box可以发现有很多框都是竖直翻转之后的框，并且没有完全框住车牌边界，因此需要进行框的竖直翻转以及轻微扩大，示意图如下：
+
+![](https://ai-studio-static-online.cdn.bcebos.com/59ab0411c8eb4dfd917fb2b6e5b69a17ee7ca48351444aec9ac6104b79ff1028)
+
+修改前后个方案指标对比如下：
+
+
+各个方案端到端指标如下：
+
+|模型|base|A:识别结果过滤|B:use_dilation|C:flip_box|best|
+|---|---|---|---|---|---|
+|PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型|0.04%|0.08%|0.02%|0.05%|0.00%(A)|
+|PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的`·`|78.27%|90.84%|78.61%|79.43%|91.66%(A+B+C)|
+|PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune|87.14%|90.40%|87.66%|89.98|92.5%(A+B+C)|
+|PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化|88%|90.54%|88.5%|89.46%|92.02%(A+B+C)|
+
+
+从结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标变为92.02%。
+
+### 4.4 部署
+
+- 基于 Paddle Inference 的python推理
+
+检测模型和识别模型分别 fine-tune 并导出为inference模型之后，可以使用如下命令基于 Paddle Inference 进行端到端推理并对结果进行可视化。
+
+```bash
+python tools/infer/predict_system.py \
     --det_model_dir=output/CCPD/det/infer/ \
     --rec_model_dir=output/CCPD/rec/infer/ \
-    --image_dir=&quot;/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&amp;509_530&amp;611-527&amp;611_172&amp;599_159&amp;509_530&amp;525-0_0_3_32_30_31_30_30-109-106.jpg&quot; \
-    --rec_image_shape=3,48,320"><pre>python tools/infer/predict_system.py \
-    --det_model_dir=output/CCPD/det/infer/ \
-    --rec_model_dir=output/CCPD/rec/infer/ \
-    --image_dir=<span class="pl-s"><span class="pl-pds">"</span>/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&amp;509_530&amp;611-527&amp;611_172&amp;599_159&amp;509_530&amp;525-0_0_3_32_30_31_30_30-109-106.jpg<span class="pl-pds">"</span></span> \
-    --rec_image_shape=3,48,320</pre></div>
-<p dir="auto">推理结果如下</p>
-<p dir="auto"><a target="_blank" rel="noopener noreferrer" href="https://camo.githubusercontent.com/b5c04d7affeca84aebcff3bf829edff2d142d26d700fd208c24b983144df766d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f37366236613039333963326334636634393033396236353633633462323865323431653131323835643734363465373939653831633538633066373730376137"><img src="https://camo.githubusercontent.com/b5c04d7affeca84aebcff3bf829edff2d142d26d700fd208c24b983144df766d/68747470733a2f2f61692d73747564696f2d7374617469632d6f6e6c696e652e63646e2e626365626f732e636f6d2f37366236613039333963326334636634393033396236353633633462323865323431653131323835643734363465373939653831633538633066373730376137" alt="" data-canonical-src="https://ai-studio-static-online.cdn.bcebos.com/76b6a0939c2c4cf49039b6563c4b28e241e11285d7464e799e81c58c0f7707a7" style="max-width: 100%;"></a></p>
-<ul dir="auto">
-<li>端侧部署</li>
-</ul>
-<p dir="auto">端侧部署我们采用基于 PaddleLite 的 cpp 推理。Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。具体可参考 <a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a></p>
-<h3 dir="auto"><a id="user-content-45-实验总结" class="anchor" aria-hidden="true" href="#45-实验总结"><svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path fill-rule="evenodd" d="M7.775 3.275a.75.75 0 001.06 1.06l1.25-1.25a2 2 0 112.83 2.83l-2.5 2.5a2 2 0 01-2.83 0 .75.75 0 00-1.06 1.06 3.5 3.5 0 004.95 0l2.5-2.5a3.5 3.5 0 00-4.95-4.95l-1.25 1.25zm-4.69 9.64a2 2 0 010-2.83l2.5-2.5a2 2 0 012.83 0 .75.75 0 001.06-1.06 3.5 3.5 0 00-4.95 0l-2.5 2.5a3.5 3.5 0 004.95 4.95l1.25-1.25a.75.75 0 00-1.06-1.06l-1.25 1.25a2 2 0 01-2.83 0z"></path></svg></a>4.5 实验总结</h3>
-<p dir="auto">我们分别使用PP-OCRv3中英文超轻量预训练模型在车牌数据集上进行了直接评估和 fine-tune 和 fine-tune +量化3种方案的实验，并基于<a href="/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/lite/readme_ch.md">PaddleOCR lite教程</a>进行了速度测试，指标对比如下：</p>
-<ul dir="auto">
-<li>检测</li>
-</ul>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>hmeans</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型直接预测</td>
-<td>76.12%</td>
-<td>2.5M</td>
-<td>233ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune</td>
-<td>99%</td>
-<td>2.5M</td>
-<td>233ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 fine-tune + 量化</td>
-<td>98.91%</td>
-<td>1M</td>
-<td>189ms</td>
-</tr>
-</tbody>
-</table>
-<ul dir="auto">
-<li>识别</li>
-</ul>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>acc</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测</td>
-<td>0%</td>
-<td>10.3M</td>
-<td>4.2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的<code>·</code></td>
-<td>90.97%</td>
-<td>10.3M</td>
-<td>4.2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune</td>
-<td>94.54%</td>
-<td>10.3M</td>
-<td>4,2ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化</td>
-<td>93.4%</td>
-<td>4.8M</td>
-<td>1.8ms</td>
-</tr>
-</tbody>
-</table>
-<ul dir="auto">
-<li>端到端指标如下：</li>
-</ul>
-<table>
-<thead>
-<tr>
-<th>方案</th>
-<th>fmeasure</th>
-<th>模型大小</th>
-<th>预测速度(lite)</th>
-</tr>
-</thead>
-<tbody>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型</td>
-<td>0.08%</td>
-<td>12.8M</td>
-<td>298ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的<code>·</code></td>
-<td>91.66%</td>
-<td>12.8M</td>
-<td>298ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune</td>
-<td>92.5%</td>
-<td>12.8M</td>
-<td>298ms</td>
-</tr>
-<tr>
-<td>PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化</td>
-<td>92.02%</td>
-<td>5.8M</td>
-<td>224ms</td>
-</tr>
-</tbody>
-</table>
-<p dir="auto"><strong>结论</strong></p>
-<p dir="auto">PP-OCRv3的检测模型在未经过fine-tune的情况下，在车牌数据集上也有一定的精度，经过 fine-tune 后能够极大的提升检测效果，精度达到99%。在使用量化训练后检测模型的精度几乎无损，并且模型大小压缩60%。</p>
-<p dir="auto">PP-OCRv3的识别模型在未经过fine-tune的情况下，在车牌数据集上精度为0，但是经过分析可以知道，模型大部分字符都预测正确，但是会多预测一个特殊字符，去掉这个特殊字符后，精度达到90%。PP-OCRv3识别模型在经过 fine-tune 后识别精度进一步提升，达到94.4%。在使用量化训练后识别模型大小压缩53%，但是由于数据量多少，带来了1%的精度损失。</p>
-<p dir="auto">从端到端结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标轻微下降到92.02%但模型大小降低54%。</p>
-</article>
-  </div>
+    --image_dir="/home/aistudio/data/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg" \
+    --rec_image_shape=3,48,320
+```
+推理结果如下
 
-    </div>
+![](https://ai-studio-static-online.cdn.bcebos.com/76b6a0939c2c4cf49039b6563c4b28e241e11285d7464e799e81c58c0f7707a7)
 
-  </readme-toc>
+- 端侧部署
+
+端侧部署我们采用基于 PaddleLite 的 cpp 推理。Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。具体可参考 [PaddleOCR lite教程](../deploy/lite/readme_ch.md)
 
 
+### 4.5 实验总结
 
-  <details class="details-reset details-overlay details-overlay-dark" id="jumpto-line-details-dialog">
-    <summary data-hotkey="l" aria-label="Jump to line"></summary>
-    <details-dialog class="Box Box--overlay d-flex flex-column anim-fade-in fast linejump" aria-label="Jump to line">
-      <!-- '"` --><!-- </textarea></xmp> --></option></form><form class="js-jump-to-line-form Box-body d-flex" data-turbo="false" action="" accept-charset="UTF-8" method="get">
-        <input class="form-control flex-auto mr-3 linejump-input js-jump-to-line-field" type="text" placeholder="Jump to line&hellip;" aria-label="Jump to line" autofocus>
-        <button data-close-dialog="" type="submit" data-view-component="true" class="btn">  Go
+我们分别使用PP-OCRv3中英文超轻量预训练模型在车牌数据集上进行了直接评估和 fine-tune 和 fine-tune +量化3种方案的实验，并基于[PaddleOCR lite教程](../deploy/lite/readme_ch.md)进行了速度测试，指标对比如下：
 
-</button>
-</form>    </details-dialog>
-  </details>
+- 检测
 
-    <div class="Popover anim-scale-in js-tagsearch-popover"
-     hidden
-     data-tagsearch-url="/PaddlePaddle/PaddleOCR/find-definition"
-     data-tagsearch-ref="dygraph"
-     data-tagsearch-code-nav-context="BLOB_VIEW">
-  <div class="Popover-message Popover-message--large Popover-message--top-left TagsearchPopover mt-1 mb-4 mx-auto Box color-shadow-large">
-    <div class="TagsearchPopover-content js-tagsearch-popover-content overflow-auto" style="will-change:transform;">
-    </div>
+|方案|hmeans| 模型大小 | 预测速度(lite) |
+|---|---|------|------------|
+|PP-OCRv3中英文超轻量检测预训练模型直接预测|76.12%|2.5M| 233ms      |
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune|99%| 2.5M | 233ms      |
+|PP-OCRv3中英文超轻量检测预训练模型 fine-tune + 量化|98.91%| 1M   | 189ms      |fine-tune
 
-  </div>
-</div>
+- 识别
+
+|方案| acc    | 模型大小  | 预测速度(lite) |
+|---|--------|-------|------------|
+|PP-OCRv3中英文超轻量识别预训练模型直接预测| 0%     |10.3M| 4.2ms      |
+|PP-OCRv3中英文超轻量识别预训练模型直接预测+后处理去掉多识别的`·`| 90.97% |10.3M| 4.2ms      |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune| 94.54% | 10.3M | 4,2ms      |
+|PP-OCRv3中英文超轻量识别预训练模型 fine-tune + 量化| 93.4%  | 4.8M  | 1.8ms      |
 
 
-</div>
+- 端到端指标如下：
 
-  </div>
+|方案|fmeasure|模型大小|预测速度(lite) |
+|---|---|---|---|
+|PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型|0.08%|12.8M|298ms|
+|PP-OCRv3中英文超轻量检测预训练模型 <br> PP-OCRv3中英文超轻量识别预训练模型 + 后处理去掉多识别的`·`|91.66%|12.8M|298ms|
+|PP-OCRv3中英文超轻量检测预训练模型+fine-tune <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune|92.5%|12.8M|298ms|
+|PP-OCRv3中英文超轻量检测预训练模型+fine-tune+量化 <br> PP-OCRv3中英文超轻量识别预训练模型+fine-tune+量化|92.02%|5.8M|224ms|
 
 
-</div>
+**结论**
 
+PP-OCRv3的检测模型在未经过fine-tune的情况下，在车牌数据集上也有一定的精度，经过 fine-tune 后能够极大的提升检测效果，精度达到99%。在使用量化训练后检测模型的精度几乎无损，并且模型大小压缩60%。
 
-    </main>
-  </div>
+PP-OCRv3的识别模型在未经过fine-tune的情况下，在车牌数据集上精度为0，但是经过分析可以知道，模型大部分字符都预测正确，但是会多预测一个特殊字符，去掉这个特殊字符后，精度达到90%。PP-OCRv3识别模型在经过 fine-tune 后识别精度进一步提升，达到94.4%。在使用量化训练后识别模型大小压缩53%，但是由于数据量多少，带来了1%的精度损失。
 
-  </div>
-
-          <footer class="footer width-full container-xl p-responsive" role="contentinfo">
-
-
-  <div class="position-relative d-flex flex-items-center pb-2 f6 color-fg-muted border-top color-border-muted flex-column-reverse flex-lg-row flex-wrap flex-lg-nowrap mt-6 pt-6">
-    <ul class="list-style-none d-flex flex-wrap col-0 col-lg-2 flex-justify-start flex-lg-justify-between mb-2 mb-lg-0">
-      <li class="mt-2 mt-lg-0 d-flex flex-items-center">
-        <a aria-label="Homepage" title="GitHub" class="footer-octicon mr-2" href="https://github.com">
-          <svg aria-hidden="true" height="24" viewBox="0 0 16 16" version="1.1" width="24" data-view-component="true" class="octicon octicon-mark-github">
-    <path fill-rule="evenodd" d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"></path>
-</svg>
-</a>        <span>
-        &copy; 2022 GitHub, Inc.
-        </span>
-      </li>
-    </ul>
-    <ul class="list-style-none d-flex flex-wrap col-12 col-lg-8 flex-justify-center flex-lg-justify-between mb-2 mb-lg-0">
-        <li class="mr-3 mr-lg-0"><a href="https://docs.github.com/en/github/site-policy/github-terms-of-service" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to terms&quot;,&quot;label&quot;:&quot;text:terms&quot;}">Terms</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://docs.github.com/en/github/site-policy/github-privacy-statement" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to privacy&quot;,&quot;label&quot;:&quot;text:privacy&quot;}">Privacy</a></li>
-        <li class="mr-3 mr-lg-0"><a data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to security&quot;,&quot;label&quot;:&quot;text:security&quot;}" href="https://github.com/security">Security</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://www.githubstatus.com/" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to status&quot;,&quot;label&quot;:&quot;text:status&quot;}">Status</a></li>
-        <li class="mr-3 mr-lg-0"><a data-ga-click="Footer, go to help, text:Docs" href="https://docs.github.com">Docs</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://support.github.com?tags=dotcom-footer" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to contact&quot;,&quot;label&quot;:&quot;text:contact&quot;}">Contact GitHub</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://github.com/pricing" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to Pricing&quot;,&quot;label&quot;:&quot;text:Pricing&quot;}">Pricing</a></li>
-      <li class="mr-3 mr-lg-0"><a href="https://docs.github.com" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to api&quot;,&quot;label&quot;:&quot;text:api&quot;}">API</a></li>
-      <li class="mr-3 mr-lg-0"><a href="https://services.github.com" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to training&quot;,&quot;label&quot;:&quot;text:training&quot;}">Training</a></li>
-        <li class="mr-3 mr-lg-0"><a href="https://github.blog" data-analytics-event="{&quot;category&quot;:&quot;Footer&quot;,&quot;action&quot;:&quot;go to blog&quot;,&quot;label&quot;:&quot;text:blog&quot;}">Blog</a></li>
-        <li><a data-ga-click="Footer, go to about, text:about" href="https://github.com/about">About</a></li>
-    </ul>
-  </div>
-  <div class="d-flex flex-justify-center pb-6">
-    <span class="f6 color-fg-muted"></span>
-  </div>
-</footer>
-
-
-
-
-  <div id="ajax-error-message" class="ajax-error-message flash flash-error" hidden>
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert">
-    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
-</svg>
-    <button type="button" class="flash-close js-ajax-error-dismiss" aria-label="Dismiss error">
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-    </button>
-    You can’t perform that action at this time.
-  </div>
-
-  <div class="js-stale-session-flash flash flash-warn flash-banner" hidden
-    >
-    <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-alert">
-    <path fill-rule="evenodd" d="M8.22 1.754a.25.25 0 00-.44 0L1.698 13.132a.25.25 0 00.22.368h12.164a.25.25 0 00.22-.368L8.22 1.754zm-1.763-.707c.659-1.234 2.427-1.234 3.086 0l6.082 11.378A1.75 1.75 0 0114.082 15H1.918a1.75 1.75 0 01-1.543-2.575L6.457 1.047zM9 11a1 1 0 11-2 0 1 1 0 012 0zm-.25-5.25a.75.75 0 00-1.5 0v2.5a.75.75 0 001.5 0v-2.5z"></path>
-</svg>
-    <span class="js-stale-session-flash-signed-in" hidden>You signed in with another tab or window. <a href="">Reload</a> to refresh your session.</span>
-    <span class="js-stale-session-flash-signed-out" hidden>You signed out in another tab or window. <a href="">Reload</a> to refresh your session.</span>
-  </div>
-    <template id="site-details-dialog">
-  <details class="details-reset details-overlay details-overlay-dark lh-default color-fg-default hx_rsm" open>
-    <summary role="button" aria-label="Close dialog"></summary>
-    <details-dialog class="Box Box--overlay d-flex flex-column anim-fade-in fast hx_rsm-dialog hx_rsm-modal">
-      <button class="Box-btn-octicon m-0 btn-octicon position-absolute right-0 top-0" type="button" aria-label="Close dialog" data-close-dialog>
-        <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-x">
-    <path fill-rule="evenodd" d="M3.72 3.72a.75.75 0 011.06 0L8 6.94l3.22-3.22a.75.75 0 111.06 1.06L9.06 8l3.22 3.22a.75.75 0 11-1.06 1.06L8 9.06l-3.22 3.22a.75.75 0 01-1.06-1.06L6.94 8 3.72 4.78a.75.75 0 010-1.06z"></path>
-</svg>
-      </button>
-      <div class="octocat-spinner my-6 js-details-dialog-spinner"></div>
-    </details-dialog>
-  </details>
-</template>
-
-    <div class="Popover js-hovercard-content position-absolute" style="display: none; outline: none;" tabindex="0">
-  <div class="Popover-message Popover-message--bottom-left Popover-message--large Box color-shadow-large" style="width:360px;">
-  </div>
-</div>
-
-    <template id="snippet-clipboard-copy-button">
-  <div class="zeroclipboard-container position-absolute right-0 top-0">
-    <clipboard-copy aria-label="Copy" class="ClipboardButton btn js-clipboard-copy m-2 p-0 tooltipped-no-delay" data-copy-feedback="Copied!" data-tooltip-direction="w">
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-copy js-clipboard-copy-icon m-2">
-    <path fill-rule="evenodd" d="M0 6.75C0 5.784.784 5 1.75 5h1.5a.75.75 0 010 1.5h-1.5a.25.25 0 00-.25.25v7.5c0 .138.112.25.25.25h7.5a.25.25 0 00.25-.25v-1.5a.75.75 0 011.5 0v1.5A1.75 1.75 0 019.25 16h-7.5A1.75 1.75 0 010 14.25v-7.5z"></path><path fill-rule="evenodd" d="M5 1.75C5 .784 5.784 0 6.75 0h7.5C15.216 0 16 .784 16 1.75v7.5A1.75 1.75 0 0114.25 11h-7.5A1.75 1.75 0 015 9.25v-7.5zm1.75-.25a.25.25 0 00-.25.25v7.5c0 .138.112.25.25.25h7.5a.25.25 0 00.25-.25v-7.5a.25.25 0 00-.25-.25h-7.5z"></path>
-</svg>
-      <svg aria-hidden="true" height="16" viewBox="0 0 16 16" version="1.1" width="16" data-view-component="true" class="octicon octicon-check js-clipboard-check-icon color-fg-success d-none m-2">
-    <path fill-rule="evenodd" d="M13.78 4.22a.75.75 0 010 1.06l-7.25 7.25a.75.75 0 01-1.06 0L2.22 9.28a.75.75 0 011.06-1.06L6 10.94l6.72-6.72a.75.75 0 011.06 0z"></path>
-</svg>
-    </clipboard-copy>
-  </div>
-</template>
-
-
-
-
-  </body>
-</html>
+从端到端结果中可以看到对预训练模型不做修改，只根据场景下的具体情况进行后处理的修改就能大幅提升端到端指标到91.66%，在CCPD数据集上进行 fine-tune 后指标进一步提升到92.5%, 在经过量化训练之后，指标轻微下降到92.02%但模型大小降低54%。

From 9b1e9ae6dd22504251dbb7190d39d5b7823355de Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Tue, 21 Jun 2022 08:21:21 +0000
Subject: [PATCH 035/134] add table_ops.py

---
 ppocr/data/imaug/__init__.py                         | 2 +-
 ppocr/data/imaug/{gen_table_mask.py => table_ops.py} | 0
 2 files changed, 1 insertion(+), 1 deletion(-)
 rename ppocr/data/imaug/{gen_table_mask.py => table_ops.py} (100%)

diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index 548832fb0d..750a754157 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -34,7 +34,7 @@ from .label_ops import *
 from .east_process import *
 from .sast_process import *
 from .pg_process import *
-from .gen_table_mask import *
+from .table_ops import *
 
 from .vqa import *
 
diff --git a/ppocr/data/imaug/gen_table_mask.py b/ppocr/data/imaug/table_ops.py
similarity index 100%
rename from ppocr/data/imaug/gen_table_mask.py
rename to ppocr/data/imaug/table_ops.py

From 4957a923fb8ccfbfdaf2255572953ed28062e960 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 21 Jun 2022 17:55:49 +0800
Subject: [PATCH 036/134] fix training

---
 ppocr/data/imaug/label_ops.py | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 8b017b3219..44c871d9ef 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -70,14 +70,22 @@ class DetLabelEncode(object):
         return data
 
     def order_points_clockwise(self, pts):
-        rect = np.zeros((4, 2), dtype="float32")
-        s = pts.sum(axis=1)
-        rect[0] = pts[np.argmin(s)]
-        rect[2] = pts[np.argmax(s)]
-        diff = np.diff(pts, axis=1)
-        rect[1] = pts[np.argmin(diff)]
-        rect[3] = pts[np.argmax(diff)]
-        return rect
+        """
+        refer to :https://github.com/PyImageSearch/imutils/blob/9f740a53bcc2ed7eba2558afed8b4c17fd8a1d4c/imutils/perspective.py#L9
+        """
+        # sort the points based on their x-coordinates
+        xSorted = pts[np.argsort(pts[:, 0]), :]
+
+        leftMost = xSorted[:2, :]
+        rightMost = xSorted[2:, :]
+
+        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
+        (tl, bl) = leftMost
+
+        D = dist.cdist(tl[np.newaxis], rightMost, "euclidean")[0]
+        (br, tr) = rightMost[np.argsort(D)[::-1], :]
+
+        return np.array([tl, tr, br, bl], dtype="float32")
 
     def expand_points_num(self, boxes):
         max_points_num = 0
@@ -443,7 +451,9 @@ class KieLabelEncode(object):
             elif 'key_cls' in ann.keys():
                 labels.append(ann['key_cls'])
             else:
-                raise ValueError("Cannot found 'key_cls' in ann.keys(), please check your training annotation.")
+                raise ValueError(
+                    "Cannot found 'key_cls' in ann.keys(), please check your training annotation."
+                )
             edges.append(ann.get('edge', 0))
         ann_infos = dict(
             image=data['image'],

From 114d3399b89e3ed762b79286de508420c5ec8a65 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Tue, 21 Jun 2022 17:59:44 +0800
Subject: [PATCH 037/134] fix training

---
 ppocr/data/imaug/label_ops.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 44c871d9ef..36017a823e 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -23,7 +23,7 @@ import string
 from shapely.geometry import LineString, Point, Polygon
 import json
 import copy
-
+from scipy.spatial import distance as dist
 from ppocr.utils.logging import get_logger
 
 

From 3ef6f1bff98c5f0038db39db0fbde8e6da7e85af Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Tue, 21 Jun 2022 12:10:09 +0000
Subject: [PATCH 038/134] add table fleet

---
 ...leet_normal_infer_python_linux_gpu_cpu.txt | 53 +++++++++++++++++++
 1 file changed, 53 insertions(+)
 create mode 100644 test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt

diff --git a/test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000000..41d236c376
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]

From 32c65b0dfd9114e055173d0fd23db9609a226c5e Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Wed, 22 Jun 2022 12:26:12 +0800
Subject: [PATCH 039/134] opt points order

---
 tools/infer/predict_det.py | 26 +++++++++-----------------
 1 file changed, 9 insertions(+), 17 deletions(-)

diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py
index 73387aef59..b9fab7dd0d 100755
--- a/tools/infer/predict_det.py
+++ b/tools/infer/predict_det.py
@@ -24,7 +24,6 @@ import cv2
 import numpy as np
 import time
 import sys
-from scipy.spatial import distance as dist
 
 import tools.infer.utility as utility
 from ppocr.utils.logging import get_logger
@@ -151,22 +150,15 @@ class TextDetector(object):
                 logger=logger)
 
     def order_points_clockwise(self, pts):
-        """
-        refer to :https://github.com/PyImageSearch/imutils/blob/9f740a53bcc2ed7eba2558afed8b4c17fd8a1d4c/imutils/perspective.py#L9
-        """
-        # sort the points based on their x-coordinates
-        xSorted = pts[np.argsort(pts[:, 0]), :]
-
-        leftMost = xSorted[:2, :]
-        rightMost = xSorted[2:, :]
-
-        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
-        (tl, bl) = leftMost
-
-        D = dist.cdist(tl[np.newaxis], rightMost, "euclidean")[0]
-        (br, tr) = rightMost[np.argsort(D)[::-1], :]
-
-        return np.array([tl, tr, br, bl], dtype="float32")
+        rect = np.zeros((4, 2), dtype="int32")
+        s = pts.sum(axis=1)
+        rect[0] = pts[np.argmin(s)]
+        rect[2] = pts[np.argmax(s)]
+        tmp = np.delete(pts, (np.argmin(s), np.argmax(s)), axis=0)
+        diff = np.diff(np.array(tmp), axis=1)
+        rect[1] = tmp[np.argmin(diff)]
+        rect[3] = tmp[np.argmax(diff)]
+        return rect
 
     def clip_det_res(self, points, img_height, img_width):
         for pno in range(points.shape[0]):

From 715fc2377e9ad2dab94873635a4346815673fb5e Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Wed, 22 Jun 2022 12:29:06 +0800
Subject: [PATCH 040/134] fix

---
 ppocr/data/imaug/label_ops.py | 26 +++++++++-----------------
 tools/infer/predict_det.py    |  2 +-
 2 files changed, 10 insertions(+), 18 deletions(-)

diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 36017a823e..39c2cbbac8 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -23,7 +23,6 @@ import string
 from shapely.geometry import LineString, Point, Polygon
 import json
 import copy
-from scipy.spatial import distance as dist
 from ppocr.utils.logging import get_logger
 
 
@@ -70,22 +69,15 @@ class DetLabelEncode(object):
         return data
 
     def order_points_clockwise(self, pts):
-        """
-        refer to :https://github.com/PyImageSearch/imutils/blob/9f740a53bcc2ed7eba2558afed8b4c17fd8a1d4c/imutils/perspective.py#L9
-        """
-        # sort the points based on their x-coordinates
-        xSorted = pts[np.argsort(pts[:, 0]), :]
-
-        leftMost = xSorted[:2, :]
-        rightMost = xSorted[2:, :]
-
-        leftMost = leftMost[np.argsort(leftMost[:, 1]), :]
-        (tl, bl) = leftMost
-
-        D = dist.cdist(tl[np.newaxis], rightMost, "euclidean")[0]
-        (br, tr) = rightMost[np.argsort(D)[::-1], :]
-
-        return np.array([tl, tr, br, bl], dtype="float32")
+        rect = np.zeros((4, 2), dtype="float32")
+        s = pts.sum(axis=1)
+        rect[0] = pts[np.argmin(s)]
+        rect[2] = pts[np.argmax(s)]
+        tmp = np.delete(pts, (np.argmin(s), np.argmax(s)), axis=0)
+        diff = np.diff(np.array(tmp), axis=1)
+        rect[1] = tmp[np.argmin(diff)]
+        rect[3] = tmp[np.argmax(diff)]
+        return rect
 
     def expand_points_num(self, boxes):
         max_points_num = 0
diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py
index b9fab7dd0d..7b6bebf1fb 100755
--- a/tools/infer/predict_det.py
+++ b/tools/infer/predict_det.py
@@ -150,7 +150,7 @@ class TextDetector(object):
                 logger=logger)
 
     def order_points_clockwise(self, pts):
-        rect = np.zeros((4, 2), dtype="int32")
+        rect = np.zeros((4, 2), dtype="float32")
         s = pts.sum(axis=1)
         rect[0] = pts[np.argmin(s)]
         rect[2] = pts[np.argmax(s)]

From f140143f318b8903cc28e98ddd91388f18921885 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Wed, 22 Jun 2022 07:03:44 +0000
Subject: [PATCH 041/134] update log

---
 .../train_pact_infer_python.txt               |  0
 ..._normal_amp_infer_python_linux_gpu_cpu.txt | 53 ----------------
 .../train_pact_infer_python.txt               |  0
 ..._normal_amp_infer_python_linux_gpu_cpu.txt | 53 ----------------
 .../train_pact_infer_python.txt               |  0
 ..._normal_amp_infer_python_linux_gpu_cpu.txt | 53 ----------------
 .../train_pact_infer_python.txt               |  0
 .../train_pact_infer_python.txt               |  0
 ..._normal_amp_infer_python_linux_gpu_cpu.txt | 53 ----------------
 .../train_pact_infer_python.txt               |  0
 ..._normal_amp_infer_python_linux_gpu_cpu.txt | 53 ----------------
 .../train_pact_infer_python.txt               |  0
 ..._normal_amp_infer_python_linux_gpu_cpu.txt | 53 ----------------
 test_tipc/test_paddle2onnx.sh                 | 12 ++--
 test_tipc/test_ptq_inference_python.sh        |  3 +-
 test_tipc/test_serving_infer_cpp.sh           | 17 +++--
 test_tipc/test_serving_infer_python.sh        | 30 +++++----
 test_tipc/test_train_inference_python.sh      | 62 +++----------------
 18 files changed, 49 insertions(+), 393 deletions(-)
 rename test_tipc/configs/{ch_PP-OCRv2_det_PACT => ch_PP-OCRv2_det}/train_pact_infer_python.txt (100%)
 delete mode 100644 test_tipc/configs/ch_PP-OCRv2_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
 rename test_tipc/configs/{ch_PP-OCRv2_rec_PACT => ch_PP-OCRv2_rec}/train_pact_infer_python.txt (100%)
 delete mode 100644 test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
 rename test_tipc/configs/{ch_PP-OCRv3_det_PACT => ch_PP-OCRv3_det}/train_pact_infer_python.txt (100%)
 delete mode 100644 test_tipc/configs/ch_PP-OCRv3_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
 rename test_tipc/configs/{ch_PP-OCRv3_rec_PACT => ch_PP-OCRv3_rec}/train_pact_infer_python.txt (100%)
 rename test_tipc/configs/{ch_ppocr_mobile_v2.0_det_PACT => ch_ppocr_mobile_v2.0_det}/train_pact_infer_python.txt (100%)
 delete mode 100644 test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
 rename test_tipc/configs/{ch_ppocr_mobile_v2.0_rec_PACT => ch_ppocr_mobile_v2.0_rec}/train_pact_infer_python.txt (100%)
 delete mode 100644 test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
 rename test_tipc/configs/{en_table_structure_PACT => en_table_structure}/train_pact_infer_python.txt (100%)
 delete mode 100644 test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt

diff --git a/test_tipc/configs/ch_PP-OCRv2_det_PACT/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv2_det_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
deleted file mode 100644
index 3afc0acb79..0000000000
--- a/test_tipc/configs/ch_PP-OCRv2_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
+++ /dev/null
@@ -1,53 +0,0 @@
-===========================train_params===========================
-model_name:ch_PP-OCRv2_det_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
-train_model_name:latest
-train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
-fpgm_export: 
-distill_export:null
-export1:null
-export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv2_det_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_det.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv2_rec/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
deleted file mode 100644
index e0f86d9020..0000000000
--- a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
+++ /dev/null
@@ -1,53 +0,0 @@
-===========================train_params===========================
-model_name:ch_PP-OCRv2_rec_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
-Global.pretrained_model:pretrain_models/ch_PP-OCRv2_rec_train/best_accuracy
-train_model_name:latest
-train_infer_img_dir:./inference/rec_inference
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
-fpgm_export: null
-distill_export:null
-export1:null
-export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv2_rec_slim_quant_infer
-infer_export:null
-infer_quant:True
-inference:tools/infer/predict_rec.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1|6
---use_tensorrt:False
---precision:fp32
---rec_model_dir:
---image_dir:./inference/rec_inference
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_PACT/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv3_det_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv3_det/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
deleted file mode 100644
index 252378e4dc..0000000000
--- a/test_tipc/configs/ch_PP-OCRv3_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
+++ /dev/null
@@ -1,53 +0,0 @@
-===========================train_params===========================
-model_name:ch_PP-OCRv3_det_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
-train_model_name:latest
-train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o 
-fpgm_export: 
-distill_export:null
-export1:null
-export2:null
-inference_dir:Student
-infer_model:./inference/ch_PP-OCRv3_det_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_det.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv3_rec_PACT/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_rec/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_PP-OCRv3_rec_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/ch_PP-OCRv3_rec/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_pact_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
deleted file mode 100644
index 3a5d8faf3f..0000000000
--- a/test_tipc/configs/ch_ppocr_mobile_v2.0_det_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
+++ /dev/null
@@ -1,53 +0,0 @@
-===========================train_params===========================
-model_name:ch_ppocr_mobile_v2.0_det_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
-train_model_name:latest
-train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
-fpgm_export:null
-distill_export:null
-export1:null
-export2:null
-inference_dir:null
-train_model:./inference/ch_ppocr_mobile_v2.0_det_prune_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_det.py
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---det_model_dir:
---image_dir:./inference/ch_det_data_50/all-sum-510/
-null:null
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_pact_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/ch_ppocr_mobile_v2.0_rec/train_pact_infer_python.txt
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
deleted file mode 100644
index a1a2a8e63d..0000000000
--- a/test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
+++ /dev/null
@@ -1,53 +0,0 @@
-===========================train_params===========================
-model_name:ch_ppocr_mobile_v2.0_rec_PACT
-python:python3.7
-gpu_list:0
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
-Global.checkpoints:null
-train_model_name:latest
-train_infer_img_dir:./train_data/ic15_data/test/word_1.png
-null:null
-##
-trainer:pact_train
-norm_train:null
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_ppocr_mobile_v2.0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o 
-fpgm_export:null
-distill_export:null
-export1:null
-export2:null
-inference_dir:null
-infer_model:./inference/ch_ppocr_mobile_v2.0_rec_slim_infer/
-infer_export:null
-infer_quant:False
-inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_image_shape="3,32,100"
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1|6
---use_tensorrt:False
---precision:fp32
---rec_model_dir:
---image_dir:./inference/rec_inference
---save_log_path:./test/output/
---benchmark:True
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/en_table_structure_PACT/train_pact_infer_python.txt b/test_tipc/configs/en_table_structure/train_pact_infer_python.txt
similarity index 100%
rename from test_tipc/configs/en_table_structure_PACT/train_pact_infer_python.txt
rename to test_tipc/configs/en_table_structure/train_pact_infer_python.txt
diff --git a/test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
deleted file mode 100644
index d7c847fcdc..0000000000
--- a/test_tipc/configs/en_table_structure_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
+++ /dev/null
@@ -1,53 +0,0 @@
-===========================train_params===========================
-model_name:en_table_structure_PACT
-python:python3.7
-gpu_list:0|0,1
-Global.use_gpu:True|True
-Global.auto_cast:amp
-Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
-Global.save_model_dir:./output/
-Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
-Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
-train_model_name:latest
-train_infer_img_dir:./ppstructure/docs/table/table.jpg
-null:null
-##
-trainer:pact_train
-norm_train:null 
-pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
-fpgm_train:null
-distill_train:null
-null:null
-null:null
-##
-===========================eval_params=========================== 
-eval:null
-null:null
-##
-===========================infer_params===========================
-Global.save_inference_dir:./output/
-Global.checkpoints:
-norm_export:null 
-quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/en_table_structure/table_mv3.yml -o 
-fpgm_export: 
-distill_export:null
-export1:null
-export2:null
-##
-infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
-infer_export:null
-infer_quant:True
-inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
---use_gpu:True|False
---enable_mkldnn:False
---cpu_threads:6
---rec_batch_num:1
---use_tensorrt:False
---precision:fp32
---table_model_dir:
---image_dir:./ppstructure/docs/table/table.jpg
-null:null
---benchmark:False
-null:null
-===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/test_paddle2onnx.sh b/test_tipc/test_paddle2onnx.sh
index 4ad2035f48..356bc98041 100644
--- a/test_tipc/test_paddle2onnx.sh
+++ b/test_tipc/test_paddle2onnx.sh
@@ -62,7 +62,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
@@ -73,7 +74,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${rec_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_rec_log="${LOG_PATH}/trans_model_rec.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
@@ -85,7 +87,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"     
@@ -97,7 +100,8 @@ function func_paddle2onnx(){
         set_save_model=$(func_set_params "--save_file" "${rec_save_file_value}")
         set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
         set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
-        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker}"
+        trans_rec_log="${LOG_PATH}/trans_model_rec.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
         last_status=${PIPESTATUS[0]}
         status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
diff --git a/test_tipc/test_ptq_inference_python.sh b/test_tipc/test_ptq_inference_python.sh
index 2454c0b50a..c1aa3daa6c 100644
--- a/test_tipc/test_ptq_inference_python.sh
+++ b/test_tipc/test_ptq_inference_python.sh
@@ -139,7 +139,8 @@ if [ ${MODE} = "whole_infer" ]; then
             save_infer_dir="${infer_model}_klquant"
             set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
             set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
-            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key}"
+            export_log_path="${LOG_PATH}/_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
             echo ${infer_run_exports[Count]} 
             echo $export_cmd
             eval $export_cmd
diff --git a/test_tipc/test_serving_infer_cpp.sh b/test_tipc/test_serving_infer_cpp.sh
index f9f7ac1aa5..4088c66f57 100644
--- a/test_tipc/test_serving_infer_cpp.sh
+++ b/test_tipc/test_serving_infer_cpp.sh
@@ -70,7 +70,8 @@ function func_serving(){
     set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
     set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
     python_list=(${python_list})
-    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+    trans_det_log="${LOG_PATH}/cpp_trans_model_det.log"
+    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
     eval $trans_model_cmd
     cp "deploy/pdserving/serving_client_conf.prototxt" ${det_serving_client_value}
     # trans rec
@@ -78,7 +79,8 @@ function func_serving(){
     set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
     set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
     python_list=(${python_list})
-    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+    trans_rec_log="${LOG_PATH}/cpp_trans_model_rec.log"
+    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
     eval $trans_model_cmd
     last_status=${PIPESTATUS[0]}
     status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}"
@@ -88,22 +90,25 @@ function func_serving(){
     # cpp serving
     for gpu_id in ${gpu_value[*]}; do
         if [ ${gpu_id} = "null" ]; then
-            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > serving_log_cpu.log &"
+            server_log_path="${LOG_PATH}/cpp_server_cpu.log"
+            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > ${server_log_path} 2>&1 "
             eval $web_service_cpp_cmd
             last_status=${PIPESTATUS[0]}
             status_check $last_status "${web_service_cpp_cmd}" "${status_log}" "${model_name}"
             sleep 5s
-            _save_log_path="${LOG_PATH}/server_infer_cpp_cpu.log"
+            _save_log_path="${LOG_PATH}/cpp_client_cpu.log"
             cpp_client_cmd="${python_list[0]} ${cpp_client_py} ${det_client_value} ${rec_client_value} > ${_save_log_path} 2>&1"
             eval $cpp_client_cmd
             last_status=${PIPESTATUS[0]}
+            eval "cat ${_save_log_path}"
             status_check $last_status "${cpp_client_cmd}" "${status_log}" "${model_name}"
             ps ux | grep -i ${port_value} | awk '{print $2}' | xargs kill -s 9
         else
-            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > serving_log_gpu.log &"
+            server_log_path="${LOG_PATH}/cpp_server_gpu.log"
+            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > ${server_log_path} 2>&1 "
             eval $web_service_cpp_cmd
             sleep 5s
-            _save_log_path="${LOG_PATH}/server_infer_cpp_gpu.log"
+            _save_log_path="${LOG_PATH}/cpp_client_gpu.log"
             cpp_client_cmd="${python_list[0]} ${cpp_client_py} ${det_client_value} ${rec_client_value} > ${_save_log_path} 2>&1"
             eval $cpp_client_cmd
             last_status=${PIPESTATUS[0]}
diff --git a/test_tipc/test_serving_infer_python.sh b/test_tipc/test_serving_infer_python.sh
index c76d6f5d19..57dab6aeb5 100644
--- a/test_tipc/test_serving_infer_python.sh
+++ b/test_tipc/test_serving_infer_python.sh
@@ -77,14 +77,16 @@ function func_serving(){
         set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_det_log="${LOG_PATH}/python_trans_model_det.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
         # trans rec
         set_dirname=$(func_set_params "--dirname" "${rec_infer_model_dir_value}")
         set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_rec_log="${LOG_PATH}/python_trans_model_rec.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
     elif [[ ${model_name} =~ "det" ]]; then
         # trans det
@@ -92,7 +94,8 @@ function func_serving(){
         set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_det_log="${LOG_PATH}/python_trans_model_det.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
         eval $trans_model_cmd
     elif [[ ${model_name} =~ "rec" ]]; then
         # trans rec
@@ -100,7 +103,8 @@ function func_serving(){
         set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
         set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
         python_list=(${python_list})
-        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client}"
+        trans_rec_log="${LOG_PATH}/python_trans_model_rec.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
         eval $trans_model_cmd
     fi
     set_image_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
@@ -115,29 +119,30 @@ function func_serving(){
             for use_mkldnn in ${web_use_mkldnn_list[*]}; do
                 for threads in ${web_cpu_threads_list[*]}; do
                     set_cpu_threads=$(func_set_params "${web_cpu_threads_key}" "${threads}")
+                    server_log_path="${LOG_PATH}/python_server_cpu_usemkldnn_${use_mkldnn}_threads_${threads}.log"
                     if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2.0" ] || [ ${model_name} = "ch_ppocr_server_v2.0" ]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "det" ]]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "rec" ]]; then
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     fi
                     sleep 2s
                     for pipeline in ${pipeline_py[*]}; do
-                        _save_log_path="${LOG_PATH}/server_infer_cpu_${pipeline%_client*}_usemkldnn_${use_mkldnn}_threads_${threads}_batchsize_1.log"
+                        _save_log_path="${LOG_PATH}/python_client_cpu_${pipeline%_client*}_usemkldnn_${use_mkldnn}_threads_${threads}_batchsize_1.log"
                         pipeline_cmd="${python} ${pipeline} ${set_image_dir} > ${_save_log_path} 2>&1 "
                         eval $pipeline_cmd
                         last_status=${PIPESTATUS[0]}
@@ -151,6 +156,7 @@ function func_serving(){
         elif [ ${use_gpu} = "gpu" ]; then
             for use_trt in ${web_use_trt_list[*]}; do
                 for precision in ${web_precision_list[*]}; do
+                    server_log_path="${LOG_PATH}/python_server_gpu_usetrt_${use_trt}_precision_${precision}.log"
                     if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
                         continue
                     fi
@@ -168,26 +174,26 @@ function func_serving(){
                     if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2.0" ] || [ ${model_name} = "ch_ppocr_server_v2.0" ]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "det" ]]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "rec" ]]; then
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} &"
+                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} > ${server_log_path} 2>&1 "
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     fi
                     sleep 2s
                     for pipeline in ${pipeline_py[*]}; do
-                        _save_log_path="${LOG_PATH}/server_infer_gpu_${pipeline%_client*}_usetrt_${use_trt}_precision_${precision}_batchsize_1.log"
+                        _save_log_path="${LOG_PATH}/python_client_gpu_${pipeline%_client*}_usetrt_${use_trt}_precision_${precision}_batchsize_1.log"
                         pipeline_cmd="${python} ${pipeline} ${set_image_dir}> ${_save_log_path} 2>&1"
                         eval $pipeline_cmd
                         last_status=${PIPESTATUS[0]}
diff --git a/test_tipc/test_train_inference_python.sh b/test_tipc/test_train_inference_python.sh
index 62a56a32ce..fa68cb2632 100644
--- a/test_tipc/test_train_inference_python.sh
+++ b/test_tipc/test_train_inference_python.sh
@@ -2,7 +2,7 @@
 source test_tipc/common_func.sh
 
 FILENAME=$1
-# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer', 'whole_infer', 'klquant_whole_infer']
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer', 'whole_infer']
 MODE=$2
 
 dataline=$(awk 'NR==1, NR==51{print}'  $FILENAME)
@@ -88,43 +88,6 @@ benchmark_value=$(func_parser_value "${lines[49]}")
 infer_key1=$(func_parser_key "${lines[50]}")
 infer_value1=$(func_parser_value "${lines[50]}")
 
-# parser klquant_infer
-if [ ${MODE} = "klquant_whole_infer" ]; then
-    dataline=$(awk 'NR==1, NR==17{print}'  $FILENAME)
-    lines=(${dataline})
-    model_name=$(func_parser_value "${lines[1]}")
-    python=$(func_parser_value "${lines[2]}")
-    export_weight=$(func_parser_key "${lines[3]}")
-    save_infer_key=$(func_parser_key "${lines[4]}")
-    # parser inference model 
-    infer_model_dir_list=$(func_parser_value "${lines[5]}")
-    infer_export_list=$(func_parser_value "${lines[6]}")
-    infer_is_quant=$(func_parser_value "${lines[7]}")
-    # parser inference 
-    inference_py=$(func_parser_value "${lines[8]}")
-    use_gpu_key=$(func_parser_key "${lines[9]}")
-    use_gpu_list=$(func_parser_value "${lines[9]}")
-    use_mkldnn_key=$(func_parser_key "${lines[10]}")
-    use_mkldnn_list=$(func_parser_value "${lines[10]}")
-    cpu_threads_key=$(func_parser_key "${lines[11]}")
-    cpu_threads_list=$(func_parser_value "${lines[11]}")
-    batch_size_key=$(func_parser_key "${lines[12]}")
-    batch_size_list=$(func_parser_value "${lines[12]}")
-    use_trt_key=$(func_parser_key "${lines[13]}")
-    use_trt_list=$(func_parser_value "${lines[13]}")
-    precision_key=$(func_parser_key "${lines[14]}")
-    precision_list=$(func_parser_value "${lines[14]}")
-    infer_model_key=$(func_parser_key "${lines[15]}")
-    image_dir_key=$(func_parser_key "${lines[16]}")
-    infer_img_dir=$(func_parser_value "${lines[16]}")
-    save_log_key=$(func_parser_key "${lines[17]}")
-    save_log_value=$(func_parser_value "${lines[17]}")
-    benchmark_key=$(func_parser_key "${lines[18]}")
-    benchmark_value=$(func_parser_value "${lines[18]}")
-    infer_key1=$(func_parser_key "${lines[19]}")
-    infer_value1=$(func_parser_value "${lines[19]}")
-fi
-
 LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
 mkdir -p ${LOG_PATH}
 status_log="${LOG_PATH}/results_python.log"
@@ -211,7 +174,7 @@ function func_inference(){
     done
 }
 
-if [ ${MODE} = "whole_infer" ] || [ ${MODE} = "klquant_whole_infer" ]; then
+if [ ${MODE} = "whole_infer" ]; then
     GPUID=$3
     if [ ${#GPUID} -le 0 ];then
         env=" "
@@ -226,16 +189,12 @@ if [ ${MODE} = "whole_infer" ] || [ ${MODE} = "klquant_whole_infer" ]; then
     infer_quant_flag=(${infer_is_quant})
     for infer_model in ${infer_model_dir_list[*]}; do
         # run export
-        if [ ${infer_run_exports[Count]} != "null" ];then
-            if [ ${MODE} = "klquant_whole_infer" ]; then
-                save_infer_dir="${infer_model}_klquant"
-            fi
-            if [ ${MODE} = "whole_infer" ]; then
-                save_infer_dir="${infer_model}"
-            fi
+        if [ ${infer_run_exports[Count]} != "null" ];then 
+            save_infer_dir="${infer_model}"
             set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
             set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
-            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key}"
+            export_log_path="${LOG_PATH}/_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
             echo ${infer_run_exports[Count]} 
             echo $export_cmd
             eval $export_cmd
@@ -246,9 +205,6 @@ if [ ${MODE} = "whole_infer" ] || [ ${MODE} = "klquant_whole_infer" ]; then
         fi
         #run inference
         is_quant=${infer_quant_flag[Count]}
-        if [ ${MODE} = "klquant_whole_infer" ]; then
-            is_quant="True"
-        fi
         func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant}
         Count=$(($Count + 1))
     done
@@ -347,7 +303,8 @@ else
                 if [ ${eval_py} != "null" ]; then
                     eval ${env}
                     set_eval_params1=$(func_set_params "${eval_key1}" "${eval_value1}")
-                    eval_cmd="${python} ${eval_py} ${set_eval_pretrain} ${set_use_gpu} ${set_eval_params1}" 
+                    eval_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_eval.log"
+                    eval_cmd="${python} ${eval_py} ${set_eval_pretrain} ${set_use_gpu} ${set_eval_params1} > ${eval_log_path} 2>&1 " 
                     eval $eval_cmd
                     status_check $? "${eval_cmd}" "${status_log}" "${model_name}"
                 fi
@@ -355,9 +312,10 @@ else
                 if [ ${run_export} != "null" ]; then 
                     # run export model
                     save_infer_path="${save_log}"
+                    export_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_export.log"
                     set_export_weight=$(func_set_params "${export_weight}" "${save_log}/${train_model_name}")
                     set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_path}")
-                    export_cmd="${python} ${run_export} ${set_export_weight} ${set_save_infer_key}"
+                    export_cmd="${python} ${run_export} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
                     eval $export_cmd
                     status_check $? "${export_cmd}" "${status_log}" "${model_name}"
 

From ea1cc9225794efff63715ea6f7126add0159cfb8 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Wed, 22 Jun 2022 09:38:22 +0000
Subject: [PATCH 042/134] add ch doc

---
 deploy/avh/README_ch.md | 93 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 93 insertions(+)
 create mode 100644 deploy/avh/README_ch.md

diff --git a/deploy/avh/README_ch.md b/deploy/avh/README_ch.md
new file mode 100644
index 0000000000..94db92e72c
--- /dev/null
+++ b/deploy/avh/README_ch.md
@@ -0,0 +1,93 @@
+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+
+通过TVM在 Arm(R) Cortex(R)-M55 CPU 上运行 PaddleOCR文 本能识别模型
+===============================================================
+
+此文件夹包含如何使用 TVM 在 Cortex(R)-M55 CPU 上运行 PaddleOCR 模型的示例。
+
+依赖
+-------------
+本demo运行在TVM提供的docker环境上，在该环境中已经安装好的必须的软件
+
+
+在非docker环境中，需要手动安装如下依赖项: 
+
+- 软件可通过[安装脚本](https://github.com/apache/tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh)一键安装
+  - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
+  - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
+  - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
+  - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)
+  - [CMSIS](https://github.com/ARM-software/CMSIS_5)
+- python 依赖
+  ```bash
+  pip install -r ./requirements.txt
+  ```
+- TVM
+  - 从源码安装([Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    从源码安装时，需要设置如下字段
+      - set(USE_CMSISNN ON)
+      - set(USE_MICRO ON)
+      - set(USE_LLVM ON)
+  - 从TLCPack 安装([TLCPack](https://tlcpack.ai/))
+
+安装完成后需要更新环境变量，以软件安装地址为`/opt/arm`为例：
+```bash
+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
+```
+
+运行demo
+----------------------------
+使用如下命令可以一键运行demo
+
+```bash
+./run_demo.sh
+```
+
+如果 Ethos(TM)-U 平台或 CMSIS 没有安装在 `/opt/arm/ethosu` 中，可通过参数进行设置，例如：
+
+```bash
+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \
+--ethosu_platform_path /home/tvm-user/ethosu/core_platform
+```
+
+`./run_demo.sh`脚本会执行如下步骤:
+- 下载 PaddleOCR 文字识别模型
+- 使用tvm将PaddleOCR 文字识别模型编译为 Cortex(R)-M55 CPU 和 CMSIS-NN 后端的可执行文件
+- 创建一个包含输入图像数据的头文件`inputs.c`
+- 创建一个包含输出tensor大小的头文件`outputs.c`
+- 编译可执行程序
+- 运行程序
+- 输出图片上的文字和置信度
+
+使用自己的图片
+--------------------
+替换 `run_demo.sh ` 中140行处的图片地址即可
+
+使用自己的模型
+--------------------
+替换 `run_demo.sh ` 中130行处的模型地址即可
+
+模型描述
+-----------------
+
+在这个demo中，我们使用的模型是基于[PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md)的英文识别模型。 PP-OCRv3是[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)发布的PP-OCR系列模型的第三个版本。 该系列模型具有以下特点：
+   - 超轻量级OCR系统：检测（3.6M）+方向分类器（1.4M）+识别（12M）=17.0M。
+   - 支持80多种多语言识别模型，包括英文、中文、法文、德文、阿拉伯文、韩文、日文等。 
+   - 支持竖排文本识别，长文本识别。
+
+PP-OCRv3 中的文本识别模型支持 80 多种语言。 在模型开发过程中，由于Arm(R) Cortex(R)-M55 CPU不支持rnn算子，我们在PP-OCRv3文本识别模型的基础上删除了不支持的算子，得到当前模型。
\ No newline at end of file

From 6778ab933ae29bdef0d6f62a9b023c47f08ebab3 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Wed, 22 Jun 2022 09:42:24 +0000
Subject: [PATCH 043/134] update doc

---
 deploy/avh/README.md    | 2 +-
 deploy/avh/README_ch.md | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/deploy/avh/README.md b/deploy/avh/README.md
index ecd635cc42..922fb4da92 100644
--- a/deploy/avh/README.md
+++ b/deploy/avh/README.md
@@ -28,7 +28,7 @@ software will already be installed.
 
 If the demo is not run in the ci_cpu Docker container, then you will need the following:
 - Software required to build and run the demo (These can all be installed by running 
-  tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh.)
+  https://github.com/apache/tvm/blob/main/docker/install/ubuntu_install_ethosu_driver_stack.sh .)
   - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
   - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
   - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
diff --git a/deploy/avh/README_ch.md b/deploy/avh/README_ch.md
index 94db92e72c..7e6272ae12 100644
--- a/deploy/avh/README_ch.md
+++ b/deploy/avh/README_ch.md
@@ -27,7 +27,7 @@
 
 在非docker环境中，需要手动安装如下依赖项: 
 
-- 软件可通过[安装脚本](https://github.com/apache/tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh)一键安装
+- 软件可通过[安装脚本](https://github.com/apache/tvm/blob/main/docker/install/ubuntu_install_ethosu_driver_stack.sh)一键安装
   - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
   - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
   - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)

From c503dc2f9352272615dc3cc11737b833036c6ccc Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Wed, 22 Jun 2022 14:19:55 +0000
Subject: [PATCH 044/134] [New Rec] add vitstr and ABINet

---
 configs/rec/rec_mtb_nrtr.yml                  |   6 +-
 configs/rec/rec_r45_abinet.yml                | 103 ++++
 configs/rec/rec_svtrnet.yml                   |   2 +-
 configs/rec/rec_vitstr_none_ce.yml            |  12 +-
 doc/doc_ch/algorithm_overview.md              |   2 +
 doc/doc_ch/algorithm_rec_abinet.md            | 155 ++++++
 doc/doc_ch/algorithm_rec_nrtr.md              | 140 ++++-
 doc/doc_en/algorithm_overview_en.md           |   2 +
 doc/doc_en/algorithm_rec_abinet_en.md         | 136 +++++
 doc/doc_en/algorithm_rec_nrtr_en.md           | 141 ++++-
 doc/doc_en/algorithm_rec_vitstr_en.md         |   2 +-
 ppocr/data/imaug/__init__.py                  |   2 +-
 ppocr/data/imaug/label_ops.py                 | 158 +++---
 ppocr/data/imaug/operators.py                 |  33 --
 ppocr/data/imaug/rec_img_aug.py               |  38 ++
 ppocr/losses/rec_ce_loss.py                   |  76 ++-
 ppocr/modeling/backbones/__init__.py          |  18 +-
 ppocr/modeling/backbones/rec_resnet.py        | 280 ++++++++++
 ppocr/modeling/heads/__init__.py              |   3 +-
 ppocr/modeling/heads/multiheadAttention.py    | 163 ------
 ppocr/modeling/heads/rec_abinet_head.py       | 296 +++++++++++
 ppocr/modeling/heads/rec_nrtr_head.py         | 489 ++++++------------
 ppocr/postprocess/__init__.py                 |   4 +-
 ppocr/postprocess/rec_postprocess.py          | 209 ++++----
 .../configs/rec_mtb_nrtr/rec_mtb_nrtr.yml     |   2 +-
 .../configs/rec_r45_abinet/rec_r45_abinet.yml | 105 ++++
 .../rec_r45_abinet/train_infer_python.txt     |  53 ++
 test_tipc/configs/rec_svtrnet/rec_svtrnet.yml |   2 +-
 .../rec_svtrnet/train_infer_python.txt        |   4 +-
 .../rec_vitstr_none_ce/rec_vitstr_none_ce.yml |   5 +-
 tools/export_model.py                         |  15 +-
 tools/infer/predict_rec.py                    |  51 +-
 tools/program.py                              |   2 +-
 33 files changed, 1973 insertions(+), 736 deletions(-)
 create mode 100644 configs/rec/rec_r45_abinet.yml
 create mode 100644 doc/doc_ch/algorithm_rec_abinet.md
 create mode 100644 doc/doc_en/algorithm_rec_abinet_en.md
 create mode 100644 ppocr/modeling/backbones/rec_resnet.py
 delete mode 100755 ppocr/modeling/heads/multiheadAttention.py
 create mode 100644 ppocr/modeling/heads/rec_abinet_head.py
 create mode 100644 test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
 create mode 100644 test_tipc/configs/rec_r45_abinet/train_infer_python.txt

diff --git a/configs/rec/rec_mtb_nrtr.yml b/configs/rec/rec_mtb_nrtr.yml
index 405674882c..4e5826adc9 100644
--- a/configs/rec/rec_mtb_nrtr.yml
+++ b/configs/rec/rec_mtb_nrtr.yml
@@ -9,7 +9,7 @@ Global:
   eval_batch_step: [0, 2000]
   cal_metric_during_train: True
   pretrained_model:
-  checkpoints: 
+  checkpoints:
   save_inference_dir:
   use_visualdl: False
   infer_img: doc/imgs_words_en/word_10.png
@@ -82,7 +82,7 @@ Train:
 Eval:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/evaluation/
+    data_dir: ./train_data/data_lmdb_release/validation/
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
@@ -97,5 +97,5 @@ Eval:
     shuffle: False
     drop_last: False
     batch_size_per_card: 256
-    num_workers: 1
+    num_workers: 4
     use_shared_memory: False
diff --git a/configs/rec/rec_r45_abinet.yml b/configs/rec/rec_r45_abinet.yml
new file mode 100644
index 0000000000..a756fead9f
--- /dev/null
+++ b/configs/rec/rec_r45_abinet.yml
@@ -0,0 +1,103 @@
+Global:
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r45_abinet/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_abinet.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 20.0
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: ABINet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ResNet45
+
+  Head:
+    name: ABINetHead
+    use_lang: True
+    iter_size: 3
+    
+
+Loss:
+  name: CELoss
+  ignore_index: &ignore_index 100 # Must be greater than the number of character classes
+
+PostProcess:
+  name: ABINetLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 96
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/configs/rec/rec_svtrnet.yml b/configs/rec/rec_svtrnet.yml
index a700e4bd92..a3d292b6ba 100644
--- a/configs/rec/rec_svtrnet.yml
+++ b/configs/rec/rec_svtrnet.yml
@@ -26,7 +26,7 @@ Optimizer:
   name: AdamW
   beta1: 0.9
   beta2: 0.99
-  epsilon: 0.00000008
+  epsilon: 8.e-8
   weight_decay: 0.05
   no_weight_decay_name: norm pos_embed
   one_dim_param_no_weight_decay: true
diff --git a/configs/rec/rec_vitstr_none_ce.yml b/configs/rec/rec_vitstr_none_ce.yml
index c71eba0cc2..b969c83a5d 100644
--- a/configs/rec/rec_vitstr_none_ce.yml
+++ b/configs/rec/rec_vitstr_none_ce.yml
@@ -6,7 +6,7 @@ Global:
   save_model_dir: ./output/rec/vitstr_none_ce/
   save_epoch_step: 1
   # evaluation is run every 2000 iterations after the 0th iteration#
-  eval_batch_step: [0, 50]
+  eval_batch_step: [0, 2000]
   cal_metric_during_train: True
   pretrained_model:
   checkpoints:
@@ -23,7 +23,7 @@ Global:
 
 Optimizer:
   name: Adadelta
-  epsilon: 0.00000001
+  epsilon: 1.e-8
   rho: 0.95
   clip_norm: 5.0
   lr:
@@ -45,8 +45,8 @@ Architecture:
 
 Loss:
   name: CELoss
-  smoothing: False
   with_all: True
+  ignore_index: &ignore_index 0 # Must be zero or greater than the number of character classes
 
 PostProcess:
   name: ViTSTRLabelDecode
@@ -58,12 +58,13 @@ Metric:
 Train:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/training
+    data_dir: ./train_data/data_lmdb_release/training/
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
       - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
       - GrayRecResizeImg:
           image_shape: [224, 224] # W H
           resize_type: PIL # PIL or OpenCV
@@ -80,12 +81,13 @@ Train:
 Eval:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/validation
+    data_dir: ./train_data/data_lmdb_release/validation/
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
       - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
       - GrayRecResizeImg:
           image_shape: [224, 224] # W H
           resize_type: PIL # PIL or OpenCV
diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index 4196ec05af..eb81e4cd6d 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -67,6 +67,7 @@
 - [x]  [SEED](./algorithm_rec_seed.md)
 - [x]  [SVTR](./algorithm_rec_svtr.md)
 - [x]  [ViTSTR](./algorithm_rec_vitstr.md)
+- [x]  [ABINet](./algorithm_rec_abinet.md)
 
 参考[DTRB](https://arxiv.org/abs/1904.01906)[3]文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
 
@@ -86,6 +87,7 @@
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
 |ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_ch/algorithm_rec_abinet.md b/doc/doc_ch/algorithm_rec_abinet.md
new file mode 100644
index 0000000000..d20c703014
--- /dev/null
+++ b/doc/doc_ch/algorithm_rec_abinet.md
@@ -0,0 +1,155 @@
+# 场景文本识别算法-ABINet
+
+- [1. 算法简介](#1)
+- [2. 环境配置](#2)
+- [3. 模型训练、评估、预测](#3)
+    - [3.1 训练](#3-1)
+    - [3.2 评估](#3-2)
+    - [3.3 预测](#3-3)
+- [4. 推理部署](#4)
+    - [4.1 Python推理](#4-1)
+    - [4.2 C++推理](#4-2)
+    - [4.3 Serving服务化部署](#4-3)
+    - [4.4 更多推理部署](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. 算法简介
+
+论文信息：
+> [ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition](https://openaccess.thecvf.com/content/CVPR2021/papers/Fang_Read_Like_Humans_Autonomous_Bidirectional_and_Iterative_Language_Modeling_for_CVPR_2021_paper.pdf)
+> Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang
+> CVPR, 2021
+
+
+<a name="model"></a>
+`ABINet`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[训练模型]()/[预训练模型]|
+
+<a name="2"></a>
+## 2. 环境配置
+请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](./clone.md)克隆项目代码。
+
+
+<a name="3"></a>
+## 3. 模型训练、评估、预测
+
+<a name="3-1"></a>
+### 3.1 模型训练
+
+请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ABINet`识别模型时需要**更换配置文件**为`ABINet`的[配置文件](../../configs/rec/rec_r45_abinet.yml)。
+
+#### 启动训练
+
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r45_abinet.yml
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_abinet.yml
+```
+
+<a name="3-2"></a>
+### 3.2 评估
+
+可下载已训练完成的[模型文件](#model)，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+```
+
+<a name="3-3"></a>
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+
+<a name="4"></a>
+## 4. 推理部署
+
+<a name="4-1"></a>
+### 4.1 Python推理
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址]() )，可以使用如下命令进行转换：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy Global.save_inference_dir=./inference/rec_r45_abinet/
+```
+**注意：**
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ABINet的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+```
+/inference/rec_r45_abinet/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```shell
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_r45_abinet/' --rec_algorithm='ABINet' --rec_image_shape='3,32,128' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![](../imgs_words_en/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999995231628418)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[3，32，128]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ABINet的预处理为您的预处理方法。
+
+
+<a name="4-2"></a>
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持ABINet，所以暂未支持
+
+<a name="4-3"></a>
+### 4.3 Serving服务化部署
+
+暂不支持
+
+<a name="4-4"></a>
+### 4.4 更多推理部署
+
+暂不支持
+
+<a name="5"></a>
+## 5. FAQ
+
+1. MJSynth和SynthText两种数据集来自于[ABINet源repo](https://github.com/FangShancheng/ABINet) 。
+2. 我们使用ABINet作者提供的预训练模型进行finetune训练。
+
+## 引用
+
+```bibtex
+@article{Fang2021ABINet,
+  title     = {ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition},
+  author    = {Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang},
+  booktitle = {CVPR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2103.06495},
+  pages     = {7098-7107}
+}
+```
diff --git a/doc/doc_ch/algorithm_rec_nrtr.md b/doc/doc_ch/algorithm_rec_nrtr.md
index d3b626d024..c619ac1dbc 100644
--- a/doc/doc_ch/algorithm_rec_nrtr.md
+++ b/doc/doc_ch/algorithm_rec_nrtr.md
@@ -12,6 +12,7 @@
     - [4.3 Serving服务化部署](#4-3)
     - [4.4 更多推理部署](#4-4)
 - [5. FAQ](#5)
+- [6. 发行公告](#6)
 
 <a name="1"></a>
 ## 1. 算法简介
@@ -110,7 +111,7 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png'
 执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
 结果如下：
 ```shell
-Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9465042352676392)
 ```
 
 **注意**：
@@ -140,12 +141,147 @@ Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
 
 1. `NRTR`论文中使用Beam搜索进行解码字符，但是速度较慢，这里默认未使用Beam搜索，以贪婪搜索进行解码字符。
 
+<a name="6"></a>
+## 6. 发行公告
+
+1. release/2.6更新NRTR代码结构，新版NRTR可加载旧版(release/2.5及之前)模型参数，使用下面示例代码将旧版模型参数转换为新版模型参数：
+
+```python
+
+    params = paddle.load('path/' + '.pdparams') # 旧版本参数
+    state_dict = model.state_dict() # 新版模型参数
+    new_state_dict = {}
+
+    for k1, v1 in state_dict.items():
+
+        k = k1
+        if 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'encoder' in k and 'norm3' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para.replace('norm3', 'norm2')]
+
+        elif 'encoder' in k and 'norm1' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = q[:, :, 0, 0]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')]
+            new_state_dict[k1] = q
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('kv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')]
+            v = params[k_para.replace('kv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([k, v], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            new_state_dict[k1] = params[k_para]
+        elif 'decoder' in k and 'norm' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+        elif 'mlp' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = w[:, :, 0, 0]
+        elif 'mlp' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para]
+            new_state_dict[k1] = w
+
+        else:
+            new_state_dict[k1] = params[k1]
+
+        if list(new_state_dict[k1].shape) != list(v1.shape):
+            print(k1)
+
+
+    for k, v1 in state_dict.items():
+        if k not in new_state_dict.keys():
+            print(1, k)
+        elif list(new_state_dict[k].shape) != list(v1.shape):
+            print(2, k)
+
+
+
+    model.set_state_dict(new_state_dict)
+    paddle.save(model.state_dict(), 'nrtrnew_from_old_params.pdparams')
+
+```
+
+2. 新版相比与旧版，代码结构简洁，推理速度有所提高。
+
+
 ## 引用
 
 ```bibtex
 @article{Sheng2019NRTR,
   title     = {NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition},
-  author    = {Fenfen Sheng and Zhineng Chen andBo Xu},
+  author    = {Fenfen Sheng and Zhineng Chen and Bo Xu},
   booktitle = {ICDAR},
   year      = {2019},
   url       = {http://arxiv.org/abs/1806.00926},
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index f2284542ed..28aca7c0d1 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -66,6 +66,7 @@ Supported text recognition algorithms (Click the link to get the tutorial):
 - [x]  [SEED](./algorithm_rec_seed_en.md)
 - [x]  [SVTR](./algorithm_rec_svtr_en.md)
 - [x]  [ViTSTR](./algorithm_rec_vitstr_en.md)
+- [x]  [ABINet](./algorithm_rec_abinet_en.md)
 
 Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
 
@@ -85,6 +86,7 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
 |ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet_en | [trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_en/algorithm_rec_abinet_en.md b/doc/doc_en/algorithm_rec_abinet_en.md
new file mode 100644
index 0000000000..3b0f6c09a9
--- /dev/null
+++ b/doc/doc_en/algorithm_rec_abinet_en.md
@@ -0,0 +1,136 @@
+# ABINet
+
+- [1. Introduction](#1)
+- [2. Environment](#2)
+- [3. Model Training / Evaluation / Prediction](#3)
+    - [3.1 Training](#3-1)
+    - [3.2 Evaluation](#3-2)
+    - [3.3 Prediction](#3-3)
+- [4. Inference and Deployment](#4)
+    - [4.1 Python Inference](#4-1)
+    - [4.2 C++ Inference](#4-2)
+    - [4.3 Serving](#4-3)
+    - [4.4 More](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. Introduction
+
+Paper:
+> [ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition](https://openaccess.thecvf.com/content/CVPR2021/papers/Fang_Read_Like_Humans_Autonomous_Bidirectional_and_Iterative_Language_Modeling_for_CVPR_2021_paper.pdf)
+> Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang
+> CVPR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[trained model]()/[pretrained model]()|
+
+<a name="2"></a>
+## 2. Environment
+Please refer to ["Environment Preparation"](./environment_en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone_en.md) to clone the project code.
+
+
+<a name="3"></a>
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](./recognition_en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```
+#Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r45_abinet.yml
+
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_abinet.yml
+```
+
+Evaluation:
+
+```
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+Prediction:
+
+```
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+```
+
+<a name="4"></a>
+## 4. Inference and Deployment
+
+<a name="4-1"></a>
+### 4.1 Python Inference
+First, the model saved during the ABINet text recognition training process is converted into an inference model. ( [Model download link]()) ), you can use the following command to convert:
+
+```
+python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy  Global.save_inference_dir=./inference/rec_r45_abinet
+```
+
+**Note:**
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to ABINet in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+```
+/inference/rec_r45_abinet/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+
+For ABINet text recognition model inference, the following commands can be executed:
+
+```
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_r45_abinet/' --rec_algorithm='ABINet' --rec_image_shape='3,32,128' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+```
+
+![](../imgs_words_en/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999995231628418)
+```
+
+<a name="4-2"></a>
+### 4.2 C++ Inference
+
+Not supported
+
+<a name="4-3"></a>
+### 4.3 Serving
+
+Not supported
+
+<a name="4-4"></a>
+### 4.4 More
+
+Not supported
+
+<a name="5"></a>
+## 5. FAQ
+
+1. Note that the MJSynth and SynthText datasets come from [ABINet repo](https://github.com/FangShancheng/ABINet).
+2. We use the pre-trained model provided by the ABINet authors for finetune training.
+
+## Citation
+
+```bibtex
+@article{Fang2021ABINet,
+  title     = {ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition},
+  author    = {Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang},
+  booktitle = {CVPR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2103.06495},
+  pages     = {7098-7107}
+}
+```
diff --git a/doc/doc_en/algorithm_rec_nrtr_en.md b/doc/doc_en/algorithm_rec_nrtr_en.md
index 40c9b91629..309d7ab123 100644
--- a/doc/doc_en/algorithm_rec_nrtr_en.md
+++ b/doc/doc_en/algorithm_rec_nrtr_en.md
@@ -12,6 +12,7 @@
     - [4.3 Serving](#4-3)
     - [4.4 More](#4-4)
 - [5. FAQ](#5)
+- [6. Release Note](#6)
 
 <a name="1"></a>
 ## 1. Introduction
@@ -25,7 +26,7 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval
 
 |Model|Backbone|config|Acc|Download link|
 | --- | --- | --- | --- | --- |
-|NRTR|MTB|[rec_mtb_nrtr.yml](../../configs/rec/rec_mtb_nrtr.yml)|84.21%|[train model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)|
+|NRTR|MTB|[rec_mtb_nrtr.yml](../../configs/rec/rec_mtb_nrtr.yml)|84.21%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)|
 
 <a name="2"></a>
 ## 2. Environment
@@ -98,7 +99,7 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png'
 After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
 The result is as follows:
 ```shell
-Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9465042352676392)
 ```
 
 <a name="4-2"></a>
@@ -121,12 +122,146 @@ Not supported
 
 1. In the `NRTR` paper, Beam search is used to decode characters, but the speed is slow. Beam search is not used by default here, and greedy search is used to decode characters.
 
+<a name="6"></a>
+## 6. Release Note
+
+1. The release/2.6 version updates the NRTR code structure. The new version of NRTR can load the model parameters of the old version (release/2.5 and before), and you may use the following code to convert the old version model parameters to the new version model parameters:
+
+```python
+
+    params = paddle.load('path/' + '.pdparams') # the old version parameters
+    state_dict = model.state_dict() # the new version model parameters
+    new_state_dict = {}
+
+    for k1, v1 in state_dict.items():
+
+        k = k1
+        if 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'encoder' in k and 'norm3' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para.replace('norm3', 'norm2')]
+
+        elif 'encoder' in k and 'norm1' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = q[:, :, 0, 0]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')]
+            new_state_dict[k1] = q
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('kv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')]
+            v = params[k_para.replace('kv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([k, v], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            new_state_dict[k1] = params[k_para]
+        elif 'decoder' in k and 'norm' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+        elif 'mlp' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = w[:, :, 0, 0]
+        elif 'mlp' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para]
+            new_state_dict[k1] = w
+
+        else:
+            new_state_dict[k1] = params[k1]
+
+        if list(new_state_dict[k1].shape) != list(v1.shape):
+            print(k1)
+
+
+    for k, v1 in state_dict.items():
+        if k not in new_state_dict.keys():
+            print(1, k)
+        elif list(new_state_dict[k].shape) != list(v1.shape):
+            print(2, k)
+
+
+
+    model.set_state_dict(new_state_dict)
+    paddle.save(model.state_dict(), 'nrtrnew_from_old_params.pdparams')
+
+```
+
+2. The new version has a clean code structure and improved inference speed compared with the old version.
+
 ## Citation
 
 ```bibtex
 @article{Sheng2019NRTR,
   title     = {NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition},
-  author    = {Fenfen Sheng and Zhineng Chen andBo Xu},
+  author    = {Fenfen Sheng and Zhineng Chen and Bo Xu},
   booktitle = {ICDAR},
   year      = {2019},
   url       = {http://arxiv.org/abs/1806.00926},
diff --git a/doc/doc_en/algorithm_rec_vitstr_en.md b/doc/doc_en/algorithm_rec_vitstr_en.md
index b6f26d39e2..a6f9e2f15d 100644
--- a/doc/doc_en/algorithm_rec_vitstr_en.md
+++ b/doc/doc_en/algorithm_rec_vitstr_en.md
@@ -25,7 +25,7 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval
 
 |Model|Backbone|config|Acc|Download link|
 | --- | --- | --- | --- | --- |
-|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
 
 <a name="2"></a>
 ## 2. Environment
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index 88c9603fb0..58e8a5c780 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -24,7 +24,7 @@ from .make_pse_gt import MakePseGt
 
 
 from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
-    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg
+    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, ABINetRecResizeImg
 from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 0366e3f785..d96ecf639a 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -157,37 +157,6 @@ class BaseRecLabelEncode(object):
         return text_list
 
 
-class NRTRLabelEncode(BaseRecLabelEncode):
-    """ Convert between text-label and text-index """
-
-    def __init__(self,
-                 max_text_length,
-                 character_dict_path=None,
-                 use_space_char=False,
-                 **kwargs):
-
-        super(NRTRLabelEncode, self).__init__(
-            max_text_length, character_dict_path, use_space_char)
-
-    def __call__(self, data):
-        text = data['label']
-        text = self.encode(text)
-        if text is None:
-            return None
-        if len(text) >= self.max_text_len - 1:
-            return None
-        data['length'] = np.array(len(text))
-        text.insert(0, 2)
-        text.append(3)
-        text = text + [0] * (self.max_text_len - len(text))
-        data['label'] = np.array(text)
-        return data
-
-    def add_special_char(self, dict_character):
-        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
-        return dict_character
-
-
 class CTCLabelEncode(BaseRecLabelEncode):
     """ Convert between text-label and text-index """
 
@@ -840,37 +809,6 @@ class PRENLabelEncode(BaseRecLabelEncode):
         return data
 
 
-class ViTSTRLabelEncode(BaseRecLabelEncode):
-    """ Convert between text-label and text-index """
-
-    def __init__(self,
-                 max_text_length,
-                 character_dict_path=None,
-                 use_space_char=False,
-                 **kwargs):
-
-        super(ViTSTRLabelEncode, self).__init__(
-            max_text_length, character_dict_path, use_space_char)
-
-    def __call__(self, data):
-        text = data['label']
-        text = self.encode(text)
-        if text is None:
-            return None
-        if len(text) >= self.max_text_len:
-            return None
-        data['length'] = np.array(len(text))
-        text.insert(0, 0)
-        text.append(1)
-        text = text + [0] * (self.max_text_len + 2 - len(text))
-        data['label'] = np.array(text)
-        return data
-
-    def add_special_char(self, dict_character):
-        dict_character = ['<s>', '</s>'] + dict_character
-        return dict_character
-
-
 class VQATokenLabelEncode(object):
     """
     Label encode for NLP VQA methods
@@ -1077,3 +1015,99 @@ class MultiLabelEncode(BaseRecLabelEncode):
         data_out['label_sar'] = sar['label']
         data_out['length'] = ctc['length']
         return data_out
+
+
+class NRTRLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 **kwargs):
+
+        super(NRTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len - 1:
+            return None
+        data['length'] = np.array(len(text))
+        text.insert(0, 2)
+        text.append(3)
+        text = text + [0] * (self.max_text_len - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
+        return dict_character
+
+
+class ViTSTRLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 ignore_index=0,
+                 **kwargs):
+
+        super(ViTSTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data['length'] = np.array(len(text))
+        text.insert(0, self.ignore_index)
+        text.append(1)
+        text = text + [self.ignore_index] * (self.max_text_len + 2 - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['<s>', '</s>'] + dict_character
+        return dict_character
+
+
+class ABINetLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 ignore_index=100,
+                 **kwargs):
+
+        super(ABINetLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data['length'] = np.array(len(text))
+        text.append(0)
+        text = text + [self.ignore_index] * (self.max_text_len + 1 - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['</s>'] + dict_character
+        return dict_character
diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
index 09736515e7..5397d71ccb 100644
--- a/ppocr/data/imaug/operators.py
+++ b/ppocr/data/imaug/operators.py
@@ -67,39 +67,6 @@ class DecodeImage(object):
         return data
 
 
-class NRTRDecodeImage(object):
-    """ decode image """
-
-    def __init__(self, img_mode='RGB', channel_first=False, **kwargs):
-        self.img_mode = img_mode
-        self.channel_first = channel_first
-
-    def __call__(self, data):
-        img = data['image']
-        if six.PY2:
-            assert type(img) is str and len(
-                img) > 0, "invalid input 'img' in DecodeImage"
-        else:
-            assert type(img) is bytes and len(
-                img) > 0, "invalid input 'img' in DecodeImage"
-        img = np.frombuffer(img, dtype='uint8')
-
-        img = cv2.imdecode(img, 1)
-
-        if img is None:
-            return None
-        if self.img_mode == 'GRAY':
-            img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
-        elif self.img_mode == 'RGB':
-            assert img.shape[2] == 3, 'invalid shape of image[%s]' % (img.shape)
-            img = img[:, :, ::-1]
-        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        if self.channel_first:
-            img = img.transpose((2, 0, 1))
-        data['image'] = img
-        return data
-
-
 class NormalizeImage(object):
     """ normalize image such as substract mean, divide std
     """
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index d3120f5950..a8b3b813b3 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -279,6 +279,24 @@ class PRENResizeImg(object):
         return data
 
 
+class ABINetRecResizeImg(object):
+    def __init__(self,
+                 image_shape,
+                 infer_mode=False,
+                 character_dict_path=None,
+                 **kwargs):
+        self.image_shape = image_shape
+        self.infer_mode = infer_mode
+        self.character_dict_path = character_dict_path
+
+    def __call__(self, data):
+        img = data['image']
+        norm_img, valid_ratio = resize_norm_img_abinet(img, self.image_shape)
+        data['image'] = norm_img
+        data['valid_ratio'] = valid_ratio
+        return data
+
+
 def resize_norm_img_sar(img, image_shape, width_downsample_ratio=0.25):
     imgC, imgH, imgW_min, imgW_max = image_shape
     h = img.shape[0]
@@ -397,6 +415,26 @@ def resize_norm_img_srn(img, image_shape):
     return np.reshape(img_black, (c, row, col)).astype(np.float32)
 
 
+def resize_norm_img_abinet(img, image_shape):
+    imgC, imgH, imgW = image_shape
+
+    resized_image = cv2.resize(
+        img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+    resized_w = imgW
+    resized_image = resized_image.astype('float32')
+    resized_image = resized_image / 255.
+
+    mean = np.array([0.485, 0.456, 0.406])
+    std = np.array([0.229, 0.224, 0.225])
+    resized_image = (
+        resized_image - mean[None, None, ...]) / std[None, None, ...]
+    resized_image = resized_image.transpose((2, 0, 1))
+    resized_image = resized_image.astype('float32')
+
+    valid_ratio = min(1.0, float(resized_w / imgW))
+    return resized_image, valid_ratio
+
+
 def srn_other_inputs(image_shape, num_heads, max_text_length):
 
     imgC, imgH, imgW = image_shape
diff --git a/ppocr/losses/rec_ce_loss.py b/ppocr/losses/rec_ce_loss.py
index b837ac27bb..614384de86 100644
--- a/ppocr/losses/rec_ce_loss.py
+++ b/ppocr/losses/rec_ce_loss.py
@@ -4,31 +4,63 @@ import paddle.nn.functional as F
 
 
 class CELoss(nn.Layer):
-    def __init__(self, smoothing=True, with_all=False, **kwargs):
+    def __init__(self,
+                 smoothing=False,
+                 with_all=False,
+                 ignore_index=-1,
+                 **kwargs):
         super(CELoss, self).__init__()
-        self.loss_func = nn.CrossEntropyLoss(reduction='mean', ignore_index=0)
+        if ignore_index >= 0:
+            self.loss_func = nn.CrossEntropyLoss(
+                reduction='mean', ignore_index=ignore_index)
+        else:
+            self.loss_func = nn.CrossEntropyLoss(reduction='mean')
         self.smoothing = smoothing
         self.with_all = with_all
 
     def forward(self, pred, batch):
-        pred = pred.reshape([-1, pred.shape[2]])
-        if self.with_all:
-            tgt = batch[1]
+
+        if isinstance(pred, dict):  # for ABINet
+            loss = {}
+            loss_sum = []
+            for name, logits in pred.items():
+                if isinstance(logits, list):
+                    logit_num = len(logits)
+                    all_tgt = paddle.concat([batch[1]] * logit_num, 0)
+                    all_logits = paddle.concat(logits, 0)
+                    flt_logtis = all_logits.reshape([-1, all_logits.shape[2]])
+                    flt_tgt = all_tgt.reshape([-1])
+                else:
+                    flt_logtis = logits.reshape([-1, logits.shape[2]])
+                    flt_tgt = batch[1].reshape([-1])
+                loss[name + '_loss'] = self.loss_func(flt_logtis, flt_tgt)
+                loss_sum.append(loss[name + '_loss'])
+            loss['loss'] = sum(loss_sum)
+            return loss
         else:
-            max_len = batch[2].max()
-            tgt = batch[1][:, 1:2 + max_len]
-        tgt = tgt.reshape([-1])
-        if self.smoothing:
-            eps = 0.1
-            n_class = pred.shape[1]
-            one_hot = F.one_hot(tgt, pred.shape[1])
-            one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
-            log_prb = F.log_softmax(pred, axis=1)
-            non_pad_mask = paddle.not_equal(
-                tgt, paddle.zeros(
-                    tgt.shape, dtype=tgt.dtype))
-            loss = -(one_hot * log_prb).sum(axis=1)
-            loss = loss.masked_select(non_pad_mask).mean()
-        else:
-            loss = self.loss_func(pred, tgt)
-        return {'loss': loss}
+            if self.with_all:  # for ViTSTR
+                tgt = batch[1]
+                pred = pred.reshape([-1, pred.shape[2]])
+                tgt = tgt.reshape([-1])
+                loss = self.loss_func(pred, tgt)
+                return {'loss': loss}
+            else:  # for NRTR
+                max_len = batch[2].max()
+                tgt = batch[1][:, 1:2 + max_len]
+                pred = pred.reshape([-1, pred.shape[2]])
+                tgt = tgt.reshape([-1])
+                if self.smoothing:
+                    eps = 0.1
+                    n_class = pred.shape[1]
+                    one_hot = F.one_hot(tgt, pred.shape[1])
+                    one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (
+                        n_class - 1)
+                    log_prb = F.log_softmax(pred, axis=1)
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(
+                            tgt.shape, dtype=tgt.dtype))
+                    loss = -(one_hot * log_prb).sum(axis=1)
+                    loss = loss.masked_select(non_pad_mask).mean()
+                else:
+                    loss = self.loss_func(pred, tgt)
+                return {'loss': loss}
diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
index a368e74816..56d2c2b9db 100755
--- a/ppocr/modeling/backbones/__init__.py
+++ b/ppocr/modeling/backbones/__init__.py
@@ -27,7 +27,7 @@ def build_backbone(config, model_type):
         from .rec_resnet_fpn import ResNetFPN
         from .rec_mv1_enhance import MobileNetV1Enhance
         from .rec_nrtr_mtb import MTB
-        from .rec_resnet_31 import ResNet31
+        from .rec_resnet import ResNet31, ResNet45
         from .rec_resnet_aster import ResNet_ASTER
         from .rec_micronet import MicroNet
         from .rec_efficientb3_pren import EfficientNetb3_PREN
@@ -35,29 +35,29 @@ def build_backbone(config, model_type):
         from .rec_vitstr import ViTSTR
         support_dict = [
             'MobileNetV1Enhance', 'MobileNetV3', 'ResNet', 'ResNetFPN', 'MTB',
-            "ResNet31", "ResNet_ASTER", 'MicroNet', 'EfficientNetb3_PREN',
-            'SVTRNet', 'ViTSTR'
+            'ResNet31', 'ResNet45', 'ResNet_ASTER', 'MicroNet',
+            'EfficientNetb3_PREN', 'SVTRNet', 'ViTSTR'
         ]
-    elif model_type == "e2e":
+    elif model_type == 'e2e':
         from .e2e_resnet_vd_pg import ResNet
         support_dict = ['ResNet']
     elif model_type == 'kie':
         from .kie_unet_sdmgr import Kie_backbone
         support_dict = ['Kie_backbone']
-    elif model_type == "table":
+    elif model_type == 'table':
         from .table_resnet_vd import ResNet
         from .table_mobilenet_v3 import MobileNetV3
-        support_dict = ["ResNet", "MobileNetV3"]
+        support_dict = ['ResNet', 'MobileNetV3']
     elif model_type == 'vqa':
         from .vqa_layoutlm import LayoutLMForSer, LayoutLMv2ForSer, LayoutLMv2ForRe, LayoutXLMForSer, LayoutXLMForRe
         support_dict = [
-            "LayoutLMForSer", "LayoutLMv2ForSer", 'LayoutLMv2ForRe',
-            "LayoutXLMForSer", 'LayoutXLMForRe'
+            'LayoutLMForSer', 'LayoutLMv2ForSer', 'LayoutLMv2ForRe',
+            'LayoutXLMForSer', 'LayoutXLMForRe'
         ]
     else:
         raise NotImplementedError
 
-    module_name = config.pop("name")
+    module_name = config.pop('name')
     assert module_name in support_dict, Exception(
         "when model typs is {}, backbone only support {}".format(model_type,
                                                                  support_dict))
diff --git a/ppocr/modeling/backbones/rec_resnet.py b/ppocr/modeling/backbones/rec_resnet.py
new file mode 100644
index 0000000000..ce76b642cc
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet.py
@@ -0,0 +1,280 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/layers/conv_layer.py
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/backbones/resnet31_ocr.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+import math
+
+__all__ = ["ResNet31", "ResNet45"]
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    return nn.Conv2D(
+        in_planes,
+        out_planes,
+        kernel_size=1,
+        stride=stride,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False)
+
+
+def conv3x3(in_channel, out_channel, stride=1):
+    return nn.Conv2D(
+        in_channel,
+        out_channel,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False)
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self, in_channels, channels, stride=1, downsample=None):
+        super().__init__()
+        self.conv1 = conv1x1(in_channels, channels)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(channels, channels, stride)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet31(nn.Layer):
+    '''
+    Args:
+        in_channels (int): Number of channels of input image tensor.
+        layers (list[int]): List of BasicBlock number for each stage.
+        channels (list[int]): List of out_channels of Conv2d layer.
+        out_indices (None | Sequence[int]): Indices of output stages.
+        last_stage_pool (bool): If True, add `MaxPool2d` layer to last stage.
+    '''
+
+    def __init__(self,
+                 in_channels=3,
+                 layers=[1, 2, 5, 3],
+                 channels=[64, 128, 256, 256, 512, 512, 512],
+                 out_indices=None,
+                 last_stage_pool=False):
+        super(ResNet31, self).__init__()
+        assert isinstance(in_channels, int)
+        assert isinstance(last_stage_pool, bool)
+
+        self.out_indices = out_indices
+        self.last_stage_pool = last_stage_pool
+
+        # conv 1 (Conv Conv)
+        self.conv1_1 = nn.Conv2D(
+            in_channels, channels[0], kernel_size=3, stride=1, padding=1)
+        self.bn1_1 = nn.BatchNorm2D(channels[0])
+        self.relu1_1 = nn.ReLU()
+
+        self.conv1_2 = nn.Conv2D(
+            channels[0], channels[1], kernel_size=3, stride=1, padding=1)
+        self.bn1_2 = nn.BatchNorm2D(channels[1])
+        self.relu1_2 = nn.ReLU()
+
+        # conv 2 (Max-pooling, Residual block, Conv)
+        self.pool2 = nn.MaxPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block2 = self._make_layer(channels[1], channels[2], layers[0])
+        self.conv2 = nn.Conv2D(
+            channels[2], channels[2], kernel_size=3, stride=1, padding=1)
+        self.bn2 = nn.BatchNorm2D(channels[2])
+        self.relu2 = nn.ReLU()
+
+        # conv 3 (Max-pooling, Residual block, Conv)
+        self.pool3 = nn.MaxPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block3 = self._make_layer(channels[2], channels[3], layers[1])
+        self.conv3 = nn.Conv2D(
+            channels[3], channels[3], kernel_size=3, stride=1, padding=1)
+        self.bn3 = nn.BatchNorm2D(channels[3])
+        self.relu3 = nn.ReLU()
+
+        # conv 4 (Max-pooling, Residual block, Conv)
+        self.pool4 = nn.MaxPool2D(
+            kernel_size=(2, 1), stride=(2, 1), padding=0, ceil_mode=True)
+        self.block4 = self._make_layer(channels[3], channels[4], layers[2])
+        self.conv4 = nn.Conv2D(
+            channels[4], channels[4], kernel_size=3, stride=1, padding=1)
+        self.bn4 = nn.BatchNorm2D(channels[4])
+        self.relu4 = nn.ReLU()
+
+        # conv 5 ((Max-pooling), Residual block, Conv)
+        self.pool5 = None
+        if self.last_stage_pool:
+            self.pool5 = nn.MaxPool2D(
+                kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block5 = self._make_layer(channels[4], channels[5], layers[3])
+        self.conv5 = nn.Conv2D(
+            channels[5], channels[5], kernel_size=3, stride=1, padding=1)
+        self.bn5 = nn.BatchNorm2D(channels[5])
+        self.relu5 = nn.ReLU()
+
+        self.out_channels = channels[-1]
+
+    def _make_layer(self, input_channels, output_channels, blocks):
+        layers = []
+        for _ in range(blocks):
+            downsample = None
+            if input_channels != output_channels:
+                downsample = nn.Sequential(
+                    nn.Conv2D(
+                        input_channels,
+                        output_channels,
+                        kernel_size=1,
+                        stride=1,
+                        weight_attr=ParamAttr(initializer=KaimingNormal()),
+                        bias_attr=False),
+                    nn.BatchNorm2D(output_channels), )
+
+            layers.append(
+                BasicBlock(
+                    input_channels, output_channels, downsample=downsample))
+            input_channels = output_channels
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv1_1(x)
+        x = self.bn1_1(x)
+        x = self.relu1_1(x)
+
+        x = self.conv1_2(x)
+        x = self.bn1_2(x)
+        x = self.relu1_2(x)
+
+        outs = []
+        for i in range(4):
+            layer_index = i + 2
+            pool_layer = getattr(self, f'pool{layer_index}')
+            block_layer = getattr(self, f'block{layer_index}')
+            conv_layer = getattr(self, f'conv{layer_index}')
+            bn_layer = getattr(self, f'bn{layer_index}')
+            relu_layer = getattr(self, f'relu{layer_index}')
+
+            if pool_layer is not None:
+                x = pool_layer(x)
+            x = block_layer(x)
+            x = conv_layer(x)
+            x = bn_layer(x)
+            x = relu_layer(x)
+
+            outs.append(x)
+
+        if self.out_indices is not None:
+            return tuple([outs[i] for i in self.out_indices])
+
+        return x
+
+
+class ResNet(nn.Layer):
+    def __init__(self, block, layers, in_channels=3):
+        self.inplanes = 32
+        super(ResNet, self).__init__()
+        self.conv1 = nn.Conv2D(
+            3,
+            32,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False)
+        self.bn1 = nn.BatchNorm2D(32)
+        self.relu = nn.ReLU()
+
+        self.layer1 = self._make_layer(block, 32, layers[0], stride=2)
+        self.layer2 = self._make_layer(block, 64, layers[1], stride=1)
+        self.layer3 = self._make_layer(block, 128, layers[2], stride=2)
+        self.layer4 = self._make_layer(block, 256, layers[3], stride=1)
+        self.layer5 = self._make_layer(block, 512, layers[4], stride=1)
+        self.out_channels = 512
+
+        # for m in self.modules():
+        #     if isinstance(m, nn.Conv2D):
+        #         n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+        #         m.weight.data.normal_(0, math.sqrt(2. / n))
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            # downsample = True
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    weight_attr=ParamAttr(initializer=KaimingNormal()),
+                    bias_attr=False),
+                nn.BatchNorm2D(planes * block.expansion), )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        # print(x)
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        # print(x)
+        x = self.layer4(x)
+        x = self.layer5(x)
+        return x
+
+
+def ResNet45(in_channels=3):
+    return ResNet(BasicBlock, [3, 4, 6, 6, 3], in_channels=in_channels)
diff --git a/ppocr/modeling/heads/__init__.py b/ppocr/modeling/heads/__init__.py
index 1670ea38e6..14e6aab854 100755
--- a/ppocr/modeling/heads/__init__.py
+++ b/ppocr/modeling/heads/__init__.py
@@ -33,6 +33,7 @@ def build_head(config):
     from .rec_aster_head import AsterHead
     from .rec_pren_head import PRENHead
     from .rec_multi_head import MultiHead
+    from .rec_abinet_head import ABINetHead
 
     # cls head
     from .cls_head import ClsHead
@@ -46,7 +47,7 @@ def build_head(config):
         'DBHead', 'PSEHead', 'FCEHead', 'EASTHead', 'SASTHead', 'CTCHead',
         'ClsHead', 'AttentionHead', 'SRNHead', 'PGHead', 'Transformer',
         'TableAttentionHead', 'SARHead', 'AsterHead', 'SDMGRHead', 'PRENHead',
-        'MultiHead'
+        'MultiHead', 'ABINetHead'
     ]
 
     #table head
diff --git a/ppocr/modeling/heads/multiheadAttention.py b/ppocr/modeling/heads/multiheadAttention.py
deleted file mode 100755
index 900865ba1a..0000000000
--- a/ppocr/modeling/heads/multiheadAttention.py
+++ /dev/null
@@ -1,163 +0,0 @@
-# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#    http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import paddle
-from paddle import nn
-import paddle.nn.functional as F
-from paddle.nn import Linear
-from paddle.nn.initializer import XavierUniform as xavier_uniform_
-from paddle.nn.initializer import Constant as constant_
-from paddle.nn.initializer import XavierNormal as xavier_normal_
-
-zeros_ = constant_(value=0.)
-ones_ = constant_(value=1.)
-
-
-class MultiheadAttention(nn.Layer):
-    """Allows the model to jointly attend to information
-    from different representation subspaces.
-    See reference: Attention Is All You Need
-
-    .. math::
-        \text{MultiHead}(Q, K, V) = \text{Concat}(head_1,\dots,head_h)W^O
-        \text{where} head_i = \text{Attention}(QW_i^Q, KW_i^K, VW_i^V)
-
-    Args:
-        embed_dim: total dimension of the model
-        num_heads: parallel attention layers, or heads
-
-    """
-
-    def __init__(self,
-                 embed_dim,
-                 num_heads,
-                 dropout=0.,
-                 bias=True,
-                 add_bias_kv=False,
-                 add_zero_attn=False):
-        super(MultiheadAttention, self).__init__()
-        self.embed_dim = embed_dim
-        self.num_heads = num_heads
-        self.dropout = dropout
-        self.head_dim = embed_dim // num_heads
-        assert self.head_dim * num_heads == self.embed_dim, "embed_dim must be divisible by num_heads"
-        self.scaling = self.head_dim**-0.5
-        self.out_proj = Linear(embed_dim, embed_dim, bias_attr=bias)
-        self._reset_parameters()
-        self.conv1 = paddle.nn.Conv2D(
-            in_channels=embed_dim, out_channels=embed_dim, kernel_size=(1, 1))
-        self.conv2 = paddle.nn.Conv2D(
-            in_channels=embed_dim, out_channels=embed_dim, kernel_size=(1, 1))
-        self.conv3 = paddle.nn.Conv2D(
-            in_channels=embed_dim, out_channels=embed_dim, kernel_size=(1, 1))
-
-    def _reset_parameters(self):
-        xavier_uniform_(self.out_proj.weight)
-
-    def forward(self,
-                query,
-                key,
-                value,
-                key_padding_mask=None,
-                incremental_state=None,
-                attn_mask=None):
-        """
-        Inputs of forward function
-            query: [target length, batch size, embed dim]
-            key: [sequence length, batch size, embed dim]
-            value: [sequence length, batch size, embed dim]
-            key_padding_mask: if True, mask padding based on batch size
-            incremental_state: if provided, previous time steps are cashed
-            need_weights: output attn_output_weights
-            static_kv: key and value are static
-
-        Outputs of forward function
-            attn_output: [target length, batch size, embed dim]
-            attn_output_weights: [batch size, target length, sequence length]
-        """
-        q_shape = paddle.shape(query)
-        src_shape = paddle.shape(key)
-        q = self._in_proj_q(query)
-        k = self._in_proj_k(key)
-        v = self._in_proj_v(value)
-        q *= self.scaling
-        q = paddle.transpose(
-            paddle.reshape(
-                q, [q_shape[0], q_shape[1], self.num_heads, self.head_dim]),
-            [1, 2, 0, 3])
-        k = paddle.transpose(
-            paddle.reshape(
-                k, [src_shape[0], q_shape[1], self.num_heads, self.head_dim]),
-            [1, 2, 0, 3])
-        v = paddle.transpose(
-            paddle.reshape(
-                v, [src_shape[0], q_shape[1], self.num_heads, self.head_dim]),
-            [1, 2, 0, 3])
-        if key_padding_mask is not None:
-            assert key_padding_mask.shape[0] == q_shape[1]
-            assert key_padding_mask.shape[1] == src_shape[0]
-        attn_output_weights = paddle.matmul(q,
-                                            paddle.transpose(k, [0, 1, 3, 2]))
-        if attn_mask is not None:
-            attn_mask = paddle.unsqueeze(paddle.unsqueeze(attn_mask, 0), 0)
-            attn_output_weights += attn_mask
-        if key_padding_mask is not None:
-            attn_output_weights = paddle.reshape(
-                attn_output_weights,
-                [q_shape[1], self.num_heads, q_shape[0], src_shape[0]])
-            key = paddle.unsqueeze(paddle.unsqueeze(key_padding_mask, 1), 2)
-            key = paddle.cast(key, 'float32')
-            y = paddle.full(
-                shape=paddle.shape(key), dtype='float32', fill_value='-inf')
-            y = paddle.where(key == 0., key, y)
-            attn_output_weights += y
-        attn_output_weights = F.softmax(
-            attn_output_weights.astype('float32'),
-            axis=-1,
-            dtype=paddle.float32 if attn_output_weights.dtype == paddle.float16
-            else attn_output_weights.dtype)
-        attn_output_weights = F.dropout(
-            attn_output_weights, p=self.dropout, training=self.training)
-
-        attn_output = paddle.matmul(attn_output_weights, v)
-        attn_output = paddle.reshape(
-            paddle.transpose(attn_output, [2, 0, 1, 3]),
-            [q_shape[0], q_shape[1], self.embed_dim])
-        attn_output = self.out_proj(attn_output)
-
-        return attn_output
-
-    def _in_proj_q(self, query):
-        query = paddle.transpose(query, [1, 2, 0])
-        query = paddle.unsqueeze(query, axis=2)
-        res = self.conv1(query)
-        res = paddle.squeeze(res, axis=2)
-        res = paddle.transpose(res, [2, 0, 1])
-        return res
-
-    def _in_proj_k(self, key):
-        key = paddle.transpose(key, [1, 2, 0])
-        key = paddle.unsqueeze(key, axis=2)
-        res = self.conv2(key)
-        res = paddle.squeeze(res, axis=2)
-        res = paddle.transpose(res, [2, 0, 1])
-        return res
-
-    def _in_proj_v(self, value):
-        value = paddle.transpose(value, [1, 2, 0])  #(1, 2, 0)
-        value = paddle.unsqueeze(value, axis=2)
-        res = self.conv3(value)
-        res = paddle.squeeze(res, axis=2)
-        res = paddle.transpose(res, [2, 0, 1])
-        return res
diff --git a/ppocr/modeling/heads/rec_abinet_head.py b/ppocr/modeling/heads/rec_abinet_head.py
new file mode 100644
index 0000000000..a0f60f1be1
--- /dev/null
+++ b/ppocr/modeling/heads/rec_abinet_head.py
@@ -0,0 +1,296 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/FangShancheng/ABINet/tree/main/modules
+"""
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle.nn import LayerList
+from ppocr.modeling.heads.rec_nrtr_head import TransformerBlock, PositionalEncoding
+
+
+class BCNLanguage(nn.Layer):
+    def __init__(self,
+                 d_model=512,
+                 nhead=8,
+                 num_layers=4,
+                 dim_feedforward=2048,
+                 dropout=0.,
+                 max_length=25,
+                 detach=True,
+                 num_classes=37):
+        super().__init__()
+
+        self.d_model = d_model
+        self.detach = detach
+        self.max_length = max_length + 1  # additional stop token
+        self.proj = nn.Linear(num_classes, d_model, bias_attr=False)
+        self.token_encoder = PositionalEncoding(
+            dropout=0.1, dim=d_model, max_len=self.max_length)
+        self.pos_encoder = PositionalEncoding(
+            dropout=0, dim=d_model, max_len=self.max_length)
+
+        self.decoder = nn.LayerList([
+            TransformerBlock(
+                d_model=d_model,
+                nhead=nhead,
+                dim_feedforward=dim_feedforward,
+                attention_dropout_rate=dropout,
+                residual_dropout_rate=dropout,
+                with_self_attn=False,
+                with_cross_attn=True) for i in range(num_layers)
+        ])
+
+        self.cls = nn.Linear(d_model, num_classes)
+
+    def forward(self, tokens, lengths):
+        """
+        Args:
+            tokens: (B, N, C) where N is length, B is batch size and C is classes number
+            lengths: (B,)
+        """
+        if self.detach: tokens = tokens.detach()
+        embed = self.proj(tokens)  # (B, N, C)
+        embed = self.token_encoder(embed)  # (B, N, C)
+        padding_mask = _get_mask(lengths, self.max_length)
+        zeros = paddle.zeros_like(embed)  # (B, N, C)
+        qeury = self.pos_encoder(zeros)
+        for decoder_layer in self.decoder:
+            qeury = decoder_layer(qeury, embed, cross_mask=padding_mask)
+        output = qeury  # (B, N, C)
+
+        logits = self.cls(output)  # (B, N, C)
+
+        return output, logits
+
+
+def encoder_layer(in_c, out_c, k=3, s=2, p=1):
+    return nn.Sequential(
+        nn.Conv2D(in_c, out_c, k, s, p), nn.BatchNorm2D(out_c), nn.ReLU())
+
+
+def decoder_layer(in_c,
+                  out_c,
+                  k=3,
+                  s=1,
+                  p=1,
+                  mode='nearest',
+                  scale_factor=None,
+                  size=None):
+    align_corners = False if mode == 'nearest' else True
+    return nn.Sequential(
+        nn.Upsample(
+            size=size,
+            scale_factor=scale_factor,
+            mode=mode,
+            align_corners=align_corners),
+        nn.Conv2D(in_c, out_c, k, s, p),
+        nn.BatchNorm2D(out_c),
+        nn.ReLU())
+
+
+class PositionAttention(nn.Layer):
+    def __init__(self,
+                 max_length,
+                 in_channels=512,
+                 num_channels=64,
+                 h=8,
+                 w=32,
+                 mode='nearest',
+                 **kwargs):
+        super().__init__()
+        self.max_length = max_length
+        self.k_encoder = nn.Sequential(
+            encoder_layer(
+                in_channels, num_channels, s=(1, 2)),
+            encoder_layer(
+                num_channels, num_channels, s=(2, 2)),
+            encoder_layer(
+                num_channels, num_channels, s=(2, 2)),
+            encoder_layer(
+                num_channels, num_channels, s=(2, 2)))
+        self.k_decoder = nn.Sequential(
+            decoder_layer(
+                num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(
+                num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(
+                num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(
+                num_channels, in_channels, size=(h, w), mode=mode))
+
+        self.pos_encoder = PositionalEncoding(
+            dropout=0, dim=in_channels, max_len=max_length)
+        self.project = nn.Linear(in_channels, in_channels)
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        k, v = x, x
+
+        # calculate key vector
+        features = []
+        for i in range(0, len(self.k_encoder)):
+            k = self.k_encoder[i](k)
+            features.append(k)
+        for i in range(0, len(self.k_decoder) - 1):
+            k = self.k_decoder[i](k)
+            # print(k.shape, features[len(self.k_decoder) - 2 - i].shape)
+            k = k + features[len(self.k_decoder) - 2 - i]
+        k = self.k_decoder[-1](k)
+
+        # calculate query vector
+        # TODO q=f(q,k)
+        zeros = paddle.zeros(
+            (B, self.max_length, C), dtype=x.dtype)  # (T, N, C)
+        q = self.pos_encoder(zeros)  # (B, N, C)
+        q = self.project(q)  # (B, N, C)
+
+        # calculate attention
+        attn_scores = q @k.flatten(2)  # (B, N, (H*W))
+        attn_scores = attn_scores / (C**0.5)
+        attn_scores = F.softmax(attn_scores, axis=-1)
+
+        v = v.flatten(2).transpose([0, 2, 1])  # (B, (H*W), C)
+        attn_vecs = attn_scores @v  # (B, N, C)
+
+        return attn_vecs, attn_scores.reshape([0, self.max_length, H, W])
+
+
+class ABINetHead(nn.Layer):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 d_model=512,
+                 nhead=8,
+                 num_layers=3,
+                 dim_feedforward=2048,
+                 dropout=0.1,
+                 max_length=25,
+                 use_lang=False,
+                 iter_size=1):
+        super().__init__()
+        self.max_length = max_length + 1
+        self.pos_encoder = PositionalEncoding(
+            dropout=0.1, dim=d_model, max_len=8 * 32)
+        self.encoder = nn.LayerList([
+            TransformerBlock(
+                d_model=d_model,
+                nhead=nhead,
+                dim_feedforward=dim_feedforward,
+                attention_dropout_rate=dropout,
+                residual_dropout_rate=dropout,
+                with_self_attn=True,
+                with_cross_attn=False) for i in range(num_layers)
+        ])
+        self.decoder = PositionAttention(
+            max_length=max_length + 1,  # additional stop token
+            mode='nearest', )
+        self.out_channels = out_channels
+        self.cls = nn.Linear(d_model, self.out_channels)
+        self.use_lang = use_lang
+        if use_lang:
+            self.iter_size = iter_size
+            self.language = BCNLanguage(
+                d_model=d_model,
+                nhead=nhead,
+                num_layers=4,
+                dim_feedforward=dim_feedforward,
+                dropout=dropout,
+                max_length=max_length,
+                num_classes=self.out_channels)
+            # alignment
+            self.w_att_align = nn.Linear(2 * d_model, d_model)
+            self.cls_align = nn.Linear(d_model, self.out_channels)
+
+    def forward(self, x, targets=None):
+        x = x.transpose([0, 2, 3, 1])
+        _, H, W, C = x.shape
+        feature = x.flatten(1, 2)
+        feature = self.pos_encoder(feature)
+        for encoder_layer in self.encoder:
+            feature = encoder_layer(feature)
+        feature = feature.reshape([0, H, W, C]).transpose([0, 3, 1, 2])
+        v_feature, attn_scores = self.decoder(
+            feature)  # (B, N, C), (B, C, H, W)
+        vis_logits = self.cls(v_feature)  # (B, N, C)
+        logits = vis_logits
+        vis_lengths = _get_length(vis_logits)
+        if self.use_lang:
+            align_logits = vis_logits
+            align_lengths = vis_lengths
+            all_l_res, all_a_res = [], []
+            for i in range(self.iter_size):
+                tokens = F.softmax(align_logits, axis=-1)
+                lengths = align_lengths
+                lengths = paddle.clip(
+                    lengths, 2, self.max_length)  # TODO:move to langauge model
+                l_feature, l_logits = self.language(tokens, lengths)
+
+                # alignment
+                all_l_res.append(l_logits)
+                fuse = paddle.concat((l_feature, v_feature), -1)
+                f_att = F.sigmoid(self.w_att_align(fuse))
+                output = f_att * v_feature + (1 - f_att) * l_feature
+                align_logits = self.cls_align(output)  # (B, N, C)
+
+                align_lengths = _get_length(align_logits)
+                all_a_res.append(align_logits)
+            if self.training:
+                return {
+                    'align': all_a_res,
+                    'lang': all_l_res,
+                    'vision': vis_logits
+                }
+            else:
+                logits = align_logits
+        if self.training:
+            return logits
+        else:
+            return F.softmax(logits, -1)
+
+
+def _get_length(logit):
+    """ Greed decoder to obtain length from logit"""
+    out = (logit.argmax(-1) == 0)
+    abn = out.any(-1)
+    out_int = out.cast('int32')
+    out = (out_int.cumsum(-1) == 1) & out
+    out = out.cast('int32')
+    out = out.argmax(-1)
+    out = out + 1
+    out = paddle.where(abn, out, paddle.to_tensor(logit.shape[1]))
+    return out
+
+
+def _get_mask(length, max_length):
+    """Generate a square mask for the sequence. The masked positions are filled with float('-inf').
+        Unmasked positions are filled with float(0.0).
+    """
+    length = length.unsqueeze(-1)
+    B = paddle.shape(length)[0]
+    grid = paddle.arange(0, max_length).unsqueeze(0).tile([B, 1])
+    zero_mask = paddle.zeros([B, max_length], dtype='float32')
+    inf_mask = paddle.full([B, max_length], '-inf', dtype='float32')
+    diag_mask = paddle.diag(
+        paddle.full(
+            [max_length], '-inf', dtype=paddle.float32),
+        offset=0,
+        name=None)
+    mask = paddle.where(grid >= length, inf_mask, zero_mask)
+    mask = mask.unsqueeze(1) + diag_mask
+    return mask.unsqueeze(1)
diff --git a/ppocr/modeling/heads/rec_nrtr_head.py b/ppocr/modeling/heads/rec_nrtr_head.py
index 38ba0c9178..bf9ef56145 100644
--- a/ppocr/modeling/heads/rec_nrtr_head.py
+++ b/ppocr/modeling/heads/rec_nrtr_head.py
@@ -14,20 +14,15 @@
 
 import math
 import paddle
-import copy
 from paddle import nn
 import paddle.nn.functional as F
 from paddle.nn import LayerList
-from paddle.nn.initializer import XavierNormal as xavier_uniform_
-from paddle.nn import Dropout, Linear, LayerNorm, Conv2D
+# from paddle.nn.initializer import XavierNormal as xavier_uniform_
+from paddle.nn import Dropout, Linear, LayerNorm
 import numpy as np
-from ppocr.modeling.heads.multiheadAttention import MultiheadAttention
-from paddle.nn.initializer import Constant as constant_
+from ppocr.modeling.backbones.rec_svtrnet import Mlp, zeros_, ones_
 from paddle.nn.initializer import XavierNormal as xavier_normal_
 
-zeros_ = constant_(value=0.)
-ones_ = constant_(value=1.)
-
 
 class Transformer(nn.Layer):
     """A transformer model. User is able to modify the attributes as needed. The architechture
@@ -45,7 +40,6 @@ class Transformer(nn.Layer):
         dropout: the dropout value (default=0.1).
         custom_encoder: custom encoder (default=None).
         custom_decoder: custom decoder (default=None).
-
     """
 
     def __init__(self,
@@ -54,45 +48,49 @@ class Transformer(nn.Layer):
                  num_encoder_layers=6,
                  beam_size=0,
                  num_decoder_layers=6,
+                 max_len=25,
                  dim_feedforward=1024,
                  attention_dropout_rate=0.0,
                  residual_dropout_rate=0.1,
-                 custom_encoder=None,
-                 custom_decoder=None,
                  in_channels=0,
                  out_channels=0,
                  scale_embedding=True):
         super(Transformer, self).__init__()
         self.out_channels = out_channels + 1
+        self.max_len = max_len
         self.embedding = Embeddings(
             d_model=d_model,
             vocab=self.out_channels,
             padding_idx=0,
             scale_embedding=scale_embedding)
         self.positional_encoding = PositionalEncoding(
-            dropout=residual_dropout_rate,
-            dim=d_model, )
-        if custom_encoder is not None:
-            self.encoder = custom_encoder
-        else:
-            if num_encoder_layers > 0:
-                encoder_layer = TransformerEncoderLayer(
-                    d_model, nhead, dim_feedforward, attention_dropout_rate,
-                    residual_dropout_rate)
-                self.encoder = TransformerEncoder(encoder_layer,
-                                                  num_encoder_layers)
-            else:
-                self.encoder = None
+            dropout=residual_dropout_rate, dim=d_model)
 
-        if custom_decoder is not None:
-            self.decoder = custom_decoder
+        if num_encoder_layers > 0:
+            self.encoder = nn.LayerList([
+                TransformerBlock(
+                    d_model,
+                    nhead,
+                    dim_feedforward,
+                    attention_dropout_rate,
+                    residual_dropout_rate,
+                    with_self_attn=True,
+                    with_cross_attn=False) for i in range(num_encoder_layers)
+            ])
         else:
-            decoder_layer = TransformerDecoderLayer(
-                d_model, nhead, dim_feedforward, attention_dropout_rate,
-                residual_dropout_rate)
-            self.decoder = TransformerDecoder(decoder_layer, num_decoder_layers)
+            self.encoder = None
+
+        self.decoder = nn.LayerList([
+            TransformerBlock(
+                d_model,
+                nhead,
+                dim_feedforward,
+                attention_dropout_rate,
+                residual_dropout_rate,
+                with_self_attn=True,
+                with_cross_attn=True) for i in range(num_decoder_layers)
+        ])
 
-        self._reset_parameters()
         self.beam_size = beam_size
         self.d_model = d_model
         self.nhead = nhead
@@ -105,7 +103,7 @@ class Transformer(nn.Layer):
 
     def _init_weights(self, m):
 
-        if isinstance(m, nn.Conv2D):
+        if isinstance(m, nn.Linear):
             xavier_normal_(m.weight)
             if m.bias is not None:
                 zeros_(m.bias)
@@ -113,24 +111,20 @@ class Transformer(nn.Layer):
     def forward_train(self, src, tgt):
         tgt = tgt[:, :-1]
 
-        tgt_key_padding_mask = self.generate_padding_mask(tgt)
-        tgt = self.embedding(tgt).transpose([1, 0, 2])
+        tgt = self.embedding(tgt)
         tgt = self.positional_encoding(tgt)
-        tgt_mask = self.generate_square_subsequent_mask(tgt.shape[0])
+        tgt_mask = self.generate_square_subsequent_mask(tgt.shape[1])
 
         if self.encoder is not None:
-            src = self.positional_encoding(src.transpose([1, 0, 2]))
-            memory = self.encoder(src)
+            src = self.positional_encoding(src)
+            for encoder_layer in self.encoder:
+                src = encoder_layer(src)
+            memory = src  # B N C
         else:
-            memory = src.squeeze(2).transpose([2, 0, 1])
-        output = self.decoder(
-            tgt,
-            memory,
-            tgt_mask=tgt_mask,
-            memory_mask=None,
-            tgt_key_padding_mask=tgt_key_padding_mask,
-            memory_key_padding_mask=None)
-        output = output.transpose([1, 0, 2])
+            memory = src  # B N C
+        for decoder_layer in self.decoder:
+            tgt = decoder_layer(tgt, memory, self_mask=tgt_mask)
+        output = tgt
         logit = self.tgt_word_prj(output)
         return logit
 
@@ -140,8 +134,8 @@ class Transformer(nn.Layer):
             src: the sequence to the encoder (required).
             tgt: the sequence to the decoder (required).
         Shape:
-            - src: :math:`(S, N, E)`.
-            - tgt: :math:`(T, N, E)`.
+            - src: :math:`(B, sN, C)`.
+            - tgt: :math:`(B, tN, C)`.
         Examples:
             >>> output = transformer_model(src, tgt)
         """
@@ -157,36 +151,35 @@ class Transformer(nn.Layer):
                 return self.forward_test(src)
 
     def forward_test(self, src):
+
         bs = paddle.shape(src)[0]
         if self.encoder is not None:
-            src = self.positional_encoding(paddle.transpose(src, [1, 0, 2]))
-            memory = self.encoder(src)
+            src = self.positional_encoding(src)
+            for encoder_layer in self.encoder:
+                src = encoder_layer(src)
+            memory = src  # B N C
         else:
-            memory = paddle.transpose(paddle.squeeze(src, 2), [2, 0, 1])
+            memory = src
         dec_seq = paddle.full((bs, 1), 2, dtype=paddle.int64)
         dec_prob = paddle.full((bs, 1), 1., dtype=paddle.float32)
-        for len_dec_seq in range(1, 25):
-            dec_seq_embed = paddle.transpose(self.embedding(dec_seq), [1, 0, 2])
+        for len_dec_seq in range(1, self.max_len):
+            dec_seq_embed = self.embedding(dec_seq)
             dec_seq_embed = self.positional_encoding(dec_seq_embed)
             tgt_mask = self.generate_square_subsequent_mask(
-                paddle.shape(dec_seq_embed)[0])
-            output = self.decoder(
-                dec_seq_embed,
-                memory,
-                tgt_mask=tgt_mask,
-                memory_mask=None,
-                tgt_key_padding_mask=None,
-                memory_key_padding_mask=None)
-            dec_output = paddle.transpose(output, [1, 0, 2])
+                paddle.shape(dec_seq_embed)[1])
+            tgt = dec_seq_embed
+            for decoder_layer in self.decoder:
+                tgt = decoder_layer(tgt, memory, self_mask=tgt_mask)
+            dec_output = tgt
             dec_output = dec_output[:, -1, :]
-            word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=1)
-            preds_idx = paddle.argmax(word_prob, axis=1)
+            word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=-1)
+            preds_idx = paddle.argmax(word_prob, axis=-1)
             if paddle.equal_all(
                     preds_idx,
                     paddle.full(
                         paddle.shape(preds_idx), 3, dtype='int64')):
                 break
-            preds_prob = paddle.max(word_prob, axis=1)
+            preds_prob = paddle.max(word_prob, axis=-1)
             dec_seq = paddle.concat(
                 [dec_seq, paddle.reshape(preds_idx, [-1, 1])], axis=1)
             dec_prob = paddle.concat(
@@ -194,10 +187,10 @@ class Transformer(nn.Layer):
         return [dec_seq, dec_prob]
 
     def forward_beam(self, images):
-        ''' Translation work in one batch '''
+        """ Translation work in one batch """
 
         def get_inst_idx_to_tensor_position_map(inst_idx_list):
-            ''' Indicate the position of an instance in a tensor. '''
+            """ Indicate the position of an instance in a tensor. """
             return {
                 inst_idx: tensor_position
                 for tensor_position, inst_idx in enumerate(inst_idx_list)
@@ -205,7 +198,7 @@ class Transformer(nn.Layer):
 
         def collect_active_part(beamed_tensor, curr_active_inst_idx,
                                 n_prev_active_inst, n_bm):
-            ''' Collect tensor parts associated to active instances. '''
+            """ Collect tensor parts associated to active instances. """
 
             beamed_tensor_shape = paddle.shape(beamed_tensor)
             n_curr_active_inst = len(curr_active_inst_idx)
@@ -237,9 +230,8 @@ class Transformer(nn.Layer):
             return active_src_enc, active_inst_idx_to_position_map
 
         def beam_decode_step(inst_dec_beams, len_dec_seq, enc_output,
-                             inst_idx_to_position_map, n_bm,
-                             memory_key_padding_mask):
-            ''' Decode and update beam status, and then return active beam idx '''
+                             inst_idx_to_position_map, n_bm):
+            """ Decode and update beam status, and then return active beam idx """
 
             def prepare_beam_dec_seq(inst_dec_beams, len_dec_seq):
                 dec_partial_seq = [
@@ -249,19 +241,15 @@ class Transformer(nn.Layer):
                 dec_partial_seq = dec_partial_seq.reshape([-1, len_dec_seq])
                 return dec_partial_seq
 
-            def predict_word(dec_seq, enc_output, n_active_inst, n_bm,
-                             memory_key_padding_mask):
-                dec_seq = paddle.transpose(self.embedding(dec_seq), [1, 0, 2])
+            def predict_word(dec_seq, enc_output, n_active_inst, n_bm):
+                dec_seq = self.embedding(dec_seq)
                 dec_seq = self.positional_encoding(dec_seq)
                 tgt_mask = self.generate_square_subsequent_mask(
-                    paddle.shape(dec_seq)[0])
-                dec_output = self.decoder(
-                    dec_seq,
-                    enc_output,
-                    tgt_mask=tgt_mask,
-                    tgt_key_padding_mask=None,
-                    memory_key_padding_mask=memory_key_padding_mask, )
-                dec_output = paddle.transpose(dec_output, [1, 0, 2])
+                    paddle.shape(dec_seq)[1])
+                tgt = dec_seq
+                for decoder_layer in self.decoder:
+                    tgt = decoder_layer(tgt, enc_output, self_mask=tgt_mask)
+                dec_output = tgt
                 dec_output = dec_output[:,
                                         -1, :]  # Pick the last step: (bh * bm) * d_h
                 word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=1)
@@ -281,8 +269,7 @@ class Transformer(nn.Layer):
 
             n_active_inst = len(inst_idx_to_position_map)
             dec_seq = prepare_beam_dec_seq(inst_dec_beams, len_dec_seq)
-            word_prob = predict_word(dec_seq, enc_output, n_active_inst, n_bm,
-                                     None)
+            word_prob = predict_word(dec_seq, enc_output, n_active_inst, n_bm)
             # Update the beam with predicted word prob information and collect incomplete instances
             active_inst_idx_list = collect_active_inst_idx_list(
                 inst_dec_beams, word_prob, inst_idx_to_position_map)
@@ -303,10 +290,10 @@ class Transformer(nn.Layer):
         with paddle.no_grad():
             #-- Encode
             if self.encoder is not None:
-                src = self.positional_encoding(images.transpose([1, 0, 2]))
+                src = self.positional_encoding(images)
                 src_enc = self.encoder(src)
             else:
-                src_enc = images.squeeze(2).transpose([0, 2, 1])
+                src_enc = images
 
             n_bm = self.beam_size
             src_shape = paddle.shape(src_enc)
@@ -317,11 +304,11 @@ class Transformer(nn.Layer):
             inst_idx_to_position_map = get_inst_idx_to_tensor_position_map(
                 active_inst_idx_list)
             # Decode
-            for len_dec_seq in range(1, 25):
+            for len_dec_seq in range(1, self.max_len):
                 src_enc_copy = src_enc.clone()
                 active_inst_idx_list = beam_decode_step(
                     inst_dec_beams, len_dec_seq, src_enc_copy,
-                    inst_idx_to_position_map, n_bm, None)
+                    inst_idx_to_position_map, n_bm)
                 if not active_inst_idx_list:
                     break  # all instances have finished their path to <EOS>
                 src_enc, inst_idx_to_position_map = collate_active_info(
@@ -354,261 +341,124 @@ class Transformer(nn.Layer):
                 shape=[sz, sz], dtype='float32', fill_value='-inf'),
             diagonal=1)
         mask = mask + mask_inf
-        return mask
-
-    def generate_padding_mask(self, x):
-        padding_mask = paddle.equal(x, paddle.to_tensor(0, dtype=x.dtype))
-        return padding_mask
-
-    def _reset_parameters(self):
-        """Initiate parameters in the transformer model."""
-
-        for p in self.parameters():
-            if p.dim() > 1:
-                xavier_uniform_(p)
+        return mask.unsqueeze([0, 1])
 
 
-class TransformerEncoder(nn.Layer):
-    """TransformerEncoder is a stack of N encoder layers
-    Args:
-        encoder_layer: an instance of the TransformerEncoderLayer() class (required).
-        num_layers: the number of sub-encoder-layers in the encoder (required).
-        norm: the layer normalization component (optional).
-    """
+class MultiheadAttention(nn.Layer):
+    """Allows the model to jointly attend to information
+    from different representation subspaces.
+    See reference: Attention Is All You Need
 
-    def __init__(self, encoder_layer, num_layers):
-        super(TransformerEncoder, self).__init__()
-        self.layers = _get_clones(encoder_layer, num_layers)
-        self.num_layers = num_layers
-
-    def forward(self, src):
-        """Pass the input through the endocder layers in turn.
-        Args:
-            src: the sequnce to the encoder (required).
-            mask: the mask for the src sequence (optional).
-            src_key_padding_mask: the mask for the src keys per batch (optional).
-        """
-        output = src
-
-        for i in range(self.num_layers):
-            output = self.layers[i](output,
-                                    src_mask=None,
-                                    src_key_padding_mask=None)
-
-        return output
-
-
-class TransformerDecoder(nn.Layer):
-    """TransformerDecoder is a stack of N decoder layers
+    .. math::
+        \text{MultiHead}(Q, K, V) = \text{Concat}(head_1,\dots,head_h)W^O
+        \text{where} head_i = \text{Attention}(QW_i^Q, KW_i^K, VW_i^V)
 
     Args:
-        decoder_layer: an instance of the TransformerDecoderLayer() class (required).
-        num_layers: the number of sub-decoder-layers in the decoder (required).
-        norm: the layer normalization component (optional).
+        embed_dim: total dimension of the model
+        num_heads: parallel attention layers, or heads
 
     """
 
-    def __init__(self, decoder_layer, num_layers):
-        super(TransformerDecoder, self).__init__()
-        self.layers = _get_clones(decoder_layer, num_layers)
-        self.num_layers = num_layers
+    def __init__(self, embed_dim, num_heads, dropout=0., self_attn=False):
+        super(MultiheadAttention, self).__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        # self.dropout = dropout
+        self.head_dim = embed_dim // num_heads
+        assert self.head_dim * num_heads == self.embed_dim, "embed_dim must be divisible by num_heads"
+        self.scale = self.head_dim**-0.5
+        self.self_attn = self_attn
+        if self_attn:
+            self.qkv = nn.Linear(embed_dim, embed_dim * 3)
+        else:
+            self.q = nn.Linear(embed_dim, embed_dim)
+            self.kv = nn.Linear(embed_dim, embed_dim * 2)
+        self.attn_drop = nn.Dropout(dropout)
+        self.out_proj = nn.Linear(embed_dim, embed_dim)
 
-    def forward(self,
-                tgt,
-                memory,
-                tgt_mask=None,
-                memory_mask=None,
-                tgt_key_padding_mask=None,
-                memory_key_padding_mask=None):
-        """Pass the inputs (and mask) through the decoder layer in turn.
+    def forward(self, query, key=None, attn_mask=None):
 
-        Args:
-            tgt: the sequence to the decoder (required).
-            memory: the sequnce from the last layer of the encoder (required).
-            tgt_mask: the mask for the tgt sequence (optional).
-            memory_mask: the mask for the memory sequence (optional).
-            tgt_key_padding_mask: the mask for the tgt keys per batch (optional).
-            memory_key_padding_mask: the mask for the memory keys per batch (optional).
-        """
-        output = tgt
-        for i in range(self.num_layers):
-            output = self.layers[i](
-                output,
-                memory,
-                tgt_mask=tgt_mask,
-                memory_mask=memory_mask,
-                tgt_key_padding_mask=tgt_key_padding_mask,
-                memory_key_padding_mask=memory_key_padding_mask)
+        qN = query.shape[1]
 
-        return output
+        if self.self_attn:
+            qkv = self.qkv(query).reshape(
+                (0, qN, 3, self.num_heads, self.head_dim)).transpose(
+                    (2, 0, 3, 1, 4))
+            q, k, v = qkv[0], qkv[1], qkv[2]
+        else:
+            kN = key.shape[1]
+            q = self.q(query).reshape(
+                [0, qN, self.num_heads, self.head_dim]).transpose([0, 2, 1, 3])
+            kv = self.kv(key).reshape(
+                (0, kN, 2, self.num_heads, self.head_dim)).transpose(
+                    (2, 0, 3, 1, 4))
+            k, v = kv[0], kv[1]
+
+        attn = (q.matmul(k.transpose((0, 1, 3, 2)))) * self.scale
+
+        if attn_mask is not None:
+            attn += attn_mask
+
+        attn = F.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape(
+            (0, qN, self.embed_dim))
+        x = self.out_proj(x)
+
+        return x
 
 
-class TransformerEncoderLayer(nn.Layer):
-    """TransformerEncoderLayer is made up of self-attn and feedforward network.
-    This standard encoder layer is based on the paper "Attention Is All You Need".
-    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
-    Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
-    Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
-    in a different way during application.
-
-    Args:
-        d_model: the number of expected features in the input (required).
-        nhead: the number of heads in the multiheadattention models (required).
-        dim_feedforward: the dimension of the feedforward network model (default=2048).
-        dropout: the dropout value (default=0.1).
-
-    """
-
+class TransformerBlock(nn.Layer):
     def __init__(self,
                  d_model,
                  nhead,
                  dim_feedforward=2048,
                  attention_dropout_rate=0.0,
-                 residual_dropout_rate=0.1):
-        super(TransformerEncoderLayer, self).__init__()
-        self.self_attn = MultiheadAttention(
-            d_model, nhead, dropout=attention_dropout_rate)
+                 residual_dropout_rate=0.1,
+                 with_self_attn=True,
+                 with_cross_attn=False,
+                 epsilon=1e-5):
+        super(TransformerBlock, self).__init__()
+        self.with_self_attn = with_self_attn
+        if with_self_attn:
+            self.self_attn = MultiheadAttention(
+                d_model,
+                nhead,
+                dropout=attention_dropout_rate,
+                self_attn=with_self_attn)
+            self.norm1 = LayerNorm(d_model, epsilon=epsilon)
+            self.dropout1 = Dropout(residual_dropout_rate)
+        self.with_cross_attn = with_cross_attn
+        if with_cross_attn:
+            self.cross_attn = MultiheadAttention(  #for self_attn of encoder or cross_attn of decoder
+                d_model,
+                nhead,
+                dropout=attention_dropout_rate)
+            self.norm2 = LayerNorm(d_model, epsilon=epsilon)
+            self.dropout2 = Dropout(residual_dropout_rate)
 
-        self.conv1 = Conv2D(
-            in_channels=d_model,
-            out_channels=dim_feedforward,
-            kernel_size=(1, 1))
-        self.conv2 = Conv2D(
-            in_channels=dim_feedforward,
-            out_channels=d_model,
-            kernel_size=(1, 1))
+        self.mlp = Mlp(in_features=d_model,
+                       hidden_features=dim_feedforward,
+                       act_layer=nn.ReLU,
+                       drop=residual_dropout_rate)
 
-        self.norm1 = LayerNorm(d_model)
-        self.norm2 = LayerNorm(d_model)
-        self.dropout1 = Dropout(residual_dropout_rate)
-        self.dropout2 = Dropout(residual_dropout_rate)
+        self.norm3 = LayerNorm(d_model, epsilon=epsilon)
 
-    def forward(self, src, src_mask=None, src_key_padding_mask=None):
-        """Pass the input through the endocder layer.
-        Args:
-            src: the sequnce to the encoder layer (required).
-            src_mask: the mask for the src sequence (optional).
-            src_key_padding_mask: the mask for the src keys per batch (optional).
-        """
-        src2 = self.self_attn(
-            src,
-            src,
-            src,
-            attn_mask=src_mask,
-            key_padding_mask=src_key_padding_mask)
-        src = src + self.dropout1(src2)
-        src = self.norm1(src)
-
-        src = paddle.transpose(src, [1, 2, 0])
-        src = paddle.unsqueeze(src, 2)
-        src2 = self.conv2(F.relu(self.conv1(src)))
-        src2 = paddle.squeeze(src2, 2)
-        src2 = paddle.transpose(src2, [2, 0, 1])
-        src = paddle.squeeze(src, 2)
-        src = paddle.transpose(src, [2, 0, 1])
-
-        src = src + self.dropout2(src2)
-        src = self.norm2(src)
-        return src
-
-
-class TransformerDecoderLayer(nn.Layer):
-    """TransformerDecoderLayer is made up of self-attn, multi-head-attn and feedforward network.
-    This standard decoder layer is based on the paper "Attention Is All You Need".
-    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
-    Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
-    Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
-    in a different way during application.
-
-    Args:
-        d_model: the number of expected features in the input (required).
-        nhead: the number of heads in the multiheadattention models (required).
-        dim_feedforward: the dimension of the feedforward network model (default=2048).
-        dropout: the dropout value (default=0.1).
-
-    """
-
-    def __init__(self,
-                 d_model,
-                 nhead,
-                 dim_feedforward=2048,
-                 attention_dropout_rate=0.0,
-                 residual_dropout_rate=0.1):
-        super(TransformerDecoderLayer, self).__init__()
-        self.self_attn = MultiheadAttention(
-            d_model, nhead, dropout=attention_dropout_rate)
-        self.multihead_attn = MultiheadAttention(
-            d_model, nhead, dropout=attention_dropout_rate)
-
-        self.conv1 = Conv2D(
-            in_channels=d_model,
-            out_channels=dim_feedforward,
-            kernel_size=(1, 1))
-        self.conv2 = Conv2D(
-            in_channels=dim_feedforward,
-            out_channels=d_model,
-            kernel_size=(1, 1))
-
-        self.norm1 = LayerNorm(d_model)
-        self.norm2 = LayerNorm(d_model)
-        self.norm3 = LayerNorm(d_model)
-        self.dropout1 = Dropout(residual_dropout_rate)
-        self.dropout2 = Dropout(residual_dropout_rate)
         self.dropout3 = Dropout(residual_dropout_rate)
 
-    def forward(self,
-                tgt,
-                memory,
-                tgt_mask=None,
-                memory_mask=None,
-                tgt_key_padding_mask=None,
-                memory_key_padding_mask=None):
-        """Pass the inputs (and mask) through the decoder layer.
+    def forward(self, tgt, memory=None, self_mask=None, cross_mask=None):
+        if self.with_self_attn:
+            tgt1 = self.self_attn(tgt, attn_mask=self_mask)
+            tgt = self.norm1(tgt + self.dropout1(tgt1))
 
-        Args:
-            tgt: the sequence to the decoder layer (required).
-            memory: the sequnce from the last layer of the encoder (required).
-            tgt_mask: the mask for the tgt sequence (optional).
-            memory_mask: the mask for the memory sequence (optional).
-            tgt_key_padding_mask: the mask for the tgt keys per batch (optional).
-            memory_key_padding_mask: the mask for the memory keys per batch (optional).
-
-        """
-        tgt2 = self.self_attn(
-            tgt,
-            tgt,
-            tgt,
-            attn_mask=tgt_mask,
-            key_padding_mask=tgt_key_padding_mask)
-        tgt = tgt + self.dropout1(tgt2)
-        tgt = self.norm1(tgt)
-        tgt2 = self.multihead_attn(
-            tgt,
-            memory,
-            memory,
-            attn_mask=memory_mask,
-            key_padding_mask=memory_key_padding_mask)
-        tgt = tgt + self.dropout2(tgt2)
-        tgt = self.norm2(tgt)
-
-        # default
-        tgt = paddle.transpose(tgt, [1, 2, 0])
-        tgt = paddle.unsqueeze(tgt, 2)
-        tgt2 = self.conv2(F.relu(self.conv1(tgt)))
-        tgt2 = paddle.squeeze(tgt2, 2)
-        tgt2 = paddle.transpose(tgt2, [2, 0, 1])
-        tgt = paddle.squeeze(tgt, 2)
-        tgt = paddle.transpose(tgt, [2, 0, 1])
-
-        tgt = tgt + self.dropout3(tgt2)
-        tgt = self.norm3(tgt)
+        if self.with_cross_attn:
+            tgt2 = self.cross_attn(tgt, key=memory, attn_mask=cross_mask)
+            tgt = self.norm2(tgt + self.dropout2(tgt2))
+        tgt = self.norm3(tgt + self.dropout3(self.mlp(tgt)))
         return tgt
 
 
-def _get_clones(module, N):
-    return LayerList([copy.deepcopy(module) for i in range(N)])
-
-
 class PositionalEncoding(nn.Layer):
     """Inject some information about the relative or absolute position of the tokens
         in the sequence. The positional encodings have the same dimension as
@@ -651,8 +501,9 @@ class PositionalEncoding(nn.Layer):
         Examples:
             >>> output = pos_encoder(x)
         """
+        x = x.transpose([1, 0, 2])
         x = x + self.pe[:paddle.shape(x)[0], :]
-        return self.dropout(x)
+        return self.dropout(x).transpose([1, 0, 2])
 
 
 class PositionalEncoding_2d(nn.Layer):
@@ -725,7 +576,7 @@ class PositionalEncoding_2d(nn.Layer):
 
 
 class Embeddings(nn.Layer):
-    def __init__(self, d_model, vocab, padding_idx, scale_embedding):
+    def __init__(self, d_model, vocab, padding_idx=None, scale_embedding=True):
         super(Embeddings, self).__init__()
         self.embedding = nn.Embedding(vocab, d_model, padding_idx=padding_idx)
         w0 = np.random.normal(0.0, d_model**-0.5,
@@ -742,7 +593,7 @@ class Embeddings(nn.Layer):
 
 
 class Beam():
-    ''' Beam search '''
+    """ Beam search """
 
     def __init__(self, size, device=False):
 
diff --git a/ppocr/postprocess/__init__.py b/ppocr/postprocess/__init__.py
index 4f900ee1fc..2635117c84 100644
--- a/ppocr/postprocess/__init__.py
+++ b/ppocr/postprocess/__init__.py
@@ -27,7 +27,7 @@ from .sast_postprocess import SASTPostProcess
 from .fce_postprocess import FCEPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, \
     DistillationCTCLabelDecode, TableLabelDecode, NRTRLabelDecode, SARLabelDecode, \
-    SEEDLabelDecode, PRENLabelDecode, ViTSTRLabelDecode
+    SEEDLabelDecode, PRENLabelDecode, ViTSTRLabelDecode, ABINetLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
 from .vqa_token_ser_layoutlm_postprocess import VQASerTokenLayoutLMPostProcess
@@ -42,7 +42,7 @@ def build_post_process(config, global_config=None):
         'DistillationDBPostProcess', 'NRTRLabelDecode', 'SARLabelDecode',
         'SEEDLabelDecode', 'VQASerTokenLayoutLMPostProcess',
         'VQAReTokenLayoutLMPostProcess', 'PRENLabelDecode',
-        'DistillationSARLabelDecode', 'ViTSTRLabelDecode'
+        'DistillationSARLabelDecode', 'ViTSTRLabelDecode', 'ABINetLabelDecode'
     ]
 
     if config['name'] == 'PSEPostProcess':
diff --git a/ppocr/postprocess/rec_postprocess.py b/ppocr/postprocess/rec_postprocess.py
index df6203fada..c77420ad19 100644
--- a/ppocr/postprocess/rec_postprocess.py
+++ b/ppocr/postprocess/rec_postprocess.py
@@ -140,96 +140,6 @@ class DistillationCTCLabelDecode(CTCLabelDecode):
         return output
 
 
-class NRTRLabelDecode(BaseRecLabelDecode):
-    """ Convert between text-label and text-index """
-
-    def __init__(self, character_dict_path=None, use_space_char=True, **kwargs):
-        super(NRTRLabelDecode, self).__init__(character_dict_path,
-                                              use_space_char)
-
-    def __call__(self, preds, label=None, *args, **kwargs):
-
-        if len(preds) == 2:
-            preds_id = preds[0]
-            preds_prob = preds[1]
-            if isinstance(preds_id, paddle.Tensor):
-                preds_id = preds_id.numpy()
-            if isinstance(preds_prob, paddle.Tensor):
-                preds_prob = preds_prob.numpy()
-            if preds_id[0][0] == 2:
-                preds_idx = preds_id[:, 1:]
-                preds_prob = preds_prob[:, 1:]
-            else:
-                preds_idx = preds_id
-            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
-            if label is None:
-                return text
-            label = self.decode(label[:, 1:])
-        else:
-            if isinstance(preds, paddle.Tensor):
-                preds = preds.numpy()
-            preds_idx = preds.argmax(axis=2)
-            preds_prob = preds.max(axis=2)
-            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
-            if label is None:
-                return text
-            label = self.decode(label[:, 1:])
-        return text, label
-
-    def add_special_char(self, dict_character):
-        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
-        return dict_character
-
-    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
-        """ convert text-index into text-label. """
-        result_list = []
-        batch_size = len(text_index)
-        for batch_idx in range(batch_size):
-            char_list = []
-            conf_list = []
-            for idx in range(len(text_index[batch_idx])):
-                try:
-                    char_idx = self.character[int(text_index[batch_idx][idx])]
-                except:
-                    continue
-                if char_idx == '</s>':  # end
-                    break
-                char_list.append(char_idx)
-                if text_prob is not None:
-                    conf_list.append(text_prob[batch_idx][idx])
-                else:
-                    conf_list.append(1)
-            text = ''.join(char_list)
-            result_list.append((text.lower(), np.mean(conf_list).tolist()))
-        return result_list
-
-
-class ViTSTRLabelDecode(NRTRLabelDecode):
-    """ Convert between text-label and text-index """
-
-    def __init__(self, character_dict_path=None, use_space_char=False,
-                 **kwargs):
-        super(ViTSTRLabelDecode, self).__init__(character_dict_path,
-                                                use_space_char)
-
-    def __call__(self, preds, label=None, *args, **kwargs):
-        if isinstance(preds, paddle.Tensor):
-            preds = preds[:, 1:].numpy()
-        else:
-            preds = preds[:, 1:]
-        preds_idx = preds.argmax(axis=2)
-        preds_prob = preds.max(axis=2)
-        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
-        if label is None:
-            return text
-        label = self.decode(label[:, 1:])
-        return text, label
-
-    def add_special_char(self, dict_character):
-        dict_character = ['<s>', '</s>'] + dict_character
-        return dict_character
-
-
 class AttnLabelDecode(BaseRecLabelDecode):
     """ Convert between text-label and text-index """
 
@@ -778,3 +688,122 @@ class PRENLabelDecode(BaseRecLabelDecode):
             return text
         label = self.decode(label)
         return text, label
+
+
+class NRTRLabelDecode(BaseRecLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=True, **kwargs):
+        super(NRTRLabelDecode, self).__init__(character_dict_path,
+                                              use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+
+        if len(preds) == 2:
+            preds_id = preds[0]
+            preds_prob = preds[1]
+            if isinstance(preds_id, paddle.Tensor):
+                preds_id = preds_id.numpy()
+            if isinstance(preds_prob, paddle.Tensor):
+                preds_prob = preds_prob.numpy()
+            if preds_id[0][0] == 2:
+                preds_idx = preds_id[:, 1:]
+                preds_prob = preds_prob[:, 1:]
+            else:
+                preds_idx = preds_id
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            if label is None:
+                return text
+            label = self.decode(label[:, 1:])
+        else:
+            if isinstance(preds, paddle.Tensor):
+                preds = preds.numpy()
+            preds_idx = preds.argmax(axis=2)
+            preds_prob = preds.max(axis=2)
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            if label is None:
+                return text
+            label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        result_list = []
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                try:
+                    char_idx = self.character[int(text_index[batch_idx][idx])]
+                except:
+                    continue
+                if char_idx == '</s>':  # end
+                    break
+                char_list.append(char_idx)
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = ''.join(char_list)
+            result_list.append((text.lower(), np.mean(conf_list).tolist()))
+        return result_list
+
+
+class ViTSTRLabelDecode(NRTRLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=False,
+                 **kwargs):
+        super(ViTSTRLabelDecode, self).__init__(character_dict_path,
+                                                use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds[:, 1:].numpy()
+        else:
+            preds = preds[:, 1:]
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['<s>', '</s>'] + dict_character
+        return dict_character
+
+
+class ABINetLabelDecode(NRTRLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=False,
+                 **kwargs):
+        super(ABINetLabelDecode, self).__init__(character_dict_path,
+                                                use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, dict):
+            preds = preds['align'][-1].numpy()
+        elif isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        else:
+            preds = preds
+
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['</s>'] + dict_character
+        return dict_character
diff --git a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
index ba6728d836..8118d58724 100644
--- a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
+++ b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
@@ -99,5 +99,5 @@ Eval:
     shuffle: False
     drop_last: False
     batch_size_per_card: 256
-    num_workers: 1
+    num_workers: 4
     use_shared_memory: False
diff --git a/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
new file mode 100644
index 0000000000..306c49d683
--- /dev/null
+++ b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
@@ -0,0 +1,105 @@
+Global:
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r45_abinet/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_abinet.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 20.0
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: ABINet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ResNet45
+
+  Head:
+    name: ABINetHead
+    use_lang: True
+    iter_size: 3
+    
+
+Loss:
+  name: CELoss
+  ignore_index: &ignore_index 100 # Must be greater than the number of character classes
+
+PostProcess:
+  name: ABINetLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 96
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/test_tipc/configs/rec_r45_abinet/train_infer_python.txt b/test_tipc/configs/rec_r45_abinet/train_infer_python.txt
new file mode 100644
index 0000000000..ecab1bcbbd
--- /dev/null
+++ b/test_tipc/configs/rec_r45_abinet/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_abinet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r45_abinet_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,128" --rec_algorithm="ABINet"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,128]}]
diff --git a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
index 26facca34d..bffd8ac09a 100644
--- a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
+++ b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
@@ -26,7 +26,7 @@ Optimizer:
   name: AdamW
   beta1: 0.9
   beta2: 0.99
-  epsilon: 0.00000008
+  epsilon: 8.e-8
   weight_decay: 0.05
   no_weight_decay_name: norm pos_embed
   one_dim_param_no_weight_decay: true
diff --git a/test_tipc/configs/rec_svtrnet/train_infer_python.txt b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
index 60fd25ad4f..a7e4a24063 100644
--- a/test_tipc/configs/rec_svtrnet/train_infer_python.txt
+++ b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
@@ -37,7 +37,7 @@ export2:null
 train_model:./inference/rec_svtrnet_train/best_accuracy
 infer_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
 infer_quant:False
-inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="3,64,256" --rec_algorithm="SVTR"
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,64,256" --rec_algorithm="SVTR"
 --use_gpu:True|False
 --enable_mkldnn:False
 --cpu_threads:6
@@ -50,4 +50,4 @@ inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbo
 --benchmark:True
 null:null
 ===========================infer_benchmark_params==========================
-random_infer_input:[{float32,[1=3,64,256]}]
+random_infer_input:[{float32,[3,64,256]}]
diff --git a/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
index 4e8151f599..a0aed48875 100644
--- a/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
+++ b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
@@ -23,7 +23,7 @@ Global:
 
 Optimizer:
   name: Adadelta
-  epsilon: 0.00000001
+  epsilon: 1.e-8
   rho: 0.95
   clip_norm: 5.0
   lr:
@@ -46,6 +46,7 @@ Loss:
   name: CELoss
   smoothing: False
   with_all: True
+  ignore_index: &ignore_index 0 # Must be zero or greater than the number of character classes
 
 PostProcess:
   name: ViTSTRLabelDecode
@@ -64,6 +65,7 @@ Train:
           img_mode: BGR
           channel_first: False
       - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
       - GrayRecResizeImg:
           image_shape: [224, 224] # W H
           resize_type: PIL # PIL or OpenCV
@@ -87,6 +89,7 @@ Eval:
           img_mode: BGR
           channel_first: False
       - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
       - GrayRecResizeImg:
           image_shape: [224, 224] # W H
           resize_type: PIL # PIL or OpenCV
diff --git a/tools/export_model.py b/tools/export_model.py
index 7ef2c36b12..e2673239b1 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -79,6 +79,19 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
                 shape=[None, 1, 224, 224], dtype="float32"),
         ]
         model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "ABINet":
+        other_shape = [
+            paddle.static.InputSpec(
+                shape=[None, 3, 32, 128], dtype="float32"),
+        ]
+        # print([None, 3, 32, 128])
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "NRTR":
+        other_shape = [
+            paddle.static.InputSpec(
+                shape=[None, 1, 32, 100], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
     else:
         infer_shape = [3, -1, -1]
         if arch_config["model_type"] == "rec":
@@ -90,8 +103,6 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
                     "When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training"
                 )
                 infer_shape[-1] = 100
-            if arch_config["algorithm"] == "NRTR":
-                infer_shape = [1, 32, 100]
         elif arch_config["model_type"] == "table":
             infer_shape = [3, 488, 488]
         model = to_static(
diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py
index 1945667972..a95f555966 100755
--- a/tools/infer/predict_rec.py
+++ b/tools/infer/predict_rec.py
@@ -75,6 +75,12 @@ class TextRecognizer(object):
                 "character_dict_path": args.rec_char_dict_path,
                 "use_space_char": args.use_space_char
             }
+        elif self.rec_algorithm == 'ABINet':
+            postprocess_params = {
+                'name': 'ABINetLabelDecode',
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char
+            }
         self.postprocess_op = build_post_process(postprocess_params)
         self.predictor, self.input_tensor, self.output_tensors, self.config = \
             utility.create_predictor(args, 'rec', logger)
@@ -145,17 +151,6 @@ class TextRecognizer(object):
         padding_im[:, :, 0:resized_w] = resized_image
         return padding_im
 
-    def resize_norm_img_svtr(self, img, image_shape):
-
-        imgC, imgH, imgW = image_shape
-        resized_image = cv2.resize(
-            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
-        resized_image = resized_image.astype('float32')
-        resized_image = resized_image.transpose((2, 0, 1)) / 255
-        resized_image -= 0.5
-        resized_image /= 0.5
-        return resized_image
-
     def resize_norm_img_srn(self, img, image_shape):
         imgC, imgH, imgW = image_shape
 
@@ -263,6 +258,35 @@ class TextRecognizer(object):
 
         return padding_im, resize_shape, pad_shape, valid_ratio
 
+    def resize_norm_img_svtr(self, img, image_shape):
+
+        imgC, imgH, imgW = image_shape
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype('float32')
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        return resized_image
+
+    def resize_norm_img_abinet(self, img, image_shape):
+
+        imgC, imgH, imgW = image_shape
+
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype('float32')
+        resized_image = resized_image / 255.
+
+        mean = np.array([0.485, 0.456, 0.406])
+        std = np.array([0.229, 0.224, 0.225])
+        resized_image = (
+            resized_image - mean[None, None, ...]) / std[None, None, ...]
+        resized_image = resized_image.transpose((2, 0, 1))
+        resized_image = resized_image.astype('float32')
+
+        return resized_image
+
     def __call__(self, img_list):
         img_num = len(img_list)
         # Calculate the aspect ratio of all text bars
@@ -313,6 +337,11 @@ class TextRecognizer(object):
                                                          self.rec_image_shape)
                     norm_img = norm_img[np.newaxis, :]
                     norm_img_batch.append(norm_img)
+                elif self.rec_algorithm == "ABINet":
+                    norm_img = self.resize_norm_img_abinet(
+                        img_list[indices[ino]], self.rec_image_shape)
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
                 else:
                     norm_img = self.resize_norm_img(img_list[indices[ino]],
                                                     max_wh_ratio)
diff --git a/tools/program.py b/tools/program.py
index 745c28b872..f61d3f9af7 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -575,7 +575,7 @@ def preprocess(is_train=False):
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
         'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
-        'ViTSTR'
+        'ViTSTR', 'ABINet'
     ]
 
     if use_xpu:

From 37f22e3e33400599fa0856777b94ab6af0e2f31b Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Thu, 23 Jun 2022 05:03:15 +0000
Subject: [PATCH 045/134] add rec_resnet45

---
 ppocr/modeling/backbones/__init__.py      |   3 +-
 ppocr/modeling/backbones/rec_resnet.py    | 280 ----------------------
 ppocr/modeling/backbones/rec_resnet_45.py | 147 ++++++++++++
 3 files changed, 149 insertions(+), 281 deletions(-)
 delete mode 100644 ppocr/modeling/backbones/rec_resnet.py
 create mode 100644 ppocr/modeling/backbones/rec_resnet_45.py

diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
index 56d2c2b9db..f8959e263e 100755
--- a/ppocr/modeling/backbones/__init__.py
+++ b/ppocr/modeling/backbones/__init__.py
@@ -27,7 +27,8 @@ def build_backbone(config, model_type):
         from .rec_resnet_fpn import ResNetFPN
         from .rec_mv1_enhance import MobileNetV1Enhance
         from .rec_nrtr_mtb import MTB
-        from .rec_resnet import ResNet31, ResNet45
+        from .rec_resnet_31 import ResNet31
+        from .rec_resnet_45 import ResNet45
         from .rec_resnet_aster import ResNet_ASTER
         from .rec_micronet import MicroNet
         from .rec_efficientb3_pren import EfficientNetb3_PREN
diff --git a/ppocr/modeling/backbones/rec_resnet.py b/ppocr/modeling/backbones/rec_resnet.py
deleted file mode 100644
index ce76b642cc..0000000000
--- a/ppocr/modeling/backbones/rec_resnet.py
+++ /dev/null
@@ -1,280 +0,0 @@
-# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#    http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""
-This code is refer from: 
-https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/layers/conv_layer.py
-https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/backbones/resnet31_ocr.py
-"""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import paddle
-from paddle import ParamAttr
-from paddle.nn.initializer import KaimingNormal
-import paddle.nn as nn
-import paddle.nn.functional as F
-import numpy as np
-import math
-
-__all__ = ["ResNet31", "ResNet45"]
-
-
-def conv1x1(in_planes, out_planes, stride=1):
-    return nn.Conv2D(
-        in_planes,
-        out_planes,
-        kernel_size=1,
-        stride=stride,
-        weight_attr=ParamAttr(initializer=KaimingNormal()),
-        bias_attr=False)
-
-
-def conv3x3(in_channel, out_channel, stride=1):
-    return nn.Conv2D(
-        in_channel,
-        out_channel,
-        kernel_size=3,
-        stride=stride,
-        padding=1,
-        weight_attr=ParamAttr(initializer=KaimingNormal()),
-        bias_attr=False)
-
-
-class BasicBlock(nn.Layer):
-    expansion = 1
-
-    def __init__(self, in_channels, channels, stride=1, downsample=None):
-        super().__init__()
-        self.conv1 = conv1x1(in_channels, channels)
-        self.bn1 = nn.BatchNorm2D(channels)
-        self.relu = nn.ReLU()
-        self.conv2 = conv3x3(channels, channels, stride)
-        self.bn2 = nn.BatchNorm2D(channels)
-        self.downsample = downsample
-        self.stride = stride
-
-    def forward(self, x):
-        residual = x
-
-        out = self.conv1(x)
-        out = self.bn1(out)
-        out = self.relu(out)
-
-        out = self.conv2(out)
-        out = self.bn2(out)
-
-        if self.downsample is not None:
-            residual = self.downsample(x)
-        out += residual
-        out = self.relu(out)
-
-        return out
-
-
-class ResNet31(nn.Layer):
-    '''
-    Args:
-        in_channels (int): Number of channels of input image tensor.
-        layers (list[int]): List of BasicBlock number for each stage.
-        channels (list[int]): List of out_channels of Conv2d layer.
-        out_indices (None | Sequence[int]): Indices of output stages.
-        last_stage_pool (bool): If True, add `MaxPool2d` layer to last stage.
-    '''
-
-    def __init__(self,
-                 in_channels=3,
-                 layers=[1, 2, 5, 3],
-                 channels=[64, 128, 256, 256, 512, 512, 512],
-                 out_indices=None,
-                 last_stage_pool=False):
-        super(ResNet31, self).__init__()
-        assert isinstance(in_channels, int)
-        assert isinstance(last_stage_pool, bool)
-
-        self.out_indices = out_indices
-        self.last_stage_pool = last_stage_pool
-
-        # conv 1 (Conv Conv)
-        self.conv1_1 = nn.Conv2D(
-            in_channels, channels[0], kernel_size=3, stride=1, padding=1)
-        self.bn1_1 = nn.BatchNorm2D(channels[0])
-        self.relu1_1 = nn.ReLU()
-
-        self.conv1_2 = nn.Conv2D(
-            channels[0], channels[1], kernel_size=3, stride=1, padding=1)
-        self.bn1_2 = nn.BatchNorm2D(channels[1])
-        self.relu1_2 = nn.ReLU()
-
-        # conv 2 (Max-pooling, Residual block, Conv)
-        self.pool2 = nn.MaxPool2D(
-            kernel_size=2, stride=2, padding=0, ceil_mode=True)
-        self.block2 = self._make_layer(channels[1], channels[2], layers[0])
-        self.conv2 = nn.Conv2D(
-            channels[2], channels[2], kernel_size=3, stride=1, padding=1)
-        self.bn2 = nn.BatchNorm2D(channels[2])
-        self.relu2 = nn.ReLU()
-
-        # conv 3 (Max-pooling, Residual block, Conv)
-        self.pool3 = nn.MaxPool2D(
-            kernel_size=2, stride=2, padding=0, ceil_mode=True)
-        self.block3 = self._make_layer(channels[2], channels[3], layers[1])
-        self.conv3 = nn.Conv2D(
-            channels[3], channels[3], kernel_size=3, stride=1, padding=1)
-        self.bn3 = nn.BatchNorm2D(channels[3])
-        self.relu3 = nn.ReLU()
-
-        # conv 4 (Max-pooling, Residual block, Conv)
-        self.pool4 = nn.MaxPool2D(
-            kernel_size=(2, 1), stride=(2, 1), padding=0, ceil_mode=True)
-        self.block4 = self._make_layer(channels[3], channels[4], layers[2])
-        self.conv4 = nn.Conv2D(
-            channels[4], channels[4], kernel_size=3, stride=1, padding=1)
-        self.bn4 = nn.BatchNorm2D(channels[4])
-        self.relu4 = nn.ReLU()
-
-        # conv 5 ((Max-pooling), Residual block, Conv)
-        self.pool5 = None
-        if self.last_stage_pool:
-            self.pool5 = nn.MaxPool2D(
-                kernel_size=2, stride=2, padding=0, ceil_mode=True)
-        self.block5 = self._make_layer(channels[4], channels[5], layers[3])
-        self.conv5 = nn.Conv2D(
-            channels[5], channels[5], kernel_size=3, stride=1, padding=1)
-        self.bn5 = nn.BatchNorm2D(channels[5])
-        self.relu5 = nn.ReLU()
-
-        self.out_channels = channels[-1]
-
-    def _make_layer(self, input_channels, output_channels, blocks):
-        layers = []
-        for _ in range(blocks):
-            downsample = None
-            if input_channels != output_channels:
-                downsample = nn.Sequential(
-                    nn.Conv2D(
-                        input_channels,
-                        output_channels,
-                        kernel_size=1,
-                        stride=1,
-                        weight_attr=ParamAttr(initializer=KaimingNormal()),
-                        bias_attr=False),
-                    nn.BatchNorm2D(output_channels), )
-
-            layers.append(
-                BasicBlock(
-                    input_channels, output_channels, downsample=downsample))
-            input_channels = output_channels
-        return nn.Sequential(*layers)
-
-    def forward(self, x):
-        x = self.conv1_1(x)
-        x = self.bn1_1(x)
-        x = self.relu1_1(x)
-
-        x = self.conv1_2(x)
-        x = self.bn1_2(x)
-        x = self.relu1_2(x)
-
-        outs = []
-        for i in range(4):
-            layer_index = i + 2
-            pool_layer = getattr(self, f'pool{layer_index}')
-            block_layer = getattr(self, f'block{layer_index}')
-            conv_layer = getattr(self, f'conv{layer_index}')
-            bn_layer = getattr(self, f'bn{layer_index}')
-            relu_layer = getattr(self, f'relu{layer_index}')
-
-            if pool_layer is not None:
-                x = pool_layer(x)
-            x = block_layer(x)
-            x = conv_layer(x)
-            x = bn_layer(x)
-            x = relu_layer(x)
-
-            outs.append(x)
-
-        if self.out_indices is not None:
-            return tuple([outs[i] for i in self.out_indices])
-
-        return x
-
-
-class ResNet(nn.Layer):
-    def __init__(self, block, layers, in_channels=3):
-        self.inplanes = 32
-        super(ResNet, self).__init__()
-        self.conv1 = nn.Conv2D(
-            3,
-            32,
-            kernel_size=3,
-            stride=1,
-            padding=1,
-            weight_attr=ParamAttr(initializer=KaimingNormal()),
-            bias_attr=False)
-        self.bn1 = nn.BatchNorm2D(32)
-        self.relu = nn.ReLU()
-
-        self.layer1 = self._make_layer(block, 32, layers[0], stride=2)
-        self.layer2 = self._make_layer(block, 64, layers[1], stride=1)
-        self.layer3 = self._make_layer(block, 128, layers[2], stride=2)
-        self.layer4 = self._make_layer(block, 256, layers[3], stride=1)
-        self.layer5 = self._make_layer(block, 512, layers[4], stride=1)
-        self.out_channels = 512
-
-        # for m in self.modules():
-        #     if isinstance(m, nn.Conv2D):
-        #         n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
-        #         m.weight.data.normal_(0, math.sqrt(2. / n))
-
-    def _make_layer(self, block, planes, blocks, stride=1):
-        downsample = None
-        if stride != 1 or self.inplanes != planes * block.expansion:
-            # downsample = True
-            downsample = nn.Sequential(
-                nn.Conv2D(
-                    self.inplanes,
-                    planes * block.expansion,
-                    kernel_size=1,
-                    stride=stride,
-                    weight_attr=ParamAttr(initializer=KaimingNormal()),
-                    bias_attr=False),
-                nn.BatchNorm2D(planes * block.expansion), )
-
-        layers = []
-        layers.append(block(self.inplanes, planes, stride, downsample))
-        self.inplanes = planes * block.expansion
-        for i in range(1, blocks):
-            layers.append(block(self.inplanes, planes))
-
-        return nn.Sequential(*layers)
-
-    def forward(self, x):
-
-        x = self.conv1(x)
-        x = self.bn1(x)
-        x = self.relu(x)
-        # print(x)
-        x = self.layer1(x)
-        x = self.layer2(x)
-        x = self.layer3(x)
-        # print(x)
-        x = self.layer4(x)
-        x = self.layer5(x)
-        return x
-
-
-def ResNet45(in_channels=3):
-    return ResNet(BasicBlock, [3, 4, 6, 6, 3], in_channels=in_channels)
diff --git a/ppocr/modeling/backbones/rec_resnet_45.py b/ppocr/modeling/backbones/rec_resnet_45.py
new file mode 100644
index 0000000000..9093d0bc99
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_45.py
@@ -0,0 +1,147 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/FangShancheng/ABINet/tree/main/modules
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+import math
+
+__all__ = ["ResNet45"]
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    return nn.Conv2D(
+        in_planes,
+        out_planes,
+        kernel_size=1,
+        stride=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False)
+
+
+def conv3x3(in_channel, out_channel, stride=1):
+    return nn.Conv2D(
+        in_channel,
+        out_channel,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False)
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self, in_channels, channels, stride=1, downsample=None):
+        super().__init__()
+        self.conv1 = conv1x1(in_channels, channels)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(channels, channels, stride)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet45(nn.Layer):
+    def __init__(self, block=BasicBlock, layers=[3, 4, 6, 6, 3], in_channels=3):
+        self.inplanes = 32
+        super(ResNet45, self).__init__()
+        self.conv1 = nn.Conv2D(
+            3,
+            32,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False)
+        self.bn1 = nn.BatchNorm2D(32)
+        self.relu = nn.ReLU()
+
+        self.layer1 = self._make_layer(block, 32, layers[0], stride=2)
+        self.layer2 = self._make_layer(block, 64, layers[1], stride=1)
+        self.layer3 = self._make_layer(block, 128, layers[2], stride=2)
+        self.layer4 = self._make_layer(block, 256, layers[3], stride=1)
+        self.layer5 = self._make_layer(block, 512, layers[4], stride=1)
+        self.out_channels = 512
+
+        # for m in self.modules():
+        #     if isinstance(m, nn.Conv2D):
+        #         n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+        #         m.weight.data.normal_(0, math.sqrt(2. / n))
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            # downsample = True
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    weight_attr=ParamAttr(initializer=KaimingNormal()),
+                    bias_attr=False),
+                nn.BatchNorm2D(planes * block.expansion), )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        # print(x)
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        # print(x)
+        x = self.layer4(x)
+        x = self.layer5(x)
+        return x

From 0a343fd3aae14b866281d897474d0669b1b26282 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Thu, 23 Jun 2022 08:17:00 +0000
Subject: [PATCH 046/134] svtr ch large model

---
 configs/rec/rec_svtrnet.yml         |   4 +-
 configs/rec/rec_svtrnet_ch.yml      | 159 ++++++++++++++++++++++++++++
 doc/doc_ch/algorithm_rec_svtr.md    |   1 -
 doc/doc_en/algorithm_rec_svtr_en.md |   1 -
 ppocr/data/imaug/__init__.py        |   3 +-
 ppocr/data/imaug/rec_img_aug.py     | 124 +++++++++++++---------
 tools/export_model.py               |  19 +++-
 7 files changed, 252 insertions(+), 59 deletions(-)
 create mode 100644 configs/rec/rec_svtrnet_ch.yml

diff --git a/configs/rec/rec_svtrnet.yml b/configs/rec/rec_svtrnet.yml
index a3d292b6ba..5896b10515 100644
--- a/configs/rec/rec_svtrnet.yml
+++ b/configs/rec/rec_svtrnet.yml
@@ -83,7 +83,7 @@ Train:
           img_mode: BGR
           channel_first: False
       - CTCLabelEncode: # Class handling label
-      - RecResizeImg:
+      - SVTRRecResizeImg:
           character_dict_path:
           image_shape: [3, 64, 256]
           padding: False
@@ -104,7 +104,7 @@ Eval:
           img_mode: BGR
           channel_first: False
       - CTCLabelEncode: # Class handling label
-      - RecResizeImg:
+      - SVTRRecResizeImg:
           character_dict_path:
           image_shape: [3, 64, 256]
           padding: False
diff --git a/configs/rec/rec_svtrnet_ch.yml b/configs/rec/rec_svtrnet_ch.yml
new file mode 100644
index 0000000000..b327edf499
--- /dev/null
+++ b/configs/rec/rec_svtrnet_ch.yml
@@ -0,0 +1,159 @@
+Global:
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr_ch_all/
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  save_res_path: ./output/rec/predicts_svtr_tiny_ch_all.txt
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 8.0e-08
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform: null
+  Backbone:
+    name: SVTRNet
+    img_size:
+    - 32
+    - 320
+    out_char_num: 40
+    out_channels: 96
+    patch_merging: Conv
+    embed_dim:
+    - 64
+    - 128
+    - 256
+    depth:
+    - 3
+    - 6
+    - 3
+    num_heads:
+    - 2
+    - 4
+    - 8
+    mixer:
+    - Local
+    - Local
+    - Local
+    - Local
+    - Local
+    - Local
+    - Global
+    - Global
+    - Global
+    - Global
+    - Global
+    - Global
+    local_mixer:
+    - - 7
+      - 11
+    - - 7
+      - 11
+    - - 7
+      - 11
+    last_stage: true
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    label_file_list:
+    - /paddle/data/ocr_all/train_all_list.txt
+    data_dir: /paddle/data/ocr_all
+    ext_op_transform_idx: 1
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape:
+        - 32
+        - 320
+        - 3
+    - RecAug: null
+    - CTCLabelEncode: null
+    - SVTRRecResizeImg:
+        character_dict_path: null
+        infer_mode: False
+        image_shape:
+        - 3
+        - 32
+        - 320
+        padding: true
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /paddle/data/ocr_all
+    label_file_list:
+    - /paddle/data/ocr_all/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - SVTRRecResizeImg:
+        character_dict_path: null
+        infer_mode: False
+        image_shape:
+        - 3
+        - 32
+        - 320
+        padding: true
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 2
+profiler_options: null
diff --git a/doc/doc_ch/algorithm_rec_svtr.md b/doc/doc_ch/algorithm_rec_svtr.md
index 41a22ca65c..c0e26433e9 100644
--- a/doc/doc_ch/algorithm_rec_svtr.md
+++ b/doc/doc_ch/algorithm_rec_svtr.md
@@ -111,7 +111,6 @@ python3 tools/export_model.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny
 
 **注意：**
 - 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
-- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应SVTR的`infer_shape`。
 
 转换成功后，在目录下有三个文件：
 ```
diff --git a/doc/doc_en/algorithm_rec_svtr_en.md b/doc/doc_en/algorithm_rec_svtr_en.md
index d402a6b491..37cd35f35a 100644
--- a/doc/doc_en/algorithm_rec_svtr_en.md
+++ b/doc/doc_en/algorithm_rec_svtr_en.md
@@ -88,7 +88,6 @@ python3 tools/export_model.py -c configs/rec/rec_svtrnet.yml -o Global.pretraine
 
 **Note:**
 - If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
-- If you modified the input size during training, please modify the `infer_shape` corresponding to SVTR in the `tools/export_model.py` file.
 
 After the conversion is successful, there are three files in the directory:
 ```
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index 58e8a5c780..437e0152fa 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -24,7 +24,8 @@ from .make_pse_gt import MakePseGt
 
 
 from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
-    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, ABINetRecResizeImg
+    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, \
+    ABINetRecResizeImg, SVTRRecResizeImg
 from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index a8b3b813b3..874d9aa084 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -148,57 +148,6 @@ class ClsResizeImg(object):
         return data
 
 
-class GrayRecResizeImg(object):
-    def __init__(self,
-                 image_shape,
-                 resize_type,
-                 inter_type='Image.ANTIALIAS',
-                 scale=True,
-                 padding=False,
-                 **kwargs):
-        self.image_shape = image_shape
-        self.resize_type = resize_type
-        self.padding = padding
-        self.inter_type = eval(inter_type)
-        self.scale = scale
-
-    def __call__(self, data):
-        img = data['image']
-        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        image_shape = self.image_shape
-        if self.padding:
-            imgC, imgH, imgW = image_shape
-            # todo: change to 0 and modified image shape
-            h = img.shape[0]
-            w = img.shape[1]
-            ratio = w / float(h)
-            if math.ceil(imgH * ratio) > imgW:
-                resized_w = imgW
-            else:
-                resized_w = int(math.ceil(imgH * ratio))
-            resized_image = cv2.resize(img, (resized_w, imgH))
-            norm_img = np.expand_dims(resized_image, -1)
-            norm_img = norm_img.transpose((2, 0, 1))
-            resized_image = norm_img.astype(np.float32) / 128. - 1.
-            padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
-            padding_im[:, :, 0:resized_w] = resized_image
-            data['image'] = padding_im
-            return data
-        if self.resize_type == 'PIL':
-            image_pil = Image.fromarray(np.uint8(img))
-            img = image_pil.resize(self.image_shape, self.inter_type)
-            img = np.array(img)
-        if self.resize_type == 'OpenCV':
-            img = cv2.resize(img, self.image_shape)
-        norm_img = np.expand_dims(img, -1)
-        norm_img = norm_img.transpose((2, 0, 1))
-        if self.scale:
-            data['image'] = norm_img.astype(np.float32) / 128. - 1.
-        else:
-            data['image'] = norm_img.astype(np.float32) / 255.
-        return data
-
-
 class RecResizeImg(object):
     def __init__(self,
                  image_shape,
@@ -279,6 +228,57 @@ class PRENResizeImg(object):
         return data
 
 
+class GrayRecResizeImg(object):
+    def __init__(self,
+                 image_shape,
+                 resize_type,
+                 inter_type='Image.ANTIALIAS',
+                 scale=True,
+                 padding=False,
+                 **kwargs):
+        self.image_shape = image_shape
+        self.resize_type = resize_type
+        self.padding = padding
+        self.inter_type = eval(inter_type)
+        self.scale = scale
+
+    def __call__(self, data):
+        img = data['image']
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        image_shape = self.image_shape
+        if self.padding:
+            imgC, imgH, imgW = image_shape
+            # todo: change to 0 and modified image shape
+            h = img.shape[0]
+            w = img.shape[1]
+            ratio = w / float(h)
+            if math.ceil(imgH * ratio) > imgW:
+                resized_w = imgW
+            else:
+                resized_w = int(math.ceil(imgH * ratio))
+            resized_image = cv2.resize(img, (resized_w, imgH))
+            norm_img = np.expand_dims(resized_image, -1)
+            norm_img = norm_img.transpose((2, 0, 1))
+            resized_image = norm_img.astype(np.float32) / 128. - 1.
+            padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+            padding_im[:, :, 0:resized_w] = resized_image
+            data['image'] = padding_im
+            return data
+        if self.resize_type == 'PIL':
+            image_pil = Image.fromarray(np.uint8(img))
+            img = image_pil.resize(self.image_shape, self.inter_type)
+            img = np.array(img)
+        if self.resize_type == 'OpenCV':
+            img = cv2.resize(img, self.image_shape)
+        norm_img = np.expand_dims(img, -1)
+        norm_img = norm_img.transpose((2, 0, 1))
+        if self.scale:
+            data['image'] = norm_img.astype(np.float32) / 128. - 1.
+        else:
+            data['image'] = norm_img.astype(np.float32) / 255.
+        return data
+
+
 class ABINetRecResizeImg(object):
     def __init__(self,
                  image_shape,
@@ -297,6 +297,28 @@ class ABINetRecResizeImg(object):
         return data
 
 
+class SVTRRecResizeImg(object):
+    def __init__(self,
+                 image_shape,
+                 infer_mode=False,
+                 character_dict_path='./ppocr/utils/ppocr_keys_v1.txt',
+                 padding=True,
+                 **kwargs):
+        self.image_shape = image_shape
+        self.infer_mode = infer_mode
+        self.character_dict_path = character_dict_path
+        self.padding = padding
+
+    def __call__(self, data):
+        img = data['image']
+
+        norm_img, valid_ratio = resize_norm_img(img, self.image_shape,
+                                                self.padding)
+        data['image'] = norm_img
+        data['valid_ratio'] = valid_ratio
+        return data
+
+
 def resize_norm_img_sar(img, image_shape, width_downsample_ratio=0.25):
     imgC, imgH, imgW_min, imgW_max = image_shape
     h = img.shape[0]
diff --git a/tools/export_model.py b/tools/export_model.py
index e2673239b1..b10d41d5b2 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -31,7 +31,12 @@ from ppocr.utils.logging import get_logger
 from tools.program import load_config, merge_config, ArgsParser
 
 
-def export_single_model(model, arch_config, save_path, logger, quanter=None):
+def export_single_model(model,
+                        arch_config,
+                        save_path,
+                        logger,
+                        input_shape=None,
+                        quanter=None):
     if arch_config["algorithm"] == "SRN":
         max_text_length = arch_config["Head"]["max_text_length"]
         other_shape = [
@@ -64,7 +69,7 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
         else:
             other_shape = [
                 paddle.static.InputSpec(
-                    shape=[None, 3, 64, 256], dtype="float32"),
+                    shape=[None] + input_shape, dtype="float32"),
             ]
         model = to_static(model, input_spec=other_shape)
     elif arch_config["algorithm"] == "PREN":
@@ -174,6 +179,13 @@ def main():
 
     arch_config = config["Architecture"]
 
+    if arch_config["algorithm"] == "SVTR" and arch_config["Head"][
+            "name"] != 'MultiHead':
+        input_shape = config["Eval"]["dataset"]["transforms"][-2][
+            'SVTRRecResizeImg']['image_shape']
+    else:
+        input_shape = None
+
     if arch_config["algorithm"] in ["Distillation", ]:  # distillation model
         archs = list(arch_config["Models"].values())
         for idx, name in enumerate(model.model_name_list):
@@ -182,7 +194,8 @@ def main():
                                 sub_model_save_path, logger)
     else:
         save_path = os.path.join(save_path, "inference")
-        export_single_model(model, arch_config, save_path, logger)
+        export_single_model(
+            model, arch_config, save_path, logger, input_shape=input_shape)
 
 
 if __name__ == "__main__":

From 31ab300ca258243a311ae08e1496faa789567a0c Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Thu, 23 Jun 2022 13:09:48 +0000
Subject: [PATCH 047/134] [application] svtr ch model

---
 applications/高精度中文识别模型.md | 101 +++++++++++++++++++++++++++++
 1 file changed, 101 insertions(+)
 create mode 100644 applications/高精度中文识别模型.md

diff --git a/applications/高精度中文识别模型.md b/applications/高精度中文识别模型.md
new file mode 100644
index 0000000000..27b63abf3c
--- /dev/null
+++ b/applications/高精度中文识别模型.md
@@ -0,0 +1,101 @@
+# 高精度中文场景文本识别模型SVTR
+
+## 1. 简介
+
+PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中超轻量的场景中文识别模型SVTR_LCNet使用了SVTR算法结构。为了保证速度，SVTR_LCNet将SVTR模型的Local Blocks替换为LCNet，使用两层Global Blocks。在中文场景中，经过多次优化，SVTR_LCNet的最终精度为79.4%。具体的：
+- GTC：Attention指导CTC训练策略；
+- TextConAug：挖掘文字上下文信息的数据增广策略；
+- TextRotNet：自监督的预训练模型；
+- UDML：联合互学习策略；
+- UIM：无标注数据挖掘方案。
+
+其中 *UIM：无标注数据挖掘方案* 使用了高精度的SVTR中文模型，该模型在PP-OCRv3识别的数据集上训练，精度为82.5%。本项目主要介绍SVTR中文模型如何使用。
+
+aistudio项目链接: [高精度中文场景文本识别模型SVTR](https://aistudio.baidu.com/aistudio/projectdetail/4263032)
+
+## 2. SVTR中文模型使用
+
+### 环境准备
+
+
+本任务基于Aistudio完成, 具体环境如下：
+
+- 操作系统: Linux
+- PaddlePaddle: 2.3
+- PaddleOCR: dygraph
+
+下载 PaddleOCR代码
+
+```bash
+git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR
+```
+
+安装依赖库
+
+```bash
+pip install -r PaddleOCR/requirements.txt -i https://mirror.baidu.com/pypi/simple
+```
+
+### 快速使用
+
+获取SVTR中文模型文件，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+```bash
+# 解压模型文件
+tar xf svtr_ch_high_accuracy.tar
+```
+
+预测中文文本，以下图为例：
+![](https://ai-studio-static-online.cdn.bcebos.com/511f6d65f0e84ccc8c5668eafbe7feb73b26c40befec4bd58dde64b59bf340a1)
+
+预测命令：
+
+```bash
+# CPU预测
+python tools/infer_rec.py -c configs/rec/rec_svtrnet_ch.yml -o Global.pretrained_model=./svtr_ch_high_accuracy/best_accuracy Global.infer_img=./doc/imgs_words/ch/word_1.jpg Global.use_gpu=False
+
+# GPU预测
+#python tools/infer_rec.py -c configs/rec/rec_svtrnet_ch.yml -o Global.pretrained_model=./svtr_ch_high_accuracy/best_accuracy Global.infer_img=./doc/imgs_words/ch/word_1.jpg Global.use_gpu=True
+```
+
+可以看到最后打印结果为
+- result: 韩国小馆    0.9853458404541016
+
+0.9853458404541016为预测置信度。
+
+### 推理模型导出与预测
+
+inference 模型（paddle.jit.save保存的模型） 一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。 训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。 与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+运行识别模型转inference模型命令，如下：
+
+```bash
+python tools/export_model.py -c configs/rec/rec_svtrnet_ch.yml -o Global.pretrained_model=./svtr_ch_high_accuracy/best_accuracy Global.save_inference_dir=./inference/svtr_ch
+```
+
+转换成功后，在目录下有三个文件：
+```shell
+inference/svtr_ch/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+inference模型预测，命令如下：
+
+```bash
+# CPU预测
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_1.jpg" --rec_algorithm='SVTR' --rec_model_dir=./inference/svtr_ch/ --rec_image_shape='3, 32, 320'  --rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt --use_gpu=False
+
+# GPU预测
+#python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_1.jpg" --rec_algorithm='SVTR' --rec_model_dir=./inference/svtr_ch/ --rec_image_shape='3, 32, 320'  --rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt --use_gpu=True
+```
+
+**注意**
+
+- 使用PP-OCRv3识别SVTR_LCNet进行推理时，不需要使用--rec_algorithm指定算法名称，使用默认的推理方式即为PP-OCRv3识别的推理过程。但是使用SVTR算法时，需要指定--rec_algorithm='SVTR'
+- 如果使用自定义字典训练的模型，需要将--rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt修改为自定义的字典
+- --rec_image_shape='3, 32, 320' 该参数不能去掉

From 6ef2c8762ed7ec3a90a6680292a6095236428103 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Thu, 23 Jun 2022 13:25:31 +0000
Subject: [PATCH 048/134] [application] svtr ch model

---
 applications/高精度中文识别模型.md | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/applications/高精度中文识别模型.md b/applications/高精度中文识别模型.md
index 27b63abf3c..31d68abbff 100644
--- a/applications/高精度中文识别模型.md
+++ b/applications/高精度中文识别模型.md
@@ -9,7 +9,12 @@ PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中
 - UDML：联合互学习策略；
 - UIM：无标注数据挖掘方案。
 
-其中 *UIM：无标注数据挖掘方案* 使用了高精度的SVTR中文模型，该模型在PP-OCRv3识别的数据集上训练，精度为82.5%。本项目主要介绍SVTR中文模型如何使用。
+其中 *UIM：无标注数据挖掘方案* 使用了高精度的SVTR中文模型进行无标注文件的刷库，该模型在PP-OCRv3识别的数据集上训练，精度对比如下表。
+
+|中文识别算法|模型|精度|
+| --- | --- | --- |
+|PP-OCRv3|SVTR_LCNet|79.4%|
+|SVTR|SVTR-Tiny|82.5%|
 
 aistudio项目链接: [高精度中文场景文本识别模型SVTR](https://aistudio.baidu.com/aistudio/projectdetail/4263032)
 
@@ -96,6 +101,6 @@ python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_1.jpg"
 
 **注意**
 
-- 使用PP-OCRv3识别SVTR_LCNet进行推理时，不需要使用--rec_algorithm指定算法名称，使用默认的推理方式即为PP-OCRv3识别的推理过程。但是使用SVTR算法时，需要指定--rec_algorithm='SVTR'
+- 使用SVTR算法时，需要指定--rec_algorithm='SVTR'
 - 如果使用自定义字典训练的模型，需要将--rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt修改为自定义的字典
 - --rec_image_shape='3, 32, 320' 该参数不能去掉

From 1fc0655d857f25704a54e6550a37c7c68f7d5fdd Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Thu, 23 Jun 2022 13:31:57 +0000
Subject: [PATCH 049/134] [application] svtr ch model

---
 applications/高精度中文识别模型.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/applications/高精度中文识别模型.md b/applications/高精度中文识别模型.md
index 31d68abbff..e3d0693ece 100644
--- a/applications/高精度中文识别模型.md
+++ b/applications/高精度中文识别模型.md
@@ -54,7 +54,7 @@ tar xf svtr_ch_high_accuracy.tar
 ```
 
 预测中文文本，以下图为例：
-![](https://ai-studio-static-online.cdn.bcebos.com/511f6d65f0e84ccc8c5668eafbe7feb73b26c40befec4bd58dde64b59bf340a1)
+![](../doc/imgs_words/ch/word_1.jpg)
 
 预测命令：
 

From 773b687f5a5a6d3115e27ac15070fd13ea6bb5ce Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Fri, 24 Jun 2022 07:25:50 +0000
Subject: [PATCH 050/134] add abinet_rec_aug and trained model

---
 configs/rec/rec_r45_abinet.yml                |   3 +-
 doc/doc_ch/algorithm_rec_abinet.md            |   4 +-
 doc/doc_en/algorithm_rec_abinet_en.md         |   4 +-
 ppocr/data/imaug/__init__.py                  |   2 +-
 ppocr/data/imaug/abinet_aug.py                | 407 ++++++++++++++++++
 ppocr/data/imaug/rec_img_aug.py               |  27 ++
 .../configs/rec_r45_abinet/rec_r45_abinet.yml |   1 +
 7 files changed, 442 insertions(+), 6 deletions(-)
 create mode 100644 ppocr/data/imaug/abinet_aug.py

diff --git a/configs/rec/rec_r45_abinet.yml b/configs/rec/rec_r45_abinet.yml
index a756fead9f..3cf4cddb04 100644
--- a/configs/rec/rec_r45_abinet.yml
+++ b/configs/rec/rec_r45_abinet.yml
@@ -8,7 +8,7 @@ Global:
   # evaluation is run every 2000 iterations
   eval_batch_step: [0, 2000]
   cal_metric_during_train: True
-  pretrained_model:
+  pretrained_model: ./rec_r45_abinet_train/abinet_vl_pretrained
   checkpoints:
   save_inference_dir:
   use_visualdl: False
@@ -67,6 +67,7 @@ Train:
       - DecodeImage: # load image
           img_mode: RGB
           channel_first: False
+      - ABINetRecAug:
       - ABINetLabelEncode: # Class handling label
           ignore_index: *ignore_index
       - ABINetRecResizeImg:
diff --git a/doc/doc_ch/algorithm_rec_abinet.md b/doc/doc_ch/algorithm_rec_abinet.md
index d20c703014..47507c36c7 100644
--- a/doc/doc_ch/algorithm_rec_abinet.md
+++ b/doc/doc_ch/algorithm_rec_abinet.md
@@ -27,7 +27,7 @@
 
 |模型|骨干网络|配置文件|Acc|下载链接|
 | --- | --- | --- | --- | --- |
-|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[训练模型]()/[预训练模型]|
+|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[预训练、训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)|
 
 <a name="2"></a>
 ## 2. 环境配置
@@ -80,7 +80,7 @@ python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img
 
 <a name="4-1"></a>
 ### 4.1 Python推理
-首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址]() )，可以使用如下命令进行转换：
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) )，可以使用如下命令进行转换：
 
 ```shell
 # 注意将pretrained_model的路径设置为本地路径。
diff --git a/doc/doc_en/algorithm_rec_abinet_en.md b/doc/doc_en/algorithm_rec_abinet_en.md
index 3b0f6c09a9..767ca65f64 100644
--- a/doc/doc_en/algorithm_rec_abinet_en.md
+++ b/doc/doc_en/algorithm_rec_abinet_en.md
@@ -25,7 +25,7 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval
 
 |Model|Backbone|config|Acc|Download link|
 | --- | --- | --- | --- | --- |
-|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[trained model]()/[pretrained model]()|
+|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[pretrained & trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)|
 
 <a name="2"></a>
 ## 2. Environment
@@ -68,7 +68,7 @@ python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img
 
 <a name="4-1"></a>
 ### 4.1 Python Inference
-First, the model saved during the ABINet text recognition training process is converted into an inference model. ( [Model download link]()) ), you can use the following command to convert:
+First, the model saved during the ABINet text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)) ), you can use the following command to convert:
 
 ```
 python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy  Global.save_inference_dir=./inference/rec_r45_abinet
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index 437e0152fa..63dfda91f8 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -25,7 +25,7 @@ from .make_pse_gt import MakePseGt
 
 from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
     SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, \
-    ABINetRecResizeImg, SVTRRecResizeImg
+    ABINetRecResizeImg, SVTRRecResizeImg, ABINetRecAug
 from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste
diff --git a/ppocr/data/imaug/abinet_aug.py b/ppocr/data/imaug/abinet_aug.py
new file mode 100644
index 0000000000..eefdc75d5a
--- /dev/null
+++ b/ppocr/data/imaug/abinet_aug.py
@@ -0,0 +1,407 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FangShancheng/ABINet/blob/main/transforms.py
+"""
+import math
+import numbers
+import random
+
+import cv2
+import numpy as np
+from paddle.vision.transforms import Compose, ColorJitter
+
+
+def sample_asym(magnitude, size=None):
+    return np.random.beta(1, 4, size) * magnitude
+
+
+def sample_sym(magnitude, size=None):
+    return (np.random.beta(4, 4, size=size) - 0.5) * 2 * magnitude
+
+
+def sample_uniform(low, high, size=None):
+    return np.random.uniform(low, high, size=size)
+
+
+def get_interpolation(type='random'):
+    if type == 'random':
+        choice = [
+            cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA
+        ]
+        interpolation = choice[random.randint(0, len(choice) - 1)]
+    elif type == 'nearest':
+        interpolation = cv2.INTER_NEAREST
+    elif type == 'linear':
+        interpolation = cv2.INTER_LINEAR
+    elif type == 'cubic':
+        interpolation = cv2.INTER_CUBIC
+    elif type == 'area':
+        interpolation = cv2.INTER_AREA
+    else:
+        raise TypeError(
+            'Interpolation types only nearest, linear, cubic, area are supported!'
+        )
+    return interpolation
+
+
+class CVRandomRotation(object):
+    def __init__(self, degrees=15):
+        assert isinstance(degrees,
+                          numbers.Number), "degree should be a single number."
+        assert degrees >= 0, "degree must be positive."
+        self.degrees = degrees
+
+    @staticmethod
+    def get_params(degrees):
+        return sample_sym(degrees)
+
+    def __call__(self, img):
+        angle = self.get_params(self.degrees)
+        src_h, src_w = img.shape[:2]
+        M = cv2.getRotationMatrix2D(
+            center=(src_w / 2, src_h / 2), angle=angle, scale=1.0)
+        abs_cos, abs_sin = abs(M[0, 0]), abs(M[0, 1])
+        dst_w = int(src_h * abs_sin + src_w * abs_cos)
+        dst_h = int(src_h * abs_cos + src_w * abs_sin)
+        M[0, 2] += (dst_w - src_w) / 2
+        M[1, 2] += (dst_h - src_h) / 2
+
+        flags = get_interpolation()
+        return cv2.warpAffine(
+            img,
+            M, (dst_w, dst_h),
+            flags=flags,
+            borderMode=cv2.BORDER_REPLICATE)
+
+
+class CVRandomAffine(object):
+    def __init__(self, degrees, translate=None, scale=None, shear=None):
+        assert isinstance(degrees,
+                          numbers.Number), "degree should be a single number."
+        assert degrees >= 0, "degree must be positive."
+        self.degrees = degrees
+
+        if translate is not None:
+            assert isinstance(translate, (tuple, list)) and len(translate) == 2, \
+                "translate should be a list or tuple and it must be of length 2."
+            for t in translate:
+                if not (0.0 <= t <= 1.0):
+                    raise ValueError(
+                        "translation values should be between 0 and 1")
+        self.translate = translate
+
+        if scale is not None:
+            assert isinstance(scale, (tuple, list)) and len(scale) == 2, \
+                "scale should be a list or tuple and it must be of length 2."
+            for s in scale:
+                if s <= 0:
+                    raise ValueError("scale values should be positive")
+        self.scale = scale
+
+        if shear is not None:
+            if isinstance(shear, numbers.Number):
+                if shear < 0:
+                    raise ValueError(
+                        "If shear is a single number, it must be positive.")
+                self.shear = [shear]
+            else:
+                assert isinstance(shear, (tuple, list)) and (len(shear) == 2), \
+                    "shear should be a list or tuple and it must be of length 2."
+                self.shear = shear
+        else:
+            self.shear = shear
+
+    def _get_inverse_affine_matrix(self, center, angle, translate, scale,
+                                   shear):
+        # https://github.com/pytorch/vision/blob/v0.4.0/torchvision/transforms/functional.py#L717
+        from numpy import sin, cos, tan
+
+        if isinstance(shear, numbers.Number):
+            shear = [shear, 0]
+
+        if not isinstance(shear, (tuple, list)) and len(shear) == 2:
+            raise ValueError(
+                "Shear should be a single value or a tuple/list containing " +
+                "two values. Got {}".format(shear))
+
+        rot = math.radians(angle)
+        sx, sy = [math.radians(s) for s in shear]
+
+        cx, cy = center
+        tx, ty = translate
+
+        # RSS without scaling
+        a = cos(rot - sy) / cos(sy)
+        b = -cos(rot - sy) * tan(sx) / cos(sy) - sin(rot)
+        c = sin(rot - sy) / cos(sy)
+        d = -sin(rot - sy) * tan(sx) / cos(sy) + cos(rot)
+
+        # Inverted rotation matrix with scale and shear
+        # det([[a, b], [c, d]]) == 1, since det(rotation) = 1 and det(shear) = 1
+        M = [d, -b, 0, -c, a, 0]
+        M = [x / scale for x in M]
+
+        # Apply inverse of translation and of center translation: RSS^-1 * C^-1 * T^-1
+        M[2] += M[0] * (-cx - tx) + M[1] * (-cy - ty)
+        M[5] += M[3] * (-cx - tx) + M[4] * (-cy - ty)
+
+        # Apply center translation: C * RSS^-1 * C^-1 * T^-1
+        M[2] += cx
+        M[5] += cy
+        return M
+
+    @staticmethod
+    def get_params(degrees, translate, scale_ranges, shears, height):
+        angle = sample_sym(degrees)
+        if translate is not None:
+            max_dx = translate[0] * height
+            max_dy = translate[1] * height
+            translations = (np.round(sample_sym(max_dx)),
+                            np.round(sample_sym(max_dy)))
+        else:
+            translations = (0, 0)
+
+        if scale_ranges is not None:
+            scale = sample_uniform(scale_ranges[0], scale_ranges[1])
+        else:
+            scale = 1.0
+
+        if shears is not None:
+            if len(shears) == 1:
+                shear = [sample_sym(shears[0]), 0.]
+            elif len(shears) == 2:
+                shear = [sample_sym(shears[0]), sample_sym(shears[1])]
+        else:
+            shear = 0.0
+
+        return angle, translations, scale, shear
+
+    def __call__(self, img):
+        src_h, src_w = img.shape[:2]
+        angle, translate, scale, shear = self.get_params(
+            self.degrees, self.translate, self.scale, self.shear, src_h)
+
+        M = self._get_inverse_affine_matrix((src_w / 2, src_h / 2), angle,
+                                            (0, 0), scale, shear)
+        M = np.array(M).reshape(2, 3)
+
+        startpoints = [(0, 0), (src_w - 1, 0), (src_w - 1, src_h - 1),
+                       (0, src_h - 1)]
+        project = lambda x, y, a, b, c: int(a * x + b * y + c)
+        endpoints = [(project(x, y, *M[0]), project(x, y, *M[1]))
+                     for x, y in startpoints]
+
+        rect = cv2.minAreaRect(np.array(endpoints))
+        bbox = cv2.boxPoints(rect).astype(dtype=np.int)
+        max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max()
+        min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min()
+
+        dst_w = int(max_x - min_x)
+        dst_h = int(max_y - min_y)
+        M[0, 2] += (dst_w - src_w) / 2
+        M[1, 2] += (dst_h - src_h) / 2
+
+        # add translate
+        dst_w += int(abs(translate[0]))
+        dst_h += int(abs(translate[1]))
+        if translate[0] < 0: M[0, 2] += abs(translate[0])
+        if translate[1] < 0: M[1, 2] += abs(translate[1])
+
+        flags = get_interpolation()
+        return cv2.warpAffine(
+            img,
+            M, (dst_w, dst_h),
+            flags=flags,
+            borderMode=cv2.BORDER_REPLICATE)
+
+
+class CVRandomPerspective(object):
+    def __init__(self, distortion=0.5):
+        self.distortion = distortion
+
+    def get_params(self, width, height, distortion):
+        offset_h = sample_asym(
+            distortion * height / 2, size=4).astype(dtype=np.int)
+        offset_w = sample_asym(
+            distortion * width / 2, size=4).astype(dtype=np.int)
+        topleft = (offset_w[0], offset_h[0])
+        topright = (width - 1 - offset_w[1], offset_h[1])
+        botright = (width - 1 - offset_w[2], height - 1 - offset_h[2])
+        botleft = (offset_w[3], height - 1 - offset_h[3])
+
+        startpoints = [(0, 0), (width - 1, 0), (width - 1, height - 1),
+                       (0, height - 1)]
+        endpoints = [topleft, topright, botright, botleft]
+        return np.array(
+            startpoints, dtype=np.float32), np.array(
+                endpoints, dtype=np.float32)
+
+    def __call__(self, img):
+        height, width = img.shape[:2]
+        startpoints, endpoints = self.get_params(width, height, self.distortion)
+        M = cv2.getPerspectiveTransform(startpoints, endpoints)
+
+        # TODO: more robust way to crop image
+        rect = cv2.minAreaRect(endpoints)
+        bbox = cv2.boxPoints(rect).astype(dtype=np.int)
+        max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max()
+        min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min()
+        min_x, min_y = max(min_x, 0), max(min_y, 0)
+
+        flags = get_interpolation()
+        img = cv2.warpPerspective(
+            img,
+            M, (max_x, max_y),
+            flags=flags,
+            borderMode=cv2.BORDER_REPLICATE)
+        img = img[min_y:, min_x:]
+        return img
+
+
+class CVRescale(object):
+    def __init__(self, factor=4, base_size=(128, 512)):
+        """ Define image scales using gaussian pyramid and rescale image to target scale.
+        
+        Args:
+            factor: the decayed factor from base size, factor=4 keeps target scale by default.
+            base_size: base size the build the bottom layer of pyramid
+        """
+        if isinstance(factor, numbers.Number):
+            self.factor = round(sample_uniform(0, factor))
+        elif isinstance(factor, (tuple, list)) and len(factor) == 2:
+            self.factor = round(sample_uniform(factor[0], factor[1]))
+        else:
+            raise Exception('factor must be number or list with length 2')
+        # assert factor is valid
+        self.base_h, self.base_w = base_size[:2]
+
+    def __call__(self, img):
+        if self.factor == 0: return img
+        src_h, src_w = img.shape[:2]
+        cur_w, cur_h = self.base_w, self.base_h
+        scale_img = cv2.resize(
+            img, (cur_w, cur_h), interpolation=get_interpolation())
+        for _ in range(self.factor):
+            scale_img = cv2.pyrDown(scale_img)
+        scale_img = cv2.resize(
+            scale_img, (src_w, src_h), interpolation=get_interpolation())
+        return scale_img
+
+
+class CVGaussianNoise(object):
+    def __init__(self, mean=0, var=20):
+        self.mean = mean
+        if isinstance(var, numbers.Number):
+            self.var = max(int(sample_asym(var)), 1)
+        elif isinstance(var, (tuple, list)) and len(var) == 2:
+            self.var = int(sample_uniform(var[0], var[1]))
+        else:
+            raise Exception('degree must be number or list with length 2')
+
+    def __call__(self, img):
+        noise = np.random.normal(self.mean, self.var**0.5, img.shape)
+        img = np.clip(img + noise, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVMotionBlur(object):
+    def __init__(self, degrees=12, angle=90):
+        if isinstance(degrees, numbers.Number):
+            self.degree = max(int(sample_asym(degrees)), 1)
+        elif isinstance(degrees, (tuple, list)) and len(degrees) == 2:
+            self.degree = int(sample_uniform(degrees[0], degrees[1]))
+        else:
+            raise Exception('degree must be number or list with length 2')
+        self.angle = sample_uniform(-angle, angle)
+
+    def __call__(self, img):
+        M = cv2.getRotationMatrix2D((self.degree // 2, self.degree // 2),
+                                    self.angle, 1)
+        motion_blur_kernel = np.zeros((self.degree, self.degree))
+        motion_blur_kernel[self.degree // 2, :] = 1
+        motion_blur_kernel = cv2.warpAffine(motion_blur_kernel, M,
+                                            (self.degree, self.degree))
+        motion_blur_kernel = motion_blur_kernel / self.degree
+        img = cv2.filter2D(img, -1, motion_blur_kernel)
+        img = np.clip(img, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVGeometry(object):
+    def __init__(self,
+                 degrees=15,
+                 translate=(0.3, 0.3),
+                 scale=(0.5, 2.),
+                 shear=(45, 15),
+                 distortion=0.5,
+                 p=0.5):
+        self.p = p
+        type_p = random.random()
+        if type_p < 0.33:
+            self.transforms = CVRandomRotation(degrees=degrees)
+        elif type_p < 0.66:
+            self.transforms = CVRandomAffine(
+                degrees=degrees, translate=translate, scale=scale, shear=shear)
+        else:
+            self.transforms = CVRandomPerspective(distortion=distortion)
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            return self.transforms(img)
+        else:
+            return img
+
+
+class CVDeterioration(object):
+    def __init__(self, var, degrees, factor, p=0.5):
+        self.p = p
+        transforms = []
+        if var is not None:
+            transforms.append(CVGaussianNoise(var=var))
+        if degrees is not None:
+            transforms.append(CVMotionBlur(degrees=degrees))
+        if factor is not None:
+            transforms.append(CVRescale(factor=factor))
+
+        random.shuffle(transforms)
+        transforms = Compose(transforms)
+        self.transforms = transforms
+
+    def __call__(self, img):
+        if random.random() < self.p:
+
+            return self.transforms(img)
+        else:
+            return img
+
+
+class CVColorJitter(object):
+    def __init__(self,
+                 brightness=0.5,
+                 contrast=0.5,
+                 saturation=0.5,
+                 hue=0.1,
+                 p=0.5):
+        self.p = p
+        self.transforms = ColorJitter(
+            brightness=brightness,
+            contrast=contrast,
+            saturation=saturation,
+            hue=hue)
+
+    def __call__(self, img):
+        if random.random() < self.p: return self.transforms(img)
+        else: return img
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index 874d9aa084..92f7382724 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -19,6 +19,8 @@ import random
 import copy
 from PIL import Image
 from .text_image_aug import tia_perspective, tia_stretch, tia_distort
+from .abinet_aug import CVGeometry, CVDeterioration, CVColorJitter
+from paddle.vision.transforms import Compose
 
 
 class RecAug(object):
@@ -94,6 +96,31 @@ class BaseDataAugmentation(object):
         return data
 
 
+class ABINetRecAug(object):
+    def __init__(self, **kwargs):
+        self.transforms = Compose([
+            CVGeometry(
+                degrees=45,
+                translate=(0.0, 0.0),
+                scale=(0.5, 2.),
+                shear=(45, 15),
+                distortion=0.5,
+                p=0.5), CVDeterioration(
+                    var=20, degrees=6, factor=4, p=0.25), CVColorJitter(
+                        brightness=0.5,
+                        contrast=0.5,
+                        saturation=0.5,
+                        hue=0.1,
+                        p=0.25)
+        ])
+
+    def __call__(self, data):
+        img = data['image']
+        img = self.transforms(img)
+        data['image'] = img
+        return data
+
+
 class RecConAug(object):
     def __init__(self,
                  prob=0.5,
diff --git a/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
index 306c49d683..5b5890e772 100644
--- a/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
+++ b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
@@ -68,6 +68,7 @@ Train:
       - DecodeImage: # load image
           img_mode: RGB
           channel_first: False
+      - ABINetRecAug:
       - ABINetLabelEncode: # Class handling label
           ignore_index: *ignore_index
       - ABINetRecResizeImg:

From 1c5e7f2578b8ac1d8b4999782ac0349a0f49781d Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Fri, 24 Jun 2022 09:56:27 +0000
Subject: [PATCH 051/134] aug p infe

---
 applications/高精度中文识别模型.md | 11 ++++++-----
 configs/rec/rec_svtrnet_ch.yml     |  8 ++++----
 ppocr/data/imaug/rec_img_aug.py    | 21 +++++++++++++--------
 3 files changed, 23 insertions(+), 17 deletions(-)

diff --git a/applications/高精度中文识别模型.md b/applications/高精度中文识别模型.md
index e3d0693ece..3c31af42ee 100644
--- a/applications/高精度中文识别模型.md
+++ b/applications/高精度中文识别模型.md
@@ -2,7 +2,7 @@
 
 ## 1. 简介
 
-PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中超轻量的场景中文识别模型SVTR_LCNet使用了SVTR算法结构。为了保证速度，SVTR_LCNet将SVTR模型的Local Blocks替换为LCNet，使用两层Global Blocks。在中文场景中，经过多次优化，SVTR_LCNet的最终精度为79.4%。具体的：
+PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中超轻量的场景中文识别模型SVTR_LCNet使用了SVTR算法结构。为了保证速度，SVTR_LCNet将SVTR模型的Local Blocks替换为LCNet，使用两层Global Blocks。在中文场景中，PP-OCRv3识别主要使用如下优化策略：
 - GTC：Attention指导CTC训练策略；
 - TextConAug：挖掘文字上下文信息的数据增广策略；
 - TextRotNet：自监督的预训练模型；
@@ -11,10 +11,11 @@ PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中
 
 其中 *UIM：无标注数据挖掘方案* 使用了高精度的SVTR中文模型进行无标注文件的刷库，该模型在PP-OCRv3识别的数据集上训练，精度对比如下表。
 
-|中文识别算法|模型|精度|
-| --- | --- | --- |
-|PP-OCRv3|SVTR_LCNet|79.4%|
-|SVTR|SVTR-Tiny|82.5%|
+|中文识别算法|模型|UIM|精度|
+| --- | --- | --- |--- |
+|PP-OCRv3|SVTR_LCNet| w/o |78.4%|
+|PP-OCRv3|SVTR_LCNet| w |79.4%|
+|SVTR|SVTR-Tiny|-|82.5%|
 
 aistudio项目链接: [高精度中文场景文本识别模型SVTR](https://aistudio.baidu.com/aistudio/projectdetail/4263032)
 
diff --git a/configs/rec/rec_svtrnet_ch.yml b/configs/rec/rec_svtrnet_ch.yml
index b327edf499..c4ce4ccc16 100644
--- a/configs/rec/rec_svtrnet_ch.yml
+++ b/configs/rec/rec_svtrnet_ch.yml
@@ -92,9 +92,9 @@ Metric:
 Train:
   dataset:
     name: SimpleDataSet
+    data_dir: ./train_data
     label_file_list:
-    - /paddle/data/ocr_all/train_all_list.txt
-    data_dir: /paddle/data/ocr_all
+    - ./train_data/train_list.txt
     ext_op_transform_idx: 1
     transforms:
     - DecodeImage:
@@ -130,9 +130,9 @@ Train:
 Eval:
   dataset:
     name: SimpleDataSet
-    data_dir: /paddle/data/ocr_all
+    data_dir: ./train_data
     label_file_list:
-    - /paddle/data/ocr_all/val_list.txt
+    - ./train_data/val_list.txt
     transforms:
     - DecodeImage:
         img_mode: BGR
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index 92f7382724..1de8e6ee3c 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -97,7 +97,11 @@ class BaseDataAugmentation(object):
 
 
 class ABINetRecAug(object):
-    def __init__(self, **kwargs):
+    def __init__(self,
+                 geometry_p=0.5,
+                 deterioration_p=0.25,
+                 colorjitter_p=0.25,
+                 **kwargs):
         self.transforms = Compose([
             CVGeometry(
                 degrees=45,
@@ -105,13 +109,14 @@ class ABINetRecAug(object):
                 scale=(0.5, 2.),
                 shear=(45, 15),
                 distortion=0.5,
-                p=0.5), CVDeterioration(
-                    var=20, degrees=6, factor=4, p=0.25), CVColorJitter(
-                        brightness=0.5,
-                        contrast=0.5,
-                        saturation=0.5,
-                        hue=0.1,
-                        p=0.25)
+                p=geometry_p), CVDeterioration(
+                    var=20, degrees=6, factor=4, p=deterioration_p),
+            CVColorJitter(
+                brightness=0.5,
+                contrast=0.5,
+                saturation=0.5,
+                hue=0.1,
+                p=colorjitter_p)
         ])
 
     def __call__(self, data):

From 6a8980e478ccacb7d3dfd9cd093c0c47132e01c9 Mon Sep 17 00:00:00 2001
From: an1018 <614803115@qq.com>
Date: Sat, 25 Jun 2022 17:28:34 +0800
Subject: [PATCH 052/134] update

---
 applications/多模态表单识别.md | 157 ++++++++++++++++-----------------
 1 file changed, 77 insertions(+), 80 deletions(-)

diff --git a/applications/多模态表单识别.md b/applications/多模态表单识别.md
index d47bbe7704..457d397030 100644
--- a/applications/多模态表单识别.md
+++ b/applications/多模态表单识别.md
@@ -1,4 +1,8 @@
-# 1 项目说明
+# 多模态表单识别
+
+[TOC]
+
+## 1 项目说明
 
 计算机视觉在金融领域的应用覆盖文字识别、图像识别、视频识别等，其中文字识别（OCR）是金融领域中的核心AI能力，其应用覆盖客户服务、风险防控、运营管理等各项业务，针对的对象包括通用卡证票据识别（银行卡、身份证、营业执照等）、通用文本表格识别（印刷体、多语言、手写体等）以及一些金融特色票据凭证。通过因此如果能够在结构化信息提取时同时利用文字、页面布局等信息，便可增强不同版式下的泛化性。
 
@@ -16,39 +20,37 @@
 <center><img src='https://ai-studio-static-online.cdn.bcebos.com/9bd844b970f94e5ba0bc0c5799bd819ea9b1861bb306471fabc2d628864d418e'></center>
 <center>图1 多模态表单识别流程图</center>
 
-注：欢迎再AIStudio领取免费算力体验线上实训，项目链接: [多模态表单识别](https://aistudio.baidu.com/aistudio/projectdetail/3884375)(配备Tesla V100、A100等高级算力资源)
+注：欢迎再AIStudio领取免费算力体验线上实训，项目链接: [多模态表单识别](https://aistudio.baidu.com/aistudio/projectdetail/3884375?contributionType=1)
 
-
-
-# 2 安装说明
+## 2 安装说明
 
 
 下载PaddleOCR源码，上述AIStudio项目中已经帮大家打包好的PaddleOCR(已经修改好配置文件)，无需下载解压即可，只需安装依赖环境~
 
 
 ```python
-! unzip -q PaddleOCR.zip
+unzip -q PaddleOCR.zip
 ```
 
 
 ```python
 # 如仍需安装or安装更新，可以执行以下步骤
-# ! git clone https://github.com/PaddlePaddle/PaddleOCR.git -b dygraph
-# ! git clone https://gitee.com/PaddlePaddle/PaddleOCR
+# git clone https://github.com/PaddlePaddle/PaddleOCR.git -b dygraph
+# git clone https://gitee.com/PaddlePaddle/PaddleOCR
 ```
 
 
 ```python
 # 安装依赖包
-! pip install -U pip
-! pip install -r /home/aistudio/PaddleOCR/requirements.txt
-! pip install paddleocr
+pip install -U pip
+pip install -r /home/aistudio/PaddleOCR/requirements.txt
+pip install paddleocr
 
-! pip install yacs gnureadline paddlenlp==2.2.1
-! pip install xlsxwriter
+pip install yacs gnureadline paddlenlp==2.2.1
+pip install xlsxwriter
 ```
 
-# 3 数据准备
+## 3 数据准备
 
 这里使用[XFUN数据集](https://github.com/doc-analysis/XFUND)做为实验数据集。 XFUN数据集是微软提出的一个用于KIE任务的多语言数据集，共包含七个数据集，每个数据集包含149张训练集和50张验证集
 
@@ -59,7 +61,7 @@
 <center><img src="https://ai-studio-static-online.cdn.bcebos.com/0f84137778cd4ab6899c64109d452290e9c678ccf01744978bc9c0647adbba45" width="1000" ></center>
 <center>图2 数据集样例，左中文，右法语</center>
 
-## 3.1 下载处理好的数据集
+### 3.1 下载处理好的数据集
 
 
 处理好的XFUND中文数据集下载地址：[https://paddleocr.bj.bcebos.com/dataset/XFUND.tar](https://paddleocr.bj.bcebos.com/dataset/XFUND.tar) ,可以运行如下指令完成中文数据集下载和解压。
@@ -69,13 +71,13 @@
 
 
 ```python
-! wget https://paddleocr.bj.bcebos.com/dataset/XFUND.tar
-! tar -xf XFUND.tar
+wget https://paddleocr.bj.bcebos.com/dataset/XFUND.tar
+tar -xf XFUND.tar
 
 # XFUN其他数据集使用下面的代码进行转换
 # 代码链接：https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.4/ppstructure/vqa/helper/trans_xfun_data.py
 # %cd PaddleOCR
-# !python3 ppstructure/vqa/tools/trans_xfun_data.py --ori_gt_path=path/to/json_path --output_path=path/to/save_path
+# python3 ppstructure/vqa/tools/trans_xfun_data.py --ori_gt_path=path/to/json_path --output_path=path/to/save_path
 # %cd ../
 ```
 
@@ -119,7 +121,7 @@
 }
 ```
 
-## 3.2 转换为PaddleOCR检测和识别格式
+### 3.2 转换为PaddleOCR检测和识别格式
 
 使用XFUND训练PaddleOCR检测和识别模型，需要将数据集格式改为训练需求的格式。
 
@@ -147,7 +149,7 @@ train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
 
 
 ```python
-! unzip -q /home/aistudio/data/data140302/XFUND_ori.zip -d /home/aistudio/data/data140302/
+unzip -q /home/aistudio/data/data140302/XFUND_ori.zip -d /home/aistudio/data/data140302/
 ```
 
 已经提供转换脚本，执行如下代码即可转换成功：
@@ -155,21 +157,20 @@ train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
 
 ```python
 %cd /home/aistudio/
-! python trans_xfund_data.py
+python trans_xfund_data.py
 ```
 
-# 4 OCR
+## 4 OCR
 
 选用飞桨OCR开发套件[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/README_ch.md)中的PP-OCRv2模型进行文本检测和识别。PP-OCRv2在PP-OCR的基础上，进一步在5个方面重点优化，检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略；识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/enhanced_ctc_loss.md)损失函数改进，进一步在推理速度和预测效果上取得明显提升。更多细节请参考PP-OCRv2[技术报告](https://arxiv.org/abs/2109.03144)。
 
-
-## 4.1 文本检测
+### 4.1 文本检测
 
 我们使用2种方案进行训练、评估：
 -  **PP-OCRv2中英文超轻量检测预训练模型**
 -  **XFUND数据集+fine-tune**
 
-### **4.1.1 方案1：预训练模型**
+#### **4.1.1 方案1：预训练模型**
 
 **1）下载预训练模型**
 
@@ -195,8 +196,8 @@ PaddleOCR已经提供了PP-OCR系列模型，部分模型展示如下表所示
 
 ```python
 %cd /home/aistudio/PaddleOCR/pretrain/
-! wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar
-! tar -xf ch_PP-OCRv2_det_distill_train.tar && rm -rf ch_PP-OCRv2_det_distill_train.tar
+wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar
+tar -xf ch_PP-OCRv2_det_distill_train.tar && rm -rf ch_PP-OCRv2_det_distill_train.tar
 % cd ..
 ```
 
@@ -226,7 +227,7 @@ Eval.dataset.label_file_list：指向验证集标注文件
 
 ```python
 %cd /home/aistudio/PaddleOCR
-! python tools/eval.py \
+python tools/eval.py \
     -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml \
     -o Global.checkpoints="./pretrain_models/ch_PP-OCRv2_det_distill_train/best_accuracy"
 ```
@@ -237,9 +238,9 @@ Eval.dataset.label_file_list：指向验证集标注文件
 | -------- | -------- |
 | PP-OCRv2中英文超轻量检测预训练模型     | 77.26%     |
 
-使用文本检测预训练模型在XFUND验证集上评估，达到77%左右，充分说明ppocr提供的预训练模型有一定的泛化能力。
+使用文本检测预训练模型在XFUND验证集上评估，达到77%左右，充分说明ppocr提供的预训练模型具有泛化能力。
 
-### **4.1.2 方案2：XFUND数据集+fine-tune**
+#### **4.1.2 方案2：XFUND数据集+fine-tune**
 
 PaddleOCR提供的蒸馏预训练模型包含了多个模型的参数，我们提取Student模型的参数，在XFUND数据集上进行finetune，可以参考如下代码：
 
@@ -281,7 +282,7 @@ Eval.dataset.transforms.DetResizeForTest：评估尺寸，添加如下参数
 
 
 ```python
-! CUDA_VISIBLE_DEVICES=0 python tools/train.py \
+CUDA_VISIBLE_DEVICES=0 python tools/train.py \
         -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml
 ```
 
@@ -290,12 +291,12 @@ Eval.dataset.transforms.DetResizeForTest：评估尺寸，添加如下参数
 <center><img src="https://ai-studio-static-online.cdn.bcebos.com/5a75137c5f924dfeb6956b5818812298cc3dc7992ac84954b4175be9adf83c77"></center>
 <center>图8 文本检测方案2-模型评估</center>
 
-使用训练好的模型进行评估，更新模型路径`Global.checkpoints`，这里为大家提供训练好的模型`./pretrain/ch_db_mv3-student1600-finetune/best_accuracy`，[模型下载地址](https://paddleocr.bj.bcebos.com/fanliku/sheet_recognition/ch_db_mv3-student1600-finetune.zip)
+使用训练好的模型进行评估，更新模型路径`Global.checkpoints`。
 
 
 ```python
 %cd /home/aistudio/PaddleOCR/
-! python tools/eval.py \
+python tools/eval.py \
     -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml \
     -o Global.checkpoints="pretrain/ch_db_mv3-student1600-finetune/best_accuracy"
 ```
@@ -305,7 +306,7 @@ Eval.dataset.transforms.DetResizeForTest：评估尺寸，添加如下参数
 
 ```python
 %cd /home/aistudio/PaddleOCR/
-! python tools/eval.py \
+python tools/eval.py \
     -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml \
     -o Global.checkpoints="pretrain/ch_db_mv3-student1600/best_accuracy"
 ```
@@ -331,7 +332,7 @@ Eval.dataset.transforms.DetResizeForTest：评估尺寸，添加如下参数
 # 加载配置文件`ch_PP-OCRv2_det_student.yml`，从`pretrain/ch_db_mv3-student1600-finetune`目录下加载`best_accuracy`模型
 # inference模型保存在`./output/det_db_inference`目录下
 %cd /home/aistudio/PaddleOCR/
-! python tools/export_model.py \
+python tools/export_model.py \
     -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml \
     -o Global.pretrained_model="pretrain/ch_db_mv3-student1600-finetune/best_accuracy" \
     Global.save_inference_dir="./output/det_db_inference/"
@@ -374,12 +375,11 @@ use_gpu：是否使用GPU
 
 | 方案 | hmeans | 结果分析 |
 | -------- | -------- | -------- |
-| PP-OCRv2中英文超轻量检测预训练模型     | 77.26%     | ppocr提供的预训练模型有一定的泛化能力     |
+| PP-OCRv2中英文超轻量检测预训练模型     | 77.26%     | ppocr提供的预训练模型有泛化能力     |
 | XFUND数据集     | 79.27%     |     |
 | XFUND数据集+finetune     | 85.24%     | finetune会提升垂类场景效果    |
 
-
-## 4.2 文本识别
+### 4.2 文本识别
 
 我们分别使用如下3种方案进行训练、评估：
 
@@ -387,8 +387,7 @@ use_gpu：是否使用GPU
 - XFUND数据集+fine-tune
 - XFUND数据集+fine-tune+真实通用识别数据
 
-
-### **4.2.1 方案1：预训练模型**
+#### **4.2.1 方案1：预训练模型**
 
 **1）下载预训练模型**
 
@@ -401,8 +400,8 @@ use_gpu：是否使用GPU
 
 ```python
 %cd /home/aistudio/PaddleOCR/pretrain/
-! wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar
-! tar -xf ch_PP-OCRv2_rec_train.tar && rm -rf ch_PP-OCRv2_rec_train.tar
+wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar
+tar -xf ch_PP-OCRv2_rec_train.tar && rm -rf ch_PP-OCRv2_rec_train.tar
 % cd ..
 ```
 
@@ -424,7 +423,7 @@ Eval.dataset.label_file_list：指向验证集标注文件
 
 ```python
 %cd /home/aistudio/PaddleOCR
-! CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
+CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
     -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml \
     -o Global.checkpoints=./pretrain/ch_PP-OCRv2_rec_train/best_accuracy
 ```
@@ -435,9 +434,9 @@ Eval.dataset.label_file_list：指向验证集标注文件
 | -------- | -------- | 
 | PP-OCRv2中英文超轻量识别预训练模型     | 67.48%     | 
 
-使用文本预训练模型在XFUND验证集上评估，acc达到67%左右，充分说明ppocr提供的预训练模型有一定的泛化能力。
+使用文本预训练模型在XFUND验证集上评估，acc达到67%左右，充分说明ppocr提供的预训练模型具有泛化能力。
 
-### **4.2.2 方案2：XFUND数据集+finetune**
+#### **4.2.2 方案2：XFUND数据集+finetune**
 
 同检测模型，我们提取Student模型的参数，在XFUND数据集上进行finetune，可以参考如下代码：
 
@@ -474,11 +473,9 @@ Eval.dataset.label_file_list：指向验证集标注文件
 ```
 执行如下命令启动训练：
 
-
-
 ```python
 %cd /home/aistudio/PaddleOCR/
-! CUDA_VISIBLE_DEVICES=0 python tools/train.py \
+CUDA_VISIBLE_DEVICES=0 python tools/train.py \
         -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml
 ```
 
@@ -493,7 +490,7 @@ Eval.dataset.label_file_list：指向验证集标注文件
 
 ```python
 %cd /home/aistudio/PaddleOCR/
-! CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
+CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
     -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml \
     -o Global.checkpoints=./pretrain/rec_mobile_pp-OCRv2-student-finetune/best_accuracy
 ```
@@ -506,7 +503,7 @@ Eval.dataset.label_file_list：指向验证集标注文件
 
 使用XFUND数据集+finetune训练，在验证集上评估达到72%左右，说明 finetune会提升垂类场景效果。
 
-### **4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据**
+#### **4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据**
 
 接着我们在上述`XFUND数据集+finetune`实验的基础上，添加真实通用识别数据，进一步提升识别效果。首先准备真实通用识别数据，并上传到AIStudio：
 
@@ -528,7 +525,7 @@ Train.dataset.ratio_list：动态采样
 
 ```python
 %cd /home/aistudio/PaddleOCR/
-! CUDA_VISIBLE_DEVICES=0 python tools/train.py \
+CUDA_VISIBLE_DEVICES=0 python tools/train.py \
         -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml
 ```
 
@@ -538,11 +535,11 @@ Train.dataset.ratio_list：动态采样
 
 <center>图16 文本识别方案3-模型评估</center>
 
-使用训练好的模型进行评估，更新模型路径`Global.checkpoints`，这里为大家提供训练好的模型`./pretrain/rec_mobile_pp-OCRv2-student-readldata/best_accuracy`，[模型下载地址](https://paddleocr.bj.bcebos.com/fanliku/sheet_recognition/rec_mobile_pp-OCRv2-student-realdata.zip)
+使用训练好的模型进行评估，更新模型路径`Global.checkpoints`。
 
 
 ```python
-! CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
+CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
     -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml \
     -o Global.checkpoints=./pretrain/rec_mobile_pp-OCRv2-student-realdata/best_accuracy
 ```
@@ -580,7 +577,7 @@ Train.dataset.ratio_list：动态采样
 
 
 ```python
-! python tools/infer/predict_system.py \
+python tools/infer/predict_system.py \
     --image_dir="./doc/vqa/input/zh_val_21.jpg" \
     --det_model_dir="./output/det_db_inference/" \
     --rec_model_dir="./output/rec_crnn_inference/" \
@@ -592,11 +589,11 @@ Train.dataset.ratio_list：动态采样
 
 | 方案 | acc  | 结果分析 |
 | -------- | -------- | -------- |
-| PP-OCRv2中英文超轻量识别预训练模型     | 67.48%    | ppocr提供的预训练模型有一定的泛化能力     |
+| PP-OCRv2中英文超轻量识别预训练模型     | 67.48%    | ppocr提供的预训练模型具有泛化能力  |
 | XFUND数据集+fine-tune    |72.33%     | finetune会提升垂类场景效果    |
 | XFUND数据集+fine-tune+真实通用识别数据    | 85.29%     | 真实通用识别数据对于性能提升很有帮助    |
 
-# 5 文档视觉问答(DOC-VQA)
+## 5 文档视觉问答(DOC-VQA)
 
 VQA指视觉问答，主要针对图像内容进行提问和回答,DOC-VQA是VQA任务中的一种，DOC-VQA主要针对文本图像的文字内容提出问题。
 
@@ -608,14 +605,13 @@ PaddleOCR中DOC-VQA系列算法基于PaddleNLP自然语言处理算法库实现L
 ```python
 %cd pretrain
 #下载SER模型
-! wget https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar && tar -xvf ser_LayoutXLM_xfun_zh.tar
+wget https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar && tar -xvf ser_LayoutXLM_xfun_zh.tar
 #下载RE模型
-! wget https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar && tar -xvf re_LayoutXLM_xfun_zh.tar
+wget https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar && tar -xvf re_LayoutXLM_xfun_zh.tar
 %cd ../
 ```
 
-
-## 5.1 SER
+### 5.1 SER
 
 SER: 语义实体识别 (Semantic Entity Recognition）, 可以完成对图像中的文本识别与分类。 
 
@@ -647,7 +643,7 @@ SER: 语义实体识别 (Semantic Entity Recognition）, 可以完成对图像
 
 ```python
 %cd /home/aistudio/PaddleOCR/
-! CUDA_VISIBLE_DEVICES=0 python tools/train.py -c configs/vqa/ser/layoutxlm.yml
+CUDA_VISIBLE_DEVICES=0 python tools/train.py -c configs/vqa/ser/layoutxlm.yml
 ```
 
 最终会打印出`precision`, `recall`, `hmean`等指标。 在`./output/ser_layoutxlm/`文件夹中会保存训练日志，最优的模型和最新epoch的模型。
@@ -664,7 +660,7 @@ SER: 语义实体识别 (Semantic Entity Recognition）, 可以完成对图像
 
 
 ```python
-! CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
+CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
     -c configs/vqa/ser/layoutxlm.yml \
     -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/
 ```
@@ -684,7 +680,7 @@ SER: 语义实体识别 (Semantic Entity Recognition）, 可以完成对图像
 
 
 ```python
-! CUDA_VISIBLE_DEVICES=0 python tools/infer_vqa_token_ser.py \
+CUDA_VISIBLE_DEVICES=0 python tools/infer_vqa_token_ser.py \
     -c configs/vqa/ser/layoutxlm.yml  \
     -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/ \
     Global.infer_img=doc/vqa/input/zh_val_42.jpg
@@ -704,7 +700,7 @@ plt.figure(figsize=(48,24))
 plt.imshow(img)
 ```
 
-## 5.2 RE
+### 5.2 RE
 
 基于 RE 任务，可以完成对图象中的文本内容的关系提取，如判断问题对(pair)。
 
@@ -729,7 +725,7 @@ plt.imshow(img)
 
 
 ```python
-! CUDA_VISIBLE_DEVICES=0 python3 tools/train.py -c configs/vqa/re/layoutxlm.yml
+CUDA_VISIBLE_DEVICES=0 python3 tools/train.py -c configs/vqa/re/layoutxlm.yml
 ```
 
 最终会打印出`precision`, `recall`, `hmean`等指标。 在`./output/re_layoutxlm/`文件夹中会保存训练日志，最优的模型和最新epoch的模型
@@ -744,7 +740,7 @@ plt.imshow(img)
 
 
 ```python
-! CUDA_VISIBLE_DEVICES=0 python3 tools/eval.py \
+CUDA_VISIBLE_DEVICES=0 python3 tools/eval.py \
     -c configs/vqa/re/layoutxlm.yml \
     -o Architecture.Backbone.checkpoints=pretrain/re_LayoutXLM_xfun_zh/
 ```
@@ -760,20 +756,14 @@ plt.imshow(img)
 
 <center>图26 RE-模型预测</center>
 
-    使用OCR引擎 + SER + RE串联预测
-
-使用如下命令即可完成OCR引擎 + SER + RE的串联预测, 以预训练SER和RE模型为例：
-
-
+使用如下命令即可完成OCR引擎 + SER + RE的串联预测, 以预训练SER和RE模型为例，
 
 最终会在config.Global.save_res_path字段所配置的目录下保存预测结果可视化图像以及预测结果文本文件，预测结果文本文件名为infer_results.txt。
 
 
-
-
 ```python
-%cd /home/aistudio/PaddleOCR
-! CUDA_VISIBLE_DEVICES=0 python3 tools/infer_vqa_token_ser_re.py \
+cd /home/aistudio/PaddleOCR
+CUDA_VISIBLE_DEVICES=0 python3 tools/infer_vqa_token_ser_re.py \
     -c configs/vqa/re/layoutxlm.yml \
     -o Architecture.Backbone.checkpoints=pretrain/re_LayoutXLM_xfun_zh/ \
     Global.infer_img=test_imgs/ \
@@ -787,10 +777,9 @@ plt.imshow(img)
 test_imgs/t131.jpg	{"政治面税": "群众", "性别": "男", "籍贯": "河北省邯郸市", "婚姻状况": "亏末婚口已婚口已娇", "通讯地址": "邯郸市阳光苑7号楼003", "民族": "汉族", "毕业院校": "河南工业大学", "户口性质": "口农村城镇", "户口地址": "河北省邯郸市", "联系电话": "13288888888", "健康状况": "健康", "姓名": "小六", "好高cm": "180", "出生年月": "1996年8月9日", "文化程度": "本科", "身份证号码": "458933777777777777"}
 ````
 
-
+展示预测结果
 
 ```python
-# 展示预测结果
 import cv2
 from matplotlib import pyplot as plt
 %matplotlib inline
@@ -800,7 +789,7 @@ plt.figure(figsize=(48,24))
 plt.imshow(img)
 ```
 
-# 6 导出Excel
+## 6 导出Excel
 
 <center><img src="https://ai-studio-static-online.cdn.bcebos.com/ab93d3d90d77437a81c9534b2dd1d3e39ef81e8473054fd3aeff6e837ebfb827"></center>
 <center>图27 导出Excel</center>
@@ -859,7 +848,15 @@ with open('output/re/infer_results.txt', 'r', encoding='utf-8') as fin:
 workbook.close()
 ```
 
-# 更多资源
+## 获得模型
+
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型推理
+
+## 更多资源
 
 - 更多深度学习知识、产业案例、面试宝典等，请参考：[awesome-DeepLearning](https://github.com/paddlepaddle/awesome-DeepLearning)
 
@@ -869,7 +866,7 @@ workbook.close()
 
 - 飞桨框架相关资料，请参考：[飞桨深度学习平台](https://www.paddlepaddle.org.cn/?fr=paddleEdu_aistudio)
 
-# 参考链接
+## 参考链接
 
 -  LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, https://arxiv.org/pdf/2104.08836.pdf
 

From 079cdf980cccaf54302040562fafde94b9a9181e Mon Sep 17 00:00:00 2001
From: zhoujun <zjwenmu@gmail.com>
Date: Mon, 27 Jun 2022 00:49:00 +0800
Subject: [PATCH 053/134] update save model func (#6693)

* save latest metric

* save latest metric

* add boader judge

* add boader judge
---
 PPOCRLabel/libs/shape.py        |  1 +
 applications/轻量级车牌识别.md  | 11 ++++++-----
 ppocr/data/imaug/fce_targets.py |  9 +++++++--
 ppocr/utils/save_load.py        |  4 ++--
 4 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/PPOCRLabel/libs/shape.py b/PPOCRLabel/libs/shape.py
index 18cc4a8e76..97e2eb7238 100644
--- a/PPOCRLabel/libs/shape.py
+++ b/PPOCRLabel/libs/shape.py
@@ -48,6 +48,7 @@ class Shape(object):
 
     def __init__(self, label=None, line_color=None, difficult=False, key_cls="None", paintLabel=False):
         self.label = label
+        self.idx = 0
         self.points = []
         self.fill = False
         self.selected = False
diff --git a/applications/轻量级车牌识别.md b/applications/轻量级车牌识别.md
index 7012c7f4bb..1a63091b92 100644
--- a/applications/轻量级车牌识别.md
+++ b/applications/轻量级车牌识别.md
@@ -311,7 +311,6 @@ python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o
 
 在上述命令中，通过`-o`的方式修改了配置文件中的参数。
 
-训练好的模型地址为: [det_ppocr_v3_finetune.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_finetune.tar)
 
 **评估**
 
@@ -354,8 +353,6 @@ python3.7 deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCR
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/det.txt]
 ```
 
-训练好的模型地址为: [det_ppocr_v3_quant.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/det_ppocr_v3_quant.tar)
-
 量化后指标对比如下
 
 |方案|hmeans| 模型大小 | 预测速度(lite) |
@@ -436,6 +433,12 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
 ```
 
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
 评估部分日志如下：
 ```bash
 [2022/05/12 19:52:02] ppocr INFO: load pretrain successful from models/ch_PP-OCRv3_rec_train/best_accuracy
@@ -528,7 +531,6 @@ python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec.yml -o \
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
 ```
-训练好的模型地址为: [rec_ppocr_v3_finetune.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_finetune.tar)
 
 **评估**
 
@@ -570,7 +572,6 @@ python3.7 deploy/slim/quantization/quant.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_
     Eval.dataset.data_dir=/home/aistudio/data/CCPD2020/PPOCR \
     Eval.dataset.label_file_list=[/home/aistudio/data/CCPD2020/PPOCR/test/rec.txt]
 ```
-训练好的模型地址为: [rec_ppocr_v3_quant.tar](https://paddleocr.bj.bcebos.com/fanliku/license_plate_recognition/rec_ppocr_v3_quant.tar)
 
 量化后指标对比如下
 
diff --git a/ppocr/data/imaug/fce_targets.py b/ppocr/data/imaug/fce_targets.py
index 4d1903c0a7..8c64276e26 100644
--- a/ppocr/data/imaug/fce_targets.py
+++ b/ppocr/data/imaug/fce_targets.py
@@ -107,17 +107,20 @@ class FCENetTargets:
         for i in range(1, n):
             current_line_len = i * delta_length
 
-            while current_line_len >= length_cumsum[current_edge_ind + 1]:
+            while current_edge_ind + 1 < len(length_cumsum) and current_line_len >= length_cumsum[current_edge_ind + 1]:
                 current_edge_ind += 1
+
             current_edge_end_shift = current_line_len - length_cumsum[
                 current_edge_ind]
+
+            if current_edge_ind >= len(length_list):
+                break
             end_shift_ratio = current_edge_end_shift / length_list[
                 current_edge_ind]
             current_point = line[current_edge_ind] + (line[current_edge_ind + 1]
                                                       - line[current_edge_ind]
                                                       ) * end_shift_ratio
             resampled_line.append(current_point)
-
         resampled_line.append(line[-1])
         resampled_line = np.array(resampled_line)
 
@@ -328,6 +331,8 @@ class FCENetTargets:
             resampled_top_line, resampled_bot_line = self.resample_sidelines(
                 top_line, bot_line, self.resample_step)
             resampled_bot_line = resampled_bot_line[::-1]
+            if len(resampled_top_line) != len(resampled_bot_line):
+                continue
             center_line = (resampled_top_line + resampled_bot_line) / 2
 
             line_head_shrink_len = norm(resampled_top_line[0] -
diff --git a/ppocr/utils/save_load.py b/ppocr/utils/save_load.py
index b09f1db6e9..3647111fdd 100644
--- a/ppocr/utils/save_load.py
+++ b/ppocr/utils/save_load.py
@@ -177,9 +177,9 @@ def save_model(model,
             model.backbone.model.save_pretrained(model_prefix)
         metric_prefix = os.path.join(model_prefix, 'metric')
     # save metric and config
+    with open(metric_prefix + '.states', 'wb') as f:
+        pickle.dump(kwargs, f, protocol=2)
     if is_best:
-        with open(metric_prefix + '.states', 'wb') as f:
-            pickle.dump(kwargs, f, protocol=2)
         logger.info('save best model is to {}'.format(model_prefix))
     else:
         logger.info("save model in {}".format(model_prefix))

From d554037ece782db092cb46394ad4aface7834c0c Mon Sep 17 00:00:00 2001
From: an1018 <614803115@qq.com>
Date: Mon, 27 Jun 2022 10:04:39 +0800
Subject: [PATCH 054/134] update

---
 applications/多模态表单识别.md | 29 +++++++++++++++++++++++++++--
 1 file changed, 27 insertions(+), 2 deletions(-)

diff --git a/applications/多模态表单识别.md b/applications/多模态表单识别.md
index 457d397030..27f981101a 100644
--- a/applications/多模态表单识别.md
+++ b/applications/多模态表单识别.md
@@ -1,6 +1,31 @@
 # 多模态表单识别
-
-[TOC]
+- [多模态表单识别](#多模态表单识别)
+  - [1 项目说明](#1-项目说明)
+  - [2 安装说明](#2-安装说明)
+  - [3 数据准备](#3-数据准备)
+    - [3.1 下载处理好的数据集](#31-下载处理好的数据集)
+    - [3.2 转换为PaddleOCR检测和识别格式](#32-转换为paddleocr检测和识别格式)
+  - [4 OCR](#4-ocr)
+    - [4.1 文本检测](#41-文本检测)
+      - [**4.1.1 方案1：预训练模型**](#411-方案1预训练模型)
+      - [**4.1.2 方案2：XFUND数据集+fine-tune**](#412-方案2xfund数据集fine-tune)
+    - [4.2 文本识别](#42-文本识别)
+      - [**4.2.1 方案1：预训练模型**](#421-方案1预训练模型)
+      - [**4.2.2 方案2：XFUND数据集+finetune**](#422-方案2xfund数据集finetune)
+      - [**4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据**](#423-方案3xfund数据集finetune真实通用识别数据)
+  - [5 文档视觉问答(DOC-VQA)](#5-文档视觉问答doc-vqa)
+    - [5.1 SER](#51-ser)
+      - [5.1.1 模型训练](#511-模型训练)
+      - [5.1.2 模型评估](#512-模型评估)
+      - [5.1.3 模型预测](#513-模型预测)
+    - [5.2 RE](#52-re)
+      - [5.2.1 模型训练](#521-模型训练)
+      - [5.2.2 模型评估](#522-模型评估)
+      - [5.2.3 模型预测](#523-模型预测)
+  - [6 导出Excel](#6-导出excel)
+  - [获得模型](#获得模型)
+  - [更多资源](#更多资源)
+  - [参考链接](#参考链接)
 
 ## 1 项目说明
 

From 1eba8ae3f89ac850fbec35f5d3b1c6e402dfd02a Mon Sep 17 00:00:00 2001
From: xiaoting <31891223+tink2123@users.noreply.github.com>
Date: Mon, 27 Jun 2022 14:51:33 +0800
Subject: [PATCH 055/134] =?UTF-8?q?Update=20=E5=85=89=E5=8A=9F=E7=8E=87?=
 =?UTF-8?q?=E8=AE=A1=E6=95=B0=E7=A0=81=E7=AE=A1=E5=AD=97=E7=AC=A6=E8=AF=86?=
 =?UTF-8?q?=E5=88=AB.md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 applications/光功率计数码管字符识别.md | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/applications/光功率计数码管字符识别.md b/applications/光功率计数码管字符识别.md
index 33c515a5e9..d6c11259bf 100644
--- a/applications/光功率计数码管字符识别.md
+++ b/applications/光功率计数码管字符识别.md
@@ -21,7 +21,12 @@
 
 第二步，使用文本识别模型获得准确的数字和单位信息。
 
-本项目主要介绍如何完成第二步文本识别部分。
+本项目主要介绍如何完成第二步文本识别部分，包括：
+
+* 真实评估集的建立
+* 训练数据的合成
+* 基于 PP-OCRv3 和 SVTR_Tiny 两个模型进行训练
+* 评估并推理
 
 ## 2. PaddleOCR 快速使用
 
@@ -400,7 +405,11 @@ python tools/train.py -c rec_svtr_tiny_none_ctc_ch_train/rec_svtr_tiny_6local_6g
 
 ### 3.4 验证效果
 
-**本例中的评估集和模型路径，可通过填写[问卷](todo:add link)获得**
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型推理
 
 * 指标评估
 

From 5f990d9cda3656ca8992f9e6473f25f13eee4ff3 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 08:11:31 +0000
Subject: [PATCH 056/134] add yml

---
 configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml
index 3833cb0bad..df429314cd 100644
--- a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml
@@ -28,7 +28,7 @@ Architecture:
       algorithm: DB
       Transform:
       Backbone:
-        name: ResNet
+        name: ResNet_vd
         layers: 18
       Neck:
         name: DBFPN

From 5ba74d56277debf930dc10dd6105e84f55247b56 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 08:16:10 +0000
Subject: [PATCH 057/134] add yml

---
 configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml
index b4644244f8..d24ee11f3d 100644
--- a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml
@@ -45,7 +45,7 @@ Architecture:
       algorithm: DB
       Transform:
       Backbone:
-        name: ResNet
+        name: ResNet_vd
         layers: 18
       Neck:
         name: DBFPN

From bc29ec0ba8e72df1db81d0dfed57a337331660ba Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 08:17:49 +0000
Subject: [PATCH 058/134] add yml

---
 configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml b/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml
index 3e77577c17..ef58befd69 100644
--- a/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml
+++ b/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml
@@ -61,7 +61,7 @@ Architecture:
       model_type: det
       algorithm: DB
       Backbone:
-        name: ResNet
+        name: ResNet_vd
         in_channels: 3
         layers: 50
       Neck:

From 961dca72cf3e4708d6ac57769a79fba1b56d5e2c Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 08:21:15 +0000
Subject: [PATCH 059/134] add yml

---
 configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml    | 4 ++--
 configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml | 2 +-
 configs/det/det_r50_vd_db.yml                      | 2 +-
 configs/det/det_r50_vd_dcn_fce_ctw.yml             | 2 +-
 configs/det/det_r50_vd_east.yml                    | 2 +-
 configs/det/det_r50_vd_pse.yml                     | 2 +-
 configs/det/det_res18_db_v2.0.yml                  | 2 +-
 7 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml b/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml
index 3f30ada13f..e1831f2639 100644
--- a/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml
+++ b/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml
@@ -25,7 +25,7 @@ Architecture:
       model_type: det
       algorithm: DB
       Backbone:
-        name: ResNet
+        name: ResNet_vd
         in_channels: 3
         layers: 50
       Neck:
@@ -40,7 +40,7 @@ Architecture:
       model_type: det
       algorithm: DB
       Backbone:
-        name: ResNet
+        name: ResNet_vd
         in_channels: 3
         layers: 50
       Neck:
diff --git a/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml b/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml
index 7b07ef9964..e983c221e2 100644
--- a/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml
+++ b/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml
@@ -20,7 +20,7 @@ Architecture:
   algorithm: DB
   Transform:
   Backbone:
-    name: ResNet
+    name: ResNet_vd
     layers: 18
     disable_se: True
   Neck:
diff --git a/configs/det/det_r50_vd_db.yml b/configs/det/det_r50_vd_db.yml
index ab67786ece..288dcc8c1a 100644
--- a/configs/det/det_r50_vd_db.yml
+++ b/configs/det/det_r50_vd_db.yml
@@ -20,7 +20,7 @@ Architecture:
   algorithm: DB
   Transform:
   Backbone:
-    name: ResNet
+    name: ResNet_vd
     layers: 50
   Neck:
     name: DBFPN
diff --git a/configs/det/det_r50_vd_dcn_fce_ctw.yml b/configs/det/det_r50_vd_dcn_fce_ctw.yml
index a9f7c4143d..3a4075b322 100755
--- a/configs/det/det_r50_vd_dcn_fce_ctw.yml
+++ b/configs/det/det_r50_vd_dcn_fce_ctw.yml
@@ -21,7 +21,7 @@ Architecture:
   algorithm: FCE
   Transform:
   Backbone:
-    name: ResNet
+    name: ResNet_vd
     layers: 50
     dcn_stage: [False, True, True, True]
     out_indices: [1,2,3]
diff --git a/configs/det/det_r50_vd_east.yml b/configs/det/det_r50_vd_east.yml
index e84a5fa7a7..af90ef0adb 100644
--- a/configs/det/det_r50_vd_east.yml
+++ b/configs/det/det_r50_vd_east.yml
@@ -20,7 +20,7 @@ Architecture:
   algorithm: EAST
   Transform:
   Backbone:
-    name: ResNet
+    name: ResNet_vd
     layers: 50
   Neck:
     name: EASTFPN
diff --git a/configs/det/det_r50_vd_pse.yml b/configs/det/det_r50_vd_pse.yml
index 8e77506c41..1a971564fd 100644
--- a/configs/det/det_r50_vd_pse.yml
+++ b/configs/det/det_r50_vd_pse.yml
@@ -20,7 +20,7 @@ Architecture:
   algorithm: PSE
   Transform:
   Backbone:
-    name: ResNet
+    name: ResNet_vd
     layers: 50
   Neck:
     name: FPN
diff --git a/configs/det/det_res18_db_v2.0.yml b/configs/det/det_res18_db_v2.0.yml
index 7b07ef9964..e983c221e2 100644
--- a/configs/det/det_res18_db_v2.0.yml
+++ b/configs/det/det_res18_db_v2.0.yml
@@ -20,7 +20,7 @@ Architecture:
   algorithm: DB
   Transform:
   Backbone:
-    name: ResNet
+    name: ResNet_vd
     layers: 18
     disable_se: True
   Neck:

From 142b5e9dfefe9f06a3b82067d8d0d50301d2f509 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 08:26:53 +0000
Subject: [PATCH 060/134] add db++

---
 ppocr/data/imaug/operators.py             |  7 +++
 ppocr/modeling/backbones/__init__.py      |  5 +-
 ppocr/modeling/backbones/det_resnet_vd.py | 12 +++--
 ppocr/modeling/necks/db_fpn.py            | 60 ++++++++++++++++++++++-
 ppocr/optimizer/learning_rate.py          | 35 +++++++++++++
 5 files changed, 111 insertions(+), 8 deletions(-)

diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
index 09736515e7..a83dc64838 100644
--- a/ppocr/data/imaug/operators.py
+++ b/ppocr/data/imaug/operators.py
@@ -238,9 +238,12 @@ class DetResizeForTest(object):
     def __init__(self, **kwargs):
         super(DetResizeForTest, self).__init__()
         self.resize_type = 0
+        self.keep_ratio = False
         if 'image_shape' in kwargs:
             self.image_shape = kwargs['image_shape']
             self.resize_type = 1
+            if 'keep_ratio' in kwargs:  ######
+                self.keep_ratio = kwargs['keep_ratio']  #######
         elif 'limit_side_len' in kwargs:
             self.limit_side_len = kwargs['limit_side_len']
             self.limit_type = kwargs.get('limit_type', 'min')
@@ -270,6 +273,10 @@ class DetResizeForTest(object):
     def resize_image_type1(self, img):
         resize_h, resize_w = self.image_shape
         ori_h, ori_w = img.shape[:2]  # (h, w, c)
+        if self.keep_ratio:  ########
+            resize_w = ori_w * resize_h / ori_h
+            N = math.ceil(resize_w / 32)
+            resize_w = N * 32
         ratio_h = float(resize_h) / ori_h
         ratio_w = float(resize_w) / ori_w
         img = cv2.resize(img, (int(resize_w), int(resize_h)))
diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
index 072d6e0f84..0d8e60e93a 100755
--- a/ppocr/modeling/backbones/__init__.py
+++ b/ppocr/modeling/backbones/__init__.py
@@ -18,9 +18,10 @@ __all__ = ["build_backbone"]
 def build_backbone(config, model_type):
     if model_type == "det" or model_type == "table":
         from .det_mobilenet_v3 import MobileNetV3
-        from .det_resnet_vd import ResNet
+        from .det_resnet import ResNet
+        from .det_resnet_vd import ResNet_vd
         from .det_resnet_vd_sast import ResNet_SAST
-        support_dict = ["MobileNetV3", "ResNet", "ResNet_SAST"]
+        support_dict = ["MobileNetV3", "ResNet", "ResNet_vd", "ResNet_SAST"]
     elif model_type == "rec" or model_type == "cls":
         from .rec_mobilenet_v3 import MobileNetV3
         from .rec_resnet_vd import ResNet
diff --git a/ppocr/modeling/backbones/det_resnet_vd.py b/ppocr/modeling/backbones/det_resnet_vd.py
index 8c955a4af3..5337e14c40 100644
--- a/ppocr/modeling/backbones/det_resnet_vd.py
+++ b/ppocr/modeling/backbones/det_resnet_vd.py
@@ -25,7 +25,7 @@ from paddle.vision.ops import DeformConv2D
 from paddle.regularizer import L2Decay
 from paddle.nn.initializer import Normal, Constant, XavierUniform
 
-__all__ = ["ResNet"]
+__all__ = ["ResNet_vd", "ConvBNLayer", "DeformableConvV2"]
 
 
 class DeformableConvV2(nn.Layer):
@@ -104,6 +104,7 @@ class ConvBNLayer(nn.Layer):
                  kernel_size,
                  stride=1,
                  groups=1,
+                 dcn_groups=1,
                  is_vd_mode=False,
                  act=None,
                  is_dcn=False):
@@ -128,7 +129,7 @@ class ConvBNLayer(nn.Layer):
                 kernel_size=kernel_size,
                 stride=stride,
                 padding=(kernel_size - 1) // 2,
-                groups=2,  #groups,
+                groups=dcn_groups,  #groups,
                 bias_attr=False)
         self._batch_norm = nn.BatchNorm(out_channels, act=act)
 
@@ -162,7 +163,8 @@ class BottleneckBlock(nn.Layer):
             kernel_size=3,
             stride=stride,
             act='relu',
-            is_dcn=is_dcn)
+            is_dcn=is_dcn,
+            dcn_groups=2)
         self.conv2 = ConvBNLayer(
             in_channels=out_channels,
             out_channels=out_channels * 4,
@@ -238,14 +240,14 @@ class BasicBlock(nn.Layer):
         return y
 
 
-class ResNet(nn.Layer):
+class ResNet_vd(nn.Layer):
     def __init__(self,
                  in_channels=3,
                  layers=50,
                  dcn_stage=None,
                  out_indices=None,
                  **kwargs):
-        super(ResNet, self).__init__()
+        super(ResNet_vd, self).__init__()
 
         self.layers = layers
         supported_layers = [18, 34, 50, 101, 152, 200]
diff --git a/ppocr/modeling/necks/db_fpn.py b/ppocr/modeling/necks/db_fpn.py
index 93ed2dbfd1..b46d7c460d 100644
--- a/ppocr/modeling/necks/db_fpn.py
+++ b/ppocr/modeling/necks/db_fpn.py
@@ -105,9 +105,10 @@ class DSConv(nn.Layer):
 
 
 class DBFPN(nn.Layer):
-    def __init__(self, in_channels, out_channels, **kwargs):
+    def __init__(self, in_channels, out_channels, use_asf=None, **kwargs):
         super(DBFPN, self).__init__()
         self.out_channels = out_channels
+        self.use_asf = use_asf
         weight_attr = paddle.nn.initializer.KaimingUniform()
 
         self.in2_conv = nn.Conv2D(
@@ -163,6 +164,9 @@ class DBFPN(nn.Layer):
             weight_attr=ParamAttr(initializer=weight_attr),
             bias_attr=False)
 
+        if self.use_asf:
+            self.asf = ASFBlock(self.out_channels, self.out_channels // 4)
+
     def forward(self, x):
         c2, c3, c4, c5 = x
 
@@ -187,6 +191,10 @@ class DBFPN(nn.Layer):
         p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
 
         fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+
+        if self.use_asf:
+            fuse = self.asf(fuse, [p5, p4, p3, p2])
+
         return fuse
 
 
@@ -356,3 +364,53 @@ class LKPAN(nn.Layer):
 
         fuse = paddle.concat([p5, p4, p3, p2], axis=1)
         return fuse
+
+
+class ASFBlock(nn.Layer):
+    def __init__(self, in_channels, inter_channels, out_features_num=4):
+        super(ASFBlock, self).__init__()
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+        self.in_channels = in_channels
+        self.inter_channels = inter_channels
+        self.out_features_num = out_features_num
+        self.conv = nn.Conv2D(in_channels, inter_channels, 3, padding=1)
+
+        self.attention_block_1 = nn.Sequential(
+            #Nx1xHxW
+            nn.Conv2D(
+                1,
+                1,
+                3,
+                bias_attr=False,
+                padding=1,
+                weight_attr=ParamAttr(initializer=weight_attr)),
+            nn.ReLU(),
+            nn.Conv2D(
+                1,
+                1,
+                1,
+                bias_attr=False,
+                weight_attr=ParamAttr(initializer=weight_attr)),
+            nn.Sigmoid())
+
+        self.attention_block_2 = nn.Sequential(
+            nn.Conv2D(
+                inter_channels,
+                out_features_num,
+                1,
+                bias_attr=False,
+                weight_attr=ParamAttr(initializer=weight_attr)),
+            nn.Sigmoid())
+
+    def forward(self, fuse_features, features_list):
+        fuse_features = self.conv(fuse_features)
+        attention_scores = self.attention_block_1(
+            paddle.mean(
+                fuse_features, axis=1, keepdim=True)) + fuse_features
+        attention_scores = self.attention_block_2(attention_scores)
+        assert len(features_list) == self.out_features_num
+
+        out_list = []
+        for i in range(self.out_features_num):
+            out_list.append(attention_scores[:, i:i + 1] * features_list[i])
+        return paddle.concat(out_list, axis=1)
diff --git a/ppocr/optimizer/learning_rate.py b/ppocr/optimizer/learning_rate.py
index fe251f36e7..8e05a84011 100644
--- a/ppocr/optimizer/learning_rate.py
+++ b/ppocr/optimizer/learning_rate.py
@@ -308,3 +308,38 @@ class Const(object):
                 end_lr=self.learning_rate,
                 last_epoch=self.last_epoch)
         return learning_rate
+
+
+class DecayLearningRate(object):
+    """
+    DecayLearningRate learning rate decay
+    new_lr = (lr - end_lr) * (1 - epoch/decay_steps)**power + end_lr
+    Args:
+        learning_rate(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        factor(float): Power of polynomial, should greater than 0.0 to get learning rate decay. Default: 0.9
+        end_lr(float): The minimum final learning rate. Default: 0.0.
+    """
+
+    def __init__(self,
+                 learning_rate,
+                 step_each_epoch,
+                 epochs,
+                 factor=0.9,
+                 end_lr=0,
+                 **kwargs):
+        super(DecayLearningRate, self).__init__()
+        self.learning_rate = learning_rate
+        self.epochs = epochs + 1
+        self.factor = factor
+        self.end_lr = 0
+        self.decay_steps = step_each_epoch * epochs
+
+    def __call__(self):
+        learning_rate = lr.PolynomialDecay(
+            learning_rate=self.learning_rate,
+            decay_steps=self.decay_steps,
+            power=self.factor,
+            end_lr=self.end_lr)
+        return learning_rate

From 26a89db76a71434dd1a759b1f4dcce36047b1fd0 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 08:27:26 +0000
Subject: [PATCH 061/134] add resnet

---
 configs/det/det_r50_db++_td_tr.yml     | 166 +++++++++++++++++
 ppocr/modeling/backbones/det_resnet.py | 237 +++++++++++++++++++++++++
 2 files changed, 403 insertions(+)
 create mode 100644 configs/det/det_r50_db++_td_tr.yml
 create mode 100644 ppocr/modeling/backbones/det_resnet.py

diff --git a/configs/det/det_r50_db++_td_tr.yml b/configs/det/det_r50_db++_td_tr.yml
new file mode 100644
index 0000000000..4954e1abb6
--- /dev/null
+++ b/configs/det/det_r50_db++_td_tr.yml
@@ -0,0 +1,166 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 1000
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_td_tr/
+  save_epoch_step: 200
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: false
+  pretrained_model: ./pretrain_models/synthtext_pretrained_res50_dcn_asf_spatial
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: ResNet
+    layers: 50
+    dcn_stage: [False, True, True, True]
+  Neck:
+    name: DBFPN
+    out_channels: 256
+    use_asf: True
+  Head:
+    name: DBHead
+    k: 50
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: BCELoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  lr:
+    name: DecayLearningRate
+    learning_rate: 0.007
+    epochs: 1000
+    factor: 0.9
+    end_lr: 0
+  weight_decay: 0.0001
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.5
+  max_candidates: 1000
+  unclip_ratio: 1.5
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/                                                                                 
+    label_file_list:                                                                                            
+    - ./train_data/TD_TR/TD500/train_gt_labels.txt                                                           
+    - ./train_data/TD_TR/TR400/gt_labels.txt                                                                 
+    ratio_list:                                                                                                 
+    - 1.0                                                                                                       
+    - 1.0 
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 10
+        keep_ratio: true
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 4
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/                                                                          
+    label_file_list:                                                                                            
+    - ./train_data/TD_TR/TD500/test_gt_labels.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        image_shape:
+        - 736
+        - 736
+        keep_ratio: True
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/ppocr/modeling/backbones/det_resnet.py b/ppocr/modeling/backbones/det_resnet.py
new file mode 100644
index 0000000000..4c75b663af
--- /dev/null
+++ b/ppocr/modeling/backbones/det_resnet.py
@@ -0,0 +1,237 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
+from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
+from paddle.nn.initializer import Uniform
+
+import math
+
+from paddle.vision.ops import DeformConv2D
+from paddle.regularizer import L2Decay
+from paddle.nn.initializer import Normal, Constant, XavierUniform
+from .det_resnet_vd import DeformableConvV2, ConvBNLayer
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(self,
+                 num_channels,
+                 num_filters,
+                 stride,
+                 shortcut=True,
+                 is_dcn=False):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=1,
+            act="relu", )
+        self.conv1 = ConvBNLayer(
+            in_channels=num_filters,
+            out_channels=num_filters,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            is_dcn=is_dcn,
+            dcn_groups=1, )
+        self.conv2 = ConvBNLayer(
+            in_channels=num_filters,
+            out_channels=num_filters * 4,
+            kernel_size=1,
+            act=None, )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=num_channels,
+                out_channels=num_filters * 4,
+                kernel_size=1,
+                stride=stride, )
+
+        self.shortcut = shortcut
+
+        self._num_channels_out = num_filters * 4
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(self,
+                 num_channels,
+                 num_filters,
+                 stride,
+                 shortcut=True,
+                 name=None):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=3,
+            stride=stride,
+            act="relu")
+        self.conv1 = ConvBNLayer(
+            in_channels=num_filters,
+            out_channels=num_filters,
+            kernel_size=3,
+            act=None)
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=num_channels,
+                out_channels=num_filters,
+                kernel_size=1,
+                stride=stride)
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet(nn.Layer):
+    def __init__(self,
+                 in_channels=3,
+                 layers=50,
+                 out_indices=None,
+                 dcn_stage=None):
+        super(ResNet, self).__init__()
+
+        self.layers = layers
+        self.input_image_channel = in_channels
+
+        supported_layers = [18, 34, 50, 101, 152]
+        assert layers in supported_layers, \
+            "supported layers are {} but input layer is {}".format(
+                supported_layers, layers)
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            depth = [3, 4, 6, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        num_channels = [64, 256, 512,
+                        1024] if layers >= 50 else [64, 64, 128, 256]
+        num_filters = [64, 128, 256, 512]
+
+        self.dcn_stage = dcn_stage if dcn_stage is not None else [
+            False, False, False, False
+        ]
+        self.out_indices = out_indices if out_indices is not None else [
+            0, 1, 2, 3
+        ]
+
+        self.conv = ConvBNLayer(
+            in_channels=self.input_image_channel,
+            out_channels=64,
+            kernel_size=7,
+            stride=2,
+            act="relu", )
+        self.pool2d_max = MaxPool2D(
+            kernel_size=3,
+            stride=2,
+            padding=1, )
+
+        self.stages = []
+        self.out_channels = []
+        if layers >= 50:
+            for block in range(len(depth)):
+                shortcut = False
+                block_list = []
+                is_dcn = self.dcn_stage[block]
+                for i in range(depth[block]):
+                    if layers in [101, 152] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+                    bottleneck_block = self.add_sublayer(
+                        conv_name,
+                        BottleneckBlock(
+                            num_channels=num_channels[block]
+                            if i == 0 else num_filters[block] * 4,
+                            num_filters=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            is_dcn=is_dcn))
+                    block_list.append(bottleneck_block)
+                    shortcut = True
+                if block in self.out_indices:
+                    self.out_channels.append(num_filters[block] * 4)
+                self.stages.append(nn.Sequential(*block_list))
+        else:
+            for block in range(len(depth)):
+                shortcut = False
+                block_list = []
+                # is_dcn = self.dcn_stage[block]
+                for i in range(depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    basic_block = self.add_sublayer(
+                        conv_name,
+                        BasicBlock(
+                            num_channels=num_channels[block]
+                            if i == 0 else num_filters[block],
+                            num_filters=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut))
+                    block_list.append(basic_block)
+                    shortcut = True
+                if block in self.out_indices:
+                    self.out_channels.append(num_filters[block])
+                self.stages.append(nn.Sequential(*block_list))
+
+    def forward(self, inputs):
+        y = self.conv(inputs)
+        y = self.pool2d_max(y)
+        out = []
+        for i, block in enumerate(self.stages):
+            y = block(y)
+            if i in self.out_indices:
+                out.append(y)
+        return out

From 6e6d77432fb2f77af6b30e638f2d6eeef974ee12 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Mon, 27 Jun 2022 09:01:38 +0000
Subject: [PATCH 062/134] fix ci export bug

---
 configs/rec/rec_r45_abinet.yml                |  2 --
 configs/rec/rec_svtrnet.yml                   |  2 --
 configs/rec/rec_svtrnet_ch.yml                |  4 ----
 ppocr/data/imaug/rec_img_aug.py               | 17 ++---------------
 test_tipc/configs/rec_svtrnet/rec_svtrnet.yml |  6 ++----
 5 files changed, 4 insertions(+), 27 deletions(-)

diff --git a/configs/rec/rec_r45_abinet.yml b/configs/rec/rec_r45_abinet.yml
index 3cf4cddb04..1e03db8b1b 100644
--- a/configs/rec/rec_r45_abinet.yml
+++ b/configs/rec/rec_r45_abinet.yml
@@ -72,7 +72,6 @@ Train:
           ignore_index: *ignore_index
       - ABINetRecResizeImg:
           image_shape: [3, 32, 128]
-          padding: False
       - KeepKeys:
           keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
   loader:
@@ -93,7 +92,6 @@ Eval:
           ignore_index: *ignore_index
       - ABINetRecResizeImg:
           image_shape: [3, 32, 128]
-          padding: False
       - KeepKeys:
           keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
   loader:
diff --git a/configs/rec/rec_svtrnet.yml b/configs/rec/rec_svtrnet.yml
index 5896b10515..c1f5cc380a 100644
--- a/configs/rec/rec_svtrnet.yml
+++ b/configs/rec/rec_svtrnet.yml
@@ -84,7 +84,6 @@ Train:
           channel_first: False
       - CTCLabelEncode: # Class handling label
       - SVTRRecResizeImg:
-          character_dict_path:
           image_shape: [3, 64, 256]
           padding: False
       - KeepKeys:
@@ -105,7 +104,6 @@ Eval:
           channel_first: False
       - CTCLabelEncode: # Class handling label
       - SVTRRecResizeImg:
-          character_dict_path:
           image_shape: [3, 64, 256]
           padding: False
       - KeepKeys:
diff --git a/configs/rec/rec_svtrnet_ch.yml b/configs/rec/rec_svtrnet_ch.yml
index c4ce4ccc16..0d3f63d125 100644
--- a/configs/rec/rec_svtrnet_ch.yml
+++ b/configs/rec/rec_svtrnet_ch.yml
@@ -110,8 +110,6 @@ Train:
     - RecAug: null
     - CTCLabelEncode: null
     - SVTRRecResizeImg:
-        character_dict_path: null
-        infer_mode: False
         image_shape:
         - 3
         - 32
@@ -139,8 +137,6 @@ Eval:
         channel_first: false
     - CTCLabelEncode: null
     - SVTRRecResizeImg:
-        character_dict_path: null
-        infer_mode: False
         image_shape:
         - 3
         - 32
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index 1de8e6ee3c..26773d0a51 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -312,14 +312,8 @@ class GrayRecResizeImg(object):
 
 
 class ABINetRecResizeImg(object):
-    def __init__(self,
-                 image_shape,
-                 infer_mode=False,
-                 character_dict_path=None,
-                 **kwargs):
+    def __init__(self, image_shape, **kwargs):
         self.image_shape = image_shape
-        self.infer_mode = infer_mode
-        self.character_dict_path = character_dict_path
 
     def __call__(self, data):
         img = data['image']
@@ -330,15 +324,8 @@ class ABINetRecResizeImg(object):
 
 
 class SVTRRecResizeImg(object):
-    def __init__(self,
-                 image_shape,
-                 infer_mode=False,
-                 character_dict_path='./ppocr/utils/ppocr_keys_v1.txt',
-                 padding=True,
-                 **kwargs):
+    def __init__(self, image_shape, padding=True, **kwargs):
         self.image_shape = image_shape
-        self.infer_mode = infer_mode
-        self.character_dict_path = character_dict_path
         self.padding = padding
 
     def __call__(self, data):
diff --git a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
index bffd8ac09a..140b17e0e7 100644
--- a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
+++ b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
@@ -84,8 +84,7 @@ Train:
           img_mode: BGR
           channel_first: False
       - CTCLabelEncode: # Class handling label
-      - RecResizeImg:
-          character_dict_path:
+      - SVTRRecResizeImg:
           image_shape: [3, 64, 256]
           padding: False
       - KeepKeys:
@@ -106,8 +105,7 @@ Eval:
           img_mode: BGR
           channel_first: False
       - CTCLabelEncode: # Class handling label
-      - RecResizeImg:
-          character_dict_path:
+      - SVTRRecResizeImg:
           image_shape: [3, 64, 256]
           padding: False
       - KeepKeys:

From 5dfcc98306ff5482f230ca0929ef0fc3c4697639 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 27 Jun 2022 09:37:07 +0000
Subject: [PATCH 063/134] fix bug

---
 ppocr/data/imaug/label_ops.py          | 12 ++++--
 ppocr/losses/table_att_loss.py         | 55 +++++++++++++++-----------
 ppocr/metrics/table_metric.py          |  6 ++-
 ppocr/modeling/heads/table_att_head.py | 15 ++++---
 4 files changed, 54 insertions(+), 34 deletions(-)

diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index a55869a641..d63901a03d 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -591,7 +591,7 @@ class TableLabelEncode(AttnLabelEncode):
                  replace_empty_cell_token=False,
                  merge_no_span_structure=False,
                  learn_empty_box=False,
-                 point_num=4,
+                 point_num=2,
                  **kwargs):
         self.max_text_len = max_text_length
         self.lower = False
@@ -669,13 +669,15 @@ class TableLabelEncode(AttnLabelEncode):
 
         # encode box
         bboxes = np.zeros(
-            (self._max_text_len, self.point_num), dtype=np.float32)
+            (self._max_text_len, self.point_num * 2), dtype=np.float32)
         bbox_masks = np.zeros((self._max_text_len, 1), dtype=np.float32)
 
         bbox_idx = 0
+
         for i, token in enumerate(structure):
             if self.idx2char[token] in self.td_token:
-                if 'bbox' in cells[bbox_idx]:
+                if 'bbox' in cells[bbox_idx] and len(cells[bbox_idx][
+                        'tokens']) > 0:
                     bbox = cells[bbox_idx]['bbox'].copy()
                     bbox = np.array(bbox, dtype=np.float32).reshape(-1)
                     bboxes[i] = bbox
@@ -723,11 +725,13 @@ class TableMasterLabelEncode(TableLabelEncode):
                  replace_empty_cell_token=False,
                  merge_no_span_structure=False,
                  learn_empty_box=False,
-                 point_num=4,
+                 point_num=2,
                  **kwargs):
         super(TableMasterLabelEncode, self).__init__(
             max_text_length, character_dict_path, replace_empty_cell_token,
             merge_no_span_structure, learn_empty_box, point_num, **kwargs)
+        self.pad_idx = self.dict[self.pad_str]
+        self.unknown_idx = self.dict[self.unknown_str]
 
     @property
     def _max_text_len(self):
diff --git a/ppocr/losses/table_att_loss.py b/ppocr/losses/table_att_loss.py
index d7fd99e695..4bdccad399 100644
--- a/ppocr/losses/table_att_loss.py
+++ b/ppocr/losses/table_att_loss.py
@@ -21,15 +21,21 @@ from paddle import nn
 from paddle.nn import functional as F
 from paddle import fluid
 
+
 class TableAttentionLoss(nn.Layer):
-    def __init__(self, structure_weight, loc_weight, use_giou=False, giou_weight=1.0, **kwargs):
+    def __init__(self,
+                 structure_weight,
+                 loc_weight,
+                 use_giou=False,
+                 giou_weight=1.0,
+                 **kwargs):
         super(TableAttentionLoss, self).__init__()
         self.loss_func = nn.CrossEntropyLoss(weight=None, reduction='none')
         self.structure_weight = structure_weight
         self.loc_weight = loc_weight
         self.use_giou = use_giou
         self.giou_weight = giou_weight
-        
+
     def giou_loss(self, preds, bbox, eps=1e-7, reduction='mean'):
         '''
         :param preds:[[x1,y1,x2,y2], [x1,y1,x2,y2],,,]
@@ -48,9 +54,10 @@ class TableAttentionLoss(nn.Layer):
         inters = iw * ih
 
         # union
-        uni = (preds[:, 2] - preds[:, 0] + 1e-3) * (preds[:, 3] - preds[:, 1] + 1e-3
-            ) + (bbox[:, 2] - bbox[:, 0] + 1e-3) * (
-            bbox[:, 3] - bbox[:, 1] + 1e-3) - inters + eps
+        uni = (preds[:, 2] - preds[:, 0] + 1e-3) * (
+            preds[:, 3] - preds[:, 1] + 1e-3) + (bbox[:, 2] - bbox[:, 0] + 1e-3
+                                                 ) * (bbox[:, 3] - bbox[:, 1] +
+                                                      1e-3) - inters + eps
 
         # ious
         ious = inters / uni
@@ -80,30 +87,34 @@ class TableAttentionLoss(nn.Layer):
         structure_probs = predicts['structure_probs']
         structure_targets = batch[1].astype("int64")
         structure_targets = structure_targets[:, 1:]
-        if len(batch) == 6:
-            structure_mask = batch[5].astype("int64")
-            structure_mask = structure_mask[:, 1:]
-            structure_mask = paddle.reshape(structure_mask, [-1])
-        structure_probs = paddle.reshape(structure_probs, [-1, structure_probs.shape[-1]])
+        structure_probs = paddle.reshape(structure_probs,
+                                         [-1, structure_probs.shape[-1]])
         structure_targets = paddle.reshape(structure_targets, [-1])
         structure_loss = self.loss_func(structure_probs, structure_targets)
-        
-        if len(batch) == 6:
-             structure_loss = structure_loss * structure_mask
-            
-#         structure_loss = paddle.sum(structure_loss) * self.structure_weight
+
         structure_loss = paddle.mean(structure_loss) * self.structure_weight
-        
+
         loc_preds = predicts['loc_preds']
         loc_targets = batch[2].astype("float32")
-        loc_targets_mask = batch[4].astype("float32")
+        loc_targets_mask = batch[3].astype("float32")
         loc_targets = loc_targets[:, 1:, :]
         loc_targets_mask = loc_targets_mask[:, 1:, :]
-        loc_loss = F.mse_loss(loc_preds * loc_targets_mask, loc_targets) * self.loc_weight
+        loc_loss = F.mse_loss(loc_preds * loc_targets_mask,
+                              loc_targets) * self.loc_weight
         if self.use_giou:
-            loc_loss_giou = self.giou_loss(loc_preds * loc_targets_mask, loc_targets) * self.giou_weight
+            loc_loss_giou = self.giou_loss(loc_preds * loc_targets_mask,
+                                           loc_targets) * self.giou_weight
             total_loss = structure_loss + loc_loss + loc_loss_giou
-            return {'loss':total_loss, "structure_loss":structure_loss, "loc_loss":loc_loss, "loc_loss_giou":loc_loss_giou}
+            return {
+                'loss': total_loss,
+                "structure_loss": structure_loss,
+                "loc_loss": loc_loss,
+                "loc_loss_giou": loc_loss_giou
+            }
         else:
-            total_loss = structure_loss + loc_loss            
-            return {'loss':total_loss, "structure_loss":structure_loss, "loc_loss":loc_loss}
\ No newline at end of file
+            total_loss = structure_loss + loc_loss
+            return {
+                'loss': total_loss,
+                "structure_loss": structure_loss,
+                "loc_loss": loc_loss
+            }
diff --git a/ppocr/metrics/table_metric.py b/ppocr/metrics/table_metric.py
index 17f3dc92b2..26f577a03a 100644
--- a/ppocr/metrics/table_metric.py
+++ b/ppocr/metrics/table_metric.py
@@ -31,6 +31,8 @@ class TableStructureMetric(object):
                                              gt_structure_batch_list):
             pred_str = ''.join(pred)
             target_str = ''.join(target)
+            # pred_str = pred_str.replace('<thead>','').replace('</thead>','').replace('<tbody>','').replace('</tbody>','')
+            # target_str = target_str.replace('<thead>','').replace('</thead>','').replace('<tbody>','').replace('</tbody>','')
             if pred_str == target_str:
                 correct_num += 1
             all_num += 1
@@ -131,10 +133,10 @@ class TableMetric(object):
             self.bbox_metric.reset()
 
     def format_box(self, box):
-        if self.point_num == 4:
+        if self.point_num == 2:
             x1, y1, x2, y2 = box
             box = [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
-        elif self.point_num == 8:
+        elif self.point_num == 4:
             x1, y1, x2, y2, x3, y3, x4, y4 = box
             box = [[x1, y1], [x2, y2], [x3, y3], [x4, y4]]
         return box
diff --git a/ppocr/modeling/heads/table_att_head.py b/ppocr/modeling/heads/table_att_head.py
index b64713898d..4f39d6253d 100644
--- a/ppocr/modeling/heads/table_att_head.py
+++ b/ppocr/modeling/heads/table_att_head.py
@@ -31,16 +31,18 @@ class TableAttentionHead(nn.Layer):
                  loc_type,
                  in_max_len=488,
                  max_text_length=800,
+                 out_channels=30,
+                 point_num=2,
                  **kwargs):
         super(TableAttentionHead, self).__init__()
         self.input_size = in_channels[-1]
         self.hidden_size = hidden_size
-        self.elem_num = 30
+        self.out_channels = out_channels
         self.max_text_length = max_text_length
 
         self.structure_attention_cell = AttentionGRUCell(
-            self.input_size, hidden_size, self.elem_num, use_gru=False)
-        self.structure_generator = nn.Linear(hidden_size, self.elem_num)
+            self.input_size, hidden_size, self.out_channels, use_gru=False)
+        self.structure_generator = nn.Linear(hidden_size, self.out_channels)
         self.loc_type = loc_type
         self.in_max_len = in_max_len
 
@@ -53,7 +55,8 @@ class TableAttentionHead(nn.Layer):
                 self.loc_fea_trans = nn.Linear(625, self.max_text_length + 1)
             else:
                 self.loc_fea_trans = nn.Linear(256, self.max_text_length + 1)
-            self.loc_generator = nn.Linear(self.input_size + hidden_size, 4)
+            self.loc_generator = nn.Linear(self.input_size + hidden_size,
+                                           point_num * 2)
 
     def _char_to_onehot(self, input_char, onehot_dim):
         input_ont_hot = F.one_hot(input_char, onehot_dim)
@@ -77,7 +80,7 @@ class TableAttentionHead(nn.Layer):
             structure = targets[0]
             for i in range(self.max_text_length + 1):
                 elem_onehots = self._char_to_onehot(
-                    structure[:, i], onehot_dim=self.elem_num)
+                    structure[:, i], onehot_dim=self.out_channels)
                 (outputs, hidden), alpha = self.structure_attention_cell(
                     hidden, fea, elem_onehots)
                 output_hiddens.append(paddle.unsqueeze(outputs, axis=1))
@@ -104,7 +107,7 @@ class TableAttentionHead(nn.Layer):
             i = 0
             while i < max_text_length + 1:
                 elem_onehots = self._char_to_onehot(
-                    temp_elem, onehot_dim=self.elem_num)
+                    temp_elem, onehot_dim=self.out_channels)
                 (outputs, hidden), alpha = self.structure_attention_cell(
                     hidden, fea, elem_onehots)
                 output_hiddens.append(paddle.unsqueeze(outputs, axis=1))

From 7c92b441a88925a82a4d4880bb21710ba0d5c945 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 27 Jun 2022 09:39:35 +0000
Subject: [PATCH 064/134] update mv3 config

---
 configs/table/table_mv3.yml | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/configs/table/table_mv3.yml b/configs/table/table_mv3.yml
index 9159addc31..66c1c83e12 100755
--- a/configs/table/table_mv3.yml
+++ b/configs/table/table_mv3.yml
@@ -17,7 +17,7 @@ Global:
   # for data or label process
   character_dict_path: ppocr/utils/dict/table_structure_dict.txt
   character_type: en
-  max_text_length: 500
+  max_text_length: 800
   infer_mode: False
   process_total_num: 0
   process_cut_num: 0
@@ -44,7 +44,7 @@ Architecture:
     name: TableAttentionHead
     hidden_size: 256
     loc_type: 2
-    max_text_length: 500
+    max_text_length: 800
 
 Loss:
   name: TableAttentionLoss
@@ -57,13 +57,13 @@ PostProcess:
 Metric:
   name: TableMetric
   main_indicator: acc
-  compute_bbox_metric: False # cost many time, set False for training
+  compute_bbox_metric: false # cost many time, set False for training
 
 Train:
   dataset:
     name: PubTabDataSet
     data_dir: train_data/table/pubtabnet/train/
-    label_file_path: train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl]
     transforms:
       - DecodeImage: # load image
           img_mode: BGR

From 78064ad9560f479c3aa4def2c44cb0539c5a0bdc Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 09:43:14 +0000
Subject: [PATCH 065/134] add db++

---
 ppocr/data/imaug/operators.py          |  6 ++--
 ppocr/modeling/backbones/det_resnet.py |  2 +-
 ppocr/modeling/necks/db_fpn.py         | 47 ++++++++++++++++----------
 3 files changed, 33 insertions(+), 22 deletions(-)

diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
index a83dc64838..ee5492de79 100644
--- a/ppocr/data/imaug/operators.py
+++ b/ppocr/data/imaug/operators.py
@@ -242,8 +242,8 @@ class DetResizeForTest(object):
         if 'image_shape' in kwargs:
             self.image_shape = kwargs['image_shape']
             self.resize_type = 1
-            if 'keep_ratio' in kwargs:  ######
-                self.keep_ratio = kwargs['keep_ratio']  #######
+            if 'keep_ratio' in kwargs:
+                self.keep_ratio = kwargs['keep_ratio']
         elif 'limit_side_len' in kwargs:
             self.limit_side_len = kwargs['limit_side_len']
             self.limit_type = kwargs.get('limit_type', 'min')
@@ -273,7 +273,7 @@ class DetResizeForTest(object):
     def resize_image_type1(self, img):
         resize_h, resize_w = self.image_shape
         ori_h, ori_w = img.shape[:2]  # (h, w, c)
-        if self.keep_ratio:  ########
+        if self.keep_ratio:
             resize_w = ori_w * resize_h / ori_h
             N = math.ceil(resize_w / 32)
             resize_w = N * 32
diff --git a/ppocr/modeling/backbones/det_resnet.py b/ppocr/modeling/backbones/det_resnet.py
index 4c75b663af..3d8ce44508 100644
--- a/ppocr/modeling/backbones/det_resnet.py
+++ b/ppocr/modeling/backbones/det_resnet.py
@@ -1,4 +1,4 @@
-# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/ppocr/modeling/necks/db_fpn.py b/ppocr/modeling/necks/db_fpn.py
index b46d7c460d..8c3f52a331 100644
--- a/ppocr/modeling/necks/db_fpn.py
+++ b/ppocr/modeling/necks/db_fpn.py
@@ -105,7 +105,7 @@ class DSConv(nn.Layer):
 
 
 class DBFPN(nn.Layer):
-    def __init__(self, in_channels, out_channels, use_asf=None, **kwargs):
+    def __init__(self, in_channels, out_channels, use_asf=False, **kwargs):
         super(DBFPN, self).__init__()
         self.out_channels = out_channels
         self.use_asf = use_asf
@@ -164,7 +164,7 @@ class DBFPN(nn.Layer):
             weight_attr=ParamAttr(initializer=weight_attr),
             bias_attr=False)
 
-        if self.use_asf:
+        if self.use_asf is True:
             self.asf = ASFBlock(self.out_channels, self.out_channels // 4)
 
     def forward(self, x):
@@ -192,7 +192,7 @@ class DBFPN(nn.Layer):
 
         fuse = paddle.concat([p5, p4, p3, p2], axis=1)
 
-        if self.use_asf:
+        if self.use_asf is True:
             fuse = self.asf(fuse, [p5, p4, p3, p2])
 
         return fuse
@@ -367,7 +367,19 @@ class LKPAN(nn.Layer):
 
 
 class ASFBlock(nn.Layer):
+    """
+    This code is refered from:
+        https://github.com/MhLiao/DB/blob/master/decoders/feature_attention.py
+    """
+
     def __init__(self, in_channels, inter_channels, out_features_num=4):
+        """
+        Adaptive Scale Fusion (ASF) block of DBNet++
+        Args:
+            in_channels: the number of channels in the input data
+            inter_channels: the number of middle channels
+            out_features_num: the number of fused stages
+        """
         super(ASFBlock, self).__init__()
         weight_attr = paddle.nn.initializer.KaimingUniform()
         self.in_channels = in_channels
@@ -375,39 +387,38 @@ class ASFBlock(nn.Layer):
         self.out_features_num = out_features_num
         self.conv = nn.Conv2D(in_channels, inter_channels, 3, padding=1)
 
-        self.attention_block_1 = nn.Sequential(
+        self.spatial_scale = nn.Sequential(
             #Nx1xHxW
             nn.Conv2D(
-                1,
-                1,
-                3,
+                in_channels=1,
+                out_channels=1,
+                kernel_size=3,
                 bias_attr=False,
                 padding=1,
                 weight_attr=ParamAttr(initializer=weight_attr)),
             nn.ReLU(),
             nn.Conv2D(
-                1,
-                1,
-                1,
+                in_channels=1,
+                out_channels=1,
+                kernel_size=1,
                 bias_attr=False,
                 weight_attr=ParamAttr(initializer=weight_attr)),
             nn.Sigmoid())
 
-        self.attention_block_2 = nn.Sequential(
+        self.channel_scale = nn.Sequential(
             nn.Conv2D(
-                inter_channels,
-                out_features_num,
-                1,
+                in_channels=inter_channels,
+                out_channels=out_features_num,
+                kernel_size=1,
                 bias_attr=False,
                 weight_attr=ParamAttr(initializer=weight_attr)),
             nn.Sigmoid())
 
     def forward(self, fuse_features, features_list):
         fuse_features = self.conv(fuse_features)
-        attention_scores = self.attention_block_1(
-            paddle.mean(
-                fuse_features, axis=1, keepdim=True)) + fuse_features
-        attention_scores = self.attention_block_2(attention_scores)
+        spatial_x = paddle.mean(fuse_features, axis=1, keepdim=True)
+        attention_scores = self.spatial_scale(spatial_x) + fuse_features
+        attention_scores = self.channel_scale(attention_scores)
         assert len(features_list) == self.out_features_num
 
         out_list = []

From 679ec78bb2bc77bbc54c31b77c33074d34d100fe Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 27 Jun 2022 10:01:35 +0000
Subject: [PATCH 066/134] add db++

---
 configs/det/{det_r50_db++_td_tr.yml => det_r50_db++.yml} | 0
 1 file changed, 0 insertions(+), 0 deletions(-)
 rename configs/det/{det_r50_db++_td_tr.yml => det_r50_db++.yml} (100%)

diff --git a/configs/det/det_r50_db++_td_tr.yml b/configs/det/det_r50_db++.yml
similarity index 100%
rename from configs/det/det_r50_db++_td_tr.yml
rename to configs/det/det_r50_db++.yml

From ec3f8ff7e83d2462ef99b538ff0b5c7311ae992d Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Mon, 27 Jun 2022 19:29:33 +0800
Subject: [PATCH 067/134] support PP-OCRv3 for lite demo (arm8)

---
 deploy/lite/config.txt      |  3 ++-
 deploy/lite/crnn_process.cc | 18 ++++++++--------
 deploy/lite/crnn_process.h  |  2 +-
 deploy/lite/ocr_db_crnn.cc  | 13 ++++++++----
 deploy/lite/readme.md       | 36 ++++++++++++++++++--------------
 deploy/lite/readme_ch.md    | 41 ++++++++++++++++++++-----------------
 6 files changed, 65 insertions(+), 48 deletions(-)

diff --git a/deploy/lite/config.txt b/deploy/lite/config.txt
index 4c68105d39..dda0d2b032 100644
--- a/deploy/lite/config.txt
+++ b/deploy/lite/config.txt
@@ -4,4 +4,5 @@ det_db_box_thresh  0.5
 det_db_unclip_ratio  1.6
 det_db_use_dilate 0
 det_use_polygon_score 1
-use_direction_classify  1
\ No newline at end of file
+use_direction_classify  1
+rec_image_height  32
\ No newline at end of file
diff --git a/deploy/lite/crnn_process.cc b/deploy/lite/crnn_process.cc
index 7528f36fe6..6d5fc1504e 100644
--- a/deploy/lite/crnn_process.cc
+++ b/deploy/lite/crnn_process.cc
@@ -19,25 +19,27 @@
 
 const std::vector<int> rec_image_shape{3, 32, 320};
 
-cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio) {
+cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height) {
   int imgC, imgH, imgW;
   imgC = rec_image_shape[0];
+  imgH = rec_image_height;
   imgW = rec_image_shape[2];
-  imgH = rec_image_shape[1];
 
-  imgW = int(32 * wh_ratio);
+  imgW = int(imgH * wh_ratio);
 
-  float ratio = static_cast<float>(img.cols) / static_cast<float>(img.rows);
+  float ratio = float(img.cols) / float(img.rows);
   int resize_w, resize_h;
+
   if (ceilf(imgH * ratio) > imgW)
     resize_w = imgW;
   else
-    resize_w = static_cast<int>(ceilf(imgH * ratio));
-  cv::Mat resize_img;
+    resize_w = int(ceilf(imgH * ratio));
+
   cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
              cv::INTER_LINEAR);
-
-  return resize_img;
+  cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
+                     int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
+                     {127, 127, 127});
 }
 
 std::vector<std::string> ReadDict(std::string path) {
diff --git a/deploy/lite/crnn_process.h b/deploy/lite/crnn_process.h
index 29e6790697..ed7a316706 100644
--- a/deploy/lite/crnn_process.h
+++ b/deploy/lite/crnn_process.h
@@ -26,7 +26,7 @@
 #include "opencv2/imgcodecs.hpp"
 #include "opencv2/imgproc.hpp"
 
-cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio);
+cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height);
 
 std::vector<std::string> ReadDict(std::string path);
 
diff --git a/deploy/lite/ocr_db_crnn.cc b/deploy/lite/ocr_db_crnn.cc
index 1ffbbacb74..cb2bf7791a 100644
--- a/deploy/lite/ocr_db_crnn.cc
+++ b/deploy/lite/ocr_db_crnn.cc
@@ -162,7 +162,8 @@ void RunRecModel(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat img,
                  std::vector<std::string> charactor_dict,
                  std::shared_ptr<PaddlePredictor> predictor_cls,
                  int use_direction_classify,
-                 std::vector<double> *times) {
+                 std::vector<double> *times,
+                 int rec_image_height) {
   std::vector<float> mean = {0.5f, 0.5f, 0.5f};
   std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
 
@@ -183,7 +184,7 @@ void RunRecModel(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat img,
     float wh_ratio =
         static_cast<float>(crop_img.cols) / static_cast<float>(crop_img.rows);
 
-    resize_img = CrnnResizeImg(crop_img, wh_ratio);
+    resize_img = CrnnResizeImg(crop_img, wh_ratio, rec_image_height);
     resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
 
     const float *dimg = reinterpret_cast<const float *>(resize_img.data);
@@ -444,7 +445,7 @@ void system(char **argv){
   //// load config from txt file
   auto Config = LoadConfigTxt(det_config_path);
   int use_direction_classify = int(Config["use_direction_classify"]);
-
+  int rec_image_height = int(Config["rec_image_height"]);
   auto charactor_dict = ReadDict(dict_path);
   charactor_dict.insert(charactor_dict.begin(), "#"); // blank char for ctc
   charactor_dict.push_back(" ");
@@ -590,12 +591,16 @@ void rec(int argc, char **argv) {
   std::string batchsize = argv[6];
   std::string img_dir = argv[7];
   std::string dict_path = argv[8];
+  std::string config_path = argv[9];
 
   if (strcmp(argv[4], "FP32") != 0 && strcmp(argv[4], "INT8") != 0) {
       std::cerr << "Only support FP32 or INT8." << std::endl;
       exit(1);
   }
 
+  auto Config = LoadConfigTxt(config_path);
+  int rec_image_height = int(Config["rec_image_height"]);
+
   std::vector<cv::String> cv_all_img_names;
   cv::glob(img_dir, cv_all_img_names);
 
@@ -630,7 +635,7 @@ void rec(int argc, char **argv) {
     std::vector<float> rec_text_score;
     std::vector<double> times;
     RunRecModel(boxes, srcimg, rec_predictor, rec_text, rec_text_score,
-                charactor_dict, cls_predictor, 0, &times);
+                charactor_dict, cls_predictor, 0, &times, rec_image_height);
   
     //// print recognized text
     for (int i = 0; i < rec_text.size(); i++) {
diff --git a/deploy/lite/readme.md b/deploy/lite/readme.md
index 9926e2dd8c..883aff9890 100644
--- a/deploy/lite/readme.md
+++ b/deploy/lite/readme.md
@@ -34,7 +34,7 @@ For the compilation process of different development environments, please refer
 ### 1.2 Prepare Paddle-Lite library
 
 There are two ways to obtain the Paddle-Lite library：
-- 1. Download directly, the download link of the Paddle-Lite library is as follows：
+- 1. [Recommended] Download directly, the download link of the Paddle-Lite library is as follows：
 
       | Platform | Paddle-Lite library download link |
       |---|---|
@@ -43,7 +43,9 @@ There are two ways to obtain the Paddle-Lite library：
 
       Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.10 branch. For more information about Paddle-Lite 2.10, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10).
 
-- 2. [Recommended] Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
+    **Note: It is recommended to use paddlelite>=2.10 version of the prediction library, other prediction library versions [download link](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+- 2. Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
@@ -104,20 +106,16 @@ If you directly use the model in the above table for deployment, you can skip th
 
 If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
 
-The `opt` tool can be obtained by compiling Paddle Lite.
+- Step 1: Refer to [document](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html) to install paddlelite, which is used to convert paddle inference model to paddlelite required for running nb model
 ```
-git clone https://github.com/PaddlePaddle/Paddle-Lite.git
-cd Paddle-Lite
-git checkout release/v2.10
-./lite/tools/build.sh build_optimize_tool
+pip install paddlelite==2.10 # The paddlelite version should be the same as the prediction library version
+```
+After installation, the following commands can view the help information
+```
+paddle_lite_opt
 ```
 
-After the compilation is complete, the opt file is located under build.opt/lite/api/, You can view the operating options and usage of opt in the following ways:
-
-```
-cd build.opt/lite/api/
-./opt
-```
+Introduction to paddle_lite_opt parameters:
 
 |Options|Description|
 |---|---|
@@ -131,6 +129,8 @@ cd build.opt/lite/api/
 
 `--model_dir` is suitable for the non-combined mode of the model to be optimized, and the inference model of PaddleOCR is the combined mode, that is, the model structure and model parameters are stored in a single file.
 
+- Step 2: Use paddle_lite_opt to convert the inference model to the mobile model format.
+
 The following takes the ultra-lightweight Chinese model of PaddleOCR as an example to introduce the use of the compiled opt file to complete the conversion of the inference model to the Paddle-Lite optimized model
 
 ```
@@ -240,6 +240,7 @@ det_db_thresh  0.3        # Used to filter the binarized image of DB prediction,
 det_db_box_thresh  0.5    # DDB post-processing filter box threshold, if there is a missing box detected, it can be reduced as appropriate
 det_db_unclip_ratio  1.6  # Indicates the compactness of the text box, the smaller the value, the closer the text box to the text
 use_direction_classify  0  # Whether to use the direction classifier, 0 means not to use, 1 means to use
+rec_image_height  32      # The height of the input image of the recognition model, the PP-OCRv3 model needs to be set to 48, and the PP-OCRv2 model needs to be set to 32
 ```
 
  5. Run Model on phone
@@ -258,8 +259,9 @@ After the above steps are completed, you can use adb to push the file to the pho
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # The use of ocr_db_crnn is:
- # ./ocr_db_crnn Detection model file Orientation classifier model file Recognition model file Test image path Dictionary file path
- ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_opt.nb  ./11.jpg  ppocr_keys_v1.txt
+ # ./ocr_db_crnn Detection model file Orientation classifier model file Recognition model file  Hardware  Precision  Threads Batchsize  Test image path Dictionary file path
+ ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+# precision can be INT8 for quantitative model or FP32 for normal model.
  ```
 
 If you modify the code, you need to recompile and push to the phone.
@@ -283,3 +285,7 @@ A2: Replace the .jpg test image under ./debug with the image you want to test, a
 Q3: How to package it into the mobile APP?
 
 A3: This demo aims to provide the core algorithm part that can run OCR on mobile phones. Further, PaddleOCR/deploy/android_demo is an example of encapsulating this demo into a mobile app for reference.
+
+Q4: When running the demo, an error is reported `Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4: The problem is that the installed paddlelite version does not match the downloaded prediction library version. Make sure that the paddleliteopt tool matches your prediction library version, and try to switch to the nb model again.
diff --git a/deploy/lite/readme_ch.md b/deploy/lite/readme_ch.md
index 99a543d0d6..e238b4ec53 100644
--- a/deploy/lite/readme_ch.md
+++ b/deploy/lite/readme_ch.md
@@ -8,7 +8,7 @@
     - [2.1 模型优化](#21-模型优化)
     - [2.2 与手机联调](#22-与手机联调)
 - [FAQ](#faq)
-  
+
 
 本教程将介绍基于[Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) 在移动端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
 
@@ -32,7 +32,7 @@ Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理
 ### 1.2 准备预测库
 
 预测库有两种获取方式：
-- 1. 直接下载，预测库下载链接如下：
+- 1. [推荐]直接下载，预测库下载链接如下：
 
       | 平台 | 预测库下载链接 |
       |---|---|
@@ -41,7 +41,9 @@ Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理
 
       注：1. 上述预测库为PaddleLite 2.10分支编译得到，有关PaddleLite 2.10 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10) 。
 
-- 2. [推荐]编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
+**注：建议使用paddlelite>=2.10版本的预测库，其他预测库版本[下载链接](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+- 2. 编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
@@ -102,22 +104,16 @@ Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括
 
 如果要部署的模型不在上述表格中，则需要按照如下步骤获得优化后的模型。
 
-模型优化需要Paddle-Lite的opt可执行文件，可以通过编译Paddle-Lite源码获得，编译步骤如下：
+- 步骤1：参考[文档](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html)安装paddlelite，用于转换paddle inference model为paddlelite运行所需的nb模型
 ```
-# 如果准备环境时已经clone了Paddle-Lite，则不用重新clone Paddle-Lite
-git clone https://github.com/PaddlePaddle/Paddle-Lite.git
-cd Paddle-Lite
-git checkout release/v2.10
-# 启动编译
-./lite/tools/build.sh build_optimize_tool
+pip install paddlelite==2.10  # paddlelite版本要与预测库版本一致
 ```
-
-编译完成后，opt文件位于`build.opt/lite/api/`下，可通过如下方式查看opt的运行选项和使用方式；
+安装完后，如下指令可以查看帮助信息
 ```
-cd build.opt/lite/api/
-./opt
+paddle_lite_opt
 ```
 
+paddle_lite_opt 参数介绍：
 |选项|说明|
 |---|---|
 |--model_dir|待优化的PaddlePaddle模型（非combined形式）的路径|
@@ -130,6 +126,8 @@ cd build.opt/lite/api/
 
 `--model_dir`适用于待优化的模型是非combined方式，PaddleOCR的inference模型是combined方式，即模型结构和模型参数使用单独一个文件存储。
 
+- 步骤2：使用paddle_lite_opt将inference模型转换成移动端模型格式。
+
 下面以PaddleOCR的超轻量中文模型为例，介绍使用编译好的opt文件完成inference模型到Paddle-Lite优化模型的转换。
 
 ```
@@ -148,7 +146,7 @@ wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls
 
 转换成功后，inference模型目录下会多出`.nb`结尾的文件，即是转换成功的模型文件。
 
-注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt 工具的输入模型是paddle保存的inference模型
+注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt工具的输入模型是paddle保存的inference模型
 
 <a name="2.2与手机联调"></a>
 ### 2.2 与手机联调
@@ -234,13 +232,14 @@ ppocr_keys_v1.txt   # 中文字典
 ...
 ```
 
-2.  `config.txt` 包含了检测器、分类器的超参数，如下：
+2.  `config.txt` 包含了检测器、分类器、识别器的超参数，如下：
 ```
 max_side_len  960         # 输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960
 det_db_thresh  0.3        # 用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不明显
-det_db_box_thresh  0.5    # DB后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小
+det_db_box_thresh  0.5    # 检测器后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小
 det_db_unclip_ratio  1.6  # 表示文本框的紧致程度，越小则文本框更靠近文本
 use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1表示使用
+rec_image_height  32      # 识别模型输入图像的高度，PP-OCRv3模型设置为48，PP-OCRv2模型需要设置为32
 ```
 
  5. 启动调试
@@ -259,8 +258,8 @@ use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # 开始使用，ocr_db_crnn可执行文件的使用方式为:
- # ./ocr_db_crnn  检测模型文件 方向分类器模型文件  识别模型文件  测试图像路径  字典文件路径
- ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  ./11.jpg  ppocr_keys_v1.txt
+ # ./ocr_db_crnn  检测模型文件 方向分类器模型文件  识别模型文件 运行硬件 运行精度 线程数  batchsize  测试图像路径  参数配置路径  字典文件路径 是否使用benchmark参数
+ ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
  ```
 
  如果对代码做了修改，则需要重新编译并push到手机上。
@@ -284,3 +283,7 @@ A2：替换debug下的.jpg测试图像为你想要测试的图像，adb push 到
 Q3：如何封装到手机APP中？
 
 A3：此demo旨在提供能在手机上运行OCR的核心算法部分，PaddleOCR/deploy/android_demo是将这个demo封装到手机app的示例，供参考
+
+Q4：运行demo时遇到报错`Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4：问题是安装的paddlelite版本和下载的预测库版本不匹配，确保paddleliteopt工具和你的预测库版本匹配，重新转nb模型试试。

From d56c0b9b7cea28bca25b5b8720c5eff9453178a3 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Mon, 27 Jun 2022 19:47:07 +0800
Subject: [PATCH 068/134] add mode

---
 deploy/lite/readme.md    | 10 ++++++++--
 deploy/lite/readme_ch.md | 10 ++++++++--
 2 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/deploy/lite/readme.md b/deploy/lite/readme.md
index 883aff9890..a1bef8120e 100644
--- a/deploy/lite/readme.md
+++ b/deploy/lite/readme.md
@@ -259,9 +259,15 @@ After the above steps are completed, you can use adb to push the file to the pho
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # The use of ocr_db_crnn is:
- # ./ocr_db_crnn Detection model file Orientation classifier model file Recognition model file  Hardware  Precision  Threads Batchsize  Test image path Dictionary file path
- ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+ # ./ocr_db_crnn Mode Detection model file Orientation classifier model file Recognition model file  Hardware  Precision  Threads Batchsize  Test image path Dictionary file path
+ ./ocr_db_crnn system ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
 # precision can be INT8 for quantitative model or FP32 for normal model.
+
+# Only using detection model
+./ocr_db_crnn  det ch_PP-OCRv2_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# Only using recognition model
+./ocr_db_crnn  rec ch_PP-OCRv2_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
  ```
 
 If you modify the code, you need to recompile and push to the phone.
diff --git a/deploy/lite/readme_ch.md b/deploy/lite/readme_ch.md
index e238b4ec53..0793827fe6 100644
--- a/deploy/lite/readme_ch.md
+++ b/deploy/lite/readme_ch.md
@@ -258,8 +258,14 @@ rec_image_height  32      # 识别模型输入图像的高度，PP-OCRv3模型
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
  # 开始使用，ocr_db_crnn可执行文件的使用方式为:
- # ./ocr_db_crnn  检测模型文件 方向分类器模型文件  识别模型文件 运行硬件 运行精度 线程数  batchsize  测试图像路径  参数配置路径  字典文件路径 是否使用benchmark参数
- ./ocr_db_crnn ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+ # ./ocr_db_crnn 预测模式  检测模型文件 方向分类器模型文件  识别模型文件 运行硬件 运行精度 线程数  batchsize  测试图像路径  参数配置路径  字典文件路径 是否使用benchmark参数
+ ./ocr_db_crnn system  ch_PP-OCRv2_det_slim_opt.nb  ch_PP-OCRv2_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+
+# 仅使用文本检测模型，使用方式如下：
+./ocr_db_crnn  det ch_PP-OCRv2_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# 仅使用文本识别模型，使用方式如下：
+./ocr_db_crnn  rec ch_PP-OCRv2_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
  ```
 
  如果对代码做了修改，则需要重新编译并push到手机上。

From 4805aa0f100749e82250a78154321b7002e4b812 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Tue, 28 Jun 2022 02:08:48 +0000
Subject: [PATCH 069/134] fix amp vqa

---
 tools/program.py | 31 ++++++++++++++++++++++---------
 1 file changed, 22 insertions(+), 9 deletions(-)

diff --git a/tools/program.py b/tools/program.py
index aa0d2698cf..9a90eda2f8 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -255,6 +255,8 @@ def train(config,
                 with paddle.amp.auto_cast():
                     if model_type == 'table' or extra_input:
                         preds = model(images, data=batch[1:])
+                    elif model_type in ["kie", 'vqa']:
+                        preds = model(batch)
                     else:
                         preds = model(images)
             else:
@@ -307,7 +309,8 @@ def train(config,
             train_stats.update(stats)
 
             if log_writer is not None and dist.get_rank() == 0:
-                log_writer.log_metrics(metrics=train_stats.get(), prefix="TRAIN", step=global_step)
+                log_writer.log_metrics(
+                    metrics=train_stats.get(), prefix="TRAIN", step=global_step)
 
             if dist.get_rank() == 0 and (
                 (global_step > 0 and global_step % print_batch_step == 0) or
@@ -354,7 +357,8 @@ def train(config,
 
                 # logger metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics=cur_metric, prefix="EVAL", step=global_step)
+                    log_writer.log_metrics(
+                        metrics=cur_metric, prefix="EVAL", step=global_step)
 
                 if cur_metric[main_indicator] >= best_model_dict[
                         main_indicator]:
@@ -377,11 +381,18 @@ def train(config,
                 logger.info(best_str)
                 # logger best metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics={
-                        "best_{}".format(main_indicator): best_model_dict[main_indicator]
-                        }, prefix="EVAL", step=global_step)
-                    
-                    log_writer.log_model(is_best=True, prefix="best_accuracy", metadata=best_model_dict)
+                    log_writer.log_metrics(
+                        metrics={
+                            "best_{}".format(main_indicator):
+                            best_model_dict[main_indicator]
+                        },
+                        prefix="EVAL",
+                        step=global_step)
+
+                    log_writer.log_model(
+                        is_best=True,
+                        prefix="best_accuracy",
+                        metadata=best_model_dict)
 
             reader_start = time.time()
         if dist.get_rank() == 0:
@@ -413,7 +424,8 @@ def train(config,
                 epoch=epoch,
                 global_step=global_step)
             if log_writer is not None:
-                log_writer.log_model(is_best=False, prefix='iter_epoch_{}'.format(epoch))
+                log_writer.log_model(
+                    is_best=False, prefix='iter_epoch_{}'.format(epoch))
 
     best_str = 'best metric, {}'.format(', '.join(
         ['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))
@@ -585,7 +597,8 @@ def preprocess(is_train=False):
         vdl_writer_path = '{}/vdl/'.format(save_model_dir)
         log_writer = VDLLogger(save_model_dir)
         loggers.append(log_writer)
-    if ('use_wandb' in config['Global'] and config['Global']['use_wandb']) or 'wandb' in config:
+    if ('use_wandb' in config['Global'] and
+            config['Global']['use_wandb']) or 'wandb' in config:
         save_dir = config['Global']['save_model_dir']
         wandb_writer_path = "{}/wandb".format(save_dir)
         if "wandb" in config:

From adeb8a17c9f67f99f4ed6021a3b50e3b0d6f02e3 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Tue, 28 Jun 2022 02:17:53 +0000
Subject: [PATCH 070/134] fix amp vqa

---
 tools/program.py | 31 ++++++++++++++++++++++---------
 1 file changed, 22 insertions(+), 9 deletions(-)

diff --git a/tools/program.py b/tools/program.py
index aa0d2698cf..9a90eda2f8 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -255,6 +255,8 @@ def train(config,
                 with paddle.amp.auto_cast():
                     if model_type == 'table' or extra_input:
                         preds = model(images, data=batch[1:])
+                    elif model_type in ["kie", 'vqa']:
+                        preds = model(batch)
                     else:
                         preds = model(images)
             else:
@@ -307,7 +309,8 @@ def train(config,
             train_stats.update(stats)
 
             if log_writer is not None and dist.get_rank() == 0:
-                log_writer.log_metrics(metrics=train_stats.get(), prefix="TRAIN", step=global_step)
+                log_writer.log_metrics(
+                    metrics=train_stats.get(), prefix="TRAIN", step=global_step)
 
             if dist.get_rank() == 0 and (
                 (global_step > 0 and global_step % print_batch_step == 0) or
@@ -354,7 +357,8 @@ def train(config,
 
                 # logger metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics=cur_metric, prefix="EVAL", step=global_step)
+                    log_writer.log_metrics(
+                        metrics=cur_metric, prefix="EVAL", step=global_step)
 
                 if cur_metric[main_indicator] >= best_model_dict[
                         main_indicator]:
@@ -377,11 +381,18 @@ def train(config,
                 logger.info(best_str)
                 # logger best metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics={
-                        "best_{}".format(main_indicator): best_model_dict[main_indicator]
-                        }, prefix="EVAL", step=global_step)
-                    
-                    log_writer.log_model(is_best=True, prefix="best_accuracy", metadata=best_model_dict)
+                    log_writer.log_metrics(
+                        metrics={
+                            "best_{}".format(main_indicator):
+                            best_model_dict[main_indicator]
+                        },
+                        prefix="EVAL",
+                        step=global_step)
+
+                    log_writer.log_model(
+                        is_best=True,
+                        prefix="best_accuracy",
+                        metadata=best_model_dict)
 
             reader_start = time.time()
         if dist.get_rank() == 0:
@@ -413,7 +424,8 @@ def train(config,
                 epoch=epoch,
                 global_step=global_step)
             if log_writer is not None:
-                log_writer.log_model(is_best=False, prefix='iter_epoch_{}'.format(epoch))
+                log_writer.log_model(
+                    is_best=False, prefix='iter_epoch_{}'.format(epoch))
 
     best_str = 'best metric, {}'.format(', '.join(
         ['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))
@@ -585,7 +597,8 @@ def preprocess(is_train=False):
         vdl_writer_path = '{}/vdl/'.format(save_model_dir)
         log_writer = VDLLogger(save_model_dir)
         loggers.append(log_writer)
-    if ('use_wandb' in config['Global'] and config['Global']['use_wandb']) or 'wandb' in config:
+    if ('use_wandb' in config['Global'] and
+            config['Global']['use_wandb']) or 'wandb' in config:
         save_dir = config['Global']['save_model_dir']
         wandb_writer_path = "{}/wandb".format(save_dir)
         if "wandb" in config:

From 0ae4f664612104227cf3e7ff13bc41b6d229dcfc Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 28 Jun 2022 02:42:47 +0000
Subject: [PATCH 071/134] add db++

---
 configs/det/det_r50_db++_ic15.yml             | 163 ++++++++++++++++++
 ...et_r50_db++.yml => det_r50_db++_td_tr.yml} |   0
 2 files changed, 163 insertions(+)
 create mode 100644 configs/det/det_r50_db++_ic15.yml
 rename configs/det/{det_r50_db++.yml => det_r50_db++_td_tr.yml} (100%)

diff --git a/configs/det/det_r50_db++_ic15.yml b/configs/det/det_r50_db++_ic15.yml
new file mode 100644
index 0000000000..c1e8e86870
--- /dev/null
+++ b/configs/det/det_r50_db++_ic15.yml
@@ -0,0 +1,163 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 1000
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_icdar15/
+  save_epoch_step: 200
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: false
+  pretrained_model: ./pretrain_models/synthtext_pretrained_res50_dcn_asf_spatial
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: ResNet
+    layers: 50
+    dcn_stage: [False, True, True, True]
+  Neck:
+    name: DBFPN
+    out_channels: 256
+    use_asf: True
+  Head:
+    name: DBHead
+    k: 50
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: BCELoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  lr:
+    name: DecayLearningRate
+    learning_rate: 0.007
+    epochs: 1000
+    factor: 0.9
+    end_lr: 0
+  weight_decay: 0.0001
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/                                                                        
+    label_file_list:                                                                                            
+    - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt                                                     
+    ratio_list:                                                                                                 
+    - 1.0 
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 10
+        keep_ratio: true
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 4
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization                                                                          
+    label_file_list:                                                                                            
+    - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        image_shape:
+        - 1152
+        - 2048
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/det_r50_db++.yml b/configs/det/det_r50_db++_td_tr.yml
similarity index 100%
rename from configs/det/det_r50_db++.yml
rename to configs/det/det_r50_db++_td_tr.yml

From 03a6ef531f82eda961183ba38e7473b8cfc9c3fa Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Tue, 28 Jun 2022 03:27:48 +0000
Subject: [PATCH 072/134] fix abinet ci bug

---
 configs/rec/rec_r45_abinet.yml | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/configs/rec/rec_r45_abinet.yml b/configs/rec/rec_r45_abinet.yml
index 1e03db8b1b..e604bcd12d 100644
--- a/configs/rec/rec_r45_abinet.yml
+++ b/configs/rec/rec_r45_abinet.yml
@@ -8,7 +8,7 @@ Global:
   # evaluation is run every 2000 iterations
   eval_batch_step: [0, 2000]
   cal_metric_during_train: True
-  pretrained_model: ./rec_r45_abinet_train/abinet_vl_pretrained
+  pretrained_model:
   checkpoints:
   save_inference_dir:
   use_visualdl: False
@@ -41,7 +41,6 @@ Architecture:
   Transform:
   Backbone:
     name: ResNet45
-
   Head:
     name: ABINetHead
     use_lang: True

From a63f4414e2cdf3dfd42cddaeb9dcd3b0d412ffd0 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 28 Jun 2022 03:28:53 +0000
Subject: [PATCH 073/134] add db++

---
 ppocr/data/imaug/operators.py             | 2 +-
 ppocr/modeling/backbones/det_resnet_vd.py | 1 -
 2 files changed, 1 insertion(+), 2 deletions(-)

diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
index ee5492de79..070fb7afa0 100644
--- a/ppocr/data/imaug/operators.py
+++ b/ppocr/data/imaug/operators.py
@@ -273,7 +273,7 @@ class DetResizeForTest(object):
     def resize_image_type1(self, img):
         resize_h, resize_w = self.image_shape
         ori_h, ori_w = img.shape[:2]  # (h, w, c)
-        if self.keep_ratio:
+        if self.keep_ratio is True:
             resize_w = ori_w * resize_h / ori_h
             N = math.ceil(resize_w / 32)
             resize_w = N * 32
diff --git a/ppocr/modeling/backbones/det_resnet_vd.py b/ppocr/modeling/backbones/det_resnet_vd.py
index 5337e14c40..a421da0ab4 100644
--- a/ppocr/modeling/backbones/det_resnet_vd.py
+++ b/ppocr/modeling/backbones/det_resnet_vd.py
@@ -323,7 +323,6 @@ class ResNet_vd(nn.Layer):
             for block in range(len(depth)):
                 block_list = []
                 shortcut = False
-                # is_dcn = self.dcn_stage[block]
                 for i in range(depth[block]):
                     basic_block = self.add_sublayer(
                         'bb_%d_%d' % (block, i),

From 9422629cc05e0fdf6ead09d6b03a18424b131043 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 28 Jun 2022 04:17:14 +0000
Subject: [PATCH 074/134] add db++

---
 configs/det/det_r50_db++_ic15.yml  |  2 +-
 configs/det/det_r50_db++_td_tr.yml |  2 +-
 doc/doc_ch/algorithm_det_db.md     | 24 ++++++++++++++++++++++--
 doc/doc_ch/dataset/ocr_datasets.md |  1 +
 4 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/configs/det/det_r50_db++_ic15.yml b/configs/det/det_r50_db++_ic15.yml
index c1e8e86870..a90c868cc3 100644
--- a/configs/det/det_r50_db++_ic15.yml
+++ b/configs/det/det_r50_db++_ic15.yml
@@ -10,7 +10,7 @@ Global:
   - 0
   - 2000
   cal_metric_during_train: false
-  pretrained_model: ./pretrain_models/synthtext_pretrained_res50_dcn_asf_spatial
+  pretrained_model: ./pretrain_models/ResNet50_dcn_asf_synthtext_pretrained
   checkpoints: null
   save_inference_dir: null
   use_visualdl: false
diff --git a/configs/det/det_r50_db++_td_tr.yml b/configs/det/det_r50_db++_td_tr.yml
index 4954e1abb6..5e26ddb0ac 100644
--- a/configs/det/det_r50_db++_td_tr.yml
+++ b/configs/det/det_r50_db++_td_tr.yml
@@ -10,7 +10,7 @@ Global:
   - 0
   - 2000
   cal_metric_during_train: false
-  pretrained_model: ./pretrain_models/synthtext_pretrained_res50_dcn_asf_spatial
+  pretrained_model: ./pretrain_models/ResNet50_dcn_asf_synthtext_pretrained
   checkpoints: null
   save_inference_dir: null
   use_visualdl: false
diff --git a/doc/doc_ch/algorithm_det_db.md b/doc/doc_ch/algorithm_det_db.md
index 90837c2ac1..8466565563 100644
--- a/doc/doc_ch/algorithm_det_db.md
+++ b/doc/doc_ch/algorithm_det_db.md
@@ -1,4 +1,4 @@
-# DB
+# DB与DB++
 
 - [1. 算法简介](#1)
 - [2. 环境配置](#2)
@@ -21,12 +21,24 @@
 > Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang
 > AAAI, 2020
 
+> [Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusion](https://arxiv.org/abs/2202.10304)
+> Liao, Minghui and Zou, Zhisheng and Wan, Zhaoyi and Yao, Cong and Bai, Xiang
+> TPAMI, 2022
+
+
 在ICDAR2015文本检测公开数据集上，算法复现效果如下：
 
 |模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
 | --- | --- | --- | --- | --- | --- | --- |
 |DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](../../configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
 |DB|MobileNetV3|[configs/det/det_mv3_db.yml](../../configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|DB++|ResNet50|[configs/det/det_r50_db++_ic15.yml](../../configs/det/det_r50_db++_ic15.yml)|90.89%|82.66%|86.58%|[合成数据预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_icdar15_train.tar)|
+
+在TD_TR文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|DB++|ResNet50|[configs/det/det_r50_db++_td_tr.yml](../../configs/det/det_r50_db++_td_tr.yml)|92.92%|86.48%|89.58%|[合成数据预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_td_tr_train.tar)|
 
 
 <a name="2"></a>
@@ -96,4 +108,12 @@ DB模型还支持以下推理部署方式：
   pages={11474--11481},
   year={2020}
 }
-```
\ No newline at end of file
+
+@article{liao2022real,
+  title={Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusion},
+  author={Liao, Minghui and Zou, Zhisheng and Wan, Zhaoyi and Yao, Cong and Bai, Xiang},
+  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
+  year={2022},
+  publisher={IEEE}
+}
+```
diff --git a/doc/doc_ch/dataset/ocr_datasets.md b/doc/doc_ch/dataset/ocr_datasets.md
index c6ff2e170f..b7666fd63e 100644
--- a/doc/doc_ch/dataset/ocr_datasets.md
+++ b/doc/doc_ch/dataset/ocr_datasets.md
@@ -34,6 +34,7 @@ json.dumps编码前的图像标注信息是包含多个字典的list，字典中
 | ICDAR 2015 |https://rrc.cvc.uab.es/?ch=4&com=downloads| [train](https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt) / [test](https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt) |
 | ctw1500 |https://paddleocr.bj.bcebos.com/dataset/ctw1500.zip| 图片下载地址中已包含 |
 | total text |https://paddleocr.bj.bcebos.com/dataset/total_text.tar| 图片下载地址中已包含 |
+| td tr |https://paddleocr.bj.bcebos.com/dataset/TD_TR.tar| 图片下载地址中已包含 |
 
 #### 1.2.1 ICDAR 2015
 ICDAR 2015 数据集包含1000张训练图像和500张测试图像。ICDAR 2015 数据集可以从上表中链接下载，首次下载需注册。

From 1315cdfc8633ebb501f56c16a0b5dff3635eb7af Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 28 Jun 2022 06:17:45 +0000
Subject: [PATCH 075/134] add db++

---
 ppocr/modeling/backbones/det_resnet.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/ppocr/modeling/backbones/det_resnet.py b/ppocr/modeling/backbones/det_resnet.py
index 3d8ce44508..87eef11cf0 100644
--- a/ppocr/modeling/backbones/det_resnet.py
+++ b/ppocr/modeling/backbones/det_resnet.py
@@ -209,7 +209,6 @@ class ResNet(nn.Layer):
             for block in range(len(depth)):
                 shortcut = False
                 block_list = []
-                # is_dcn = self.dcn_stage[block]
                 for i in range(depth[block]):
                     conv_name = "res" + str(block + 2) + chr(97 + i)
                     basic_block = self.add_sublayer(

From 671415bbd878bd6d4bc392b22b7a61afa9e6048b Mon Sep 17 00:00:00 2001
From: topduke <784990967@qq.com>
Date: Tue, 28 Jun 2022 15:06:53 +0800
Subject: [PATCH 076/134] [New Rec]add rec ViTSTR & ABINet algorithm. (#6414)

* add rec vitstr algorithm.

* fix cpu_thread and precision

* fix svtr tipc

* modify vitstr name

* modify vitstr config batchsize

* [New Rec] add vitstr and ABINet

* add rec_resnet45

* svtr ch large model

* [application] svtr ch model

* [application] svtr ch model

* [application] svtr ch model

* add abinet_rec_aug and trained model

* aug p infe

* fix ci export bug

* fix abinet ci bug
---
 applications/高精度中文识别模型.md            | 107 ++++
 configs/rec/rec_mtb_nrtr.yml                  |  16 +-
 configs/rec/rec_r45_abinet.yml                | 101 ++++
 configs/rec/rec_svtrnet.yml                   |  12 +-
 configs/rec/rec_svtrnet_ch.yml                | 155 ++++++
 configs/rec/rec_vitstr_none_ce.yml            | 102 ++++
 doc/doc_ch/algorithm_overview.md              |   5 +-
 doc/doc_ch/algorithm_rec_abinet.md            | 155 ++++++
 doc/doc_ch/algorithm_rec_nrtr.md              | 140 ++++-
 doc/doc_ch/algorithm_rec_svtr.md              |   1 -
 doc/doc_ch/algorithm_rec_vitstr.md            | 154 ++++++
 doc/doc_en/algorithm_overview_en.md           |   5 +-
 doc/doc_en/algorithm_rec_abinet_en.md         | 136 +++++
 doc/doc_en/algorithm_rec_nrtr_en.md           | 141 ++++-
 doc/doc_en/algorithm_rec_svtr_en.md           |   1 -
 doc/doc_en/algorithm_rec_vitstr_en.md         | 134 +++++
 ppocr/data/imaug/__init__.py                  |   6 +-
 ppocr/data/imaug/abinet_aug.py                | 407 +++++++++++++++
 ppocr/data/imaug/label_ops.py                 | 127 +++--
 ppocr/data/imaug/operators.py                 |  33 --
 ppocr/data/imaug/rec_img_aug.py               | 170 ++++--
 ppocr/losses/__init__.py                      |   4 +-
 ppocr/losses/rec_ce_loss.py                   |  66 +++
 ppocr/losses/rec_nrtr_loss.py                 |  30 --
 ppocr/modeling/backbones/__init__.py          |  18 +-
 ppocr/modeling/backbones/rec_resnet_45.py     | 147 ++++++
 ppocr/modeling/backbones/rec_svtrnet.py       | 102 ++--
 ppocr/modeling/backbones/rec_vitstr.py        | 120 +++++
 ppocr/modeling/heads/__init__.py              |   3 +-
 ppocr/modeling/heads/multiheadAttention.py    | 163 ------
 ppocr/modeling/heads/rec_abinet_head.py       | 296 +++++++++++
 ppocr/modeling/heads/rec_nrtr_head.py         | 489 ++++++------------
 ppocr/postprocess/__init__.py                 |   4 +-
 ppocr/postprocess/rec_postprocess.py          | 183 ++++---
 .../configs/rec_mtb_nrtr/rec_mtb_nrtr.yml     |   8 +-
 .../configs/rec_r45_abinet/rec_r45_abinet.yml | 106 ++++
 .../rec_r45_abinet/train_infer_python.txt     |  53 ++
 test_tipc/configs/rec_svtrnet/rec_svtrnet.yml | 117 +++++
 .../rec_svtrnet/train_infer_python.txt        |  53 ++
 .../rec_vitstr_none_ce/rec_vitstr_none_ce.yml | 104 ++++
 .../rec_vitstr_none_ce/train_infer_python.txt |  53 ++
 tools/export_model.py                         |  40 +-
 tools/infer/predict_rec.py                    |  70 ++-
 tools/program.py                              |   3 +-
 44 files changed, 3550 insertions(+), 790 deletions(-)
 create mode 100644 applications/高精度中文识别模型.md
 create mode 100644 configs/rec/rec_r45_abinet.yml
 create mode 100644 configs/rec/rec_svtrnet_ch.yml
 create mode 100644 configs/rec/rec_vitstr_none_ce.yml
 create mode 100644 doc/doc_ch/algorithm_rec_abinet.md
 create mode 100644 doc/doc_ch/algorithm_rec_vitstr.md
 create mode 100644 doc/doc_en/algorithm_rec_abinet_en.md
 create mode 100644 doc/doc_en/algorithm_rec_vitstr_en.md
 create mode 100644 ppocr/data/imaug/abinet_aug.py
 create mode 100644 ppocr/losses/rec_ce_loss.py
 delete mode 100644 ppocr/losses/rec_nrtr_loss.py
 create mode 100644 ppocr/modeling/backbones/rec_resnet_45.py
 create mode 100644 ppocr/modeling/backbones/rec_vitstr.py
 delete mode 100755 ppocr/modeling/heads/multiheadAttention.py
 create mode 100644 ppocr/modeling/heads/rec_abinet_head.py
 create mode 100644 test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
 create mode 100644 test_tipc/configs/rec_r45_abinet/train_infer_python.txt
 create mode 100644 test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
 create mode 100644 test_tipc/configs/rec_svtrnet/train_infer_python.txt
 create mode 100644 test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
 create mode 100644 test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt

diff --git a/applications/高精度中文识别模型.md b/applications/高精度中文识别模型.md
new file mode 100644
index 0000000000..3c31af42ee
--- /dev/null
+++ b/applications/高精度中文识别模型.md
@@ -0,0 +1,107 @@
+# 高精度中文场景文本识别模型SVTR
+
+## 1. 简介
+
+PP-OCRv3是百度开源的超轻量级场景文本检测识别模型库，其中超轻量的场景中文识别模型SVTR_LCNet使用了SVTR算法结构。为了保证速度，SVTR_LCNet将SVTR模型的Local Blocks替换为LCNet，使用两层Global Blocks。在中文场景中，PP-OCRv3识别主要使用如下优化策略：
+- GTC：Attention指导CTC训练策略；
+- TextConAug：挖掘文字上下文信息的数据增广策略；
+- TextRotNet：自监督的预训练模型；
+- UDML：联合互学习策略；
+- UIM：无标注数据挖掘方案。
+
+其中 *UIM：无标注数据挖掘方案* 使用了高精度的SVTR中文模型进行无标注文件的刷库，该模型在PP-OCRv3识别的数据集上训练，精度对比如下表。
+
+|中文识别算法|模型|UIM|精度|
+| --- | --- | --- |--- |
+|PP-OCRv3|SVTR_LCNet| w/o |78.4%|
+|PP-OCRv3|SVTR_LCNet| w |79.4%|
+|SVTR|SVTR-Tiny|-|82.5%|
+
+aistudio项目链接: [高精度中文场景文本识别模型SVTR](https://aistudio.baidu.com/aistudio/projectdetail/4263032)
+
+## 2. SVTR中文模型使用
+
+### 环境准备
+
+
+本任务基于Aistudio完成, 具体环境如下：
+
+- 操作系统: Linux
+- PaddlePaddle: 2.3
+- PaddleOCR: dygraph
+
+下载 PaddleOCR代码
+
+```bash
+git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR
+```
+
+安装依赖库
+
+```bash
+pip install -r PaddleOCR/requirements.txt -i https://mirror.baidu.com/pypi/simple
+```
+
+### 快速使用
+
+获取SVTR中文模型文件，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+```bash
+# 解压模型文件
+tar xf svtr_ch_high_accuracy.tar
+```
+
+预测中文文本，以下图为例：
+![](../doc/imgs_words/ch/word_1.jpg)
+
+预测命令：
+
+```bash
+# CPU预测
+python tools/infer_rec.py -c configs/rec/rec_svtrnet_ch.yml -o Global.pretrained_model=./svtr_ch_high_accuracy/best_accuracy Global.infer_img=./doc/imgs_words/ch/word_1.jpg Global.use_gpu=False
+
+# GPU预测
+#python tools/infer_rec.py -c configs/rec/rec_svtrnet_ch.yml -o Global.pretrained_model=./svtr_ch_high_accuracy/best_accuracy Global.infer_img=./doc/imgs_words/ch/word_1.jpg Global.use_gpu=True
+```
+
+可以看到最后打印结果为
+- result: 韩国小馆    0.9853458404541016
+
+0.9853458404541016为预测置信度。
+
+### 推理模型导出与预测
+
+inference 模型（paddle.jit.save保存的模型） 一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。 训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。 与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+运行识别模型转inference模型命令，如下：
+
+```bash
+python tools/export_model.py -c configs/rec/rec_svtrnet_ch.yml -o Global.pretrained_model=./svtr_ch_high_accuracy/best_accuracy Global.save_inference_dir=./inference/svtr_ch
+```
+
+转换成功后，在目录下有三个文件：
+```shell
+inference/svtr_ch/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+inference模型预测，命令如下：
+
+```bash
+# CPU预测
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_1.jpg" --rec_algorithm='SVTR' --rec_model_dir=./inference/svtr_ch/ --rec_image_shape='3, 32, 320'  --rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt --use_gpu=False
+
+# GPU预测
+#python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_1.jpg" --rec_algorithm='SVTR' --rec_model_dir=./inference/svtr_ch/ --rec_image_shape='3, 32, 320'  --rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt --use_gpu=True
+```
+
+**注意**
+
+- 使用SVTR算法时，需要指定--rec_algorithm='SVTR'
+- 如果使用自定义字典训练的模型，需要将--rec_char_dict_path=ppocr/utils/ppocr_keys_v1.txt修改为自定义的字典
+- --rec_image_shape='3, 32, 320' 该参数不能去掉
diff --git a/configs/rec/rec_mtb_nrtr.yml b/configs/rec/rec_mtb_nrtr.yml
index 0426750085..4e5826adc9 100644
--- a/configs/rec/rec_mtb_nrtr.yml
+++ b/configs/rec/rec_mtb_nrtr.yml
@@ -9,7 +9,7 @@ Global:
   eval_batch_step: [0, 2000]
   cal_metric_during_train: True
   pretrained_model:
-  checkpoints: 
+  checkpoints:
   save_inference_dir:
   use_visualdl: False
   infer_img: doc/imgs_words_en/word_10.png
@@ -49,7 +49,7 @@ Architecture:
     
 
 Loss:
-  name: NRTRLoss
+  name: CELoss
   smoothing: True
 
 PostProcess:
@@ -68,8 +68,8 @@ Train:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
-          image_shape: [100, 32]
+      - GrayRecResizeImg:
+          image_shape: [100, 32] # W H
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
           keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
@@ -82,14 +82,14 @@ Train:
 Eval:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/evaluation/
+    data_dir: ./train_data/data_lmdb_release/validation/
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
-          image_shape: [100, 32]
+      - GrayRecResizeImg:
+          image_shape: [100, 32] # W H
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
           keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
@@ -97,5 +97,5 @@ Eval:
     shuffle: False
     drop_last: False
     batch_size_per_card: 256
-    num_workers: 1
+    num_workers: 4
     use_shared_memory: False
diff --git a/configs/rec/rec_r45_abinet.yml b/configs/rec/rec_r45_abinet.yml
new file mode 100644
index 0000000000..e604bcd12d
--- /dev/null
+++ b/configs/rec/rec_r45_abinet.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r45_abinet/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_abinet.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 20.0
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: ABINet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ResNet45
+  Head:
+    name: ABINetHead
+    use_lang: True
+    iter_size: 3
+    
+
+Loss:
+  name: CELoss
+  ignore_index: &ignore_index 100 # Must be greater than the number of character classes
+
+PostProcess:
+  name: ABINetLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetRecAug:
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 96
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/configs/rec/rec_svtrnet.yml b/configs/rec/rec_svtrnet.yml
index 233d5e2765..c1f5cc380a 100644
--- a/configs/rec/rec_svtrnet.yml
+++ b/configs/rec/rec_svtrnet.yml
@@ -26,7 +26,7 @@ Optimizer:
   name: AdamW
   beta1: 0.9
   beta2: 0.99
-  epsilon: 0.00000008
+  epsilon: 8.e-8
   weight_decay: 0.05
   no_weight_decay_name: norm pos_embed
   one_dim_param_no_weight_decay: true
@@ -77,14 +77,13 @@ Metric:
 Train:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/training/
+    data_dir: ./train_data/data_lmdb_release/training
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
       - CTCLabelEncode: # Class handling label
-      - RecResizeImg:
-          character_dict_path:
+      - SVTRRecResizeImg:
           image_shape: [3, 64, 256]
           padding: False
       - KeepKeys:
@@ -98,14 +97,13 @@ Train:
 Eval:
   dataset:
     name: LMDBDataSet
-    data_dir: ./train_data/data_lmdb_release/validation/
+    data_dir: ./train_data/data_lmdb_release/validation
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
       - CTCLabelEncode: # Class handling label
-      - RecResizeImg:
-          character_dict_path:
+      - SVTRRecResizeImg:
           image_shape: [3, 64, 256]
           padding: False
       - KeepKeys:
diff --git a/configs/rec/rec_svtrnet_ch.yml b/configs/rec/rec_svtrnet_ch.yml
new file mode 100644
index 0000000000..0d3f63d125
--- /dev/null
+++ b/configs/rec/rec_svtrnet_ch.yml
@@ -0,0 +1,155 @@
+Global:
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr_ch_all/
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  save_res_path: ./output/rec/predicts_svtr_tiny_ch_all.txt
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 8.0e-08
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform: null
+  Backbone:
+    name: SVTRNet
+    img_size:
+    - 32
+    - 320
+    out_char_num: 40
+    out_channels: 96
+    patch_merging: Conv
+    embed_dim:
+    - 64
+    - 128
+    - 256
+    depth:
+    - 3
+    - 6
+    - 3
+    num_heads:
+    - 2
+    - 4
+    - 8
+    mixer:
+    - Local
+    - Local
+    - Local
+    - Local
+    - Local
+    - Local
+    - Global
+    - Global
+    - Global
+    - Global
+    - Global
+    - Global
+    local_mixer:
+    - - 7
+      - 11
+    - - 7
+      - 11
+    - - 7
+      - 11
+    last_stage: true
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/train_list.txt
+    ext_op_transform_idx: 1
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape:
+        - 32
+        - 320
+        - 3
+    - RecAug: null
+    - CTCLabelEncode: null
+    - SVTRRecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+        padding: true
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - SVTRRecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+        padding: true
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 2
+profiler_options: null
diff --git a/configs/rec/rec_vitstr_none_ce.yml b/configs/rec/rec_vitstr_none_ce.yml
new file mode 100644
index 0000000000..b969c83a5d
--- /dev/null
+++ b/configs/rec/rec_vitstr_none_ce.yml
@@ -0,0 +1,102 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/vitstr_none_ce/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration#
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_vitstr.txt
+
+
+Optimizer:
+  name: Adadelta
+  epsilon: 1.e-8
+  rho: 0.95
+  clip_norm: 5.0
+  lr:
+    learning_rate: 1.0
+
+Architecture:
+  model_type: rec
+  algorithm: ViTSTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ViTSTR
+    scale: tiny
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CELoss
+  with_all: True
+  ignore_index: &ignore_index 0 # Must be zero or greater than the number of character classes
+
+PostProcess:
+  name: ViTSTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index 6227a21498..eb81e4cd6d 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -66,6 +66,8 @@
 - [x]  [SAR](./algorithm_rec_sar.md)
 - [x]  [SEED](./algorithm_rec_seed.md)
 - [x]  [SVTR](./algorithm_rec_svtr.md)
+- [x]  [ViTSTR](./algorithm_rec_vitstr.md)
+- [x]  [ABINet](./algorithm_rec_abinet.md)
 
 参考[DTRB](https://arxiv.org/abs/1904.01906)[3]文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
 
@@ -84,7 +86,8 @@
 |SAR|Resnet31| 87.20% | rec_r31_sar | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_ch/algorithm_rec_abinet.md b/doc/doc_ch/algorithm_rec_abinet.md
new file mode 100644
index 0000000000..47507c36c7
--- /dev/null
+++ b/doc/doc_ch/algorithm_rec_abinet.md
@@ -0,0 +1,155 @@
+# 场景文本识别算法-ABINet
+
+- [1. 算法简介](#1)
+- [2. 环境配置](#2)
+- [3. 模型训练、评估、预测](#3)
+    - [3.1 训练](#3-1)
+    - [3.2 评估](#3-2)
+    - [3.3 预测](#3-3)
+- [4. 推理部署](#4)
+    - [4.1 Python推理](#4-1)
+    - [4.2 C++推理](#4-2)
+    - [4.3 Serving服务化部署](#4-3)
+    - [4.4 更多推理部署](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. 算法简介
+
+论文信息：
+> [ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition](https://openaccess.thecvf.com/content/CVPR2021/papers/Fang_Read_Like_Humans_Autonomous_Bidirectional_and_Iterative_Language_Modeling_for_CVPR_2021_paper.pdf)
+> Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang
+> CVPR, 2021
+
+
+<a name="model"></a>
+`ABINet`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[预训练、训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)|
+
+<a name="2"></a>
+## 2. 环境配置
+请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](./clone.md)克隆项目代码。
+
+
+<a name="3"></a>
+## 3. 模型训练、评估、预测
+
+<a name="3-1"></a>
+### 3.1 模型训练
+
+请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ABINet`识别模型时需要**更换配置文件**为`ABINet`的[配置文件](../../configs/rec/rec_r45_abinet.yml)。
+
+#### 启动训练
+
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r45_abinet.yml
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_abinet.yml
+```
+
+<a name="3-2"></a>
+### 3.2 评估
+
+可下载已训练完成的[模型文件](#model)，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+```
+
+<a name="3-3"></a>
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+
+<a name="4"></a>
+## 4. 推理部署
+
+<a name="4-1"></a>
+### 4.1 Python推理
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) )，可以使用如下命令进行转换：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy Global.save_inference_dir=./inference/rec_r45_abinet/
+```
+**注意：**
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ABINet的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+```
+/inference/rec_r45_abinet/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```shell
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_r45_abinet/' --rec_algorithm='ABINet' --rec_image_shape='3,32,128' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![](../imgs_words_en/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999995231628418)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[3，32，128]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ABINet的预处理为您的预处理方法。
+
+
+<a name="4-2"></a>
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持ABINet，所以暂未支持
+
+<a name="4-3"></a>
+### 4.3 Serving服务化部署
+
+暂不支持
+
+<a name="4-4"></a>
+### 4.4 更多推理部署
+
+暂不支持
+
+<a name="5"></a>
+## 5. FAQ
+
+1. MJSynth和SynthText两种数据集来自于[ABINet源repo](https://github.com/FangShancheng/ABINet) 。
+2. 我们使用ABINet作者提供的预训练模型进行finetune训练。
+
+## 引用
+
+```bibtex
+@article{Fang2021ABINet,
+  title     = {ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition},
+  author    = {Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang},
+  booktitle = {CVPR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2103.06495},
+  pages     = {7098-7107}
+}
+```
diff --git a/doc/doc_ch/algorithm_rec_nrtr.md b/doc/doc_ch/algorithm_rec_nrtr.md
index d3b626d024..c619ac1dbc 100644
--- a/doc/doc_ch/algorithm_rec_nrtr.md
+++ b/doc/doc_ch/algorithm_rec_nrtr.md
@@ -12,6 +12,7 @@
     - [4.3 Serving服务化部署](#4-3)
     - [4.4 更多推理部署](#4-4)
 - [5. FAQ](#5)
+- [6. 发行公告](#6)
 
 <a name="1"></a>
 ## 1. 算法简介
@@ -110,7 +111,7 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png'
 执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
 结果如下：
 ```shell
-Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9465042352676392)
 ```
 
 **注意**：
@@ -140,12 +141,147 @@ Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
 
 1. `NRTR`论文中使用Beam搜索进行解码字符，但是速度较慢，这里默认未使用Beam搜索，以贪婪搜索进行解码字符。
 
+<a name="6"></a>
+## 6. 发行公告
+
+1. release/2.6更新NRTR代码结构，新版NRTR可加载旧版(release/2.5及之前)模型参数，使用下面示例代码将旧版模型参数转换为新版模型参数：
+
+```python
+
+    params = paddle.load('path/' + '.pdparams') # 旧版本参数
+    state_dict = model.state_dict() # 新版模型参数
+    new_state_dict = {}
+
+    for k1, v1 in state_dict.items():
+
+        k = k1
+        if 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'encoder' in k and 'norm3' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para.replace('norm3', 'norm2')]
+
+        elif 'encoder' in k and 'norm1' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = q[:, :, 0, 0]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')]
+            new_state_dict[k1] = q
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('kv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')]
+            v = params[k_para.replace('kv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([k, v], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            new_state_dict[k1] = params[k_para]
+        elif 'decoder' in k and 'norm' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+        elif 'mlp' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = w[:, :, 0, 0]
+        elif 'mlp' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para]
+            new_state_dict[k1] = w
+
+        else:
+            new_state_dict[k1] = params[k1]
+
+        if list(new_state_dict[k1].shape) != list(v1.shape):
+            print(k1)
+
+
+    for k, v1 in state_dict.items():
+        if k not in new_state_dict.keys():
+            print(1, k)
+        elif list(new_state_dict[k].shape) != list(v1.shape):
+            print(2, k)
+
+
+
+    model.set_state_dict(new_state_dict)
+    paddle.save(model.state_dict(), 'nrtrnew_from_old_params.pdparams')
+
+```
+
+2. 新版相比与旧版，代码结构简洁，推理速度有所提高。
+
+
 ## 引用
 
 ```bibtex
 @article{Sheng2019NRTR,
   title     = {NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition},
-  author    = {Fenfen Sheng and Zhineng Chen andBo Xu},
+  author    = {Fenfen Sheng and Zhineng Chen and Bo Xu},
   booktitle = {ICDAR},
   year      = {2019},
   url       = {http://arxiv.org/abs/1806.00926},
diff --git a/doc/doc_ch/algorithm_rec_svtr.md b/doc/doc_ch/algorithm_rec_svtr.md
index 41a22ca65c..c0e26433e9 100644
--- a/doc/doc_ch/algorithm_rec_svtr.md
+++ b/doc/doc_ch/algorithm_rec_svtr.md
@@ -111,7 +111,6 @@ python3 tools/export_model.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny
 
 **注意：**
 - 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
-- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应SVTR的`infer_shape`。
 
 转换成功后，在目录下有三个文件：
 ```
diff --git a/doc/doc_ch/algorithm_rec_vitstr.md b/doc/doc_ch/algorithm_rec_vitstr.md
new file mode 100644
index 0000000000..ab12be7202
--- /dev/null
+++ b/doc/doc_ch/algorithm_rec_vitstr.md
@@ -0,0 +1,154 @@
+# 场景文本识别算法-ViTSTR
+
+- [1. 算法简介](#1)
+- [2. 环境配置](#2)
+- [3. 模型训练、评估、预测](#3)
+    - [3.1 训练](#3-1)
+    - [3.2 评估](#3-2)
+    - [3.3 预测](#3-3)
+- [4. 推理部署](#4)
+    - [4.1 Python推理](#4-1)
+    - [4.2 C++推理](#4-2)
+    - [4.3 Serving服务化部署](#4-3)
+    - [4.4 更多推理部署](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. 算法简介
+
+论文信息：
+> [Vision Transformer for Fast and Efficient Scene Text Recognition](https://arxiv.org/abs/2105.08582)
+> Rowel Atienza
+> ICDAR, 2021
+
+
+<a name="model"></a>
+`ViTSTR`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar)|
+
+<a name="2"></a>
+## 2. 环境配置
+请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](./clone.md)克隆项目代码。
+
+
+<a name="3"></a>
+## 3. 模型训练、评估、预测
+
+<a name="3-1"></a>
+### 3.1 模型训练
+
+请参考[文本识别训练教程](./recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](../../configs/rec/rec_vitstr_none_ce.yml)。
+
+#### 启动训练
+
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+```
+
+<a name="3-2"></a>
+### 3.2 评估
+
+可下载已训练完成的[模型文件](#model)，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
+```
+
+<a name="3-3"></a>
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+
+<a name="4"></a>
+## 4. 推理部署
+
+<a name="4-1"></a>
+### 4.1 Python推理
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) )，可以使用如下命令进行转换：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/
+```
+**注意：**
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ViTSTR的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+```
+/inference/rec_vitstr/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```shell
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_vitstr/' --rec_algorithm='ViTSTR' --rec_image_shape='1,224,224' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![](../imgs_words_en/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[1，224，224]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ViTSTR的预处理为您的预处理方法。
+
+
+<a name="4-2"></a>
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持ViTSTR，所以暂未支持
+
+<a name="4-3"></a>
+### 4.3 Serving服务化部署
+
+暂不支持
+
+<a name="4-4"></a>
+### 4.4 更多推理部署
+
+暂不支持
+
+<a name="5"></a>
+## 5. FAQ
+
+1. 在`ViTSTR`论文中，使用在ImageNet1k上的预训练权重进行初始化训练，我们在训练未采用预训练权重，最终精度没有变化甚至有所提高。
+2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果需要使用small、base版本，可将[ViTSTR源repo](https://github.com/roatienza/deep-text-recognition-benchmark) 中的预训练权重转为Paddle权重使用。
+
+## 引用
+
+```bibtex
+@article{Atienza2021ViTSTR,
+  title     = {Vision Transformer for Fast and Efficient Scene Text Recognition},
+  author    = {Rowel Atienza},
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.08582}
+}
+```
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index 383cbe39bb..28aca7c0d1 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -65,6 +65,8 @@ Supported text recognition algorithms (Click the link to get the tutorial):
 - [x]  [SAR](./algorithm_rec_sar_en.md)
 - [x]  [SEED](./algorithm_rec_seed_en.md)
 - [x]  [SVTR](./algorithm_rec_svtr_en.md)
+- [x]  [ViTSTR](./algorithm_rec_vitstr_en.md)
+- [x]  [ABINet](./algorithm_rec_abinet_en.md)
 
 Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
 
@@ -83,7 +85,8 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 |SAR|Resnet31| 87.20% | rec_r31_sar | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet_en | [trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_en/algorithm_rec_abinet_en.md b/doc/doc_en/algorithm_rec_abinet_en.md
new file mode 100644
index 0000000000..767ca65f64
--- /dev/null
+++ b/doc/doc_en/algorithm_rec_abinet_en.md
@@ -0,0 +1,136 @@
+# ABINet
+
+- [1. Introduction](#1)
+- [2. Environment](#2)
+- [3. Model Training / Evaluation / Prediction](#3)
+    - [3.1 Training](#3-1)
+    - [3.2 Evaluation](#3-2)
+    - [3.3 Prediction](#3-3)
+- [4. Inference and Deployment](#4)
+    - [4.1 Python Inference](#4-1)
+    - [4.2 C++ Inference](#4-2)
+    - [4.3 Serving](#4-3)
+    - [4.4 More](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. Introduction
+
+Paper:
+> [ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition](https://openaccess.thecvf.com/content/CVPR2021/papers/Fang_Read_Like_Humans_Autonomous_Bidirectional_and_Iterative_Language_Modeling_for_CVPR_2021_paper.pdf)
+> Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang
+> CVPR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|ABINet|ResNet45|[rec_r45_abinet.yml](../../configs/rec/rec_r45_abinet.yml)|90.75%|[pretrained & trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)|
+
+<a name="2"></a>
+## 2. Environment
+Please refer to ["Environment Preparation"](./environment_en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone_en.md) to clone the project code.
+
+
+<a name="3"></a>
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](./recognition_en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```
+#Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r45_abinet.yml
+
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_abinet.yml
+```
+
+Evaluation:
+
+```
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+Prediction:
+
+```
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+```
+
+<a name="4"></a>
+## 4. Inference and Deployment
+
+<a name="4-1"></a>
+### 4.1 Python Inference
+First, the model saved during the ABINet text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)) ), you can use the following command to convert:
+
+```
+python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy  Global.save_inference_dir=./inference/rec_r45_abinet
+```
+
+**Note:**
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to ABINet in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+```
+/inference/rec_r45_abinet/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+
+For ABINet text recognition model inference, the following commands can be executed:
+
+```
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_r45_abinet/' --rec_algorithm='ABINet' --rec_image_shape='3,32,128' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+```
+
+![](../imgs_words_en/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999995231628418)
+```
+
+<a name="4-2"></a>
+### 4.2 C++ Inference
+
+Not supported
+
+<a name="4-3"></a>
+### 4.3 Serving
+
+Not supported
+
+<a name="4-4"></a>
+### 4.4 More
+
+Not supported
+
+<a name="5"></a>
+## 5. FAQ
+
+1. Note that the MJSynth and SynthText datasets come from [ABINet repo](https://github.com/FangShancheng/ABINet).
+2. We use the pre-trained model provided by the ABINet authors for finetune training.
+
+## Citation
+
+```bibtex
+@article{Fang2021ABINet,
+  title     = {ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition},
+  author    = {Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang},
+  booktitle = {CVPR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2103.06495},
+  pages     = {7098-7107}
+}
+```
diff --git a/doc/doc_en/algorithm_rec_nrtr_en.md b/doc/doc_en/algorithm_rec_nrtr_en.md
index 40c9b91629..309d7ab123 100644
--- a/doc/doc_en/algorithm_rec_nrtr_en.md
+++ b/doc/doc_en/algorithm_rec_nrtr_en.md
@@ -12,6 +12,7 @@
     - [4.3 Serving](#4-3)
     - [4.4 More](#4-4)
 - [5. FAQ](#5)
+- [6. Release Note](#6)
 
 <a name="1"></a>
 ## 1. Introduction
@@ -25,7 +26,7 @@ Using MJSynth and SynthText two text recognition datasets for training, and eval
 
 |Model|Backbone|config|Acc|Download link|
 | --- | --- | --- | --- | --- |
-|NRTR|MTB|[rec_mtb_nrtr.yml](../../configs/rec/rec_mtb_nrtr.yml)|84.21%|[train model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)|
+|NRTR|MTB|[rec_mtb_nrtr.yml](../../configs/rec/rec_mtb_nrtr.yml)|84.21%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)|
 
 <a name="2"></a>
 ## 2. Environment
@@ -98,7 +99,7 @@ python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png'
 After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
 The result is as follows:
 ```shell
-Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9265879392623901)
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9465042352676392)
 ```
 
 <a name="4-2"></a>
@@ -121,12 +122,146 @@ Not supported
 
 1. In the `NRTR` paper, Beam search is used to decode characters, but the speed is slow. Beam search is not used by default here, and greedy search is used to decode characters.
 
+<a name="6"></a>
+## 6. Release Note
+
+1. The release/2.6 version updates the NRTR code structure. The new version of NRTR can load the model parameters of the old version (release/2.5 and before), and you may use the following code to convert the old version model parameters to the new version model parameters:
+
+```python
+
+    params = paddle.load('path/' + '.pdparams') # the old version parameters
+    state_dict = model.state_dict() # the new version model parameters
+    new_state_dict = {}
+
+    for k1, v1 in state_dict.items():
+
+        k = k1
+        if 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'encoder' in k and 'norm3' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para.replace('norm3', 'norm2')]
+
+        elif 'encoder' in k and 'norm1' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = q[:, :, 0, 0]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')]
+            new_state_dict[k1] = q
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('kv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')]
+            v = params[k_para.replace('kv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([k, v], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            new_state_dict[k1] = params[k_para]
+        elif 'decoder' in k and 'norm' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+        elif 'mlp' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = w[:, :, 0, 0]
+        elif 'mlp' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para]
+            new_state_dict[k1] = w
+
+        else:
+            new_state_dict[k1] = params[k1]
+
+        if list(new_state_dict[k1].shape) != list(v1.shape):
+            print(k1)
+
+
+    for k, v1 in state_dict.items():
+        if k not in new_state_dict.keys():
+            print(1, k)
+        elif list(new_state_dict[k].shape) != list(v1.shape):
+            print(2, k)
+
+
+
+    model.set_state_dict(new_state_dict)
+    paddle.save(model.state_dict(), 'nrtrnew_from_old_params.pdparams')
+
+```
+
+2. The new version has a clean code structure and improved inference speed compared with the old version.
+
 ## Citation
 
 ```bibtex
 @article{Sheng2019NRTR,
   title     = {NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition},
-  author    = {Fenfen Sheng and Zhineng Chen andBo Xu},
+  author    = {Fenfen Sheng and Zhineng Chen and Bo Xu},
   booktitle = {ICDAR},
   year      = {2019},
   url       = {http://arxiv.org/abs/1806.00926},
diff --git a/doc/doc_en/algorithm_rec_svtr_en.md b/doc/doc_en/algorithm_rec_svtr_en.md
index d402a6b491..37cd35f35a 100644
--- a/doc/doc_en/algorithm_rec_svtr_en.md
+++ b/doc/doc_en/algorithm_rec_svtr_en.md
@@ -88,7 +88,6 @@ python3 tools/export_model.py -c configs/rec/rec_svtrnet.yml -o Global.pretraine
 
 **Note:**
 - If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
-- If you modified the input size during training, please modify the `infer_shape` corresponding to SVTR in the `tools/export_model.py` file.
 
 After the conversion is successful, there are three files in the directory:
 ```
diff --git a/doc/doc_en/algorithm_rec_vitstr_en.md b/doc/doc_en/algorithm_rec_vitstr_en.md
new file mode 100644
index 0000000000..a6f9e2f15d
--- /dev/null
+++ b/doc/doc_en/algorithm_rec_vitstr_en.md
@@ -0,0 +1,134 @@
+# ViTSTR
+
+- [1. Introduction](#1)
+- [2. Environment](#2)
+- [3. Model Training / Evaluation / Prediction](#3)
+    - [3.1 Training](#3-1)
+    - [3.2 Evaluation](#3-2)
+    - [3.3 Prediction](#3-3)
+- [4. Inference and Deployment](#4)
+    - [4.1 Python Inference](#4-1)
+    - [4.2 C++ Inference](#4-2)
+    - [4.3 Serving](#4-3)
+    - [4.4 More](#4-4)
+- [5. FAQ](#5)
+
+<a name="1"></a>
+## 1. Introduction
+
+Paper:
+> [Vision Transformer for Fast and Efficient Scene Text Recognition](https://arxiv.org/abs/2105.08582)
+> Rowel Atienza
+> ICDAR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)|
+
+<a name="2"></a>
+## 2. Environment
+Please refer to ["Environment Preparation"](./environment_en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone_en.md) to clone the project code.
+
+
+<a name="3"></a>
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](./recognition_en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```
+#Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+```
+
+Evaluation:
+
+```
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+Prediction:
+
+```
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
+```
+
+<a name="4"></a>
+## 4. Inference and Deployment
+
+<a name="4-1"></a>
+### 4.1 Python Inference
+First, the model saved during the ViTSTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)) ), you can use the following command to convert:
+
+```
+python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy  Global.save_inference_dir=./inference/rec_vitstr
+```
+
+**Note:**
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to ViTSTR in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+```
+/inference/rec_vitstr/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+
+For ViTSTR text recognition model inference, the following commands can be executed:
+
+```
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_vitstr/' --rec_algorithm='ViTSTR' --rec_image_shape='1,224,224' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+```
+
+![](../imgs_words_en/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+```shell
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617)
+```
+
+<a name="4-2"></a>
+### 4.2 C++ Inference
+
+Not supported
+
+<a name="4-3"></a>
+### 4.3 Serving
+
+Not supported
+
+<a name="4-4"></a>
+### 4.4 More
+
+Not supported
+
+<a name="5"></a>
+## 5. FAQ
+
+1. In the `ViTSTR` paper, using pre-trained weights on ImageNet1k for initial training, we did not use pre-trained weights in training, and the final accuracy did not change or even improved.
+
+## Citation
+
+```bibtex
+@article{Atienza2021ViTSTR,
+  title     = {Vision Transformer for Fast and Efficient Scene Text Recognition},
+  author    = {Rowel Atienza},
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.08582}
+}
+```
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index f0fd578f61..63dfda91f8 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -22,8 +22,10 @@ from .make_shrink_map import MakeShrinkMap
 from .random_crop_data import EastRandomCropData, RandomCropImgMask
 from .make_pse_gt import MakePseGt
 
-from .rec_img_aug import BaseDataAugmentation, RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
-    SRNRecResizeImg, NRTRRecResizeImg, SARRecResizeImg, PRENResizeImg
+
+from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
+    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, \
+    ABINetRecResizeImg, SVTRRecResizeImg, ABINetRecAug
 from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste
diff --git a/ppocr/data/imaug/abinet_aug.py b/ppocr/data/imaug/abinet_aug.py
new file mode 100644
index 0000000000..eefdc75d5a
--- /dev/null
+++ b/ppocr/data/imaug/abinet_aug.py
@@ -0,0 +1,407 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FangShancheng/ABINet/blob/main/transforms.py
+"""
+import math
+import numbers
+import random
+
+import cv2
+import numpy as np
+from paddle.vision.transforms import Compose, ColorJitter
+
+
+def sample_asym(magnitude, size=None):
+    return np.random.beta(1, 4, size) * magnitude
+
+
+def sample_sym(magnitude, size=None):
+    return (np.random.beta(4, 4, size=size) - 0.5) * 2 * magnitude
+
+
+def sample_uniform(low, high, size=None):
+    return np.random.uniform(low, high, size=size)
+
+
+def get_interpolation(type='random'):
+    if type == 'random':
+        choice = [
+            cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA
+        ]
+        interpolation = choice[random.randint(0, len(choice) - 1)]
+    elif type == 'nearest':
+        interpolation = cv2.INTER_NEAREST
+    elif type == 'linear':
+        interpolation = cv2.INTER_LINEAR
+    elif type == 'cubic':
+        interpolation = cv2.INTER_CUBIC
+    elif type == 'area':
+        interpolation = cv2.INTER_AREA
+    else:
+        raise TypeError(
+            'Interpolation types only nearest, linear, cubic, area are supported!'
+        )
+    return interpolation
+
+
+class CVRandomRotation(object):
+    def __init__(self, degrees=15):
+        assert isinstance(degrees,
+                          numbers.Number), "degree should be a single number."
+        assert degrees >= 0, "degree must be positive."
+        self.degrees = degrees
+
+    @staticmethod
+    def get_params(degrees):
+        return sample_sym(degrees)
+
+    def __call__(self, img):
+        angle = self.get_params(self.degrees)
+        src_h, src_w = img.shape[:2]
+        M = cv2.getRotationMatrix2D(
+            center=(src_w / 2, src_h / 2), angle=angle, scale=1.0)
+        abs_cos, abs_sin = abs(M[0, 0]), abs(M[0, 1])
+        dst_w = int(src_h * abs_sin + src_w * abs_cos)
+        dst_h = int(src_h * abs_cos + src_w * abs_sin)
+        M[0, 2] += (dst_w - src_w) / 2
+        M[1, 2] += (dst_h - src_h) / 2
+
+        flags = get_interpolation()
+        return cv2.warpAffine(
+            img,
+            M, (dst_w, dst_h),
+            flags=flags,
+            borderMode=cv2.BORDER_REPLICATE)
+
+
+class CVRandomAffine(object):
+    def __init__(self, degrees, translate=None, scale=None, shear=None):
+        assert isinstance(degrees,
+                          numbers.Number), "degree should be a single number."
+        assert degrees >= 0, "degree must be positive."
+        self.degrees = degrees
+
+        if translate is not None:
+            assert isinstance(translate, (tuple, list)) and len(translate) == 2, \
+                "translate should be a list or tuple and it must be of length 2."
+            for t in translate:
+                if not (0.0 <= t <= 1.0):
+                    raise ValueError(
+                        "translation values should be between 0 and 1")
+        self.translate = translate
+
+        if scale is not None:
+            assert isinstance(scale, (tuple, list)) and len(scale) == 2, \
+                "scale should be a list or tuple and it must be of length 2."
+            for s in scale:
+                if s <= 0:
+                    raise ValueError("scale values should be positive")
+        self.scale = scale
+
+        if shear is not None:
+            if isinstance(shear, numbers.Number):
+                if shear < 0:
+                    raise ValueError(
+                        "If shear is a single number, it must be positive.")
+                self.shear = [shear]
+            else:
+                assert isinstance(shear, (tuple, list)) and (len(shear) == 2), \
+                    "shear should be a list or tuple and it must be of length 2."
+                self.shear = shear
+        else:
+            self.shear = shear
+
+    def _get_inverse_affine_matrix(self, center, angle, translate, scale,
+                                   shear):
+        # https://github.com/pytorch/vision/blob/v0.4.0/torchvision/transforms/functional.py#L717
+        from numpy import sin, cos, tan
+
+        if isinstance(shear, numbers.Number):
+            shear = [shear, 0]
+
+        if not isinstance(shear, (tuple, list)) and len(shear) == 2:
+            raise ValueError(
+                "Shear should be a single value or a tuple/list containing " +
+                "two values. Got {}".format(shear))
+
+        rot = math.radians(angle)
+        sx, sy = [math.radians(s) for s in shear]
+
+        cx, cy = center
+        tx, ty = translate
+
+        # RSS without scaling
+        a = cos(rot - sy) / cos(sy)
+        b = -cos(rot - sy) * tan(sx) / cos(sy) - sin(rot)
+        c = sin(rot - sy) / cos(sy)
+        d = -sin(rot - sy) * tan(sx) / cos(sy) + cos(rot)
+
+        # Inverted rotation matrix with scale and shear
+        # det([[a, b], [c, d]]) == 1, since det(rotation) = 1 and det(shear) = 1
+        M = [d, -b, 0, -c, a, 0]
+        M = [x / scale for x in M]
+
+        # Apply inverse of translation and of center translation: RSS^-1 * C^-1 * T^-1
+        M[2] += M[0] * (-cx - tx) + M[1] * (-cy - ty)
+        M[5] += M[3] * (-cx - tx) + M[4] * (-cy - ty)
+
+        # Apply center translation: C * RSS^-1 * C^-1 * T^-1
+        M[2] += cx
+        M[5] += cy
+        return M
+
+    @staticmethod
+    def get_params(degrees, translate, scale_ranges, shears, height):
+        angle = sample_sym(degrees)
+        if translate is not None:
+            max_dx = translate[0] * height
+            max_dy = translate[1] * height
+            translations = (np.round(sample_sym(max_dx)),
+                            np.round(sample_sym(max_dy)))
+        else:
+            translations = (0, 0)
+
+        if scale_ranges is not None:
+            scale = sample_uniform(scale_ranges[0], scale_ranges[1])
+        else:
+            scale = 1.0
+
+        if shears is not None:
+            if len(shears) == 1:
+                shear = [sample_sym(shears[0]), 0.]
+            elif len(shears) == 2:
+                shear = [sample_sym(shears[0]), sample_sym(shears[1])]
+        else:
+            shear = 0.0
+
+        return angle, translations, scale, shear
+
+    def __call__(self, img):
+        src_h, src_w = img.shape[:2]
+        angle, translate, scale, shear = self.get_params(
+            self.degrees, self.translate, self.scale, self.shear, src_h)
+
+        M = self._get_inverse_affine_matrix((src_w / 2, src_h / 2), angle,
+                                            (0, 0), scale, shear)
+        M = np.array(M).reshape(2, 3)
+
+        startpoints = [(0, 0), (src_w - 1, 0), (src_w - 1, src_h - 1),
+                       (0, src_h - 1)]
+        project = lambda x, y, a, b, c: int(a * x + b * y + c)
+        endpoints = [(project(x, y, *M[0]), project(x, y, *M[1]))
+                     for x, y in startpoints]
+
+        rect = cv2.minAreaRect(np.array(endpoints))
+        bbox = cv2.boxPoints(rect).astype(dtype=np.int)
+        max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max()
+        min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min()
+
+        dst_w = int(max_x - min_x)
+        dst_h = int(max_y - min_y)
+        M[0, 2] += (dst_w - src_w) / 2
+        M[1, 2] += (dst_h - src_h) / 2
+
+        # add translate
+        dst_w += int(abs(translate[0]))
+        dst_h += int(abs(translate[1]))
+        if translate[0] < 0: M[0, 2] += abs(translate[0])
+        if translate[1] < 0: M[1, 2] += abs(translate[1])
+
+        flags = get_interpolation()
+        return cv2.warpAffine(
+            img,
+            M, (dst_w, dst_h),
+            flags=flags,
+            borderMode=cv2.BORDER_REPLICATE)
+
+
+class CVRandomPerspective(object):
+    def __init__(self, distortion=0.5):
+        self.distortion = distortion
+
+    def get_params(self, width, height, distortion):
+        offset_h = sample_asym(
+            distortion * height / 2, size=4).astype(dtype=np.int)
+        offset_w = sample_asym(
+            distortion * width / 2, size=4).astype(dtype=np.int)
+        topleft = (offset_w[0], offset_h[0])
+        topright = (width - 1 - offset_w[1], offset_h[1])
+        botright = (width - 1 - offset_w[2], height - 1 - offset_h[2])
+        botleft = (offset_w[3], height - 1 - offset_h[3])
+
+        startpoints = [(0, 0), (width - 1, 0), (width - 1, height - 1),
+                       (0, height - 1)]
+        endpoints = [topleft, topright, botright, botleft]
+        return np.array(
+            startpoints, dtype=np.float32), np.array(
+                endpoints, dtype=np.float32)
+
+    def __call__(self, img):
+        height, width = img.shape[:2]
+        startpoints, endpoints = self.get_params(width, height, self.distortion)
+        M = cv2.getPerspectiveTransform(startpoints, endpoints)
+
+        # TODO: more robust way to crop image
+        rect = cv2.minAreaRect(endpoints)
+        bbox = cv2.boxPoints(rect).astype(dtype=np.int)
+        max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max()
+        min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min()
+        min_x, min_y = max(min_x, 0), max(min_y, 0)
+
+        flags = get_interpolation()
+        img = cv2.warpPerspective(
+            img,
+            M, (max_x, max_y),
+            flags=flags,
+            borderMode=cv2.BORDER_REPLICATE)
+        img = img[min_y:, min_x:]
+        return img
+
+
+class CVRescale(object):
+    def __init__(self, factor=4, base_size=(128, 512)):
+        """ Define image scales using gaussian pyramid and rescale image to target scale.
+        
+        Args:
+            factor: the decayed factor from base size, factor=4 keeps target scale by default.
+            base_size: base size the build the bottom layer of pyramid
+        """
+        if isinstance(factor, numbers.Number):
+            self.factor = round(sample_uniform(0, factor))
+        elif isinstance(factor, (tuple, list)) and len(factor) == 2:
+            self.factor = round(sample_uniform(factor[0], factor[1]))
+        else:
+            raise Exception('factor must be number or list with length 2')
+        # assert factor is valid
+        self.base_h, self.base_w = base_size[:2]
+
+    def __call__(self, img):
+        if self.factor == 0: return img
+        src_h, src_w = img.shape[:2]
+        cur_w, cur_h = self.base_w, self.base_h
+        scale_img = cv2.resize(
+            img, (cur_w, cur_h), interpolation=get_interpolation())
+        for _ in range(self.factor):
+            scale_img = cv2.pyrDown(scale_img)
+        scale_img = cv2.resize(
+            scale_img, (src_w, src_h), interpolation=get_interpolation())
+        return scale_img
+
+
+class CVGaussianNoise(object):
+    def __init__(self, mean=0, var=20):
+        self.mean = mean
+        if isinstance(var, numbers.Number):
+            self.var = max(int(sample_asym(var)), 1)
+        elif isinstance(var, (tuple, list)) and len(var) == 2:
+            self.var = int(sample_uniform(var[0], var[1]))
+        else:
+            raise Exception('degree must be number or list with length 2')
+
+    def __call__(self, img):
+        noise = np.random.normal(self.mean, self.var**0.5, img.shape)
+        img = np.clip(img + noise, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVMotionBlur(object):
+    def __init__(self, degrees=12, angle=90):
+        if isinstance(degrees, numbers.Number):
+            self.degree = max(int(sample_asym(degrees)), 1)
+        elif isinstance(degrees, (tuple, list)) and len(degrees) == 2:
+            self.degree = int(sample_uniform(degrees[0], degrees[1]))
+        else:
+            raise Exception('degree must be number or list with length 2')
+        self.angle = sample_uniform(-angle, angle)
+
+    def __call__(self, img):
+        M = cv2.getRotationMatrix2D((self.degree // 2, self.degree // 2),
+                                    self.angle, 1)
+        motion_blur_kernel = np.zeros((self.degree, self.degree))
+        motion_blur_kernel[self.degree // 2, :] = 1
+        motion_blur_kernel = cv2.warpAffine(motion_blur_kernel, M,
+                                            (self.degree, self.degree))
+        motion_blur_kernel = motion_blur_kernel / self.degree
+        img = cv2.filter2D(img, -1, motion_blur_kernel)
+        img = np.clip(img, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVGeometry(object):
+    def __init__(self,
+                 degrees=15,
+                 translate=(0.3, 0.3),
+                 scale=(0.5, 2.),
+                 shear=(45, 15),
+                 distortion=0.5,
+                 p=0.5):
+        self.p = p
+        type_p = random.random()
+        if type_p < 0.33:
+            self.transforms = CVRandomRotation(degrees=degrees)
+        elif type_p < 0.66:
+            self.transforms = CVRandomAffine(
+                degrees=degrees, translate=translate, scale=scale, shear=shear)
+        else:
+            self.transforms = CVRandomPerspective(distortion=distortion)
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            return self.transforms(img)
+        else:
+            return img
+
+
+class CVDeterioration(object):
+    def __init__(self, var, degrees, factor, p=0.5):
+        self.p = p
+        transforms = []
+        if var is not None:
+            transforms.append(CVGaussianNoise(var=var))
+        if degrees is not None:
+            transforms.append(CVMotionBlur(degrees=degrees))
+        if factor is not None:
+            transforms.append(CVRescale(factor=factor))
+
+        random.shuffle(transforms)
+        transforms = Compose(transforms)
+        self.transforms = transforms
+
+    def __call__(self, img):
+        if random.random() < self.p:
+
+            return self.transforms(img)
+        else:
+            return img
+
+
+class CVColorJitter(object):
+    def __init__(self,
+                 brightness=0.5,
+                 contrast=0.5,
+                 saturation=0.5,
+                 hue=0.1,
+                 p=0.5):
+        self.p = p
+        self.transforms = ColorJitter(
+            brightness=brightness,
+            contrast=contrast,
+            saturation=saturation,
+            hue=hue)
+
+    def __call__(self, img):
+        if random.random() < self.p: return self.transforms(img)
+        else: return img
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 39c2cbbac8..312d6dc9ad 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -157,37 +157,6 @@ class BaseRecLabelEncode(object):
         return text_list
 
 
-class NRTRLabelEncode(BaseRecLabelEncode):
-    """ Convert between text-label and text-index """
-
-    def __init__(self,
-                 max_text_length,
-                 character_dict_path=None,
-                 use_space_char=False,
-                 **kwargs):
-
-        super(NRTRLabelEncode, self).__init__(
-            max_text_length, character_dict_path, use_space_char)
-
-    def __call__(self, data):
-        text = data['label']
-        text = self.encode(text)
-        if text is None:
-            return None
-        if len(text) >= self.max_text_len - 1:
-            return None
-        data['length'] = np.array(len(text))
-        text.insert(0, 2)
-        text.append(3)
-        text = text + [0] * (self.max_text_len - len(text))
-        data['label'] = np.array(text)
-        return data
-
-    def add_special_char(self, dict_character):
-        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
-        return dict_character
-
-
 class CTCLabelEncode(BaseRecLabelEncode):
     """ Convert between text-label and text-index """
 
@@ -1046,3 +1015,99 @@ class MultiLabelEncode(BaseRecLabelEncode):
         data_out['label_sar'] = sar['label']
         data_out['length'] = ctc['length']
         return data_out
+
+
+class NRTRLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 **kwargs):
+
+        super(NRTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len - 1:
+            return None
+        data['length'] = np.array(len(text))
+        text.insert(0, 2)
+        text.append(3)
+        text = text + [0] * (self.max_text_len - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
+        return dict_character
+
+
+class ViTSTRLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 ignore_index=0,
+                 **kwargs):
+
+        super(ViTSTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data['length'] = np.array(len(text))
+        text.insert(0, self.ignore_index)
+        text.append(1)
+        text = text + [self.ignore_index] * (self.max_text_len + 2 - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['<s>', '</s>'] + dict_character
+        return dict_character
+
+
+class ABINetLabelEncode(BaseRecLabelEncode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 max_text_length,
+                 character_dict_path=None,
+                 use_space_char=False,
+                 ignore_index=100,
+                 **kwargs):
+
+        super(ABINetLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char)
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data['label']
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data['length'] = np.array(len(text))
+        text.append(0)
+        text = text + [self.ignore_index] * (self.max_text_len + 1 - len(text))
+        data['label'] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ['</s>'] + dict_character
+        return dict_character
diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
index 09736515e7..5397d71ccb 100644
--- a/ppocr/data/imaug/operators.py
+++ b/ppocr/data/imaug/operators.py
@@ -67,39 +67,6 @@ class DecodeImage(object):
         return data
 
 
-class NRTRDecodeImage(object):
-    """ decode image """
-
-    def __init__(self, img_mode='RGB', channel_first=False, **kwargs):
-        self.img_mode = img_mode
-        self.channel_first = channel_first
-
-    def __call__(self, data):
-        img = data['image']
-        if six.PY2:
-            assert type(img) is str and len(
-                img) > 0, "invalid input 'img' in DecodeImage"
-        else:
-            assert type(img) is bytes and len(
-                img) > 0, "invalid input 'img' in DecodeImage"
-        img = np.frombuffer(img, dtype='uint8')
-
-        img = cv2.imdecode(img, 1)
-
-        if img is None:
-            return None
-        if self.img_mode == 'GRAY':
-            img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
-        elif self.img_mode == 'RGB':
-            assert img.shape[2] == 3, 'invalid shape of image[%s]' % (img.shape)
-            img = img[:, :, ::-1]
-        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        if self.channel_first:
-            img = img.transpose((2, 0, 1))
-        data['image'] = img
-        return data
-
-
 class NormalizeImage(object):
     """ normalize image such as substract mean, divide std
     """
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
index 32de2b3fc3..26773d0a51 100644
--- a/ppocr/data/imaug/rec_img_aug.py
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -19,6 +19,8 @@ import random
 import copy
 from PIL import Image
 from .text_image_aug import tia_perspective, tia_stretch, tia_distort
+from .abinet_aug import CVGeometry, CVDeterioration, CVColorJitter
+from paddle.vision.transforms import Compose
 
 
 class RecAug(object):
@@ -94,6 +96,36 @@ class BaseDataAugmentation(object):
         return data
 
 
+class ABINetRecAug(object):
+    def __init__(self,
+                 geometry_p=0.5,
+                 deterioration_p=0.25,
+                 colorjitter_p=0.25,
+                 **kwargs):
+        self.transforms = Compose([
+            CVGeometry(
+                degrees=45,
+                translate=(0.0, 0.0),
+                scale=(0.5, 2.),
+                shear=(45, 15),
+                distortion=0.5,
+                p=geometry_p), CVDeterioration(
+                    var=20, degrees=6, factor=4, p=deterioration_p),
+            CVColorJitter(
+                brightness=0.5,
+                contrast=0.5,
+                saturation=0.5,
+                hue=0.1,
+                p=colorjitter_p)
+        ])
+
+    def __call__(self, data):
+        img = data['image']
+        img = self.transforms(img)
+        data['image'] = img
+        return data
+
+
 class RecConAug(object):
     def __init__(self,
                  prob=0.5,
@@ -148,46 +180,6 @@ class ClsResizeImg(object):
         return data
 
 
-class NRTRRecResizeImg(object):
-    def __init__(self, image_shape, resize_type, padding=False, **kwargs):
-        self.image_shape = image_shape
-        self.resize_type = resize_type
-        self.padding = padding
-
-    def __call__(self, data):
-        img = data['image']
-        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        image_shape = self.image_shape
-        if self.padding:
-            imgC, imgH, imgW = image_shape
-            # todo: change to 0 and modified image shape
-            h = img.shape[0]
-            w = img.shape[1]
-            ratio = w / float(h)
-            if math.ceil(imgH * ratio) > imgW:
-                resized_w = imgW
-            else:
-                resized_w = int(math.ceil(imgH * ratio))
-            resized_image = cv2.resize(img, (resized_w, imgH))
-            norm_img = np.expand_dims(resized_image, -1)
-            norm_img = norm_img.transpose((2, 0, 1))
-            resized_image = norm_img.astype(np.float32) / 128. - 1.
-            padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
-            padding_im[:, :, 0:resized_w] = resized_image
-            data['image'] = padding_im
-            return data
-        if self.resize_type == 'PIL':
-            image_pil = Image.fromarray(np.uint8(img))
-            img = image_pil.resize(self.image_shape, Image.ANTIALIAS)
-            img = np.array(img)
-        if self.resize_type == 'OpenCV':
-            img = cv2.resize(img, self.image_shape)
-        norm_img = np.expand_dims(img, -1)
-        norm_img = norm_img.transpose((2, 0, 1))
-        data['image'] = norm_img.astype(np.float32) / 128. - 1.
-        return data
-
-
 class RecResizeImg(object):
     def __init__(self,
                  image_shape,
@@ -268,6 +260,84 @@ class PRENResizeImg(object):
         return data
 
 
+class GrayRecResizeImg(object):
+    def __init__(self,
+                 image_shape,
+                 resize_type,
+                 inter_type='Image.ANTIALIAS',
+                 scale=True,
+                 padding=False,
+                 **kwargs):
+        self.image_shape = image_shape
+        self.resize_type = resize_type
+        self.padding = padding
+        self.inter_type = eval(inter_type)
+        self.scale = scale
+
+    def __call__(self, data):
+        img = data['image']
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        image_shape = self.image_shape
+        if self.padding:
+            imgC, imgH, imgW = image_shape
+            # todo: change to 0 and modified image shape
+            h = img.shape[0]
+            w = img.shape[1]
+            ratio = w / float(h)
+            if math.ceil(imgH * ratio) > imgW:
+                resized_w = imgW
+            else:
+                resized_w = int(math.ceil(imgH * ratio))
+            resized_image = cv2.resize(img, (resized_w, imgH))
+            norm_img = np.expand_dims(resized_image, -1)
+            norm_img = norm_img.transpose((2, 0, 1))
+            resized_image = norm_img.astype(np.float32) / 128. - 1.
+            padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+            padding_im[:, :, 0:resized_w] = resized_image
+            data['image'] = padding_im
+            return data
+        if self.resize_type == 'PIL':
+            image_pil = Image.fromarray(np.uint8(img))
+            img = image_pil.resize(self.image_shape, self.inter_type)
+            img = np.array(img)
+        if self.resize_type == 'OpenCV':
+            img = cv2.resize(img, self.image_shape)
+        norm_img = np.expand_dims(img, -1)
+        norm_img = norm_img.transpose((2, 0, 1))
+        if self.scale:
+            data['image'] = norm_img.astype(np.float32) / 128. - 1.
+        else:
+            data['image'] = norm_img.astype(np.float32) / 255.
+        return data
+
+
+class ABINetRecResizeImg(object):
+    def __init__(self, image_shape, **kwargs):
+        self.image_shape = image_shape
+
+    def __call__(self, data):
+        img = data['image']
+        norm_img, valid_ratio = resize_norm_img_abinet(img, self.image_shape)
+        data['image'] = norm_img
+        data['valid_ratio'] = valid_ratio
+        return data
+
+
+class SVTRRecResizeImg(object):
+    def __init__(self, image_shape, padding=True, **kwargs):
+        self.image_shape = image_shape
+        self.padding = padding
+
+    def __call__(self, data):
+        img = data['image']
+
+        norm_img, valid_ratio = resize_norm_img(img, self.image_shape,
+                                                self.padding)
+        data['image'] = norm_img
+        data['valid_ratio'] = valid_ratio
+        return data
+
+
 def resize_norm_img_sar(img, image_shape, width_downsample_ratio=0.25):
     imgC, imgH, imgW_min, imgW_max = image_shape
     h = img.shape[0]
@@ -386,6 +456,26 @@ def resize_norm_img_srn(img, image_shape):
     return np.reshape(img_black, (c, row, col)).astype(np.float32)
 
 
+def resize_norm_img_abinet(img, image_shape):
+    imgC, imgH, imgW = image_shape
+
+    resized_image = cv2.resize(
+        img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+    resized_w = imgW
+    resized_image = resized_image.astype('float32')
+    resized_image = resized_image / 255.
+
+    mean = np.array([0.485, 0.456, 0.406])
+    std = np.array([0.229, 0.224, 0.225])
+    resized_image = (
+        resized_image - mean[None, None, ...]) / std[None, None, ...]
+    resized_image = resized_image.transpose((2, 0, 1))
+    resized_image = resized_image.astype('float32')
+
+    valid_ratio = min(1.0, float(resized_w / imgW))
+    return resized_image, valid_ratio
+
+
 def srn_other_inputs(image_shape, num_heads, max_text_length):
 
     imgC, imgH, imgW = image_shape
diff --git a/ppocr/losses/__init__.py b/ppocr/losses/__init__.py
index de8419b7c1..7bea87f62f 100755
--- a/ppocr/losses/__init__.py
+++ b/ppocr/losses/__init__.py
@@ -30,7 +30,7 @@ from .det_fce_loss import FCELoss
 from .rec_ctc_loss import CTCLoss
 from .rec_att_loss import AttentionLoss
 from .rec_srn_loss import SRNLoss
-from .rec_nrtr_loss import NRTRLoss
+from .rec_ce_loss import CELoss
 from .rec_sar_loss import SARLoss
 from .rec_aster_loss import AsterLoss
 from .rec_pren_loss import PRENLoss
@@ -60,7 +60,7 @@ def build_loss(config):
     support_dict = [
         'DBLoss', 'PSELoss', 'EASTLoss', 'SASTLoss', 'FCELoss', 'CTCLoss',
         'ClsLoss', 'AttentionLoss', 'SRNLoss', 'PGLoss', 'CombinedLoss',
-        'NRTRLoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', 'SDMGRLoss',
+        'CELoss', 'TableAttentionLoss', 'SARLoss', 'AsterLoss', 'SDMGRLoss',
         'VQASerTokenLayoutLMLoss', 'LossFromOutput', 'PRENLoss', 'MultiLoss'
     ]
     config = copy.deepcopy(config)
diff --git a/ppocr/losses/rec_ce_loss.py b/ppocr/losses/rec_ce_loss.py
new file mode 100644
index 0000000000..614384de86
--- /dev/null
+++ b/ppocr/losses/rec_ce_loss.py
@@ -0,0 +1,66 @@
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+
+class CELoss(nn.Layer):
+    def __init__(self,
+                 smoothing=False,
+                 with_all=False,
+                 ignore_index=-1,
+                 **kwargs):
+        super(CELoss, self).__init__()
+        if ignore_index >= 0:
+            self.loss_func = nn.CrossEntropyLoss(
+                reduction='mean', ignore_index=ignore_index)
+        else:
+            self.loss_func = nn.CrossEntropyLoss(reduction='mean')
+        self.smoothing = smoothing
+        self.with_all = with_all
+
+    def forward(self, pred, batch):
+
+        if isinstance(pred, dict):  # for ABINet
+            loss = {}
+            loss_sum = []
+            for name, logits in pred.items():
+                if isinstance(logits, list):
+                    logit_num = len(logits)
+                    all_tgt = paddle.concat([batch[1]] * logit_num, 0)
+                    all_logits = paddle.concat(logits, 0)
+                    flt_logtis = all_logits.reshape([-1, all_logits.shape[2]])
+                    flt_tgt = all_tgt.reshape([-1])
+                else:
+                    flt_logtis = logits.reshape([-1, logits.shape[2]])
+                    flt_tgt = batch[1].reshape([-1])
+                loss[name + '_loss'] = self.loss_func(flt_logtis, flt_tgt)
+                loss_sum.append(loss[name + '_loss'])
+            loss['loss'] = sum(loss_sum)
+            return loss
+        else:
+            if self.with_all:  # for ViTSTR
+                tgt = batch[1]
+                pred = pred.reshape([-1, pred.shape[2]])
+                tgt = tgt.reshape([-1])
+                loss = self.loss_func(pred, tgt)
+                return {'loss': loss}
+            else:  # for NRTR
+                max_len = batch[2].max()
+                tgt = batch[1][:, 1:2 + max_len]
+                pred = pred.reshape([-1, pred.shape[2]])
+                tgt = tgt.reshape([-1])
+                if self.smoothing:
+                    eps = 0.1
+                    n_class = pred.shape[1]
+                    one_hot = F.one_hot(tgt, pred.shape[1])
+                    one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (
+                        n_class - 1)
+                    log_prb = F.log_softmax(pred, axis=1)
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(
+                            tgt.shape, dtype=tgt.dtype))
+                    loss = -(one_hot * log_prb).sum(axis=1)
+                    loss = loss.masked_select(non_pad_mask).mean()
+                else:
+                    loss = self.loss_func(pred, tgt)
+                return {'loss': loss}
diff --git a/ppocr/losses/rec_nrtr_loss.py b/ppocr/losses/rec_nrtr_loss.py
deleted file mode 100644
index 200a6d0486..0000000000
--- a/ppocr/losses/rec_nrtr_loss.py
+++ /dev/null
@@ -1,30 +0,0 @@
-import paddle
-from paddle import nn
-import paddle.nn.functional as F
-
-
-class NRTRLoss(nn.Layer):
-    def __init__(self, smoothing=True, **kwargs):
-        super(NRTRLoss, self).__init__()
-        self.loss_func = nn.CrossEntropyLoss(reduction='mean', ignore_index=0)
-        self.smoothing = smoothing
-
-    def forward(self, pred, batch):
-        pred = pred.reshape([-1, pred.shape[2]])
-        max_len = batch[2].max()
-        tgt = batch[1][:, 1:2 + max_len]
-        tgt = tgt.reshape([-1])
-        if self.smoothing:
-            eps = 0.1
-            n_class = pred.shape[1]
-            one_hot = F.one_hot(tgt, pred.shape[1])
-            one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
-            log_prb = F.log_softmax(pred, axis=1)
-            non_pad_mask = paddle.not_equal(
-                tgt, paddle.zeros(
-                    tgt.shape, dtype=tgt.dtype))
-            loss = -(one_hot * log_prb).sum(axis=1)
-            loss = loss.masked_select(non_pad_mask).mean()
-        else:
-            loss = self.loss_func(pred, tgt)
-        return {'loss': loss}
diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
index 072d6e0f84..f8959e263e 100755
--- a/ppocr/modeling/backbones/__init__.py
+++ b/ppocr/modeling/backbones/__init__.py
@@ -28,35 +28,37 @@ def build_backbone(config, model_type):
         from .rec_mv1_enhance import MobileNetV1Enhance
         from .rec_nrtr_mtb import MTB
         from .rec_resnet_31 import ResNet31
+        from .rec_resnet_45 import ResNet45
         from .rec_resnet_aster import ResNet_ASTER
         from .rec_micronet import MicroNet
         from .rec_efficientb3_pren import EfficientNetb3_PREN
         from .rec_svtrnet import SVTRNet
+        from .rec_vitstr import ViTSTR
         support_dict = [
             'MobileNetV1Enhance', 'MobileNetV3', 'ResNet', 'ResNetFPN', 'MTB',
-            "ResNet31", "ResNet_ASTER", 'MicroNet', 'EfficientNetb3_PREN',
-            'SVTRNet'
+            'ResNet31', 'ResNet45', 'ResNet_ASTER', 'MicroNet',
+            'EfficientNetb3_PREN', 'SVTRNet', 'ViTSTR'
         ]
-    elif model_type == "e2e":
+    elif model_type == 'e2e':
         from .e2e_resnet_vd_pg import ResNet
         support_dict = ['ResNet']
     elif model_type == 'kie':
         from .kie_unet_sdmgr import Kie_backbone
         support_dict = ['Kie_backbone']
-    elif model_type == "table":
+    elif model_type == 'table':
         from .table_resnet_vd import ResNet
         from .table_mobilenet_v3 import MobileNetV3
-        support_dict = ["ResNet", "MobileNetV3"]
+        support_dict = ['ResNet', 'MobileNetV3']
     elif model_type == 'vqa':
         from .vqa_layoutlm import LayoutLMForSer, LayoutLMv2ForSer, LayoutLMv2ForRe, LayoutXLMForSer, LayoutXLMForRe
         support_dict = [
-            "LayoutLMForSer", "LayoutLMv2ForSer", 'LayoutLMv2ForRe',
-            "LayoutXLMForSer", 'LayoutXLMForRe'
+            'LayoutLMForSer', 'LayoutLMv2ForSer', 'LayoutLMv2ForRe',
+            'LayoutXLMForSer', 'LayoutXLMForRe'
         ]
     else:
         raise NotImplementedError
 
-    module_name = config.pop("name")
+    module_name = config.pop('name')
     assert module_name in support_dict, Exception(
         "when model typs is {}, backbone only support {}".format(model_type,
                                                                  support_dict))
diff --git a/ppocr/modeling/backbones/rec_resnet_45.py b/ppocr/modeling/backbones/rec_resnet_45.py
new file mode 100644
index 0000000000..9093d0bc99
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_45.py
@@ -0,0 +1,147 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/FangShancheng/ABINet/tree/main/modules
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+import math
+
+__all__ = ["ResNet45"]
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    return nn.Conv2D(
+        in_planes,
+        out_planes,
+        kernel_size=1,
+        stride=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False)
+
+
+def conv3x3(in_channel, out_channel, stride=1):
+    return nn.Conv2D(
+        in_channel,
+        out_channel,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False)
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self, in_channels, channels, stride=1, downsample=None):
+        super().__init__()
+        self.conv1 = conv1x1(in_channels, channels)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(channels, channels, stride)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet45(nn.Layer):
+    def __init__(self, block=BasicBlock, layers=[3, 4, 6, 6, 3], in_channels=3):
+        self.inplanes = 32
+        super(ResNet45, self).__init__()
+        self.conv1 = nn.Conv2D(
+            3,
+            32,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False)
+        self.bn1 = nn.BatchNorm2D(32)
+        self.relu = nn.ReLU()
+
+        self.layer1 = self._make_layer(block, 32, layers[0], stride=2)
+        self.layer2 = self._make_layer(block, 64, layers[1], stride=1)
+        self.layer3 = self._make_layer(block, 128, layers[2], stride=2)
+        self.layer4 = self._make_layer(block, 256, layers[3], stride=1)
+        self.layer5 = self._make_layer(block, 512, layers[4], stride=1)
+        self.out_channels = 512
+
+        # for m in self.modules():
+        #     if isinstance(m, nn.Conv2D):
+        #         n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+        #         m.weight.data.normal_(0, math.sqrt(2. / n))
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            # downsample = True
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    weight_attr=ParamAttr(initializer=KaimingNormal()),
+                    bias_attr=False),
+                nn.BatchNorm2D(planes * block.expansion), )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        # print(x)
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        # print(x)
+        x = self.layer4(x)
+        x = self.layer5(x)
+        return x
diff --git a/ppocr/modeling/backbones/rec_svtrnet.py b/ppocr/modeling/backbones/rec_svtrnet.py
index c57bf46345..c2c07f4476 100644
--- a/ppocr/modeling/backbones/rec_svtrnet.py
+++ b/ppocr/modeling/backbones/rec_svtrnet.py
@@ -147,7 +147,7 @@ class Attention(nn.Layer):
                  dim,
                  num_heads=8,
                  mixer='Global',
-                 HW=[8, 25],
+                 HW=None,
                  local_k=[7, 11],
                  qkv_bias=False,
                  qk_scale=None,
@@ -210,7 +210,7 @@ class Block(nn.Layer):
                  num_heads,
                  mixer='Global',
                  local_mixer=[7, 11],
-                 HW=[8, 25],
+                 HW=None,
                  mlp_ratio=4.,
                  qkv_bias=False,
                  qk_scale=None,
@@ -274,7 +274,9 @@ class PatchEmbed(nn.Layer):
                  img_size=[32, 100],
                  in_channels=3,
                  embed_dim=768,
-                 sub_num=2):
+                 sub_num=2,
+                 patch_size=[4, 4],
+                 mode='pope'):
         super().__init__()
         num_patches = (img_size[1] // (2 ** sub_num)) * \
                       (img_size[0] // (2 ** sub_num))
@@ -282,50 +284,56 @@ class PatchEmbed(nn.Layer):
         self.num_patches = num_patches
         self.embed_dim = embed_dim
         self.norm = None
-        if sub_num == 2:
-            self.proj = nn.Sequential(
-                ConvBNLayer(
-                    in_channels=in_channels,
-                    out_channels=embed_dim // 2,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None),
-                ConvBNLayer(
-                    in_channels=embed_dim // 2,
-                    out_channels=embed_dim,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None))
-        if sub_num == 3:
-            self.proj = nn.Sequential(
-                ConvBNLayer(
-                    in_channels=in_channels,
-                    out_channels=embed_dim // 4,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None),
-                ConvBNLayer(
-                    in_channels=embed_dim // 4,
-                    out_channels=embed_dim // 2,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None),
-                ConvBNLayer(
-                    in_channels=embed_dim // 2,
-                    out_channels=embed_dim,
-                    kernel_size=3,
-                    stride=2,
-                    padding=1,
-                    act=nn.GELU,
-                    bias_attr=None))
+        if mode == 'pope':
+            if sub_num == 2:
+                self.proj = nn.Sequential(
+                    ConvBNLayer(
+                        in_channels=in_channels,
+                        out_channels=embed_dim // 2,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 2,
+                        out_channels=embed_dim,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None))
+            if sub_num == 3:
+                self.proj = nn.Sequential(
+                    ConvBNLayer(
+                        in_channels=in_channels,
+                        out_channels=embed_dim // 4,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 4,
+                        out_channels=embed_dim // 2,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 2,
+                        out_channels=embed_dim,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None))
+        elif mode == 'linear':
+            self.proj = nn.Conv2D(
+                1, embed_dim, kernel_size=patch_size, stride=patch_size)
+            self.num_patches = img_size[0] // patch_size[0] * img_size[
+                1] // patch_size[1]
 
     def forward(self, x):
         B, C, H, W = x.shape
diff --git a/ppocr/modeling/backbones/rec_vitstr.py b/ppocr/modeling/backbones/rec_vitstr.py
new file mode 100644
index 0000000000..d5d7d5148a
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_vitstr.py
@@ -0,0 +1,120 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/roatienza/deep-text-recognition-benchmark/blob/master/modules/vitstr.py
+"""
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+from ppocr.modeling.backbones.rec_svtrnet import Block, PatchEmbed, zeros_, trunc_normal_, ones_
+
+scale_dim_heads = {'tiny': [192, 3], 'small': [384, 6], 'base': [768, 12]}
+
+
+class ViTSTR(nn.Layer):
+    def __init__(self,
+                 img_size=[224, 224],
+                 in_channels=1,
+                 scale='tiny',
+                 seqlen=27,
+                 patch_size=[16, 16],
+                 embed_dim=None,
+                 depth=12,
+                 num_heads=None,
+                 mlp_ratio=4,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 drop_path_rate=0.,
+                 drop_rate=0.,
+                 attn_drop_rate=0.,
+                 norm_layer='nn.LayerNorm',
+                 act_layer='nn.GELU',
+                 epsilon=1e-6,
+                 out_channels=None,
+                 **kwargs):
+        super().__init__()
+        self.seqlen = seqlen
+        embed_dim = embed_dim if embed_dim is not None else scale_dim_heads[
+            scale][0]
+        num_heads = num_heads if num_heads is not None else scale_dim_heads[
+            scale][1]
+        out_channels = out_channels if out_channels is not None else embed_dim
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            in_channels=in_channels,
+            embed_dim=embed_dim,
+            patch_size=patch_size,
+            mode='linear')
+        num_patches = self.patch_embed.num_patches
+
+        self.pos_embed = self.create_parameter(
+            shape=[1, num_patches + 1, embed_dim], default_initializer=zeros_)
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.cls_token = self.create_parameter(
+            shape=[1, 1, embed_dim], default_initializer=zeros_)
+        self.add_parameter("cls_token", self.cls_token)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = np.linspace(0, drop_path_rate, depth)
+        self.blocks = nn.LayerList([
+            Block(
+                dim=embed_dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[i],
+                norm_layer=norm_layer,
+                act_layer=eval(act_layer),
+                epsilon=epsilon,
+                prenorm=False) for i in range(depth)
+        ])
+        self.norm = eval(norm_layer)(embed_dim, epsilon=epsilon)
+
+        self.out_channels = out_channels
+
+        trunc_normal_(self.pos_embed)
+        trunc_normal_(self.cls_token)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward_features(self, x):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+        cls_tokens = paddle.tile(self.cls_token, repeat_times=[B, 1, 1])
+        x = paddle.concat((cls_tokens, x), axis=1)
+        x = x + self.pos_embed
+        x = self.pos_drop(x)
+        for blk in self.blocks:
+            x = blk(x)
+        x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        x = x[:, :self.seqlen]
+        return x.transpose([0, 2, 1]).unsqueeze(2)
diff --git a/ppocr/modeling/heads/__init__.py b/ppocr/modeling/heads/__init__.py
index 1670ea38e6..14e6aab854 100755
--- a/ppocr/modeling/heads/__init__.py
+++ b/ppocr/modeling/heads/__init__.py
@@ -33,6 +33,7 @@ def build_head(config):
     from .rec_aster_head import AsterHead
     from .rec_pren_head import PRENHead
     from .rec_multi_head import MultiHead
+    from .rec_abinet_head import ABINetHead
 
     # cls head
     from .cls_head import ClsHead
@@ -46,7 +47,7 @@ def build_head(config):
         'DBHead', 'PSEHead', 'FCEHead', 'EASTHead', 'SASTHead', 'CTCHead',
         'ClsHead', 'AttentionHead', 'SRNHead', 'PGHead', 'Transformer',
         'TableAttentionHead', 'SARHead', 'AsterHead', 'SDMGRHead', 'PRENHead',
-        'MultiHead'
+        'MultiHead', 'ABINetHead'
     ]
 
     #table head
diff --git a/ppocr/modeling/heads/multiheadAttention.py b/ppocr/modeling/heads/multiheadAttention.py
deleted file mode 100755
index 900865ba1a..0000000000
--- a/ppocr/modeling/heads/multiheadAttention.py
+++ /dev/null
@@ -1,163 +0,0 @@
-# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#    http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import paddle
-from paddle import nn
-import paddle.nn.functional as F
-from paddle.nn import Linear
-from paddle.nn.initializer import XavierUniform as xavier_uniform_
-from paddle.nn.initializer import Constant as constant_
-from paddle.nn.initializer import XavierNormal as xavier_normal_
-
-zeros_ = constant_(value=0.)
-ones_ = constant_(value=1.)
-
-
-class MultiheadAttention(nn.Layer):
-    """Allows the model to jointly attend to information
-    from different representation subspaces.
-    See reference: Attention Is All You Need
-
-    .. math::
-        \text{MultiHead}(Q, K, V) = \text{Concat}(head_1,\dots,head_h)W^O
-        \text{where} head_i = \text{Attention}(QW_i^Q, KW_i^K, VW_i^V)
-
-    Args:
-        embed_dim: total dimension of the model
-        num_heads: parallel attention layers, or heads
-
-    """
-
-    def __init__(self,
-                 embed_dim,
-                 num_heads,
-                 dropout=0.,
-                 bias=True,
-                 add_bias_kv=False,
-                 add_zero_attn=False):
-        super(MultiheadAttention, self).__init__()
-        self.embed_dim = embed_dim
-        self.num_heads = num_heads
-        self.dropout = dropout
-        self.head_dim = embed_dim // num_heads
-        assert self.head_dim * num_heads == self.embed_dim, "embed_dim must be divisible by num_heads"
-        self.scaling = self.head_dim**-0.5
-        self.out_proj = Linear(embed_dim, embed_dim, bias_attr=bias)
-        self._reset_parameters()
-        self.conv1 = paddle.nn.Conv2D(
-            in_channels=embed_dim, out_channels=embed_dim, kernel_size=(1, 1))
-        self.conv2 = paddle.nn.Conv2D(
-            in_channels=embed_dim, out_channels=embed_dim, kernel_size=(1, 1))
-        self.conv3 = paddle.nn.Conv2D(
-            in_channels=embed_dim, out_channels=embed_dim, kernel_size=(1, 1))
-
-    def _reset_parameters(self):
-        xavier_uniform_(self.out_proj.weight)
-
-    def forward(self,
-                query,
-                key,
-                value,
-                key_padding_mask=None,
-                incremental_state=None,
-                attn_mask=None):
-        """
-        Inputs of forward function
-            query: [target length, batch size, embed dim]
-            key: [sequence length, batch size, embed dim]
-            value: [sequence length, batch size, embed dim]
-            key_padding_mask: if True, mask padding based on batch size
-            incremental_state: if provided, previous time steps are cashed
-            need_weights: output attn_output_weights
-            static_kv: key and value are static
-
-        Outputs of forward function
-            attn_output: [target length, batch size, embed dim]
-            attn_output_weights: [batch size, target length, sequence length]
-        """
-        q_shape = paddle.shape(query)
-        src_shape = paddle.shape(key)
-        q = self._in_proj_q(query)
-        k = self._in_proj_k(key)
-        v = self._in_proj_v(value)
-        q *= self.scaling
-        q = paddle.transpose(
-            paddle.reshape(
-                q, [q_shape[0], q_shape[1], self.num_heads, self.head_dim]),
-            [1, 2, 0, 3])
-        k = paddle.transpose(
-            paddle.reshape(
-                k, [src_shape[0], q_shape[1], self.num_heads, self.head_dim]),
-            [1, 2, 0, 3])
-        v = paddle.transpose(
-            paddle.reshape(
-                v, [src_shape[0], q_shape[1], self.num_heads, self.head_dim]),
-            [1, 2, 0, 3])
-        if key_padding_mask is not None:
-            assert key_padding_mask.shape[0] == q_shape[1]
-            assert key_padding_mask.shape[1] == src_shape[0]
-        attn_output_weights = paddle.matmul(q,
-                                            paddle.transpose(k, [0, 1, 3, 2]))
-        if attn_mask is not None:
-            attn_mask = paddle.unsqueeze(paddle.unsqueeze(attn_mask, 0), 0)
-            attn_output_weights += attn_mask
-        if key_padding_mask is not None:
-            attn_output_weights = paddle.reshape(
-                attn_output_weights,
-                [q_shape[1], self.num_heads, q_shape[0], src_shape[0]])
-            key = paddle.unsqueeze(paddle.unsqueeze(key_padding_mask, 1), 2)
-            key = paddle.cast(key, 'float32')
-            y = paddle.full(
-                shape=paddle.shape(key), dtype='float32', fill_value='-inf')
-            y = paddle.where(key == 0., key, y)
-            attn_output_weights += y
-        attn_output_weights = F.softmax(
-            attn_output_weights.astype('float32'),
-            axis=-1,
-            dtype=paddle.float32 if attn_output_weights.dtype == paddle.float16
-            else attn_output_weights.dtype)
-        attn_output_weights = F.dropout(
-            attn_output_weights, p=self.dropout, training=self.training)
-
-        attn_output = paddle.matmul(attn_output_weights, v)
-        attn_output = paddle.reshape(
-            paddle.transpose(attn_output, [2, 0, 1, 3]),
-            [q_shape[0], q_shape[1], self.embed_dim])
-        attn_output = self.out_proj(attn_output)
-
-        return attn_output
-
-    def _in_proj_q(self, query):
-        query = paddle.transpose(query, [1, 2, 0])
-        query = paddle.unsqueeze(query, axis=2)
-        res = self.conv1(query)
-        res = paddle.squeeze(res, axis=2)
-        res = paddle.transpose(res, [2, 0, 1])
-        return res
-
-    def _in_proj_k(self, key):
-        key = paddle.transpose(key, [1, 2, 0])
-        key = paddle.unsqueeze(key, axis=2)
-        res = self.conv2(key)
-        res = paddle.squeeze(res, axis=2)
-        res = paddle.transpose(res, [2, 0, 1])
-        return res
-
-    def _in_proj_v(self, value):
-        value = paddle.transpose(value, [1, 2, 0])  #(1, 2, 0)
-        value = paddle.unsqueeze(value, axis=2)
-        res = self.conv3(value)
-        res = paddle.squeeze(res, axis=2)
-        res = paddle.transpose(res, [2, 0, 1])
-        return res
diff --git a/ppocr/modeling/heads/rec_abinet_head.py b/ppocr/modeling/heads/rec_abinet_head.py
new file mode 100644
index 0000000000..a0f60f1be1
--- /dev/null
+++ b/ppocr/modeling/heads/rec_abinet_head.py
@@ -0,0 +1,296 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: 
+https://github.com/FangShancheng/ABINet/tree/main/modules
+"""
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle.nn import LayerList
+from ppocr.modeling.heads.rec_nrtr_head import TransformerBlock, PositionalEncoding
+
+
+class BCNLanguage(nn.Layer):
+    def __init__(self,
+                 d_model=512,
+                 nhead=8,
+                 num_layers=4,
+                 dim_feedforward=2048,
+                 dropout=0.,
+                 max_length=25,
+                 detach=True,
+                 num_classes=37):
+        super().__init__()
+
+        self.d_model = d_model
+        self.detach = detach
+        self.max_length = max_length + 1  # additional stop token
+        self.proj = nn.Linear(num_classes, d_model, bias_attr=False)
+        self.token_encoder = PositionalEncoding(
+            dropout=0.1, dim=d_model, max_len=self.max_length)
+        self.pos_encoder = PositionalEncoding(
+            dropout=0, dim=d_model, max_len=self.max_length)
+
+        self.decoder = nn.LayerList([
+            TransformerBlock(
+                d_model=d_model,
+                nhead=nhead,
+                dim_feedforward=dim_feedforward,
+                attention_dropout_rate=dropout,
+                residual_dropout_rate=dropout,
+                with_self_attn=False,
+                with_cross_attn=True) for i in range(num_layers)
+        ])
+
+        self.cls = nn.Linear(d_model, num_classes)
+
+    def forward(self, tokens, lengths):
+        """
+        Args:
+            tokens: (B, N, C) where N is length, B is batch size and C is classes number
+            lengths: (B,)
+        """
+        if self.detach: tokens = tokens.detach()
+        embed = self.proj(tokens)  # (B, N, C)
+        embed = self.token_encoder(embed)  # (B, N, C)
+        padding_mask = _get_mask(lengths, self.max_length)
+        zeros = paddle.zeros_like(embed)  # (B, N, C)
+        qeury = self.pos_encoder(zeros)
+        for decoder_layer in self.decoder:
+            qeury = decoder_layer(qeury, embed, cross_mask=padding_mask)
+        output = qeury  # (B, N, C)
+
+        logits = self.cls(output)  # (B, N, C)
+
+        return output, logits
+
+
+def encoder_layer(in_c, out_c, k=3, s=2, p=1):
+    return nn.Sequential(
+        nn.Conv2D(in_c, out_c, k, s, p), nn.BatchNorm2D(out_c), nn.ReLU())
+
+
+def decoder_layer(in_c,
+                  out_c,
+                  k=3,
+                  s=1,
+                  p=1,
+                  mode='nearest',
+                  scale_factor=None,
+                  size=None):
+    align_corners = False if mode == 'nearest' else True
+    return nn.Sequential(
+        nn.Upsample(
+            size=size,
+            scale_factor=scale_factor,
+            mode=mode,
+            align_corners=align_corners),
+        nn.Conv2D(in_c, out_c, k, s, p),
+        nn.BatchNorm2D(out_c),
+        nn.ReLU())
+
+
+class PositionAttention(nn.Layer):
+    def __init__(self,
+                 max_length,
+                 in_channels=512,
+                 num_channels=64,
+                 h=8,
+                 w=32,
+                 mode='nearest',
+                 **kwargs):
+        super().__init__()
+        self.max_length = max_length
+        self.k_encoder = nn.Sequential(
+            encoder_layer(
+                in_channels, num_channels, s=(1, 2)),
+            encoder_layer(
+                num_channels, num_channels, s=(2, 2)),
+            encoder_layer(
+                num_channels, num_channels, s=(2, 2)),
+            encoder_layer(
+                num_channels, num_channels, s=(2, 2)))
+        self.k_decoder = nn.Sequential(
+            decoder_layer(
+                num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(
+                num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(
+                num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(
+                num_channels, in_channels, size=(h, w), mode=mode))
+
+        self.pos_encoder = PositionalEncoding(
+            dropout=0, dim=in_channels, max_len=max_length)
+        self.project = nn.Linear(in_channels, in_channels)
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        k, v = x, x
+
+        # calculate key vector
+        features = []
+        for i in range(0, len(self.k_encoder)):
+            k = self.k_encoder[i](k)
+            features.append(k)
+        for i in range(0, len(self.k_decoder) - 1):
+            k = self.k_decoder[i](k)
+            # print(k.shape, features[len(self.k_decoder) - 2 - i].shape)
+            k = k + features[len(self.k_decoder) - 2 - i]
+        k = self.k_decoder[-1](k)
+
+        # calculate query vector
+        # TODO q=f(q,k)
+        zeros = paddle.zeros(
+            (B, self.max_length, C), dtype=x.dtype)  # (T, N, C)
+        q = self.pos_encoder(zeros)  # (B, N, C)
+        q = self.project(q)  # (B, N, C)
+
+        # calculate attention
+        attn_scores = q @k.flatten(2)  # (B, N, (H*W))
+        attn_scores = attn_scores / (C**0.5)
+        attn_scores = F.softmax(attn_scores, axis=-1)
+
+        v = v.flatten(2).transpose([0, 2, 1])  # (B, (H*W), C)
+        attn_vecs = attn_scores @v  # (B, N, C)
+
+        return attn_vecs, attn_scores.reshape([0, self.max_length, H, W])
+
+
+class ABINetHead(nn.Layer):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 d_model=512,
+                 nhead=8,
+                 num_layers=3,
+                 dim_feedforward=2048,
+                 dropout=0.1,
+                 max_length=25,
+                 use_lang=False,
+                 iter_size=1):
+        super().__init__()
+        self.max_length = max_length + 1
+        self.pos_encoder = PositionalEncoding(
+            dropout=0.1, dim=d_model, max_len=8 * 32)
+        self.encoder = nn.LayerList([
+            TransformerBlock(
+                d_model=d_model,
+                nhead=nhead,
+                dim_feedforward=dim_feedforward,
+                attention_dropout_rate=dropout,
+                residual_dropout_rate=dropout,
+                with_self_attn=True,
+                with_cross_attn=False) for i in range(num_layers)
+        ])
+        self.decoder = PositionAttention(
+            max_length=max_length + 1,  # additional stop token
+            mode='nearest', )
+        self.out_channels = out_channels
+        self.cls = nn.Linear(d_model, self.out_channels)
+        self.use_lang = use_lang
+        if use_lang:
+            self.iter_size = iter_size
+            self.language = BCNLanguage(
+                d_model=d_model,
+                nhead=nhead,
+                num_layers=4,
+                dim_feedforward=dim_feedforward,
+                dropout=dropout,
+                max_length=max_length,
+                num_classes=self.out_channels)
+            # alignment
+            self.w_att_align = nn.Linear(2 * d_model, d_model)
+            self.cls_align = nn.Linear(d_model, self.out_channels)
+
+    def forward(self, x, targets=None):
+        x = x.transpose([0, 2, 3, 1])
+        _, H, W, C = x.shape
+        feature = x.flatten(1, 2)
+        feature = self.pos_encoder(feature)
+        for encoder_layer in self.encoder:
+            feature = encoder_layer(feature)
+        feature = feature.reshape([0, H, W, C]).transpose([0, 3, 1, 2])
+        v_feature, attn_scores = self.decoder(
+            feature)  # (B, N, C), (B, C, H, W)
+        vis_logits = self.cls(v_feature)  # (B, N, C)
+        logits = vis_logits
+        vis_lengths = _get_length(vis_logits)
+        if self.use_lang:
+            align_logits = vis_logits
+            align_lengths = vis_lengths
+            all_l_res, all_a_res = [], []
+            for i in range(self.iter_size):
+                tokens = F.softmax(align_logits, axis=-1)
+                lengths = align_lengths
+                lengths = paddle.clip(
+                    lengths, 2, self.max_length)  # TODO:move to langauge model
+                l_feature, l_logits = self.language(tokens, lengths)
+
+                # alignment
+                all_l_res.append(l_logits)
+                fuse = paddle.concat((l_feature, v_feature), -1)
+                f_att = F.sigmoid(self.w_att_align(fuse))
+                output = f_att * v_feature + (1 - f_att) * l_feature
+                align_logits = self.cls_align(output)  # (B, N, C)
+
+                align_lengths = _get_length(align_logits)
+                all_a_res.append(align_logits)
+            if self.training:
+                return {
+                    'align': all_a_res,
+                    'lang': all_l_res,
+                    'vision': vis_logits
+                }
+            else:
+                logits = align_logits
+        if self.training:
+            return logits
+        else:
+            return F.softmax(logits, -1)
+
+
+def _get_length(logit):
+    """ Greed decoder to obtain length from logit"""
+    out = (logit.argmax(-1) == 0)
+    abn = out.any(-1)
+    out_int = out.cast('int32')
+    out = (out_int.cumsum(-1) == 1) & out
+    out = out.cast('int32')
+    out = out.argmax(-1)
+    out = out + 1
+    out = paddle.where(abn, out, paddle.to_tensor(logit.shape[1]))
+    return out
+
+
+def _get_mask(length, max_length):
+    """Generate a square mask for the sequence. The masked positions are filled with float('-inf').
+        Unmasked positions are filled with float(0.0).
+    """
+    length = length.unsqueeze(-1)
+    B = paddle.shape(length)[0]
+    grid = paddle.arange(0, max_length).unsqueeze(0).tile([B, 1])
+    zero_mask = paddle.zeros([B, max_length], dtype='float32')
+    inf_mask = paddle.full([B, max_length], '-inf', dtype='float32')
+    diag_mask = paddle.diag(
+        paddle.full(
+            [max_length], '-inf', dtype=paddle.float32),
+        offset=0,
+        name=None)
+    mask = paddle.where(grid >= length, inf_mask, zero_mask)
+    mask = mask.unsqueeze(1) + diag_mask
+    return mask.unsqueeze(1)
diff --git a/ppocr/modeling/heads/rec_nrtr_head.py b/ppocr/modeling/heads/rec_nrtr_head.py
index 38ba0c9178..bf9ef56145 100644
--- a/ppocr/modeling/heads/rec_nrtr_head.py
+++ b/ppocr/modeling/heads/rec_nrtr_head.py
@@ -14,20 +14,15 @@
 
 import math
 import paddle
-import copy
 from paddle import nn
 import paddle.nn.functional as F
 from paddle.nn import LayerList
-from paddle.nn.initializer import XavierNormal as xavier_uniform_
-from paddle.nn import Dropout, Linear, LayerNorm, Conv2D
+# from paddle.nn.initializer import XavierNormal as xavier_uniform_
+from paddle.nn import Dropout, Linear, LayerNorm
 import numpy as np
-from ppocr.modeling.heads.multiheadAttention import MultiheadAttention
-from paddle.nn.initializer import Constant as constant_
+from ppocr.modeling.backbones.rec_svtrnet import Mlp, zeros_, ones_
 from paddle.nn.initializer import XavierNormal as xavier_normal_
 
-zeros_ = constant_(value=0.)
-ones_ = constant_(value=1.)
-
 
 class Transformer(nn.Layer):
     """A transformer model. User is able to modify the attributes as needed. The architechture
@@ -45,7 +40,6 @@ class Transformer(nn.Layer):
         dropout: the dropout value (default=0.1).
         custom_encoder: custom encoder (default=None).
         custom_decoder: custom decoder (default=None).
-
     """
 
     def __init__(self,
@@ -54,45 +48,49 @@ class Transformer(nn.Layer):
                  num_encoder_layers=6,
                  beam_size=0,
                  num_decoder_layers=6,
+                 max_len=25,
                  dim_feedforward=1024,
                  attention_dropout_rate=0.0,
                  residual_dropout_rate=0.1,
-                 custom_encoder=None,
-                 custom_decoder=None,
                  in_channels=0,
                  out_channels=0,
                  scale_embedding=True):
         super(Transformer, self).__init__()
         self.out_channels = out_channels + 1
+        self.max_len = max_len
         self.embedding = Embeddings(
             d_model=d_model,
             vocab=self.out_channels,
             padding_idx=0,
             scale_embedding=scale_embedding)
         self.positional_encoding = PositionalEncoding(
-            dropout=residual_dropout_rate,
-            dim=d_model, )
-        if custom_encoder is not None:
-            self.encoder = custom_encoder
-        else:
-            if num_encoder_layers > 0:
-                encoder_layer = TransformerEncoderLayer(
-                    d_model, nhead, dim_feedforward, attention_dropout_rate,
-                    residual_dropout_rate)
-                self.encoder = TransformerEncoder(encoder_layer,
-                                                  num_encoder_layers)
-            else:
-                self.encoder = None
+            dropout=residual_dropout_rate, dim=d_model)
 
-        if custom_decoder is not None:
-            self.decoder = custom_decoder
+        if num_encoder_layers > 0:
+            self.encoder = nn.LayerList([
+                TransformerBlock(
+                    d_model,
+                    nhead,
+                    dim_feedforward,
+                    attention_dropout_rate,
+                    residual_dropout_rate,
+                    with_self_attn=True,
+                    with_cross_attn=False) for i in range(num_encoder_layers)
+            ])
         else:
-            decoder_layer = TransformerDecoderLayer(
-                d_model, nhead, dim_feedforward, attention_dropout_rate,
-                residual_dropout_rate)
-            self.decoder = TransformerDecoder(decoder_layer, num_decoder_layers)
+            self.encoder = None
+
+        self.decoder = nn.LayerList([
+            TransformerBlock(
+                d_model,
+                nhead,
+                dim_feedforward,
+                attention_dropout_rate,
+                residual_dropout_rate,
+                with_self_attn=True,
+                with_cross_attn=True) for i in range(num_decoder_layers)
+        ])
 
-        self._reset_parameters()
         self.beam_size = beam_size
         self.d_model = d_model
         self.nhead = nhead
@@ -105,7 +103,7 @@ class Transformer(nn.Layer):
 
     def _init_weights(self, m):
 
-        if isinstance(m, nn.Conv2D):
+        if isinstance(m, nn.Linear):
             xavier_normal_(m.weight)
             if m.bias is not None:
                 zeros_(m.bias)
@@ -113,24 +111,20 @@ class Transformer(nn.Layer):
     def forward_train(self, src, tgt):
         tgt = tgt[:, :-1]
 
-        tgt_key_padding_mask = self.generate_padding_mask(tgt)
-        tgt = self.embedding(tgt).transpose([1, 0, 2])
+        tgt = self.embedding(tgt)
         tgt = self.positional_encoding(tgt)
-        tgt_mask = self.generate_square_subsequent_mask(tgt.shape[0])
+        tgt_mask = self.generate_square_subsequent_mask(tgt.shape[1])
 
         if self.encoder is not None:
-            src = self.positional_encoding(src.transpose([1, 0, 2]))
-            memory = self.encoder(src)
+            src = self.positional_encoding(src)
+            for encoder_layer in self.encoder:
+                src = encoder_layer(src)
+            memory = src  # B N C
         else:
-            memory = src.squeeze(2).transpose([2, 0, 1])
-        output = self.decoder(
-            tgt,
-            memory,
-            tgt_mask=tgt_mask,
-            memory_mask=None,
-            tgt_key_padding_mask=tgt_key_padding_mask,
-            memory_key_padding_mask=None)
-        output = output.transpose([1, 0, 2])
+            memory = src  # B N C
+        for decoder_layer in self.decoder:
+            tgt = decoder_layer(tgt, memory, self_mask=tgt_mask)
+        output = tgt
         logit = self.tgt_word_prj(output)
         return logit
 
@@ -140,8 +134,8 @@ class Transformer(nn.Layer):
             src: the sequence to the encoder (required).
             tgt: the sequence to the decoder (required).
         Shape:
-            - src: :math:`(S, N, E)`.
-            - tgt: :math:`(T, N, E)`.
+            - src: :math:`(B, sN, C)`.
+            - tgt: :math:`(B, tN, C)`.
         Examples:
             >>> output = transformer_model(src, tgt)
         """
@@ -157,36 +151,35 @@ class Transformer(nn.Layer):
                 return self.forward_test(src)
 
     def forward_test(self, src):
+
         bs = paddle.shape(src)[0]
         if self.encoder is not None:
-            src = self.positional_encoding(paddle.transpose(src, [1, 0, 2]))
-            memory = self.encoder(src)
+            src = self.positional_encoding(src)
+            for encoder_layer in self.encoder:
+                src = encoder_layer(src)
+            memory = src  # B N C
         else:
-            memory = paddle.transpose(paddle.squeeze(src, 2), [2, 0, 1])
+            memory = src
         dec_seq = paddle.full((bs, 1), 2, dtype=paddle.int64)
         dec_prob = paddle.full((bs, 1), 1., dtype=paddle.float32)
-        for len_dec_seq in range(1, 25):
-            dec_seq_embed = paddle.transpose(self.embedding(dec_seq), [1, 0, 2])
+        for len_dec_seq in range(1, self.max_len):
+            dec_seq_embed = self.embedding(dec_seq)
             dec_seq_embed = self.positional_encoding(dec_seq_embed)
             tgt_mask = self.generate_square_subsequent_mask(
-                paddle.shape(dec_seq_embed)[0])
-            output = self.decoder(
-                dec_seq_embed,
-                memory,
-                tgt_mask=tgt_mask,
-                memory_mask=None,
-                tgt_key_padding_mask=None,
-                memory_key_padding_mask=None)
-            dec_output = paddle.transpose(output, [1, 0, 2])
+                paddle.shape(dec_seq_embed)[1])
+            tgt = dec_seq_embed
+            for decoder_layer in self.decoder:
+                tgt = decoder_layer(tgt, memory, self_mask=tgt_mask)
+            dec_output = tgt
             dec_output = dec_output[:, -1, :]
-            word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=1)
-            preds_idx = paddle.argmax(word_prob, axis=1)
+            word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=-1)
+            preds_idx = paddle.argmax(word_prob, axis=-1)
             if paddle.equal_all(
                     preds_idx,
                     paddle.full(
                         paddle.shape(preds_idx), 3, dtype='int64')):
                 break
-            preds_prob = paddle.max(word_prob, axis=1)
+            preds_prob = paddle.max(word_prob, axis=-1)
             dec_seq = paddle.concat(
                 [dec_seq, paddle.reshape(preds_idx, [-1, 1])], axis=1)
             dec_prob = paddle.concat(
@@ -194,10 +187,10 @@ class Transformer(nn.Layer):
         return [dec_seq, dec_prob]
 
     def forward_beam(self, images):
-        ''' Translation work in one batch '''
+        """ Translation work in one batch """
 
         def get_inst_idx_to_tensor_position_map(inst_idx_list):
-            ''' Indicate the position of an instance in a tensor. '''
+            """ Indicate the position of an instance in a tensor. """
             return {
                 inst_idx: tensor_position
                 for tensor_position, inst_idx in enumerate(inst_idx_list)
@@ -205,7 +198,7 @@ class Transformer(nn.Layer):
 
         def collect_active_part(beamed_tensor, curr_active_inst_idx,
                                 n_prev_active_inst, n_bm):
-            ''' Collect tensor parts associated to active instances. '''
+            """ Collect tensor parts associated to active instances. """
 
             beamed_tensor_shape = paddle.shape(beamed_tensor)
             n_curr_active_inst = len(curr_active_inst_idx)
@@ -237,9 +230,8 @@ class Transformer(nn.Layer):
             return active_src_enc, active_inst_idx_to_position_map
 
         def beam_decode_step(inst_dec_beams, len_dec_seq, enc_output,
-                             inst_idx_to_position_map, n_bm,
-                             memory_key_padding_mask):
-            ''' Decode and update beam status, and then return active beam idx '''
+                             inst_idx_to_position_map, n_bm):
+            """ Decode and update beam status, and then return active beam idx """
 
             def prepare_beam_dec_seq(inst_dec_beams, len_dec_seq):
                 dec_partial_seq = [
@@ -249,19 +241,15 @@ class Transformer(nn.Layer):
                 dec_partial_seq = dec_partial_seq.reshape([-1, len_dec_seq])
                 return dec_partial_seq
 
-            def predict_word(dec_seq, enc_output, n_active_inst, n_bm,
-                             memory_key_padding_mask):
-                dec_seq = paddle.transpose(self.embedding(dec_seq), [1, 0, 2])
+            def predict_word(dec_seq, enc_output, n_active_inst, n_bm):
+                dec_seq = self.embedding(dec_seq)
                 dec_seq = self.positional_encoding(dec_seq)
                 tgt_mask = self.generate_square_subsequent_mask(
-                    paddle.shape(dec_seq)[0])
-                dec_output = self.decoder(
-                    dec_seq,
-                    enc_output,
-                    tgt_mask=tgt_mask,
-                    tgt_key_padding_mask=None,
-                    memory_key_padding_mask=memory_key_padding_mask, )
-                dec_output = paddle.transpose(dec_output, [1, 0, 2])
+                    paddle.shape(dec_seq)[1])
+                tgt = dec_seq
+                for decoder_layer in self.decoder:
+                    tgt = decoder_layer(tgt, enc_output, self_mask=tgt_mask)
+                dec_output = tgt
                 dec_output = dec_output[:,
                                         -1, :]  # Pick the last step: (bh * bm) * d_h
                 word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=1)
@@ -281,8 +269,7 @@ class Transformer(nn.Layer):
 
             n_active_inst = len(inst_idx_to_position_map)
             dec_seq = prepare_beam_dec_seq(inst_dec_beams, len_dec_seq)
-            word_prob = predict_word(dec_seq, enc_output, n_active_inst, n_bm,
-                                     None)
+            word_prob = predict_word(dec_seq, enc_output, n_active_inst, n_bm)
             # Update the beam with predicted word prob information and collect incomplete instances
             active_inst_idx_list = collect_active_inst_idx_list(
                 inst_dec_beams, word_prob, inst_idx_to_position_map)
@@ -303,10 +290,10 @@ class Transformer(nn.Layer):
         with paddle.no_grad():
             #-- Encode
             if self.encoder is not None:
-                src = self.positional_encoding(images.transpose([1, 0, 2]))
+                src = self.positional_encoding(images)
                 src_enc = self.encoder(src)
             else:
-                src_enc = images.squeeze(2).transpose([0, 2, 1])
+                src_enc = images
 
             n_bm = self.beam_size
             src_shape = paddle.shape(src_enc)
@@ -317,11 +304,11 @@ class Transformer(nn.Layer):
             inst_idx_to_position_map = get_inst_idx_to_tensor_position_map(
                 active_inst_idx_list)
             # Decode
-            for len_dec_seq in range(1, 25):
+            for len_dec_seq in range(1, self.max_len):
                 src_enc_copy = src_enc.clone()
                 active_inst_idx_list = beam_decode_step(
                     inst_dec_beams, len_dec_seq, src_enc_copy,
-                    inst_idx_to_position_map, n_bm, None)
+                    inst_idx_to_position_map, n_bm)
                 if not active_inst_idx_list:
                     break  # all instances have finished their path to <EOS>
                 src_enc, inst_idx_to_position_map = collate_active_info(
@@ -354,261 +341,124 @@ class Transformer(nn.Layer):
                 shape=[sz, sz], dtype='float32', fill_value='-inf'),
             diagonal=1)
         mask = mask + mask_inf
-        return mask
-
-    def generate_padding_mask(self, x):
-        padding_mask = paddle.equal(x, paddle.to_tensor(0, dtype=x.dtype))
-        return padding_mask
-
-    def _reset_parameters(self):
-        """Initiate parameters in the transformer model."""
-
-        for p in self.parameters():
-            if p.dim() > 1:
-                xavier_uniform_(p)
+        return mask.unsqueeze([0, 1])
 
 
-class TransformerEncoder(nn.Layer):
-    """TransformerEncoder is a stack of N encoder layers
-    Args:
-        encoder_layer: an instance of the TransformerEncoderLayer() class (required).
-        num_layers: the number of sub-encoder-layers in the encoder (required).
-        norm: the layer normalization component (optional).
-    """
+class MultiheadAttention(nn.Layer):
+    """Allows the model to jointly attend to information
+    from different representation subspaces.
+    See reference: Attention Is All You Need
 
-    def __init__(self, encoder_layer, num_layers):
-        super(TransformerEncoder, self).__init__()
-        self.layers = _get_clones(encoder_layer, num_layers)
-        self.num_layers = num_layers
-
-    def forward(self, src):
-        """Pass the input through the endocder layers in turn.
-        Args:
-            src: the sequnce to the encoder (required).
-            mask: the mask for the src sequence (optional).
-            src_key_padding_mask: the mask for the src keys per batch (optional).
-        """
-        output = src
-
-        for i in range(self.num_layers):
-            output = self.layers[i](output,
-                                    src_mask=None,
-                                    src_key_padding_mask=None)
-
-        return output
-
-
-class TransformerDecoder(nn.Layer):
-    """TransformerDecoder is a stack of N decoder layers
+    .. math::
+        \text{MultiHead}(Q, K, V) = \text{Concat}(head_1,\dots,head_h)W^O
+        \text{where} head_i = \text{Attention}(QW_i^Q, KW_i^K, VW_i^V)
 
     Args:
-        decoder_layer: an instance of the TransformerDecoderLayer() class (required).
-        num_layers: the number of sub-decoder-layers in the decoder (required).
-        norm: the layer normalization component (optional).
+        embed_dim: total dimension of the model
+        num_heads: parallel attention layers, or heads
 
     """
 
-    def __init__(self, decoder_layer, num_layers):
-        super(TransformerDecoder, self).__init__()
-        self.layers = _get_clones(decoder_layer, num_layers)
-        self.num_layers = num_layers
+    def __init__(self, embed_dim, num_heads, dropout=0., self_attn=False):
+        super(MultiheadAttention, self).__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        # self.dropout = dropout
+        self.head_dim = embed_dim // num_heads
+        assert self.head_dim * num_heads == self.embed_dim, "embed_dim must be divisible by num_heads"
+        self.scale = self.head_dim**-0.5
+        self.self_attn = self_attn
+        if self_attn:
+            self.qkv = nn.Linear(embed_dim, embed_dim * 3)
+        else:
+            self.q = nn.Linear(embed_dim, embed_dim)
+            self.kv = nn.Linear(embed_dim, embed_dim * 2)
+        self.attn_drop = nn.Dropout(dropout)
+        self.out_proj = nn.Linear(embed_dim, embed_dim)
 
-    def forward(self,
-                tgt,
-                memory,
-                tgt_mask=None,
-                memory_mask=None,
-                tgt_key_padding_mask=None,
-                memory_key_padding_mask=None):
-        """Pass the inputs (and mask) through the decoder layer in turn.
+    def forward(self, query, key=None, attn_mask=None):
 
-        Args:
-            tgt: the sequence to the decoder (required).
-            memory: the sequnce from the last layer of the encoder (required).
-            tgt_mask: the mask for the tgt sequence (optional).
-            memory_mask: the mask for the memory sequence (optional).
-            tgt_key_padding_mask: the mask for the tgt keys per batch (optional).
-            memory_key_padding_mask: the mask for the memory keys per batch (optional).
-        """
-        output = tgt
-        for i in range(self.num_layers):
-            output = self.layers[i](
-                output,
-                memory,
-                tgt_mask=tgt_mask,
-                memory_mask=memory_mask,
-                tgt_key_padding_mask=tgt_key_padding_mask,
-                memory_key_padding_mask=memory_key_padding_mask)
+        qN = query.shape[1]
 
-        return output
+        if self.self_attn:
+            qkv = self.qkv(query).reshape(
+                (0, qN, 3, self.num_heads, self.head_dim)).transpose(
+                    (2, 0, 3, 1, 4))
+            q, k, v = qkv[0], qkv[1], qkv[2]
+        else:
+            kN = key.shape[1]
+            q = self.q(query).reshape(
+                [0, qN, self.num_heads, self.head_dim]).transpose([0, 2, 1, 3])
+            kv = self.kv(key).reshape(
+                (0, kN, 2, self.num_heads, self.head_dim)).transpose(
+                    (2, 0, 3, 1, 4))
+            k, v = kv[0], kv[1]
+
+        attn = (q.matmul(k.transpose((0, 1, 3, 2)))) * self.scale
+
+        if attn_mask is not None:
+            attn += attn_mask
+
+        attn = F.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape(
+            (0, qN, self.embed_dim))
+        x = self.out_proj(x)
+
+        return x
 
 
-class TransformerEncoderLayer(nn.Layer):
-    """TransformerEncoderLayer is made up of self-attn and feedforward network.
-    This standard encoder layer is based on the paper "Attention Is All You Need".
-    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
-    Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
-    Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
-    in a different way during application.
-
-    Args:
-        d_model: the number of expected features in the input (required).
-        nhead: the number of heads in the multiheadattention models (required).
-        dim_feedforward: the dimension of the feedforward network model (default=2048).
-        dropout: the dropout value (default=0.1).
-
-    """
-
+class TransformerBlock(nn.Layer):
     def __init__(self,
                  d_model,
                  nhead,
                  dim_feedforward=2048,
                  attention_dropout_rate=0.0,
-                 residual_dropout_rate=0.1):
-        super(TransformerEncoderLayer, self).__init__()
-        self.self_attn = MultiheadAttention(
-            d_model, nhead, dropout=attention_dropout_rate)
+                 residual_dropout_rate=0.1,
+                 with_self_attn=True,
+                 with_cross_attn=False,
+                 epsilon=1e-5):
+        super(TransformerBlock, self).__init__()
+        self.with_self_attn = with_self_attn
+        if with_self_attn:
+            self.self_attn = MultiheadAttention(
+                d_model,
+                nhead,
+                dropout=attention_dropout_rate,
+                self_attn=with_self_attn)
+            self.norm1 = LayerNorm(d_model, epsilon=epsilon)
+            self.dropout1 = Dropout(residual_dropout_rate)
+        self.with_cross_attn = with_cross_attn
+        if with_cross_attn:
+            self.cross_attn = MultiheadAttention(  #for self_attn of encoder or cross_attn of decoder
+                d_model,
+                nhead,
+                dropout=attention_dropout_rate)
+            self.norm2 = LayerNorm(d_model, epsilon=epsilon)
+            self.dropout2 = Dropout(residual_dropout_rate)
 
-        self.conv1 = Conv2D(
-            in_channels=d_model,
-            out_channels=dim_feedforward,
-            kernel_size=(1, 1))
-        self.conv2 = Conv2D(
-            in_channels=dim_feedforward,
-            out_channels=d_model,
-            kernel_size=(1, 1))
+        self.mlp = Mlp(in_features=d_model,
+                       hidden_features=dim_feedforward,
+                       act_layer=nn.ReLU,
+                       drop=residual_dropout_rate)
 
-        self.norm1 = LayerNorm(d_model)
-        self.norm2 = LayerNorm(d_model)
-        self.dropout1 = Dropout(residual_dropout_rate)
-        self.dropout2 = Dropout(residual_dropout_rate)
+        self.norm3 = LayerNorm(d_model, epsilon=epsilon)
 
-    def forward(self, src, src_mask=None, src_key_padding_mask=None):
-        """Pass the input through the endocder layer.
-        Args:
-            src: the sequnce to the encoder layer (required).
-            src_mask: the mask for the src sequence (optional).
-            src_key_padding_mask: the mask for the src keys per batch (optional).
-        """
-        src2 = self.self_attn(
-            src,
-            src,
-            src,
-            attn_mask=src_mask,
-            key_padding_mask=src_key_padding_mask)
-        src = src + self.dropout1(src2)
-        src = self.norm1(src)
-
-        src = paddle.transpose(src, [1, 2, 0])
-        src = paddle.unsqueeze(src, 2)
-        src2 = self.conv2(F.relu(self.conv1(src)))
-        src2 = paddle.squeeze(src2, 2)
-        src2 = paddle.transpose(src2, [2, 0, 1])
-        src = paddle.squeeze(src, 2)
-        src = paddle.transpose(src, [2, 0, 1])
-
-        src = src + self.dropout2(src2)
-        src = self.norm2(src)
-        return src
-
-
-class TransformerDecoderLayer(nn.Layer):
-    """TransformerDecoderLayer is made up of self-attn, multi-head-attn and feedforward network.
-    This standard decoder layer is based on the paper "Attention Is All You Need".
-    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,
-    Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in
-    Neural Information Processing Systems, pages 6000-6010. Users may modify or implement
-    in a different way during application.
-
-    Args:
-        d_model: the number of expected features in the input (required).
-        nhead: the number of heads in the multiheadattention models (required).
-        dim_feedforward: the dimension of the feedforward network model (default=2048).
-        dropout: the dropout value (default=0.1).
-
-    """
-
-    def __init__(self,
-                 d_model,
-                 nhead,
-                 dim_feedforward=2048,
-                 attention_dropout_rate=0.0,
-                 residual_dropout_rate=0.1):
-        super(TransformerDecoderLayer, self).__init__()
-        self.self_attn = MultiheadAttention(
-            d_model, nhead, dropout=attention_dropout_rate)
-        self.multihead_attn = MultiheadAttention(
-            d_model, nhead, dropout=attention_dropout_rate)
-
-        self.conv1 = Conv2D(
-            in_channels=d_model,
-            out_channels=dim_feedforward,
-            kernel_size=(1, 1))
-        self.conv2 = Conv2D(
-            in_channels=dim_feedforward,
-            out_channels=d_model,
-            kernel_size=(1, 1))
-
-        self.norm1 = LayerNorm(d_model)
-        self.norm2 = LayerNorm(d_model)
-        self.norm3 = LayerNorm(d_model)
-        self.dropout1 = Dropout(residual_dropout_rate)
-        self.dropout2 = Dropout(residual_dropout_rate)
         self.dropout3 = Dropout(residual_dropout_rate)
 
-    def forward(self,
-                tgt,
-                memory,
-                tgt_mask=None,
-                memory_mask=None,
-                tgt_key_padding_mask=None,
-                memory_key_padding_mask=None):
-        """Pass the inputs (and mask) through the decoder layer.
+    def forward(self, tgt, memory=None, self_mask=None, cross_mask=None):
+        if self.with_self_attn:
+            tgt1 = self.self_attn(tgt, attn_mask=self_mask)
+            tgt = self.norm1(tgt + self.dropout1(tgt1))
 
-        Args:
-            tgt: the sequence to the decoder layer (required).
-            memory: the sequnce from the last layer of the encoder (required).
-            tgt_mask: the mask for the tgt sequence (optional).
-            memory_mask: the mask for the memory sequence (optional).
-            tgt_key_padding_mask: the mask for the tgt keys per batch (optional).
-            memory_key_padding_mask: the mask for the memory keys per batch (optional).
-
-        """
-        tgt2 = self.self_attn(
-            tgt,
-            tgt,
-            tgt,
-            attn_mask=tgt_mask,
-            key_padding_mask=tgt_key_padding_mask)
-        tgt = tgt + self.dropout1(tgt2)
-        tgt = self.norm1(tgt)
-        tgt2 = self.multihead_attn(
-            tgt,
-            memory,
-            memory,
-            attn_mask=memory_mask,
-            key_padding_mask=memory_key_padding_mask)
-        tgt = tgt + self.dropout2(tgt2)
-        tgt = self.norm2(tgt)
-
-        # default
-        tgt = paddle.transpose(tgt, [1, 2, 0])
-        tgt = paddle.unsqueeze(tgt, 2)
-        tgt2 = self.conv2(F.relu(self.conv1(tgt)))
-        tgt2 = paddle.squeeze(tgt2, 2)
-        tgt2 = paddle.transpose(tgt2, [2, 0, 1])
-        tgt = paddle.squeeze(tgt, 2)
-        tgt = paddle.transpose(tgt, [2, 0, 1])
-
-        tgt = tgt + self.dropout3(tgt2)
-        tgt = self.norm3(tgt)
+        if self.with_cross_attn:
+            tgt2 = self.cross_attn(tgt, key=memory, attn_mask=cross_mask)
+            tgt = self.norm2(tgt + self.dropout2(tgt2))
+        tgt = self.norm3(tgt + self.dropout3(self.mlp(tgt)))
         return tgt
 
 
-def _get_clones(module, N):
-    return LayerList([copy.deepcopy(module) for i in range(N)])
-
-
 class PositionalEncoding(nn.Layer):
     """Inject some information about the relative or absolute position of the tokens
         in the sequence. The positional encodings have the same dimension as
@@ -651,8 +501,9 @@ class PositionalEncoding(nn.Layer):
         Examples:
             >>> output = pos_encoder(x)
         """
+        x = x.transpose([1, 0, 2])
         x = x + self.pe[:paddle.shape(x)[0], :]
-        return self.dropout(x)
+        return self.dropout(x).transpose([1, 0, 2])
 
 
 class PositionalEncoding_2d(nn.Layer):
@@ -725,7 +576,7 @@ class PositionalEncoding_2d(nn.Layer):
 
 
 class Embeddings(nn.Layer):
-    def __init__(self, d_model, vocab, padding_idx, scale_embedding):
+    def __init__(self, d_model, vocab, padding_idx=None, scale_embedding=True):
         super(Embeddings, self).__init__()
         self.embedding = nn.Embedding(vocab, d_model, padding_idx=padding_idx)
         w0 = np.random.normal(0.0, d_model**-0.5,
@@ -742,7 +593,7 @@ class Embeddings(nn.Layer):
 
 
 class Beam():
-    ''' Beam search '''
+    """ Beam search """
 
     def __init__(self, size, device=False):
 
diff --git a/ppocr/postprocess/__init__.py b/ppocr/postprocess/__init__.py
index f50b5f1c5f..2635117c84 100644
--- a/ppocr/postprocess/__init__.py
+++ b/ppocr/postprocess/__init__.py
@@ -27,7 +27,7 @@ from .sast_postprocess import SASTPostProcess
 from .fce_postprocess import FCEPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, \
     DistillationCTCLabelDecode, TableLabelDecode, NRTRLabelDecode, SARLabelDecode, \
-    SEEDLabelDecode, PRENLabelDecode
+    SEEDLabelDecode, PRENLabelDecode, ViTSTRLabelDecode, ABINetLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
 from .vqa_token_ser_layoutlm_postprocess import VQASerTokenLayoutLMPostProcess
@@ -42,7 +42,7 @@ def build_post_process(config, global_config=None):
         'DistillationDBPostProcess', 'NRTRLabelDecode', 'SARLabelDecode',
         'SEEDLabelDecode', 'VQASerTokenLayoutLMPostProcess',
         'VQAReTokenLayoutLMPostProcess', 'PRENLabelDecode',
-        'DistillationSARLabelDecode'
+        'DistillationSARLabelDecode', 'ViTSTRLabelDecode', 'ABINetLabelDecode'
     ]
 
     if config['name'] == 'PSEPostProcess':
diff --git a/ppocr/postprocess/rec_postprocess.py b/ppocr/postprocess/rec_postprocess.py
index bf0fd890bf..c77420ad19 100644
--- a/ppocr/postprocess/rec_postprocess.py
+++ b/ppocr/postprocess/rec_postprocess.py
@@ -140,70 +140,6 @@ class DistillationCTCLabelDecode(CTCLabelDecode):
         return output
 
 
-class NRTRLabelDecode(BaseRecLabelDecode):
-    """ Convert between text-label and text-index """
-
-    def __init__(self, character_dict_path=None, use_space_char=True, **kwargs):
-        super(NRTRLabelDecode, self).__init__(character_dict_path,
-                                              use_space_char)
-
-    def __call__(self, preds, label=None, *args, **kwargs):
-
-        if len(preds) == 2:
-            preds_id = preds[0]
-            preds_prob = preds[1]
-            if isinstance(preds_id, paddle.Tensor):
-                preds_id = preds_id.numpy()
-            if isinstance(preds_prob, paddle.Tensor):
-                preds_prob = preds_prob.numpy()
-            if preds_id[0][0] == 2:
-                preds_idx = preds_id[:, 1:]
-                preds_prob = preds_prob[:, 1:]
-            else:
-                preds_idx = preds_id
-            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
-            if label is None:
-                return text
-            label = self.decode(label[:, 1:])
-        else:
-            if isinstance(preds, paddle.Tensor):
-                preds = preds.numpy()
-            preds_idx = preds.argmax(axis=2)
-            preds_prob = preds.max(axis=2)
-            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
-            if label is None:
-                return text
-            label = self.decode(label[:, 1:])
-        return text, label
-
-    def add_special_char(self, dict_character):
-        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
-        return dict_character
-
-    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
-        """ convert text-index into text-label. """
-        result_list = []
-        batch_size = len(text_index)
-        for batch_idx in range(batch_size):
-            char_list = []
-            conf_list = []
-            for idx in range(len(text_index[batch_idx])):
-                if text_index[batch_idx][idx] == 3:  # end
-                    break
-                try:
-                    char_list.append(self.character[int(text_index[batch_idx][
-                        idx])])
-                except:
-                    continue
-                if text_prob is not None:
-                    conf_list.append(text_prob[batch_idx][idx])
-                else:
-                    conf_list.append(1)
-            text = ''.join(char_list)
-            result_list.append((text.lower(), np.mean(conf_list).tolist()))
-        return result_list
-
-
 class AttnLabelDecode(BaseRecLabelDecode):
     """ Convert between text-label and text-index """
 
@@ -752,3 +688,122 @@ class PRENLabelDecode(BaseRecLabelDecode):
             return text
         label = self.decode(label)
         return text, label
+
+
+class NRTRLabelDecode(BaseRecLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=True, **kwargs):
+        super(NRTRLabelDecode, self).__init__(character_dict_path,
+                                              use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+
+        if len(preds) == 2:
+            preds_id = preds[0]
+            preds_prob = preds[1]
+            if isinstance(preds_id, paddle.Tensor):
+                preds_id = preds_id.numpy()
+            if isinstance(preds_prob, paddle.Tensor):
+                preds_prob = preds_prob.numpy()
+            if preds_id[0][0] == 2:
+                preds_idx = preds_id[:, 1:]
+                preds_prob = preds_prob[:, 1:]
+            else:
+                preds_idx = preds_id
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            if label is None:
+                return text
+            label = self.decode(label[:, 1:])
+        else:
+            if isinstance(preds, paddle.Tensor):
+                preds = preds.numpy()
+            preds_idx = preds.argmax(axis=2)
+            preds_prob = preds.max(axis=2)
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            if label is None:
+                return text
+            label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        result_list = []
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                try:
+                    char_idx = self.character[int(text_index[batch_idx][idx])]
+                except:
+                    continue
+                if char_idx == '</s>':  # end
+                    break
+                char_list.append(char_idx)
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = ''.join(char_list)
+            result_list.append((text.lower(), np.mean(conf_list).tolist()))
+        return result_list
+
+
+class ViTSTRLabelDecode(NRTRLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=False,
+                 **kwargs):
+        super(ViTSTRLabelDecode, self).__init__(character_dict_path,
+                                                use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds[:, 1:].numpy()
+        else:
+            preds = preds[:, 1:]
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['<s>', '</s>'] + dict_character
+        return dict_character
+
+
+class ABINetLabelDecode(NRTRLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self, character_dict_path=None, use_space_char=False,
+                 **kwargs):
+        super(ABINetLabelDecode, self).__init__(character_dict_path,
+                                                use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, dict):
+            preds = preds['align'][-1].numpy()
+        elif isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        else:
+            preds = preds
+
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ['</s>'] + dict_character
+        return dict_character
diff --git a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
index 15119bb2a9..8118d58724 100644
--- a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
+++ b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
@@ -49,7 +49,7 @@ Architecture:
     
 
 Loss:
-  name: NRTRLoss
+  name: CELoss
   smoothing: True
 
 PostProcess:
@@ -69,7 +69,7 @@ Train:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
+      - GrayRecResizeImg:
           image_shape: [100, 32]
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
@@ -90,7 +90,7 @@ Eval:
           img_mode: BGR
           channel_first: False
       - NRTRLabelEncode: # Class handling label
-      - NRTRRecResizeImg:
+      - GrayRecResizeImg:
           image_shape: [100, 32]
           resize_type: PIL # PIL or OpenCV
       - KeepKeys:
@@ -99,5 +99,5 @@ Eval:
     shuffle: False
     drop_last: False
     batch_size_per_card: 256
-    num_workers: 1
+    num_workers: 4
     use_shared_memory: False
diff --git a/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
new file mode 100644
index 0000000000..5b5890e772
--- /dev/null
+++ b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
@@ -0,0 +1,106 @@
+Global:
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r45_abinet/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_abinet.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 20.0
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: ABINet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ResNet45
+
+  Head:
+    name: ABINetHead
+    use_lang: True
+    iter_size: 3
+    
+
+Loss:
+  name: CELoss
+  ignore_index: &ignore_index 100 # Must be greater than the number of character classes
+
+PostProcess:
+  name: ABINetLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetRecAug:
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 96
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/test_tipc/configs/rec_r45_abinet/train_infer_python.txt b/test_tipc/configs/rec_r45_abinet/train_infer_python.txt
new file mode 100644
index 0000000000..ecab1bcbbd
--- /dev/null
+++ b/test_tipc/configs/rec_r45_abinet/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_abinet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r45_abinet_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,128" --rec_algorithm="ABINet"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,128]}]
diff --git a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
new file mode 100644
index 0000000000..140b17e0e7
--- /dev/null
+++ b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
@@ -0,0 +1,117 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_tiny.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 8.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+    name: STN_ON
+    tps_inputsize: [32, 64]
+    tps_outputsize: [32, 100]
+    num_control_points: 20
+    tps_margins: [0.05,0.05]
+    stn_activation: none
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 100]
+    out_char_num: 25
+    out_channels: 192
+    patch_merging: 'Conv'
+    embed_dim: [64, 128, 256]
+    depth: [3, 6, 3]
+    num_heads: [2, 4, 8]
+    mixer: ['Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[7, 11], [7, 11], [7, 11]]
+    last_stage: True
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/test_tipc/configs/rec_svtrnet/train_infer_python.txt b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
new file mode 100644
index 0000000000..a7e4a24063
--- /dev/null
+++ b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_svtrnet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_svtrnet_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,64,256" --rec_algorithm="SVTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,64,256]}]
diff --git a/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
new file mode 100644
index 0000000000..a0aed48875
--- /dev/null
+++ b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
@@ -0,0 +1,104 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/vitstr_none_ce/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration#
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_vitstr.txt
+
+
+Optimizer:
+  name: Adadelta
+  epsilon: 1.e-8
+  rho: 0.95
+  clip_norm: 5.0
+  lr:
+    learning_rate: 1.0
+
+Architecture:
+  model_type: rec
+  algorithm: ViTSTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ViTSTR
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CELoss
+  smoothing: False
+  with_all: True
+  ignore_index: &ignore_index 0 # Must be zero or greater than the number of character classes
+
+PostProcess:
+  name: ViTSTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
new file mode 100644
index 0000000000..04c5742ea2
--- /dev/null
+++ b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_vitstr
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_vitstr_none_ce_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="ViTSTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,224,224]}]
diff --git a/tools/export_model.py b/tools/export_model.py
index 3ea0228f85..b10d41d5b2 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -31,7 +31,12 @@ from ppocr.utils.logging import get_logger
 from tools.program import load_config, merge_config, ArgsParser
 
 
-def export_single_model(model, arch_config, save_path, logger, quanter=None):
+def export_single_model(model,
+                        arch_config,
+                        save_path,
+                        logger,
+                        input_shape=None,
+                        quanter=None):
     if arch_config["algorithm"] == "SRN":
         max_text_length = arch_config["Head"]["max_text_length"]
         other_shape = [
@@ -64,7 +69,7 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
         else:
             other_shape = [
                 paddle.static.InputSpec(
-                    shape=[None, 3, 64, 256], dtype="float32"),
+                    shape=[None] + input_shape, dtype="float32"),
             ]
         model = to_static(model, input_spec=other_shape)
     elif arch_config["algorithm"] == "PREN":
@@ -73,6 +78,25 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
                 shape=[None, 3, 64, 512], dtype="float32"),
         ]
         model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "ViTSTR":
+        other_shape = [
+            paddle.static.InputSpec(
+                shape=[None, 1, 224, 224], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "ABINet":
+        other_shape = [
+            paddle.static.InputSpec(
+                shape=[None, 3, 32, 128], dtype="float32"),
+        ]
+        # print([None, 3, 32, 128])
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "NRTR":
+        other_shape = [
+            paddle.static.InputSpec(
+                shape=[None, 1, 32, 100], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
     else:
         infer_shape = [3, -1, -1]
         if arch_config["model_type"] == "rec":
@@ -84,8 +108,6 @@ def export_single_model(model, arch_config, save_path, logger, quanter=None):
                     "When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training"
                 )
                 infer_shape[-1] = 100
-            if arch_config["algorithm"] == "NRTR":
-                infer_shape = [1, 32, 100]
         elif arch_config["model_type"] == "table":
             infer_shape = [3, 488, 488]
         model = to_static(
@@ -157,6 +179,13 @@ def main():
 
     arch_config = config["Architecture"]
 
+    if arch_config["algorithm"] == "SVTR" and arch_config["Head"][
+            "name"] != 'MultiHead':
+        input_shape = config["Eval"]["dataset"]["transforms"][-2][
+            'SVTRRecResizeImg']['image_shape']
+    else:
+        input_shape = None
+
     if arch_config["algorithm"] in ["Distillation", ]:  # distillation model
         archs = list(arch_config["Models"].values())
         for idx, name in enumerate(model.model_name_list):
@@ -165,7 +194,8 @@ def main():
                                 sub_model_save_path, logger)
     else:
         save_path = os.path.join(save_path, "inference")
-        export_single_model(model, arch_config, save_path, logger)
+        export_single_model(
+            model, arch_config, save_path, logger, input_shape=input_shape)
 
 
 if __name__ == "__main__":
diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py
index 3664ef2caf..a95f555966 100755
--- a/tools/infer/predict_rec.py
+++ b/tools/infer/predict_rec.py
@@ -69,6 +69,18 @@ class TextRecognizer(object):
                 "character_dict_path": args.rec_char_dict_path,
                 "use_space_char": args.use_space_char
             }
+        elif self.rec_algorithm == 'ViTSTR':
+            postprocess_params = {
+                'name': 'ViTSTRLabelDecode',
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char
+            }
+        elif self.rec_algorithm == 'ABINet':
+            postprocess_params = {
+                'name': 'ABINetLabelDecode',
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char
+            }
         self.postprocess_op = build_post_process(postprocess_params)
         self.predictor, self.input_tensor, self.output_tensors, self.config = \
             utility.create_predictor(args, 'rec', logger)
@@ -96,15 +108,22 @@ class TextRecognizer(object):
 
     def resize_norm_img(self, img, max_wh_ratio):
         imgC, imgH, imgW = self.rec_image_shape
-        if self.rec_algorithm == 'NRTR':
+        if self.rec_algorithm == 'NRTR' or self.rec_algorithm == 'ViTSTR':
             img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
             # return padding_im
             image_pil = Image.fromarray(np.uint8(img))
-            img = image_pil.resize([100, 32], Image.ANTIALIAS)
+            if self.rec_algorithm == 'ViTSTR':
+                img = image_pil.resize([imgW, imgH], Image.BICUBIC)
+            else:
+                img = image_pil.resize([imgW, imgH], Image.ANTIALIAS)
             img = np.array(img)
             norm_img = np.expand_dims(img, -1)
             norm_img = norm_img.transpose((2, 0, 1))
-            return norm_img.astype(np.float32) / 128. - 1.
+            if self.rec_algorithm == 'ViTSTR':
+                norm_img = norm_img.astype(np.float32) / 255.
+            else:
+                norm_img = norm_img.astype(np.float32) / 128. - 1.
+            return norm_img
 
         assert imgC == img.shape[2]
         imgW = int((imgH * max_wh_ratio))
@@ -132,17 +151,6 @@ class TextRecognizer(object):
         padding_im[:, :, 0:resized_w] = resized_image
         return padding_im
 
-    def resize_norm_img_svtr(self, img, image_shape):
-
-        imgC, imgH, imgW = image_shape
-        resized_image = cv2.resize(
-            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
-        resized_image = resized_image.astype('float32')
-        resized_image = resized_image.transpose((2, 0, 1)) / 255
-        resized_image -= 0.5
-        resized_image /= 0.5
-        return resized_image
-
     def resize_norm_img_srn(self, img, image_shape):
         imgC, imgH, imgW = image_shape
 
@@ -250,6 +258,35 @@ class TextRecognizer(object):
 
         return padding_im, resize_shape, pad_shape, valid_ratio
 
+    def resize_norm_img_svtr(self, img, image_shape):
+
+        imgC, imgH, imgW = image_shape
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype('float32')
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        return resized_image
+
+    def resize_norm_img_abinet(self, img, image_shape):
+
+        imgC, imgH, imgW = image_shape
+
+        resized_image = cv2.resize(
+            img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype('float32')
+        resized_image = resized_image / 255.
+
+        mean = np.array([0.485, 0.456, 0.406])
+        std = np.array([0.229, 0.224, 0.225])
+        resized_image = (
+            resized_image - mean[None, None, ...]) / std[None, None, ...]
+        resized_image = resized_image.transpose((2, 0, 1))
+        resized_image = resized_image.astype('float32')
+
+        return resized_image
+
     def __call__(self, img_list):
         img_num = len(img_list)
         # Calculate the aspect ratio of all text bars
@@ -300,6 +337,11 @@ class TextRecognizer(object):
                                                          self.rec_image_shape)
                     norm_img = norm_img[np.newaxis, :]
                     norm_img_batch.append(norm_img)
+                elif self.rec_algorithm == "ABINet":
+                    norm_img = self.resize_norm_img_abinet(
+                        img_list[indices[ino]], self.rec_image_shape)
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
                 else:
                     norm_img = self.resize_norm_img(img_list[indices[ino]],
                                                     max_wh_ratio)
diff --git a/tools/program.py b/tools/program.py
index 9a90eda2f8..aa3ba82c44 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -576,7 +576,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
+        'ViTSTR', 'ABINet'
     ]
 
     if use_xpu:

From 04e710419468ad25afa5c8e478d9c88c23ea0d5c Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 28 Jun 2022 11:45:21 +0000
Subject: [PATCH 077/134] add db++

---
 configs/det/det_r50_db++_ic15.yml  |  2 +-
 configs/det/det_r50_db++_td_tr.yml |  2 +-
 tools/infer/predict_det.py         | 19 +++++++++++++++++-
 tools/program.py                   | 31 ++++++++++++++++++++----------
 4 files changed, 41 insertions(+), 13 deletions(-)

diff --git a/configs/det/det_r50_db++_ic15.yml b/configs/det/det_r50_db++_ic15.yml
index a90c868cc3..e0cd6012b6 100644
--- a/configs/det/det_r50_db++_ic15.yml
+++ b/configs/det/det_r50_db++_ic15.yml
@@ -18,7 +18,7 @@ Global:
   save_res_path: ./checkpoints/det_db/predicts_db.txt
 Architecture:
   model_type: det
-  algorithm: DB
+  algorithm: DB++
   Transform: null
   Backbone:
     name: ResNet
diff --git a/configs/det/det_r50_db++_td_tr.yml b/configs/det/det_r50_db++_td_tr.yml
index 5e26ddb0ac..65021bb661 100644
--- a/configs/det/det_r50_db++_td_tr.yml
+++ b/configs/det/det_r50_db++_td_tr.yml
@@ -18,7 +18,7 @@ Global:
   save_res_path: ./checkpoints/det_db/predicts_db.txt
 Architecture:
   model_type: det
-  algorithm: DB
+  algorithm: DB++
   Transform: null
   Backbone:
     name: ResNet
diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py
index 7b6bebf1fb..394a48948b 100755
--- a/tools/infer/predict_det.py
+++ b/tools/infer/predict_det.py
@@ -67,6 +67,23 @@ class TextDetector(object):
             postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
             postprocess_params["use_dilation"] = args.use_dilation
             postprocess_params["score_mode"] = args.det_db_score_mode
+        elif self.det_algorithm == "DB++":
+            postprocess_params['name'] = 'DBPostProcess'
+            postprocess_params["thresh"] = args.det_db_thresh
+            postprocess_params["box_thresh"] = args.det_db_box_thresh
+            postprocess_params["max_candidates"] = 1000
+            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
+            postprocess_params["use_dilation"] = args.use_dilation
+            postprocess_params["score_mode"] = args.det_db_score_mode
+            pre_process_list[1] = {
+                'NormalizeImage': {
+                    'std': [1.0, 1.0, 1.0],
+                    'mean':
+                    [0.48109378172549, 0.45752457890196, 0.40787054090196],
+                    'scale': '1./255.',
+                    'order': 'hwc'
+                }
+            }
         elif self.det_algorithm == "EAST":
             postprocess_params['name'] = 'EASTPostProcess'
             postprocess_params["score_thresh"] = args.det_east_score_thresh
@@ -231,7 +248,7 @@ class TextDetector(object):
             preds['f_score'] = outputs[1]
             preds['f_tco'] = outputs[2]
             preds['f_tvo'] = outputs[3]
-        elif self.det_algorithm in ['DB', 'PSE']:
+        elif self.det_algorithm in ['DB', 'PSE', 'DB++']:
             preds['maps'] = outputs[0]
         elif self.det_algorithm == 'FCE':
             for i, output in enumerate(outputs):
diff --git a/tools/program.py b/tools/program.py
index aa0d2698cf..620c61f093 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -307,7 +307,8 @@ def train(config,
             train_stats.update(stats)
 
             if log_writer is not None and dist.get_rank() == 0:
-                log_writer.log_metrics(metrics=train_stats.get(), prefix="TRAIN", step=global_step)
+                log_writer.log_metrics(
+                    metrics=train_stats.get(), prefix="TRAIN", step=global_step)
 
             if dist.get_rank() == 0 and (
                 (global_step > 0 and global_step % print_batch_step == 0) or
@@ -354,7 +355,8 @@ def train(config,
 
                 # logger metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics=cur_metric, prefix="EVAL", step=global_step)
+                    log_writer.log_metrics(
+                        metrics=cur_metric, prefix="EVAL", step=global_step)
 
                 if cur_metric[main_indicator] >= best_model_dict[
                         main_indicator]:
@@ -377,11 +379,18 @@ def train(config,
                 logger.info(best_str)
                 # logger best metric
                 if log_writer is not None:
-                    log_writer.log_metrics(metrics={
-                        "best_{}".format(main_indicator): best_model_dict[main_indicator]
-                        }, prefix="EVAL", step=global_step)
-                    
-                    log_writer.log_model(is_best=True, prefix="best_accuracy", metadata=best_model_dict)
+                    log_writer.log_metrics(
+                        metrics={
+                            "best_{}".format(main_indicator):
+                            best_model_dict[main_indicator]
+                        },
+                        prefix="EVAL",
+                        step=global_step)
+
+                    log_writer.log_model(
+                        is_best=True,
+                        prefix="best_accuracy",
+                        metadata=best_model_dict)
 
             reader_start = time.time()
         if dist.get_rank() == 0:
@@ -413,7 +422,8 @@ def train(config,
                 epoch=epoch,
                 global_step=global_step)
             if log_writer is not None:
-                log_writer.log_model(is_best=False, prefix='iter_epoch_{}'.format(epoch))
+                log_writer.log_model(
+                    is_best=False, prefix='iter_epoch_{}'.format(epoch))
 
     best_str = 'best metric, {}'.format(', '.join(
         ['{}: {}'.format(k, v) for k, v in best_model_dict.items()]))
@@ -564,7 +574,7 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR', 'DB++'
     ]
 
     if use_xpu:
@@ -585,7 +595,8 @@ def preprocess(is_train=False):
         vdl_writer_path = '{}/vdl/'.format(save_model_dir)
         log_writer = VDLLogger(save_model_dir)
         loggers.append(log_writer)
-    if ('use_wandb' in config['Global'] and config['Global']['use_wandb']) or 'wandb' in config:
+    if ('use_wandb' in config['Global'] and
+            config['Global']['use_wandb']) or 'wandb' in config:
         save_dir = config['Global']['save_model_dir']
         wandb_writer_path = "{}/wandb".format(save_dir)
         if "wandb" in config:

From 9bb4ead8eae06b1be3366594a61e416b610adf51 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 28 Jun 2022 11:51:55 +0000
Subject: [PATCH 078/134] add db++

---
 doc/doc_ch/algorithm_det_db.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/doc_ch/algorithm_det_db.md b/doc/doc_ch/algorithm_det_db.md
index 8466565563..afdddb1a73 100644
--- a/doc/doc_ch/algorithm_det_db.md
+++ b/doc/doc_ch/algorithm_det_db.md
@@ -66,7 +66,7 @@ python3 tools/export_model.py -c configs/det/det_r50_vd_db.yml -o Global.pretrai
 DB文本检测模型推理，可以执行如下命令：
 
 ```shell
-python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_db/"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_db/" --det_algorithm="DB"
 ```
 
 可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：

From bcd56d0054cd251d9280cc43d10348981e52b783 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Tue, 28 Jun 2022 12:31:14 +0000
Subject: [PATCH 079/134] update doc for case

---
 applications/imgs/det.png      | Bin 66713 -> 0 bytes
 applications/imgs/rec.png      | Bin 13765 -> 0 bytes
 applications/imgs/sys_res.png  | Bin 291470 -> 0 bytes
 applications/液晶屏读数识别.md |  86 +++++++++++++++++++--------------
 4 files changed, 49 insertions(+), 37 deletions(-)
 delete mode 100644 applications/imgs/det.png
 delete mode 100644 applications/imgs/rec.png
 delete mode 100644 applications/imgs/sys_res.png

diff --git a/applications/imgs/det.png b/applications/imgs/det.png
deleted file mode 100644
index 6f21ee170ed272e6ae9c3ef475f60855a064cbbc..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 66713
zcmYhj2Rzk({4aivWABhXLnv8C_Exe&^fj`HlgQp%va%1d9m+^{$lg0Lv-jRBd*Ao@
z{qBAI{~nb(j^lIQpVxSf*Q+oMwHHMAwD=GRgh)y8nI;4RRRn*(#>EDI-WQQv0iQ6P
z<dk0Hf)8)px1r$Qc=n2VP7nwI8TvPLc-HqCJSgh?T-RC4&eGY<)X@U+!PME_#?INs
z%8c37!qLgf&X!q#M}P;;&HT>U*<Os7_y2nWkDViemzClf76if!QF<o(+C6QjLC?l;
zGG+d39>GR}h>d^tp6z*TN{;4e(O#pl&rl?@?C`tHSgI-p=5Mj7IKm$x9-+&5%)tQ?
zY{b`ZSFrI&Vf<p)WUP^y_qzCF8oQs*)4u*M<+*aUrRLXXu@5{HGaS|fE}z_<krjL%
zt$L;Xv;rc_HX!9yvUqhne@8bqrIeEEIX_o*w(H|HKWAIxH_71sfB&g7?d9+Nzkk0<
zIb~#?|KBGy<~HDT|L@bQO04#w|IptEzGOZ1|9^S|>@AY*P65d_qJ*4LL9W>O%@6qf
z?~6@at27RXb+ak`Lt3)kSR*qOFI9)<6-SQ8&c?$2?+Z)*+E11B<=!+($#t?y8DqaG
z%a%URxH@QTu-l)$`jw}Bo`G^x95%nY3Co@j<yA(mAdtSNYvUwG7xhTC!+}b=qtz6_
zm+<lP`Q?4T5ufYi`7zfzx9gfS*2L*wFKZ0cl#rk9YF<d8ba)a8#&fJvYF;4O-d9a0
zqncFw8iv--@2cm$Xs(2P*0<Nh7*mZCdskf<^k<B3JNLcxLhP<?wl7yuyQgU-UZ=J`
zceUG1N}Wx;Q|IuTd0SNF^sXw(ps}&{$W=elZfo3j{x1BNoxa<xUF93_BjztN_vx6=
z{tgS?)MTSnG|O~VuP!e!G!!-KW+^lAN-N$J9XJ(fYh`KRj@y3tfHlaKpq#wVLhtM6
zi};%Q$C8v;YW>10<;Upze9ipt<>gvIa<YbohRE!_J=>GxWAr1dj(Qztul`tSU3eQ6
zX+NH68hVF*_{39EqHfl<0DPwg=Ud2E@Y#UJ+Q~_)!;GKx0DQ-Zry;V{J*hz9Th_$j
zpnDXI@C^Fhy7JBWEZrGG0-PH7QQ$@TNPHZa?W(3t(=NQF`uG)~2a{7%_|`3V*E2Rk
zqqbH5JxcidzA}DbJrKEm<T3jGX+p0><(nc^KfN}=m;Z56{IwBsd-@ezSod#R?Ed>h
zEnVwL9_tx}(oJ2#Q|p|K(gLbYbdjC<*4?`FxYvz^_m8Hhrwg<Ta{l~znPc_CvYck+
z^V9tC?ZtI)p_b+OCnql8jc=^>CW7OAFG#zX5rJ|T#(A53a_HdYWPf@nOdH9p!V^;H
zWwX(^*QkVi=izZtG~i}EFDxUK;N0#jeZ6KkKVe21fz^LM1k3W3xR}`A@$p2ch>~oU
zZoR^{dzh$wH)rSX-@eI9N=p8C{RsS09(kRh{M^UKCyQj)RUe!+QUIe5uJ@Wsq&Z-{
z>>Ss<#z4FB4N+#W$zcQaAd>CXw;}g!kG{xxj>KM5Sr}wkM_7;;H%F7}7&TDc1Sera
zWgjFzu7^S9&s<Kg_))i$wsV-~j-5pI|5S0_XW(=@J3C{-uuR}x;Ly$Mb&c1;r49(2
ziZ$~;?bn({C#TUlGx7dRVGF|#lG}j$nG+ds@ClG&8F#tX6`7S$cUITb<lo-<{#P(+
z&*U8FCGf@P>eR}|#}Hh3TU#44J2p19kNyLQw^@1KvGMUXdJ{#%elz69dJ(HEyNH(6
zFhrJ0q3F%$<O4S=aBjX2H$jG*Phod9@odh|Z71H-rELvBvKZ*=>p!9l2M6@N^39L&
zZ4jn3Rh3m$I^f{8wzlvIOAL5$PB}iPb1saIj@~a+Bjnj5tHbIdgBclwXkrL@t520h
zfH+K0X6G(07sO0DR1t4ou<4onSi!o{$Bw^dE6!b|%N>k>Va_3NG;kP&iTg9?NlA~B
zSR7EZ)KARUJo?+~FI<jkG#oSYN8dv^V`wHy1|D0l|K2*H4}ADJy48%Eo|%7zctNV^
zFiq6yt=h%>c+b`F`^vwQlQb()9^cfjKUiC<fIH*8w$S&~|LpS8GnfT#w5B*EBzTNR
z0~<{8f)r^N<ZBmT$rO!mgMb6?Q?|=&jV|z^qk2=ypIq9=;NPo`N#rC!BK~_hJ|&Rk
zTr0Q|QW|+Yz_!YMqi&Tlcl~9*-Xi&8wxxyYM_rvhh)Ix#Adp@}(GMLQltwX0XS|YY
z1l0-7Be$cr?kHMRQetglBD1}{y|uA{E2sG=;iJ7t`eyj!ka@V#O44lg^Wc1u>KcL_
zf+*%8dg_@E>*D+)E*24nm1bp}O5ipG6m`tIK$epGZ~3vMs*XpJ$!+cn*XVLvu7x#<
z##tB?7QvYzVYlbrnSz`}C1E8&<%~E|I4^KK5F*E2G~PB4`e(C2uB_WX_Z(OIx?6BB
z4(*ALT3tZxFi@9H=8b6Vr7~aen!~wdsGHcU;U7>!et-Q4i+Q<r;+y}u4G1if{5M6|
z!*RW<9~}ZMhK2o#=dbGcdiEu;lXEvw%9kVi{}`yax#(V)$$}~_Bqa2AbTmgUD-&cl
zgNVHhIXU@Mm7Q^EBPUaHVCB&Sc6byi3_CS7wFL<k0JWrAV1|5g?RIag8l&2}jE>bY
z)4nQY5Zb&D(K;C|O~CG89j*a0;-l3O>r^S^)#fao{NO|XZkF@BoS~k2du&WQi5Exr
zlE*@pu%lUq&sLbbzm?5`kYp#l@#5Tkd1s;9$6(X2h6WYlWFNPEQY$7d9>{%Z@IVVv
zMHiR^<@cKz*Aqo`Z8JopgMA=Fve$br3(_Ps7CRws+GrrP5kg0T_)YwY$8DxbgA-jz
zK=lB3-Nb~sFQ?+qM`2;%lckRE2u9K7F2>(R(<EjbFD0zX5;~thO6+|^f&WHNZ)oV-
zi<6-jOjh>x8lbR)TAZGqZrW<T-Se?bgELD14drmMa>1x({%icqjpA>QU-gxm$)0vN
zb+%H6Nw(n6oH%#LaHhSi{U=7gXa~ZZnzX7`ijq34d~xD_xS>9dNCuXc8&!f-ZEw5#
zMI$^?mqrh6EI#IBZT*-$oSldAjHQ_NKr{k6w%q4mHs4003_6sF)}sUgpJVsZURar6
zqR(Y?Rjcj{rwN}=k>T3;jDFqh*vyQ9yEeVLr9oqV<1d4{%F2Yzyb^1|pQXBIoab~A
zQkI?l-sZV)K+Bk)FB!>G-);{hH<gODhzOzk+wjYck(mKg;%W^dCy}SY`Od}VzpiMe
z@en3o>Q5;h2W$%ElZ1B$-RZLYW&K~%@WeDc>xnuPICDP8Ztm>#Rlnk7`@(0bJv_~f
zB{H_{vU+r(N*}(>uwsDgR=#yCY+1rtd*&dFWqNxvSC{~I_AAEOxTn1oP^7KGz`(Fr
z<L_U$QHUy66;|%QRn9>w=V+cYWSZUT8l&ulm5C!9TD#)dYi3+E+XLptKRWO+_0j#_
z;}#L(E)-rqD#Lhqy|=xQ8E)#L@JHf<tu6L42Nx{6ESw>b>8A!rYY#}G-uO7RjMYYN
zNf4b7>TEWN*~yTT!ujIg>cNh7L4s)l|ADW**E*C8`fvV%DF}lZngvJnV?MXff^om%
zQ37#6qp&5nW<k-<{f^Pi!F7>fxp;kT&9d{qLx|4~V|ge~caR)QsZldc?{2a5fd3E%
zg>5BGI4q!?X5YN;v*hU-<o>|7>?(nLHcW^{Cn24iM#OR`v3Ep}l;>T;+ipqByGFVP
z8C{67*h80f@fp<JW?bN-?G+v_ne~W~uC+x{94g^qYHr~QMwpA9+b_B}$XXq!Cyo1J
zq`lt*wFm_Z%LAKw%X)Fl5A7qmqz#*9+K%|52CDq*Kz)_I*;Jp$cyB!aKR4Nmaouro
zW&KT2F6i-*k&##hw8pHvrF?EL=`#%+4D7Exxf9Co>slDRk!N5G-&A6t(srxTZP)Ft
zo-1-T(C<773+vqb@%Z!ODvl2tF}gj}tS(WE7&04B>^1yOOit*AR9>02x_3`Yd&1}2
ztkA-^c%sfOenqChuCj~Niny}K1aWL8;V404DmlDHwK;X+8EOH<hyL1huDV0l+)bO`
zdYq1nGRoDV>j$bh+MNt_=f|@|iCr3o#e8GMOveMWr@Ec7L+-ZA7tdZ$6Py9el!8Ky
z_r!6)JNb$AcXAqzAe%P0&l7(h(~@#hhSSo};LXd^7-pFVFT}kp6mb1;KDq1(pcN-y
z_<Bbjw^*w>iqmM#35Nnl(7$`XW1uul@a5>_WHQu`gk5CA-EyFpUBYtR31h-5)L)lf
z>46+>KdDa?1EEQ2F?FM@L7Ra`fV<6#!BBm6_x*XOzZAa^2~o2M4a{lXl!Pz={-1A`
zj0Z}hto8aSZR7&7u9q6_lvDxx2v=TX@>ur7Q9<05*|ZP1ecSQhJnA<PA?;Y?af=)h
z9mm4i6v5tLVQ%H<;}`4F)k_tz4NE(8O34*--q6@kv0*!#n)C3VdC8w-Kd`5)>F3vk
zjg9@9C-J|Mad_B$H}zzDoM3R1L3ndfx(X{!z53!>m!5z2)s9brG8F+Ob|@SudR%G3
zW6z^1g-xNY`F-EHYA#_6nt@ujKDzMEbx)rqW2EX`b)1@+iO(Lyxg^G|`C7|jRaMn1
zhzo7+xG}m4eh}ZgN*L!Lw1V9_tBQlj_(Oj+)0%zlhr_0BD@x6;EyeL=^Yeh(lgo|m
z`qY2}_uSr>HtjT20?+0Rq|F#)izT(YFZVj<>Pr~EILZ^4bcIojb+vS#{AqYVB0V&q
zYn&!iT_wj!L=#o;XvPACLXm{&6TYTzz5I3B4`b72D0Nyq8BPGmw56q`R6{k512c)@
z*B>;pf?A|9|JE!RZEb+k-Nwcr4x~S=&cRAb`D1C~wRbuCzEU-|`te(O*6i<JusH`I
zOW>A|jF3;zwD!$F<*2_Xe2X<L$cVSEVqASONV+GIx!<~9EY!*fi-c<kzwRUWy%Z61
zax#(05$?_?@ONk^HkRXZTc0mqF_cKa3A34nP6>&Lg}HM<Jbo@Q>~@B2(8B)_^bt_%
zXqeX!!oT-gABcM^m|+M&x>p=UU7VfY>i#JAOEkZXhz)OJKD_=G!nkzs6acM%e^Nxm
zPx&FAGwd><ryHaYBH5Viva&ML;KO~<>6=_|4%zW8P7)bG8vL=P<P(FSyIKz4(||M+
z<d)gYOdD!a;sh^53{rJG5hK_v48A@0wWok$o!DgVe`MaqFE{gEb-XD0y65Uum!z$^
zhCoTXZo2mkM-uoviz(e&bri)dy6dNJchDnY)tK1sd4SrM%OYJHkIaIRnxC9F`mZ%I
zPT89{F0+5wzqbFiqQrPfX=m&-(5_-_BFrdcAKkIup(p6ao1iQ^+3>bcwoNau-2H9W
ztSOGgzI@+7b-($FEgM&v<FAcBE+f#9$Pazm=`LNWf7e+plOPOYL9u2#{WjCPzoHl=
zX^#in5MCdosr^^{P%;lO!n@bwN9?<wOQA^QcH_wP?UkU^^r5MLa%*dAKgzWeMs&#4
zT+>ou<^-@@z$FRP)9|tVGhJL=m8bF>R7YNeThS8qkp<EmalhGARns**>9IiB^UH2E
z9A4cuFFq`3>St4v*rE5`a$iW7ZV)^6edDmKX4=b3{82HzODIxT2r2AuhSC-vo@WK(
zV-L~r=#P7+8+w@f*oK-1uCkC_V)m|-A}840L@kJu?QmH46JgOI8ZmHnv0As)Lv;>J
zMp{SVNt&;85BA}A;?{37W?kGYH^R6Y!`04jVw_gGPJ(MpeI@?>(;;St;^ytnk-E+(
zKP(=4_s+-X=1F3jS7%s|$ijB>Bd#>pkG(r%)YzRT)x-T6+3A;xRVl~$MkNTfoW+{s
z5b`kk2@v+BX>7j3f}@?C3JSP}vajebvE!h9-gZ1J5JXZ$;7&bUGNp^tT&EMYYce)8
zbU0Qka)mJn!8l_~Qh9H>3I8KcbS=-AX83bNS<~_=zg#%waTcsT5c)e5Is-lu96EC(
zyW{CJGq6%!qOp?QayPMOCgZwnd~B#OGrBxkqmixGt;ewh>W-ar^k>fM11`k!`}~%|
zk3P9li<@ItGr*ScBL!bpR#e3E)6Zkk=P*HJ9?;wJOEyQJ9E3e%2Im@36QVB_AB9TK
z(k>W<fWqLP$RR@NZ&{u^`4GEz@q(E9+L)7~bxc<Mp-0+?o_(6B3l+XFeb%9BE}S+=
zg(tDM{i#FV(lC+P$I3TPsF^d?Bf@-HHh7Y7`CE67`fm8ifKU-<YBE?{&uE;Tg$rT!
z?=?6|D=I1?l<}I6&CxEnzx2gu=FicR`vdG2i-;?7?3Or_Cx*A)Ql|D9-vetd)E<rF
z3ICHS*^8>fS<E2viG*)Nq&vjbA6>DcB4*{a9Hf1U4I4L7y)gK%T>YQtCdsSc^5bmu
zOijHR<l=3ZvIYnQ+4-PjPbl1wn^JzLs*7mWvy+{;_?bTustnQ?Q%p*11lStp7?d9&
ziW>rdhI8v$$Dm7MOX2W3+Io?Pwb_U_7^x)g@QAyS4ktcl5VMC1QOJ1b{Cx2f8_u)%
zV(kKpPti@{@5D;ss6Fhu$I<*@B`%pkhZhoX-_Ca3naozQn@jMLYppHgvdxkZ+MCbT
zI|ePuFLW%|dgbq_VN$~e)~BkU_0iOqx6HOap>`s1;3Ki+mT+Eq+g{@2HcKS+hn$62
zaU;l!NeW^6w?W;YcOq=$r*A~y;UVu!t&_7egd^!Z7V~@$%&DokW>85F3znH}L$B(Y
zYE@&s5^;|u&$|u2zDHL6F5^!E#Y8IrSsEJBv0YeD7K=kkZ1S!;cb;PSM=m9B4(-O&
zsfb<DH*Qr{jas1|tBg`Y(fghf%9;u-v(ygimz$N=#Rq~J+j|!Dy-p7^a;;1(ImyR?
zNP;f7GIM$r6;A)imRFk1+@8|N)2q95xUnwTMElLaU{rUJC(enp=5~9hwFRMlb!qN+
z30(>4$zNG*H8$#^!>;PWcO#;2A!bH+vM95IZ~F$zwL7L9h->4#JBN8m)2*y9OQ)+H
z<yN}B8>4A)k{CwvV`j%fcQ*b~WN{2*)j<z}Eb9z#Z@0Y}Jd5XPJ)tp&(r26$_&>A@
z(9}J=&Fx}7sJX<HMUrg+68zKI#l`BwhbQLjIqB8N!1}u42%JQkfOfukW%ds?Hs9;&
z#Bmb|?hialP|kW+S6&Wp>o(B9V-PVZBW}k^mKVn74CWBlaK)i1ly*`%ih+I{m=Q?q
z$CPyKy1qJ3o}PH=5_6!NI#3fB5vaorJ8oMJAQpnWN{c-ab+DxIQepIaHpln7FNAJM
zWO2Ev+%;pxS$>@m;LqSdx3I-au@2AOJ%RD_8$<Q#s74QI7I?<7{CFJOPyh_)cw^{W
zpX4b%a$iD5d2%PI?un`f5ZsK7=k>$a#|(ZmI{(HS)};J|lJ?68Gk15WgtJpTP$DZT
zEUd3|=Z|snbE`4HyO3`5wX))Z9UGjv;8T|`>r!hF>v-bC%yKsp9NGGr&twD~i%t=-
zUGF=P&Q698blFN@NnBowf18QS5jp5UUFjMGFP%72l7~>}UQKkkYDCILQNNMU?xZ>E
zMQvK|I8u2%X|JWx*_o~GTI}qRC$LR3^ff3pw`)oME{r2ya=0S($;CUZ*z0(7p9&_r
z-d4-<Ke;;0G4FRa;8c@VTUQ6ikk5UZyURcEN^tIdB?dV)H8t*D0M16v&M!F~TkfCb
z_YfbBd!OXqML}x#7Z4MLRN`LVoIC;cRr3k`0ipBeS+10uPE1`%^w;h}XEs0{<!iBB
zK6&cw|If>uRO1~E8wy9~k@4fPuBLSb&b_8+Duw#gA^SuzR}S--FY$Wg?ou;x_RDm=
zwbHD&ljDB^Mpi!~lqSc;MJC!8zs<g#zgj(7RES)^=)X^0Y|JE_^usP;cXyY_!b@Vh
zYi%39j3pU!S)t44kDJ%6?QS}bVwFrk#5>k3DEpY*)i9Z2;N9ZqwWACBDdHaJryLMs
z<sTkdzTQ1`MFzn%#Kpzi7Hc8Ywh{xodwXat?kqdnCe9fO-bWT;Fm)cR+K%IyqWLd*
zCq9?nUbpYzY`A+o#77D}pY^Rj(N)bwe{r&JzVGOt82^fA4UhfcPVaFd4%gMc(plZd
zsG2LR0!Ndu<?nk%5N;zQ@^X%heddj`)5K8G5sMrCkGBofua|#gzLm@v0C*R$M}CR^
zQ-;4cN3#;dyk)YMhK=X8g>>D?bOw#<wY!egIu6xPHv@a@oh;1b6rh@I>9fTCFEc8B
zcPRIO&nT3>>k+BkLnPK|ADdh!9QVD>qx(0DOt)D8S^&T+;G)l_)E&`z!^+dO9^Z3*
z;dPvTPUIl{JDx2l2&h%2c?5-21OzgG18Y#WO_S8s)*fG;Y(sAS<#ROg{ClrSSJt!F
z=H})O>MFj)QzVcNytRco$fToax<Gq&H~Lm~^O+E-fvl*l{gIW$!rtSWub!hh1iluU
z2UKb+z|61C_DS-h+cY38#l<{8XB@;)vg6V!7{x!y1A;UVt^i|wV`M};zielfvIW#y
zT3TBCK39-L6z*{p|7!IJEx?n<L7IWup|74HG1T3ysJH=o9oDb;_my&ntTwnZ3$&+{
zS2Xbj1qHG^Dmh3-3HQuE91?5>$ae|Ko}M*&?zG%#m$3LM=+*|!q?=ov;Q$ab<xenw
zVoZqEz2)w{1GRhNGCb`-f~(Fc7<Kd9@ml=^=U+!dL*Y(c1rw%>XFUtHa$&SS^%hCP
zN5+NNpJhJ`coI!rZHquoJxciOdw1q1%}4j?GvQW&P6>v|<sMb@uh)-)c5V0s$SAL)
zm_6_8@5=r`WNQqFGr76=FfqeaxxyHdp4yBOv_J*?Dq+jZcASlD=n|=!k4c*2o-vTT
zKRWuKi<=wHbLJX(QU*%Q!Pc2qpR=;E4&J;<A~*_g;}a2~PGY)q4`f!wU=1kft~ZmZ
ztE+=Gb0~ey$Y2EA!s^{SMI?qfLhb763U1Q3543tZI=ZaCghEd&&V;L`#r5ZjnWPGf
zi;GQ~B`{Z*TYV;G+o>TwEA%AY+t8c0U%jPDDl4C!oStG&zSU<P*x1~}o3|3AH4UM?
zJYMP`0Q&kThx_<g;Hp2l2|nTF4H;))E;%MLj8uR6Z)LX9di)sScZ#4fbUqEe_0au%
z^^fLhZU5_Jaev<`-Hq`|pyQZ-;cW03EMPl1Ik9-3OKMBWDJ45)C5ZuRlf$i{8UxzS
z(>6^#O#Bq6sDFtr_j?~7N$#kwzt!qR+P!~YH~QX){auFQ`cwWc|MiQqb60<$G*im1
zpV$gQ_E~*42hM#O*|uX+RQ(yk*hgk!LSzC2*LUXTtUHco3^NTN1GY9dd-&+rhRCH$
zUrmT|r}-9^lpucn6Lo5_KENElysTxj_?_pS7v76w-F01?y7PSEwb;4o{r+=_U%%_?
z>NfX8#xVqsUrmKsYc?+J3Haq%-lt(XUSzk}c->$$zP0r}h9ODn5}U4z(^DkCUZuuh
z3_O-=_cqA`yLLU#66%TQPaER(Jq%ZM?_RDh)8AVOS(`uHhBzn={ZBWp5b>2Y$1pKw
zgh5}Mec9w5+K+Us4qD6bwgV?Y9{mqe-Q|{Tqd-IzEW#c!pvPA1Sw7iU-#uI6l#8cW
zt-qcbr2?C%MrO75WmP!Uiv5l-;ZH0bm@ou7yey5xZE()%14rTaM%+Zq@U?&dC`eF3
z$ndyh4SiBL)dY;P@%Fe(yI>QfrEI>rhuN&I3Uz#TbG(FvtTf0Vn0#P|2%ndOw89q$
z2q_Oop)m8v>}>j^;ffw2k*agi>z{+3q(}GV<ylU;&q4o`BB9y2L}wSAnacK=ME^^a
zF4Y)8I?|PplrbE>dj2P~Tk@!FM&#<Wl-5Z?P%wLSb#;M%r}8EUDxid9Mc&3{!;qRv
z(&{6}Ou#2JR=$j8YjAhkoL8!vfBT(H>6SprPfJr}*bkYzDLvg=ZYJycJF=Oa!_3US
z;M$Lu-Ze3N-V_Ho8?&>svL+Kakq?|ml#-7i9@t}3vyUlB7M4Yc@geyYUcr;g2#M75
z*vG`J1N(;r2@sQL;m%@SpJ%%FoR>np*PPU(!$eyPC7zX*mEDu+zCT9JZ&lk6GDhPc
z9U{6uiUJyJiE$G#4}2kd9m03>FR`hw8=0sKq?NF?^R%IsVoxuxz@hKj1uf>#ikUzb
zXxmXJgB{4BY#s0s6Brdm3^Hot%y<86PONK$+`+q3FFk@cu{Y?f?W(1}|F!9gt%-T&
znz#e9!-d>j);%x=%KuteD&rVqVu{vdCA<ckO)?@_KQh)u@5Dt;9f&Oz_4U`<WLwiy
zC2h>9$lG$kmraVu>=(lc;Mguw2X5;GU*6F*i6|lA$lKcNa-;(LjkFR{HP`C0Xy3%y
z#2Nn&!nP&ZM#rCTLYVw>rn0!QnjxD^@Cl|&_D*<^ndyR)lzdD>jlEX{=UnSr2v@rj
zE0qX}ZZEJ`MTiz^={tkpRp$+Pse=*mnIOOE^lw*7`O-O84<<tHTI~*4kBZbSgb;0A
zzUHN5-?jVp$8wGo3R6AQcjsNw{cl~9g=O$e7OrfwlZ0{9#&)+QNLNOM8JM+fZIa+}
z@&hR8aVlf)dj{@=HSnj15C`mO^ImG#{U$HA>cyRhau2Uhm?6ksHtd)(aq{_ziSG7G
zL!%3xBekzw4@D)$w^<F>Zk#zFv51=Wq~29Q5s~jeO%6G;-v6J-dpUxoL}2G@1nlZx
z+4A!T3DPmgUBG>2n^i}yWMoIP-&|Bl_7;VVcr}NbNKG2xh<yqxKdpDh$a;}asylBL
zK*~I!D|xw!A1WhuJ_tcl@fu-<tL*SdJrjC4!d%GNe|vKUA1j%)4Y96P@B6&TA+yTL
z)Nb(qW&sYBxxDc4&5sP_`{*!M@raqKvYA(QxE(2}O7!d7x)Nf~Sqll<ur!p#c^@6Q
zJg`})dz9qln}^|FV^Sh%9c5M)+?beJ=`6HQNX6HTTTPdoJ2~RYhXfu0O=9#RP)$G$
ze`~Z2CGYfyz5~7q%gRCmE;V~+C!Q8@u0QE^c)(kR&;av;nlWft4Yc~`BAsg4^shww
zstLZ+$!@`MuRO!8&%StVv!{}sS85?N>&af(_-P_ppek~uI-7*e=fNfi@|IHx>8CTi
zK4qEQJB2-9JUO1RkqZ59c)DYBxZhRoA)dXx&hVXRrYNgSv!3pfld9Ryn;|A{F_Yxn
zd;nMj$;uZPyXR+S{w6%u0TyiD;h_UCkI}7>CH7~YwpacOFt2d96m)rY1-m*-y0;gP
z$WlrXq08L28{_U%s(kYw0)ZfPMdX)u01oaj$|uu5OdF|yE32^;kSShjhG~`<<?(W4
zms{ct(?4w%`Vf*7KbuOTf<Xw>McmYtqYFqpx^Xr!8rf5kuB{xJ8N9pP?d;WiXq_;y
z4lj{jup3tU`iJ~s_XAUkkxCijm=`2PgA!`}^TtiS3~M;QRIig9JR*h$^j;#_;<Naf
zy6ITx_VDhtfcJGgeG+@vv8pN+62XG~B7)Adbb5~wM-y@F^`S{f=AqK?lTE+4V^pqH
zZ3<hDbo{YfB}3wM9r9LR>2A@jetm%LMpaWGfUN=Ri!Q?8Cy>0}kLyyq)k%b}ehk7m
zKU5=*?`NKUx>&<V73V6j%so|(7&r!Y(Pn?TxM_>`3t0H>A+Oas=gq~1$83|IF_Yi5
zL#W#&-t4eiG7<CoA@$aXw_6~bn3`96EO3@HFpw{^%?-3G<F^K={(d#XT4yx9CHFF~
zXOxGfT~mt+3_FNGsip(SI}}8LzZm?-9Md#LO5Mm2Yul8_9(F@}8NjmQcI#o*LQ~(>
zHS;~Q+XySRcU2CXW-`zULjoyhM&QAk(c{?y(J%S}dJNE6vXhf(e85G^+D`5MDmHE^
zCg)IltJWZk;G5dOW_zZI{kKugRl;bCUwsdU?sVe2mnE_H^${uQxJOoX;(h)uU&{3}
z>=-2^MI}+xok~jX-Bl`7vTEQCN9k@o)flBLW@e1``t2$OU;6QEt|MP_Q;vw7hVqFi
zdSDZsE)`dFsizIOYovxb**oT9A2}q-OgX$XFxATh!w0;L=W<PlbWJQuZ}j!GK+7L3
z(5VEbI*HV|%y@}03FNp1hiT(?!sD!m`!h98Tf4iXW#(PZCtITttG@6qC_O|BL4sRK
zK=pf#hbRoAgl5Hyn7aShCV~I?VhGcFGz_jD-#&?P_iqqWkv&i5pkN~X<Z*SBpZe)u
zoyu-s=d738k>U@<VGfk{R{)|0t6ExQ(?y*gmev4^UtQ{g&7b1zu1DSXt6AB1;H@_+
zD(#3eo4-}ks9EYH3=?hff2Fm_ucDsh!HGM0LdHS(wAd|6W3bWZ3h47E$D3LVqK<6c
z1dN|T$~Kw;GvFr^qn5J<rwkRku+GIE)?$}rC&}Y^!?}iIj#Dzp;(6PuO3)N<6`0P3
zebc#JA8h)?JBz;@X|nTU@*qdWXZxuPM+!^dg=PDtvGvT4yCT=Jv(?WoB_7KEu65V_
z@F^P4;mxiCE6zO~UmTjvU7t42`#XW|<an_YyH=HNJkJlZjOyL9m~J*+K)vyUK1$tP
zI}wqP6lhG$MLJ&mv-s2B{{tAQV5|kqU~KL0j{~yM8MM8zF?f7j5k)Ubwp5Ipsc+vI
zjM0Hd8$XIQ(i;Sc0JjlPF+BO|R9xK#-<vm)3i2gJ4VpZOBfozuCwcL5C%T@l{8T&a
zd9nAkBe`47)T>L$8=K9#&qo4xKLb{o525N@SdNwkV&Cx1)mb1vyS!1s=<zbsoruJK
zIBz5Fwz#k`C|ZZjG1B0AZ`SN?En~Q5{8nHJ%7nBB$K`C$wm*OVK;|@ENy<VV@M$on
z9KO-Tzo|I|5#xDzoG*2?180!-HE?orx;$TwmLLAE!MQP6j;QrID+V?1si-KIRTt~$
z$h;8Lj12TEwVCkWh=Bb|uY|O0dqdTj8JTZw#~sPvIY{$8TwH4tPq+EHQHqytY=)qP
z6$txXi6d{pP*!5at{G@<OieTge*i29%hsRj_rEK0CZVyh;+Q2d;fEIEO4T^n3-RPC
zK<!^-?<ia+Dg7&lGtbwwF1gq5_3}2o2G%aXig3uU#>V1sD&8N!(MI`RT4|&Jc8I<=
zL+>`eMN19;Nfnl_A<ZFl;3iu3&xBA!r0;%?AQ}JS<@m?eK?5@qk!wY3U<0uQ{86Xa
zU}$XYi|@ZUlp-=i+OIexBLh;1^8Oe0zB0x4%*{`DCUxE3kAfgZ;hxmZvDVFTp7b9G
zMrF!h?^>P3KG`YLhhX@G#Y-vTZ+R!h>o^g!#>WYtA0_<O@4?LYk0((c>InGdG--M|
za~?#mOqZ~DYi1@5{e!?IH+WxYq5*nyvrLS&f8bs0tLS}?q;6E?;Tl|cr|=FulgFSs
zhW`gJhoc!KhrxRXnf(~N)2+cQ`C6Z=UsBfxZ6@@M*Z)))#GDmzNEqWQ7{=VM&&w2E
zya4H#vprs74OklZg7@R>Wqu=>KKKbE)4HF!J-_AIrFVa_9b$fFK5WcrswA%C_Kt10
zQ|s=SuJP!M?WSS>JS{@gbC7aOH~+OjMaI6;Z{0`H!aoNy;vFt`BPdk%Z)8j*`&*9C
zbhs^q#M<3mq)mfMzSiR?S3yC+d1H_Tlx}p<2fPqX)A!6k$S*B@1((LorR(PZ+pUsM
z8}*G%Nlu$-klAB-N7OZ+M*8r9Kg6=5a%5x#RJ@TKl|+G%QcWa`)EiXj%YVJRF56>I
zA|oQuN*_ELS0Zx(k3`^x@hC`Cq}+@sn}PkfhBL8uu^FQ{_E(ivabY3mLV$&P+HZNv
z{x6EEBJUdA59I*<B5AIuuKo_jk5o^oEl+wlQ19H`b(>D~n}E2T!}=m>qcf7O)_GHN
zqUt3^7=n)xN5ePR#qPt0peKQ)E@MgiBqSu(uCAv=hB}OOwnC|5_6z>R^dh8d7vL75
z>jDt6CnT{wYMj=iq8P+;!IT`}u`ge~+yrTrlAew((io&)0~sGSw}nksY}b~Y*xs@F
z5trUNfe=w*Q$O|5zmna_K2%EhS%&BDubZ9dA_8wsU%qlMz&SWXWG3x=ZQgx0Q7=WX
zTH||D3y!qS*NCM1TQ!r<@hj39U<3kd48ntZ4aPST*rxSQ_vRW>9b}7*8Xj38=AV8^
zx;3u0XK8!0nTnW8f9fvKzTUi?RGI<a{9~QsjV|e12L;8NsOHARx->~IZJ?aQsY?wF
z4#sqLbp<Xuxa9%I{q!N=6mOLcb>WD}ZW($VODT;|K<rjNJ%x5_sE)yy11!ithLS(d
zAO(~jeU^&u26e74(TAS}jYb(Z`5IcKV0ZRNeB0>zmiV}%mYHiO=hQ~P5#M~pE&|L5
zy|=Vf08?V<;{?F$81yU4iHXgTJeRA}4?r_`Z)PS3E)3K#aG?LGt49k73qJ+uIwvQG
zd7bqBuYO8=EOb(+sHmWe95KTbmu{HL17bhsk-eUp!qT<w#{%6me70v|bdkZ}=4A#2
z1x3i-{5lPUC_W5Ta%$$Lz-Q@SrTnCH+FVGG+EpwjCMKXioAf%qbJ?xBa@SQ{OJ&@~
zmdT*3vN9);*BD5^hWzz*#B2Tip)OM1=jd@6R#sLqm#x>`#jOu>^YWhZ^WXb~9D65N
zQBy-R5m6^>D{9&U!Iq)bw{9lR)H5*n3&OpXKGrnv4kUb)?d<Txgsurc5V62yo0#KD
zC>oy_HT+myT;$`RW+qezLpOLa#y49!w2X|z3!vIOm6FPM#jShaU$KRMnVm;DP3`xW
zNkx2%=bS7k4@Z*c@-&hc=*GLy0t3OwmDsxpj_=KzH^1ABO{2gp=K26i2VhzI2I;Gv
z^5#aq_G?vo{-{b1>71SZhw2YHO_A>_e*k#&Np}xqyvxR5Rw@5i&HQ)o-z&r6@XO;7
zb(hWI9KIKoZ(g45&yV9Ud1nHYvOQL$Ps_wKo(xCG`5DXudio!zC`M6r)x`xH@H1H{
zz6aNR=jo!JVd`n!?e9)cPq!|PV&vzvfD1IdDG;|$*rD<=?fD{)`_2#ztBtZ}on>E<
zVDrL_jP3>T&V!d)>5^VTZ*6NB18^vO1}erGYrPtPVn{&6qd3)5>b0o>WY_^$)5UBQ
zxO(H;e~Fj7^S8UIr*KNe%a*5oPq5pn-jne<uX^2H*JMa|eC<u-eVo{9Wojym7y#x~
zqw8*pT80EoD>Go;>L{sE(DboCL3GF%!^6Wp!S{gf-r3nX_gg{?7orhg3BjetdJ7xd
zSPUyFdg61rnJ3$rW(cY8r#BV&1WCFAxG6UF9%#r`Ab?4jr2c~HpZm+s)m2*wxdCnm
zO|Tazah1~_Kuf+ZY7Z%?lQy0b6;uD+xRx-VlWwTjJ=510fd(5AA5{{0tXy4%;{U6O
z74tqX-J7kWS~Ih;QMI<VUOc{?n=@ip`da6<zx6j?3vH%!(x$h_w5txZ8AX(jPR-q`
ztgcQP@)_4}a*MdvsBk=a$3##6x38~HpZe(KF3yp*yj53@*}a_f2yTp(RUysO51##)
z5Y>kxXLbL$Z?hu$G1;+%pyv=*e`*JkPvZ$B9qgh3);~cyqzc<%hpVE%NFpUAWvg4m
zEtyQb>rAm$o|=`nw?uo7hRBkP6w03)HKd*)Yh!I|t0pfm4<%)k_ALacAD}O3d_Qv|
zJVZof<mOz1mnn%S=={J$`M3l{exkQuW&$n8v8K>zy&r7`k#junNsDx^A##fBWvDZT
z(DP4%0o}^V%05#nTaC1u3nxjd7yq__0aNR=t|s#L)4t#B#X2yUYTXabV#d*T>2wkS
zhCBX(>fL5ba@pY{THH(iv@gBxn>Ur{19G+N%hTQI#iXb8K+teLKUh5A$=%$vnyq$_
z2hLSAlXQRm*DTHajf<mo5a%Q-D}Bju^mzwIK7IBpRwr6qpeZ9+9m%Fevgx^I@rt5i
zKbQcV=4T!zo?eqKqbXpENxA!;g|Vs-C@xU`{BHF1W9r(|x12;6n*;KY^!#Sk>rdo+
z&ok*CzbVo_`M26j`l5}<`O8AGMQ4ZJXFf6N;>Wci<bZP8w~_jT8V;sxl{~LPy*n|E
za+xxUu=r@^Bk6>lug?z$QC??zXuZvpAQka-Q`jLtKc7p|?^c5I;BdLCvbL6X?@v~B
z1ROQ6eSL8>H8rLA;zh{i<t0$yX$6)t{12+L139r<l5y}u?`>{v4Ff4&F}O5p2gmJs
zU=V*3P(Q)_?G8jtf;=fgN9TN_513W0^*s3jg5e<thrAq^skZX;e1mjeYz^wegR}kB
zCHYyx$a-27R9MEbTWnDC)q1=*$VmnaFaY&9ZIZv8{);kux|Giwz>&p#E-Tc)OD81U
zkBo`}E#N`i%ZH7TvhmrVDq&uHsG0_fD=-LxjT$`fXRNYtXM{^=v8gBPr<sS+d^h(u
z|HQ#$u8!lemJ9|caH*!XE+809<#qld58{+E=iM0EX;x=HPW4liY4>B|ZfMQ5NWVXp
z&{9>xFNWVFxE4NIAgunPoMX?(#@u=S;>C-=uOLE|rnum>4$Hy+F6tjXd?<jIC+@cQ
z&e2iet-?Lh&Ih~Iro^57QKT}i+`w-I1GU>bljX8pJ%cc6PWCTe_V#?_ogyw<@gPy;
z9*$XKy_yOL2*{K{R@lu;J>lm5{-y|x^|IKoPZzUr6UpOgf=mr4&dpUD?)a?L#oU9L
zSi>dSm0&V^TcZWCjg|-mfb02eO2-%Lr~&|w0lf5h-1|qu^jRAJh0?A+k%e`)WDiiL
zb3h<qX>0?@-D!1u(nd(09#uof$k_ksduWY#I8HM^BhO<{!$}ly8N$maAPsg_$7P<#
z;PnGZDI9J+Ekg5@To&2RDUpI%VAj)Q$NLmyRcJj?{|SSh>;`4r<4){NKKrXUDfb!@
zv+h1a1_^%Zc@cu@Tyo&W{M{RxoeueBmk&$RXfA2IR0heP#ak`TNxBoV(zmvj3yLtA
z+SS_07hR2L<5vQK1PYW?YnXiZN?PBMP9#$FQ%iusFNn-q>7)&I$5#|mj=dRmcY#6>
z<1*6Izcw~Tffx9>EhJ^66Vv-)d=J3IjqX@hbO+s=Zz=(!_5RKCVSBbuw?k960{n_O
z%=-<>I_p0})iDN-VMqq*Svc4&7zR`u&c6gq5}%V%9hQ3kMMsn7B;{$lxnpH4#7W}^
zijMX5j*gDf)QDEQ8MkpRnP`u*ubv32W1{gHdCqQ3|M=W|bQoSfn`o6Jg7llm+J0hq
zHp_527_*M4tM1&Xg!D47dAD%kWwkF6m|dg7wfnc%^rfL<SvQ|ZAy^Gb-gYBb{`r|t
z)NS_KUd{Kc{DI`igpH8P+1{*4P}a?!<rxsy1hS52;k$f<#STK3fE$2ZHUOvy6#M`W
zWpP`Uj_p!D#BiyJ^lw7ppQG=se0_}o79Wq}P!|LecnAp-w$M2k0VbXP7%N0eB<Xvt
z8$!&W365}VY^>tv&#z$bL7?LC<Ht1qV8~wU+tA>@e?LEWd<8vRe!fAhCGd`7@W^0|
zd`$64ZeVuB(Vp$Z*ZuEik$evtwIO-gszUcYVDyj?(DMMUe0$k@?Z~)>207ksnG$n1
zzzZym>m(AvI=5+QZ+;LCP!Dk6@Ekuo0NUWEt1~ut(k(BaA?rHnPzrq;qXfyHwAos0
z3sD}~O|$?VbQ7e~VmsGca`o<$pxuOiG6u-(tScoQ+$~yiSZ8b``ss|yl`JHi1462N
zyT?v(_!w&9?_jU{JEMlv!b*Z>{Cy=Ye&wv##j`G~(hA^4fd&<XC;{GImQAIr(NcS>
z6KZpFQ&p}jD9{1p5di_g8)M_~zkgqA*hgz!7_j*JIwvsEFKFveVSgMNdIg&8(9n>E
zzv@GGz=huOTm4x(Rm%eV58w-+BwMV#&=kRs#MlJ_Jea%q6c>4&R1HWPsnB85x4*~p
z$#nRiwzpj(`80KzoZ1>DW!ofb4`pZ^*kd{T6{mOv2iB{qb}p`*V$wD9vjO`J9`RUW
z5b$keQAiXi>?(rVv_2pf0=x-&z>f(^4Yi{p4AMqTMMZ_yxWdB1NHabM{8k@avfG0$
z{>gN!U)cj>M9lrbgs-1*`gz`x0F!SNxhOTv?Jiw!xI}kc@^B*oZ+^^|6tc;WQ+3XW
zI@D!oadewbdd6K*5jT4JM&Q#EU>aaWk(iy76M*#<0e;FkOPwO^r{dzNdL|@9g9xND
zXA8(@K&64M#@OllZ>2|7v<pmUPn<XaplCu3nu4XJC5FBF%a;unFHO~sVUU*3S_C+J
zSrMSEj<vTx+X#Udf+GY5E#?`XHpCxnq`b4TQUdx5@@JO^za8M1X}@4}0WIY`g@;G$
z12BIGVWC5PN5A^XX<|XvM5&bjK;+|KJmy19fxQ|K(V%sJy84#PQt*sNq?p*`a8}PY
z6YVyIiYbbwI%{N0M=G%Ly&kjl_BKGnGj!)XT<Tz!k06Kv7@e>)F3@L0aP<1x8=|)M
zc@Q7FqlKKLMP(k%>0vm;Z&z50_1`w~9~U1D6KBQ#EKDpV4Q?g5b@8$60|s(H&mMX%
z7q3(!FvOm@@NGldk0Z%=c1y-adFUBYN4%@6O=e8HfG)Sp6zuf6-v^KmIkG@nIq9+~
z6~UQSbSVOZ`7M<SVfPe2mArnmd49N@+|X;40(#+6ktRNM8luE>Kmztlb|KA+*}M34
z(b%ccOtjRcBG{Rzw}s@ac=;<AX^(&w2}mP+9Nqid1s!%b3Gm1$ATj_Hy)iT#1-w<@
zP(jMgEqouAU0qm@2wE55upzX!IEg3;M6&%E7{Fl-{n%<GnUUPPias5ac5fyTf%X#z
z?h&FNA><O-3&Fu&_C7%Bo0v!furAGX3BtQ7;ZNj}RyYsHFM7=)C<qHwW}w|K+ze7>
z+;4LILW(Z_U%q_#J3UP&1=H}^f$(m3iOkNvh}Pp+NfC7_flk^F!vg32XFLqgo@)3B
zr`#^Tay*^=aT05bc~$#ftN7E9FwgpqS!2vESf_Et!0<XbKQvdGjC)(Z(@w){ekek_
ze6~Z!P6O6n0A&FSRdI1~TYG!Hy)GDiW?z<i@R`C9{*p9Uum=$!nbFycE;PrRBeKjn
z=@WS+A$3<fDITXg$pG?$Z8H}Ds`@5_7_}6_d?YylzH$HPVU|9-+Byh5;EbmGPp47v
z<;lUKJU}a!6I?QTd)GiA2iikYn8#Jvv;I$^cds^^)A*!o40!4VTFF^$-D^tdtZvsm
z0B`bBX#|;r9<|7wzE5OsZXUqdoR;|IOKZcA(I0y=HQzwEhj<-&PU%<KZUV@eyu){d
zJ?JUW(?)u9bR?@FsrQNapSQH#@alcZLqx4w9goLE*Gc}YEs%sbBiwk=!!2BOqcIbh
zam&TO0>8!22CuSH;8G}u&{58}8M2dbnm_mPE3Rmu=JW&|0n~ihd98}PB?I5-ry-!R
z!I<ILIPVNPJR6F^IK&p9=8OKi6Q9pUs#q|5A;0eS`PC86N}~Y6L?0n&NnlehB_(Az
zUEHnpoQA0{)WupT6)hJ5iG(sf<IGZ=(_A7n8O&Y;?VRoG6ao){YAp~EtMoh;&$N1b
z&4HU@WZBjKQ3(%FxEw?*Z9y{w<k<6g<JB796qrgR)82D6f98Fgo0}_O&VG12s$(29
z!QZ^SuvfN3O)crsnj|T;vGABH2A7&)8|(~%!btIm4{`|5(CVrMkhg<G=@}Wd^Is1G
z<qb^QrVK@j@Gcxl2?^y+PEPU-^czVY2kx|{8B*$5?=3Dn$p?hLOiuTDWw36V+JK(p
zxcmw)V-pK)>WEv7Th_*kS@4q?a+7j5Xq%+qS5!(C?p!2O#8<vT_i3+FZf>zj2ZKHk
zVPN+lh(BwE_!cN-=n)jq;@G(;0++uM6A|U-<-Mq~n;*NruIDkT{|>MvW^hl^?1`BK
z?t|6_eaPG_l~o<_$Ur5gAlP667yUzUmaA2v)NZym2XuK$vsScncz%AaZyizDg=H;B
zi(V1{JuH{kw~$^L=Jm~QFINel<rB3GzYwR96bg+E*gkL=CJDPW88>><7(6dQBv1(K
zP(PY^mH0v_IqAEOf((T3O+=e4YZOD**-3RiovBrq{vu6pZ?CDZyDtC<zkh!QVJl#y
zjzHuE2L}sEN)~}F8yt;1<G)^>|M*{Oq2*?fe#w!9rE1c^hC((ANKVprv6;8E&%_<e
z7th8mTm=ROdR{t@JJq@YsP2E<!-Bg<_)#hDrz%P5VaFLEN?%aj(!K1_;e#q=ZUIOw
z-3b#lwT7{=vBsQkvm~~nG?=?<3B(DsZy~E0!y=?Kc$DD0HzSG`BzcYML0!^)m_Il@
zotoU`*}!O)#d!CxkJ(S&fkp=~9<&PtQ~(^X5GIrhMyUWJ;YysA$~#2cTJ@k}1Qaa>
zGe~;A9`3hVFHDs71Je?mbj%Bt;Zi8bV<k?C_JD7r?}IMYr09U|fp(9Rgl@c#H)5xD
zci&?Z;$+?dH4BIw>&fK7x_$zS?Ul=eVIfj6FH%C-A;e61bP|bAapJ<l@Fd+2n7a*t
z@ZZ|{aczBl^qrvd&O}+CgIRfbdGhd25c6u=i|H$*(3QxMUC`>E3JOxpakPn@Z^J>2
z_>!2|w%E@Ta`?vMI^h<2didt+HwVQI^GJUVM!KVon+br9C_clXd+#}v$cW(<KbICB
ze-3;e_cO%Rvj-O8zZSkaHUm$~NNO=W`xGk8ifvi-dG8Ta87VL}dd6Bcy=#9U3ralj
zK7cL=HpvKPYh4~ktb<cuoDhr_7abIk(s_5em4$mN?eQ-gs1u1wjsW>=v<4A~x$UJ%
zoiD(Em!tqxr|}>HDgmB5P*hwNngf{6jTF@gqgPZx)d7>7YuvJz?PoPO*87JABE-Ga
zNlv$}o5Sp2s;rqbUs;B7!cZ&!b?YLzfjR@1lhe)wKZCfdDnPjd?PJ;L=ap~~U4iL)
z;BeY>(^i!xg}p9+{TICsaKnF(?t|ALtOk36n?M`@%n|R*PAFB@$n>_G9vdtzgE4iO
z3Q!h*ul@w`9W166@q&x&a2m~=mdHw=o?F;~zHr-!SaJ&I*lh;ZC2x!YcEPobg{n$W
z3u<#*g4*7+NRd8zI8i>rW8#fM(kdmVQ>yZP#Efy*9!@6W{+5@=@eNag#57A#af<6h
z^KT{b{o|yJjDB3jl-w-bw6d~6uw<B4?09*+SxOKh=OO)cwhaqBbTphN;!8z^^KGY|
z?4wU3CrqK89LpDZ1yUE?JGYok``g(~$-#y;P!2(NBgX$NA~D3y1zP}cpwXKe)BK<H
z?4M^=OYfOR?*G48fOJip5!ToKnnI~B(Yh(<Naa{$G@wJmU@)`@3o^LvDCvcPP6l3_
zTeBA^!R`+i){lAz1bO8U83m(1`ZFX8z!hu(<p3@B0T2PE9g?8LYw2Y*I>A(A<IA<L
zmUh#C%_K_<2K+vz)vk9=x&P;15*@--FBrX@R>-W+D3>Xk8eyI`li0XdiC$n@*dXLk
zLTV&-eAN0*NJ@b}z3=VN8$`FEeXsNx^xgvm8RUDsq%EH8J3(52T!<{5Km@^}o1TBd
z6TkQrMf@wL{rM&)1J%x@d6zR&XJ<kGi;WFHs=wII)#n3H9<+aCIRCz*n6PsZBWQb-
zhZO}Tv$sKjz6Y%dsP}-D3q+9FGnZe2X0nN9ksklvR)>sJ|KoN`kc?kAe0WEdc*4wD
z`pV71u%(Cc2<-qo6%(WF=UI6fQQ8h<((N`9X~Lf|04RabUeH`S@?el~*C1h(7z7h+
z7|isNo`P}72<`})W<Z*OTpSOyraU@fTTIaqJulo4{sZB?7irGGp8y5ZR0qFP82kO6
zdQ*(vmMpu)SZicvT?w#v{_aj!$wwe{XvvLtcXs{)sTHkLRn>n<!fbbtALmyJpi%kl
z?&Zt7@_@Ldz3di@c}ih#QnvcJ>JV69E^t;nzn4r(L7g~Z{WzWLvRp!u1m>*a`p^SZ
z%s+j7pUR0nzEs5JAGza~HG!%4_ars*&CULRW}}HskUTa$ci4Xzm`3`#pvZ!E*&GFt
zprVu+&`T)at{yO07Y_xit1gSJSVhIfXq+I6hXNEo!|QZsvRMYtT%<S7RgL5Un{0QB
z^%#9Oe?%i3X5nVi)m@{qj=v{fg6?Vnj6r7JX4Rk`=V#js^n*Bi%}H?#v@q3&i3cB1
zF)NVWV!_~C4NgqfkKFFTX)7!;a*~Df)?jb7p`ih7Wum<cBX{c7<;{$W`0Sh<_6E<B
zM}1VLB+*1)>8j^JRYyOIm{H<;W202Mh{J;!)UEsjSCz~b-(?Nj$V!*C1u(IoOZYLQ
zFeVG^2m^$MP#mv9j~UJixF0;V6!*va(szH_&<sKA|E(f!bJcf-?$JQ&0ciAc;gE{y
z4*)kK%)<8G$0@O6uXd?J+kthswY@z8I5n`H?@=_oDGG+cOj^MOLubIgFG#2>ZB74m
zxOBgqB>6Ut+gd8W=exF`8t`8$0RM|G7%iTl4V3}lRkS3m-|nfg4Eoa*q8F@xjvtfA
zuO{0>e0rQ-=Y|wum!dD~Y@qqJKXkI|(->8353Chn8;k*`>S1T-R{@3k`-HNf`=6h?
z^9=&7KrcYZxAb|$+q?0hl=fxPfF0_Y-__}^hHP`187^2v0*ZJ(z@0(o)&-S@G_6^!
z#M#9m)Ff|nCTD#8{wyt-+0Q~PJ2wQBQHC!VDw1HG%bgV8TFxbvg8MsF1zju6A-CwA
zEjFUX?5r8wnkISbZCXU_j0-Dvm%~Bb{Zvg9B~HXW!6jP8!QA%Xvx8x`TQ{LKS#Qip
zZSbEtjEuzPM}PZ;d?RMXVY_*#iSyzT&!5%cHi`k~5tSe${f*WH$<fOA*K92<twp%-
z!~Rf-p_T!jZfc^nzzbLPfa($im*G8KdBXc;Qi<Qr2Qk)~$;BJj$IbtC%ctZn`b^k*
zSCiZvV3PeGn!W@Y>ivJ8WZ#!E!k~mAdm%Fj#fYec${Iq-8b)JZvlj~4Z<e%JqU`%J
zLWqRy#u^!0#_s=`@9+H2x%ZxPD;b~9`~6y;*YiBD&+4Q7cjKsps-qWDs|Qw2C~mtK
zk&g#V#o#(#cAeIIXFACx7BCOx?GzKm5${IRGo=KtouC&p8{R;EI{rFxzv9DkiKbn^
zXgIE_Fw5NuyRT9hhiWE#%xKIzVyS=*`$+)MnF4bzVlH65%j>D?#M@-%9Np=2eNw;s
zuPE1{D(!LHp@8--8p9iZcmUkGA%eTGZE3FwT7o~SGWDu@BPuwmmyPwmut@3+#hZ^)
z42+BsYCPf$!Rr+~#|SI(6oeH%2}Vo8zc1Z~ee&PajuV)aJ&hNvyJph7tJl4wS+oD+
zqM5Q~Hl@qY&-Xao--&j^%BVUnS}{u~%b?@VO{rflwOyfPSFOYkhRNKP)u`#^bm;3o
zr|OgR&$aoF@0KR}QXF@q>CUO`OPT?14ol{`yDC{XW<SrX=7Y<LF(2W2yr$@lwDvTk
ztHU?=2Cc<~KTl#lRj@+t#Iahh`ZI+&l|A^y-C4)0oMb!bNYSss?$@K(*iu^)O&s}c
zVH7{-)h}P0n763q>~}O0e|Ysxp6J<wD24j#ef8%SXHxxD6E;~%yA&av!9^8)l7;~x
z_><OA0<=%f?(#QG4@Ff^>=Qq-26|X|Os(34B=J|?Z!%dBZI~SCb&1+B{p~7HJRbeS
z+Cq{~On5m#W%sCyN^Q}O(@oVR86+n2o2{IZ(&vXh$(>Sg>aI9QzAPZ%J}*e5VdfT)
zZhdtP881vv%a1!2wPTfzxh3<Em@g_*Gtrex${%=S8|xbES~*3+o_BrF_E+L0VKQ{(
z)>oglz2?h`%62RZ`9;-o4qjg<|1&9kfnnls*V#Di)oMsi%2NthIjQ1)W?@6XIuyU>
z%`ynFcHN>X>VP)hZ~lAFQEASRai9O;9PvnO3^`mBFTAK9ZzHVNAL?Ai!#O!@&Dy>@
zX;y&JUdeQqIvP{;^2FQhX*U01WAd8y5xOgA&h^u&JeZLsz>;$8ebqnDL7jM~iMn*9
znZ!XFgROpWK7si$Q?vV5ir_qd3Te^?Ldw@wAxg>NQtH%9)~y1rkXX)AUvt1gKHV=@
zZ)$th%*;lrcbkob9wm7=DEvRynCdbXx{~Z>x=DvDMbp7YPq=TpBW<&bYc-z#O@-Me
zLl0!T-$O$NUS3`X_gI7)?`$P2bltrr@@oWKG2jCLHn-bx=-d{F>#3<&E>ZbdDW~PH
zvJra4a_^yJ(rssF=VV@5ua(A8Wy@BYL_32U;Xaj(>&prUn~@BK9%frEj6LoyaWMrw
z$^d2%-P#UFzYk?8WM6IiLp`ChOYA3G;ZV^$>k;P1Zfpo);{I0akMIBGrR+0fEkD(Q
z&nfrf#!i}d)J+G}%a!<WUsbCZ1iqN_L><LbNlm2S!|Vv6;BgoMPNhpoAUr(Na@(it
zv1_N(Ub_+6n8_)=a&$|Grlvp|!$D@UdzCRUF=_wybPWp+Z)UgMl=55}bCE8h(_P+L
zS7M^oxLq_zE}_Z1`bUq>KhCW2KdJ-tM_yJop3<0Be)O>d4bt5@CfOO0PL(Bq60}Y6
zS<b*;^pNIL3+#bPAGp;`err9dN2pCln8(5gyJuoT_mP!)KIH6#V1O$WgLnM>{Vksf
z_-$Lw*%X*M?J1(8yP#N?lPnwjt3+uTJIXvhI%=4E>l;OBrc;Xpdf5jw&K42UC=}ng
z=FsqP0Svv!BnB82Jv-R-$P4>wUbF=0amp&?Mq1DOMLD?y?-Xvwv$kI4N(RrYAAfW=
zy)*4TS+$_sKik4nOoapABmBzW+yh&RnR=j6_X}gTuJ6u+ThkhAzr5Z=z<kF2cwyE4
zuZ$Tf!cz7gucueeJ7M69=IgQ%$RQX-a24=wRPZ=OGv=3?@Y?0oIl&|e689AzXDH`j
z;2Tn#N9OLR9Bvb&UqgZpmrVdz7ntV*OaG#Ka~$4lgtLS$d#Kd=BBvraMVqh@AD%IF
zQ~j=Z@7F-<PUSAR%FoFCcpJbJz$&IRH^_Ld>3%D>#RB}8Ri1zEwdF!#_y$V!mId%!
zfl(CyZlD2oRBQ3xF*Tl3r}`!*Ew7zK)iimfA5?uH()6O)9h7Ojw^yym&^i!cFZ=Cd
zUnBtI5SkBi6AZ5jrfdu&3im(#{YZq;lw5Gi1_{j^^!=YG?pyIkGAjLV?Z3wpWs_mu
z8p4p*dxp}4%kEtuv&jIiXNYzJ9~YpVPtnOI+d~eYnXyp}n1PJML$rtSuyi++Y+qmR
zo>GNwGWrupmICTZy_3Xe5|4c6lW`Z3*CNeuBsYY@`cyWh-OGt3?-velRD3s%9)O*e
zVIMv|gK$|CBYWCm?fb0$S^knD0l$s%?F`kVO?M1{kp<Ni$nXTuke?SK72*Ob?yNC=
zrutOxvGRQ*n2kd`{&S$+N$M%@^_hqpX^QMCol{@wya6I8xSbD@xHEu@o!9Gsp_i7L
z+P}SB5pPpnHV^Ozd8|&Z`oKiU+$vyc##O(e)r^I!tu$`s^p21BFCLnp)El41S65xw
zs&sPygKF@RpPx$00AQMM51!6-rAR9P$&SJK(M<YO2+Qbzhsb1#VdjXUnEv$llma>3
zfb`?sYD}pcFWUb6c`v*O-||s;`9yUo5X+$7hXFkRG`dipJ=<EoYx$&atFt8ojXRk!
zyY>qZn2^l$E4%DYuJ!?+d2Kkpo)Xoo`PhyY+Ev>QNEm|v@d>7?(pp+^$N^|*FJ8Qu
z3)MWCBLQ|YWbzN-UqLU|jz1?Bze!629S5suSVBtz7sFp6@o~{OfBe6$8*1B1o+V~R
zDRM=*uRj0l++_3?dR>>a<wyzsM`rPE4sQx;cXlk2HZ3r_oS37GoTJ_EM@RV7qWy>*
zC?95kHXfEUIC0WHaeVw;&ja@QRYg|L#VbVM?{*7Ck74RXqqO*iqRWOyz0aghBw+HK
zixXmEm@&!$Opb}3vLDmBn%Q@N!~w=Y^v>d%CdG%Sw?#$znZ8?hrP;b&fD)fc&n>Ou
zg)@y9y>+I{k6Co-#gP4%dqOj6+LYXJd&H4)eR%wRz!(A+CCqd=SdP*&Gkc=rJmSAv
z+@Mi&+rIk7$!^hgHeIjv*vdzRZeclrj3;N^$DEv#M9gZRU9{GP(~`Zjvm<CtCc@0j
z<d7RSQ>XbMS=H@V69+D|0bhq7KY~cve*9t(vXn*>H8tDD>q%Vs#H$*YlJD>y?iiUD
zf>q$N2t#p2UDeao4OwjhWD3UF>h^(}wgB^Wb?GiJ_DtG3LR(BJ(C!0w2Tnq;WXZT>
zcmRLN-G)Bv{1y7`&KW+j%3pVLf`zk_6jQ1Ox0Ynvsp1b8QzOK+cY&>cXn~3M5n0Rx
z*@(hfbmHdVxam>;Z^^aF{J*l9N0w5ouHvVkh|A4aS$U&qgw7=tfK^9erS@<|=)}a`
za}?LC*By|Yft8h6&&~xVNmC?ZKAZJ5QRWrPdCIPuCjl?XK(lKD1}w+KvshsWb!E-+
zm6HU5d*cM{JUTE~woT?9MT1-iUr0mR?i-i=!?#*-V1?GS^nm9GdM!Db4!9m+>6Zg2
zi;T77h2`+59y5KL7OFMVxChU+2AgFvwt+n2vJft6G@2g##ck*~{v&($TJ$!sbT?;{
zEZ&2>bzA1z$XJ6(NyL+NWyeM2iO&o8!yguHQctKJgWX1-JhRm?JF}WAYrnrZlA!`9
z3XUnQMk^nHVHz~dIWXpd>yrmFGMMAo+*n!TP>S=w%h=Cc{G~O+LbKwy+$H1rFa4!5
z<HK$r<=OHt7L0=Z^63drU6bmiFuTTHJGk?U&c1{56b}n^)8dzn+N`hp)druPAj#6M
zu=nGP)rYe0dw%?nh<L%}+3b>mjGTKMvwN{2nYcwFIa*t5kg-JI@3ylL4sU<%1A`0m
zS&Zs}*K(M)a(<s6!d!lZ0l$yUMo*BEEkArY*g}21<HoAv0FYHaz#im5I%F9L)O{eo
zYr5X+G+M#_YtZc3?u)#U5f<+sukAvtZC9(V1#q=mMSp*|@rjg(t?A&TCxf~0^%rGj
zIn-LLt5oVMT0$O-4rKtD0VEPa$R!~vwU+Av(*IF><h`d*9E~TuT-kL)KTF6fAai-*
zuYG?g*CKM)f@L1T&p!YiH<_mf@d@~ZwLoM9U?y7?abm@+z_1Mb0x(V3;Zp}|Uvi5+
zCh&UcH?8V!w%1~chOfq<vzIPU^=$rk;Ot1;w>nW`iDCFTIvjUD1ffXyC=oSJTD2=F
zu28LDmSztbOgTgPKELrMIk~r~NnND>x%8QRGFZ{uYd{eWP^pQa?U}w2soL;w7f&%m
zuwA7=8~0WSNg$6JWvL{P%XUsE8YnJ-?3P_wSt(rz#Bc!DFGxy;*^$Qpla0Y-O5B`x
zS7r6(3`HJb905iGBf}4X*a4@~$cs46tQD#B_JwC+v-R^r!|vZZ%F5<x5ztSK)#8kS
zX~%M~NF=h}7fl%fYoY~sP&^8mN*<xJ8D!R+^8}0Np<wuObC~z%Cmh~xM4GVfs%N94
zV>))%y9Gr>M4F~}AP$q~p4y&&C+aQYpF$}E*?6qlOCKidw~dXHpqi4~VAz(FlzjNd
zEh#AphHe(*$vJrd43=0BHE8&Fy!h(Ya&-8PMf7;LuAZi{q#&)k5?>d&2)^P{w}g@Y
zzL5s4ikCjR;RaKzV{%r}_4lh@areP)Pl$2;z#gg#ijc8C(v|r3_UABbZ~7*^EI!w3
z60f-V*62fMUnXR=l$4a!H<ag=&V!TFAE?u=@PJ-KTbjaih2^rlTi9P2sl@&SekXgk
z%D)MB$upD=KN?zExyC)fHquevvQ&|Wzs+`m7MaBU8>UdZ{~XgVd*~=QjrKw~MWUbO
zeEWt0iOyK1yXa5j%f;-3pue)*E#d@^H7`fpXc3U9;n@Ou_zPIX!LsY!DKM%)`)o^7
zu*+OfDCQegb1fsiN;ij_L8&IK33vhsZMzY^`mx@hDKB+;1PE`mPwW^=oNUJE*9oM0
zPk(8Hl7`H8?(4gy%|=FFiZ!B1*&5+Z2v#`PP^B@musi}r3|6ite@y)Psu9XCaNJU!
ze`fo>>DSo{G*|qt&84HOL}$i<uneTB@if0?7R8Q<PkLSdJ$S}yOH$+*Y-D<ee)<Sm
zJ-<wORmq`P+?1!>1WF?SB*F0zhJKKh?*L{E<6FhUKQI8-i6?*Zc(NLhO&<7fOpzTG
ziZ{cS(!G~YzV%3oJ<R$^5m&7Wx<0tEZ&Fhe%6xs$S!>I~axlCM;b0h2uoJBkb(w6S
zBug3kxw%Mvxz{O$zg5B1(-e4k@cg=M1TjwDVj+jvuE-u~u6S7-slYZZE$t_idw?rz
zoSqvhv4&^&3xW%PT2G+?w=_Vd-ta^Oy8;yuJbGMi?E#ufmUXQ9aEOevPW;o{FjL+~
zkFPK0${`qS>gww);Jwn>qPp-v4kSkj5b}0CjP%6qoi)q7g&_+Dm-PFncdoTz?RMd)
zgCiyxAk;m2nO^a%DMs<~%-tt5cr>rVf=@S=?#}W__nb=Fiq!lsWHYe%Kx`UiIsKAa
zfhM1LPzUj3e#0$gv5LLp*3&zO@z1zg7+<2VVJ<xI@+ub&-9@XeU~UbG7>PX=u(*Uq
z^MC%ee?jjy`R)P)0PS8m@JB#a5X{@`!^F&txpj@gz`Q}mQ`r}n9Bi33xteSOJaI_F
zavq65W2pGK;y34}1UFTcLh$0p8fU)dEtu%W#KqCRb%WyUVyYhbtlwT~Qk+hXFTI+f
zmr2QS9#Qf7qUH|Z_GEIp@UZtMhyT;gG+)*kDY-4xa?L1NA+ys|GJmApnRc#ycV=V6
z*$V|R;f1(LGuxTlw>(L-H0kpcu^BejF%i6(D!!LPT&2Yza0o@W_^`0Bz@R+T_$+jE
z3qTY4v%JjI<^@!@|8+&WA5105hH{{R&2B(t_JsFI7E!o1cOqV#JzUYMYORVg{TwSz
z7>jpB-utiKwAtr>hHHJ+5x_tZFg+!b6rojoAfYJ&kc{$a`|g~abjw1#pp_*>I_(hU
z4FsicoVCxbOCk$FE<p3?S~CckglV3Ok_&D@OR~K7OmpKKB{yl!6yL7sX1OxS1{H2P
zB;_ttejy1P=h5RcD@f%G80E@kb@kXZp5wW~gHTnQoQNWP>Jqa4w|#EqB12d5IesPA
z>6ZJVMN|Q%gtJn}+XO*&DWL~?3BvTR*!gHm>gqD+6=vKjU$9xl<fg~V44tM}mK2pZ
zuX5`<WnyLX01^BYmO4oY@ulAr69PBpTz|FsG>g!S7Dwfq7Xc(I%}IMz1WWVrl+yH*
z_>AQosg-zKS6WZ6@7zcXt!&E|r}p0Y+A;uEbJHcbbdXee+|fG97(TA%D^yc9jRzDa
zhp{KDId%kWH{N^#!N%$}O>RqYPV+<!-AJDTvAt!&+R}Y9F{nL_jEu-0fA>C}EF}F_
zjH|2)z^9<wDB5*~5k-vW5A4iQR9!V}TGPsYQl?gMKXe!lNEdVJEB^TO5L^QsOv75d
zQj{qMV_?ShR9a-&wh#<pqqSGAC~-!1-c8G(wDp}<aIE@Z0YZ<i5nGNyvgOF~US4M1
z7tZEMnm{EAO-kCiR6iL{T7isUEjGMlit-92y<i$)NwX$?C)&$m=h+@(Rn305@89lS
zRQ95+qxqY*(PtN*i$1n@azQ)o7`L^)9#`QHMT905VqM0)epUQ?^;yc)E2>%jS)ZM|
zyPS!RotF8~y$=J11%`&7jHqB#Z>{jFG<fqn)k6F+%QM*SahD^P7*jP2at-A$U(6_x
zA19EzWg@&WtsZL7M&#XcRaDd-rh3k<&23<p@=+=>x_3`Gl&_uNa8Wff^nT2bNoBB6
z-A$j^Nw++HSHF!V;dI`-)u}^D>5!+-(Z(MJ1l>mkM7np^n7BAS4bQ)7+Ivr_XBV&H
z)!ks4F=|GwOA)tyw><|>gSMI%l0Y%cwKG?Aa??<25UUWuC#+pQ6QcVm$62>0ok!#B
zP76d+?!?SaWD?xgp8TbyeBtUqcWe@q2$h2ib5zeii7@f<qn<C-wJxIsUL+T78XI2-
zP1x1Pp`rh>N;6p(0*K(-dgnKN*WNdov}ZA`T5F4MlhUzY4La&DGKK6Ot8>*h-w215
zO?a6zc4spL3D1T;igPsaa#IlcZrgpcL*m@8Hm6VfNsoX<+_`IBgkeCUfRLaYlO4y*
zCIRkUG|vkjlu;<LHp-{MN171wlsxBtOl6)@5v#=LR_Pgrw+kpOD=qD67EjZjX$X{Y
zNg5iY_iko;CQGlRC19eQsoF#la(aJX&n6DHihMVu;TjwxSt;cH>y(_6GtzD3I&OYR
zFn3v5lr6HmySt0=HzkkAOB1GCeu?L$h(o)~Lb+N96E`~7h~AdNX3Q($z17ErbXiPF
zKKu>zJ)~5p^Di;+SsLr^?j8M36H<G$(w41B@v*GgaFJIh(sA~2|6at))kADW6Xg&y
z%LcthA{STt*ki6iPNDAYoq3Y2O1i7!_0vXOL++KP@02!A8O+&$p+U2bPuD*4$v@Gf
z^UwFzQMBF))1#R?uCKeiMma5Ox-ErRTj;Hm5*^W=f9ZMWXh_MKi^^_MRuA{rdNygt
z1oZ<N*EPA36g}?g%^%k5*Z8$#JHN_Vwaw5R6mh9t#%5QDWy@XUVpgLx1TeMw4uz0~
zi=AJgni$teG3wg%yhNWCs|0Sj|I_)w25Cf8T=>;oeWyxFC6i!Lw$TgGW1`&8yr?Ky
z97Ic+&s|$76bvy}jx6Rp1IqHHF@UWAMWE<uW<sU*7&v+DVQF0D+M2zTVrYt81vmY+
zy<Z?FlRO_}Axk%hN`pn#o9#AvtqC+^Xb#g2c+}<@Xa80NBHvr0_thTEp=f-6ad=P5
ztwc~>F{-I7u^nFw4>v=k;?OepLu@baJv}4zH$=<i3?}8z3;BbiN?iF+$JdMXOR7n=
zK6mS0N!|I{>6K(wxHN9(UPZ6ZcJA0^_W!P^zuTbrepR0R8<oEBg%?3VL8sW%8Q5b^
z6`Z)I&flb=u6q0m<2eQ?=FJsf&w)a>$_|p_j^BOrt<On=Ggbr6ACLHE-M@Ne8om2-
z?(WyYH|||Bb!%ov6AEqG!TBpUn8y8>IK8KTz7@(3<mD}LtNi}&Zpe_6E!vitaguX(
zJF$9h-=8yMszJGDa;5(@<BM=#v5R&T>M4)Dy#4}%C98EyPko6-?dXd%rM1LD!<uo9
zSzdy?N`_*WM~&h8@G+J?!rfxe!RAwK#K@6PN<p*qlx+)ap|~*fuR*d|k2@T$^2W<<
zt{&oi#W$8CpI1Jea#wM=;-=!}{L}kq_HpCo3=@~&V`FTO!hI2Tj}2F2P1!ALd}}Q8
z5SHP$_xTGAr|AEa3Aai!=4*4=?_o~iJ)a+w({&I%Z2U<kY^eIsp>4Fpq{hIwyW5y$
zX>@LP2CZE_Bb{lj@3iK7Z0uBZkV_~H<+sGDV^wu^ES;k{%P(1FRHm(B27NI9W}n2%
zA{j{pjB91+#=oiz#kJ#i_?+IlZ2uKX7C*5_5T+$eBa4Qg=H=-l!^XM(Yi+0}gv198
zE=5MTs{4mCI<0GuM!B($ahf9dj5$!m?8@+0y+0kAS&3`yB1XStR2aJvcF)a{Z<^fD
zVtT=Qt>{Xq1<KSm*1q{woQB}&ybE2=nrhqk*&kAAC%4pR4-nY=>s|h3-2KSiAFoyC
z+rAGzRQ^fGs0$xbIJZQsR`I4jw|IuJvrrg$Yl?b1=lZ`dJ${p~4DsXJ<7=7)m)O_d
z=q1zqSB(!frXZZQ6r&-zp>*BzghL*DCfHC#<DHvSepA>zy{cI&Ms#5v=oESI(9=1h
z_?r-AezwWIn(F`01qdHS%}26axpp^*QO<9d&Du!e>?c$D3O9GDm)>qysib<Z@x}Dg
zu*HOT)I?nzD8Wy^46)1f&No!utG@T?73yVl_LGIhY{e(oMea?j=-yaQEDQ0ct&qyu
z9}bJ-AH-SlxwZRYb!#7ug;y)>8UT#dZ{yo2<~=7l*NwKzXEh$Z)Z=};xn_KBSAr`i
z@!uq$R!E93Pl~nZXpc>$LdQ&^PHvoRsfm0~qMWPr2fGZjbK^P3`Mb8htr7WTCw;ME
zuXJYLc9lB6nYw$=?y}U5Yf)fN4O{Bko(cc_tbaja4WblRhpba~Ch3^E_ztEkGN`%s
zOGWb#pHgP?_XIp!@2owK;qh~&YI4p`#l;p@j9O2-MgK(e6l7Fru<!Zw<@FC}ctoFw
zj9FWI{IS@ra<_?&WYB13UHKqQg(WWbLF-~<NBRk8TG3;ze+;5TcOO(Ku{K2cgefeW
z3w^tYR~=hubrZm~sxIw5<`UMOG*1iv8X_a6k;q*?=KeHv%0wVqd_bK5GZ1OP<bDks
z)V1GL=j6D>+3ke6GDUBv-5qO&DFK5CA16^K)-Ij0GAj=ES+b?BZkYOe3=<-Duvdm=
z?pP7s**SCiq<-pM;01*&*M!$QNoV<|nY=f0R4gLBs>3C>)fTEfSZL)`(Bz1<R+4&l
zWa?eTzPXlFToclsnd{%%Iv=34rOiOa@@euc%DQdQfU=X>jAoU|ySvs9!_z!Fp)4_N
ze!9$^XHrQ4l<d~YP4i*%JZDIiNpWW=zo!i%{G-hd3)Te9veGK8KZbXSN8doW_g~{e
z{+DGrcFXAeE8VH`aQ{Lzx=>-;8YPhz5&1<nU(?xZ1a2Mp#AAJ;J2`oymD>yt(hGIR
z9?4x|6|#2bCXP4|5#pPXOTTJH&$r=f{1J@mx7${9|LSzlcku9EU5fnfFpo$dWWKCt
zpo@6Giu3KZ@0sFr?$I#%opT}n`P%SL$)B1RTrcDqv-ZAjW<orkr<`ZB=hI+tTD7j6
zROpnC&_*riIhRq#O(sr$bXha^uh;Urflof{c0W|D;nvdb4eGSr(2og9R2jEIe7wLy
z&#p-lq?1Ec3J%Em-|`&@H#Baii^Aa~*<7`k#&h)xC-m(Xv6<Fb<h9FTvjHgUfC6pd
zE1a0Xb3FCp)ROpXXg;m&N<7c+nBef6g*H5nX_KyVAtELC<<;YaJc~pw?|1XPVd1Y<
ztSVgJ5>@O9_r+i7XkWrE#?-Th&ZyI8su6~+ugy?x5OhLap(5+5vfpmBV%Q~`e<3uf
z;KymgUXo;uGBuLUp^UIy-eV`e;2MyJF!(+Ap`U+_l(o#)ku_0Bw-}tnNm;aAQ~7q1
z@HBVFcTluQ(29X{9vmVWnVyG+X#;0UxA6X<qVDqx*Wdi=5Fb)gdoQk?k?WhekSdXG
zShPYarg5DAtC;tvs|A-%dGinUd*W%#myI}fo8+HSy^9Iq7_mo)QcymB#Ja@A;n+`~
zyHC||R5I%|+dkSFP+h(#+T;C@Tc7S!h3uM5syH{vKK-khp1u0JMIP+#T;ynpJvPQm
zxn*)(*uNV8C$dxM_*?_J%%r3@N%5JBil^=D2mO7sRc@jya7Sroq6n8z@^j`Wt>2a&
zsp4wmv(@rStgCg&qG|arL=~R(CfQjNRbsWBaK7sd96=S*x2@?if|9V_-aJj?Kp~Bz
z*vov)?;bX@cW`hJ?L`@^c#Y;)(`@S$(n<Q8uI2Ka+%?ks(SFySR+(==LoGg=DyLk%
zS1~`A9gd4T4F2rdy5#He0ZF)45{}RilB~I5ERfj~k+j%?;IMWXl`zMBujV_i_Ixo;
zJYMnXrt6wH>6mY9MUCRb2>c-9l}9b~u_pADH)p2ro|i`jwz{<^k1Jf!#l)LY&bUqL
z;B8snD5A3{i8d1fLz8>k4~T<S=fhL7m=A21M+I=vDB?b+@|E&Ra>lT2rr}mApNWkD
zYB$Tx(;+V&anAF`R4L4ivuh&34~_Wpk9ihjNcgMC;9FnkIzOc|UNG0-dyPAd=cGoy
z&r<X0v*P-_>F!iv>b=Uss+4B6)|a+bDEy$PLq^Q*mVY>oYMhpIe)<L79I4WCDs0#{
z^kvUoJ~WmleQn*Gm0<RKw4sHtQAw%rw5P4vdM<G$yI*AMGnRH^5=Cf=n~qDOYKv63
zR3~<ob>OBSvg-2RTHnmuM~5PB7iyL|DXfcz77-ig2(!*@QC@r+^o81Eje+5&caJ_U
zB^bX@VLJ>fKx?|CNo)xVDvlWMXttb;x9jQY(HP@38Z};ZQ59m$ei@FC*_uWs;&cz4
zNcKMI65r@o74bFGo|Y#ku1nWtPOOAtJaUQIst0jv=r%4KzX~mKy+ZAWls;ccUu*3M
zViKPa$()oEmzXJjiq&}ROf6IryvCCe)s(Ztsi&F!7T$_)rq<{gH>{>lEM@S7wIm(2
zZV(RKRl>HQ8kl6jzoI9$yj*p~(Je6~98HPG-7+nxb`9j_-*tW76;o;f8w~!AOTT3G
z^;C>t^xhRBUp;HD<Zs{?%rkxc?nfHiIYn*LX|+9Rx+9RFbo*~<GAB$~eF*N=y=8jX
zG(jgqXk5R3w<ZL)pB^8il`M?W?sDnTXZ)MyeNZuwoSl&K)hayv%dJ``sl&8O+ue9w
zt7mWWDAy;I+ad#+dp9|)t`ti%AwAtP`@5TMN%zbGs|Oc{-P)Qb;Lh`f%ypbPg>Sns
zUi9Nc5vdtJyXh_!@8#{^C_g6F62+eLxFBYOk7l%1L?L#Ce=pf|Yn0%H^mLZlde6g~
zS$V~Me1m~wu5an=7&gb@&6=7uY6#ym>%J7*@uW!gYmjcd4G@4chhTYw@R1e#>ku!@
zyR~CcZ0Co@#P0}W=Hi8SxSe-0nBI-|_OiWxau&m}s%`O+m?5Zn?Of%9h|-u_V|(_F
zj<;c6cg;ab<Mcvd3*7*pMc_;9px&<^%|6`MXeX@1ifj`;QxgXV|Dn&;nu=uh*u_SB
z2*q{Eq`l03X{T`Dix420H#9EPJFg23kT#yiB^dO4st@mwY$zk>{AzZ?H%(G17_P=7
zeQ)_&I(w3)Vd=89#7NxUs?BrUuNvN6u;(z4`(yo7BgSsoBe77zKy0<zx&-msqyV=$
zoy`AawQlJCnf3oHkgnd1=uB_i4QKx?5p2ZH`f;;;wDqdWX?GNN^O?erwY4Lr{*S3)
zKZWU=JXM|kO|5u8i2~t8qW|`8{S?_rYbrce<>jV$e1^{MS<I1Hym0w%(!NL*;;a@=
zRJi}jPNOLu$|{+AyswUPWpYx`W_DD$@jdSOw2*@D>bE!zTVweeRim&;iEbvY+k4f4
zr?3LYqKbv=D^35hoi}bPY`3z{Xx+0kb(;0=epP$_;nB#eg#N)=-x@xZc_otRXfuI-
zKUPD<*hb7hKcaqwxtPR%l_lmZEI$DE=mReR@4$K+n;$1*bgb|x>@P5=tCQ`D5zX{A
zSwSseKjNyo6U6i~#f?uEj>14(j^WKA_CX(7XGU!NWoBnO9sVI<C+b9|7IO*hVwuqN
zs$bQyxk%%p)|&0oLe&Dcy&lXx?f$2;cB31e@rO=t@@D1rl7zos-_3XKuMkI1ybJ6M
zO=-M6+VOgYE-vY><yA9w%*E-GQLWMV`N(FkFRo#+2}}}!Zq$>=Q;~8ksyWL>#y&Uf
z^(?H^SSv@BqZs=uCo}}4MU%gyl@?}>&5MU;(OY}Y&RI|UO4pSy9-B@+Q;!~<HZ%9)
z=2Zx19qn9oRoa}?K@eKdGM=4<y2zV`=T88R{g6Yhq?~EYH?5AU64x6$vc?qo+p7>f
zo>r)wvD*>T(ahKRyeooSTY1H?OmQ2M>kG(ATr+Y&iMXn9gYsF`4>u~_{&DZq8P_x8
zl8uidurGf66OZ>BJUkM|#MfxY&y*(!TMZ5iVH`bscEfvAs^F4*0bdnqSo!i)it$+S
zokY1bIgdBLwh|K_g@;rpwH&aB;XF^Zp_G)qSw2HH;XD!7+UGZTYGPif+;FK-XsaE6
zeP1P{IW&NTAGWd>TylJY2E(clTyNd)=1?x`EB9F@uX~8!$aHQJ%cdr^>9RcWIrM3!
z3yF)rz{-2jZp?L`&q;}Y?DppA%^}5Y#Rq9o^ZVtz_s0dSpTtuCXHREgEph5z+iIM%
z3*~`Nide@<YF9r$q%5f$SF(0&{;A)lSi@rA>5V^B1$A->;bn#P>4)YFGrjQ|wvt-W
z7hdzs%P8pVd*&{?FGbW7tNom)Q-m=IIsSM1j@GB%P9$!s^~DK4!R!|K+o!@v*62-M
z{foRssfyLc5F5K46iZU^SFRXkjg}Oto=ER^eFFopO5Qo00CTu&SMcV!hTL@JmUL&L
ziQ^_85o5<LBj<FGA~7d|h`#d~3B1&H=NCG#*3mJZY?qw-V-mxaUG0~3BqT#4&_;Fh
zewRE~Rr;Js+SbJ1BU*IBRwK+7U+#U!j<8(Eer^pNuC>h1$6;hHtj3O-1W#3J&0Ou>
zq}}xf2F59W(HidliC0LUj{U0Bg!JBXu}`;0gb;Cw4YK^=9{I<Emt?;=D<qo&;-mk2
zU1b~PefNd<KmR`DOlH!imctvT>$l$IjX1n9&v&@@YE$(c@|v^usgn6HRi7=nBCIr!
zbzpH%yEJE0&qE5fgrnRzR!grQyzz0DBbc_ybzEe*Uu!?wOv&9lRQ<~7p`jg%f2Q?w
zEMu=Q3%`97(aeNhdaL$nCy_v!4IFn)EX8q~BKvk9A@MB>ap9IetTxjm$VLp4W)V1S
z^aIMH$P+2zT=wk~*e0?2B%hpJ77dlm*v3}szsS#(f&X@9%<3W{wRlvBC?aZ}{v`ga
zHIgtr#-h=heaYIY+xN>$UOCd{SZO`|nb;*U2kw;TBW}d*w<i>Oodhce@uzr=+x{TW
z`&S3m{rxAJEqL^wM9qzOo8G!T<Fme`DdI%JCR`19V5$f?-0M4XPvx$d56(X>W8K4Q
zB^cs*%|m*knVXgRW`@s-kT}uyvkyHNCoh42#kMz3+*6WtYmrfXGK10AvZicm8~H1>
zq-p45ovZUtw4yO$<%i-VaXdyYld;sFI8OOwmEYu~_!WcCbi5fZe4np-NDZH%N6B+b
z3Vg86S!s0_$o8yxaQ2d^6PrC_PokINR(l0(k7&R-bavKb!?{)7x{Q|QPOKtir=LjQ
zsB3m~&yCQn^qRjpp!Z~IO6dB^@MS}n+hrW$a;IfTuf6k|O$xOS{NoU4pZuehhV2@>
z_1@jV9*-IIc;P4Uhu)Z@I^TT@5zP7-OhJxq-Sz&>JMqH)`hAX=6BvIby?)}I*&9qv
zdLb$F>JBmvmWjG=#I@*VdaAiK?Hj*&nM)F~+#i3*IW_(b6YTWQl)`keLU7OM{^nq_
z-Z;@pWKZx}CTgbo9M>$qmW^Ax#;}J1_wBi;DXJwt(P~q1X38yi7V~^4c<Vp&o?AK|
zvkL^zxW?LUljJdkVqI#_yv9>DDRmbslIxD5X$l^TaG!t4FeM;#i5PJrnD?4MwZ=u$
zSpGefqUy7+JDZ#GgjzUI@ox}f0#R{|doA0Yv!6XZJ)ipd6)r6~l0|f<*}~P+z8<pJ
zxA9f>6MXT4YllN}|Mo@+|HP)gT5a{@0&xJH@RQef%XvR~?<g;)eurw}rIib98d^A{
zq(tWiA70)d<Q96=gZg@P9b>E<#>c40(D>wq3m4)aMQt9SPH-?o%|kI?Smu0rE}ovW
zdVZ@-F;2-STIQjMEi*F-7R_~eOG#!SAN6@55)1}pE>3(DGU_gy4{M@65f*y5N!i6|
zrMc~V&nqUp{h;gw@9DXhTRgmkY25FMii-B$evO;k-;G0iGly{5M}?-Q*;SOD`VmTA
zXO7jP-?)yiJ?}ViI6Ir`>E(41h586Wz6ND1xD-}fJvr*@iw6?Bs@r+fz}vYc|F>I>
ztY?A|HJL;f8?AM*_ZyzZ%%`iHG_D$c!c>ItE>Qduos+OL^f=-UIUk!=*#djmZPW`M
zeKBw7*t+9NA4T9*;ko6zUQ=;G#7Zg6(1My`%6lG2QErv@BFGE?8ZmR2dBe!oK$vVR
zq9h{ELWOE^oBdi|o-|CwnZ4Q@bvUi}gQc|>8@i8e2kWCZ^eOy7F2wdJCJrN4^qI2F
zT{LSt;JTs9h1zD671FDG>YU<FxQsq(j*7#*xU8h~1DtSS+78nLvgSM?ffF|FrKj_g
zjjo`o07sgtnwp<5tu~qO!TEmNq@7&~9oZborXX#hbYpi0b|&*)$*<fz&4@BCFnsFj
znhjrjBI+s(Y^jkz5xSLa(9U<F<m=ZwI74LZ>Ow<0eKh9w?F5kh!b9e3tFNg^YZT<c
z1Fs@)5~&i6pK@;|tvcw&VMkrf3PnBkiK-q9eYMmgVn(J4m4~rWEk*;PrX9cTR`vxe
znZDyMZ{CBB{MpfA?MZI#V7IRM4f4LUsc8@~0^5_U|GaqV()&-J^gy;uw%f5eWIGiS
z$<~$)+@8?M$~K{NL+iBR;o)c!WvLFsv6U{)@HkiBbJlipY@`+$*}9C~>FLupS!<RC
zlF{EvJ8pT_Uz>Yn`=W5cw4wrLTMt=s!~IsMm7HQ4tGtLAF!L%aFV6?3KG^LMx(0?W
zuzT<>j8H(!18zw0UvSc}I+vSNtmvPXo=#>qkQH$-5+C|GoXsVrbcH>O36(;@KZRL9
zS`V&eM8g|^(dHKrZ_0xp4wzNoV_|3iBJbj3)y$p`CKB&KqzE(s`2Bc8x$W*fZ1Md3
zJg6z0+M2;6$`EL$3yd3s<Ku;(babgxbel`vueuqKj}`r4v@9XL1r$xtv*~UxmrXf)
z1v$C<uH)-2I|&Xn-^G8N{w*!S@X%dCgatijmh>lRaA*kedVcCG+cV{ol5X%=0<i?_
zy?aJlF~j1)9&ezxH|y1_S6N)_w=i&cejXmyK>(4LmS#)-UW$i-a_Q*O?k<@$3gp@j
zv1e@|Zl`O!^oJG~r_A0w>~v0h2?w7Zzq3;bKTX+}FDGKlDI(Ph6fRu~lk<DP_s&%E
z0|;7$E#dfk5XWoPI)j6$fJC9Eum2%C+mopognCugj=5!!dM~3;KY=wf*MgCvTze6}
zvxWmJ>>!S-CO1)_M!OPN@4IJ2AQ0gE)9cXB66qyXPbG(No?ye{@mIcS_knQ_umOZI
zz+HmlZwOLm8Y<H4d&{o<zf)(4DrhwGNTiH8hh7;75jfkTe++?v4G<!eMFsx@GDELj
z4KX%0HqONm8&cJ3TKKh*v4foo=i!kHem7*Pt4#BtB>vVVb+umzjpz-I&;8s+jFtOY
z2p;#E5Yi=b1By;`?_4O9)qEQFqn3m9W&=CgalQwSxasJ)q?JnD`%u!jQ+MU>-$k&P
zr3?lU-2muVAU+NB^{xMS{f~ti7&{QL!-~`P_)cZ1jYK^~?}YS)I~e?dJ#f-ouP0oE
zj9sD!n{U^s{DwU=QWkY!g%kj91)SxBn!KpLSEo&V7}#}4>^*?@O6H;b`}gIBzP{Yl
z0?2cL=?ThiLKn)@!z1*uG`&n>BqQra%rWxk0GhJ2JgmBd!5zHBTlep$8W|gtFS>da
z_#g;c!Yvd!4>~)R15*(6|7hs$j)jr_CR|f9Q_~RJbVL$3EP(nF)Jo?uk7!%}5{YCI
z83<`R#F#Uqu)GS1)Trq&I%j&N3V1U(T=jUJ?+5s+1484#v1on$7VJP%6B~GR2PUV$
zkcEG;`IDj$q9`<)%)!1UX^;%B6MosMks5ja$8n@m4Mn=s`__N1*4i~bjMPLPKPldI
z7yi5TfB$_G*M3eYyg2-rb}>bmlq`0V#MYo;g5TzA7JR!nm0|a)*6%UUe}M~@!toN0
zW_nth{gWqp;8g{%RXmr(hMqWmgFD#X!*-@@AY;L?qPf<vu|5D8`BiAZP66iy8!y*W
ziz7jrMam}Bbe8t@w#p!Py|7rsIP=^NxNJnf@?1r;JV95U#CfYRghWXYsHkwOlUMSR
zhd3qSM|rll^$83RGHzxRCZ>4Jt^E0OUij?762w2?8F9US_4)o+!`lX83|vt~sjqXu
zz-;m&lAqrJL~UdNH`s2%dm?}LkTM98@$#Ub<KXQguC0|oUJD9uQM?&seuX0Q@(tdN
zQenu{p)yQhz-MQt1qN$@H^##TFP>xwZLovi0ftULNPFo@!G)1baRNs=ELSTnYbHFE
zhYv+U+S_Tdl}TcOzRe0OuodFM<Rq_61E&>>R>iTWaU-u*o)qaLgQHU@rnDL?u{7V^
z>ftXCIX>kj^>gXxgVonTkMz&04Iwhm?CZt<Ub&>}bp3i1kficq><=U@*fp>SI=pOw
zGJWj~;V;5h;a#__%SwiDj!Np?W@ct~@b_1Ji^^h+e0}MGOWF~-#q;q+z1y|^N9J%;
zJLWeY!Qln+HcgaUO%kadN0LnF3Yx=C3zuT0LPxL?kmt!u+=wwnV3Lrfi)M_uV)Gb&
zF=eBihw~u(p~OG3DigsfglZZX8~fDOw#!aU-oI5>QGrka(SI)xj^F|PJ3PSELhK<k
zbb;FSZ|*xUkg;??mkEIszQIjk-1qc!chk&1|KWmFdmo;qiDrqxJ|)ZLU0NVVlJ|<e
zQ7i)+y8>|2fQ;%0<m(3dkDT>YRb<XDBteg(Lto(NIR2j**&};<1IRpC7f=dt3lj-Y
zRv0|b&~owf81`6x6(jdxXXA05^$XDP_3-ohDdd&63aWZjFDkxGyrAMK5omqWbq({v
zGB;<r5G}?>duWBKxuP%FqVuy>P{POB`YqXHY&;4C4%rjH<>JY0G8wP;FA~PUa@7K;
z8ZcnA{|wxiS7D<sut0Mm7=Sq#*?$5!$xFZiYb0#^cmI(RtizNWe}Dg8lQ;>;Px`GS
z<)@5Y6JnikY5MzbqrKOKgPCu|s?U)DY<Jz`JkH0d-_AUBOG`duo}ypWx_O-0H#ltN
z)s?usJQ3fG?+PiqM#f)}Rda5Y#XQFe)lkB~l2zdPfyXp=;-B&fzYjo63;6Q2tjuOI
zwX#xacEg+O5~;862rF_yxb}N^_#1Yj5bJZtIPIgGRD>UC>(_fCkXj(fg3}?Q%?nB$
zFjl^(paABZCSd*YY;`ga)<w+C&22(ZZkT{m1((Ze=)nv3FBpmwQLVb_)OV@#7C;fa
zF^SRFr;`ouV(<7dJK#-W-;TcWAK&iO%e?SEt8H)%R9eEM>0Zj!aSoq&LzG}H#q*}h
zzDz{5IF;+aPTIn{(XkD$)xx=lO=4keOViWp5bG!<{7Eb6;OG+c8HGZDlH{hldntUN
z*tobm5+^SL8LSPMp5hoX?;XH8gD|iRAb-GxB3>r@8Q#l^e-WA7SCy%99WI>Fo3J#&
zq%d>jyb1qWQ}Zz_jOI&uIoX#FwDuupaOxps?7pMm-8bm?g^bs>N_*Vr0?z~70}W2j
z{!2n|<Wmw?FQt@1aRK^RdpMzlKujN4?$QjZ3*Nm137uRHf>;(DBd?O3u7uX7rmX0c
zz(oOk%7V3wVDZ!f*{Xr;HVS_P{5OF7Nw4$$RYlVC${?*vquT7P0STAo1DVKWmY1v(
zQwtF5peR6-FLZc#IGOr~wdjrO!K?!&l%A3C2O6BoJ~2531JKz)^iSD+FmvgWy;B<M
zSR`I};J-EqHmj>93Kflq*+N^8fpZ^UJ5eOEB|ZIEDkIDK)JkZRuG7w_tzAs9T)uvM
zPFIVUhX?i93NSnT`}YYb>+Qko6)xwMKzQKfR|3XOKztm8EBSk5L_>glblTcv#T{Y)
z83f6PDQ7#Lz)^eq_U)CMc*Q7O-O534x87Y%uDZ>&z<)Q+;0wtLUl0aBPI>p`%N=+G
zDe8|BKWp$gJ(V`M?Bp6D++puKCR6diBaSkAwf&eaZN@yN7#{b+^0EZh8>&6{Dd91&
z9FX1tYFq2QA%fL|8qD%BEN6kD@-pN@I3Z*Upp_Ki2>6iXS{Y<5mrH*<hR{m(k$v{;
zy@K;N=<~G#i}-b1Mq|9ERj^<(cn732U|>!E?JU<Mu=p-Qg;zWPg^fXqfy7CV`EGtP
zp%*ee1iagC@=n1PA|0?TC2#mJyxA*&())cE7?<Q!aRe&1#l-@6WU%=Vv=BcO<|Bj)
z__sZLeE8}%R!64<Z3bebHeFO)(398Z_6mE{uRQxTaNQ}5?oeS|DCqC_D;$GUtsp{&
z>VfjL=?eNMu`w~Wkp;Y`PRYy22o?Bsr0XL`ipj*;=7kUgxRNQqyP)i&$f#bY!av)X
zi-83SP1XD_Xozo#w>+tX<wTJ*2M^z}K00^S;o-x%Gq&ML;uuKpAmz`-RzU12y~y9e
z@wNqlJ`L5!PjK>qvJwj1tdbHL@~RY>7V`%vuh=M^4n%;-Io#|1661JQm3gu2mchap
zc_k&DCJTJ4BC<KD^?{3zoL$j1wU?lI|6q8YTEFvpndwRUW``#B3A2{}&jqNtp{omJ
zuOqlFL5R_bzo+*QS|_rg+r=e|K(L0qM*Wu2qN}_6(c{Mw!bR^gwBt45<bkr85mOJ<
zCR`wUFw=x`1oGP*Ev<0qZRIM+w@rGzd&6~smzOy_SbD|DCdJXdYw_4&y0X>^-{_^T
zLzw+r+q-&3)mvc`Memy7wQBegk-BF%)v76aO?t4lR7@kIqaR^;tA~e2L3z0X+xS{M
zR!H^N?-QS>R?MpeG4L%@X1JQJ?7_!@vvLzEm4-;R86Z}vogvAi3Z{r&e5+u?VTFQz
z<sKNY!kMA^CGK0}t>x(W0ZyOnc&^-AUe3(>th|AUEY)103=}ka@reZ|HfOWV|4i*q
z)leP&8}pi2)KcBOX`lj47Uk<~qFAe`B=>3+oc~EoqX0R${3~94)ib6r;lnxp-T6?&
zy{5F3vj#QSAw<U&R8I9&?KE2zh0;s4>{C`CX({caJA;;!dkbfUVpdTAPNpMRRYR@T
z$Q~fFD))mAy8uQ>kgTx+o1}NG{Ha3`uGFVQpDt93-|#<ap7c3vGdyHC^Hq@WardYd
zt23ePVqYrW4ftCRoICmY&%A@$V|=H<yF$~H09bWEt^}ZP${vqW!P*BD3uFF+R7Cb2
zq3h8@N?ZRPA2$Mt=Czk^M}bZY-7GMN%c`q+W&h9}!w>d-4X^3_`RQ^8``Pbkh&&M3
zf5KfNi`bI?lamv_y80rSllu{A2BARgp5J%|Dljngpdo39$vnWSi0SFKu=6N}Etn@0
zoNV@_q@;prWx>LCc=#@u4rsp*_uv1!!D;>E)|UP4jF&lZX>=o)k(X+&9K7F%z%4dt
zt5HU`p11;m<l&;lvbJ>6K5;+h0kxbnlsd4t79a<o1JE11FolL3h^S<@_T>7`&g+n3
zY2R7FVI>^MNI4bzMvvRoxuYpq<&2RsSeeRBd%o_yG@procg3{z*|x8hnNCK)&(_!n
zTCM`a^S1B*rW_BUYjlfaZk?)L*Y02&Iw6<x843;Z7AcU5fQq0qWWY5A1C4<&7?kS4
z2j^*>;UU1c9s?RQEiEniF0-Y34*vS}6#jJa!z$>ZKYX|jUKPe*R+}qanB&XjvqIew
zE2DFfT60cT?=~bqvaAy<#IpHU71Vt2E0&r~Tc(9|9in>ulWRc6rBH478T|n)1XiQ#
z$dR@Som5ol!a9DUw&?Es4s-@&yK(U4{H`W5=^|`~5HOL=_=J1oR4ns3@iA2Yl3h#S
zP4)BV&&1(uNbI)(6QC0h=aX{RXfqjI3c7^-)%(&DQV6tQcRu}M*h-h|y0iQV<ZTYn
zegM`I=(WlM`_EJ(ABr84jby=A9SRjWn&bGeIw4LO^Od38&dcnl8G49#`%3kHUe#{<
zLW6eMg1)EGOrCKJQq-rt7j$`1gsIousZCfQ1u$N)Enm_aG8hG4El?HJ%kWL#9p77>
zYF@ojqpp7D_eUc~-Z6-M{|h^uCQ%3bJMLx{gO^Sszz+Fem|sJZ-P*yr!I`fuv9+#0
zXfi@@LuFUJU*Zh9hgP1ng+<lZZlm-Z?HUCo%4V-KS4wpk;*<~@Kl}PP{@fY+(4zfc
zB8Zk=pPQOS*CeDr_Cc|HYpEasO7gA{3zk#ZqM|g{xBAopiDtJ~P{@7`r7YIE^~fA9
zbAe6+U*x3zIJg``1)3803Yssl{6d`6*ba(;fC;aXwx*wP^kpbw#YL)*u@&RAbD}dB
z_pAL=e|YDK;t8+RrG%p3#Lx4?yPB=Z)W`m`Ozj~2H(#`VQa$b&MW{&6F}M)|OOZRk
ztN|PA)s(Dq4$FU-BF@p%;J?<&4S|6yGXyn(JjiN*!l>Q=`|P25xA*cg{-0CaX$|h<
zgK@Sl`Z8_vY%D7T_Zo(bKqqsx?}#+>l@v8mk}w~wbr;ul1^J(^b?dx0t^qtluAIS(
zm-|wui&U_;3+qJG`^HwRf>n($TqnDb6qiiD4W2JAVlOCPn%lq({{qBtvKAyP(O10g
zWw%e;w>CTYhaZI{-MDdMbL;PYIHA|z;##S^poGzPDsS_L58_XpooQ!BfpH6;d2Xtj
zSjSj4{(ly<Id9f4D=Hd*7Y`W?J~}*LvYCB#Hg>+6b|IEwHF@Jkmp<J6aNYI#qu8kr
zBc_?5w2S6oXm&4MeFlNI4jSXYt$hmaIb`|$bapyID1?_<&qSAQF<-DsRA@j$tr$TH
zuw+Vp?%GHPmVZp+IdR6f=u~~rP}niFaU?Zj1!`6M@Vu*i`iVjPZA(LZ79n#h-+tU!
zE_t@Fu<ZoxQb#8`ze2Nx7<WA<>jNPX;GJxF>a*0kgc>$$ybY<Su(0WeKv#Hpi&zMN
zQ)&rNBH7&N)#TYIub-%OBe|j9Or;|+Xy09V><*wxAce)gyPDfo0NpVBZ>QlcMd4%m
zR7qLw+$T>85Bza0FE5u9{{WCF`B`K|TjzHBZ1-X2X$82ggD$xu{1e={lzF@7OCEUS
z?)eVd&0cwSrv`U=;L$;^yWO(<r%b2z<E`7-qHa0MG_@;N{~JPSinu>F7>9}p`f(l<
zzV=kXiY?is@{VL|l8j0Xx9nd)q5-<>gN#9Eaf+8$PMjB|G~;Whlz!ik^{Dtyw|t?Y
zFH_y)=sVIY^p)Y~23Ywu3IeQt=<mMCb7<JaDoBU*YrhwmJehQ{mEp&Tl>{<OsRn%3
zVBTG;TgEP4lI#Vgdr;pN$2kQ-?5B8j86}^VWYZx~Y{_1?N58SGS^2l4ybbV<9vQ3z
zL`crc^UJ*fu<06r69dtgtS*B?zq?XR{NDkCXa_+4$KmneVfyt~odcgqPz^#SN+%?Z
z`~!fqFaVz_sGcEV-=|`yHrn_EaZYK|oo`<&z9e+8QO!<O_wR(057M{L;#QNO`}dQ-
z(YHI(Vh?;yW-^g%Or;4-5=1vCSm*{j2{BNNLhkFrCL!lAU%JZW!VN8)`K^VazY&t(
zh`D$3d?5nbPI7ZVPOZ_@``<L=@v{${WzA1QRyk4Q;%p5+85w8Y-E>&^fcu6F0;3?Z
z$PXI96pr(AHt^d}PmrD730H*NS&0wFN98=?WV?-IuW`3v@8*hsR}AFdpmF>JF&LVy
zW^CFeF_y1#R3D=7zeY#-njX368L`8t+Uh-Fq!6*X3X>J6?&AZc`pUj?a#RaJ_enN`
z6z(3SOYF3T?UiJ4Z7`9%G7xO4x+tGm05-N(LMb(CiljNC!4l!!GW9ZBix!bJv=JPO
zWUq2(6W8FHXT=QpoJ0y!{gX4{Q?#4pN>Ue&<#eIEQo(kan+-qX^OouyZvWxql(Yg6
zv?dBi0Pxl{Q^FNV-<lxwvzdxwL^!%x!&WI<+*bYmahfUcPcz2haI1!v_fYJtwa&*n
z14V0S+-+^oQr%G_!K46iR@;TZpzNZ<!vAU9mZUv*y}Z2(pks$(PKwifC$a#vbmYBQ
zXQ#h>Dx92R=59(w(NCE1w+@0h<<iJ=@1BwmOH22j%mF+C*tu**E^iV@g8%>|tBg#O
z8RNz}yFf@Pr~DLmr50yiw}rHwRy2c`mF#s|d{^c2<@W~%zF;a!Uj4`E4ZR)|0`{)1
zGKt}{ZePAj!IzRvnaRR&dx~8?K0avE#B7RhhG<G6&$S@FY!vp7@C&T`tDK&8RyeLA
z;K%B|tx?n7(-f>Z$@Y*my3Lz+)x?y3Jq4#Ac9Qln|HwuRtATK#!2{c+Uun}*@}G#s
zdSkB>(JhR5*-|ylj#UrDBRHVKAmh&~D}^9L+kYuKX)Z))NOgaVuchoTFn9sT19;i8
zUi6{ZfvHY*z!;~_q;9Oh)Q&#XMts+AHQXiFCJ=iBDX;>)5_*n%2B5-%D@OM226R*G
zZn#F)hr!>iPhAjgLL9{w#cFtTwgS-y5D4-z6>`T6Fz;<eebt=b>vez=0OVpDnVYT(
zz#O<D=RzETnkrPK?G8e6p`?-YPW{%yGX${1Exyh*l4)jP;RrJefTSR=oc7->uC4t9
z-8%Fsw1-pM5m46)O8tB*f_aju9WGsSy0O_YlQrpnV!g&~goS_Dp0x&jty2^7<=BXM
zW@yl6BaUH$nvV1)!ZkatS!R)-XLjHgrm@(Pa2;3jdE+?cF|8B^WoD(5&DD*3J8K$o
zfHPeXYll|(#rIH3(-&!et|@Wh4x<_;FW%^FNE1uk*l_1ZBHtGkNtF^$6L#lPotu=i
zZv)zqZD0WO7c#OV0K4TH`=Qo_E@)3|S$%=*1l{p<NS8$3*h6q3j^pQt?~l)#3sy78
z_)V}Soxt4s$}&jeO;1U=Sz_`N_N0JL9Xy&7r!Jmm_rn!1`qn|SH$I+0R<uD&POf2d
zr++Z}<`72-Lb4>ViX>x?Jhq1bS@`bECh>>{F^jLhoAMgJ`r~L}!T#D7K<i`-7(P8r
zqiBla`C|Y=X_^yv*j=NKR}ttQiFL)_AFDi__*pcniZHp}MUpY!(fB-6ek&Ifzwa3@
ztPC=DyZ=YicgIuRzyFuLsR&sKA)B&S6dlQ`jI3luviBZwQszM@Gb^{1kiGW~8D(#>
z_vUxKKi|jiulpvC!#VHQxUT2*9Lakib3XwN%E@NsNh6qdu=>E*9cWqx@O{-kLbb=!
z=PvN`|K3ilt@A5UG&i_)tUnR2!4c4l+>$U0B0oEM6fCFLn^2CC@2jA0pMUdM{k1@t
z|INy*$CZVFyXNWQn%F{aq+i$t2+qbD=NR(U2JVtJFOiw<?(YZf6+;07D+2-8Z$$TO
zdV!u0-WLo?pwyxj_5t951ov>2vKV5bq8FhR0Y~)4q~I=HfFHF{@b<6#_%6V>W)#Qu
znA|PihyV5AyYr&#KrXh)?H;>g0GZ&y!ojBm@QqObsSkN4zWRC?^QTTz1R;g?@iVv>
zJV4n!_w#9`A0V5ov{F`9MzUiNg*2dy-?{@u#=g16k%bZV{;)vuI90v3ufev#rAdWO
znA`Xr(^2UCEpG`@y`HKmdV_<bkMl&LzBS_Mv-a{#diyhXK5kT6=O|r-rPuN-yQ~(~
z-d?^?9@rHDcF!B8^NSqJc~$x{xr~oOzm3u{q>b)AI$pte-`BXayuym|=6f>Gn7sE&
z>nhp02SeiQGlCp0S#m>$Lzhbvc09K~25+fdo)PCz?Z>Zo@G}w2aE~mT{ZfUlx%rNF
zceh?%G55T^Js6tf2GN&TTJL#Rtx3NBGoV<zmqLJKzrYG+%hn`%%6%QtLvunzK=}Dm
zxLoq)KW&4B7J?hkbDlHM(|?0brn*`m&TMyxA%cpb;qz<qco;YW`j-d!8#XmZ3LuMb
zf2qTWB{cl2tJBc)bp3e)OO5U!)XGpC3R1kI<NTZP(hJ6tU<_YTR)+L)AZ(VetdYqi
zUPZ=Y+0S5Sq{9&p^Cjq;xEc!>uQVT)?nrws9sD4N>6fV}eV7;u^8n`NnDm;O8j_?k
zI6+vmrwhY$w1A&QPL7t_guz;u==2*~OOlT}OoC_utPZ;k%+ZdULwQ|Y#jOyF&Uqhs
zkz>cv6-^?+Cj&^LnCtEq-{a#Vrt@<qwlkJN$~pKdu~;ES2KZPi7}fx$^)vA2B9|sP
zJYYM+q)(nDx#0UZO><9BH>@?WSM6m4OLj8rGR|C;bN#0Lo?sTW_r!GEA@axbbIAbv
zChTdEX$DiCU2L{Q$W~sw-bzDCc&(+2Y*a(PFMvusKzz2@b?%8#_^=B^+aUEblupdS
zQQmc$5E`Bj70CDDDe%@N_(OjIYT+<IG55f};osiMFo^lp0B`lZ30tKhsL{}C?>WVF
zzUEtc4)Tls<h`3;3f)oRu>o#0J@g28^*6Y7Q7U(s5$xe$1OsS3lB)wZC<p3aWf)7u
z?jGE%prsjtaUH<fh(`oojlT^;gB_zw<MIT@+0N1>mcrUvrv2Ae*4D^N09drQJ}v?1
z9Ig&F8Tb<Yf>&_z($}pRfddIDV%vPu_I^n2`R)*to<a|+CE!}!7`9G*ZFH8~OqaY_
zwgKNt>JRq(SA0UkL3RX87hYdh7VB@G4BQ<*-aPO?oiUv=Vv5ZU@_Y_gv%JpxVp!V?
zP{$?YKH#f}rq-sU>F7<Kmg7A7`}*A%=)0K(wNu&6$Nq^a%o&OKFUCDD*VBZH?{+#c
zqn)0ri>;{(p)n4DvOZfa1oOrp{}F-m=;5PEk?Ba*JDaDF^015cIYb*#JA|~sS~GiG
z0(O9q6%+V<LY|@J@ht+j_7T7|VfdR;P<t*;BbL3M*Q(GpP01T_?JT%aX(2gHEQQ$M
zyT{i}th-qnmp`ZKMX>@n70W2?iHsVcK|~YbE>fWoe=NYuphkiNZWnBb-nnhPqnDi9
z*u0_FI+4hpLOd@I@Ed~h*4O*NxEGn6^<-^5=Usn}L{!<>@VUN&dJsBf61t0-Zllv`
zla9qYW<owUW4$ELy`}DI6Y}!vWJXTcXB3S*cCPCN%Wv~#B3Yk6(GE^bxmY{qr3bng
z@5dPEVKqv}c$+vL6r6_3KhUz~|7>o{;Ij22+7&~QL{nOoy1_L4N<*qDtkH&s{_@s|
z6#XYkN$KAI{0koK-FT2iWfaI|sQR+mYtb8Xg_Zli!DAgZ*8&1T#|Np8iF96TSxdV>
zO@hqAb+>E5*9fYs)bwDnkVf77FOvGnF7Mi^XSC~5@DDs*H61kwWu6;EUzdZ@J{!7*
z%`s%4gvnJ`#k{aL0q^~&vjgF2T#-vv!$<uLXi<y%(GAPtR*f)ep(MelQy3sChVT+1
z_UjPv`~uPvz$Jxg+81Vt%<2_Ceq6P5(l^+;9jP^G?z?aMivH?HD#O($?v#XiWn)*i
zoNbOTPV$o>EfR=kahc0QXbShm#zi`e!fMSldGgzDjF_Dxnd6#RQL-l7G!MM+2r9gt
zis~<vprn%H9i%GYlB2dISY^X!oJrkOvxLwkfCn!KPHayFTN)Gk&KL^i%$4nyR=;1y
zYibl(52Y1sAtD^4-HA2yMs$ocwU9FdQFtUt%5XSd+7-CX$Xi}21>pDorz)7fA<;lE
zY`GI7EiM*WQzQ`?tslF>__?=~iz2Ji<I~m;W!KP28v56Zw&+HtU%DFF%S}n2Y92?+
z-&tp9(3MW+P(mBdP+(5zvvxfWD+>#MZX?bU%u(ZD*be9FO>*EmB<Mp7GlR4J<oG!0
zhhK}7FB~X?IwLyi*&ITeW6#BXM43DuEEcDniC$@vJUVqJ*kC><-@4_)h<oIL@qXTK
z<9T>i`*+cY4TY)Ad4X9t47hqW$#%Bbj<zaI_qpF+FuCwho3ILpU^)I#9GwD-#f(w9
zE*TByZTIcVX-!;qopY(**s)%zc1gv<U0@b{g=ieNla@GRfiXw-@di1q!X(w-7#{83
zCznJIW}Dqbn_TjLEz96f?&y$sHNELFUc$2ZM5?22_hb1#>P88nqOMTy#@(%XQSRxx
z(^8JsR#v~ih}d9-LeCBmRwh)gP%Q-&<pDa`_-DaDZLJ0Ga|M4zpY42`1MEHKY1=5j
z-H868I?s(O@4`QBf&S+ErmqQUI~^tE1n<s_I{p=TC_d`=mbIW9_&?u$#O?%utM6Z`
z<gXhf*3Y*z75<KMIZ8O%noHk#9iqNXh9*hyF~+^{8qT8x%ct6$&j-V?U@BqxXK6BG
z_W-H^7q+}d$HOvX=j3c>xTqbB66So3JW(1hBAKY&3-7WGn{=kD!lkZ0huu@-{T@a_
zBOi@F_Ef3Njf~ES(0%Y;I#6W9I+HDIl4x?}@P<M^qFVl!=eA@Y1u@~<R)Z*GMHr>V
zSJPj4L^IjgDq4<1YV;ubtFE=7eCLQhXGKGnjll+9_ZH>&qsU9!GcM%j#x@O6R|HI^
zcc?rHq<uUmVZlIp=`G`DXEuk}8|3a4t+RN4i`EP4&5~d=A~?eC<@k_*&y}W|W-KR6
zUc<Y!%w7+(aP~pKE|!>tY9mcr69c8lZ$JoOwgq!h;8X`mpHwCcCm8ApGw6nqn&}lR
zbu#J~R-hS6=@h-!&n6ct-d;c9JC80nA0s`;#GG%dS}c~bo$cqjJ2%b`1rR<sl$!mg
z{LkP9&i2*I=CbP#gT>@4Y~5T}D-+97^b8HrQncfvq3ltw^r-D^@D<wU?eARwcp1Nn
zbM@es)A7y0z#)0o)=|ljFw)0*-#^|%_qc`_*seCL`7bCj&65l*cl1}gEnrooJWA-R
zCX4Z|++u0kq4-<94E-mZQpy4q9zh|aR6GW;;XM19iDtjpq?3NfM4;Qd=#2fsLIP@z
z@>LDuo=AjTJKx-zEIumcvI6eC5`bPnF^KhhQ`Vb<`?_~|{cY;zEBu>5RC*H@y`g!t
zE$iw#i7Nr3AC7BQinKrSEVC$>24V%6J%4Ae#TM9L&Pn&51ol><2YX>@b9%W;or4C$
zY-Boos0g|MkODJ{G%QD$vEkefDhGHiI@4n0%|`P_Y6L_?_yOQb$iv5n-(_cG15gKN
z2)05a*JX*lbi%U?K7Q5?UkZ97r&B8W{^}B>4d&&A=qn`%M*=nkOtAZ4%p;oHA<wXY
z-a&z}PRaN06dk@VPhrS_x)+3CGnL-}0YNgUrq38vDyWrpK^bb$A5dq+eZ)yg<Yx|l
z67aAxfYZ0OwuUe$ip26~wKsna8~QM~4#ZsAX)<Hh=q8>imX>qdVA)yf9T2`au`#I+
zzwKNQSp!N@+nK59=~M)Mad!Ts)8A_2tii5q9>#$E@IHc*r+r5T|BS|~4{teGd^W?l
zw#Cu8Ow5b&R;>m#UvrcsuI3j!_aS+Hmb+4<V@Y7dOn=WGPzR&899=Ubs)gFmU@R&+
z`MI-E90&K+k7=9_Bz&FmlT^{Oi2wsyzIj7@vYpC}M4LfF`CYQJGiZ~MBMvAT1SyBB
zt5!4~(Czj0MMB!86A)lp-nksZtn>Rpvd1m&)J=Ylv6ZAt2h%+Ez)a#mS?Xv{M6vG8
za*EdCu$`@yPMX=4BbzL04BGv8HjA$NHZ}z?T^odh3!o=kR~iF~Gp9hb;+&G4hJu~D
z{_EVPkIi2pI9b$|cw-ZYO5pTMumfCp14xo~y>eUyOTyxFkjXT-Rt0PW@srT}rO&~*
z3u1#u@hn_+iI3t69?yOi?X}`vffh#d&zVLaC5c-2czF@a&XtwFWKwW!_+?V&$!i^B
ze=@ieBd9=`(Q=B6Jzxe05E)4oPGUmutME~l+pfb>swRYu^{(TcPD^d0yWiRpzW{Vw
z<-hx#6NRDR74Wkg`S+6dodXx4@v*Rx0PnQ&`&;|Z>9l3Y3IEX9xw(nuW=R3f0NOIO
zhC$0U?im0T5HbV|%fPA=@C_{psj@2s=>0>1x(O`F&++knfQZ2RAJHzUu9o6n4zTm&
z@}l{$iKAdj%^apEJEEpJwvodLHShmen|FT;yNrYF7xd4M%&Ne#$m+8lQ0RbQ26K-<
z?0L7+g2%iuO7&4<Fv*5W4G#6+$lciViE>C`(Z1YQDDBeb-SI}Wi{4H8FT1_)L4i{j
zWmrL700v>0!QgTkaP9#}@?b;tCP@(|N#G9T?!&sTKh#icvzP37gg8#Y`)Z>P2o!g`
zYV5yGWuH8JqXi||ssbe*+)bI#wIZCT#$q5#fnDfVu*mv{pbYj{OrWq;`@J6b5|#zj
z;Ek1PV-KD{HwqOvgd)m3CJ6W!4}kBA^4si2y{-I3-jzBD#+gdLy>DtAGru=MpQUPW
zU+z*V<4YA+tD3OuU!$sbaEOxScSnx5a${XzmeZ%GVpa(qHJ%)b=3s|YY+xwB9QL)R
zryi_602hdc$^bj$erqhnhwe4h*AHCF>c5L*b@uf3KF`wxv?r$i>6Y-O^{d`rtq#}1
zT1Q?<=0f5bv@A%fM_d4OIBIh|gRwYpEyK%7?k<kn)I#^^m0?qRrfKN86^LO4^04~|
zGk9k+y=qx#>1u_f41{z>a)T(WDN-Ig%<qVIT7!XWf0Nz_b_o*m2bnZ5%S0#<lYq@(
zKWI>S0DMr4WvkvU5F_s`y6!*+0tGM8ADgQdP@%#63096e+~WK<;EmC4aB8m8NJdA@
z+PbQ)-z>ui4uu5&Dhx+3mYqn;34U*1k=bpwvWm*Txw!|x_7Inm83y}1SFmg;0_tMG
zsk*@BTH$;5aOtZ7<A+ktFCVz92x3;;BdVWI%XbPvN{yT!>aOKKh&M+wzsjnM$>UOT
zJ^6_+Gc&?_>-K}UH?Pm6RJ|&APblZBMw0IeB0rUwYQp*;Nt|0v_~edLvEFxKVp&?M
z1#F9xlM_e>k9o$I1dbSefOLW{SimhT#jke*H{I;*0V+rEx_3+PU-(hKc%YmM$<5<-
zkxk$~jDG(Axd0w^f<%p%-@J<urno98`u7d+)sXW&3a9rno&uqZqa#@BTf(aWNCQNR
z(=|qLBOy79rnDqeccs*+6q)z;S1bD^Agq;c;NH*+pgqUf{#+EFNV7IFI|Y&f3}p!b
zwz1IeMi^{xNdRoP4CLoJ3dn;(l38G?CHL)o7VtHekl$N|i&BzVFqu$hU~XYr@;f>7
z@$k?EmIAFC7SVk_Y9d@9Dx+t0!J3IKOjR2QAAtLVB<(%Odo%sjf%Aayv~w)=CSXCz
zQ_jt9Q+zoJk5YQ(eTeH=M_*|?cO?ToGR9}xnPUAV{b-IVy>euoZAPq|Ur1o;1^+Go
zbswSo)~2rbt)WUzQw$rpxPCXqhVJ`p(GVJjhlOG7q)|ffF5NL!5V+@ZqG0VIhZR8{
z?FD#%!a<R^tec@Baq%}wAfH2jNmA18Baf+SPt?B^OU@fIfZjr^PJ@J4)Vn9_33k9z
zf#&|@FZfJgtN|YjZmPxWZ#;f!jD4zdyP@IG<_S$z#~!<?OB5f3ybkelv#Sn=ta!2F
zfZ^@!`Nv3f6s>7;=X`?#Jxq~}c%j*(Ib7M?R72nxlP@q*3fP8Vok9o>0Xpj!osl}d
zl({1&w(hS%(B;!z3);Ntsfy_@907SB0~1pYy!6w{Gbm&%Z~O07AGn#l_LYDCqUsp}
zgVQ^8$@y#N^FPZ^r|<Z%82d9-Q7?b4h2$aesz1xr*#qiJ7i0U!@PS6_*DNyQ%wILI
z_8+EU=c2E3o}T(pDBj9-g)BR%$~HWT5N`d!=H@^MlyS7RrTAdAm47GZ@1H@KIA0dp
z*)fImRWdjS!EO)$HYmuwo-xj$h?W|pYJSqqa}2)O=ZLSD*kQ#E=DCVPKzq0KDeJQe
zp9^n(9LAO+!pogKfYENs!c)pY@Ok0!&V!~aSZW-U{;dy24Zerkvb^SCTZlMl<TH4I
z!;{LQbW5eF+&*Kn3yFAE8*&w>31{zx=RpciU{LnJoOJpT)=etTO7^c@3SSvsujbfP
zIa7%F@ugl_GgbpQ3_)PwYzUbycxL%!|27H^3LbE0SZa@Ahl0uIW6`Hi*I^F20v^lo
zHFKAlF)@nyXtrhPHs7Sr$r$q_@4>p5I4^Ox<c4q3l*hVL*5wR0m>2Wy*qz@0&*r?t
zhJeP0+PqD#e7k&OhuY&sK4rC6v~h^ToP9~mO&}Ny!=Ves28@H#wF(R@5WBr<7y%eH
zw*kxs{0_K4vtS2Ju|f6?1gx_}Uz+Z2zGucEr$s?S%Bd^z{?$u*js=*|P7#?~I^Xd3
z%gW5WD<i}7?%&~|Hr(;`bYSd_s0HAd6Y6@5ClcG44b%qMr14!VfR>Ex;o;gkIZ;jF
zQQkw5lgFx^76*QH!fhVoH>Y_@z5+9;tC9w&czgE0Q<SnYU)TX>gw!1*ivz*02M71m
z1>B_n7gJof47(UOR#y&xKTcap)_PWURbF6re420^hm*emOlpT=2AwhwYX@{5jfM2A
z8Nf%_+g*MNgd`;29swxRKes}-%2OmZ5rztvli5=7$P2M0Di3944m7=_%lP>>zj0}}
zzCS4+wsROJoiNK4k?gO0>&oM$z?eR3=^1peG|h^$$1|m$igIecN=)xhXZ~R8$5`hB
zLZj-qGT?+?s)PB9Hbh-312p;%9&5aK9&i%j*o2}Q5E@mWc3rOIq9g+Bqh0%DX*9=s
z$eb+s`O^~k&2TZ$zOIjN|M;Pr{8uPb3uWVV*lx^@bU*_xt^{Zlj?j^2^ml6u3(n|>
zNB5d{pm?-h|1dCkfQQrhAhJ@{u1|5YpErm!g~}+)TzXvh@xWg*lA3GeXajX)Q6Oag
zRZJ11>8?Ho5H3VccG}Ru0FFcsR>dmDp%Vs0KD~2>vsWQBSE{roRIwc32L45873qy|
z838n@UqKWWaNi;XG-+u@Tjy}e?IX{+s(Ml=eeYSDH?&3&Z-yz}+f05Y(nI!yVo#Jo
zl30b#`aw#J1iwh&pVDe+Vq8LO*TcC~KjSxck4TNnvheG#&`Um|Oyqji`%HpjwTpQ$
zR()E^KwlpjZ`9kv4$p?c3%_*Wq=`RpHR0ecsi|Q+_Jo-qSi!v{m|6R&Z7c!vHjvhQ
ze{i*VhB1q)4npS3K6eg&XlXy(slITf;T)9p?s!*c^(XfxL>O+q1<dF-?(udvcu*pY
zP$&&q4coBNR#Qr%Nwc)`+sF9%!lc9}>$r+%KQKM)p*Juvumsi%98L8cs~a2WEr@u8
zVI&HP%Z090P(k}Y276~$*KhwLA=4htfDW7E*%Bvb<F(yBdD%r5ayoSq#)tgEEpw06
z{?u5#e0j628J?xKZgOS`&R@6yMQaf!9Ug=d;b{Zn1E|^%Lfu_}^17mn%GLcBtMFnz
z;LNo&TD4H27siMaXK=cns3-13bP79Gg2fZZBXSu}H|ma7myT&40@=eQIlM>?`qqwm
zPlETl{Jk`SSEfE#H>>4CD+9$1;Wee?c*hSGUJYlf<dVN(-vc)qt_dzz8~(q*F<wI4
zxuFvioOKNPI=P$ZEkvgJ+Lzg_I!>M%=&zPGHoegr%=A!p!(Z8{?ebktHNjPrDb?;H
zdAHQ*gYRD4G?xyvQ?8sj|C)%pW=G%*dL2~9)E`8)@Ivi%)5)u@ein*j#md60vckgD
z_V#xE+dn_e`uyEnLVKS`R%!+IH8d(%02L93Am}P<^o=CuEY21s<oGH^SIyar0Y!BQ
z&Iz#5BwHMnrMADcS-tVR&b*D@(=MuNCvhNb>*ORQH1?v*p)uiP#M*(a6qoDbq5_x<
z1ArnyVTiceGuusT4i&kr4@ofJjC*Ey(BXEIa!08oBe?ePV=Ya?+Xk$-dt@-HO?}&1
z_wr8Ft+-Zs-K2vIzn)9lwR-smR^4&*j$zfrf1Or;$67TXlAX=F9o$KboW-GV6qS%5
zE(wt)S7Ca2*?g51)ld@{CVxcZsV6W#CVS*is3yT6xtf*UR878FJ~Aa9reG*})o!U-
zN_v5jd!*5aq`FtaEA?U`$+&UiCtdo;$fG?f9-2$ZHaE%0y_TcoDP^%4EnZqxU-cJP
z3Ll9>AKKU02VTIn4_h_pi1Yf#8w?C*9JlPxL#q}3F8brwoFBh|fATKKJe`ZWtKnRF
z6~myS?5(#n-d`4bdH<1r-Nrlv)V-<IBmYDd{IAY$+YXI;k+11lRAAYK1Ye9h%y^5C
zSUVY!HMrPdGq*gJM>UfyIASO*PxLHKT+ig{cd7;$H@>i98>jkkSlbr5_DETMY+X^f
z%%aFtbsi-xu8(w~RQ*zxKC#5b;0L0o$DA4+NBnxlvp(S$?@ow1<PFDlpY2|*v48mf
ztGzc&v%B8S7gEnVI?i^F9ahObcFFfesqJsx_%J?Sf7DO<S@E4f=BhpOU~j3NY5PG{
zBUKumJ)a1Iil{~CQK)sevsyPNmcP7X*tXG8ANX4e=hK>4>Ji=LDWTRiy`!?tvRiwE
zb0@A;zQ(y#Q4zRo7}oaC9Qh^eS=`wuOAj$j+%|uYz@NA);EZ=WclPLu<QqOYm+*;6
zzinWAK3TX#j&>gV?$-7Y6^gxYFu8DWyo_FtFPF0?@0O7-Z)HsPF#VO>B_p|HAfirQ
zMHEIf?sk2ejdhR2OE15{Bk|`Z(cOh}qzE~Ynz2>9R8#-njdIQ&H+(x$W@VZu^0WLU
z4)d0W7qd!~Egg$nXN){@$)fIdt*tC<$-UqV#)aFC({ycVh-#+UNkh-tV*I}*EFM#l
zy$5RQS6y%K`u)6eY27LJL+~yJFWW%(k*>EkwU^Ss-Bs}bj9uUq=pH;JS$n7K%NJ*r
zda5lO_;B#~;_SrrXDG&&@tzP4meFfD5i}WLk?+yu#7?`~_k@OXeG)BpnBFR^@D&Q@
zjXO_35AwQtGbMWnUpHMoZAmy<WK^iQ{4QwSsiKvPys4c4&Eh)LO3I^mRFUd9zi{Eu
zELlgN^1y;UY=7*^6m7!giRrVQbN6TSZw}5RTnr9RRWtAjjXZ}ft4IlazaP#&2;1U9
zF>xx(yjAN+y3+mbSb?f)QcB7E=D`ow0yN6*+ce|-1U)%{rlIKGS9k~CrOjR+UjcuD
z%6tOJ0j0`yRAot)ln74o!%A0zn%Q#fRqx3sBY9kWDVL5Hl*E%<jvsFKNO4NOIq2ud
zN{V@yI*r?c_r-bE_8LuHugXj*`cR-}#dE`d-8Rs}GXHRjMZ}$kCcId!N^({~T9da1
zJ7(wm4$kRSnGILI(>0%y84}^{%_G|z<r=1B<7tj;OD(%VC(7{O4Gz!ytgF%MT)F*~
z$DBk9kL*z;)^-}H9)%uf8vV!(<&Geuu-Hvnxp(&O?QTWWE#9lVY)ejxOC+^Ky%>D0
zVS>`q9RpqVyKPcFf#62=&8aU7HvujQ^YCxrJ2;HfYTZl;eb)Q6mH9>PgV&3<<sa!L
zZS4`W(enx3G@6gg@(AgXCrIU@+tj+l?)f~LxL6CjEOpj~PcHU{#Dr@%SFwEG)vc*O
zA(0;WUj*^*Q<)dU7=0wo4a|uyZlRN&JJ&chJ`UD}!X#5AelZf&`)=opTN<AV`BKpI
z*;A@l9?pV{Wx6lT?~t>_y>tD-LM^MM!k;5lGb>N<6>WQ7{md}*ny$SMb@H{V<rrI|
z*%PRDAs*$?e>MxgE}e&NGO0PwCYi@m@v=V7ezrcUHfdrem{{x}Onp#ReRJ3@|8U%z
zUsAq<h}7IHtALA$JVwg%L*ug+94|fE>BMRMiYu>p?#|lwUNzXX{Uq)2HSm4o5k~Dx
z&DW4bG=pVXyU?wwedlU=QfxP`ME!GW^wyTA?BiC;qp<z)U#>EV<obUqsn@;2<*i-O
z>8kW99(2|uUds8Gs!3JF|Idt0sOn-PM0J1VcE;RDjE(p|gY?}(k(Prn-F{xt)XLSK
zxGHKWIJZ-4^Q~yZ(I)5K2jU#`ip!OKwh4SKy(!plD-Jr7KB)}T^!wKHEA6r;(NWf*
zJyiI76ahwD@D*sySmL^u!>CdI%f3A8!<0CAlM!U_NzG`xdLB50zZiS!&`VokJVt;0
zeP~?ivO<g0{Z`>EO1CVoCumDOfo8m`?q!1e6r-kXt+RI>X36H}jAzej3U<}Tk3ywI
znTm)bI%O_AG=^HI9Hu_>u|zUmHS@Vn)E|7?lvvLA@@BB}%v4`}u?QzA4x1*09nBq6
zmnT~h*Y2l0IrzI2WlmjwYEz2g?F#YBea&ray_HHwqFS3^aQ>yF(sL3h0)Jfn=Azkg
zzA?>9F)CGl-8yI*NeMU%yWiY9^`s2TNdCnFW|7~ZurGygtNmk`$M{@c3u6f%nQXjA
z+_Ha{XewQCzGhDRQ_|^s6P>R41qA`OGDuYo%1WtwomM$3M3&U98owjAh}v?)TFX2m
zmbBVuP3Ce>b~wIN;2@Se(_8UX{%4g!oU<?Un1i|G7+q)Efj+|jA>8`)qQiP%ews{q
zPE#E3!|X5W(ur}eY+<{El4-Bai{&g27L22<noL=rygd%CDv5#$=C8eQZe^?lr0_^f
zme$rL$D8zs`L#m@<R5c8;DfE$KDDmrL^5kIOxpqk?3JH%Bu(Ix0^|iM#-uBFOSeo*
z?>MOTtOzp<H2tlN7|?pxHBHF7X{dEGU$4H@lYSTf;gUAbPEeu4(xr;jwv+;oANjwn
ziP+pkjGa=Qo4vif7@XN$aMzr|*RNb_h#H#_`!0p%pZWFqVmvBv!CMu>YNU7w!Ze6t
z5}shob+(CN3a2dNcP`%W@U?H=A+ixoTINc2Z6Q5-qTZldS)+9y3?fER7^pn96TS8g
z*-+bHx<m~;oikR{y62&%Bnq<w43h2X#oAJHZ{Rv~g7H!m>yd8z+vjCJa&EqS)yfe6
zNcDkjWQwnyu@&~$<8=9F)G7o3JLG`hJOGduK+A#f6Vi_-v>3EZjaY#yx7R*)D)e4}
z`KLg;-lZ}?@Fe>gkz-0+qO^nmtv4|MU}CS0Y(p2p{j291fkt%_C`Xk5LF<^eq#cGN
z0Njeo;eSct7iS=6>$=#w6Am1KkeNtIrO4OB&5Nn_Ri6XRGgg-|@3utT7<wMpFb6kM
z{~G()RXrhioqGb@0HixsEAPwx>AMXMJyRRC?Lu=d;08eRUOTvfkK;C3Y1F?7CN@Jm
z)_Kv?@NfZs=hl1>h6yT!dKVU4QUWN_5-8Ik8%!-I{JJAhhGv@0=UqA1v7%VcDaGfT
z+^=}Tf9}G1_7+uZoP1W3Zm2p?3{`^<^YSp;*&z8OPrAi%8r|QcC;!W6(Jd(|@<H`q
zIDb*8V4oQpre*!vw*IFsgqMVZkZr-Ykp0H%k3uE@OCl+*G?nnFx;iWOARXSAZ`U@U
zhYS4tD40kwD^CK1he($?w-}ev(far6f4&BY4|L;3x?a@jpy&FYbMJa0&nAKiO`wTO
zSOK8`N*&s1`B!pdQyjJA^Q^U&76pD26y{rPt;}yVSj8aY4>~W@RnO2E&G3Y>I-t5j
ztsgxi<Wz81*fD8o)8tz7AG?_k<};6)t97+#b|W3!W;(A~7l&Rp<*7!Oo-+6_dL-wM
zh|ZC|$6|Zbw*2dT>2$n4Nf6(|2iIN6?*@_q&TO>IgokAS^C2Vyx)>Na5TyrTh=qyC
z0}x9CLj`F5<g=U3Qp=R=%VV9PZ`{XcodY)_<2|ZN_j+LwXIdPGN3qg|OP|QBPRc~=
z7M$e@V_0dUe}i2Xljq?DJPWRh4FA(j<)8>g=hh+8X4b(*5gdOFhRG`}xLv&#u(5mk
zpy>eR*<e%CJ7o9nH5J+V-l*?ce6wih8w}0~Z%j=e&N>?qw_YDBcTfbDke>j6rHI-;
z#RT}F4H7U4E&<Km-6xIKvlzr`0m9P0)2U%t;o<y`1>Bh{r8@8opyK#$nZtarg9p0u
z^j(=&o$>Erpa2mEYM@_4)PxgP92~s7@|KsE>q2sC0QOfJ`1_Ac*!<;z<?*sl+AhQF
zoW#N1lzWo`1|boAYde1*Chd{>Pcye}FQGlyc3VqVX6#?IMuaKSdbUW=FjYN}p*7Df
z9tj<ndb5DG71y-iW!M06j|^>q@#^jl>@6TU80Az#m5CGJ;PhPL?@!RHz&<1GYun#<
z#@lgb!{0bbXq~R&dfeJzOg}Jc%W~XWe$MK=a|9o4@BYOD4C~}0@yOj|SlK?P8<=A2
zgIwU;v;YOrFlrYlYCW$#%mojK;vX1}Gn;<2DEvXj+nO2@Ep_K^axWK{MF3qI<KL=~
zce6{b3?iAqoSo^I<kNHBSWpa~b=7!<_z8&H_B64hzQWKW28hUOYn&kZeDz8W(;yu{
z|2TpbHNs#9PYBUtLf!;N2&ig6T&Gs9x6ZOcE{_OVR_WTxXV&xTHu)hLg(z!5jD?Sz
z+jEZjP_VJNx%p-B2xr1MGr#qvqsjZulo;dS34Ce}YT;hxEQ=de+*<sba`c$9;t0ML
z#%fhYqRCy_SoG%!CJ~CTX*)jZj-->*!z%}#v#e3GaUu_`tPY7M?RsH=2cG&Ms-y`E
zkS#A{Gk}CT_|>{6S3b2yYLM79GfLgvbXQy7!b$I~9q1heUB3fo8`yOjgGtZD06P39
zYiOoLMKWtUL^pv<FX{Z3GX&J}plaC$op;KY8Q(GBMzfa8XkoaOf1)j|GsZtxY^n=K
zK>)q8b0**(U45kVX||r8xj%pYBryZ=#nqtzrIM0feu3mhz1>tb4wDWl(bdk61_cEL
zuy$|Iq!kt-1C7g`$XNo*Gr*<l`}YWx-b<R1z#D;_oKp{6->&YUwWj4D*{uXdXe5`2
zYEo1012K|TlFWGC;o+@p5MhGN0E|xH%z6q_9Z&`#!b^B$rb9DdDXXLhVv0N>cpZVN
zyzx<b8YSm}qUCFkVc0NQAN#U9%dO{@_AX8fRZmft72rQ3#1tjZyIJ0-WLM&gE}#0$
zOD@f{`H3rXq`dB%n@M1ZzJWnd8%Q|Kg!usOBJ{X1_41|zaE7L<3;V$Gy5&Fp;Fjks
z>wCvSl8ZYB^7;R>7Mz>J>_UQ&!!jDGsABHy)9H<ZC{~?_UA6w8bR!SX!$$Mc3cZzC
zyE=Lf5}Mx|p|~v@;e62?8Pxpuy5^$r9X=_sApt2Ic+^d-*`|JQMhT6&_9Y`{taI^;
zJUsqt*g;30YtI@X9R*N3WZr}1%ZQ8j@9+Oa@KjJ-a!cqwW<&9$i2_OubRYbA5g^ok
z{74KS>^o9JqahG=BFFQ<q9n?1b9T=F{@+-YVIqQ37|jEq@QIfpXaC5^Hy|=9H?Tc<
z3R6RvgB!xV1ln4JxTdDYFYe#7Xy<lK;g|fg<2cqo47F}8%RN<19lv72*>xmWO(oY;
z#V_i?)HYw?Jwcw0f5=iAd-$uyn}fuEjwYzSa_RnML19ZJGQAG?`;3og#-OG7JO(ae
zM8o`h?#CpjL+cg^D0k_d7id<VXtS4J{^#$-tv3;*7*R~yczsKvX5-ixS<&98ejnE5
z<i`WWCdrdAj$py&gkg=sWU}#OTDV8Bwj(BwA(p7mK2Q67JVp=Pu<Cgvl*6^IRQ{`Z
zOcphpB0;ju%v0iJfX{A9oH9`SDfcx{2mxz1^Qne$Az)^0%08l#6`=OihOeI^gJnRl
zm><Oca|%tmRPJ+81B3p6`91q+xLaa6=cmfUAg2ZM5sV6u1TcmS<Dh^V7*7G>#vv{v
z#v3G&z>AQ5t+oFKF*_7A#a0w9RKv&&`CdX&^!9^LVuday>%%4Xs%hW^fMpS<tIWD7
z2p<zLf}RVhZxSeMInbY<kX%t3eEdz%O=3gL@&1?1EE}czllrUU5@B`~G-k&Z6j5B(
zNr8+n+Yw3X*A4J*u;tD_ILiDr5tA0)$3-*7vWtsreK&z`_UOaWDX2Fmb8bTQrH}H_
z)G)-bu3*k9EOl1QP_-A+^FHvQ8e(>Pm5%997(LW~O2>~aK5GyhGJX4W#pXGWn8ONA
z=<b%Z6Jt<wYN(J@*P_8qgGlsa>k`SkF!6#RafgY-!Tx@O&Y-IYKV_w$R?c(SvIwe5
z;2uJyem&q7OBBHbHp?!gR)(0%%Lf2)!JndzUV8%;Acdw~!ksl3tKs8K?}{w}iIWQl
zvN&O^g($nQGT+%B;NbY<h5`#g=D$0#mz6p=3V`<e^QSPlsRWJZAyTN+mvCmp@ASm$
zhJ>+;R{gt479jP`xqioltRa^5!qCI~x~|rz#6=&&o&b*E-_JgOxMRsX_?wsJY>I|3
z!gVP&Y$u}F(27-oghe<4r%j%YjDp3N=~WY9M3e0=1-hRBDH3=IsZ0r?&J=ZS@5k5I
zpFP)lbMs-ahPtsjYqML~L6W52`H8pr@{b44OLf0`ipWMt9*jBc8+hdToV{7gcH<@6
zVCQ1pnW`)g#>>JMn14wtgxf^8c!eW($B`GD$Qlj=-Z?+&;*Y7TODG~vBp}vdOC(Dq
z+)IIBck)<!L^4^&kK2t79%aOC^qy~@i(pAGT%K(AwrZ}E$zC;~JYm<>JKq*QIIl^c
zoLnffc~nuH2AT?C^NfS*MqQ(EJ>(=Lb31ZY%GVR9gf-kJ9;-@=iN%u}7`}Lc`7Qdi
zS;+^|hLAkThw3K#3-4-I{+ZkTF@h%`{daJUe-@%?*lgz?aU2lp=~zS@#2MdxQ*Apf
z=RFk3qF4--_{7A7=jsw{sU$j!I_L9m93MP`1fP<^TtVFH*s;gO-Q*0SBi#asxiVGx
zSqTo!K?lSnHci<8w?qlBT(26x>={y??D0(%cZg;zaW*_=%xBz|h$}Ky3Cm5VYBcih
zJFx$!LxNApi-%euoc&f^yHZbvDmQFfe{q>Z#}r4tqwPRfSU_fFSxYi=$jy{@FGn*6
zW^y$U`KR8|+>(&(GjKjeHXP-<;$5_^$ekkZ=<#PV@=VBShx*1VpW2U9tqu8umSW0w
z=glNp8rdc{XPdMGf__WlaDM`K5*QQkp<MyHW(!p1Tn0>7HTT=MXE)M*L!IO0U&77H
zyO1qtmJhkvy*mwKMr7)_toUT~ze4Fm93hXK#4PD><>0S&*N5)c3^NGr01`COgNO-}
z_ZgDNofCZ{ssqVMo@om1Zy}WC&ehC8$`hzZbh1gqpwB^gQ@gvxFtEG}*;}I)x7G7B
zkrYS>Zi7N`tS1_SWE+BpW4z2x9tgq}aAF}|Z@Q({7BK(D798NBcbvV%u;Yk#<3PQO
z?%MaS0bp*z6WDI@U9Y4uA@jGS!)TUPi{DqxqkDoX0mV*DvMU>I>)G}7KkUAoGGa07
z2-bIHldHs7;?4-@-^|95>oJDT^*=b)p#B5$N84U0@6gH~Mv`0PfJww4%;NIsok_Z^
zNF5n%^8n6KEB>&7xFk(yT-m=L#TRdLqSNi-8Kr^>qU_#&7<~Qzxc~~PZF=dX-5pgw
ze|`#QFX^Mxqj`m_O5ki9-4zWY?jjYi`S<kPsw-h)7L22kdlQ#AA4U>B?odJnOct4k
zgODYH9IHqVP8Ub08X!LF0lZ6bhXrL}-XLyGaaEfcb2Iqd1KiK{hLcCyM}225QwU~2
zFz4KPrPInMcott)tRsZW+`ps4>dq*$g88Gvr|$2Y2bmfd8E0@CAF|(zr}m3nb28MG
zA&vf&*l3$&@`~dT-W$dBD>mVn9d0@@r6Zx*SaHSMbPkCoFMm>Ji)FNV8U2Y;*A45E
zfH{uY?e=H>3`C|gKeeSlU0K5@hg(qjJLfp!$XjRPwVk`XaBhBKm0figr4if={y(26
zp)6o2BF;|m&5|VyJI~J=Pye0ZFE4M-*pmnxdhP8)Myh?6^aV?DXEMLn&`@AL@9dp=
zVmoK_k7*>8-{tzCh}&(3xL$0}wfEl)99@#kj)AuBMUUoJuyX_1LTucRoy3~!qzMX-
zh8RgxK&D+#z5jv}tqDs^SXlE_U6)ve;i(nc_grSdeU|e2frvxeCp!Gfw*RU=B?r+a
zawY!oX)}1$rZ{fWgAshfIj@78P(JNKtK7Xna`>Bkg<`V%?UHr@P|4Xfu7ASTVQ`9b
ziUq?aBmFYl>ENUK^=qZfe^W+h)6}^=qDWU9yU;BJzEf?SPYP|8ekH?_31FQMgHU1G
z^dY=JQGe9XR7vNqf`HYwplA`v!dubjZ5=OtHy87{qDCgH>NalALKoCdyvragQQGbz
zp6)VQr7)UG`sW{V<KMFWI4O?{kxMW(@1SeEQ8s%i2p?@#Cahw8K6O!k;<h++tS6I0
zRrIqf?wtRn1i3l3b27h=6-!Z%ZE;lCyPkgE>3y*--Ve8gq$Wq9L?7TaOSJ1iybPtU
z#s-k$*soBqJ8IpD(Wuon%l}fKI~k|ASB-;O$n6$1_5u@~yYR?fnR$!U_u|wG&qs&r
z+^n|+Zo_aaC)IHLl@F$Gyqp=h4NHT+#Fu#$?+_&XmZ$!qBwmsgqklPK_j=dE&yk|s
zrX%fDfu!FwbLPW#@`W^cRYUe={qRLbn9Vf1igYZ}7Z~5uKXhb&o)bDB#Hem|*f%_^
z^~x2Ggv^RU*)u$z37?uKNN9E8?rlK}F5dXqn8~%2E&jWLl?#%}K{Klo!>w-*s+6&)
zcWi`3!x|*6zC7>1n{l_LP}Eo!sts^5o5%rcJ~#{AzIPreD01OhV;3`oo9gBX4@FIa
zJaX~awJko4bVW5!^C_5B;pu0IlX!P=3Eg|}(zK%fEqm(A8i3*WB<^)WcXpyvnkQpc
zo?nFjiDu!(97&Vj^vt<)HFs&-dYsJS)C@%WB92Qy6VH|Z`BWvlNcr2OGtN&RalXv!
z2LIi^At<qMpZl(dg1hGg*FBMp_fxaxQKT^Q-RIzv#|jt(YaUKT<Cm{Qmpn*s;D`^0
z`)<7H6Fn1?7Yq9>*#0(i`KR{sH*Ie`d2@dg`gc=Pk@o}Frkaeon4VtA4-l!_ELy6g
zulQ_tl3*qCu_rvF&&&LJgJj9Dt5C?ISAM~T$hBK1FF77BPMtf-MCzAM<ZpmeB_ieM
zk=U5q04eoCF;S!E#fJ2O2dU!__V98?441gex*n&!z|AA5+RieZH5B6%9YI+2$n=xK
z0)vBZp1lvrWOo?KO!>tUWiH)$(n+ZL=j111{35+IVY&Add{ntAv2OC<xZhEE>8uFc
zvKWQ(F~Y8gO5If%q|xP%uwrDuM9q(kK|CE|2?0M7b0HxeTR}?=E`r?+!ho}A+60rk
z_`DmZR%831qJ-FHrT{9fWyp7}x_rq+ug3G}M^Ca$@=EUkrv3Uu<0xXfY2fR}Xc+77
z!hoKQGhAtu#zGB9U2yPcT;`)AcQ6DT#;v0VEcAkcf*eow-|APn78Z|f!EemCOe-O_
znB4ZN-E7<%?V*89QDux_7>eQ2a-43!MfF3@9HZk3pf@75M^CDc7QTS1PTR{~DIS^Q
zB^D%G5*i=3?;oG4bH6Fjf-vj0sn==ua=;?$(1O8!PA)NHC*t#fSMD1;azz0HyI0jE
z20EGE3X(XT%_{8&VqXhqjVB`9Y!>_yRhVagp1<I|va6?sp~g;0!(QR`i?18D{QglL
z70ovKBkBj&rG9_A{Juo<q1jDLIZ+T+`l!JDw-24-<_YtYdbM-zg}dnkak9|3Rf$NS
zGEn+>DUHMg<fMO|0>MO(j{LqK_oo)%lOZ>>Uvvm_Ce0XLQnQumLP~H<Dx9iFHROF-
zw!b!-m7mY6UvAG<uuv#>oQQ-G12uZ*0<`2vWFr(CPE&P${Z(Q@%7PUW69)Q~&QFkx
zONhYTe#S`WH#!c)*l?M1ZCF7-!0V<n)q@wK`;QgEi8Wzzj*w7*ee^{!03>rrO_56+
z$zYo@2RshM%MKnU5OQLZ-5~pM3?FXMVc>*0!pZ>hi20?T0)o9%PSJg#C!gpf;<li+
z{?uc;@ZXIZw8EMTKfC44ROzVdsA-H}WjM)5&FyT*4ps)1r}t{C>x3IkYx>@n=-U~g
zCl#?89WhQH)yY;k?y)EANEBcBOEhVbIAZaK>JqBaVZy)ov#;#N6_<_F&8p{dF^?=&
z679U(L-IAbY;bhJa24Y2fkr;hZv=`xu$3MLB4m*waEIHCy5Sf%Fjp3S-E2g##(5R4
zcZ7jkS0|f9)I@q2u(WNWFJC#7x(0gpz2NhS=Y*Jc&k6&jP$>rEM+gBBjGbcT=ok+y
z=7l-_sF(AeC+!3RQ$3m$l-cGt1*6g~Lpr9UEkCee-K+E84CkBPq@|%j!o`7Imj?(E
zVj29ijekI*{jfC!!2I*7pb>>~0Plcti|~rrTFn=S1_sjn{QSB^$9;xG!gl-LuenOu
zPB&yf9MWBs&e(Rh%;br?m0Eb>&>ICKb4{@J=E9J;PEa&esC-&E2i?FhxbYy;Ey!73
zSOkD&Q7o^ZYWoB1FQvWjHA9x6CCgBI4#If@wNA7~KZtt~M*rJX0K=eS;BgpamGXAR
z5y_-43no^5V{&6*V2&w6A1lI|a)-X_C~{x-wXoX4@F@v#a=Ao6IzA2w0UYVTZyp97
zr}cOR8%m74RRsvEutTGPcExsYqyfBcvh&R-$<kKG$E5{DvAY59C3{Y*65&_gIBRwp
ze(j&i-mb2>wIK~9sBGQ6nX30-f0N}Q#Lu?Rq~L3?k`^wlVQ6S*Ak_0LAh{OY{vOBj
zTG-jCBQ-0)`c7+q@4@H0K%r(y12}mpxc=BPeHPQa)6K%agu?_n<(?iD0D4u1^YuHG
zBd}$F14df~!nkeUVEM(qzCMx(JsuOUEnazy<K|TTs@gdI?<d)wqMzqG_ysdjlC^z5
zULE_&b`tojN3678qmgQJUFf0(5CXK%D@bIjo?h3^`}Q_^9UZ>B9hG|R1)%MIPD&br
zlW(xEZ=^if_xg$y(AiiN!@dDvU?X=61@9Z-EP=_Qiq50??$>cdU%S4U^OBU(e%+C%
ziVZW%GS#}oDz6^YNSR5SnJwWEc^;A~P(LKM|KBAUA7EJ(yxp&!NUh``@;cpWW#fkh
znhdH2L?8)WE?j;6&jO*rx|-4#H%5k51cXg&7l5lCx7#j`N}M*gjcmgopOM%ir-zOX
zXP)g#gW(`(4}|I@Ld>i?f3Upndo3jh?4|hl_p#Y-$6<efx@|L|>;eoRG&z%|I%S33
ztb((wiSp`>JemwLpp@X-JtXth2NMGl8&y#&g1|k7$pN71%XNO(8Ik*7`~&2~IsHL+
zAQAEDrkm#{8?btiq-0G)!(`Y4S)~*1Ih5VMw2i$owm+Mz8W*NTM(->+ZmwTj*b=Ri
z^-|%qFg4AHR@#o{8<m9xVTOj~%p<@lK|wa><l?ddzX(3|p1$j&RBuCHu1&%@tK)-2
z5&I#YwEzD77se!W{y=oq#g-ymm>%ETb@Rqfy?{$+f=L&{M*1p|uZgyj-;KB#)~&b!
zBk7(xibm$!t>d&O`)i#YH}B!pnD;V1zY{trG$GNKx&$7IK>I7UnUt<R{+E-%DF$f1
z4kry>WuusYmgSH8DB5DS`R~6!)MZmaEj*!l4xiKVzrh^$`4}yw0BkE6k@PaR9kU+Q
z34rxf#ewkFqkOZb)4~@{1jOYP0q;Eab1h3gBbp89dEbC<C=)sJJ(vI7;M`SN0Jwxf
z+EbK|_MYqYhmi+~^dOT3)Op(hc(n04Ng}p;=P3a44a0?s%nVk4!HR!YFzmFu)OTuy
zebN0zWba>2!Gbdwfye@1(oKz9T&pwcX@g{?|I=U^=0l(Cs+tj5)9MuC*rc(WqWNt|
zTb|_LzB*D!-(&wLlQymXw`@D*ySI%_((o^4mGCr?y$@%tbVp+^BRzfBp2_;}9|$^b
zByIXy>;<m?$?Z$We{hcCY)sAt#pBI6p)AQ*_>W*iw6V#$OVySrAkczbk#7dP)!AI_
z`(%<g@){)<J6fhmA5>SzRez1zhU8|Osx1XhnMdJWgg-O}X2lE@DCT-$cW?iAFg$65
z4&8+#4M;|VAjPz3id)8CkXzZSSf8v-4sFL)2T=%{4I_T8Xu~pUxNkHnRNP9+c1qR&
z#e7c~r`&zd(tykYzYJE@Do%l3Ibu1;AnwRfbYI2+Y)|0WP|7Z3GE+jat0p}w`8$%>
zK1_0VC=00IhzCQ#-UdX@Ls&hK-7_2(dt$aH;ZnG}_ih!VPQ*~2NU--GXeLJ|&ws))
z%!lg}>4QNn?e=@p7nN9;kjS-2U<Xbz@L+aw@fO+J?aPgVPJoy0_%4xf=NzCaxAHvv
z*P@{x#AG_7f6nW9;oMQVC?wT2NT^x-iLrAqsTjQ2((lE>RKChil_)VNhz&C}ogEi5
zdCfSu@C=_TgDQquRjGSkfBhYldCL$DjFfhGe<LHWK<+p=otb<e9T{mutBWa1z{>l-
z^k(KnF^5bX+pd6q?JlzM>x-OBGTsRe@@9c|L*lA_l6V?)0q-(>VVh?|Z~Z9~j|g3n
zPGj&jc2*Qpd}KV1RZ>B>C>fWgmah*N81|-qI@yML_P=2|c3$z}m2OY4)v=!(z5K8d
zF?R>GIN~1&CyAN4X(MpEdjlk)WQ2EqJQYCh2*>>3iWA(pE66DW)!?(7!QE*z(|@4U
z$Sa-D0C9=*>83pL1KabRpfzukFpDbD4(mf*qhD^yE-+JqE$Wt?d;gES%FTO+JMqQj
z3T$|qwY?-UZW=|}Sw@olibKs$Zh!dt3-*!YLRWllZGO~&nkgKV*6e@M-*R20$54^w
z3jVhvblON6BQsmXaS-=n6b1Fka!>kK{sXpb9KnRlml<2tcvxJdJ|rACzo8sBKR+k&
zL`<7m1_7LL`_~6+WV;<YlJ)g<$qFkzUereIg?EzlgI^&^vl+p#MiI+$i|^SQ7Rv6@
zj!SBq%0i~YaAU~By88`YVB6LYL0dBfRPCL1dw5TO2)azbP)*;S(ToqCEYCc4?fG88
z;&VRU+zX&fLSbG2h^dwD<nd6cl1dxlnU;auLk(l4@D<zB!!3<6Gh0xZbWC|eaqZ-E
z&qd(Q2RX1uFmq-F5D7VhCR}Gpp*mNG+d|yuylTSx1S!_l;b$@@+!h&N^YM1307sLl
z;J&e0kBWv_tD~}AesW@5V!`Be-`f=q2Gxvynl-1-enku<M9Ptr4~=DQ;L1=8qi<Ia
zNbk2-`r$2PH$3X}bKR8+1VMvfJ0)$vod4(N8Le9NiI+O1j=5O!a}Behp+cH=fs3~h
zChvZAEI-(s71F<zBAvJTfnv1H=HXX!)Ax$FEL5ce{cTH;>D$xF31;72HVi<8Oh1$r
zZO_}8xUwunmRI$$bTQg3D=Jze$v{X@5VNvIw)1Ej&I_}b(38Mz{tQIl6E?jo_HRl)
z!=Y)vBnqYPj>&|qyhgB}n-j{%rxqC8PDl~3vez%Z4$l~*x(n`ZWTr|eG#Nw3wZU@B
zcOCi#!#g-{u|H-Ru8voh;<N#TRIV!3vD*F@x$~+g;;x!B0=sn>%7mR=Dmsug7S!w{
zZY-4y8+5Y#Iam43f|c-;r{sl$T&YXbf<_CFxHVXff29_ESB~4+o{ETPrlnlGEy#}g
z^lvtxy(Vp40}bO`nDgj-kOS0UJF^#d4m>c=d=dOT4NLlKf>Kh6_@h-yNcgqAy`aS5
z(=3&keOlWKxI7o<Q;`kIGP&n2Mb2?GoksWzYgr;5s-v+vn<?|rY0{$*Q3kHRV*?%e
zmmXaY3g{_`UvNlD+As<W58}wi$=4KJj0T6@^|A7_zC7a><6>Yf)+MG`+sQbbvUvI_
z^zc`zJgh5NtuR|329>2lQs-WDv`Y^yEWNj4@K;*QO*jech(h>z{z+Tgd(e7-ZQPgM
zFgvo%wn&!$hF73Pu`^SWzUu7mE>8CZmI;_QJkAaQMzZ8{aINC8aYjt}s{U(v`+ci+
ztA!2-<iA>(d=eM}<XyAtMXQvfa=WK@;CkwDa^qbB^u^+CtwDh~f9|t86}Hp<h@lPe
zuE87vbOK!Hec_uz_YP`LM)45Sq((IPqUR#0rV+I8kFXB&Q{t}b3NPnJ-ssWS`_J-(
zg+Gr&jRAcgXb_UasdzuMY~6r+<T*T47`{DrP`23gHiN&J&ZbJe{|q2l<P|w})}IEj
z=w!35{V7t6<<-BFVfAmXf>*!d8PPDI{cT3QGsIs6?*7;W|KIP)xJ7~*xobogERVjB
zz>z3dndx9>xOSJk{LLjncj_X={KWZwj)i@iudQwx=gGwrL=Mo$t&WRJ&`CJzwoMdf
zXk^c27z@+OH0Cx9pBx`U3$Rj9n-u@?cf#*=D$$4Pa~@&>dX^C1r%Ug35H_8+eHS{z
zBT=nRzKj;4Yxz%Q+;?B22&L|Hb3VrNmKokolLYdtuoc-=n&P}Eyzpp@^vR#zg=J2>
zyfnpdAf^iyJ-IzCh2OvJGX9nZIZ`Z0?K3p`Kxz2p1U--d&O*CTf|!*9c^=Z()!;m^
zp#5Y|;m`7Wot!$$h~}~mBiK@MF??9?;AUEhAa^jXZTZ$g7M3>QY*38bp=kL-i|pH>
z4*jW}{;j1lX|ketX`_)&g3w`<%Bn1Fs?G+XttVcib4P62ob{KB$^SKyy;G-2)xM#g
zPT)kyCY3F-`1H@m$K8Qz8b#Pu$1MyL%Vm=r%?G<J^xUZIa<@A>eCuj(3&*_o<;gCt
zXz1OO5++*|Ka<R4YP0Gpo2`;4dPw(aw0pB<qa^ZCcU4q<sFMQ&NdmWL&`JCDR3>$o
z)V5HE$nA^q-mz~DoOs!iPrCNLz@~>l2130LU)wrW7<!uTV;|gyxiw<?B^qH`UcLGz
zXdQ%w&WxAlZr^~i>JRhkWNqzTjz#Gi*<@2?+kKa|PaJRlNhiPkZalK`lb`&tu7gim
zUT)M$04I)5^4?dnYlq)1P9>xpf1=QjE@de@zc6t1Wfe;I!`#!^;-FD+i|9}cXh}4`
zb;&j&l`7)TVI0e}p72m8OOLQ{c7pj=cJp&qaesnSO#+qN%*9fBhrbb~vNui39yqH^
z?VMH-sCAVKc!pX{P8NpnTpJp7pzdB;)z${X{Zp%;k9a~hdNVE$FO^syIVyGb3&Ni!
z?Ja{(@ajwar2QnVY~p_EtgVmAlPB|}B4fjF<N{UysXC=Q=zAZF%PuY+j}Y>Ef<p=6
zi>_hyR(<@@@OfEtQ+A5P)Zb(GIsG$N78Z{-4)sd;#C7(+JnriJOJ;l=U_rRjt<srK
z)129d>LA>RW1luz`POzWO>1y>BJLzbg!jYO;=Y{8#m)8==Oan1H2h0?_%dkJ8^cBE
zuxFa3tvPNaAsRvjuY^%k!KuNqPd`6tQuo0M4t-&%tX;}{HJzJX$gDF$i2Pf+v+H@6
z*MgYWxy3!z!Lvu^3vbp{iv&=A;#=k<pYp7~A=^GW>PdUyZ7qR+DgIwdfU%FpZR^-r
z(VD#>T^*)4iCz0{(2AXmRWwIG?HV|;$q)T$ZEKL0uEDe=vSC!K;r&zK#|s7FpS0Jp
z_$GT3EKd!Tbs}_*eFSy>5!2QtmkgYy6=VA<bDn7`EgLB<%j=!YNB1X2_j~BdBsZGZ
z#%WX}{lB)pI;_d)Zyy8!l@93=kdy|IMoI<}Ly?vkrF3@+C>^6DMhpQ3q@+`%W0W8g
z6OfV^-SInozwhsQ|9N-qpKZ^zJ=@N6&gb0s{kd;wV=DAa$N0Lk?Cq2;<n^oz(M!_@
ze~D|!$k>fy4J!-ocp~R;N&QTq{=BuZk$LYQD_4*UM}8L!Fa3!DNn}3kse~Vb7e}(p
zLx=|NI^=TvQmXwM5^_T8_6d(rL|`f;jOIf2m5eraA2DAMpeLl0P!%6>ZVNe4T5Sb7
z2lRckm51YJ#4vt|9SeX&T8tzkfs-y+g`4+!Bvn?v9~p@gUh?6rT3O!hB}D1wT?~Ti
zM4G!56gXgs<jAdJ(q18BUtA<OF0)cnC1reopgxYQ;Y&x<Xk?J@HPN!3RQK-C7)-lV
z1O+hf2BTfzq-@kxZw8hPX#Q$G-9)DzVAwAm>#x2>=bxlCCqNsy*6$^mrn$cuV=jUZ
zI|f_;7CQoIrxuh0Ju}pyOhX*GJ{}2Ou`RL-q<j~?I=s?bhlw@)fq5@wE%No*(y|DF
zqaMFzy|rdngytXDT^ZtKT7IF4_fOcr?pxdTVj#;qC=6$M7isTSe1BJM0yZ74gLc+T
zw~jAwWl<YV{hVoC3z&GbTa13D2=AiB-O&R(ab^qa4`Fez_3h5Y#6)r|k-_Ejb=&l?
zrHJmF$Txp>&%AMt0CHk#<n+hdbVhP@-)<0ZY4$s?PX%TA8o(S*163Js$m_&fSAx`K
zun^v88v#OigM%1v0|<LvoBtnsyr0K4A5cVsTFa^BP%q99OxLEhc8t-%UCz^6QOJk?
zB+eiI<10wBx!qeN+?&sJr8Krlj5zjDd;KWNqJ=9eGpStttfLIEp{z5@6=k6~_Ns1s
zPKR|Pij^@WLFI?E@xi}AT~dj)#7q9_r@jMvX3AZExJ(HXi5SStPNlcs_En&meqNtv
zSi=MF*VPwc?>FIv^r~<>MFijpYJIW09gO>Apu@rJI$dXct<TyxQtz~?7zgS`-edsz
zG7*A6*}lV~MpL`_mK_L8evr4)>m&;l5yG*vvDq&s5h)^(yU+%h4sZD81j(wpj9d~X
z83~+Tn{iSOn*q@go0GBniK*V(sP!H-{~Exd8-4aaf%O$wv4MWn5+Gy*G@*p5To#$R
zNu|sV16v*Nco27jeQ(WcJwJE<n=0ZFBCd+3mQmq^pkgS*@Z+D}<1QqDvf-lJJhsx2
z@y@7&$!}Zbb$r}~i09k^Y=s9zVazR&iX0_&aJ{2+)hH&keMqkSQtZN~lo`xQb+LUQ
zwG4FzlOm<CP?b?wtRgqgo59_Rp%+-fa#JZ336~lr4bh@$C^GgZRgQ{+-t>Zx@P)mI
zn+01I&Su{C0^!5j)}?gBJHmw$deOW5dHbOA9D%S>9Y}~Xr#lamw+eonVqvC_Irgn;
zvr=jQ#y8#O@<Q^82Fvp51dN%&09BJ~Fx;TF7+)Ynv$e2zuxf8<^eB0HdYWO6S;AQ!
zz(ftL0CK)TQ7%EYY&k$Kx8|B27a^dK8<ZJvFY5Ph8++A_kH6>pk}gshOHM)4@nbCk
zRYZhA`Wpsc5UE)v&8NC+jeQO1U#;O&*6}alit>ui_fE@WoPs;fz3P7{G`RcB_JHN!
zBW{&pS*ll(3o9FYg%U=cjkV&eDKL#Ttj!5S=f+d^)OXmyl9-<-2BG|E@oD_#QKjic
zmaYyVOQ0-mX7m(svd@!9$U}&-NA`s&iKNfLZ*=v(@Q6syT8BVrN%qH%;ZwQ(oGKV`
z#&Gy%|K0(5cC$lXEwVQuf!K%#p<tPvr|1==ujhV?fhKy-3O(Kj-Ing2+}cI@;=ob;
z8DNQCn|ykQm?cpR65B=Lcf37x{k638@4>r6E6_XdUMo{TD#;tlrf98V49|Em934o$
z&uP6`<IfY!7qzIML-{SA>t9l!ZY-%`()KgS5Tke}lbP)^-Fs4q&EPzul%CK!3;l+A
z))J{Owq`v|Uh3FzM71VUBCIFzgUf`cx?`a<Z3=bOqi_&8H!)lIK6`8Zxt=L6{(hn(
z7@?^}u@QZ~p&<%c!w;)qnqvR{z1gN@YN^BS!W0eq`ugt4D_<~8b4+s`%0oXqY2^1$
z!|oCG5PkYk>yqUQQp9S`t^&`YpI34yef8lT!9Pv$zk(*8*JVm#9D38q^Dd;lEl!&l
z0q!*p_L%ApwK*eXXW1IzC)Hr|WW7$MGrt0SfO=dJ)NZqB{n*iKjP=fK1L?!a%jfbO
zANE$Wc>eBD53}e(zWzZY68@5H0}r&PkP^;9cgPs6+UvA~)udr0T(c0~_>-ph^tVVe
z=FUHj$>lh)Uq60GhZ|8U5Brb}p4f*{5{;u`PalO!9J|EaarTth=g9hGe%t0vJ7j_C
zVJphv4?iCs$BO>dOS=EB1t6duHp)3@$HT5m#&Ss5m6Cb3;(6lZk67-+j*uYgx$L6F
zA>1Ie*ruME((E`sJ$IDq6q*|Hk&@%wCjNT_Y(3?KMmnDb=T$1&AJLv?xIP?#Qa>^Z
zT?cLt*{|@0z9~#P7&M?wY(&I}cMKdEI*!jgj?xZ?8ae>}y2`jajlb>1PhpjP(;<vT
ze+8SuH^8R!riOk-@;V3vj0{_jm2bK0*XIUr6$Oq|QaInK(uwCnY4J#9tn32?=Bu#=
z_?SCn-G=@04P<==%r@+oVrtjkU&zpj2YyNX>SxSJfq~I>dK{ObNIAH>@)*WpuBm|B
zKsQVN_vq-Z(%2Mh%)YHlHC9ksX0jx#lI1e<5B)y<AhQc?YNx-*7aEJ&e-n><R>}Hr
z*6p@25B8<XIwTpziE$9Gm@Za%^>Tj**DdyGCR$2`mv74BglwIa$)~9>oL{O-LAA>I
zT2<8zpW$CH<BioBx9HKQ9~z){9@1l*5CS~*Rsu+(^Lr!R5(^WD^aHcd{Z{WB@n@`6
zx@S!b$D|?O8|127wAL4%xwt<iM9nb|#;XriJNeH^E_<oRI{#{F_QNj#a}j8x1Pc~-
zq5sLpuTQ|9MKrFW9-z&sKLL^{225^8%a+9aN9F+Ma3k6H+e-%ix3s-c_asN0amvjw
z?^wd*@$todt^wPgnUo^UeeS5m|0ta}XRkqWZDH3T5NzX1eSXQMqt7H?p(BLL%KNuj
zhOW3y)RG3U0ePck*;PblIhqSL1`BH*b-z=IhNJkz6vl%nNbjhW7IZk-*_DXC>}9!0
z@&hUm*z^zQDYVze4)e&4)KgA@j2QXVMj8f(1Jj4Al5ROg8{e%0u7*Bf@E#!wp_IyY
z;|83kkBYn{b2<loTXO1h*(8)a|9CZ!*b9PkKuYA9u+1%rI^87}YQ;d!Eti?5#>T2y
zX|?(GcK6`yw)4Suo3LX_X5tmncHm%JZrqTxApt;tu$TcY+^WcX|1so-`DNX<-j`UD
zJm5b9XKZ!*{<Zb&bNTDGa)3(8R4P8%>OJ;*jRNhaD|V`0+p~RNB5t;*we$s&ZDB`>
z(~a2^$||N4HXETc_2YZr1;pI*(yZTW6cpv_0rVT6(1+;rU`k>~;^KIMEooxyIRi~x
zt>i`3En!vL+fl4`{>Oj*{GrIH0_>&V(WVZNg|2gr!J&55cS@Rh6GLLR9_W4>&wKiD
zx;_qYZ9Y)xPRl8&2qWVCrQ8fE@BapL(3TSYf;n6bb%+FQOS;L3XNN1A4!pZuftI}3
zJ8~JmXIMM<B-5md2~U9TZHp<2n>6g50=#PPn82AoHPd6D@jK~!&$pxy1G#202p}%3
zdjRgx=)MlMt}d^p{p=Me+<wD0|IjQmx={&<NIqrm83p_(;7u4BfWjoNCIXafw(jmV
zqV;{E6uSD*uwCHLVxwJiDIB2K^x$N)wm7XOF&l1zlb|PGEgTWxSVQzqeVvctR+2(b
zwv(sx$<9J)4U898h&Pn;G)YWftxUh1!p$3(m^em%eTZuj*G@RKHwN@Ys5foZGdVVK
zI_lmBXxnshvO;}MYtP-)0KSGoDuu{?MwH$|Gs6a82cJcx51k*4U@aMc=>Mt^+UMs$
zuJPoq(U+f9Zht38R-Vqi2F-9cV7n2Dc5qXgkD`40y^rS6<MUyJOGQ_AA1xNtr-YYL
zm=-$0DoIwdo=*7pYG~r!RIy0-Uu8T(%Z4|R)i`UVr;Zz>AKqoB&uGb1#*xlu;KUwQ
zwu;<(5GF>fpX!fy=ULYDZVlkQlfG9zK7R5DHy>Dl46gAB^5)BAOh_=N|FI~j2SB>)
zOn4q$C!So|odEhzonEWHG*^iRRh=2lHmyR!dQ&|tQ{eXo_&VT(;KTI~2_#d@JbMLy
z628lQ!?9;JvbX1o7a@`Gx7MJ~$B`HTOlaKR^NL}2Zj^5^7~4ctFbrDU4Hv;Dl#C(H
z>JdHPYiVUlzDSiij2(x~YE=#eM#Q#(;$rO1Re2q+W$O?zl1jUK<Tr6vfK|eNV3u*A
zWrKd|Fm<EM;IvCZuRLazcBdIS;IY5|NOi*mC>YhK%L-Z%OSrIp_9sl<!t1{rAM>sl
zElERblg)<Qe}auB)-yi>oP<s--aN(Ik&kf`0c<qe2a_UPx;S0xi5EGc`?Kj%D!u*U
z8jVIHq~^O9<1DrZx~;<VUG)u!-sth6F~qSAkg%kT(2VL0v(9+7in8>21r&sbo5YvY
zyUcsOA!{KyC0XUKos{*L!On&&Pr*RqR~sWcgOo%}Oo`tpoE?8RcCdL8y}_#mbw3cb
zrsAdg&$<5b7-#^suOf<&M+Yy4hoQD@e^|<gF3#g4-w67UtZmzC#VGqetEDW^w=t1g
z!a;ig$b={kI6q2>!xJy2EhY-Y72b%7CpFj14s)s{k!0Za@sJa=`XkM={~@(WSI50Z
z)-_ae?I=OPNGpBAi!ptLWgks;Bt1TdDyiS+BjhcenDvTlAdH{z2}H!^+)DlUv~KG9
zWGVi}8U?7I!I`d%I{EVXHGv0eLsrqUtQFvd3r&e>2Y=UpiB93sbgw;Tr*N0LGzkkv
zuO3OHsaRRm+l`2G>GR_*0x+F?<C`mpf=LtC*zqDHkd-O|VmFfZgi#%zfcTJUhlA=Z
zI1Ravq$Z3JTv4Y=ijm2KLoUj9J0Up1X;_AOcrqKe@eeNeM3kjke!VVx*r?t=97=%+
zh|eJP%BT~`vt-Ntt@fjD)bSh$1cvW4UTsrHTxq<l7}l84mz}iPU_r6*M0FQ8?=NMY
zV3<|BV~~Rh4r!oo2mm!;hp|GjP>ZI%x3PYmQE=c@cOLkqYJ&#(VqP@ON5#)4GF~IX
zNT`$m_Wr;cKwIO3gBpD}4b?KATR_ViG`%nG4_5{Dcf1wfa~1)5kc<`&G?}35Asycw
z*x>+TdvAJTp41xz1|UL!`DNM7gcS-L2?3S#xNe$lBJs9<YHDN$I(SGi!y<igHuavW
zm1X(+2I1{iVB2kM@e1mu{ekWGppaVsXs}kiXuaViUKkwesXch=0wKkZm0%nU*lu*Q
zD?A_1<VZHJqJep{P|r|yD&o|JOrpK&`?Ex1rT!IPuduKdFa=p3Ikr?@AXi*k0;zSS
zhfzI&=ERw-br^?~>bP|Zv=;rcsCbp>G3T<a-@^jY+_OM__URi8u20jo=ZAj;_Gr}b
z1zWftrLwgjq12E#W48!~GbO~#+?>_Vo7i12=*NeeR?RS1;&H4dK)v3u$F(QMRFM{~
z*MR9grR#*@;3-<iW1SJ;RgtVX8V3pMuTak6Ta$>s5GEP$22_Sz+HfAgx?Nl<-BDmq
zoZ1qNSr!{WsaSGa<5ZU#5#=ZN2wk|k7^O#4`rjePeYygBBp1h*X)<k~(rz}4+FN!~
zQQr*=!}a(*<((aDnQb1VbXt)!Dl|r1^0*o8vD%(+{$S<SWi!p5>o;nQ#|+VqzbTy8
zBN<1+MbL)RY6=TUl|badsm-cZ^2p4Wg){!q58!W-MDFtWeJ3=MGI7CKmY#^aN;6`D
z;iu;u?f3lvJzqzrH{sPnBB$HVnK*=j{3%_f9mjsFZ|IP2fy4JQF1ru90fL>%u}Wn4
zVUBt%Ik~feFXmKCO71w)3dA0MjMagvsW>k6_eX&^FIV7zJIu0xO+by4f$powWaSPL
zd8xQ+(_q9u!Cf~ubs12_T00lCKRw!;VR_tA_PvjQZ&P(1btdw%LolUQ?ee!{U0{`K
zTV}hapx3S!fv<H_vx))KG>|{!(g4pE38W#oio6u~S~Ru#d5pwDJxiLvG<jz7d}otz
zY`54Rxm#JCZ)4;AoC8^8MJy&8f5`4<`F>H{_c{MG5yFN8DNO$Nf|xEDvG8FmG0|W>
zEj1M~E?~jcCC#&Op%xP&Z1%Y@W%Ja8P9k<PyjxSy)tvqpr%oI{13T-l2F5t~(pUZh
zZcdetU<ngX&{{=`OEznov=woqE(VT?Ayso^6@0w^xE*)`<B3C$D=*6q<nI3qQ|A?y
zKWFo-@wNU^#fxWW=yTUM3{VK<8!Tqs+*o>eoYQJejQK*i6^Wx#O(P-(LcJ_BaldHR
z@{F)?70Dy$9ivqp`h@3pVXzm^=)yL7U$WxTiwN{b<2_Y{O=m2hO1?@<a3PmIX|@$D
zrQWyHV0JcT3>odhQPH4yHe{z@-Ah=KA$8PpwS|zdo538cHE#5@T9wBgetp4H$nhT6
zJaB5Ea-;uPwa9EKJ~158DI0?ua!KZqWFFw2kXTm5{41Vt$%pw+Kz1CVo3jmss?q2U
z1(u>&!=FK%R<8t~g=XKT_ByCkwJFvuf!B=}Rr-C#&QKgosTXN5m0F4YWlqRU_&p0%
z-`Z*_Ins+^&txtB234U?oh}Nh9;bbJIz4+ldYlmOY`MW2b}Q9JM|AvdWoy~5Y(!s2
zyk1OVT*&2?Z2aKhmQA|#lQRqrU3HF-u$k<rtNG;x<7D(8;#4RIK^U=BmDl6f_uEp!
zNrt1EZLI>m__>UxDv9hY1FjPjV<B@fFHtpk#i<*O&?&!aWo+F_Po!e3JI_&yuq4y6
ztQCs-rRK9h%T5$K7_CdtJA2V~@V8@$P?5<0g_-c(h?k?CwZg7ErUwUCOq~0TtvVg{
z6UJQ~>%7DCWAdl`yy+V&=tSBP8vodK)%Y!ABV12mHeMh9I3&4LH21BmqiTYaL@Sz2
z?$r}7+_^1%TC~&Akaortx5%)+Gqd*V)1|k#gtE!^XlIWri8dAg8N*ak5gQYbuVvS%
z4zPh%gk*c`U-V}6FNFL2g(<J9%7}%jxNzSuPZe0H-yqKZDZEd!{~k#K6u4WLtbd^W
z?7Wr}ymY;Bct$ZWE%8r17b%qV!KLxGoh;D=t1gTqkRSW!;o<=Wjz0rmeyA=h8;wg%
zmd3bnlm?6Ni>lBQSsw9aIz_>@84uV@#-6iKd-ql0{`(JJZ3BpDmOIw4hovMP-EEBD
z&qTs<7@0kyGKbFs%2hR@Nu^78aS4LZi4flVO8BIz+e;)En@n!4Nfe5G%@A!eIJvkc
zULbso!&r-^E#N<{oax2vGWyFgGI+H6_9Y&qYvwZK$H<j5?_jBEZB_J}CYTeY<ghsf
zMoeUVrINWf4TWVvPwi>w^Fw-)gYMWL<odT6<Y9lzeNzzaR<qbFNZ;aCJQ|&(+~coF
zcxm{Ym9^qWMVdo&LC6cUz{qY|BB@a1J*#%l<aP43wR!?l&z#bG9raxO=hkPk18lc6
z-Ni-iH!S=Sha*9{rkc$9wZ?ZLt1Cy1Odu+QUqMuIn4otvbIqA1LkLCPxz-CW!Jojc
za{P_6ylm4HS1DjQ(QR{X0z*t-Uomf-*X(`AfgR8hk|7_ixe#%2CaNzV)@&NDq)JB)
z^*a{AynBiFfuGTBg((PR<lFb2itj245z8B7Sj~hZZ&M7IZxK*MA$l0nk>n>zVN2Un
ziDEiJB@-}XJ+n^_jY%|vs&pFp2{-P{2Z}F0&vo{rwIIjhd#$NhMQE>$OEMJ_@O0(x
zci%IU*0z+ZXG7msQ$^fnal6wy#e!ALqQogY1`#@{OV^bqn{2Xajh(bKNruAal18Zn
zW1F1+?qOn&lh?Zy#(F8e`WZ|Bl>^TiY*TU?16uOcy(g;FnE33soy)y2LJGnxTswMX
z(AsWqoS_h(r!j6tFEFhEf%aqfVHcU^<qDIB4pjfne1y;{inZ9@KW{UV6TxG6Lcd_*
zryj6bz}s!k&dXB$s$|M3Xv@(Qu}L^Iuqj6`UKT-bir~3pIOh!g47I_7?0amafsyh`
z83}XA&y$II&b0N&n!Vf5RLcww!>)_Ms%`O!bXJO=<~|Fo6D&GiF=juSsu?#Y=d+Zw
zxKXyjV3LE?YRiJ64|6j7qI1{V=FD*|cuU=cvP?Ve_=Gl7KQ4rBFklN+-MMCb)$`*;
z@T*lpN>1GcHv%}k0I_PLZEl-;g>D}_>&{A|Lhl`{8W#<SQ{+lGCAF{1ioIR&2vv3`
zb$2IRc#wPMBx#J+>J%r~X@K8Gbonz7CGS<n>8sAIx(kjbu<aw0J!=#_8P20nY%}>H
zF%6Qf@$Z7Ch&;DdH?n4IPDjb=T+=7+NFM8%bz>@+V6z(EbWCG-*g~euH|c`iRe`CR
zMC8a|hT3x{hEC2<)h|zAFCIil9K0a3v;4KNGq)fhH5#5d+MUpzz2WzO;=vbl0;LFr
zP7d|M<Mc1cN+kh|Uzvueq}07h1mVYE5JRBzaZCwsr4<l?mFhz~X}RjDH2bk-Si#+p
zp0%fctTQc6p5?Mh#h}=eKM`v)X;4g`HWC|CQ69dF>FsJzkZd9N+&b)cI!lm~DsujT
zxKsGO#*kz?rdlh2WEmA`gU}`^sJ8tnKE<}uv##;L40>OSS3p`m=>t#z{WmSBvgXTs
zjOD7!bAB1})#uY|TrYwi(}~~SLiI8Q%<=jUa+z?(4GK}Zs=oa4le~RTL`|zgk1l4z
zOaSAh<hdeA4qGRrsoib4k|Tbq&u{Gw^9m$ke2yzFDIEF&79t93OCAsiAfbIuIBqFB
z<q<V-X>nEgQ$*jQc)*{*ElK0KzLU|=0vk)R4uyYTvjGWp+Y~yMUc6enhagLclcwM<
zIj*MP1GZQnty^>(y<9w+v2J-X6<ev)oI9#AZ^gp<xIY)AP&2U*>Ei{1U}X@71egD?
z+I`|AOyp}y9Y#}R>YVw=i`c8&)BNhge2RU4_aAZXj6Ui0^<HmpC@TWf#^`&-6c<DB
zw^Uk3VTtnIt%Pv`=>8whYTKiGGt9JaMT<}PtK5Rn;kkd#$S+PzZ(%du@5nej#Pm+m
zL^NR11)_!XMI5_E*s5(b`3vEXADCwrQ@CVSul>5vsLjqD?dNvJN)Yd8>?R+_E007|
ztFBDfu`}A^XS>9}YH?Qd$e<cdq|JSX1#v!^P1$hpRSG%Xcy3s5hvntry0pSEXR&*+
zS$V-|GOBX0@expqX6d_#xJQ)Zu_|Rz-~L_tj#Ne%5v*GLPf=yH#8xEapQPL08O&e)
zzn+rrSb@|jkLc703nYA*I=>T00--9%cHqi>ORfmUz5SzzYHO1#sfZv^i-Bk?b>}Fv
zrP?1C?5Y-)IU`N9UA`x7jsf=-!t&U&iLJ;z;KH1w6wiJIRXX6A)Fz`Hy7o%k+<(VN
zb2aI!{!5T^y6romd}IYHxY`7AtlxG%Jo2C5fxLwWf`tdH<(EbOVJvmI3y*8d<3GXm
zcBk8VyaHLo;-XprFZcL^-u|7gTgS!<m=ItOm`(JH9^2VUg8^oKD@Rn$Q*=Kj=d6>x
zdgCv8P`9guCy}bMx^T}H_GOmo@#CHORNTin_$gs{&joi@4;oA95msS4#B-@a_$AHg
z2l&IYSG}Ky$}j5h?@On-smnq3rAtn@Q_@zzgcNOGNW}(wwz`oOwRq<Dk1QmcMKj=o
z{X`YZzS}R=1k2B1z6)&`<h#cf{R_##c2cYAjCpaV0Xl63feD4mwW_pAEB%!B5wv|u
z8uyeYx4kOSB~tDTn?+npfI*NE@;+6w;w0PyX7Drul*+qaQTg6YTv6a*EAl!fp7YW`
zhf*$bqm6YPqb^x}sO!7p$-$GS;Nt12chl0a>hx9N+oBr>mO_}E@#<SWrLX8Wk#E&F
z6jySRvs@KeqE-H&FJ@4J1p3p(>@M#{nzW2G`Ec*jVRg(Px{SG1^`Go;lu%k%UnQE5
z?y$`qHqG9j%6n-`Z;%nx^oa1@P|cnGP5l1s<Ff`AZO%A%-zm0;YMoa`Ei+xE^Oz4f
zuW-k-Zt)*Ezzq2lmm6Cc+MVT|j6KMFi1SJM0S#FetwTN01h%l=WTr5u^PXvc@)4cE
z+X=G%L%<b)xFWQW1Vjq!zZ|3e7gVkOA3)TmX!)u@)~VDKm31!cMvLb#kh4swGIp7&
zl&?)qM$ADMmGgcWnSg^))l)Mb5t@5(tCXr=QN#t<=O6b~5>52635(}b456m?*HCpZ
z{XIit1M@DRs~!wvg!vffbz17Y{bnexAH+n(eM=p8u2`PX++NgOl)G@u{PpO_1H~OF
znKby+gU*Mxjco)^T`t|eh`T04wvRy&Co=tESmpsctJ3~47&mh+%(V$uwFGrCCU;n{
zjC7f@qt6WGx$pEzeD-xYwdx4BQ7x;adq3*z8NM3*PK*1nJ+&fcD>&&o`EP}6jR^Yn
zm!F`+qv(bZ0Gm;GJX-D4(QE&dH;IW-St#Uf5fpN*H5Q5tGNkOfPIT886Wz|3Z=9?`
z!UOfjP>|(j<HE6DRS4CD$qkRWMi2GJj<Yuzlz`{~8SZH)Q_ySByW09AgfIdN_a2Wt
zVP9Wu>83PgPM$wuh6C0?Y1#LOaaqX<4XvWhy@GGP8FH`w!&I0G4c@Ivdoj%0$dskk
zA6O~W#0XnF^lcHH<D`=!(RH1%m2gzV6~iGrk2?F_XoM%qj#k&}_UE@6y5e+Bc$Az+
z;mo)@()WyKxj!IVUQxz&+1pU4NKfa3`WIPYbsMQW44Pp(r-6E~3|w|%QWZT&s71z)
z1L9@+5QO|1M)K9x^<}4fETepkWR6~|QSgOY%Jh<q8Ykci;m|W|wQI4$gR^Sc8m=hd
zn~=l~yauQe#b{`ToC#88Mc<c3_%Nbu4p_Bs+BLAx<XLmVhq#{%c6iezdaZh1784M6
z6!fS{sps~RhVw<V4<EK)Au#W4Je0yYwV@3r&^L~B42pR)k}3lBBT*2w2BlAL1ee)g
zhc+W9GT<+m%sDc>@+g)>xmq*r64p)#YV1Wi=C>Aqcvc~|7<fB^S?E1N&giq1c}Joy
zw~xAZ(ui3?1G1`=w3nP@Q6BG~oa+Zkk_APg3T5=Ev%UZy1%c=B7I`tJ{B2>Hy9WJ5
zc)f6F5weYA9pAr<E6PJ{!zd_7q)?Q5y%Z^x6t}d|Q&+xKT(cJ;%HXN=!WHe*VepGd
zIi~bS5Iw_?bbtlRlr`whe)fi-<)+hnu$GbZB!J1l#`cFGTd1FCRC=AIW6}a|OCfq-
zglEM$-FnB%;>5Dzp**=pZ)pAWt&Z>LX-zvHX-D4fo-6SIq2i=(r2d!N>0!%mckCzV
zsRONhlsn_FD5qPZ<=lS1bCX>E`@pS0Q+Z6@Mwzp>_LSC9#A;#nME@1_WcV5FbdFls
zd8)b8##XDLiOG;#xI=Q;q;p*o^9-3=pO8|~b5eRw!&Z8{(<u&<M^O(Vhq(*GlB}ci
ziwohVx|V_~Dms+a+=b!rF&mD{3?w4yz{%?@zn6dUMNXMl8u`TmSE9l*q7XN>h_%?l
zhXGpL%blMt8dAO%6!_RmNp)du?98uYi|8|taEPH?^MuB7iJ(AIMC%QLav_t0sJrq8
zKeUTd#_g8*2wh!9ewecvIB$yFDn#(6Q0&2aJ!Q*=FF&Yor{E{fn%0G*U;&nabpVdM
zT9N{Empr3Y$FcImzeinM?bl5Rf2U26{=U_FLOd0YKU!<zbeK$4na6qD1>Nx`ILYt!
z4|eVCU0!Xi4HbEwX-oC9+~P^2PI|s)GQoX66rm1l;CFwf^SX{#>vk`E>=pavOEr8b
zWIX8#O!}`=4{pmmk=XA{y7Gb8=|WA$P*lnb*f~MD-S52-$m`a_>Vl{8`S{WEF2K(m
zRhIOE?!b20T%u$)hklEnn>9qA=PO*$`P4+(ABfE1SP33+J-0yBKL};q_VV-uPCtT?
zKU!B_U{Pf~;aK@<-<!kpV_gx~ETZ5F(XXmEVAEuh_~e{cG^&@tfcX>2MG$P#@wK!_
zWL_TPt+xAXCDQ!;_X{*3b6uB1&%zs2S5g9oT*al_jEsl`mFBA54y$Ril>%aL9!+up
z6@Iv8u<v*7H*w_B+UwQjC1@f4RX-@C@VO<;t{RA-hsd~G&RxeC@`?0k5PvKk-8u{a
z>M$CF-k4_MflmtX@U`&vu4k9`eM9X#Uc4iU=SQ~Q_s)UP#v;<Fw=6vsF$JQj480gJ
z(JU05ICQn0&QqNj>(-$uHs_B%^+Pf;GLVvHzJ$^wJk=vr62@%2G*(YwR08yV?gHoU
zo_v*f->D6Ejpqw?IJOZ|I-Gs}F9I<jif^C?h$1`d5R41!ds_A({wKF%(5viz{h6EM
zRh|(+F{!|N$*qk{$wFxf<F*nuP2LfAwk~<nWURk%ta~J6Y=zdTomiJxjvFsB6msg9
z{4M1XIUFlK+!`-w8IXlQfI#g49>_LbF!I6a<4$AVa5t`lBp5%8wk#$Bvih`Q`5R5n
zh;#3Q2lX@2%2ta~iv*7TBpA!QMI}!q<+vyE^yL#UOY5Im)OI7a-!oEvub)U=Fk*KH
z^%(QThP&pC7w9h7a?jXKgXTNy%UnL&XSm#2RMi@W;dbu&qAzY??RySa8<Ynk@R`%^
zKiF`|u0ILkj_w_luo!qha7o_L6P~+$&Am~)FT|5qt|S<P?2?>#V@SppLg3x}_$XwW
z-=?7EZj&X?Rv|VRz4^$us3s<KkE{IH;C}g!<I0+u^Sw0c%Av3Km0fo19&bhH5`Cpo
z-rc@=@e2~(I_iiSFuJPx_UiYX!#q{y%Nq42aF>oB!|V05O9OX@en+fc0U69)e}sBd
zi-6bZz@(`ib=_$=Gx9&Ld-|AKJr?|*gIA%yv8MUA#`w3Yp8J?%3l|<(au>cWsl<t*
z+up~*!bw$AQh0GnjqnO+eZ?W>X?E@TX0!e(w(M;uccC1F`?>w)e4B?%VfXe1kjF72
z;j;VwZ`-Pq3^7ly2OWU2?qKqg0%&>ynH;9ora!f`WkDO0KQ?Uy5hE@a_58jT!UHL%
zyuycz;F?JIl^0~)6|yeHT^LmFak;!<S?=ns6UgrI>?8c(Bhpa~quzMZym2V=W(JPL
zS@``Ga<({$ehohmsJ*(%qI)};j06TM)vG4gjJ3Ow904k0k%-#ZqzI0=UuP{r8<<N}
zlLKmG9!wBghO>Sy>Igsf?B8SiadXK1B-6x$jIy@lXLRT{Q|ePa0z6?C3Tji)`Sm)K
zT@gs|Nf&PA<F8?-%}Dh0*}2!mti$DgQed4eTrQ<XK+Mx7C{KOVm-@`cVJRMRR*=sH
zCgt?x1J&OPLX2ok)pNf(UPRi#&wQ>)@Rr=`jdW+NZ{1tDqWplW=j{eYS?$%tFRb&z
zjU{R-mUy)@g6p6~;=fI$lu1`MkoC-xO1DOKM*gV9`APIK2jl$Nxm8kUb?vT|?YV$`
zIJ|AySy#ZoPZv^<G_PYx-<eh(|N2F+f5rd_7&f$Oa{TM7x)8ZRl4zC9T_~#MY+yu1
zjJSk8`Jk3gYU}#<e%bJGD-w?PQ=$gLOr>1ejxtS^4b;Kji$hYJTxWQ)1SP8LP<qt{
z%E1p_PY{$bNx09ST{>?ZrKu64es`P99p!U@+DnHwDxZJ-@y|7zbq+fxNQpX;^F_R}
z&dH1M|DMl<o0<(20?M<wH}=El-vG>jX#}4Nh4@|6A(;~o^5`7)3Ffub&z!&l{(MZ!
zA@&xi*eu@(HkC=grN7o5i$}tboWhIbsyrNC!4r6|?ueZkys0?`I<g%27`K?`+;yX8
zA{a<XNCEXmC}mgMcU6Jc%yJl=d@UO8^DMdo%jBfMb)TtKzll#%6ExZWugAP*`PJBx
zuF{Zq)Pnfzyf=K44mOO0Pim1~2rqiiU&CLE$@CmQC%!t%&>B}!3(|^JhyjAf?S027
zk=kQ9`){i_Vpxk*v)AfITP+XQdt3rutu-T=p=|zF^8eA0`q7@(w@g%+{4O;4ZI!u~
zAMxwvLL1{pwbWo&{m--id)Y$hS!Te2g(kJ!4#D;}SF-2yr~m7~{%d?>0(_>0JWmJ<
O+|-mcmCB!*h5kPSJgL$E

diff --git a/applications/imgs/rec.png b/applications/imgs/rec.png
deleted file mode 100644
index c9cd069e42198947cf65aeb14923765be296ab5e..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 13765
zcmb8WbzD<#_y%mFC@6w-horQ`C=nRxNF^LC0@B?eAPqxsFltQLMyP;vgGjd{q;tUN
z8X^6j@9+2fyzk%dK4<5goj&)T=YH<{x~@A$2ck+wN>6(2+BGtDHD$eP*RJ~j>#y$K
z0p_M}I75NOEsvM#26ur^;N5r8!0&smYQ`SduHAq9-}m~^Oz=6dQQA|*$W!0ty{C_*
zyUjHxOHWrv7f(lfD^_nCcMp3PXI4=`Q9&^QRy$8mSCEj<|960(i@U852bJ>eYu8w>
zsVgfQ_-1V(-~M>F<~4R@KBJA)oXbn6n-G(tNd3$bt@vg!S~l$~EBDJKu2`$j^DW8d
zFGwFtw7q;8#_h}Y@skdiJ&c6)dZ@}f6<YkKv9^0|k7YW#IlZY8`wtFk>IbF0*3DQR
zzP>664nUgD<O_LB4mTn^J->NxU6}r@7QCu#>V-(#J|EFhB71(|sJme=`86k!{zE|y
zw|0V?O98O%f2UAno9w?M23enXH2$}Dl@8NUBl+Jsa^WMi$NzJ@?PuoRntZr?hTf@(
z#@Zo#Xi&Bg{D0T=g)@6012hVstsQe+L}OeuPIEDd%UK~i-Oo$@_im~=or>FYLTiUv
zM3&_&+vH%yp+trB_y2dk)QDfa?HtFN_U=Vya`1fcLq<l%rH!!?&1Clh4PIU>rZbvZ
z!*~x=aw4y8QuHQe$7}GPN5dOKL(>5Ec3Y_T%ufB}%Gtlo@JWNtsc{syqSqGd)#46K
zVLgmn+P3Hof2Bzg_vGc-Xh^gF`cg!*X+?q;zEfU7jofDsF}K5g>=oqqsGT5?SxQ#)
z=KGQKu;5Ii8p?_;$=1tD)(+~kcWAj>w4EMzU}I?P(y-UDl|}52vTw!H9^cp1Sffmq
zXHy?5HXY8FKjf{Pa_T2|VBsTM-jz_nN+SVIb>hkSv_-{~lNX2y*#4ms?(ydPsNVO{
zp%0aOq&sK4tPMg~%$g^kJ-=sg@bDYp3YAk^?t|{N0w}}pDU(}O?gJ(Q)IHMQ$2ROM
zO;V1~bJ!x?kBDBILWf%D2YHlc^5mAcy^BWS2+p}$({J;8%3su)z{T7}+{+CR^v;?3
zY<m7>|AlAcrg{Ew;qJ(m#Nagf@t0*+E$X<4H{W0L4tmufmpTPWsF}h}GWGa^wmd-@
zgM{TeKfK7k54|9<&$_ct)?1n&nzv24Q#nn@dY;+;AtAj)m$_dErXA=EJOwbhxRe|e
zjef8fr+*^1Xr%KLD&L#;pu*%CV2)?h_Ycp4)Q3DluQk+>5LH`S+q$4L@51J>jkU>2
zhjs6ILPywsYw5)E)IG@~q)oc*%I(8e>tFJ3A>D6#k%uGRAFH@QWp^6_6>L6A3`(>G
zC3<%p*6g6pm$<Kt9;zo=PtScmAZ=>)@hEyJO&H^Be5zj3boxPpa(cJ>ot`eU&rt02
z#vv&O&r)XD{qSdS1DIV_?u)%EtI`4<Zo1x6yLdaOeYBSDiK#Zy67zFEn4az^6xELa
zy?aDPiNfx8)N!CLGSB+tA0dI()K4IK+*qGm{8ul)aQC=&fQMKcO~tW;{*rhRGq(C*
z;{A>2aYInZQLhFAStZIcNLK>A=QATP?BrN!5@9Re5!ar3Bl+K41NMi9>;#QE){NNl
zu9Z+K76SEYo|~Hp>0vq{T$VG5V{k2c?|D_d_Nkq2?!+y3R3-T=-JGItx#QQqSM<fD
z$H3DWcSr7l9L()17jv8sZ&GzYU+O1co?bA2b(_EPt}oj2{ZU>nHdF6Y;yD>DPxxVI
z9`w3(;iuIm&|(w^ym<Jq28uco^Hgn9)h-UDMm$)|#YK=;{mlN%f6v<f>z^clioxqU
zdPvvlT9165$f<ghqLGb5Pft&StXt?5vl7Eeo23s6tB3yBKtA@{sHD3QeDq>OkzB>x
zXQbW(wIj_&0vcyvbkYj37KDV=%@iLUAK!o2Ps1KS&v2A8w?os~QT3=+wL)2?Sd6x`
zxzS9!d}4Fj3-(sI8~zp+Vlck?bvzPyg;)8*VbiaG-MmE{eWM?+!ObO*eEs739bB~5
zQ%ZHc>Z#Y&Q={)hb*rbI1#NhOu&-9`otY4>p|t`W^)#?EscwluK&J!tmMPz2=*<_w
zvQ-6kx}hn{wH`70K?O*Vko>P!)spl9H`Qb`df-zF@_p~ilkGoaS}b(d(7le;Q`pA%
ziXVf{<-P6!0gc&!0T-~SW>0*!hZzQb3BwGkf6zQV$s!sG?n1d<FGA8{hSw^>^v=(G
zWd*|+lmzG{^r;xd`oh?XZo7-7Q{DcIL8H;41ys&_BZA5-a<wEAH>?djyuI_XJXG$T
zUNixQ;<sl(C@$E3vRaN`yU!;aAsyNFlwk{`)Fxl%vQ$q7(>?3%Ru>UGJ3FH%k0N~5
zpD_}7JzvUAKV?(1p(DHTok8iZuH?Wu687nCl59RM1{YD`b)Yq+m9Njl!a@^1KS!p~
zHy7M&9v{Gz99KvMqhNZYT)%0kT<NJN;2@6zyvw)YH7ahviN!wH-r9OXSLT**svJD5
zsDg~8I_VLyxEkl)X!1f@t6s!F<nlb$`p}G9?(wN9cV8NT;hiAkVtl*GYN`9SG{HRe
zp!cC$NJJ^|1d7I~4X+sl1O$AjG~qJjaZ+k_u70cN&HOGvS~NCX`=J_UVFNhqOSdvG
zZ#3J<qIsGTt>rV7JM{N3-EYRMBp+vI@aa!M2kcjLTZ08;c6BVsdqUZO!b_FI_7=qx
zzB~-()PTlMCbOSU)@O<_g6vLs7-UiQ`jd#eydaySh>MuJ@pBip@}*`+?SYjGaDxSf
z-sd|qNQg>tUx3VdyH$=^iH~6Wl7x}j=F-kgy`eqI4}Xcu%357rRf~#>BJ9t(J(#HG
zNu=QXDW}}i#GOYMgj)6#rllfy1_u5B`rU$S0m3}k_#(JDS2a}^AygQa%rP!JH|<;%
ze&@^SudX=ojP4&h@u97$YJS+bT`kZNriQk>pQ8pVtE&w^#!am?iNz-Js+%Y33U@~H
znh)q3U%oFlE^KX9AZ^4`eoIPoHx@*2$$CXSeE5(boWirPJyUOUjr`%m(Jk*mk$le|
z!#GEYVMhsmik1(R-!P^;R9Xyaw3ly<ogS_0h>~HSgp?ZDfsk(=v{?x2d>`c_w-Kg=
zA0)SKc-NZ_J*M^qNmC|dP(+idC}RUVwx%Nj@fYVaLG+28SO`KAu&WJGh@OL9W<QA%
z>UFnp;Af}Et`EKa6*$U~xxLg@YRg?}ZzoGMK?+H|Tv}|p<z3$;bhaQLP#lBrY0@LN
zW9`P9dO$YKk?8p2I{7i{i7G)O61ff-$nLIOVUK%S73bV~yj*7s_^&KsXGsK!oYLg(
z2?hn>k3`V(M;WN&o9^e@FDN@-PTo+4xPM_e9`26&gJtjuFyZf>m`I0cYXheA^7SoX
zkaRX^z!IS@%@<YM)3s<k-kL0`HLNQB8av0DoGLoWd8V#%d$&W^h&5i`V}Sm_9hFB$
zPg64Pb{(G3v4iJ;MsZl=5~Nu^p+!PM;y)NLi{>{hHqiM}A(7_Nbn@P~-<o?xffOC_
zhdlJ9B}SV={eX_*^VzSn+}uuAAJO#Hg98KnUd>nIzY6JEq+!^{s#J@jOzW8hW^}Z_
z_1k{5Y1hp0o-~-p`)I{Df0);)0_vLm;@gV4u{^QCi$XOEy>{pil9J*9x8h*L&DI5-
zE$_Fgs;V|-;T@uRY+d|d?`<I_23<ii0bRMI@{<<_&f#*5<S8D4h2Rj|6Ol!<mRr{P
z>2DD(dLRGPhW9;vPTqcu0ds?cj`zLB19P^ORiQM}r2-&Mb8EVldSuOYJ9thMR#S5O
zpejpv`O`fQQ}FjUck{GX?Xg=QjUOE!A8Te84X?rL@Q;=J)vEag6U#pf2UdH;e3A1O
z1fMqf$TB-fWTwPWKCHQ&!p6lx!^fDy&lkHBr&+(aW6%&?jOP^YO<e8TDtF8Ks^?1g
z#P4_iU_LxnKjB2ZI?B(5{z~n~Sl76d_|=btZ4#y&zojdKF2Dz$s#{YcHMB02BdQhM
z)}IQ-l|wdnb{vFh?LtDD*=g=D&p1~vHlT@%<Mk-$o9@^4!I5iHVO;y|hqdPuj&I(m
z_P-Z<^^|gJYYVB=C{dxMP}Yb$`6t;^=JmOGOhxR{c^+uiJs5GD?yX#X3yoM!Z>EGc
zaf$>NWO-~^_gVkVo3F#bAu!EWmPo;w8Wa*~FUw4`00M!AQuqz|zF^<ylAV629J;)0
z!Z;21)<V~pmN1{Hph>tjPjUJLapuI)50yR2*b3^$`RaNabpgl49e6k6nBkq{K|eS9
z?++yI1ZS5YQTyg+oNtbPH{>exwSAb^`}6wF@!6A#h1KwlIeXcHkAkR<r^S$%yNZ)5
zr)yd3Iw?%437^QyxY4F8hWoCMr@Z#)cM`UI-r7Z@!yY|J?&DTf^v?54dy@xPaFWxB
zLniVf8F&`1LQKyz%I+6>IL67g-d;8^v77h=G7K2!wVtY0T$?=Zm&oF-?{e0CJ5h_D
zq2A5of(RwN5`MFwZQr5sXYr&T#g9J)XLW~T6eS#iRbP@aI%8OWJGrNeylUebz1$>h
zJve6|)GtbbsUV$Zh&Q4Nb`7<YPeHqDp5Izql@9$Iha!PS_Iz8~78r>(08_C_2wlZd
z-lcx@@YGex0O$w>^P8JLK81z1%xCRR>C&=Et7_l{i;xLplVavt;^2{uxfuwAimouR
z6aQF5stpE-!+q>Mu&CcL(;MG8eop(Kr|qGxE0YL)OhK}><!uiJ+7-%ICuZs?uOub*
z1-`&O@iGW*S&4T9w{?Y&tv7Ybw+DljX^DmLkAhfI{x;VN<XYw)NvS@XdY3zFAuaL}
z6Y5IUQ^Cmbktb6i^Oz{Fz``WuN6j%gP9k%&_qy(Gv4rc7ZW{U%9+{?rfa%#fJCRxZ
zzNrwJrYb71XxJzGCyx-ko+lGiSIRZMDCSw57(0p8l-TCQKJK8nF7`|WHORkEUGZEH
zsxQZ!=;iLN_h-I8Q<A^Orp*QWt`3NlP;U_%pqdfQZ}k4^PZm<X@$`jxySwM&>@Yw$
z6dr!FargKAl!I4DpB+yS;4lhC_6|!z8b8$<mY6*|=>xiez2xnwwx7c`L%c@P>x?E^
z&uZkni;Jho#{zfV%ZG=DNr%cUWES1i4Fn=Pyg#O<+i}*^bsUdeNcO-We0g_h7izAu
z3G<)q>t_TTY%$a^cM-UllaANa;&2IOEjD^G)e71K?8X5!RNq{ymsWP~B$_ek8OZ2L
zn%(=IzpJaxb*AF6D<(PV1ZAc}lwp+AH!n^KeRuFL&kBkcX@la6;Voa2kG?Xg^QIiV
zy(F`+TQ!zMYmPB=Du5}V5z9xgPP}NxQuA_R(ggYsHs6M}!!0Z!KAw#s0YL)$-Tj*L
z!Ct|;;?^6Mo_%mts9)=6e*gF6!dF1Df`klWInE#8Ft_E{-$3Te99O0%@}uFt3Qsa1
zFrFZq9xDZGQSRfzju;*O=coiVF236w7|^92d9=VxWOUu)zjSCb8JS<(=gB(~`(etn
zqX{;Q8z4?Q=FM2h>aVo2ca6?sUSW4+*>}gYqgQ7TbzQ$;UAki2fzmX}N@_aRwXlw{
zI~n+H15>T3(jw&=9GhE7w0^ZTp^O^EKP;u|WVQ4{iNTKn&D+Cv_EP>2tlgc*dK2F|
zgr)cK7Q`Oj!fOgO8h|tsJG<XN<Uu6Mp+_G<$GuFMRFO4VeS?qtajwEbs%^_<uLQ%B
zJ`VFU3y?p>32;zCMvQYv!b6{aQFkI2rx!zLb96O)fR<VjOlaikmumyr?Lj7xEWbW0
zjZTB(!0)eTs$YEDC^@n4T(7Iip0UZyhbKiln+k@44Thoz)>s8vMcNlivAP!9>01Zp
zK)~HRC*)M#f}#AHlLIXeQr=w3ISea&0K1=1+Z#{bjmsP^SP@}*drLg^6}=4_Kl<S{
zdrM7hl)n)-S}TcysosTxRrHBex)vXftTUAyHgUTrty5U==2KzTYx_i8NBdEIWBJ4n
z#|L6A%n^ce&p4cLqzFe5ZTbYE1gC>l<!8{KpZxHJWEBQE{()j>!NxaAz2LlsZ%XfQ
zUq46B&XDAn@ca`A`t6VYovcb@uWE2mw};-x{n4qF>waaXTR_+Shpm0ui_t?tNx9?9
zTKw_yM_8=23P!tWBxf>uMEcUv+Ebzw8k~O<NV;u;qt_IxP+K`#9o~PX4c4h_BAxPc
zmcB8r+f6@ls*T2xqC`Y0yQ90QIup<LLuOmcvYgS@tnO8%@$QT#ZD(k*#Tqc{{GeAM
zJuGQTBWkek21n-Vw4eq;KBs3U2&^!4@@$9q^;q0lxjr*_rNF0wDYtxk#?N*?u|x-R
z-f6Iobcz16cv3qx-f?g;MS2p+`7?T2(n4H9LW5sR^sfdhR5Wx#;ffmh(GP#tuLOFj
zpotZg7Lr9mG{?8eAk6n=EcrN)c}RMFNt{~Evw<Xy(R;Hz4oQ;7cqM^w4JIV~Wch48
z%ga+IThm=;weMMV{L2}^#F*ac(f43O02Gm$S=HHiLz^`Bm)qPrGyYv;?5AHZCk5p^
z<x+X)Xj*ial#u0d9b*@0w<Q;6lI2w07);r(P*{Yn@ZqqHSXlJ6&|V``@N}1o3>Esc
zZX~le!Tn75z(7JorDcUvlQ9q?s^a2&t3{LpcYn$#(_S9VOIUy09<6hN{Jm7AIr@sl
z0DK(BX5mPQm`fs&7Z%7Biwr1r@}yQi@&ZpdxfxSz4UM137mJH%ptPPoAt4k^QTyMQ
zx+*6}ztA2U&6!WDi%GgM#emck7dY-+sOytr`Q^UodnTzT4)N3xA&ozrDjl95{#lAF
zbtyG3b<z@Q6SwcnBtYdrt*Yc{3SM1OM@UfXJ@rHxn@_yTEWCYnwDv?g5NQwMFiX}l
zAP8C3?PjB}h$-(amv?=hIyM@xj})aM?f`jZnUVsSw~3&;Iy5YLkbAKGpeT*QD=K03
z>>^nGHLaZ5>k`8#(p8@58!<oZsQPrq7Kox!_c%v6l&=UcCKTIk3$m2$1<(*-w7cfp
zHdXnbre2ZAEs>rQJmrQZy>%@#bvaL-rd3n1=-lANU&88lXZc^}u5ymwJA2kx5_Jul
zK;q_z-!YU!%0J;G3NE(4{Zfr+miek8w!}+nGi(6Z=C=ndV98qC05^XE(*IhDq*%+y
z7r%rt2VEsFNXA+&J*`-kVzeJ_nXot+(U7rYFeE-_cc8c0U^h%lN9mI1HgtH5`fbbM
z<T+fcQ%={t_ti-4<#abp4;*O>Rcc{yJ&|qZhetc6Up#E@S_=*6B;JX;*%nBhk`}gj
z@4jdl@4Z|rc3E>tqCFKaZPuZWxgBOJ@W+Dz1XyVxxR!Z^P&EJM>0e{jO>=QaFC3VO
z4@S<%?x4@~E%jSYlOF|vWYw{@p-AvXjAd~zTJ~Qsza?@n&Y%;;^Hw|+_hR40&=mI(
zoKK&WuFoQ?7Pp!NDS$*;4`f|%2wMbLb6UaI_h*<lh6Clt*47I1hV5-y*=R^>gi3cr
zO8P1WftU_R;K4FI_sHzw{db)Ujs-H;MDjjiolf)Sb`W=KVfkK=5t{)5vFBEEh1p`-
zkbledv!0t2{rBxN%N(kwQs5juOA4f>{w5u4agEq&6S3p26jAXus$Bp*@lKiC7*0E>
zm0BQ#R(^I{nF?#$aErK4%vSa`Yo|1%(&35Dj=$sBXD3w?HyjnOPDo7fV)1+9r5q^2
z&dL%fM=#DNS5i*~U3iI}x1}9zYsEsTb<6$Mf*-ZiL+6y<(O3V}-wO&F^GIdjH&yNu
zS^hOjPT{BWiP0updeam~>BDwv9L{poF4|g~U*~<MFf3dwG~iMT$&>3!1Y)(}VSOYL
z`6Ru;d&|Tgh2~*5+3}k{dptb|8gcaV{4gRD))<&{EaG3LxI`*6=W)Dk+W)>htRTqy
z;UFad>eYE~Wx!yt*e@{_y+<EW^)tmlmik}vK9rnV4rJLtC^*yLrw@1+<#M0x>B3b@
z+2)Sk=Pt7ih489Q`OepJG}!!P4@nPv+b5Fo*hjkj6ev&mYHsA9{fQ65wSd>=)ojIa
zcMCWlG#hb8#2l+0C?443A5HDJkLN{y?GHD-AcU|F89glNl;2CFQZfbQME{`GIGe4R
z{74>y*T8q|*?x<5g4u2=%Li<t@%fHz2FMcK7OR7nGInzglexpplQy1V**H6RU>*f*
zKta%E$@kP>xeLcu#^kICHw~xX`Ii@5qZP)+U|(#Xe_nH**8INSo;xinasQ2>V5k)|
z{>9}Mdk@8lwa!ad3M2*@bHXqjy8V3^Ucs@n(fUVUT1tRjF!wnm?U7rDshw6Ktq9Qw
zSNEtl8ue0R0pAe^ptBQPE1@3a%1Zc@TUpBS9TNo~>nSIiK&>l;X+vTwB?O|bjHl&Y
zAwr$S=W#zfUB@F6Pi*gVB<g!zF~!t6&GP>WZ)F#Hm2I9Q@W(=@rJz8I;t{JWbX+Jg
zajM_0r=wSvaO#_Tl*Sj`F@3BOEH-^yE|Ow%ZOQKV;GhC-^T+~OrXL&hyxDupKc0ND
zjGo}y@Fkhz;h2B90`oDD3&W2lYHH<K`WFm7Uj>lVd9_<xoVZB?ZZGS_I%sAVcms0t
z^1u@*x9U#*I;gz1v5k!Bds#eoY$K(*S~{WU7`*#{e`mEyDc8(+z-i=yoo1@Rr~T76
zgV#?<If*g#`@z&h(M{3+toas#)kX(J4kT5u2{>hO2`HBmS9J?IjFPu>vt~V?Tuuz+
zT5k>4qdDz#m!kG+PS0gqKCDBSs=aM(;=f@*r!1_@FEVCIr~xdsijz)K&`SFJ@EoFI
zz4X+{yhkyglF4DId8%VI-pTzeXl9<B9?IaE7b(yT7fMx;*Eho(>E#8QD=&)BR|xFW
zfo}n=b_v;Nkbb%%sa5KTARtzKwd!Y_xt?s)0Zv4kK|{~Yw>Z-Rhwn1r<aVuX5x7h$
z5scF`szpyk&dB%`u>e5QlST`^1lfZe`xTt4lq|P4O%9NF2kr?GtM$C>^=Cf7qr))7
zlI!L3l!uwoi|VCX^cO}CpMWR9G_sBT-=!8|wIHld5ZLy?Xl{5;PR`!svQ}=6ErXrZ
z`U6|#lt4lH1==Rc7}fCK3LFX$R(GxSf3ImwYXqyR2XGz1XEL2*UL6e%Gi6n#PVZ_x
z7taUJ*7d+%PVX-oP?1yQABn=y*YBg+-C~0Fq(tTB3Ax#6J|j=6`a5P`2!i>b!H#bq
z`RsXvknXS~T;;}5kAS_*^)-o6WaFR#Ibv4SD%KsZupz#6pWBMbcsI?hxpQH6_h&<A
zDE^6!lRRo;t9q{ZV`QCl^8X+=E`^eH#3SoS<~IX1@js@vQe!yzDk8eRluGPoe-gnN
znON}t5Ey*<D~3UyWmpBwG%dRr>^4e*{semGz=o(IF`w`o7s*hiu8G#b^HCyp{v?sU
zxJjK*d$uq)ZkShQ_p|FQ!$Sxx;P*rr1NZMPnnY<bO1~Wm6>*Iy{0`$(8o1Cy4_xdV
z_b`*VWNiAOe~a_#<WvsC2<8_#q(QXe-2D5VA-2B^KUujq95y@1R)koSUh*-9GxWwI
zjpMBx`=$+>K2G5`Mo2j{6Oku`>O>Vn5>aAsVC>Lu=N40j-z#B!iY$DOHbqx(^BjoD
zaUz12oR5|@dnU*gJa%T~07v`!T_r@4Cm@Sv-$TB-^SFs0cAoOhA-YhDqHz1=7r}6?
z{FZWewb64g({Vc7{RCs>c1IWGG78Jli%V8cb+)gi)uBKT6Bv;a&dXj{cRjp8tfY1`
ze!zLmxAdaHS-w;Lp9}>=w(}{dzV3AGuC&?ZKxrZ=fq-^zZxU^6scH{zT|YIKd@nA}
zNU(b+1(tXO`?rW`i0P8+v6f|oea#abisi4g3P1`;BnGkbA_%2}a7|)3d8V4&`irH;
zVMgbu&<uiQfwlO4>9iCi)u$T88h&>|_MLzHnv#NpRr+N=Zwi3g53C>zVhpU?TI~?l
zHT<B!M{-Z_f==9|_@zVxq@CC5(lSKPi>BcA*lf1LY<hX3AEc`D#G2!F2toIkvyfq9
zO%ZNY^uj~I67g-ro3~)I`F4cev{SqTvD^B%{iUPXBO4JC4Js}t!zXEpfeZ;7XbLxD
z`N)qA?`5>W4sOYV_?>%=Y9fu`<)4qiT-<YOtNp6{m=LsEVn;7H#p|C>E?Ax53l>%~
zb&VY=Yg_H|0^fwBTXCxYv}&pmbdf4+qT3?2QvlfX+{u8ChJTCD2!O@bxi4$n=8@Vz
z%5WFrt&kr&`6owC@kd4YuX9OanFU`TBy^3kH>UHmE`MmGpG!K&zxiAFP7j>!C!E-c
z^TEd5HHAO?TApc_bea7z&ZbK3Ow$?hwe_ekJ=%0_i3F#gu9Lt|&55M^mqt9eFmqaH
z!TNBC#@P*NS#_ygPpSfQtq5EDE(a@Dbjt#ymXr^OQ7#0NmiqIn1|OyNe+1<BJ2R=Y
zjcJ3<2;|G@cpF=zu{qE?joXVgKesMfn8?RYrXG!{h`*8L`-qC8)WhM*1lXU0?weUR
zOI3MY#0I^~v>uR7Ri}$dVj5-_=~r3OZgLa4*88=^Hq(gN=P!5yBeK9p!|*iMXiS>~
z-!x(2@g&7?-Zv1tMQ#kf_kHvT%7s`v+;j`14Kf*cHtZ(5*f{Nhzsx9!{{f{A&G1wH
zYEMv`^?&^qZet&}X!w1rE=b0(s!%Y_jiT-VPZA%NhB&D5iV~rc5VE090H@I0_ITM=
z`(lO9p0%Kh(e-{b1?TV+5t|nL-2y4<8v|W3h`#ibx{+{{rJV9xGxC+sGX(=qmJ>s7
zaM8}W$#1(?AJILhNNF#K_{~HnlhUs`)^S*C5Vk6A|7e41%=#DTWUg&P9AhB3Qg2Nz
zeS}&?x+}G2dlWPBb$T6p=6R71Tele8LMtR**XNgD6_;`o&G?eMNyG7c-O6(^z5<#U
znv|$!VV+eL6F}ZD&3i8<_u925)c<J#e$MS|0LYg6(#Nw@JwAJy>fWSpCzdyRBS<d?
zc=*>Hn0DRHXvuc{u9%U`1d&`>oxi?MANhUAeA0>XP9p#H>ri)XD3N8CRWAlZoLb$W
z3cV9MJ9gvo*yo|u^b>Jf11kG7GmYHY=v~g&zh!f2Nk3hAyz((+mMjkNxq@nI-TT9%
zYH4B8e$FM7o`|JnPc#itrYmkHzXwLsk=<i@dPb9E^J8#uPeEy9RUG}e{EA{tmE<KF
zZ2PLIt?MHUJ0M-E(9mJm+I9rqcOl<FQ&j%=<JUnrL6!YRB9@*dN#p(r26>@s&=Tv_
zNebKywNc^zb(%-)VX*JM{}?(Gk0J*|u};b_rD0_@s`K)y!5Io7hqNvCZG>S;6?BPp
z8XcZ&_-wt?NDQUArg~f!F-lKkjOzlRH`Re{s6OdAzZcS67z{$(KQ@vTyTxBZPEO8O
z`&=wJ$MFF!ros39O{IJ_Dr$BqK9-QU$UirFQkcL`voZUqjl_}*cX2TQ7^EMImXCQS
z&8rrXn^f(#8$@x>FA!>o6v@NKhdEuZBua$XzmTmmaNZBaD>j`5cy3e3m2Z{GhU9y^
zqkR>%^+a%mjZD-}y_Gp7t?Stl*ZzoC#%lDmVg53dT=Aw&#sC|bY(<W&0CB%g4wfrj
z96TT!52-JPm^ZuOBQb}`?EvHUrUwX=V;N0$pvORVL9~E$)&y<l6SknU?zJPnvc-Lf
zr1AG?yBi9s+*t=0wAkQgSDFU9E2~{ugMP=f_j<<Frt-Zs<_P9H#4BtTk;G$wWis%J
zPTjmqy4ie)Mq-WyYI(OcS7%5`>`?L6BPCsS717`Exr&x)F0qj>C;7NiQk{rmSpseh
zjRIoy6-JU%Ov4}nA^M+oOY6asd{(s-Tl%~9jA;G1Hrs9vv>58K`T;$K^OB6No>p9%
zLY8X<aUSC;E>Rl%x8GkXhi3VgQw@uXylyu>tyFf_TH01a7@QVQYxkLh)#@!YGW6rH
z&B*N#iEHWo?`@W0S5tpne^^oV1-q4}a5y3~l~yV@puAKzuk_p;M_TTRpQnpzXRD_B
z1PP9bQXf7C{gf(JKJ%InbMJ@_U2b<rC-_tOXI1niYMyqj5StMP&0H7ZXL2t}-akE+
zN!;|}3zN1HOPn>;qhKWD^3EGDJ_H%MjZdj6>CtTKyRNB4k7T{y`#?d@AVYH`w%>kN
z6eQZrId$j7Hwd&t7za*W{SyME8Q}Q{DId6-!EWB?<4ns|A_r3rVPrU#Hj~tzC>e)n
zM;bHRxt<5yx+lQ{o|EuQWuhnPWlBl8BiM600fKa^UL^W}^C@h~9*i*|A5usp{3xky
z%tn>Pn;@h@CDXNfK%S~Ssazn3p1X>aQ&uzVq6uqSJv1ouqQCc9vg+T{bFjZfZ21+M
z_No_!!@1Xs4u_nW$)iAWaz%^^x&-($O@KcISl<jtJ}{*aS2wODpRFAYx6~t{_Cy{}
zU3@mz#UNZS^}yqQ^blztmq#ndSKMpuaa8Y?MfI)i9Kpiz*vHaTq8y8*X}m@HWM(gE
z4W9`2iK^Us!S?LI$IsI{bmn-Xb3os5DM~qDQ_SO++v8{K+*bqrU<0)k#t8^WCGY0?
z=O3QIre(S;SL^P7-2||wlCCO*-+NrK?5G_YrnbR`u{#6%YM+e8BJ1|JGq9ntz*xe`
zZX3%EcJ#9OCqqWIfh{jj+NC>NLZ4Z4?2tiV^3n@^m%}{mZb{!hvVyg7wQms@GQ&lG
z-DGI;Ug~LZJ!>sT%I1wctPZ!=92*QM@R1dOk;CVE=kV;e>0N?h?vyrZ?xx(W50z-1
zZ!5q(f;q4U`J6%W*n^N0U<`A4a3!t58}xFBn{#b_^gBS{@G%(|{ngCQiXwiXW3YLc
zedXb&Q$O=}VZl-a0?pr4%<O-^oc-m?m!9qCVndcW1~X*cO3_mD%$)2NV2b1(3G0&d
z!;{BHt5k3D^3EH^Xjm8-0Zz)icUccWv~#cF?$$IbYmkRuKp^}H8-+^W1vNn-X>=!+
zp~M_@gGhjwA=k%aT>E5HZ}r_<SnY$(0iqPf`<lwk`t;($Z&+1L4LDVqraczsa$Ac$
zI1Sc~T>gB~kdX>r(Vc!g*d<gI+7Z`Doe%Fx4NS}dNCZJ!dVTa3khudoA|@uLLO?Vm
zOly#O!+kWB%uf6R;~Z%B<u%*vxB2#){Tz93(VLmvsX}Y#VI<Xqx*2}mWMKeZxtX!-
zz+T(+`-SLAk5eZg-vGEO6_jZ;L5tlEUd78z^pwc72O+6DsSS~)aKU%e9s@3pmvew!
z!9&NI{$khz_koisUMzxZW^!-SA=&H;#?Z`<YiDQ2@b_}{T=3h@&Q72yj{<C&ng1Hr
zgK2;ML~Ci;y^zK{urzD`@5W(I=$S;F#1i#oczF0Cw)QBMw0p)Mn)4D~tvL}hAE=il
z8QE~Ml4&A{P?1^`6}o?jl<@(Gt={)B0OL!=-RjeNio_<9P6CWINJi%ST_iHh&kEgs
zWRis8E0Z)g%0BB8^Xz(<JL4;t7npWGZx7bf>mRdjq%bvjD)*k^6;9W8fAUbdRa=#r
z%#X?CoQC+%F_G+3^VGG|&paUL7~R=&m0skGv&<WFV(8mDwDGA{RPX#Me&ib<8<r!>
zNiIY{dt#d<ha;%|w<RC(vCkn5$vl!jGGR&kRx<2P$G)GVypy%azHl>_^SC&NFUR2O
zmeU8uEIK5jekGTw`2`vIe|6L5-O~K9oTV*+x{Aewu&*IFFP^Dic@nRh%`q<-B=0*P
z934UXyz(jr=$Khp5LsWV4>kA(Shs2MzlIzrMHuL4M}<=qcQkK5?ZG5f2#bPtH-iQ?
zx&;^l0|J`N1PJRfu4YmFDT>ZFlXl;U;s6tR9&gY57UExWl@b_e<G&ui^h|4otk<Qo
zil4;D5WsH^LS^^<`3<zdI>82$%x)U0-O$F94=2$%{GCyxR7x9m|BNI$VNS&HS{CI&
z@u6F{1f%x1&amqfN_n)m{U5w9PlrGtBhJ-chwGy~<3Ht-&usi;WMqcByJG+%9t%i2
z5N7`BTSrF?$-Wp7T&WH6Sn4zuk{a}M4qYsE%Hb-iP~#G=izX|=^nh9CY<1?RW%V7H
z6`Y%rt9VUk38Fl2Wls^m5<AAqTI6K3LPd65viF50k8xG;`}6Ycey}KP9P2R+SI~QX
zMb)8@KIyAO_z(@(1pz+P$#qeRK{cpFc*Q=d4(|Lj8ab(1nEP{rR(*8l9k@hUk@`nX
z@~48iV^J3^#e~&_8!SoTa>0EG<&6SL-<n=Uh4e+$?@$#)*O<~U>GXArddB0G6b~-q
zec4fqwg^jG9^~B5_I<7TA$qqy-hb!X@Im+srVI%M=^SuyTXZXxKG_dcS5|>!kG$VI
zdEF|WUXj-;gAzu6pRhdNuy(!39JYns>W|{#-@BzAe!nV4dPORvy^Po8srIyg=tn<l
z!aS`huH_^c6r5kqSnA83%4-r_wJoDPD9xVN8-E~O%q#>#Nga9Kz2%HM(M_bob)3zK
zEM<FKa;VQ4dR#cEhn(p0*;tpD*1s7j>XP0#+L}r^&mkHBiUYunVPpYm2|sBkH0^e8
zm^c#Rb#}C+M;;+f{;TM1HEFERIA`HfGe_*tFb`E$rQeWzG>w^W4lOLJ9B>>IH#ndr
zQndht$UEi`_{{e~$^?nCb51DYn}csqUey^CfyT5f*Rz@CcUntiuKoLGq<4Y4;VqS{
z!93QpaF9fh36#HgrUutgf~f4eD9?)5d0Jd1bcG-|zI=<4qIkD!ocF~_+TcxaUV|V)
zGXaby{WIUjR(*DM;yVjFb1_(Lqyomul&w=`Oll)7R{&!T6ri8{@KosD97=H`_qP6q
z`U<%WgpK|HW~(<Zy$W7c!r^(AE5`n~dT+%;>)0)ZJCsh_``v^V*JC3i?jOB>g%a5g
zqt%e-nBn&ooAUCcrl%94+iBdez|To!d;h`3(co`$I}R=`+9I~vT-aR7*MMYj|FZ3$
zHys2c9kHlbs^~dlVPIh?!N;y4t}#zoa5cK;;WF0*=qF!4{SG)>K=tW7U6TXoh!$LZ
z1x<AbCqex@5#MbDpwfiLE46c%g9%c&KXk9$9<YCtm6Ln+8)8|(o}s|a&7IQKIkxq+
zF$HN&mlU$;!=y|8Fa2)Jg+~An<tBQ^CN^q7h+e!3Js|uTDHG%s7zmA*;a;zqmLxzY
z;H@H_Car2Sk!&D3%1+Btqg8;J7ftl=^)-@?9Xk|G&@fAyJD=(JBQztdm*+-&3wI>R
zdXJ?{G1#um5!-c>lg$pNNE{BoB^B?wT0kJy!V(h!<t^XZYwZt@Ienp~#ZJmuKM!N&
z5K6A^33xZx%+<SVA6X)@dmN79*)iX8GfE#Q_CL^m@Hj>D8{@2Ys0aBlzvP|pmRcJF
zYW7G+5sZ61fSgP`)38?MTJf1vRb1Nm$4-iNKl4$bY1Kt+c+WKk7y<(?@PA`>VDg*s
zj-UEHj&;9f!)<4Uv#N7VN*jK6-_1IvtW1BH?O_3%-m>IWS+=fUguREMdTj>Mqz%-c
zVA`jjCt6~E{~fi#w~R94YT9kp3``i70X$sBXB+<7z}bAj81R$)){Gpt{w`OY7=Xc9
z+hU>QezKYU2Y=tX@DhFixacYnU2w&Ex7o=FCqhd)>jiB$LFZwP+7`()Xk5Xy;y2&1
z-wjLn2wv~DMl}PkegPuQ`>cx_>s@cF$L6%PwR__l{oX$rbaWdb0@qZZ96%UX2iBp@
zN$f5&34nlh`ExlQkJlLS+nH^AJY!bx{R+Si1pJj>NYEz$Dm19si*0LIQ7<4;%43em
zj&;ko`Xd|=RFv?!;`t?X$aW!zMqDIb6fH+9Lg?8qRDRGBv|}MG6_sJtD9(BgLECFX
z+8j9W4GIZhXrp3l>$wa?Bw3wbl>>}5^7j%-K|hBhqg-(@QY;be*WL7|tLADmJm$n?
zKXJ^uo<k-3g#P-?DZE>i+hluGnpL#BMs0M`EfhL@{u2*EwoIPL1%9SGFm9+1w@KBd
zxu+waEBLRe&JJc%UL^4?HK!mX^>)64qd?5jmaHSgrlTXL^ga6-pOe1YFD}z<D{4M~
z!svkNjP{J#$p|TzyS*Ru&GT_XhrXtC)y}G37q_~uj&I}C*81t;MmueO#Y|&oYfv9F
z&j3RdBJ5hQFfl#BGY(E!6c68YWV8BYG3aDj{X*v-M{b$B$s;+JM@aZXuGo%0a}$E3
z1;c^>P74IuTsF({{_1p0l($7!2<W7*k3@ej+>g7+>V;Q9emNA$nTd3E4-@<D(~r?+
zXR-v(p8w3P4PLRlU_5KtkM48QiGPDm$~)KNU2{CC2jZ-kvjL;H!;LXbX`s0R0?u|t
znOF{b?~oaZ{6~!}zped&K6eD$-F!+fP{kFz;sYkXMdR6kQ}>Mr2$hkYW6hox(KfYT
zFG6_%6)DHo&JIFOSi;@&6SL$_qKMH~@&J3@7G@5|jc8Qdi-E!RyB*f+)vbO_z?fev
z>dfs0{Z<@SW*uwKd$U5dzcxSm)~vb)FH-Gv68rk!%qt<ax{C^iJ<}ItEPqFKEj7u6
z{wm9#ul*-__xo3tbT+nsG+#={(9>wf?dC3o7cR7>(wqHQb355x<!<?(vVrFZrkek6
zdBIv)!kGgCI_D$yyunDnHR}2Ez;V6SxaV@zEP`GyEC07Zpf~5~v*G_KQg|?gr8|ZX
zsp@N(Yw|W`>(%|?Kbb9p>z$s!hl=1s2)8YNjWr@KR$=_SxvDl>Pvh!CD`HNa_aY*H
zY@b_`7i$PS7*M6q2p<C06dG$UW%sQ<>zx863ob6cdx?@`&D_`nhfMh^D~-bH+S$3l
zkRamOS@!!glsFbX$&39jj`b#9_${|e0$0k^JJr)zlav_Cjt+*#0x?-7*M)<+0^i!r
zWcgIR(_c0t7x2|N^;5f=stsg0u>B&lSzGN?5;4-z`!xd7jPSVph<T`aI+f`b!t5uK
za?mPG*!gmxsLt|=H^^v6;mbj}G(mx54Tec{*}kg5$YgFP)YtA9=b};%UI&Jlov@mo
zUyMA=XPK5bhg_(iDypMekF_;V4}cq2Tyi#Q4<Yt0fDn5&FxYtG`Bn0PN3#4IpSkj*
z{KNDsh5ZKN%ORbitE7XQCN6qwj>R-`Ne9yE3N<Iw6Vr%n?<thd)m*)ZiGsjddGZjk
zE9w=hiES;ybt`7I(*|oZ`@c0AnGS=LTmFPikCx0=QWpfSi;Qf2Ge<oN^P$rs8x-Oo
zDtUV)D=6~L*<0VzlZ$K&17d2?C;k$6d!cUI&(nX1-6ZyB!RvEc8-~dm+G+}d7mn$X
z+gAe<wn=&#B*~ZGG0mAC^5ldjS;>YjX(6s4VXWutAzRLiiTr7dW<Gq#JjIiraGuGF
zO_bd_3Z(hN(TqA(^fd0eAX6`yKMRl?%jWhME1G-$-5`jtDb}9}*E4izy<~A8xN`D1
zsa=ZTB?KA{m3%-3*JS94c3Fko7m{c;PdYf_k~~CyZRS}k!NP&prm2ARpKHp61b#LM
zQmFh(-rRsP@t3`*5x`%nsPD}6MIUos!2j-81Z8*aWTkxD$-3ZD5AhJAytGt5ew5t!
zowtbXv^nKqG>fRvsXfRc<m~*YUQZ*yWKb^fwFAc$W#$ySP#+rg`GDW=i(QbU{=mtw
zG~sw9Yc9W%>0e3~aqf>t2Z{Yf@H%`bN5KJD3T>#bb#g%%&(kn-W1zu*4^c<T9Wv8n
zSDb58$txy=cKKqj?A`mxmoJ8{l0AbwT!#F)a4Z_9rs^#IlUpHSf&bWwJYt(U)-*Gl
zSppz_dj!slzmYV67&}p7tQoL~roZV>NYcn0l*O^Bx#{F00Oidxdp$c)dB_%JDDQZB
zK9$*AGA=03dtv8YAn|_%L7O^G|G(;n=<&1v+p+bmrx}3CqsVN<5Agv&J<`Y@F|{j}
Z>2m3xtLkl?K*`lLbrpzm@k@)a{|Bz*xt;(3

diff --git a/applications/imgs/sys_res.png b/applications/imgs/sys_res.png
deleted file mode 100644
index 6a9305c62d529f87a9ab063f9933b686c8c63228..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 291470
zcmZsC1yEd3lQj@zupoip?(PJFySsaEcZc8-Ah^2^kl+y9A-D_{T!II8xBszQTYuH=
zN7dBuX5PHp*Sh<hJ~vWXQ3@3a9|;Ny3ROm0Tonomz5xmf1`h-ed;%){r3t*jx{AuE
zgMh!UAhQVI8qrBw#}x_+8T0K0J-zDx1boQrCZX-7=4j#OY3yPSWpC`}WasE+XKh02
zVeaB;?dU+t&dkot!ANT5=H~R7h2{U>!0hN^$wGl84hIEA3MC^hqVAP_l;hcKD3Nm~
z#Nj&fxRxjhf|Zx1B2%3utpmX#gQme$Q4vwWtp}~GCd(2M;+*`c>nE+P2lViwh{2ru
zU*08<bG2CCmL$STJB%=$KIXbw_y>$-T;}?^1~I{!WGm(J-}*s(N5-;UOTLz~4*Wg$
z;wO$9-1=W{M`J_ttx*5r3*c_Z=thd}fBY&phq!w7;y*v|$_t$Of0}u_)v|Puq2WK-
z;Kk1;VEAL_e_8h5PhZ?$B$H(#`&}y<(JvZdE*imd6k+=x8}@qfKp#lgx(OAf{etMk
zSp)Bdn{FDda~EYe^uIg(U%$Wn)E9d;`x#c-B6^>5fHSZ>D9N+i{Xg3mKyd^D&#iSI
zOy6?dbN}B_e0$(tfW(=ufi~z|<!{mbd`4e5su)Nl)tnYQ#2%xXXyNs{#~4%MC7^7b
zzYQM7*U{+UB`WQ1d~1=dna1U8hUS7K+H2o$9DUG-KNqKzAL-5$7{n(o4yO*&SA5fj
zR(A#t%g*6rjnBQefy8H!puOrG=)*i5J$Y~b++**ATXVjQ-h5QUU0|MHbiJ>pq>NYx
zj$T|Ij{(DVzv%Bj9$whLV;y+Hy<-bR?<MqDD0tfa<QZ>{ckLr8Cl~he(#m8;3XMuW
z8u#%FOhr+~04Ro)uXlg!GSft~J5ajz)~L_pR;R9`ufSg;YVUp-T=B~#iqTj<ZdP3G
z(O|cW!L%zY$Hs%Vd6pP5A#(4B_cQ^T`Fcy}kXNv#1gR4D9-Bv*KxGK^O7I)K6NFqY
z#h1`Q{jo{u>=fMOk)-Q)9wZt(Zd#w9h~xkK52SGEjVg=zlXIQ=c%|9y-ENwA8+E@Q
ziOsQ^WY^d!F@#|-KI$j?eVk&{)||LiCijTH^o^HcCd%Ult+d2tic0bt-K`P#yeQrc
zwpz7^w*W9buMy}bI{}tmL<O@xxastv=@A+h<M3;bE(q7Rw=-6=L+oPcWnZUCe`uC-
z;-hw>s9KqM5G=iHcQkkiL=$9O;#K3>@Gw_zB9wi&=<-+DTZB$06U65mCalRCo_7I9
zmiUs-<m^t38T#YI?KKF8ppV~L@u3Y}_|j&sBaFd6&3>8h695%5V!Zow@+n0>aaWA3
ztnYeB5n?OepbPHm%Hvc(3=Whm3Q~=FdGQY>wL<bC3mC|eruRyiU>Y?FG4X(=tQJ+^
zucmLcd5sDTike7MTRG)SQh+eaZ-`&8i@2fO{}c<<9N$qnqB<dl$Zf8rt;`aHVQ>Gu
zXro5n{|7b5jJRe#x^;e3*S;ysWGYV|dguH1y<NZ~hv^sxi|5Kht*J~REEr1#E1nWh
z1YNoOkFxC4FR{Q~@zu(wZN=UDzur{n_?eAVBlcbxBdUSWm_bCEf{=A)lNwec_z}&~
zESI46v4FIDd9^#KnJsVYZQ;Y-V@W8stGzmyu`Y`?az&jKA-w*TFKvM}H*pDKPuqDY
z5L>ZIO$rc9G@H1JIPjkg=0Gfwo|iLMn3AG*0X$;lw@frXb4{2~z8d)4`DjsDI%}VU
zo`6x?QDja5z6dgsax&54mS%MGp7y@1d2fCeg`z8+l$UoBuenp4kDguFhWdaDtN&gn
z=3;|q>hR9tF^{m`xbN~;W8=HblM{z<RDy6ecsdb~>33c2rJwwi1`fVovj1dSYKac5
zKzzFWNDc(kQnWdRkUQ%8y)Tc*Bgi^&YyDtR5r@7OTCWK93TCc(%mJPKWyk#cB7&g*
zk*nD@&0><fdUnaY77|K}ht-~rA;&EFK|$%YH(~{Ffz=qQvZ3f8vVSKh7TZG!P<9J7
zWa|a(x9I1v$fuS1W*TD&R3187*Tg^D$i+N#Slo*`N>leVOh9_7drX_Oy)|vC7tgbQ
zw*-hG@vuAUmpj6eh4+HIN!8HRX1rwy0T94-GY`F2v3XX)5XF?h=5}0@x3x`aY~+Tj
zV-%yG)J2PCr-4c!7dH``#*6fpVIhVXQAl1bx9d`qH+4X8E+-C@Q#;WHcK%*K!uC=u
zX_c66wb0gBAjruhB|GX+R<Bk*{46qsnZ_Ju`+m*^@G-dQGgZ-ah?@LuAM3Czrz5`l
z-xY={co-SiJ!N5^*$90)YnKF(4Y3o%6joNoP|Ex}KBkHyr%yqegmh=ifTgiTO$BOY
z;f>7EvM+Rp>ut#%gPJph6kt)V<}SPpDMlV<rgg1qOyp9yX-ii}YT{}=nXiHE?xHV3
zqYSD<NUI^i7@+TJbRD}MVFV@{xn7O7A({<@#l*w}8zgF*b+^1DfsQ#jzNQT|6rS@p
zYv_gA@9ECqy-L&VT<Caz8*N6(UwO6c0Qp7y+NZ3+OC&YS60}A|M&BeXrTjp~!0?(b
zsSGr@zvu0QGG7F0g87?LS;xt`y(L>coy)pi*Hk>`#lOgU*Z&0zTqF20ES&cG+ISm^
zt>8<^MkXhwrhW`XpS9u2L}I#)ZoOyGMy!GF6$$sx3L0-IEP^ZRvT;9e*EQ<~<yTVH
z5*~%^%!#h?jb75L>ZWFiWR3b4_s1dG^dfCBK#ODgJX0K=w^PDI)Yy`z*|_r2RHU5^
z(a$OT*BY+{x=pHOhd32Y6DK|bHFR_%Z~BYt@P|5Qng7A!rMfynj2%Hiu()1-YU0nD
zUQ}^m3S0CtD0@zJrDDbe%FlE+8ojw~3EUCE5$dJ1KZe{iK|;yOR-C%F$?723jNzzb
zOr5We#oXuh=!NK=q2!bleaR(X2XK>>&04T)+vbFW6kOZH6U`7+WkoTdC-bf_!Z6^~
zlA(o>T2>n`T7J!ivQu)=iGr}AqEU<`Bt6!j^4cs&gG~ZVQFKx<;HCmSzMF+gnzCA%
zE+g9wVb8yu;Ac!It1kLoUYMETnode$!js8#1SM*q!<)LjR3uw=g}~`j9csi`A}HUW
z<3+P_2c)viNyvp`To+3kNI$TN(<&$G>QcNn2`qedI*rM{7Zknpb3TnkhoQ0~N&W{*
z1{(yLRQ&RtRxsrFldApWoc6#Kw@+k|*O#1&&fYK5^~u`(#l_>c1Y)Exm^rs_u$0tf
zm>3Nel4YV;QQ@^JUG<E>!`az?&GoavsLV19W0GShWMWZaVakYoiZBy5NfiA@gD$bC
z?<*l*+>%mfn-!S9u=^lDCO&1^#y#OCNqKy(ld7VE{g5aK_0t$TvRLR4fmEFg8rofr
zW#9vyIh%zQCv;DsFL;U;>xT?t7?+TOd@$zEe>et4Wy)7H&%ceI3U4Q_BUKj~oo?Aj
z>k7vrWkBM}{6bYIsH>#$J%P?rq*0Bs?y`(qg-=xitNtf1ZMN|2xHWmUo;k0s=G?eg
zjh#Pg2a$}c50}OTx(ZOfJ1?zL1d&PvN=GJBjtJ8rGNsVRaQ6Hi_)X2iMepT7<hC{#
zDkA^$c*JldkWwK*^<%h7kg8e6_#c~Jew__-p$X)dXc2n>kpv;&WL#5FAFGz62s&)B
zweI%VwC;y+50tV$HtO^#CLLQ3j;1KiggCbF^450DzF)AXX%E3WsfeOC@8p7*NxNBL
zdNeZyOms%L7N9Dy2AtabMwdtMoN?V^)8yv1tL{x6TaL%qQRy}%(#k?^J>9)poP0xW
z9aAg41uT)OWe1yE(}w$s&=yPhqB<_SG=z;gLDto_c|UOUuG)2XE}tq*W^b0^x6d&G
zC)R}!{Y2{#Wmv+If;AEKE)6*9OirxO-2H;-8i${}6=j;T)LNfcTW!w#-;Lxz(pI^M
zc1%YfQ|4_>Ja3o(Uc6sa4B6cXsph(C4>>eq4tX@Ra1t1KL}_^adqZ&0E=KgJIS0PU
zuZ2*oL8Har;h6XB67u%~wX-w?VqiGaIV*5)+xwXle+is#AR2EjjRj2=rYWdz&MBpr
z!qCO;LEcjhq=W6jN0fegbQRf?+)}BvNK8XRb<wO9`Pp)9WmC0BBAX)?2iNe&Ouh+b
zS;)||-HrCk%{!@jJonp4KvEM%KG<LD$cZ0nY$Zl0tHa3L$0}}XVCM;wSXVcUD0AAf
zM_fotE`!<2T0CWi4hyYo8%#=yBKBV3Y7%;|^^y)Oomjjegd%ou{h9uD&~wLakawk#
zQYK2m&DZxaP@?Yp{Asdc6|qbrx`c{~mS-p9XjE8Bo2CxN&Ows7!p1M#kb<h6gPjBA
z#t4r+mL}7psV|3fl}9amxxm`^6=#fgvt){|!UJE#rCSR6k)UZB=v{lAHCP`q#VH^)
zf|W}d$PG`2eZ6K_7HaozQ9f>LC^7jX(=a`>%zDY@v!|?|4yI}|#5(iibIKvdNb1^2
zz3NXdv{m2jHBT_swNDOh)KObjau}IOpmEUQw-vsVwM)RnI-VdRZ4tvn?V@0wMwk}H
z(aGm=_U7AFgV8_ZIa+iY`qk*NdoaB{^G?rE46X%NkjFe1p@W0628|ji74Z&38c=Yd
z!ir^Jv15{k?#_A)+q;Jw+9=imtY`g}<5l+&clT?zV?0<9wwGofS-Rne`ohYQJ(oc_
zI++=_Kvpi$5qd6H<Hy^@7)ys&p4jh)$pM~_Z?=xNdKr>MB3KgdPE=Z$Mj*N9Cfx#z
zaU3UWVa7aviCg!6Bl;Qe>g$AZ52=5k3Xe3sDN#k;ufATk4I|5#vf@yt)RRP+sNY$n
zipNw(pIq&ToSX9qRlb@(s;8%#J*2Cw#PIV&-K71Uj<PY0wvX}yhW(fZJ@=Kj3gT)i
z0%TzdGf#`QH!Tx~BiHt^fF)p6<K;&CUrkLj>+43*de=UFd?Ag{Sogee)b*!9;J7xD
z%x^Y2GOFd6Y;X&ozo0n#RTEsUG@l$Kn|oAh{7?sS4qVUC=en%wW;QVY$F#wG7l9F^
z&bo7x`;{sbha(2H+%#HnW7dae?_k?QNeV5WJ=c<pr6Ica1_aHuOO-JruCTXl1oMH=
zJ64?d{k)_0zlzZ7!YI%L3Ftfq&5$gY)6C()*@@K^-SobtHt2@6JBw`F&?*RJ@8QQr
zQ9P#1po(M?o*Z_HIG0R+qKcX8g;bdA)th@udkZKVLZs<_%pZm&2z#S$ta;MpP3e9Z
zbat*d`~VN}{LUr3R@(eBZLr+2Wu77FJ-+*(PNPN#=^veU+M^q_A35xH5lW^U4!p);
z{qk}=v2$=bay((c9FnBgVw+~JfATOv+82Xx;-748zT>|5*MGS|gZFn8n}u-R@e-sV
z=P+dSkZ4}BLAMfDFUM{oMJU$v;QAp@a4CkU>iy33Rh2MFW2Gs@U0a6*t16v|uo<50
zP@4W3&t6rI_}L+VnX75}RMt}&ez`=W;})S)i-AH^<99INIcj5?!FjA}(#mX{_;{*Z
z-MFF2PX}6En;Dk5(0`X*Ux%XcICzeK&A3QJVsn41%|UPY^=lzaCZ3@~J#Y@um~e7+
zH6}UHxG$Vbj11;-4^moR@5)D&K;SkS-1ZpI6eyEK4ed05XPMs{Q86V<5n0~i#DR3b
zG~d;^8hp~<-mp?w#pw02vL`4dCx<M%`+*x)6e-v1iOr$&-XfOogy%iq5vv}vjB`#H
zGR{XBn@__T?B=qa!XpqctfX|X^u-tlq{ewmGM4{B+AVQ`lr|`bw*OChb-?o>m1A2_
zQkRS0O6aIU?27n5zf)flwNgbY8O-ToRBSV8Xqilwd{-sxXmLB55jSF8_B<V5H9JUb
z&(0Mdl8ggF1r|~yNX!kl*4|dAgDI*EGro9IEI8UeJsnmollnK4v&<CA=UZlw9tFNm
z_(ynCeCeH!Fz|F)$e$<kVRcf;VS>={*qx%jq5SzcQ_7|K)5Q5hKeVo8KTPOks+4Wi
zv#2=^p9&!vJ)6M2D6-khN$><tw!2H*qO?kN>F)wj28KIX%p4Z4Oqd2r8BjTyp9~CA
zu&_zAu&hKtCb_BJe6D0p>di+e{QE*wAZ2VLg~1D-y_BHh8NH7teIKt{2iBx;sGd}-
z{y;0jS|RJ$p@_TmQY@rk{i9KNzE#;P>=^q%w}^{Vyt6DZ@w0=mwf*0U{84)AViZ@2
zSjLew^Ki!U0&<DMioqYpEF8hK`K5+5(PE)cm6e-@prm<~5Eo=l-&M&zpClQ&1QoSr
z>a8dL2KB``t9<gXIchC4M0LBzC&|>EJ<kDTmq60FGN{T63GSuXekL-yFY?UYmB0JR
z^+9b$=pc6E%qZy@;r=N99CY=NLV;g%;!M~N^<%N6DP`K~G%}*Juqo0Qq0oqUF`a@-
z^cu<PS91?4#;!<ejY!1VdlU7x2^Q|^`DKm20c;q+1O`(Ts(j-Nk|Obr5<|;Dlq`s$
zR8FL<-B<`_rIb@M)W<$JZ|6~@ODWlh3er<jLIW`D4|Tv_TwPyd!@COz2tc6_a7Qza
z0`LP2;1JzP$y9Uo@;~IIb7vcU2)rCEgmqZUaPVZ-J~lfgi;jGMvA!7njp??4r&(^@
z%Wd;o=8IjdQq-l9-cNQkM=KUF$F1qAM_peVwRtohIl*Vy`=cS~QG>HMxw9yku<&JP
zyxT*&-U8cnAet=dr`Qq4g4*}vktsVC1-hZD_j^@Be$X;R(phU?!%aH41CKLx(&i)q
zYD13ha6Nv3D@g>GfbX-ivIZGK^irO2_dkIwhMOK@-U(z-_)V~^j7nMBm9C45Rrp1x
z3$c(ngf@hXs{czw1iTk{ia<TQFJ4O>ng+G!doAXm{1RJrGjm}B{q2yRH9awJ{uU5&
zRW2gp2Kw+Yyi@@0h^d_u@9Is#Qt8+irLt(0NW$rN4rc<+c?j!=J;(gB?g&P&{q586
zzN^n%MM0HT!~q@{cu`nlWvH@0kFzV_z$y%Jq;Di!8-1H3%jth6^1zEN{0;hF4R8wi
zX+SWI4#!1clgg_iY4U#V2ZBRuYwMumAmw02g_etv`wHC#Bs^`FB3ZeiMDu`+taYDg
zfEP*xUYKNXBDm!0l8N?`VRhRcYu~BGJ~;Kalhk=%TEu=nptTs!l*la6_gFNSh-h!d
zFK~ZX%pCSpb~u{8Di5z`dMECk)4sXV8JNW{vf0L9L=811H6q|Xjb)krhXpR$Van{4
z75z6GOnULegjD^huG<i~yG@)ft5LEn=9Q*Ze<EYXmgkAXef@JR*WIxm+`gY&ws-fC
z0Ksm!)s|p6y8DZ?E|kE;a48bMWIDAHD5n{w%l*T{t-ug!wy$?*D{=*r`WjD{Ld%Ec
zPG1B@3>|h@<8^JL<ICb#zEc>lX$Wn~89gLkW1-`d4GFYV=yep}RgESS)rQ{i{^IAu
z76sY(J*)Sok3S2P*+j>pxZ8XE>p{?N_b|sGr%+If6@0Lgj3w0ZGJzLS_R1`l2>}QW
zhmPwx^6V81DZ(DX$0M=C5?B@KBCPjHlH*PeGO5#l{s7}Omd+M}67rkWg(dB`)uRi1
zlzjJ3jq<x=$AIr_;ij>?skNqqg)t)i7=FxO3$Tipd?*E8pDv~8@OWJ)b)yQ$AXUDp
zk0))_XXL(*%j*-lz9*4m9paCdH5e>*UxV6CTFXKETNUdD3o)c1p6z%sTp@pcKh#5e
z<G!XPZru(aE}2AD>uE|=Q(;THm0t;br|tD1Z>&MKQ=j>^GR^Ye&hlcGxK!5vqA92B
zj@EwXc&_PWFNZB42M1cul+?byJT42$&>wgYk>UKL;J_!d&(o6o@Hv%ncS?!`tXrvC
z{)gvchywbpRd!`*ss!07qjIuP+IBxJfwoixVlSvlE(9Vsbf5<;vT)vAD@RHpr1>h<
zAEZUJ8$RY*X9lZY&Dk*OH^<ZDeXRqnx}5bO)Sq@flJ`L+l~~b9s}g8Zab;T03yyZ&
zuIYucXY;yJ1Mvv%Xkj{UK{-+~P(5ml2f}8HurOZ(3B{O0)`gAwZS``s_f1*$+%w?T
z6W|*|v8-U9Ws0mU3n=4+p0Sh!@3#YMx}bbdJ3dLZUpv9j{X4GSo3eEoof->&9ErXi
zc-EtgCU10Dmi1i*{el(|c0{W%xFsAU^kXF8=7$jb9!;Rm8}$*abEfDDP$1dAR|TU8
zbm&DE&CNG|%<YoY7s{ZN*)XOG%?BKzPlsZqrWjp+1O`PEy5jNiao};Oudgqj;8(BO
zyyV2SHN#Dwl8%lXn+Bw-cVy#)nqdA~(QRJ5(TzB&UiBaW)+1yA4e@MWvK>W92!Bgz
zm(Zz@_Ut*dFEJc7T!9PniP__5%T{&sCbc5pF{SSEk>lmdA@$mMtY^*b-7ep3JzQ&B
zQ?1=1J9Eo6rbD|kw0$3Lyvs;2sciYg`pN|<Fh!)kM^hsu3aH%nOTTx=vlf3@f5_st
zH#u2pVbrb(98Qv=2!8S!hC-;JB7)SO`kO*|Iph%Od8`Z6%?}ebudr@Ztv1p9U9A_I
zSoLJF`m7rnw9~dTAMiKE@vEsyBd|@N481ns9v&XFn;eMJ3_S2Yt-5_kDm}Sb_w{;-
z6Hmb^gT^ng>xlk!3-Oujn{52~N_5HtaQ<<@1vBS6>l{a}2f<~bxbzwMQ3XT_wz?P5
zWBT&e&^quxud-Z~xCR8zbe9#!kdTi$B|>rk@UIQ=KNZvt(wal6xrVxlvu=-1@P!_F
zx?B%_luir!^S@hXP<ddGWv=f%Ni)yCaHW!|f0D53u%O{NZ4=s_RZ}vDoNryG-vkL5
zaWq;PzK}I?>2<1CugH!58FWXHjQ<umRIb`8qeDE}Xw}r?7h#LziMGP4;IBAy6dIZz
zaeXwhcf*k8dFFURxMOQO)1FG37a;sCEv?*1ujEcy-xH^9U5U$Bg4~9iJPJB1PbWm?
zgE;OkS(NL_H~xo;v%bK=two8&$YfnHQ`jMX`JC@WD_Vze2^f-i*AEtjRw!MqFZ2A*
zS2GgwYn%!h-1hj-&(9TlP47)n0-m-=_D@cB@3!Jd;)fRdwGapfe&|+y92B})FkNwl
z2&k89q%s@mhK^)?cBgj?cw9#K9pTv`HDU}6Y&nOIEQTlx{bvBQ#pSY5$MC2&tVSDJ
z+@WO%7#ShMS;i%es)!%xk5)`9R{j{I5#e{&N3y#o^vdf@I5RcX(~n~`_B--}PKS?s
zf*P1`bb+aNw57qJb)Py}RX&3q3~E`Jo73Ee1BS-wuD|CRFI#Q2zsGX}D~8Q(r8!}$
zNGi$P9R&|J1PfXew~i4_j264(tkwaZspd%Z11>}{208xScpz8*Oc_^-4K}+ew$?Kq
zV$I)7Td$i8FeN*~+8(EZ&Yza-{h3nNhyBtIJF;{MEG7=k8}P&dPv_9*!}AsYOs0!v
z#xl9S1C6OglK*O~-8_8uQ>VukrPBFj#x_5er-U1>-WM48frk87!4KM|)0c<q!+XG0
zk6!GIBE#K6{GTWcUmxDR!F~}Cy@>QuO>a9;Q4lY=Kks@a5Ct)}pNSTuulO8RLb-W(
z0D*J2(C7(5;=yV<fTGBME}pEaUKo*^OYHwp^hz7*%WS5m(?b^e<$m)$bMs#qAif|5
zCI*wDEYujk(|S58m8RnbVgVk91<K{_xas~hSOUclXX$GF^GD}N5d`OpVVY&Hqk@>0
zwV9c|xYy@(`0YEFTeN)FVG|Ro3^r4EsFdy-HC4;d-jlsvbA8wGKW^^s*ru}Ru=@2r
z*iav+<^KZ99SjDC{Zc8Vs`Hj9o?1C8Kh*hUrINs5t20PWt(ZNuXqgwoj$dFOF-(gh
zL*Q;g#rAo$PZ8QWcnYs+#p~Vok4*H*K@GYw*rTcJ-WO(Tk88Id_f%Em!HA#Mo!$`<
z5lI*_RB|z-u`nd-TIRTaOu&i<{}vIZ(4x1v_rd@$HAV!Z#D+^w4+)7^;QFd3iFGEs
zjW*v*i9&spI@Pc~Ee7=AoWLmRyn!CBLvyyNb!MH(qN+Hq(=1s4f=(1=zAoxy%<ou&
zKS-izN5?s$8hd%;=-#LN0nf$OvU++10JbhtB?@9gmvmZbQ#5{Vb*PiCZE4HfiVZ+x
zXv-C~mC>zBwy`CWOf+M0hKsYbie{U&X(0bJ^tPZ2CBF0QWxHD0+M2#R-Dmrrv=F)s
zkpcA$Anw6$U*igX<>~0?=s!2c{Ww6lH5g~v^?cKJV{_6qGZc?pMyaY@#GE7b@bPED
zs4^-HLF&Y$Iussi-s`j_@6Oc+go+1o*w3Ro3FK*X@E%!7Ge6N|h3>T{zMk4m2)%v<
z){DJ-_UByFwdeOY>jAIir$D8Fq3yVlz~0_oeVJlOGIfcO<$J_R+S5b9?ktC*$-~wD
zQRA}1TOz*E_l?8n$|^jLP;mvH)x2x!hVZkHKYG);AA^=bTV(auo%$7zJzzhLr3?DO
zFlbiny*!@oNeMk*#PRzWk^No1y*9NXGOV>%iHgu>HkvK;f@IKexF4tkYjn6wr3eae
z5y{Vj7Ffp=dg=Vne^jiXi*;gL-r)<&=6thdMs$2m-&-7ZvKUo`vJQ@*;0*taTGf0Z
z8(zIS(8`pB3#{lpzs(N{;4t)+mqR1hb-!C+8{8Z%LbY!&Ia;42F?fFzyv9NkxI!o4
ze|;*sGNiNOZ2o$ppR8)P+!#|7CzvT_Ic`JmUsv&C62{jqwbo}Ge;}H~u?bA)`Qm8R
zL-;>hfZb0k?(+tr#zS<a*}Y^^gers=tE0?YN&ycu9=tev2A&4>9ASxx!(R^Ex%qDi
zeHGJ{^|vfv@AIH44BApVZkE2I1#DA<sUCWIu8-e#y*`^%y?0tL{sw)9hGVEr>wVIB
zt$6}C&rtMi_c@*RjlUt^*WEfM(EQKd0W}zz#L2a_i58M)D;3L7TDtGDNaCgPFgAKk
zF5_JDlfbSWNnzjx7T7|-?$~DYnnG5(P=FvnsQoF|;_$t(^X(dm>TB<OhZU))vzKhA
zI(`281OL-mU;C>V2|Z21_U-N3=$lQV4wT^S2#VOhAM)<yFIS9tl<ylDzStQYOvfjF
zGaEBvE}e9d-7Z)VWCa6M=MnyqxNKz)qWv(!bad_-@sh4bQT5>aO{U{MX*GogDvMQD
zg66}UN1aAH9gHmh>W&*xy8g%YSLj4a<KaXqec$5-vy5Gq0Po<~SgNCXh~QWTN9kE&
zglqW#P<}eB&xJYty2o}@JE-l@vh(qzWlwM*h9HgL4z~Sql?E$UR<U3wv}Wh4R%wn0
ze43Fz@0X4W$v9#)Mi+Z|pNK-)89v*1%+U$D>|o?*3`!YId1<I#)&X2GES+*c&tpyn
z+)oZ9FZ+4S+pULHMfLR`=%~uIs=;0ByB`QXIS?k%C;=7kI`h$gM@M4<Q%9?y;0V@e
z*C2}G9~0P~WY#}b=*N;|4+{Y@wa~@30B01<0^^>41RYi-K}vcAitZg>mx!@hpyKc-
zWmLA<<_@t@!FR;K!An{4vB_Q+8u3pss+GeB?mvy?(cy6Enq6Kr+?WgooX8N1$>l2M
zEx8&`FMCV#Hr6mGyqA~PG->gna=j0%DT5hr6%#>;J|R>@-}iObhG_)P5jR=Me|(DA
zN|(h-;>OJNyU`)`yThPQP$+wx3<)qf*rsbCAHako(`&Tb=<<K47iC*Z)XPT@=`OU)
zEdi_UG;hV$dpBLUe53z6;;luFQ7j`%D2Z$8b#YDuwVXEKET5_^7D$Vtb2TQK%52!H
z)p>!PC@Jb|OVAsr$ibd8y|!fWn8!&m@o165P@-GT1gpxmOwznp)%M-j`L*M=AlXF1
z5+_KV)21=5{Etm>26nr^<?IZss&g+Qjq+*J^O=3qN)%=yr!}>p2eFE-ty|{Q-1JiW
z`VwiFaOwE=#FrE=gHb+YGNUpWg6b~g$9<sdEl+%|_ivyN6@*#N(+)KDX5Ml&lCQmS
zzU!jPu8}u30RG5^u_I$7-{qKx*2XyIfj#zBBTazJbP$-<c<u7R;<fw+o5QZJ`|MSn
zHiDMf&hR3JYeuV1MSfMql<Lqj9V#H5G$tJ+WkR(|&C0~8<YqltnQ3WwO0zkg0RZUs
zj#Odi+Y990jc&<jwX*1>ew1Z^D&o3bac^R(u3d+V=T0Q=aZdyf2~SO9CL|_Ohesg^
z6Vy?95BLAXh(`pq2Rf$p34|Q?NluoDCH7DmJu@2^KvXLm3zE$a=M1AUzV6oJoUe{{
zg(KOmt5nZVex{1u`fqFy%@(%N@92L?(D{7q_^!tP*+;Wdw;w2!H$NPd|M+8bb#;YB
ztp6pYRws38dO8?A7|?L^vQkEox3;z60T>GaFO+k^0;kKNu&SiAIt6l}R-fMml8rY)
z>jlP;E@2p>Wz<SCYw3H#Ir5mHt=%-@J)P04Z<y`{RAiu0LuRy3WOnj7HQSD%g`Rgl
z@$45T@$3r(SX}rs5&BMxj6F}R9A9MfUo23hpXn%b$7x$slr~NjF_tA_cfFr;*|bKu
zoNMY#VbW^^D9rIoNVUoR!F!7?cW#6+LIry64?|D|7p{n5J^3+`i2bE6=O`9kp4<U<
zmz+0zR<Ka@kc@sH%s{q>liqTgHw%>oR@_DvvfX-RtT<JcCEaS=hx8fXLx8r%5E0H|
zJ=;Tow%)i(s9fKwq`lV0^Ul7t!RSxXzsu}zxI2(?l|7Eeu#54FIfaAgFsb@_%Zjr5
zOYgP3zcE4HQ@m@GJGOui7&mCxjQAX$$^P~3JBj7V;7erRpdt+->Yo3ZKKH9{uh7>6
z5>{OQ*Ka~+*RP(uw?Go2Yx`d4$qDLh(ajF44a*N8ud<vS8og$o0*ol&@4oqGU~XSJ
z272arksG~Sr14&yQLy1Ee!i!CF&9U6%)PhJs}kOQ55DnBkoLcuA-Q}}G7!cW{;5X%
z^-84kCZuHrfS+bM0~I>;cl|8>J)LSdaIlw-U6Gd%p*-iU0Wz@J)<DeD%WW48Vk6{u
z+eQ5OD(#ZznSvxAW8oNxi4<SYzZ;D~5cjfu-wOzR(s$~C8xp=as|)-3uRf0v1rD5i
z)cJS<RSaz=c?tOLTo%YPb=TwSDNxqxLlep_ukOs0`9PhJ5v?{Mm^8}V7L7MknXrZy
za3$6WtlYcay)LMpzfWNbah#u0r77!$Zj5&)az;bXKP=SQBO{P}Wo=sXVNffP7a`}}
zn<3e%8)Y`_y+gEHYt3kC-(gz$dO0r9ffQ_!%N)#)kT$FwWYTv@(0o2XaKCojIjCcq
zBO$=qf0=TSjKG(>U4)|xm&$>^2p|+0m&zDW-}Li}JE*3!)tV}wl}Jgo?W1bl*|dv5
z02}r80KuBto6hFvk(n&h{PO24;_|;PviJ#T3fJKl&-JS-uGcI81|ib%gB39g3wq1R
zc1BVtWqN^#4|Mg>?q=ChsPiJF+UdAqzRqkIb9y#>>tSiWbKq)!hWB!e<GV!Z2R9z2
zstgXxi>n!>nZ?CnAo-6hjF2qq0mzYI>(U&_B4aU*Yhb5FmB@lcfszi2*hfZfVc=5}
z5Lwiu?nI)LtPUuVMm}A6_Rq0UI^!siKo*sjmMUY>f&U#G(DQe$7-`M%#PU8K)yvEz
ze`o1jK&sT!1?W^lufIWrPTkM|dYGP?QkiZ0fh&hxjo5TIr;CHsELo&NIpSexH&d#_
z!vB8cn)VwxED{UKo&p9ev|JjqIe;7PDJ7Qs!QB7G=(~&SHZa`rwZqoGQw!c0|KR6)
zZz;wkD+-gOgs<Sh6NaU(r<-q_w7cdq%EWkjAa1fAcAi87qp1vdEc*p@x|Y|tF3UN-
z^;i_qT_ay6sD?2g=~$9aRcBW#?cN+fVsDk>GD_5CJdw={`zPJ!Wn#^1E5^;u4P64G
zK{AFFi7lwun4Mu_x@9mLLqA|9hSEwL3oCV46vS<}&<zZ!_Z>@|inZ;|Ssz+cE~k!U
z2}f$>KhdmcJcRb#R>hXxOk2`le&1As>1Ik7f8Q?D2pAX`6zg7&mRD~ANe8uzn&5Lw
zz&23sAVujb3^+PI?#Jzm!e$Vrvp8xZVSkU2Kvef&%Vk?L*+5tX`+-J@JgKI}lng!u
z2}`Y10Sr(sv{4In=E%r6%;GR;0yiL<Y#(;r(lXXBGV?B&oN6V@a9Tih1@Ob`>uB8o
z+co}gs)$ToA`hnwiXs&C2YG>J`FVFAT0(w;^HA}MGARfKpg5qkTU@BDQ=NAjv@V0t
zzIG$=1g*dBLvv=*C6%%Q33;;WTUv?Dpjl&Qj$@mr2T5l${ikG~zRFsq48gXv_DAFU
z{ISkWf`(4hn$!DYyEoDG{$dA;abGu9Gq|-8GMMx|m#ks{V6~_fbEOFf!!RO-wBqDi
zQ|m7lx#Y8XrUXgoN&np3V+@cM1xlupC8P3rkGszU1W;Y?jFJR&1EUB7fQ7aTIH0-h
z4@zvu=~aIjpzK8<-73mwamPGcbVdTm_YY>7O>(@Mj3_A;Y?Qj?JqQ2UGvk^FgjuF&
zO86%LVC{da@HxYFHE-JJ>Pbzmr>RyB2<P~$I`y+uZMLNa#C*k<Yybxb9=8t*<i5V2
zc&AC%4|Fu0$C=!AzG`RV8woWC21t*A-O0jueR%CtTNPotygqm6o~$|s5hx>wfq!!^
zMN)AV0DWr55<YrSqdv~`8GOf_a>(}~zl(KHQ*^4fIONTZWjS_c2rsrgH|_anK>|)L
z703#e&8WqYc!i;mHM_Giq6G$zti>KoQhY|!SoS8`PRyPkt^vX%_WE#sW_fw^jlKP4
zJri{)OZNRzJ1C0|HbVrx2Jj3(+adsf1okLeWo0D*AwDSa_xL{^OOb{^%QDs=QQqnl
z-A~2&hpI+tlvik7?U&|FXt6xd!}Qc2<~5imD&DW2`b*)46W&&BU7(+i_lGbf^^n%u
zCZ#ZF)d3+*G8(sM&G!_wyT-T=NCTyay+faG*IzYMEO-;c^>DtgAhXm@F&h2?g(T||
zL*8$L-AW|3TNx#^yZjGidd0LLmfK<LQ1+}KMS*9|H}6a%bZreATy1oLD)=(og<w1*
zo_)Q>3#I=wE_b)66cCOTh0zwX*?uWFlIadf{D%S-MTsx)-W;a!^62wf^SQ$?Tkw2Y
zK?{hG>vX4^H}Vw77zT;QpE}BABuq8fP;JpDl1fmRch7e^vynV6z<~-tF#Y?MIm4e=
z9zP1boX4q%Iu9Jxd!0T)8pE4$nLCbDJMU6h_BPh+%ZnvHfFdWV>wmfO?)m={3oNes
zP}S}uxZZxG*8)3smdl}N3Gj_yk4*Erq4azjov)@&>6d2dhL0@KMbBg8x`s<%&pAK^
z=+I-lkAC`B*VX5jdC%EQ0?Tivmnh{W$jx(}3f;aVbp8!wU#@u#1Mk2uMtjbB?=VP1
z=?6lkUM_HR7_?aZpJ11kzqu}8d~+Rjbl957d*Z|k7ZP}~oU8Cn?HG(J@>_6vpXW~b
zY1Icg@8zoW`wg@uX!5HGJEtlwJ4%r!NuQLUi=gd%bsq>?G%>zFSsHIr?{KfIXiYF5
zb%hto*3OzU2;@=OAe?C&yi(KC;atGwHJt5A?t#Rbzow|zBZq<m=*cd%#uN)@F~c+p
zKcMJ2GVvXatPt=jk@WJ#3MBa5tVTrV_L*VG{$jJ9W_=@6-r`g$<31o6(X2Y!oR(wR
zNXmPn1jyvETtDgo+^XhHBp{8HK$iw?`kN@nY;mzY3^kT3u)O<#-m)6YJ(l76F-_k^
zrYcY-qA<`h*E{M{m$yYhyEF6Tzyhnv1is(>1sK)=iane(o>E5O7DgIFw04lV?Ghjj
zYkjUqHWz54A_~iRs*y3u8cOP27Xw*Cq58g|Nr4E2*TM@}i>*UQm5YC*>`K$M{3{Hw
z0JsuHgH|juSS6un%6X#zB2iF@LRRXxJGmithO=6G<A|+z$s*0{(R36#Y!$OZXjX@n
zwKvBY8yglTJpZQ;k#T1B<S=w#qIbO|&Nc^Q(Ae28L?agtniN*YhILxI+`pD#aox)U
zZlv&C!Z%yi3<aAFQ3M$IWltl;&yQn(ls)wCL3y<WK$+!J-o9?!5_)eZ@VA`d0Z`KH
zsCl04@#q$tw<#1IAjP;C4cpVh<UZDeo4&E4L~f<q?m{fjH~Z)C+f|x)o9f1P+f}*8
z$}Sgx^qkSC1t!FJWP7-MZ;T^KxzySQTLTEZdsUC8J7-Pfg+jHwU}E13XH{c#Rd$b+
zLsDhN;{(PA@DH_8my!KaROL=1@0}Dzqt0vzxwLp(h42FV-%g4?kRQu*T`Zk<CID5M
z%b%eJ;mKhfEdzUZVM(p{3?5TbYyE9f_r7ggAD>x&_$Y#Y7M{WJc4;s0`I5D?fzud>
z5fQ9^s~%5whA85;PaaQ;s#R!E#3WA~`Ur{quBO7IR2F?3b9bVAsD?wwNvQe)bu(*T
z!uE8Ue>dL#9!}yhojB^X>e0cN$odpFF`=01;jRi%MF}4y2_Veb%f#n9aev!cR+TNc
zYrZ<Q7F^eiJ3uR2Kp<wu@*mZzB9+0UmQNS{>ryHsB4#-xeBO%q=(MC``LTBz4*BDP
zWsj9Z+vWRMfg9~mBvR$s>Cg;E!St;-q0WEbtJ^PRKlAY&yVAj-J>8St;0wi;VJ6}A
znJV$evH=~nRE6{n;+v*>0t2VlI2MkHl6<W`dy_E=3!?jOiSYNP?Ua7RE_qwPGFv6I
zK>p&%DP~0&P(8z7pu;BW8lf>AA75SHVqXVL-A+}ndB|+?P|pP4+-Tw>w!a=J)fzv;
zZ(8vHp~e^@aI@L6H-FhX;7*C!*A~l{<9lKVC2V9AoCx2Lqth~Yc>yR*b@lZ2kB@f(
z-@X3=@NGc;UV@u87)JsDLI@O6e=0}(q%McaZ_qSnLxGM|41IEWkz`C^bv28Y4VCa>
z5^LNS6(lKIY$~cW(P3HtXbpe78A&Cj0RVzzJ>3m{avAP_9e&g6S5PQY%6V*E=H-oo
z)BOKP1`a`@Ypvx&t&lnsRc%r?j_nzBn}Y$Sy5?U-4FB`C>run}A<y>TjA$ahRDqj?
zkT+D){&HO{#ojJ`xj9|k%*k`Ik~8X~{aUfRYa}FIG?m}g)rHGp+kh`c=nShfXF;J-
z>&MA>+f-?u9wVB1wL8ugPT=SjXwxu4uzn{wX{u}j%anrH?oj>7Sf1cB5o*b5$*POB
zfxb?V27?WM8}&|V)(d}k#H8W6a`im#L0M7MMF&1C`|l_Q1w+k{p|BId?fkulVpcg~
z`d|s`dJv2>DpDCN*-RX7>Yi3r(6BG21gVeBk)bsOcW~wE>eI$eTl4h++;7daQ5BAL
zm~PLm7X|v36<SO(D>1HyQw#O>*+j1I0P8$v_*=rxW_--g@+`9LWE7Io#RulsxZE4x
z3Mqw@*c{!C6MVY(wtTvodOwe`6K>|ncc<L>8sj*1u?59GF66g46Tj%Sa5z^XaWl8*
zyM5B)*LYk{fwxUmv3<Z$z3oTSF*w(QOJ3v!nHoG7owpf?!G9+Tm>T_~aoHA0giqw$
zTeyk|ifrNP>5|`R5P-BhT#cy8g#MHAou235-;%8CWoKXa=sy!bA|rI3#{`=7WnRlK
z`7u^{6Qp(oHKAGBbP3%Le#32eJn{CUy2jz=aK6fIjA^jg5Q}5CTVNGm^Y<g*X1}(t
zu0Ph9sOBP_*YtmJX%?=u{-f!EMZne_(L(jXn<(pXX)80w*-Z9+zylkHs<I5mxDfF>
zPbWMD9AY@B=O0W95M-F2aYK>(?%In#S1UQE*pY)&l@+?Tne+^k6_kV=40m;<XT0y%
z_56V(yAy4R67Z<xmN_@h;((+CvqXL|t=j+kFf5Qqfur;F`E@W1#}Kr@PvN@T9><D;
z06P85rv-He)cukGhxZn=CiqVI2jWQ3te9K>*@h{r+NOVY#|P@S49VJxcRCZ+=1jY#
zA5$+#FV4cqtZgbI@nhio!ejE=eC1B|DW%n7=1^@KBsZ{k+B1)OWfRTeHyE)SEM`i^
zSDXh82wlh6q<>tnFEsx8F4P1{1o&6N&g)9Y1WLq#2y*p&CJzUVjBk>&ONJ~8mtksb
z#_^zni&j0bA@0Ea>9{F=j9}<L7T7-%@5biDwvFNI^}Q{tefzO5d2Ux+hcWpUk+H0b
ztczu;NRy@E2DVe@1_TCL45BI)?ORRq7~=M@C2ef6Q-(`?g|?Lfaz~|1I()=we`G~}
zi1ouUY0~^@UrFfFO)!b7aKum2V0?>Fv_CsKr8|SE>Al_a?brl0UYFhY)K6KL90OL8
z4E>1L$5V5c?9|^E(m@C~Y-xM#mi8p5o@ALvlW61XgFmwz3cg^}48%~C^l(3;7pL{p
zgp7uC-{($__>3AqX#p|B;ltY;k@-Z$P?WpTJdzDmN3VC&pd|*Tp|Km^<3AsBzaFvB
z&Y<*hxNlO5k|FnV!c)hre*>yEMWNfBZ6~`xm5Z;W05No2CZ&YV7~Sd5Q~6PChQ-ak
zrEn%jYOyg)5IZ{3GE1YzPxTNa9KU^|sG1=55BZh7?fB{ojw5cG%z8l+oY>eOa3v)A
z#mU9Zg2!3x#+H|y_KrP)^GHA>;RgjJHm<@oqNTzo$G4oJzuYaaV<5HNslq=By$X!>
zp(*CEr&NJQ(Yt5jgnC2}h{Syx;1aAw>G0N;Bxxx#Fi^@Hac+MwIY?Lp#|K`Q&pZEN
zsiKs@6uj#}V<(I27qXx3%ziD1!yIIzkc$14x=RNuiGgdLM?lBBVK7*9*&b^2d?BGA
z7EMX<-UVWahq0>`!YSi@NcK%dg-sMW*XMxF-*v`5P~aA5JG9F3xARyO*Z`K6S}{E$
zdD>?OOh*PS`zZ|V<UYb&oUxK7WE#@|((KX0B~6#XW&6=|FcyqSE1pOm({OTB{>RGJ
zp6a+-1(kx`eiHN7sLnBY#y@@oS)|BM%IhuYhzerVI`~~Z@hQm)nQNVI@=u(u`OEEv
z$fLo3={_d*c@7;rXuqeV@muqF_Xr{V>vxQM*_XFZK0Qc3Q&gVzgA_l1TxK;+D*0uq
zPVzY>k|Bz67>Obi;3HI(psks7B?JR7@x@@s#gJpr;WIQnzUp}+0rKOvHiEze3MpNR
zZ~8sk{CU~u$=}oS8<-2RIq^LNR{_S6tt6rjm@c4jv9yGwEUpQpeM31rz#<h<7Uxx#
zt##TECHDDOK)wVctfhqmoWvuD|J114urZFqsI83aA#nFM1c*S5w3mQPLL`yWdrf+A
z?Apv$Nk`jqaQ5QVA(!BbGIw(71kdZ!?Ip10Qk=Iuwn$#jVcrx*z>~?=?kD=t3-dT2
z{4se?+cXli?p7l&Cta`pP#>9ee|zucXoW5U+-X2xG#-g*o}b5FecrQn?V9toc2{Ro
zlA@2LGcn!j5-m%c_3kDAL7AA9J$ANJFJd$<8w?q&Od{4TxTz`4eDq_zNk0-0%fC;W
zegP<%!ltHqA34N6zuQ$KS~PG4o5PC3(_z;OgQHhpQ=nk<)Nki-^k;6jqMtan6m$}u
zg4T+=7UMn~DN(zj9~5yX1*@KPZGpo;W7A+FpvXD}RDA%HYlvHqb^B9&{gtCI2vrbV
z`7{^012P^Tf@Z#IJ=BwU(Q0)s7`z-IcLA#{${Fh1(u4+T>jbs-kiv)rbGfy>NX^%N
zLOQs)W2kCPbQQb;U}>={Z_1qJ&trSiW4@-x1p(-Vnsc~DyH!>3U}2E$ue7drOAD+U
z_7vflD<v@=S##1<IxQZk1Wk4&b7eS^P1`Ll(8S(PEMwlUixR5<-%>ceBzt&fJf}Q`
z>p{{+b7<t1dosQnO~_lv(=?Yk+Vv_Rc$op=TvyOI;46+%ygV}^HoaLqWoQig(i8Iu
zYkpSO^*Z>_j@RsDKWwA*5XWhd$n&fhnZ@Z}!M6)Qyp{X3U7=l{nUBd~m#kowZjR@B
z%8_fCx7=NFq7aSG73Jvv$R6vn$Jczf`JMN$qQ3cxZ+%zKwafHW&Yx)hTg#189ZlLd
zzuELvdk4#$+P_P-5xPIKT^_7~b1Yx(9E$Adqc}Gn4y&ak$!I$EX4@Sx`@J!i4>Pkr
zWwCK1IGWh~Cib)w$=7yzhH+>+<_1`F@5`+9F`Uk`uuV(4Rl?ZKQ)ZF1ltEqP9$s_g
zQ0RU748Cq1*QM;0;ZL~#2>0HEcPM<3s)t||Su(g^4T~rDH_O$VN^#E!^o>c`La+Cb
zU$M$>6<3=ES>UW;LqmhIq6}RIpC_}U&rBB2N%IyUvI0bs&P&#F-~R%ZvLx#Pv}plQ
zxB_Dd-484AoOQqB-APuHo=IM;w{%@JPfIYz14@GLby884zwpAPJ)Xq2TusYCM<E7X
z4vf@AA69qe?u@44csCa(hZ<{#in*;5#7Ne%x(t5KE?o#g6O^{Jq`}d5f`$Uhx?k4X
zy`|~ydvQ9Ix1V7Md<N-8RvJbN(2QbT{`Kfq4vc?V*AIP(pY{oEKIApeYfD`&{Tg6!
z*F!0z-C%=_W8j8C^7$5%Iwr|(xizLSRaWqRmJ@$Ja7!Rpi=ecbUTvZtT*G^;8w|>^
zkS2=<t$A!SvKNsJmu(kB_Ogl-0HiNB`p)x!@i*d;cWC^=WtmZjb2iv1Sv2KOIdSTN
zimIEW;xu$j&TvJjc`uJjn6S{!`a1&zj>U8HQiOw=*#%rGbW=5$@R1X)42Xr@-ipD=
z!E-jz!<fOWX1{B=3x!py=e>7-(10kVfAD-6IMiwPVxxwH!A)w~i_n}HG>jzKU{SbL
z70+w(!oug6OM^jRQN-c8emP)P+E3i8#mGRR#VU?BL4b(mUo{!3ne~YCU^_AxD{_vT
z-vE8vmbA@g+InQV|2sy@{#`<glil@5O8Sbe!uH4XG{4K~Li^m26eU2r#=e9L+t|t6
zZkSm@ONR*VEyfE{{iUG?xY98fVsHYIi!;TqreB5V@>%TI)}-i({m3{NvJ#m#njY@h
zyU?eCH`?+Rd1=aY@~H)(r0PgHx1|NDM;s2>$4NQDHKpH{9ol00zkkGlhj#QH)l+Vn
zPhtx%&h<sHytx;*t?D{yH5r3VXY>1<M5D;a|89#_eA};Km|d41+sD^e=D=N1@J)O2
zMlS2S|IxVNMGo#=Q<HsM5x>7wIX$J3ODr@@3)tfW#DGRdBn+SVGB~X<Ejvy-eyfB^
zSFlo|h&f3TtD3mE5;Ath^=o!{Ee|jH6ME({I1^4UEk&_~12Q&xXXF%FxOmP<Qd8A0
z)0DO?_Ivt?XaRS9#dKc*UDDo!UVvPx1wh0DRO-9#11>aGTM#SV^E2H5&*@#D{c<D5
zVC^_`a?o*;OHhO|ZqnZGNtV>`<@zGve70@XSrE`fvMQjwn4X`PYS3j8HE)|-)@yQj
zBXWVD7Z4znlar(GwGn1j3BWU8A7X>CVkARkM`ue1{znS{?#)`ayDWdZOG5hGrokr6
z(_3iSigT#0!b@_+9qKIv%~k42Nl769^fsV7xO@XsfN9v5C#xRIs5M*DfS5T8qARo5
zI7sZ<Zl|Zc73H$<XLm)Hl#TA|O)tb{&3n)JZxDh>bOE6KyT9CfQ$__m&j&cq=~z1V
zz~Q_}2b}MB(l~7ADhfMg=>SbFTsMG&-=f7t!@S`rK$8M8bn6-FR3^OvKwahicEZ%t
z^Y+#9bo^Zf{lwL)QqB%@$IN(PS;Z=8L||Z4p`?m<KT{78`yyw?@b4*}0WSO^|Cj-*
zC*DZv52l>-NH**?h0Um$Wo4wno6~Ne6zavqPFFp7<@{dD)c)*oD-NuWea(?Rc~n~O
zl*{?-KJnw{t>4<Mx0R#gr_NF1Z$tw_2^2vu9WN6(eBOZMXn4oPVho_V-;US<yhEt5
zH2t4ymWsia!ouLXI(B+``gA~)w(o{d1`DXOH#WbsSR5_X>bne6wZA+ZW_#~uKvihh
z2D^PG(&#XLD0s#g_bI5Bzzqk&X06o}IX|SRtt~4y_B~kaVNMTjc7I7}it|K6L!)@c
zdSrAI5WHm;u*9l0+X~lMNjzrN(?(R$U6YIA+3AfNA=+W5LlF3%pGW|GB@b}00^!SG
zERni~Muw#p<Uapq9w;He&rV+;JaBXf<pBbzn3$N^`FRfiCodl#pSy8Kf77psr~`o3
zxR2xosvFNT+qjhlT-?y0p+PJW7>(XLCbnodH%6r@yR@`qWo7l{?rg*Ny#(hxKY$tD
zdIrybpmdZ{atd%lU|=|aV%|EXdWBI}E`WzF%8j0p0%w$ja`IlE_Mr5?p7#Ue(ii<L
z_EUcnS8;N-=bHE12=QM{ihbSVTi*aSig|A+qQx<6^Zjm?POS+Xg#R*{mxai3+Vq$Q
zs84R7SzB5CedA!#x32a+t$9%Z{dp59QLFSg6S`&-IJPGMk>YYt=>O4lmT^_D&DW<J
zL?opVkQV8Z6zOh|Qo6fa4oWu&sDv~MNQ1O=cdC@M2uMjkv+w8ge{o)%qrg7w{kvvn
z&3CQqX1wc7%y)TZ#cx#{_napVOM7~rF{@;Df!?e}i1yBXL>NKIAkgkYTGyHUmL87O
z&ZK&sC-rw2^Q;}rYJbtB8qOf)l-bC#8;+kcTs|K<jjB^+c)*nlHLH5`&EEIGQA>GR
zf%1JON)KWrYE)j8FgmmDD3a!*5mCi^-LB+(HiV!+V#BtDFWP|+4N(y{-+QA}dhHqR
z?Ct`X&Tdd`g+K3*<{>gHbLUCd`DS*7hYZ6)A~zmBiIzYXQ-(5d#_`1OqP?7eAZ~1=
zXjc!^(bY{-Eh;H5FKlk6TCBq-Yr_}|SbCT}3R^n`knWLw{P^MXL5?SbF;O0|L?#eg
zMv^HZkDy1W)s3eM_QMTZu_$H&_ZC*7CmpWj3I%Z*pT~VRs!x(<JP?a7EoD<wR4gtj
z$*rlGQfIM~<8D14w5qPDt%d$h)!MoU+h=F^N0aBS&|h_yM~~z%68bGzI91pTOPif-
zsH1gjMi($p496<ISmx?e5v9Evy406mmJdxIUcttFo`rTjnnmCT?Bm_hBXU7iThS7v
za-Qq&Nd8-6wn$IP6|@ZuD3E87a#3o)Rk4mPWww!bPgN)s-ey#KwEgBI0zh4KkrFpy
zRSX->jdP44SBZ<Q2ONVhriwKpc;t6qAv4Q}lZ8IiROHEs$yMDea?~No@^kY0xANN6
zc$Sf7i^Ko30hS#XwCr~(9kZZ5j}1{|is)jQ$Ca_y^YNP)8&jL)D@@PL*-#bWQwqEd
zvX}hnO7%1NHW2zK1di{}qk_p`Jj+jzt|~A#qTk$ExBN5GQUV1QQHd89e)sR+x3RT-
zOj3&k8W>OL=yR^#v0Rz+%eGtpa6&w4k*Vi0tOdeF0pe;<b-b4X>nSE5zwCKyU(9~T
zL&A7g7OUfA)>MrxwX=$2XJ;qmMl$qfxWRQz2sSbh=K$IlTP1xe0~Si5>U+CC(+vb4
z#p5gu&?8B9XDbBHe!r#-viU>ZQdka;pKLVk%Z~Jso_-nq(ggzP1oJo9gY?fg@A<B+
znWY)$TVAm92D;{*80nF`;l}Y|O&84?UTq*PUrTb)nyXV00NXmsyPE$T8=DWV?0+xR
zX0>}OFl!3M-P_)}`$LW^qgAZ`Y36C2DpJtpverBA4%>W9(hp?IO$vV-@qYZEP*hS<
zA&ILDpLZWnr{lm4S0cBaAUU!VH)019JyLo(2%x}TTa=a@)9_-Uz(+17Ew4Pu?{wZe
zcI0<EBi~>ZR4g(iox5KBOzCeMD=RAEwb?8;%<!AB=*ge>@jYrjKnV;CL|_R|8Kh)n
zP+|Q21=7neSp18}t&tNxOq74?Yr21S;-_w4pg@QQCt_gf;X&X7Q`-d$EbNAxrffI@
zHkXA6mj3k!_E}X&^mxK9SqyM9gg~yXr+0L{ADoRiPY&05-2nhPSbr4Un6t12y9i5Q
zQ(=5P>cFJswj3bFz`x%mOJ>}aC`Z!|(0g%d=_&#(q`g_8Rj%~T!KdH}2YVYVtYsqb
zkf(UX;Y-%k(4-`x=)#r_PQ+KQY5d;USG3H8gkEo!hkb-&@gW@LoA9FwzJ6T{D8I)t
zvF#yMsc-7t(u2h8VyL6j9g0cwW=#l>M(nLJ^EYsZ;IM=vFDx$3*V>Z-|G5S)?biT%
z1>fdJWc8L!Lg}~HBe&W49eE^s?$)T}>=;3XIK@w6qSg2AzS>HT-7nur)iYTEr^VyZ
zvY$phXWtJS3qE}sYM$}G6{=hpBAA|@<}j!VcMe#ijI!*k5`wwK2aw~~XAMEErhz9q
zNYM2YAh~~6LwSbrC;Q|Jt<N$t(41Yjl_I=&GA03svVWR(xsI}g5<Xs6l9~8~nqIAv
zNd_|<JqkV_46Eb%`kUwGJo}bHJtvLk!PXgl&1n8+D+q68h~`cCy+|sEDes(vv>uBg
zAP~NnV1IFe4@>rw5uN&y-Zf~SyQ(-qabsv|YWg>z*ZLx<8W{mXJ1lMh!Xg4iFe>!n
z-$W#!N`rytY?QOymGoj@EihO*_Pv6~nlIZZts8;ATBaNK6r`KjU3}_wbk6ceZLg<`
zN2yyq#G>R2Q)V>AJl0dcAeQ=^NAlUmzbKGPl9L{-KfgUXy!DbytYC8GQoQbE7Sfk;
zrwJ)>>UUGfO%^;5*#M<oMlFx*$@_V?AD**^S4X3WQQWPO3Q@oigm5pFJ6=WGjR>WH
z!*ZID)8p%LWW&9&uJ@MLey}5ij|y8VL-weakrAR<Qbsxjvg01~2Px_4l<G>4hWx4#
zEh1Rd;3ywaCCRsSbhMoLWH2=~Ra_cxa|EvnQ8^p83$x&e&vI&t+%5ka=OAE|;7eXK
z{f{q%)GX{{og+fSe-?YcyQJttR7#SbX<%w*=6%#EreuZW_a7HWt*Sk|y?NpH?)7^P
zu_R8j)X>%K0sfuh@e4FkPc(B$@pPv5idveQ(p<B0<kBL>Vf$qN{fs_sdntYD%qaUi
zkz#8ITj$Ty5j_w2jPNepfPBkWZ^9=FekFFtuRQ4ZVsMb!_OYlwJz%baK2@2S<cImw
zSwbuw+0?_oih67lOgInjzo$FsAo6K_UBfB1y=HlN5_?5Qm#D2OP3RLNVS`w-TN>@E
z(imL5#pe4A4Oh>J3LCctdsLicdxQ$*7b3gzjK@bVDCXnoqZGqjNO!kY49f3~$05bV
zK9rGWG+6p3$1;GUv`y<9I@%kHkEHz8gHV-C!I`TJWuf);S}D^m?7KA`|Dz|Few8AS
zyq*w^#G21{J&Z0MO=}ufLqgr{i%W(ymOd!Epif`B*8A9Q3hN}ulr|?!Z@=xfbs*vU
zT<6StQM`m-3#38h%lar1cvxt^z9B7s>3@wZBw$WR=P-C+!XofAbge-jJ3XLSe81mH
z`}Q7}xMoSwR^#xOFUf3#Wuw+(2_1Gpa;tLEBjsxh4$*XYI--G^Cl3v0ce3u6INLp4
zh*}uMMM1~<kKVwPi4tq}@k`^SDJ&MutMT4+7sR2r>CTv}(4+Fs?2qGMUEgbDA_^n9
zsm6tVw3fzCP1eS*Ti;!@Y?g^%77x^eZ-Ce5zzk+8m7K;eUlIxXd+~Sm8+WMVvc&I&
z{<mg`CD9c((<pf=W2i|i9xZ#x&VQahmP1HkthIPIvDJ|9hY>+qB@26P9|nCRRlwSA
zu$u2YUDMaUQ5f@Mqw~Lc9U5?An3N6z(IZd-&Z^jj_pf-d@}6cW3*^`Esn~Wt8wk;6
zlhV@@EGSr)&rK>J4lFI$`XVFyP0Ql&!a-&KGjb~+UVovULtBLC(@jfI?L_*O_iy<n
z3rf)Wh0>In@r`p;d-UrZC;+`}%!u}wkm1T~%Snhe+Zf7V{<knajUY9U$i;jr$5{g|
z_v(>Ta|%$yOSGe)<dm8X4h+N;P6b=#;<*~XU@EaIKkakhjla2rOZvS(d>e;~61YBo
zets*l@~?kH5Q+=N9o_p$xTG9r9VoTTJ=bXsFi1Hfi9s*-9Z(9`7X$a`dwYG>czbh+
zcqS3D1JrM8aB3hJji(IerL`V}N6`9lU~!vxjLQOH^p^1}6ouE|u6PXZ1{*s&0AID|
zj_+<1moUqB|90I|s|+}IBM|>*c5iGt^x*GbH+THiP0Y89qSprG5wI8ck8j-`)hmCF
z-QMhbR1EE7Y3TstOxO12T6;!G#L{3`sVw_~ByW;Vz~2d$)qN#}H+0H#zY;$g-qH_#
z@QKO(9{4)<8sD-H`GMS6lRp`~CKJ9P>7w@^B)SY$4dLRb1o__YUg}RhOKU_@=*^FB
zTzf~YPO*E-aP%0#FY#hE|5ZwG#9tK?pF{MxY<KA?gWM|;rE6+oKjOtY=Wn<HREtt5
zqK4i{YzkUW8ZZd&e$D#zg(<obOmx@`EvBLEcc%Q-<_Lm=uK88PY@Kc&=e0={6Okag
zzcjH!A@83jT4>VmlT+*h(Dm;g$*{<rikbxrIQ_I3F`s%nUgKGld_-JZE?N1}^I^fd
zC0=RC#(SMItVHDukyOHd1M6&+QpVx+bkmWNqHL>?JxVf@K0&T#LWxrOdQPLq51yt!
z!g>{UHX^d=fB+9#PfsecT#+Iz7v-0~AC0d49elLOZhBpsw-MiSZEt$(ck6uY(_M}&
z&7w=pif!?RiNoDt^7U5cAKm~{uRnJD+OfGQ`dE{%Te_bjc@a1AJ5*m+*#(cE7P_(+
zjz{U%gq}7c)gt}%l~IauWH-6o4?6qWeD=q1;?GS)P$g7l3y&_|UodP8XYvC<K-%jH
za~U_ClE{0m;_$JNoydnD`ru7&fWZnOYTSFMY5^ZDFeLp4b5c?Jm+6oUujaM&eZ+8*
z!}EA;LG$9RIe8U;x!<}g%H9tQ3@DT6iDwJ$S@KZ8(5HO58AmNT7kqmS6J~>J!dh<|
zj4s<-TR9~qk<ju!lNI5(fBY&a=U_6kob)ho$wGAf$QOt=Nq2W%05+iWN|a}0;pF6}
z(4BW6ZQmmQ{j(+O^wg`)Vg5zw9aZ6V;^`tu9oqb4g@8wrI3(ffhQ>`e1u*IU9#+RU
z{BPej8HH*V-Bd^4uv|u9H}QlR2_41s;@4l{pxetOEmPBYOM{^&wp_vzat!Y!4Ws;N
z+SA36kz0?yg#YI<_3Wo=!#|#<bO)VCw%B2Ge9Lu?Mn1U%5guAxop-}gDmi)1pHD3&
zoNvAxmfPLw|K<M4-Y&oOO9xu=?{NF7>%XD#k7P`YZX!fJq{Q48#aQ%o)lzV4wYFht
z=5EI&LCR-O&u1bgd_b3|dKU*P0S;awv8pYN`~sBcMen@W=19|AD)WTJ*xuz-7g4Uv
zy-|1heOR3;??j%!rIkwBbTOPxf9H{3_{N46xG6j)6vFS;2URf+>&%*b92wU|c4Se0
zx;1Py2mVCg(U;rNkBXN~`3mPq&-)Bi(()72)Ysxg)L1{xinvdB9$3EFU++Js$qh+M
z!{__t3?oF(OdMjImS-)ls3?NN+fyZ0?B?sk3_<4)hP?^lC|7iOhb~iB?~k4&YkW9*
zGQh^(a5cd017Z!lz1%t-kD9ND<Y)qkdGZ2k94RB|5|s~I%E~R~t1Lm4x=$@{ZWN$@
z`<POoWXCkH<tZ^Amb6YR2}bCuOFSDE=d8x;h-crk&0WTO7Jr4cLVc!dXWx3NSImo$
zpKpDPN8n%%j3@Z`RQ#40zG6I~?IY(6UMY;`xH=RH=CZnXzhBtbBS+V>d=Ey>pH#SN
zf0#Z9y#?tunu3_xkM#80BhTX#3_^@>dMrwd<L`^GVM&|J>StBCL#3#wogk6DXVDSn
zT)j1(oy^7ltuCjmIyNeb=dW+Uu(pQAGxG4n9(F9&3*PYp$703k(|LpE6}qpK3}qCu
zFv^Z;o?Gu)R~w-E*kWMLdiGmYqfcvP_rOSroa3>oE06_RTWV92NWfB2S=nHSZRQVT
zv+Tgr$D7MdPEzP0g-Ex29tyhovv05&%L02cJ?;Pf@orPf*s3_4SWb0ZYzArLlSxY^
zG-Cb#o;A6t=sm)p3W<8S<LT{PZ9mHuuw)IBNDR%B%)&zEH91!N7YOS#Jl-1%_7(hl
zUF<|Psi~=lpw}dT=V@|grUz1Q2BB+H1hW(ZY*0~Af%zq?wdazo;wBJWDGflDLLA>~
zVO{te@U;;u(prV~lR7h2$}dy#p^?+j=s?(a5Z$qehzKmIj*w+l>io}~5J>oaN=nch
z1RT4zXL4^paE2MmE-nts=!6h|`fb=I$HvC@x!`jLgT(%q2Xh^-+Rpj!T>pA_ce-L3
z+H(;3Rdc<D^ds;o1~R?%c)=40=hmM8J}UR++*)NG_Sl5Rjr!`G*VAz)^F_PzdfDLB
zawatA=2H<<u{Gs48%cCnK8NRI|N7N}UNq6D<&i&eBYq$k^Q&dr9$Z`^nu<S7k7#-W
zSM=9$pG{gid*9jrq~AZ*ArsS+m_juBr{Zbg&e?5`vG<RHCd!JSDhHkG^DNAUlNlqO
z)qFhPoAXU!@HO_YyW`P{zuWxa9|t-Nv6237ryt#YsL$T!mQxxHF7Mn&{19HNA*AJN
z{(aXW4Y>!e5>z-TDd4pL@szyzp=-A7g8%s<>)Tz&#hx#}!-DLdy-8LR&&a<1Z5=}^
zK@*anl~YtknMXeNz0n=P6P0}a++F85(N8?ziic9b|4A?XA32$g^tdS#*~R$xM;E=U
z$^G2WYMc0NGXX{x!Gz}c)r+^GsptE>T<xah!-0yH&dy>a_80%TN;1`BrOb%2nd<J}
zSrwhuU(rRBki0!lL=kaV>=}KH7DQ}8g0sxZ%4!Zn0a_!QOEt3*FN_nqx>m(89ytyG
z^&s4}J&i5!IZ4n=8eGSzLgrUV9y!Pobf!5uISGIO(5dO^#~xeiK?kpB59VIRe=+}t
z!IOap#ii>ku+Hg+qO<cOgu4=?xt|~HltgP>sml<nqGs6R2*rH^hb^Axe;X;Y21QMu
zeWodjRhB0Tl)$-b1r9i?BlAT8nC9UL)`ZEH{xy{}Bzl4Gswa1nE*axS{PXtVL`Ygn
z(jD=C?J>^o?$yYbeEct(e?PZ0^Jmu5jUNV`7J;R9P3xJeKss2mtwoN2qNdq9XU{*~
z?V95^-a|8~hV<cITHN0FNxXUU+1;09Y(PopvE{>*WI`Y0Xy8ym?ODDQFR@5MIPJb!
zQ<buaNJpj*Q{Wf2AAPwyhh;%{KcRPY{R<=GJ?y_39B7Ibg2xvfj$wpKVOUHVA56lE
zSD7XC?AR6xg{$K?VT6>9AQH@TEB*YL2~P$Uhl!djEa3z{c1~K|NU!L)oW~$6Amjl@
zSb(-Aad`gsaBL!jG{+13MS&;ppN!jMvZnq~8N9G^peOfj$5Pa?{n)>5NrE){V-dGK
zEjBB`Tz+rv3nR7cIh#6hJPM0KA)>pjXuT-KF^ouRbXWUzK=GPECh{1Sar|78?{$fj
zk)~x7hLohGfpAvdW0kqM@qvk>GA3f#XN5(2+vc*f3KyYne+r%oI+SOp2ua&UJ^8!%
zT3&?9)6)~-@k7Eg=XD&=rV09Ijv(1z?^mq<dl}D*EF!0GX2cfFYq#h*{hVDolG0az
zsh_xLRFq`t3}2p+BWu>=`W|Aot+Tv`p?Jo2JY6elI}(W|)t<zpgalm+ow;c9O_78<
ztO*GH#GPVVCp_QXxX1QjBj;`=?+$`G!z!42L9Cb*XaDDpoPHl&7irj-)-Ju2{DiA3
zXV%zJN1cl!<*ciGLi-n{$eLFqu^-E)CZ@9Q4^7;@%G+jHFikc<y|Kd$9YYholH8+y
zuBTCSdg3=%7ya_vA65P1gSM4*ZO<2MI?o6%eCACuvjiU(tb*_6k8e^xksC$y#01-g
zp9oSy=f-~p6_Zuwlpmknd6?C&rC|`ef3m`yit&Q*!pfgM4=ePI<~S!YelCL>>w>kV
zE9w9Ajg-_>4O3I+)Bcv07Dgg$%@X3T-us0$H7WEm5x^QKl1BhCI#nsWg~pCAB6a)W
z6#Q+`e?(`0=W3Ot<W5Yt{#QQYMg<iJ;UKY=Oa}kV?AXRaon!GKX!k?HT_*<lDa@`a
zs8_-L(iEg(Pe=!fG9Ld?msX@3(DQ{VE(|jV{_qMcNc#|BxaG!Qeb!U;-a=Cd83<-l
zi-$o{5|aO*syqgpE5;p~q59Sn&Y+9QY=(mTs;VjpWahkMW6-G8o}6e)jq#Xp7F1X&
zC}1P(dGJLAO--45i}lrzm4f(M@HfD*_T!_PE3Gu9c)i2Cw2BHon78t`-9ev2_-91D
z|L{$vzy^!I|23)stB0wlX54NQh>Ou><xNo_Hoz%?f#@1RCtHCRY_-}Ii5@g3sVHL{
zDnQ7>0Xn?E<U+~a2=sAN_eG^Ay`^e00Ryi8`cDK8msPwaO&BWYT3$X>yZv+a&TXP!
zzjO5~+3me}@nde@!qN6R2b#bwb0n_ok?yEXV_Hc?Dzy2Zt>nYGet$+JE~!*Swi#Rh
zo>WZMRj2AWS1doQ|7q1)zq3a|<+^45B~FQIQDaXja<}1Xt^Vc%W^kay<eiKN!m#dm
zj@7E*rl++lm}UkJHVzE7qW|XKDToo~@=eR^fA&lN7%x;6?0h#ICp1BE_e@Uk#}HU*
zC_)v&%1^z5>VLVaif>ya)x9E7B>2Gn_~QNY`ljh}bM=UBEINH>RRetQjSkbL(U@(V
zDL=PEBCg9KB^QU+kA_5|f?GCC@;7A|kj>?_u0^b9kN(3V^pNcQ%C0*xHf9zQFjzO-
zsGn=Gp5;s*Hl`iha&kfW?x^4#3ylHG;LDXt{^m=$*1obwx1yT?wyky#B5FRFoD(3n
zDP#)1bUH7RDjiRZB<O6*8jVt`WwF(q!cx|Ff*d~EJ;L&}-9EbdfsU)Zn5M3CVl_M$
z-6@NFothZoq+Z-r9wM)#G&Y(bEkS4{b9bH%CYn0Eg-w(x%@LC@3&zLyu+T5&q>JhD
zEbn%q#AYU;T?j|%fgupe0lt|hc!b}68W!4g5-WT3UNw2x%GS>A1t2$k{QR!a$hg5j
zL=p4|Z4_iO*!Un^<6(5mg;`FWp2j0j&Ypkd%(G<=CsfWBryf1RTfT>e43Kdia%oLk
zY~kr-lfZ>LJ)E`nlYzn4C$qw!_Ttt!C-vv|v!MTjrc{|phCCu7N<&W|z;ILP>;VDK
zlanbeq-TT7fAa6-uk>vm6{6sCrzD;Sw6@cy^w~5kVp))D3>Vvf9Ib|d<<qcs*cE1q
zwPuLgn>ReU_xde}kG-)2vYDc3EVYg>I9L^gxn3)OfPFAkWEh(^(!=$Jfpes!^62d0
zM7JqkhA=`s{bfx3T*kh2g#`kFPIW%CabD_;Cr=4i^#ycUT>q1Od#3&D0h>`3V@8*h
z%j)knR!&Y?#eM`h#*k<>o+qoHJK#qt>XtGgAx(%uuSJj<PNE2(F3t`qx(9N-I7|~q
z3i^pZNl>=WdnhKd7d6IyFMaUQ6b^579+90>)$ad@60UT6AU7$Ho6nb_`H~K;?nk!d
z!FS%^WeB5@L8CYS`EtayT6lHrTjkw<Qj&_V41&yRvhznSVv3)_2#-e3gEEu9S#}gx
zrBXn<JcmQhrocG8DJ|SUaNWVNCwc7QkyNQ|soDBSMrqQIA0v0PreEuVJvL9c!E4G!
zfZv(!IN;{`KQiI97+zu~&cuHK3{qT85@=z@i{4P35j&NmvdmG=h&$BJ;2Sv5_nld~
z{I`>}-eAF#!IH(JC?-kxtX-?K>hLd(t)Nt=W+o4cRA<$2EjEKOaxGU$&X-p?jJZTS
z@CzmUh!Iz%ySTXMO7%J_3a6H4{Rx%Q53+xuP0ob>S_}?Qs)8`|8p7}1QOen--o~ss
zE+xloh&r9M3Z}3m*1i4D{Wn@%GF{eTztxfU9k=c6&{xdV=y@rU8ztp5&p;lHLuIl|
zedl_SE_5B^d-bQ^y3a~XvtP>N>^NgBEgy3{ecS4fLeUqmJ<)tEY#s{QS=Xs|6zJ`a
z=9DdusEHFb@P@0RG(txgRxkTN;HUQf-T_u%Jj?Hpvo{B3w8eU7Om2@KABu{DjSLMN
zc1p{z!rK2~w(cJlNnH0^q`#VAbu~`^X+OW0`+qFJPm_u58il?IvzSd^3myvCEc-$}
z46Furs9Um~Kr)6%0)7W<CMKqki8fE;S(~+<v&u|A^~t3hrb*<<H`F-nx>y#uD3Rvm
z6JR3JkfXMY+xMkex|peFf!-2`O)+>_(Ea$cQ-;d}!@8r%+`CBtj)fg_G7#}D@a;1c
zF+V$cfy8#yj(Ue!ySuwqV^v29>VgG?Um%qI{OamtGSh%W7>IxR?Q#NJK>cxv{5JM9
zneR=F)$o0Um658~9NZR2P&-qsvjfv0@K4jkWCXBu7vpr=n}gTTM&55@`ssB=E=7pQ
zB7kkBeuE1=L`XCPa)spZqbU_(M{-+Ngwx{+#n96o{{?o8zsi4-<58#CbCUP@Xi3aw
z_j@<r5_8d(`=oCMyV8EVIngNfXF9Kufu#*%qZ|9&>wF!t(0=Ye%>T<x*or!k59#5D
zfGRs}#X9Gu5qKe(G%&Zp)-LdDF`jos=oM_9Y3b4AsNM0}F6)C)2Unq6T{qqT8lUvn
z&7wFYhY7q^dtcZ%107fka=(4Oiempa;@$cUw$yUW$>uZTa_*h53kdr4Ej)@;l=dH^
zCTl01x|ur1hreSF;>laMkXD4E*o+m^+RmBVE>0?Ze~rrxKc5G6GrjG~a&oljgKvcB
zep56mflvs-VQjf5pmkCZ?|#UX?g9vw>mI~2%+B_%fHw?#d_kAjgJssln|i73#*aYk
zK(cC{10Ns%>guOIZhJ@Y`T3~t$+VuZL|y^El>ye`#up}~e<#jC-@xY#*=xB7e4tSw
zGFiFm-(hd#{>^0ur0>Ay&%?vhb@y}q{TZAmg%V#kX%NJR?}d;iL*C=XDqmh;2-XFG
z<|2;(g%0s1Q*pmQlkZ<s6Ysx0dvx0gD9U`J`$HqxZ~ghYFQ5><W%sk?v*L`dp<(Ov
zqxa~?_S4-5U-NLaYhcgi+%Ha+Lt31l)OV*_@%X54F%37=jS7+Ylq2;@o;f*U@d~j<
z8BF6u<)#~l@mK{tg`lR}R;ebru$q7Nv(f96Bv}1bLy#2WWdf;U|MX4mx3Z2LpYBOe
z#A1+Y%Sk8WK8$e3>gn!5CIq70)6tO?c`wjeaDNNYd&R!3Dy02h=a_{Yih<&BUwdHl
z2(j&hCnIZZ$Bu7r%&zkH<d)s`lC6sprn;1+P#mRT9Ei?gDBuu@DiKHZY=K3uqPMpo
zz&W6rEn1A&(W$*MjC>l_j@wQY5yYL`zak`S{l(^1lvy@cXjk%{CNguJ<%+&QOYZAf
zx}jhHek7|^m6$)7NQ!}P4KwMy&6e#CEP6C}-grZaM~`4zJug%3ygBk7qd5DCzn2tq
z)6DMJiynA%W@q&zhtYNnnqhSZ%W^GM;s3DjF`F&c*?I#307D&O7sai5t4oW!ob!rc
z)w}JIy6!VXk=Sq_y$?F|{_~vz{X`lx_ItQD^6q~$6;*|#To*xJ^RW^$8&Nou-K8*0
z>jpYxcxSo{`F~I#?D&TQ@;o*8=z=Fz1XqLzVCps-dAFvrm|2eCRfmdrtw;FWOUqZM
zZ)t(ag|^^sL~d@^gnP5ss-W{vhnWo6w1R)jt~v-({|3}O`?h)v+k#p$;vs=dg<Bi*
z=va7bpUbcYD@5btGX&bpOCya1Q0(w-qx<ISrMTcy0B_tHci@Ew*bD91($<LeFRlng
zw&?Cr#hC`Hd?vm#vZL%B7~fqDv%VEX4kICAP^A?W9343k=fYC7{mPss_ZKfJRV*ge
zBs9`rIy`UIuX>-K2XD`FESM%Ll!f>aUa-3zx8!VuFiyXsYHE@Wa@}vM^7qDAR7ApF
zSokVmFxcnc^6N>qa(R6H&!E>1BM&mz9%S_=sovo^vteFa5n6Xxs*5bExRAKonz46v
zz*HhcUfh{aznkzVTsQI!3s!;p{gCW!NfJ&~YCYJ;0c(IrqXRZ6x1gY#F&<WWkWV59
zBfF)grIk3Xu-h-1%O&8CCx?<zPtrgE##8^8JaC+;Z(z_r8g$AF;UhPZPZ>H87?1o>
zEiK3H-X8x^3q~m%3m-O1c77`NXye)1HFM+HIy>KLuicIK&n!(`H-!MSS%kDA?7rbL
z$XQQpvx!6hme-o#HMN(1#}R3Eg{No6Ou+@Dcj0@UPnBq4e@lSnaBiF3eAU0{ilFYZ
z*0U#ec6Q2bs|s$1{P&hCzY(nc@LnpYt4o7%0sce?bc>Rb67#VfDQ#WdXPXOZ!3DF0
zh*b#!KZ4L#mkEW3$lMlvgh&t;#|M+0@Xx?AV?d8!`;P=K?K_odc>$H$#^mlOg8PCo
zr5)HW3LXnAgb6l>ueP+bbPmdlGWzp?-yWvXX1e1y%$(1naqjYa9E;psVYYX@dicfG
z`vVRmNpAbS>uCuq0WXI?t!FeMBBCx-ud60_jv6P{YKwnbvzT6P$Oi0Xk(IoBcA=DL
z-BHhVBX{$svjSns4XX%54^`h=B(;~&A8b&#haG<5^gY~>UW0=$7u|cw|IT~k_d;<p
zwhAQ^|9MiM%+`?yA4qvvUdR_sM43F!+cCN9Fhn8{vA{8J9xlEm61pljIQ~^txPKw)
zhlflcx+6~@VMqG;t04uSje5#2W}My!D}R%}?js(mt&ClF+;0gF+FbB><AtNZ3kUQD
z$;$-Gv-o<#tG|M6g@BLKr8R%N{BPH#ll7tNYO~hdd{Z*K0}g}F7_jr)GyQkOHa>{N
zQ;T}58ym-#C1VY9=JlNaG`)?2FptT$e}<2Lhg|>LktO>(9nHld6vQ>bFS`2nqwv+1
z0W7!FH8gTdhhCh;K4qZFQX7>xKIb;lB9KefNA9XR$GIWaRw4Zi7}mm1r)*qxC1GB2
ze5SqT|IFoiRu0BwqHP>BvLbvc1)M~-wuJ~l`~2?m4WBCL-wMDXz1rGR!RSd|@^FGj
z>^4E2K57A8#K>sz;e_IrN>+If6v|-7ck8&1n#Tgl7spjAEe!w#PlEq9p!%S#p-)wO
z?DbneGb;-!IcUBScW71NxmjOd_wZm5p`P^wF<c2b+y9CRD69<S%ufUGmu4zVo!bGh
zG#4AgA2_7!!==5WzVoTSAwm!MqF0!AfZclMAriYLG&M6b0jyD+4s7<p5-ndei6|l_
zCIpg7N=hC4+9;Kza{}~90phQBnLoBWyEbwy+{U3lR#&>^Pi45vpPP_SNrf`{%?({)
za3xZXZ2bH87|QUNQc9(fGoOuli6q4WIJhl%`ayO~WZ^8keny4V4tk=E{!&+7cYLC1
zk;+(!<xAZ2fDcgVcx_zpgNT2h1=7;XpfCm<>+L>4mk-Mc{yDJ0^X$$}hV^aV>l51Q
zrJ&1|CN2mbLR*wJ467c|3C)eCoLTuV%4<7;V)E`B2FHDVD?CcW8Pu`4W(0=x5lFSc
zDhMayTTsP}p;Bn7nL=7!tlP_Ry~enb<neN#e)al~3>{Oa7dE3%B=LN6e6sp_lPZHO
zbJjX;FLivfeBvN0=Q}yFM^;dGum&|rlv+n=v;9Yg%njKSi<|W31bmk^hwGOLJ<psx
zsXq!&odwPa)-MZjVN$suz1;1QOYzod>m*$CIta+V`>}`H&w(05O}=`E7vnFZS|6Fj
zE=)~5eIpAexCx!G&j9_^;$(1pmGKT`rjJAaJ)pLTS(!k!u!MLLJ5_ucRgraKomdVu
zqdF%hkZGO13ssZ)NlCuRe3X@$<?A!fAZt%CZtpl(iOfp!=DvL7<|UeDq;nfk^uO)w
zc_otyv!-4?_8TJK>511fC~h4!V%J4-0YdG;f9iW(L{uE?ILuhN@8)biInFX=jo!17
zRHhE;L4@*syOU-AtBDJCCi-dJ4H_>IhC@0sx)2A`h$wCBptpjP!Bl2(qLqnMnqOY*
z^45i>;`-|b$r+59Uj@oOVfoa#TxH3cAc)6gpq8L&1cox|h7tAV<}S}<I?k#L@c}l)
zt?h-84C}x|+S`AmH;0SqhpQ9m3h8*vS=-SXA9SE2kcl)JKy5SV0vB9gIQF-`O-7~y
zuJh*@PnBPzdhTN%1s9j{N4IZV+ZkfwUX-jtmFWe14csO#X3LF{5Kk6#!AnEdhh6OW
zE}^AYA|Su}ZWp8<1ZK++2%~ManbvN6QdJ#OcnLiYqy{1m?8ypBr-qJCQksgeo4`38
zYV!M(m)FTi22f(6Vga5~$LZdpLeb=g*bQ%W&08~zYs2T)f`zJR;~q=pyO~GPsCJT4
z!N|3zsAo&T5=9^U*nlDw*aWE+5x77?LedOW<i<6~z6<IY1Ze&jz!O*_$kgN0Q$(Jv
zzrQG8HSl+%;mR9Ox(A^5Ygmb(6at7)r7FG-kV%Ag1l!C`*dUuArTu_(kpKv;-FW*l
zZUq>uU!%b{yn=8l(X{`3*fO((h(Oui-p++n6}}^?aygJHKfZb+bTnv08WC$-F<&?2
zoPx!LvYO{0=C_%lE6{w2;>3Ni5M0$Zpe_GYM2b`a=R|h*k+!Rzj_%&!2aI-y!JDqL
z>D=iKr3t2#d?l=cOgTPzH~<=!!*Fq0DFv@TI8|5Wxy6sJ22YlFNN|%3a~jH>0<3!3
zC9u|eoqvWw<J`AQ?;jxGBE~kre){h-%|j)9ywdx}k&Z|WabJ6VdV96oXhbZwXab{y
zo#)2Pv*%iGoH|t!ij|3*1rJ0~<4Zc&uT4&J?8N)UZLdyoj_6kuH7BBS3@Fu!(|c|s
zKS+q`*4C!SJ^SB+g5vUfh@lkH!crBr&X42Eq7SQUcV<?s**o^`Sa0W8eAgInS7lBH
z3^fJDUnA$1ZfFu4SEJZ*km!bA(V!tiM%(??eO~9EVsNK~I)b}^UIbPvkKifgrHj4!
z?DeniS*YCmRBSdSb<x)Qgo@0`aNJfo{SYxUG_-MYDy*tX8W>Q7Y4X~Sgl^{fNzT@F
z%{?qO`{}blK*Vgrk`^+i4W9$G{+OU_<exXyYq)vGy<8+g*yVTaOVZh+36#l0$qqp8
zAe9>*>W;%w>w8cVl3BG)O-+~8>uYKNjs>z1KBpuhwuM9;@#R_6I{NuqAu51s0nZX{
zt}&Shtvl)L_BrDKFFTq9-d*qx7f&FP(PsDs=@}lc!CC@|IAj3k`WYi<SJ151d_S!4
zwZTOkFv9Q&2xLYT0x(iFVm+K`iG)rKop55w>M>K@IuQoBB&D{CGYlMsJjIQT^V-zg
zQL{jhW-b9`GV-R=numg8h5VuskbQUHlAk>wf>P8Vhi*vT4w#cuNN@<E70brpCbfXM
zrEt|@Z=p8e>af@D(|=~LRQ~;SzcuRT<3uxHArb$1quvnnQ%Onxb9PoHMCu4Y5x`&~
za+neHEcozpbCI{QX;1h8(t-Qp%$fTrisYtWZV3^9Io1*mcKtM28JTYncZhN5^()On
zGi2f+e;OeI0C<l890JE2l~_y4%03nrVz4p|4-bo7RZb8iNJK=)3KFa;!0;9Z!a;rt
z;O9RXIkxPUx$zK65}pg6c_$h$*NK{lt#j4<Cj-nA!xdi|_{sdc(~zdpkBC{*OzO9I
z`^3LaI8HpGHR9su@7-NqsTJsYE6oSSr?Dwp_p>~?Xt40UC9HguOE^5zFCO}7(aC>S
z{|1TQ-4W9QQz^HK;-0@z>Bi>NMQFK@_wI)i)uueQRK=Bb)YLAbutzInGVysWEy?Zo
zpORyBs&Bg)=BlXaVe0Vn54{&n4?NTH@L<;~FAUUsSmJ@6OC-XW4l5As0Ewqq{hhe2
zomtMD-Cxk>9VA6knX$wy9D;K<@^8dme(_e|3*b_=e~uin08D1~1{|^0aNJ_FBSta*
zRSyvrut$|>4-6@Qw(~4_S9Ix|0IVZGu)43s3I>%10oX@5Fx9|JW)@OWUe5K=XxI0H
zg5>0`8(0(|(>3tYtCO4WqV22K9uLx7@N9kx7!#`HX8V0J|5g+h79ug;?>oq;t+N{a
zFboEQ4u}xNlX<m&sr;5Z6rLFb3%vTw_`Tw98xV70-iD0jc6HjG4&LhvKxWNh6h)BH
z+TMuR2I%57jg3_d4CZMfy%xMFM7=obmx4rLEUU)8-*)wPaN~Qt@CJD&_b+%JECG(h
zU@-!I3IaX~w`HheqAp{ip-%wC7-3L@qZ#!+4Wu~0I-J*Lj1C5yp|iCM13!HB9UD*x
zy74{{mrX@=?iJ>BNt=~)T0eidu*UEOtD_ok@Z0TT5S#jNPZ*uWl?{4lnaUc{P1o)A
z%dU8>ca%}5reBa03G<JVh#XG78$`*r%iX7TqMP`#8*B{P1F)gh2sU)|s8XKd#-}$8
zk>Cj*XL1U*g|fujZ_rofe)<$+!DAcpvEjidDH2xDeh$%KntYHoT=lY-0QM;kaYgUb
zm6=~k=0af%hY#}2zct^(Zn8tMQK6^>g=kbU07jdz4YL92k!<TpwKM=t@8}xL2l3#X
z8Xl9dZC9L(*<LQ*ZGtH|W7ai_B?P3hOl4i_<r%skFzmgw(Y3R;M}+1g&Qb)IjnD`p
zA~4q8+CH4FbF><HXIt9g;qUkS*q$B@qATx1WHJ{Lu(XKJORwr=cDHR=jZ8%ck9sP`
z!w)`rjV4iSoL!Lvv&is^J`h{d{sX^>>{@hEo9%RUR0Co=M;JT<MlqGln22hu{jjK%
zkUN8HwqiUgRbC-$Dti6-ZoB1##MN@F`N4OC0$Of-^;}yKImirx_|+lzV&qM3yu+4+
zR}WRpUk+N)A*d#1_a9gjujx(C2lRZPg#vR3_faMW=*0KCug;I~J(E75jtH&cO-Iz=
z&1o4L4*W;P*7Qd2wq{7%dM6~s(qL+OM#I?H@ff!tg(~2n12Ut~M~{TIa$XHJ7umij
zNU^9k;BCXESJn&_--p47chtt*=gI9>qoUTP?fVYI#Ed45smW$1xeW>kPfG0}j{cKJ
zySKYLJO3uqTf1OmCnqEGDtC8sx*^8~)AKfPX%m*`1w}=#5xeZ?YCrBQ)*)i=!MqrX
zodf}EW(H}w7BR2vKufTI4LDeJvL27S`@-A+HdCs|r=20W2w)gK<b`n9e!`FY=uri!
zO33X{hhVXngw1GGZBX3<r(uSDS+d|UITVl~ZF04@+#Szf`6Ts%$z=3p{F5FKMRAZM
zdtHT=Lr@{umiYBhCbk!f1ZBblx^1EG&@K`JlN7qJnDNZ6(udz}!x?I9Zk-*>3Yu>u
zROL!I$P-k`P^2z=C(d-4C*?Xh&95&=L0NmsmfBuV<z{dP>!6vV6Gy(_xrmqZ+Qn|U
zW&f%|bG;Sk-q(f7?~L=xqFA-P@{EdyEl9Wubb(-+&~FXkm!lRYF~_~QsM6Q9;ltFq
zMfJ#6)o&vD+<<(}p;CdIGouT)$^erhGYhKy;6klEf`QEN=kHD`w4cUC_oMj=VC7kH
z<+)OtY7wHrDld-%E&1+X6`}_MPs_>@1V$qxoJG(sMo_PSJBcmCV1Mv`7OVSkKwA7&
z14{04zb%_u)axofb{-)eS6N`w1|Hvq9d1ae3y7GO$6Mpz<)J`Q->1f47I^12!1^4P
zN$-4CBN5Uv1PczSr|({0B6g{;nt}$<qA!6SaRLFkDd6<I@897XZW7<$^XA;GCvaoV
z7LasBMZMbfpABmf!?%pm&!0czQj!qX1JI}_E-dKLpkv*rZ@d1ifZ+$s3UZ87tav(t
zh==|KVLpJ7?k(8yf&B@ex3)3uui}7S2%1)qv?hN1P=R?B7S96qdr-2KpcKFpjBpjg
zPq8$(TE>(x0~i%*H8GMUD*H6_n=Ruwb_FlAir?5wh-tU?AEG2k_hW_=QAQK9AeWE6
zP?!!$O1Tz!Dk-&b988AvFFx+Bgz8-jmWAN%-w9%gG>m3lCY<)8J2k6STK^Mb`#^}*
zWk{5i@00t|L5IZ<%UsE+BQ6ed6pFQn*^IH+JvFS3-}BWfx`E!__5C!n9YcWqKq(vc
z7nu`uT;N0WLq|b`Xy^vl7v!*_?8(jKHV@mbCm^u<@o4ER_N}K_wbt2#5vk9psUj}L
z?>XM1FD)+GkEmhhYn2d#A(QqXkapHH18!1)-z0N}auwkBYc{G-pTHFsP6&+y!!t`1
zh+&md?9b%4S3W^C3;QeLQ(885ENnMi^^`#)r{x~4=wKBzU4>6(Jyj~F2G?C2)4A0}
z_(}TR!YhH7vlz<>ij4u`xuSbj%9A#d>|(B5DK2D9O*ryLg(P%*Or0I1*E$Io=Y-OT
zaZiiaf#u13Ouw6IYO33xFkI+y%n+`oc|_<b)Wip5#ln4c4pogDW{Jv-2|S}6lZ6xG
ziz=|_>-?p_mc^b*S8=lE@VdMeZ}WL&2%Ouzk3t3}Xut+YF)Am-g)P`2dwbx>G032!
zTV6WO(YIc0=j3K=cdC&>Z2iGg0Jif$p)Cw00>krSAcfT;pdfX<DJsl!CwFpt<TP<~
zGsB~gIDj#TTq6sIzex3ohsgWDCn1><aKFbhIwn;Sg`Pnz!9z*az7a(GW<?Nd-J?`;
ze2%y?#a0JBw18!2M9g0t-mYh_l`xljVHAYj)zJ6F&GkHikNnaY;afYrcLY~>%>Od>
zsAcv+F0=|t7y-e^+s85*7!mYd6R}~R1d86d-cK%~?PRZLezD=m%JiDb4GRGs-I1Z1
zrN@$CR+9)HX{7;V?(eb8L7pqVH0@xW&}G;W94Nvt<fo*hz~ytv%qlxY&Q%;dt)&h}
zN66OigSi067pDz42&I@s5bo|9%AOLk$1x^bYnVwdH>uR*oCOhjwJq1gmp(t5I{SNu
zVUiK@y{pA*t&)0$>Dp}F#~H8hVjZByCnSWIwNF8Pw(dA@zCK+n#F?*ijKHJlgfubS
zkbNw-(0F^JaIC*PbjK*)!;R%_(Q&>f9STM@$pv#3?%USNEoyy}Gv@ZJT_tu;b77Ed
zY7=>g_4ik^SEssqQaK%#m|G)_LodBfW`5wA*H0fuvb443LL|3*O@(F_jfdH1@(s3B
zlwyw<GivVgNTMk8N;7I=A-9q_3O%4vV#Rp66D3tL`W<73o)a#%xFd0u04EIO@i@b8
zMdEr}4S~N6Uz04|_AVk;A|^)rS`x9_zZyjUG<+Mg&jx=d2vV>m2f?6-0<xwx6~(U(
zWDt_<qDV;|c$2mQ*WHA{9RTO8IrN;_hFZH4y!2>@>=UrJ2^PI`Q|z#Iy?&S@Y3)vD
znM2q)cDdJfd-EOs5SmDLRacerhVB!IT2jEf;O3lQa~LN=@?KJj`EWx@KNwt&KGI&l
z0iGSvyIq3Cw*Aw#Yj9r`$HX=t^vd~-jTs<T^h%nad`+3KamS{A3&<KYx6gXDvrc^>
ziJJz*xY6Z~R|v}_43j%Mb_ick2#VbaTt=4ht4oNyqU_i0#6Mza+L9IWN(t@Tv))P*
zS;paqqBOasF`^!$FvaDc@fJ@_{AWQ=gc~j9OT?H^cOnsi95C@X<o0llc+Yukf^<)k
znf@R&;pX)q^XWfqV`|$Q^b`+idIRhKkK<7za6GsIv2}J_wRj^X-a1$&w_Nufj&xi4
z=q7G)TGhM*viir*^b;pw*etOB$j(~cdSM5XDtwTAk*AX|GQs5tvm7)eh#MD@L(7GC
z*K*=#)FSKeR@6o5xos^0u}T#AG*sq38nUR@Z!?g@{kP1(2GeO#intE~<||Xv>|C^F
z*cDfDM1$gR1h%P4P#sXow{b98nIK|y6DU%iHS$rv6R~)eEP;%u<c5e)0?Dpgfg16!
z!sCfRg0Ob&Vqlr1wR=1a46p+O4U!bHVCdz^csFnitj7)6&`r2aW18>eeTn(!Y58T6
zrjhx1pJJ*<{xK*BM)s|DxB^a#v5YZvgr*KOum21&x0u}&c~5+{ndE2g<GY<bb<<ac
z8!;)!V<+x{@Lcj*Zz(cAc{E?AVr$C^W?7I|K~=IIQP9)s&YlJUmJDe>BsVSk)z!8~
zCKx&s78VS(qzaA?E;e5V-he^?--D|6arX92cL_27+}$FBDR+Y6fjqc0Ba2@evvf5u
z7FFd!6hX;f{W~mlG@dTFZ7R3#Q5)5AZHe(obprbY+-C$g15qR>##ethI=Q_#0Ut+i
z7&`7y?Rf8ahwReXa6mco!}m2vZ`mFVRtZ5|J!1Pee7~rcGP&PEiX^K%2aytkxROo^
zz5K&1|MkZY6|d-)YbI6Mv3z#y3-`$eHB<c&p19$QvJx!Mh~TRx2$@a(Mj#p!AnyDD
zJ#d79HlfBioObmEDqHBcJE*@NqG1n55tar=`56Ysk{Z9eZTBDn+i%ww>c76Vv!Xt3
z)V;J4S+Kc?3m-CPBKjBPKfbjW`hP6IQ{HsC1OJc&i%zEWUD*f%p5sIYMnZzRf+TAf
z4wNOC@W7OO>Nt@LmpCyu5x6R8yZ^<!xm0%)Mm1Vf(nIac$nLQz5cx(GvsBR<0jI@1
z&5D<}5rRw6gyLSF4jR_8r4P~<@J%cD;ZLOel9ZC3eP{rib6}74(=3bkK`YN1<^k>I
zTNf~X(5zbh!KhV*Byjjlys&XENI{qeMmIrQk2d%Dx1WiTv>K9TvX6nn;oA0KgaG-P
z#OT7exVV@_o}6kfEy$S;#IhC;hlae@%*NW6?o5qM%rX(E?kqvr>^Gad?zWHd!ee))
zrbiDSkEa@f4d|R^lT__(jD37JdpBm7!J$A@i0=5Q*Wkyn^*VBOI-;<b=08)yv5dx)
z-k7dVMBEXQ*Bd*^IMzPnTBu&l(fQQL@@3<Zt^c)N{P=CQ$Vl)lx$LD;t!ltu1xnh;
zry)#Z-;t#^V`Y8yM6`2HY6GHj{zv7uOhPA=mN%HBpSHjvRCf=ZCRHnO*y<ddrfMSZ
z3t=Qgn12eXdRcSI#t8c1(ySqW=sTGS%U-p^_{b76UXIN4?l~;wAO*Gwvt?<uI~wp6
z_#&Pti10Q|<1l#fPEqAG7C9R+zR2{<45`iN)^grV6OlCe{do8w1(hulNml~-fXMBE
zt=y3R9kGKDY{DQ&h<S?D0exfx%y_W2f}%+IdKYqjmSY8f(q)fk#q=t1HFlvP(F7iA
z+-pA;8%3kzr(RfeZW~5yf&E=K)v%-j16$66;fz98E%Umnu^I(Ca-V@|QxZ4c_V#Y>
zCq`I}gNrdix$w(gDE~DqbgZqtJa+N(1(T&68D!`Z6<6emyPu5|ai~wnuo@-c<;?z-
z7gew@A&Ll{xRht^KU@EZq^d62&s^}I<JVu8e;GM(=^4}3pSdUOs2d7}Q!F9z+KC@3
zEOPOX-AC%%Kav-<Gdbc-j59-u)OW$SIC<p2>Kcf?p+J#9B)~(NKy=?l(@~L?RjGJr
zhntKj!pwlcTtHGWG^SMevP`{We?;?x^6DiKrj&R#Z#xz;TB!4e`>591O4Ck#Zc*Lu
zy3b*{PINB+&D2)jH3{DEpsG1M3PPLECzR(<BD$ZWIo!tN6h=p;ZdcYOwVW8UrJhTB
zrMa9-ht>IJChE=hcjno2Uia2x6h((x(*Mp0sW&G=CD1dGeIqUB(UGM{T&n5O7QOl{
zc=9z_|Ne-IE7R8M@Y9f-z?C<$3rh1<$Vg^BmhM!ivwTB1?sK}Kupp3Dlh<B`Btdug
z$6Q=;&*;UD2&Lh(@jDIY=6tR9IdCgENak>>cMMA|zjf@6WCo$2oo@*mK;ad&y?W=2
zlBz&oiISylLtjm!TB4>|y6mxqITxowXk4R)QW4G};vCE=P^oC-|0A3DHzy}6dd}CP
zEDIE(Na98ZNpc(-pXFJ&cs^1>ye$RlDzC#<khZqArjCXpR0$>~)-cH=`G(_@b;t_1
zF+Cl5;#6MRe700O^V8qz=b4H;!wS(;dTcWEn!G+8!U8i5wl|P@R8;q*wpQ2Z($-Op
z)u);IQSBV==jl2c_TnOzFq<;cyToV%vXXp^;o(Rq%#157!$UuJ-EQ>#_vWj4ojaL2
zm7XsJ{2X!|1yZ%e+UoO((dd3hQ`pIwquJnw0+LlB|DBuTyDnN(9%IN<hZCmK3UA-N
zvs`g{sb6N_c))nVCSrbl27?89^uPoDokZD=s2O3s$aRGz?Z&E|(@!v_r@Zj;O4BWq
zZCkRovx{O>D1E18-qG{Y95B38cInKE{nh1X+>akiwJp)!_ntapV=G`AlsZ6hB6kN5
zASl&kEQRmKO-@GjCO1TXlVRYVN>gVgK0E1VKe-7c$JK6g#ZN3VdB)<+(pi`GiY?`d
z9sLXAE1VJH6ZzkCdDCRW|A{}X{$0t!G{womU>lWjQe3IgOT43FDs&HPv&HG|67fDm
zqH!d5sL8Ftr=SP1WOZ(M(j{&vCW$t^3&Uj%Urj5sbkr0q0<SMIFjjruKV+AVU%=$9
z$!778`+%ije)pcEVWVuWMoL>t12tDoRhL1vJY&`F-x$K${FFYA15d~ImN2h;GVmtZ
z<6DPCja2luqwbv<OnGkd=5{le!rH>cJ%aN-GFc7T#}sQ1)0SYHkN=tQQ^nt8%_^7@
z?3N7DfPY(HD;hv2d#>(?>jLeXZrC(j6cWkit*7OKi=tON8xL1}<))tHihEwB)Y{Lw
zL5BU`&ET8OeA@w-&P)5g1fI@ge&)}<{JvRw_747qQ9%>i6_Eo4L&@}^anSiX;vO)W
zUb`4`5ZE9y)MPsU4Dkckk;^vo@4iS_D-yTCJO8@#zLn5-eD{Q$l5*`8?G>H#?Pqu4
z9j1zc*}VrTpL9E{3@DA1e$hmVr54}y4aoV;mlp46Vix=0aixU=UjEDdL#BMJ1yU<l
zmXfh&$t5wJ7XdmiaeG<b#iu0Z#SpNFe)^A#@MTz<-S}k7jz6OgE0MwqPJY7l>5M^&
zx+RZQq&atot}GW*1V?|uzq9ooBg=3e^sv^wWG_t)FZHyeWF^)c-#tD}`KBjrtHw-G
zx(hGHG+P4>GzbV@OXnW4Z&-f4B5=Cg%z6Iav<_c#f$MYsJDN-noF#su$Zp;JU3-ZR
zmVscReh2aB0pb!x*L4Yd&Pe6bI*h*PcQt##sPef5hOe4%3{p#eEc+L;;26FqUGQ;z
z7K2-*DaJQ~r&3xF>Ad8^;ko<Jp3^||TdF`z9+Ui2P2&f9vS0oEd+JJes22F%d7EvW
zJ-DJV;FvkzcN{03KD6ffW)^`Vw_x*>P2N(`>F<uH5mu<Q&gJioAh)_{Tb;ex;jbQ5
z1cgKI8AZZ6iEFgW>>j-e^=hQ0XX&-hpUBSIZXVR1VX&8xD>UQKrTp6*Qr8|a`oZQ`
zO|+Qf)JKMIYZ_;pJXd#)#9zLAN!(~|Wfj5LzvZF%R~(XaV<axVEm>JHKh&W3O#16d
z0!|WI)c5qqUutA4OzFm(m&ZuL*@l%xiQryqLr&WFsj>A24Mth>^R`yj*7J~y$Bt|q
zK7>wBG;8>YmxbhENw(tEs=%!%xA>p`M)O$Se)v#MnrvyHyB>+7@lE=lSS^TljH5?(
ziMmOQ%4%wpPW7+R-dxm*MwwK|6-}00j7g1Q-|e-{OxpCVJ>a}TmTAnAHmR&^(TsAc
zA3Q!DiPkHQ7My+&_P6n1(SWRBqiUe+k%+zO)Qo$mzyq({OPtH%*%DiK{YQfDMh^^)
zJ#fSOj{M4pU5qO1uj~1-98PSu{KRV;pVl|=J`-y7)ArO#D2o^Tm_sr@ZTomDc>C{z
z6x59HiHG>9EN&dC_RL8!dVd1B(Xk2C|H(S4k*M5V-n{#SU(x1`hGF7EnW&GFW|cAY
zwyA-doL4XUF7nHB>MBRWk_fGb84~--2Y98Ys%FAItAxt;BJ$*)P?lTl?RYLzY4$rZ
zYlY-?JR^+Tc@bTk1_>_aeNo(>@*y}6Vl>8_>J4C#V3r!pDWM%0LZVB6@wK!PrN$tg
z4F~xxa2Ez6#nVydW7Manv@Fw@X~SofFMXgr`JGk%DO)oND71pAE*y61rQ<@uL%1v?
znDUFl8!4_wn^bT~lU{9;4ps*O_F$&o5pM5fTiCGc-WX8GsfxgkhubK&Py86|kZsv>
zm-Yz^9zA1F!|e3U_WAn$yyB^%X1*HUcU}&8qEE?ho+lf~Dix>`2MLg(@*pQl>s^NZ
z#}X)t^f{O6BA57ArfZ_K6McVWbwM8|)0|=F%iJrxidwoso{oKX{m*RZ3!#(xpEGRg
z2Sx1LT2f*tB&V9bb%rIhKG!g-ea4K<81<jsi^?AkD`)nceKkY(9bINFNU{p!<%zg0
z;iC7ST511}tg{TOy4fB!APph{(nw263Q_{n-QC@dAdPf)Nq0y~N|(~zDP01Rn~wkN
z=bYzU*L%GmP(HEuFJ@-VnzioRbO;_5jp8&W;0u*$DskgIO6|61U9cM39D8W80YwVG
zaD8?ZBKl_ObHZU6-)dA|%es(61zIxL;(K?jAYcA_EP^9a>Fp&3SwDF81>vNIFB%~S
zH^(C6@FoP6tKxNw<Lm~gt8e^2deW0_i|W6irmJp_kx^d7WnVn5GE32ZgL1a|%~C5l
zY5j~ij&3y?V|+HHY|uk`-~3Sa?=c#y0&e`Uy)l}8$GDn1hLpij=vr9a^AAzZk6t_#
zsS-=`boSWlndekpE_VS#w2YrH_t)m@5qFoGGwT(#$O?r7c!cwC*v>mDgrve37xi}2
zikM9&6!+5b(BDL~J><l=ExS4uuU^fG&q@v6izcfvn7S5`$tvM8D#(o4m>PM#Ec()F
zKlDgCEc2P*;FA{<EbNh-RIc;w1KkrhVvSk|BZPF-M{oQul}3639=9fgp^Xj7l(KT&
zqH%zf<e=9wTtE-c%Zs^wWf5;hmC$afsYKV74<<vIrUWZ?jf38%6@^p#DiMQbW?q#Y
zYn+(4f!n5|^(3M;fWg+h!X$*ulZm0MtPIj&R(uo3S7rQGp>@w6Hhend!61iEA1AdB
z*XIqm-QdD+0H8{Oo_EfwL6?NDY^RqS2nuoZ%Fwi2tQQ5(y&<e0tIde(b!pk_H1YG_
z9jN=tTLgHJ9GK4aMa!}}s4UC~rS+aejFDX;L|k1&@be@}Q$Iz;%%Z;3Zuo>^K|??h
zvH}9s?J(j<_bgN$Tm2bt?!BwOk;m&h{043UihI#~FUxZhxT+Mk2?K=+CSRl)T2yHC
za~!Vj>PNu($4Qu|Q$Lo9QGo7jkT>{$|J(UsB(6pzFG(%3A;*2A>UBhveI@^Or4xl1
z_nTRlq{Ca4re&JX2X1R|40KzBb}gGcv*lXP%nV2-BSWojt3r~Td-3H!*U+0ch;ayp
zrn(xMsYRCcgC~GcUwLLW<9%{ahF}=W$8|sB{W}J*y9fVJ+4zDjjJgXxDJhNYlojc^
zH)OMBHm|-?EqQcwl#Q2Hx^bzH78Q8jt7%;LyZ=m%i;Z#vj?kdh6HxmxFfczf7fNtN
zzmqM~K_I`HD?#R|;1E?DKXOe?O*I7Z8PuUXlK!5ZgX5Pp#tmlQYiGF3PyVTuMJrm!
zOpq>eRTaH{1Na`oii^3Z>l-BP5Y=yx{^sV9ZQ6*WH@vsRe=QJ>l~1`}t6trlJbeGK
zV|7d-dds1X_&%Em;=z9W)5|_xVr!r}KLbW!?7(;W&#hTzn)`FH@A-3-^3R<YUh!v6
z=L-mB`mbn)E>>yN&ZQ3Z9^qv+wPTCbE1zGvntNsQCA0mxBTSOEsyusr%uNbeabC16
zJ_XhD*t<oYvgq;CiE;iuNbG{Jt}hFvS@Y^s%j_QB%6U!Lb=Lajf5}WpXJGTau3Bmz
zm1Mrp*1!;*f52?06HP)Se;1p)X*#%fd?FXs5xT)M``_o?&;E@3=g4SN%ArMzyO{uf
zAaX>E{*`|q={1T-wfkO-@%e{-Efi*t-#)F^H+12j()}#l?5}~kq)w~Cs-_H!{A{sP
z1(_-~Xqj2dt&d@s>#=0_`GWLtl3<f&2+uDW%cD4DO)`GZm~Ajrq_*OfC?a8|zIx&1
z5tdPnzFgMm#f3w&PB?g=;f~v&deGEQc_-V&t>?a+tEz{4iz3Rso-fTxGsKovyXyzY
zdAASzQYeAgs&}?iuZ3a(S2=q2t8rtr@L$$yMAlw|^9MzG>i4+hI6@MU-+vcM6?NWM
zLrV0KSiDA1$mw|tZ!aO*^WV2eqTd+rM0{Z*eRG;-%UWvt+FI%QjmuEhE%$Fl5{ji_
z?yfl;zx<WG`^{729{J=Z{;3h~=|*cN`qT(F+an#<I+%PWM2=qrHy3T1W8YQuRT8OC
zMw<q+o{KR%eQ~p$AxPr2tX!3MyR8gUyeqppO~YLrxjoKsW<q?Am2-ky@rAS@Y1KEE
z$hg;GCE&+xgXehdi-f)Ud6Ox*5^)p$^=W514WU6e%$%st$d|@DC+bg+^^||b4H^DB
zLTe6K<V7l+F!zOAg!-1%DC87D1qAl+=b8@-_Qxst@%}>A@N`xp{u4zOvAZ`4fhvnu
zth^sa=C8RLlBXn*{m}W(O$#P<7(@>e*cXr7SQrXLP=h!{U#%zpjibQSFpopFpn6kz
zoS)@smF_g;MZ3t){at*}#3JYQqO5h4Qn*itK;q}}*C;eW-8=!+JrEl?q2SP3o+TM4
zY7EI<o2r<2+X9L4Q8saUj0S?buwpZzZ?6&XN0&+lrf<Is{0@sHk;NQCVzq1^&rr#<
zj<@ohPKX@iFeKo2h8(2{ect-6I7WgTFxFEN_Ot(f?c0y`-EaqzSLBKv`cJQazdT)B
zY&t9%iO9ph{cD4&eZK8k+oS=U5x;3N4!&;IbL1@Gp?fblwAV5cc@Q_)inToPm87g3
zeh%@7<e3xbXg1x(=w$vobo#J_8w=;-m*Vhv3(}YB#OKAG$!Bd5UzsD?DR>hqI=On~
za()Vq*KGCQ=2*mjX>P5*cD&38DE@b%+#(p@ds)vIN{q>`gw{&7?$xVE6Bu((r#~44
zgL^}KWy&PC@>SU)nl&?sF{pbtTbw%fGr6p5{@x2G*WgpD@!~)3!i5(%{5_=T#2+)x
zx&QWf&tLku7jP*iGk4-()OF&ADCI5ojXQ>>o%e@+uXQ$vlK0}D<BSRMP{s|Sl!WgV
zq#q=%CfM_eMN*YwL6liwvi#%R9^4#si_FT!d4O0IB(Usohiir{30GkQ%M~=cb|@%F
zm}|@#z448tl|!`Rc1yyoZFDmgIrRlY`!1Giy3p9v2<`WaFgA6OkLSy?$73my3+99q
zHK(b+qDN;&uZiE$wC^s8vpacI_J^NnW!{{uG&6k9U;T31ouRlf>t&djY4gz7_PSo*
zcZ}ALimjPc#lFg?L;o!%q05KE$kpK6jj-FBl{ks^{65}Q{@TdF(r~Gg+gdsQ*plg4
zM6xfsTTfJ9V6T4WO;DcW(cpR%hdWpW{c7e{g17E!YWIAgnE3bZo2VOSg=&4Fm3mn=
zOZORN9w*`>eFXN0r@Ex;Whpq-a6w5<Eg96!p*$9iw-aLd3dl4Q4LMVi1I`i1;i(tT
zRBVZ`3E@!LwGxrxA{r<W>tr2AxHDX+!hUK=Xv0~PgkjMlXubFeSe~I|Fwfx=JWQs?
z`9+#mHZ8A!4TRw!Q%&3;k`#l$IIc2MqMJNPgpT=s;Ae#P?Coth*=X3h<7aOyLAo$!
z142dF09RQ@&W~^C+zls0Kent0N%cDiWH=8$PNQl;!pd$%B)<w;E~;|oD;W5Me&jW(
zl=#JC*nPIY^?m*ee{xuHQIttrtKyus_fo9++p~F#@WAzR%p;^-X5N~UWrAHW4hbw!
z)gk@!Bw>GL*R9QG*ug0i$xhzQgHGcm(W|b9Pl_Gw;-WCKRZUH}<sxxD@AAt%4<;sD
z46V%@OxwymI?9{43JDx_tJK+*$_C_kcax?Bg`7)v7?CYmhv9Ne&}_7#VFgQwb<B6C
zDcS5X5eKIy@wLg4gu~%6HHemZ#M;;+D|Y)nTmKtD)YN``pO=h{OGA6~PN+}&3$}@-
zKPR_4>$7VwDM^Y>Dk>C>6lMHuR`<$*ASz`5u8nMTs1&r8Ez75siI(|Eh9!jx9}rf_
z()1^z=Z9fg+s>DzACpXuo@IlXJ(>~9C{2G8ObKu|k6uQRORA{qsYW~dWNZwUeB&Pc
zyzTl|k_pGw6h(9$X=qV0lAZROLh4WafF3k^HMbsL+j`PNUcVm0^4z4tq$r~Y4WStm
z+YNo>2pf?mWO9pvmHu;LV|XPhsxSd*t&k)COtFMwc`~Qb-DlR}vFr6bF60LTi!b<-
z63ih7KGR#%Q?d{2OZRAe$zfARqIy9Cb49X-%EOU-oN3rAvg=o2JAx7S_im*>LyBls
z@vyreidGIT1FLszTN_-NAUev<ich2H6I!yobH(Wnx&+1z3`Vx{5}C|>Y^3-zX0oi`
zUca2fX>jfB&K`eHMD1^1wW_!Jbou8u-zbfoUn5DPV42y>+7tX*hESN=jeoCXNHv`h
z&lA7mVW5Y|4tDZ%i@SlsAoFkb-lg`KR*!>}ye}3X@c$d-EzE$2gv*4GlEhud=p&c{
z=buU-JL^>Nt#M02H6pM2;+#psFP1Q=$Qe~j2pJyE{9O}MsRdTj8;wl`CAF~uHX{aP
z;RfPY8s?JP3pvfBH^`K{U$^EbP`8^`Fwn7z7z#_n9F)g~X82z3d*A+fDPMg=re(95
zVjp5rQN|X<s%ms`l2WuE>yV#V`sSuEk+yKWf%7}H*WL3lQn)$})x`AJ#~Y7nwYXh;
zWBdaXWCSg=@NR?1BKFztuQ%)#_Hv=qZ*9{UJkz`PjMu-*;5@ckhQH7M8#1;xC8>*<
zDm}e8Vb0OoF>1`M4jH)Q(tMmJFr`$_c8j#-e%iabTk>w3v9i5_^a*9{3df|+iCwQO
z$IVz9Br1^PTe>pEn8J$bUf74v$85LUe9L(q$B}|ulW&~)B5vqB$U$#6+|If|Rrh3B
zGfzzyo;2yIuIXt9uEFDjm}vYF-%EC{<0u}p2nM`+axdQ@o)8?20@C;Uc|lUU@=shf
z&aq(#=x_qB9V8M}R0fLu{P1~h7sI;M{yTh3)Qx<GU&H)x<(4>C)P_S7dPa~b2PZVi
zKr<mi-&y!+5iUuvN<cO*-U(tfl2)SDI-90uU7lfc`E2U9O$?*;W~6YEEP0xtTvsn<
z$Uq3qD7=qL1|}Ri!Vc@!8SA3_PXgoq8rJZ<6C+$QQ>`83rP(Z+#P?R1Ohk9?HOimh
zB=Zmxe0%W&F>!+4$nt7zz9s8%3{|o?kF$WQcrIlo{LJT|6=uHNc*dup?z3Y6AFU8@
zI3d;6dKV;q;-OnaE?<f3Kc8NF5gdk5s6wsOO@?KV_~Xp&ji3+~q{)r`jTvcK$@!-#
z9KkG!%(MwTmMg2nA~)_;in%P5>8097&G2D;*I<i!%E@;tVxgtJF}Ok5KS+{7%$cme
z?>yX`WI`fDxM=d>18zhxRl2Vfloe*Q#x$?+4V&NQ`-mSjLclU{i|orgH2TR?$TX+v
z>Gmdb$j4z<uj4Jfp;7uD%|I9t|F<rJbnLW27{pX?rL8(~x}p5NuV`~&!fH!faR&w7
zJwu96G$v#3JO6iA)X$#aPGN<TI4q&uk0za5q=8zXyyfoqiEfsM=LiAD!d6n*kLm7M
zlLt{w$@zc?)d>p@Yg-hSZV&|(t-_|<!`r0drvvvshL$uZq*~G0w9q;W6Qu!0G&nYZ
zN5v?n#P*5HJ<s`*DrORK;yN+aB&l2cH?0LYMk+QE4>0H_byUj4s5%K%$F#^&8eJ3g
zSgb$a->}b;i)bWU_}vD^mnY>9YI7Hl4C^EiK$cieU%n}hWW_fw9hhy-;RwGq8S%>a
zaVNSqP<NxVYcp*!%TCU$MRg`q{J4<LtoR{LH>Jr(APZITJ)42~L;LIsRu(cZ7CouF
zFXhFHbS8{>6HISq=22LRuB6LAA?ko#$&$k#0lc40=7b6ga}dT`qdD~I;4F0Vlh$4l
zEBtQ{sT**^N=XI3xP_0V<o`XGZ;ZgYxK69k!ssBOAH$fW*UDzy0Y`$Bm|zS0O-fmv
z^@LaeTSI+BK=3%Z2%Qag;0Z4266K+>CqqoxnKi_DJ!~<G2$ID;zoU`%XOY+@UEjnY
zj%(0kWaiJ?nY~N2F*|t{!o!Z2H?w>g_H|r&JNe~BCiNB0a4c(7O3$^2-_fv(RUB4T
ztv!FK)b+?Ls~k_Ykn1A*@1k{9P{K0{p|Ns}%HvGL##PZVsFmIzZ15LO4p{dnv1NZ(
z5Wbo8@207;?;f*00~(lU6vS(wbyHt>aD!$TllAT(WY*9I#V}f1c<NaXH2ku2@{YSE
z&>V$p1W}F8|E+E={*t=^H*ShifY0tty4~!^2Sl_U+q5dhRgck%qM4OTijbWeZuF&@
zm|{!3sb!?7T``iO64%OO;Tg5{7@UPt?~-2Y(o25!vw447V62(ds++Z_7&vdeozkiW
zdJh<nw%0JA^x7Phb5%J-A)xjC8!>H3l9yQG@yB3=rn#x^Ko)zY<k&E|@*vwxAw<+_
zUDbtaC0<$$UdoY^;s4D89BK`}P(+hR6qXv)k&Jh3838#pC^m_;h)fb`Lios)$olz_
zYcWnEPK612Ar8oC7VRt8<;UWcMmu8|%TU|BOGu8v(wTgxab(L7i$WfJk943Y0HNB;
zr|AA;T0l>bDxYKNIEyp%Cnl<PiEl3QHMWq-(bO;j3m#<in>c&aAU5@a+HjimT4|%|
zL&jFKK(PZSt@z9${kKJ0;xg7>3M9fS0Z|4WtK!zO9jnE22NOrGWlMRqd#)`O6{%@y
zaBJZ35=!OzNLL?qjrVVUgU?Xp!+R|B3}jS6G6_UDaY&X!v7IFTLg82Eubfi-9(+~e
z|GvayGCOf4Nl9U=&xJH<7=`J6ZEaCuWAmG<y?O``%IR!4HPkJw%Sb@Zjb;1z%|v2a
zD!zPC9rJy`JK;R+K&o+P>R5~bf7^tNXI$SaG$^;Luis%#1WUi(asUi65!GbI7-=;j
z6<TX1Md|Q~m*%;Z6>8QJEY{&#BgQ4WDX5TRh{w!B97PVv+r=oJoh~0np%Oocev4|m
znOJ+(%%5ra&H}9aK!+Jtj-W;dbS>-hdcNDed>m0llw1fWU3eZ=#~EMM<&mhWHcCFL
z{J$#<=+9oGti*Zt_SGZazqJ4#^Xym~Gjp{aj-pp4bzi{h?Ukdbj2QmW*<CqWgZ%Fu
zBIs{@SznG|O)XXUrS~fwIq~M^W@J<p<u;iV-OwU3+vIe_YauQkV*<%|bWy9BUukAS
zH_e@jcwfXm;92Ypl`fGW!WLQ%;ay_Cp1aa#e!m<{ESLM&Wx}$OPj}?(5BGO5&1Y5>
zjW5&wEXCBIl#^CDM}vB0J4*Z!7ca;|qz7Lk?Ym_=q3dN$Lr-SfZ?+AjiRQ2zY-#iv
zo31Uoq9!9COT^CJb8+!*YHI4@?qM*GWow#{l(gy(&LNd?-T&@@<R%Ojv!IHRh?3`b
zavS86JoTIQKTZ9WqFjCJ_hYXr-iMsLstWsexHI*Ft9C#cMU7X1L2;j85Dj6CrxKRx
zXVcd$RMJ|&SCWcXp$H@&(xQ?SN;WO9l!`|ew?a@fN?+kTSDc}U*}*bP<MhcF$!X3}
z#fwZNd|UJG2q8T$p~I2OywTm>o412h*|XiF!RvXv$ZUgcxIW}+dWnX(IEIbA$YR=n
z>y~jjTf5rH*^U=;F<wh0y~6gGa=wPqw$4-+o(TO`Y#Rw}3FIXsS?iW;L0<C6wFOHs
z2=h~m?iVQ=c>0K+WL&)&TK}$PG_re~_SlZEf^A%*Tcr?1gvUb<g(U9hi#U4;nlb!s
zP4~|fNnfjW@7@z!{%;(}gJPZ**?62?LA)HSkVJnfNePwS>v_1qcXhnaz8jWLb#V)q
zy>-;+<wXlQ+-%J$m{>$3aan3t2=0K~k#3b5jqar@!pS~xOIM4-({<-KacMd$ix5^u
z2o4imY!Ws)x9OGjA%7p4vlr20*2LSlszPv-N7lD;s~{H{5c`?pMJx>$pkRJ<@*PDB
zov<KFdYH73wRcG=95eS-!&?Lv!#Iki9gTEFOC>3-3S7%HwT2WWsT#(z29ya`%d}+I
zWRzL0@Vt`B%>TW~VN*yd1evBo9R;|K*oVkN?Zs#t!DOh~Eo|!<i;K3H&FBFVm7lI{
zlX6s9VuBz3y?gx$X1_1&TqY<L`I(buaZ?MnJc3UVPz&E>mIu)D{@K?Ev5|}jX^By&
zLv<?LmtZz>&lVCmJ4nx<8i5HDF^nSGlAh#Nx-EH=gj`P0a7N2fO}`pHOK2|5(A+&I
zkL<_UfYSv};g`3UK)+4tE~2yz`>gY3SsrFqvjAOD0I?7rS$u;r=q3Mqjg4xd&5SP1
zu$(ACk5e%&#E;sN1V*X`x-a$@eTJ={*AQbe6t;HKV3#9&!?K~9z20LDm`$Tn<kB=4
zw5am?_1ct35-TYO#?}dI8%f38PEy;LD?KlYI-&M!_v9#BT|K@b3)kdXyt3MQ<S%+Z
z8Lzqo`t(71VL!1ZPOKnyvhT>Kf$zxR7yVpNDn{Yk@{qd8hF~)te(SL$5vbv4&<ILq
zxJuvvpVoUrZQ)X^x&|&fg^0yZ196h!a#~agHIXnj&&KUyW+=M6wlZA`$wCowU&>L~
zqGGgr41AV|XjmOxq*r3x+C()fON<9+XDvrp3A^q|!f!*%*Imfr-LupujNP*miY#BT
zCQ}V4nB*yekT%YxLM=kCN?nvq>);oV`LOg(!y9TR5mdgkwyoD?Cna?AV<%Q#>8-IF
zqAL`!j!MKTR8ckG)lqeIBi4IyG=6*~IrS<1L5*~&xzonhg)ujSOlG#?ov`;o<GOcl
z`^RaU-?NmrUrI*~8VUl4J@k-|po!48;dz2wep0WrcDJvT{5edule1T0QA2*B3gU1r
zY}d_Y1`Ze{8vGMzw4hfG(=hmZk%y%8r5eUrzZqDNib;R{Bg~O~bY3Ps`)$^8k-S>*
zZ#=Xs!*O4mX0bqiRqr@;vQ{<K4#N?W*CCI}dso5U!H=fiFE#hXgYI&f)sOpv9u4<~
zg7`JhZ^n@@yca!4nG;{#Hj!snDp^^O8g}IVHM~>!ZNSJ;UGu_wWSk$;etB8G)x){|
zwdV%8Dk!6~LJh5gL+fDB#V7<hpa#x#h`$tuM(9FaFpevzpX4p|enduU!CGe<#-_sM
zg>hzSx1Pe=m}W0o|AhJjyhK>k>y`KIcz#vZX+^5qF<K{@_7OW@DeI&k+H!|ayey7S
zxW-$3dT!7fJIB8k+I8tyNtftz%?Y~jK|FTc^)ky8E;qS%1tA&E(TqbZy(+sa+~Qz0
zpRsx#rTE9Vq&oYSV$q@P-4w0H!FgK9L3vg$IJxso$vFEVOf)m5>nwxM|Cx7Cz}6Il
zEAprBO&l2svU%CBvPv9RG)Bok<2*FjZxdf{c8@($GyeX|^q-ae(0`an(Ylhhgh~=q
z2*`@i-nc%?sW-@=m@3gROrn-WnEfKd=A`?@k4jond3C`>%^;{THa#=5;>_-qsGkg0
zq>Cxip!T_C#ROu3gX-?P>0cHh3fqEcnyxwCyWjOMEUAxb&T?y!-^i$-jTgl*o(mFe
z#!AJO;`ZX~b(3L-VPYi}SSTBOr%lR<kL<p>Qcxljk$p#o&O?G2go%~8TcC+qP35j<
zfKVWGeH#3wH9NtCpqG2{4_DWIpo&M=&$VhTN>ibD+Q7F+Y0ZLKo!4^q$@DBiq!!=M
zZF6V-<ebB$zQO4wQ}uTa@UxGv|9S+8^=ErX5;_36EehG#QNy~JKqsz5oc0ka<W^d?
zrDN4hFWrV(q1ZPl?v;|0=$nj_^nogC#1G3knYlBT<P|xq2UCMs=n7l9TB$!F#EX`_
zeSH&)SiOQ3YGpNs7Xe0SMqGqdlFm+K=wdvXa@3sL?sCObxpN1qFOA7Nv_lbzh=_!(
zb)Qdn*Jn@-#JVZt{8GSZVkvpitywmcw|cA|aL+hYdKKs=?%j!v7`EPYYa-}vfXz>7
z=<o_r9MWa(hVFARXH`!RTg8u$US+N+RyEh7rCl$-B2<(0p1CeXsdUl2#IyP<`V#kU
zFk;k5m(_#DpH<j({2Ipmr6aXI(nUJtXb4Iq8F6q;w<Gr^sCaN)YW&!Vwn0%O5c*HV
z%NChvKms3v5cqy6IlkwTI}kyJrRN=-UssNlq|Kz96F?SC9bXO9MDCJ!u}Jeuf7dYK
z*3{`BeI4>!;>XCqK!3x+W$u2+ytNTCjLx=>A~vN1vIr!thdd4kr=VcwAin`GH1{Fs
zYWhG2)alyg4-!GNESN*e2JvTz(5g6=m<|PqwY~Kl;%Of)A<oKG!VmNB(h7)6W2~~l
zAaGq0Hjdp^9B^95FViICp~4NJAQz^n{<MFZpIuSZmQaMwbiZ*Yra=jZct+s+8S=*y
zR8FFS`4Eu*z6N=QuB#Y!NlvUiSM)zHnyO)ME#7JU0U__nYqQ#HfFoTtvp#+fCOJqB
zS6^pfVd~2qpMe?Y)F2N|6v%UzlH*#<-V!GlO}mQcis+3=3{O-`4f1Um>$3{n8g?mW
zf_pl!dCnwvv|=+Anq$?vT(8kOyf3r)F1`P=l>ni2N}igx5x1;zFE2Kg5$3Pl`d(1*
z-zy!#go>4_fJ0b_6{vD?<}YH3Og>-dP6>*1dQ$ztmIIU&LHjZlBzZJGn%)=G;K^cT
zr$|YZb8;PIvoCra4c&NH8n>;%+W45D5pgb$`P}h`thj1_>7gj3rL2*X=GPR{e3>V)
z6>BkJ<+69>6i@YwC#Z*F8h=r%s6o#8>EQ)=ChfW2^QzKkEE*XM8ck19Z*U{mGsu}y
z*>W-Isb9iHbZSH-qA0UZ{YY(ZbeEyl+{ic7a%%k}UMgaMzSXOM<DKz6GA3qX5Q`0A
zcDvx)PeO4+G%7fAj<7-6x-M8zz1V0MmRUn5<_)S+(GHe^3Qod+SRaez0)j;=Cu@gF
zGDVbbKRVRFME(t_YaWzbiCSFk2EM`k1a|OOVb{t<r2<AiTCiZ0vq<b>j^55V4d%t8
zQzHolVe?lrjbMqqk)UDoP%grX{*o&}?wc9T{ha)n;VT<Vff`0iNh}cm!zWaS+s~A$
zE`yfz*6Ux)JF&tTAL&#p-VL9(V!6Zi_Vxy!bp-7d^Np2BR4eR~mGc2LpU&qM7vj&$
zj)$O5KWS~V>S2A4kB<7mhtyc6o{c3`H1|U{YoKvPYU+A8L*@LNKi`-XcHKv<_3yM6
zk2oKEI)*dn-i)1>Ms+_1VP1H&{c)>dB+@b^3almDv8>p?x?|L;3k8pNco-d&?95z=
zag8?Do6s^7>o+b{sL=(h=6vLWGIa)_L}Nu9a3_o-39K5;erkT0WJHdE0h{dtP)P8t
z^rvJ6hvmi*b3I*N$*X@hcqYi^N3lQ11Mn$TIObXcK~#ITeD>M*>izPkDOtX&rm=0{
z2`HM{XPG%T-Dmbm08pG5K>SSv<f0XsNLdwA#70C_?K;y8I?~4AKM4mJ0|S!3Z+cg^
zmR4_Fsq5g=jO5N)>ohtHKY%V)N@+IYXdyhWs}}Y1A)1FXPN8TQTxzZ5`W>dC;$rvH
z2t6GXEyMKNN8vX03Oyi(Vvv{-M91D1h}dpEsO-V-L1E@KY&jCJ2r>LvLFW^6^k-wL
zX*YAIOb<dI{o!`Lv=YK|6~SNe2`*%Flyr4~d`!SZq5#d5(EJAgGQ}z-n?BD@Xxdv>
zD9qWu(#EU*lNLomCt87v;xsw&n=-v@zE8Y2cAS-W2I?9mZ2}$RZkM&$&pV;D>w2C?
zzw%!Q3<8R_Gqe?>bq#@eZROafkrXL1>D6;#wxI^01Ou{JlWO-(HJIPmMD|<S>W{mE
zEGQ7V7i0TQ6?wEnj6*Bce|@rQJF4_co_uvqWqr0eifx=j<x>9UV$4>wrGTQQ^A{iR
z%RbH?Fn+vxJMcq*wCu0WNLL842CfkGhRQa<kIO~UoWzY1V?C`IcP7@VJpyQN!{geE
z-}|fbws)W*WZ2yO;Xr?^TNHaK+<4u`O>g$k<h(j03zG%^XD=E4Ux`nsc}863Q63nI
zc!%(@kll&M|8De(<8P)AV)Jlb1$(ksrcE9|=53AMYrS2yRF@R8d)~{19{f)FWq|~U
zv{9ccCwTwWyTj(R`H=zuK$wA|T-vgXpb+{m=m|aNKIsg|#Z<+hB~OZBQi7v<B~d*}
z246BQK8Ak)M!8#1#NF`|U5fXSQ=n{9G^s<U-w>25=`~t4tbGTI)2uIMhl7LgyGFrn
zqFdALIyakKHN^abHEh!?@<h?x5zq`heNin_^GirN3i#=<gw*OSq9L>ce?RBz9Rhth
z)mt~V`FW(IGV1Y>lxWSG5%gO&us9zNUxrUAu}Yex%1h<Q6l1|KBHD(342)G_^qw`+
zdq~qSno=u}^k)K%LIAkE==o#4SDR}KA;e+tJbO&TJi{k00T1_xF~%+vj*OL^Yq#IU
zPdUErQYV}aUCA#A^&3vNmfA{v@ZSq`W}k_53*}g<@K%<B1+FRGkq+NeX&Vm{X%`<@
zu;UDqkcf};Q8YMJ<W|$BGOgq&9?*(U(0Alx=G(g2deC%;R*|&HwH$&^t*=h~Mse2-
z*|Vj;hwL8_Huzd1t<%g0z$@;7UhZw6&wxYZd{w4iC5GGuB)iabQyCi+Kn`j?SU_$o
zQ!a#WEaVXJpj!&l#{Yw1=wmbB`Du3eeCWmh+kR&URh>ae96*|@(nb?%@=-#1>XXFY
zLbc&DuoO~DC!^Pg)F*>7_CRpERdHXtNcUh;<HA>WJRWqIf$<3?9fPXU!$X@_D+qe%
zX&D(?_D`4g6I$vJv?^77qI!u65BOB2aL|Z(1mJ^pHA?7;j$pr;95t(D*M4KxewI`P
zZR~U2Ni_$$rr^BcNY?w45P>bAaROp5D1jGD0#KD#NPCr#idyJaDOyCX+`2OaFo-hm
z{<g)K3+yfc!s*`(ud@M(Z${A#|JkzUzr04!!{{(VbC`_F5hTLG!BXH^Z?>-71g%lD
zzW05;!=9@U^8l1hL4vN+HtWju3iGz4tJd8Qt%uuv0SKyEh3a>TaH4Eq-6};*Kbdsv
zd!JQld}S25fR{|(s1v1X&no>cB#togadt2@Xx01R*a;Nk)@*bpQL$~`t={sp6Par?
zf1(53V=T~2Cje~NEY^iZ5xsW;_y8xskvX-;^Y&TV#0t2Rax3e7k$q<v&37gNMeXg3
zq`Pm$iG$K=6IWMUa(M7`cfi$tsN%XzDr{C(YbVF2&4&t`x&aZ|?S?@f9x3~#SpHnQ
z<9tbYN$-9k;s(;#YRUp@Pdw)N_x$>-!#j0T-|3$0o3jNPk}>@^k<^%{;}~af+iP78
z)Szt*;3J_-ruuPf;KzUpPdBeA8=1Rj2og+7H^|q;)vlsyxz>-Ef)L6BMe;rgMO+&H
z9=l|r2>H32)qa~+R%U(i;_2clBo7oMtbeY&#jwAIuwN$#++at4*qL9q5LAsYDRaq<
z;y|X06>{nl@IwDGD+J1FMT=KX(d2`E4s^B47%9}K7*OFfgQ~S=*ZB7x+cr>8Ery0W
zOfO+&)4CL=?1z5-yTfOlJud`?p?GHC;JjHF^XY|lU;rEqbOLGGJI5next`}1X*yAL
zF`;AZz9D_o7vA*(YtKIrOS?w2PSC8%1qc^hXQm-9>JJn=C7*s+7vJ<FkgoZ*pTF-H
z;jC7zsy`6$eI#|%99{C+w8cN@vk}5r?D)+IW0F&0pFi+Tk{>8}Yd-BO{l08y%vC6<
zGlOU}92<w_VNopfnkif(p)<{6Zd=4r%s<jtOr&eWMEn!j)kw4Zio6Y`DeT#5Rti_4
zv+q;8{HBDUndbKj|A;IBzoJX>wbzNE-H%f<;J7yDzfw)RrWS4JMDcS$)Wv`8Zi1$+
z>%y_xvcm(Ip4`P5UVLlk?$nx6jZ3CRj6`R3%yq|yriHdyN+@$)PoB#jL$;=MFUUx&
zXP2S4K<-VK;_qYpo`$la<CarLu9o#${qW@^^BFGYo`3`z-14_QhBAgP{uNR=l_YlM
z{f@6As4m;g5JAb7N3|XdQrHe`VtB$v&7hLhUOvqyHFnIn2*;kl#Y)XH5o%o84X=8l
zN#~R|G7WYhGZpL5&#5mQ_jYK;vNgf7S)ig&{B-W0ZRZg<RCz7$bCmIZZJdjyKkDPR
zz$NvH9rN_Yv)ZyMLno&+<7htAq{8800LPZ@=-!%@gu&X(H)$4?TeLfs4r%+csqacG
zggjf>k-n15*o&KH?|onef0E&?UelK)z4zJN(xvxn+CbBWz+=RS_M5g>X#NWOdU|W+
zy?-<@QFEz^bTVlJyHPNOjH0$nkM)Mmb`Rx?XS%tgK4F!M4g@mF8tTOEruq2BCX~95
z*d3&>$Y9EYIw}|_<Pa*$0s}<&_kqjPCvtcoAgZ!CovqXzf}&>Ae_s^QH-7_VkS`5f
z(pwB1@!)}67TTW%t=*07xB-tqv(b(Obog*VTRf>T5ftgCzr7~EY#Q^9)~kY61jBYh
z1Gr?<`(OI^qOV494q*d+$tvO?lb2f5#nu{tsMKpum~;aa=U23@JyVi;+_8TwiZrl!
zzCby<RQ{-r*I1Bv|FHV>0l;bWW|&-7ugL0jfei`zcu!9q&CJZM56uOJxOX$y6B`XY
zJkq3@+v<2yR_szho4}dh3$HKv(LCY$H%a<*9lRH@5W{&S78QHzx8O?q2>Naz2Vi0N
zkj`nlG3RqM7t(rA!PMZXEBa54U}5G&O8%`ht=%50DVMBe&!J}{6`SZrC(q+i9<#%Y
zs=DL;wmzjK((yY&6``aAby~NDus-K4*1j_UY)uGx|DpR}j3<EYh>;!^%dfWZ@@2S^
z%2-8<`7{0Its!EUX+@b4SIxw?f(2OG7@GdB%1MfH4%*K?@@n{WN<D70o^#(MDqn-f
z{C=X<Pq-niYeDxt0^VurV1a=`Zy(fOcQUtKQ$v3TFvLutT+OZ><tf8r6(XFp%I<Qi
zFe#cdKQ=!el8)gIHve#-$*A)q7C=W6%zW}|S5M4*cCUW3hK_$!aVcMvzP;pn7-n~e
zl<D^*t+u6Ibv$|bTqLOKc<gTj8b%Ok1P*}yKxqW|(I29pVd(rlh(}IlTnTtkD-SF%
z^tC`*jIY&`Pe0l{Y&4<<lan!MfKkn?9cHHByL)cG^WyO&Mi9Mm`?5sDpBo;Ax?pab
zz9o8cT^4nCuaVDZ@Y%!e(_IzR^XCPzPPyY#xMm=^pCJ`fCzm=3#FpPGF^*Z+J@&wR
zVZYeF-{HZNczinHiK-i6k|5-<<i7S7a&=ND{Ydf!P%V=6;-;)V-?s`bu<?k{NQtgq
zBbZ6is>B;Pl=H0#j(&LJ1MD^v{<FB4li;}yKVkiAIH>F0p&La{V;QtcYB;XJLhltQ
zNDcUKwVcp#Kh&`Wj21;DGb7?BHw`Bfj*t3P{hB_VYZ)L{04?{IRi=us9p#Vs+<vF4
zA-MGMkvHtZNZw@vt+Hj^!)LS6C!u^BL6RJ?_^E-{?%7)L3xS`nJ&(7SK4LJ^9c-s+
zspMBRbgzBYsE&O)O}F1HNRZo@-_-9VALA{E&#s%H{Z=h#7)~$b#H7+BIH&qO<8hOr
zwK;s^w;^>a7RB4bFWsRQ0{_%?^fS}4Qjz_Cg}l)Bne90BeVi5~Lrdum{;snQwPg3t
zd9H6NTc%t7&iu!bnEnMb-KZfF%An)ebNRA|mljnzk4<IPXnSxQ>;at>!^mW@&03~t
z;jNGL`c^(y{o?LgtUtKE>uACLNAA1LaGAgI8YlwWqQ0Ic2MEzJ`uixJob<{<?Ycgu
zIq^9-;v}a^ftY}S_Cl~Tg}EsauVYJ;*!SKMW_Wm%g|h)B|9rB0)H)n&GLz^pwHol=
z|8^o;Z9XFcSl4>K>l4cREt=1hRxWOTov+rv8pm8rXnk=Y7r-N9oJDa!$@%dotqj_@
z_``aJFd9&!dKcd^#_oDgmjBwi0JSTaY@$W+b{)Hv>n!~uH>&pAJan6ka<}LC1tKmu
zYDAoND>7C6l1b0jW5c?+2ORr4GP3zEjy0?SDMY*JuChh5`AV-`@3TCC<jz=agHMG<
zay=dT{mCV%BM5Ec--M?>-l?HS2W?_+EymIR`q};K!ZVo4I`?g4R17FL-aepv)oazw
z(=8%JN{~?aTTCu7cumM4lIiFbe6OaN`JIfM+@!}W7ADp%ywn>iCeto|PFE%MWOu<N
zUxqCS55~Zw8pchiQyKuca70;gbnAfX4VO_vVS5}kPAP`(3>k_&o7Dnxm!6dN3u-z3
z8xve+JsGf_=0}2*Fw#*=#ybG-1DpfEhp8)yNXAzKhNG%l$b(NdR0atey-#=EcQoDK
z5B#~Un^kKF>&1tbSzEoL?*t8|V4l}ijRC}LCvde!6%|fqL0&p*l*w=2y!nSu4Z!db
zvuR)@_zH%NG0{?+!2@19Y2ED1_h5@^kH<ELbW?&hy($pEr@k8PqfXK3miIYJpy#Qa
zJ8%Ri)gY~Q(}~32>Ast(E~=~R(1kGL1K2nQ;dustD6U}*<>4`;_f;#^(uUL1sE2x?
z&B=1T2e=`MgJaFHX3x2}U2kZMKNJ-k?Yn^Tc%3#-O!?25&HV-zMHcN{biD&&Smz}g
zPWZ9cxm0EA?b7{_`4fVlU`nzPRAmt&a8I6)kN`DwL-jwH7Wpsf?l+5>x8B-&qj?^;
z(DV^|qp$VBC`n{E;ahueiH<0hv!pdX?o^%c`##`6A(OuM#lAm7_%Fzl|F+i+wisJ2
zEv1R3Ng&2?3!$hH{}9M-xvqugU@Sl&eO1e~uUfU%=6hTGjG=DvC^);){MX035JLp2
zT=~CuS3(#R(l8lp%dWc>I-fVM+<|)soM-3JqC$gP@Lhq#r=^U3Oi@dLW^hh_##w<|
zgyOo9`Ci}W>B^Us#&FC#8K^iOGoid@=f}QuZb0~f#)um3<4)}!H>MU9+~+Rld0KD`
z^)a;ZRtCAnUMBJ1`AhX-)U51Uw>@qJPBs%k;^T}D+SUl=N!4Ff8m>NEo&LohFm}<e
zMJ8ux=^>YvGW^a(%-<}=6%y~CL7ehYZj>x+23wp5FyD2$>Q~#MBltZXA+4;zD$mbc
zANqXnk@RmTQ+}Luqm>bNZ!W-R-?sQ}ytQ`+x41phIGVet+|zPXZ)bHh@$-zs=+f9z
zFSq^X`IE;>7$7SlJNDRZx&BGR-v#ywrR`1h%o@AIwj4j~BgrN29R=*S&u=DopGe=Y
ze8LV7q(;m7H9(Wqo&B(ot=b2pGKvoK!p<#p`>v7<yP`DOaOq=~k$(HT@bs#VAkwWl
zc5dzgKz~HNc7z`B?`He*U#&O`Fl5zYiLzCrkWBbMTdz}mZ>*t7F|Vt^GQb_;fSPbM
zTisaQv(DQ)u2*4nZn9U`p=#j$ymT(W<%B`BRmx<4|LzghJyA9OB%>{RXj}i0ko0c8
z?|QnJUv*n2xm!Fo_;PvTD>r^NK6yws1Le=eL6*YpQWZFCqe;fd#y5zoA^$mh^vV4y
zo|jyvS>;FDq9jEJLbyx>l0{>FFTDo#BHe6?_a!L5WPFpCe6EFkKJ~?XpPjI(*I@Zh
z<azjAJPD81+WoFB2o0A!a1EsY)`OVusnx2qdVq#xzb)gm$?hQE6y@Jq03V3DZxZNy
z!~2NIqErVFX3l@xxZ=n=_p~O4?juv}c8UevjQi|Y_rP@dA(3`Mb-B~|3HMhcjUbAy
zLIimvuEhKpKDYy{con0Hj}l=M+?w%FH>KWFF#Gi%(-f!ux%O`VFzUKKEpSS$=Xq<1
z&#S$SfGszDIR79BpF+{^Hc}?P+l}^wk&av-Az139fJPC7fYwJB*~`_(Ka`wXGrc>B
z1*f`0&EADL>24hW8gD)~$WI)zH@To8y{?W|qwjomabDTj&A=Fe7kSFt@xb}_hcjTD
z{Qd0y$s>H<<pVg+myG7Vm4L}=o6ABbzy8ayy4w{(YGkvXN3TA8A(Jw&z~tT`o<^6!
zN%f8FC2C+%%9Vr?e)=8@H+w`vG5|;i(^qI<QR5S(cjC@Qkk7_j2aKY5uz&b8A-?3F
zno0-M#QHU^;S5O5N}wIV0!@?iV?5PTsMaIlbN_+psw0E4tD7<3teG<PI}D$e5inl4
z&xhIlK4ukDNzxB>kR^k859Bt0*Uj?>7=cX(l`U?^o~zPnMM<LO)h*Amdk!M>?<v5R
zh7<+_ydWHgiV^B&p;T+GT@CPl>R)%`wQg@N*0(-A9Gl;qZMQ-y>`CaI&CH618;iIA
zcj_M@s;(;mjJlpXH4@1u%(bH{sEpAScpB(ms43#8rUz+Pa6q|l*<kqv>_oTIjX$-k
z2N0NSyb5q^(-8P^tu$iwBTo5?M1R3of#ucu<RVGR)(^K3C`U~*Jlir2^zZ252{{00
zRV|<nfbE5LW;W=?ZN$0-EtWUeeFAc6MLM8y4~FkzTo~HDo785!(#sm^<OgPi;LRA`
ztcVUbR**G-JE2&ec)T0@HKF;RvMs`BwiIi{opSmK3m(moZ@+O!p9Dy^!|i=um7i9A
zRyqSZut0Es&*M{C;mP5jjwAHf8kX!fyV01pfXJI0X4Z-y@&fNHZLT#KI+kHT{X`vx
zx3WFrZp&hO%vs0dVOz%KA7$<>cO-UJ)~*nNYuLw=mE#Rc(i>!`M>Bh^0a0-@I^D@n
zRWR9`$J+bp?|gpDMoG$Rzj*(!?-=U(5R$}`hcYmEeE};hbg6R&I*ae_Q(~9DonJ(b
zK6JN~>g*~THhp}OM)sp~(tBX2G6E0k5PVG^R{e#JjI=Z%E?*nC2$%vKE^`GTmUqT_
zm)xKAYrgtE?2dh5HwoJqvFA%ymj6({{Bu20%z*nWG{pg6vz&JbiZ>&BakwQHl5@qa
zrH@6Z;o<C4ZNoFz^$z;-rIQ)I-S!9}ua0q~F)AV-jtdPO8J3>Yg#0XL3)6tm1YK}P
zvvLt@;e5?B5&W?oR{?fmdi`Kl_nX1^*4kEh#tUz7Fe`qO3Al_`0uBPc59)X<vjpee
z7ELPxzPI|&m+J>0HSR~&e}5Ai;ASnbsf&T&-*J8J%zZHY-r>Y=?3sa$O(Ys%HT*t}
z{(SO1#yisV?ogM`znO(<VC@zM8B0|eorD-Nna_ROpWoG$7Zh>x$T4?jc;^cPm7f4;
z1<XI$D4@&sKK3Myu~{B&zdJmrg(n+tB9~NkY0dSrCRXDe<=ceJaLd;!>D2nz6)YVY
z)>BH?{P`e!q|Sd8{kb)u<!^HY>BF9QT`dN92k}5WzZDmiuGwnAs{b%(O6;4VAKJL;
ze7PI*`{DKP^Q^0r@?R}bb9X`f*Y3Q*RyQH!jNCyEvwEV^tB**;GRbxG)sOtiRr4KI
zyvMq9gD!<jjelOiS+9sHz<0>(3Tmjxo+yo~-tCS4{BKRpeIA>=NCl7HG*-V8$GE>E
zGgF|XdB`s+vZ=)V-{_z3xB7IR0*Dl)9LGN*Ih_sA)Ef49L-FR&{M(GR4>}9ZOwE1M
zo#Sm)d$OtL@h?MWH-0?*L-hz0_!Qp<qWj;+Z6t{zCGf^?g=p<M9$f8O_d8NE_ty&y
zPB*B*$qH05?9k)88Y;CmF8spOri*Jg!x^JYtzrl0m7pjlm}Q5}isg&d#<NMvn6+Qb
zK8mRU^W^b??XwbeASY-a&Ho3#vgz_pMt;P|MyjR$OIe5K+ddLscv03(-eWZt6xE<=
zu54OV$-&PZpX;+r#yn+YWO8Jy!WsX7gP?g^*sG@t*h}|I=XuZ7500a8rUXLQNnt|L
zlhaj%?1d$T^<vht_6<Xz*EJK}8kH(7>^m;cF=Zr~^=H~Me<&SC7=tI&-uG^4x!&Ws
zSRvzr_X}t`2}T-^*#vxjD5Un&9e0JAmgpSlA<Gz)sD&ecE5scyGI=Z)XBIC<lRuvP
z{ghM7I;*}<3EDTz@G#!j+iOv1n}NO8ilmfRA-s#={=QF3RsbrQz9LT=ODMz&qeE3d
zE%=2JP)O_i)@7iEpfiF9vMf8xlU?VD1XX)-SkHpW0w@L-AbyN(Y-}tr2^a5o((QM^
zF&QBYnAVx3=zxivxcz3t-l^yLdxOaJcxk3}S$mHug})3Z-3ws*$zFF1j>#VHj>kZ@
z<)x6rgeczW(}L$YZCSYU!;^CT6~FKcblIs-?DA2=w8*sK&*Us{euk1^G^vM4zT1?I
zWc5J!$wlob6)6_L3Mzu%v5G}ueZ+)E!i@dCZ)Zl3Rz%>zWBx+5L%`~IrI_RWfbKzy
z&ro_s-%Qzy;;TyCDQ>lybN(@2>sdxb4B~!(kiUVl@}2lBM*wg;mG?6fnCLgrz#h8&
zxUZ=tucRA5V?bZ*vrdrATxb}HZ2)#ox6S9km>kYkoV-u@%Kj(iD=3Wv*B6(hFASwf
zWe{^t>f2WvLf|CeJ)={1G~IbBelI12eED~UsH`xrU+4=IaEOlI31E+#j<9=kRiwZ%
z2~8bANzUGPm)8Vs9uT9ktPdB53)pnj*0bg7pzJdTa9`VtfSnFHzP}syIe|O(%Qkuu
zo30^`Q{+!0GSGDcL~s`PzWJX{zHR-1rl?<ofBHJ!Co1(53rXioY^<5p+}{@qY=W#5
zw0cx6?`exh0LrR`o=t{JW&tL-AHaL2Km+6o?Z(L}eF4D5Ridi|!W<~R6b8_6Nf?-&
z!mukinA<L}kY2pr0w+qF4Ei^yP>lvRaGkh;&xIc5{Qw&{Zt%?)>1Oi1HB`hd64I5`
zqz0h4M*5^Ypjp;y^Kb;NCM8A1kBxRKv`kEhYr6tZ?lNZEmCq>6{dG8e6NfHFd5?d(
z==zgc>^3VXD0Bj!0N1BIOkntU(>F!4h}Hya%QI?A^srLH2okd&JuVLiF7~H3&GB$>
z#sMDkdR0-|IiQCBCJRter>9wbAH4v9RUnC%GgFQoh(9r2^GUe7-w23WY!8zOlkLf`
zVsF<>?!{6mn8n;Yv5W)l^y}TC^mhM5F~@*eWP)h@>jTf#J0|Qm8DPzA{j#aCXYqlg
zK_E#EAUxD-9oE@_m3%4jSqYh-mKyQJ%_Qklzu&XFAIFcLH<Qob!>|Bu*J#$eFc5nf
zLcziy(=`GxZm`vvu}Tw!fuTb~-tD02#{GEZO5zO(AyEt~8I_aV2N%9J&nqZNDbFE`
zN#le{D+HCh;yVcg)BEHhOJO-xb#=$xOy{L2Zp;{aAQ_#nCZC=mDF8(5A1k&u^O2<|
z0NZDJcX<Tl5m4K@!g8+gvVf=S@eTl|Zi1iprrH43?&=7csRg#irJK_jUm16I?&K9Q
ztN_9O*u;jTSP5lhx=kLhUEZf(FMQodV?I2(wVdKU9Zp9Mn~P|har8JL?z9RsgeF=_
zk(o%iT1v8=rt*yrf^;>2hQ;tb`gWR|y}~5nw>F(-rietBDxP|SP=l$F`|zwmRyr18
zZ9KQ9_36>7ZWcVg5VTPNyp#RaDw=~i?=Pr2#Fn{*A`E^?gwi~~tB4jTMjKNzDW~H+
z!tsS1!D|(yQq>#>>573%4FJpUPvw~%?=`StL`6j-KbZ0XGey^xw_2snK(H*4X7eW$
zRaI55$C`b3DRzq!1H|XA2}y6ObeiI4cf@{sAA74+Y4!{livvXue!{xCv<-vbE?V#l
zftQ6;L%)4eI_E54Wupi%=16~G@zeS#{T!qsf2ItN`i_{_m1-GBl5GK3DffBy0RmMf
zj<@HP>Db?AIoKtgVq~Yfk`((nN%qsa>-5vfVs8^ABu!=uMepkB5OH3)V$6Jk^{cA3
z4JkhkN~c2bTVV9DJ3~<Ublf%h>Nu{N7)%{#ou%6xUns`SHut&YB;4TV8)TKJXl1mC
z8!3ihnW@zI?#p97O`;RB1U}rw__+!K0h`$~@}MB<DHC?s{lI`SxLV5^l0@opHot>g
zV(wWPQFrcd{uDZm2qbo;bvPD<6O)+a*(yk)=?3SE7!`D)%e#KXv2e8*FOGFvTy14G
zK<>1#U~1n@nIy@2sW)pKXaO*an?``$vZTEHXM-)&Q;3R+ssXl-U?u|3@-C8u7aWcO
zo}mfIogUd-(=#x@0<IQf2nI=BLj&&W)k3{Bp7ZYLHWc0c>(?(GZ^(=UohGoF0S{c?
zmAC8Rd>2r@<FgpR)S8Xl;yAto0wb86vFug5+BSgi29%wdzc~~XJrwccQj&nTW`pw8
zbN(en925T2(PENt2*xjy3`~C&dQrAio6odZ@E^<T!rR4RfEjq|Z#89#AeInRr1NVl
zcQy@rfSQ*_0hp`{ZsinMm0+GF;dO<vMdFpB1HNIGl^^Kj@KC%a6d#`&K~8=zDa24J
zZ1*HV_pP*aV=F;kEU2`uWVGQ3lU9hA*m*BDMr!!KFTXFxeSoq7-6Bc|6=y89A<fM}
zp6IYA1S9t4OG>x%5$aYDR`Ol1^#H7A;Acy)fbu=wb?gG5SqnP80zsv%O_A&!UkY?n
z1I{W^7Aq9<Ab>(vxeg3yub5j;;Dj`mUFKGxm}{`74~>rpOo?l1fP>?PO6<M@JRgST
z)GQ~tjKTC~jKC-$NE_GCCs+c!X$(}F3KpDQfb(VJ7$x@x?(gbqe3nYQWa>>wzAX*E
zOIso*kW+!9noR*frn3-t+tyRp*B=Xh-uu!JK+!IMDpj{FJw4qW)RJv-WX-05B?q46
zaywrv^Cz?KjdYQGwjZXNh5(Cd!f6bq+HzIXgJJpJ@a%nQStZa-*yR9Pz2mqxQ=$@p
zLf6`@h=V^GfGy7#=ksv;Yx}XHli8M!4)X}P;_FM+a6&2yuQ4pRFfsZ#b_U9t^<%OP
zD~gtZ&{{12U-oa;Jy$#Gj~0oY?Dr=NVG7XD@OixO$2v|tjWr3PtNJa+E(p{efR%&o
zE|?iEH&b8Yr$q%g_<4)wbAtnPniPa1D_6gtb%Hk(8jVvI>R$g+M*GTXOW3&Vga!z^
zb_@a(!V@+tHK|bV9{^X6=L#d`Ck~luUQKw6@}0}vUK|X8&!{`FuHXpVOqHNU*_0D5
zq|P<%AU}8N^O~Jee^_A+$#B%~0J05N7LN-H@Fp5(5vl4T>v`PB%VfIJJvpxUI)V9&
z&1RlOXLWN&yIoq-{e6)R&np(#qz|5#CVS(#d>8#aP<SIy9I!iX{B-l$h-<8ve|+bm
zR@Z24V`tRGoV~%j+D7mN_8H>+1NT`YW*K5WUH+m^;5%m=L8)<AgH-Oar`5R1{{;w7
zZ@tVU-kOZ&X@|V@lKL{Kff6<gn}x?TPK_Uf9HHBCH9479I4<dYscCK~jSNTn>W54w
zTdJ_ja}Bw865gvC%OTEaVs0_a>kQs~o5Zu1Jc4Ckz$L=50LPG8m7c7s>MN+BL`ym?
zUHhZ7l#~=cfP;g1S&WBM>;IVg3aBc#u3b`68U*QX5RmTf7NiAf>6Gs74(S%9W78oh
z-6GxHsdU`6zrQYHcpT>p1<vN(Yt8v&)~~gnk+}PlMMw@0|JaWnZ%qB3ws1lA^by|*
zE7dC%zJes>EM923A7`evh+bD>-{$DMcdomj&k9;5Z`LC&HwYT^tnJTl9Y<w~@m!DI
zDe0!O-H3;U<VZ1|J-#{3F!kUNhm%S=**2dhy&;?(RWr~rw&f0?$YG*7d|}4*Wryz7
zs>4Qir`z6y3I%dd0t_C7+e0qaPoxJ>$!CFoTDc-NYrniV*?~7ZoS;DGq=ERh$wHn+
zLyq#i16`%ZVk=_(XvsUFg7{~ehxgS{j}ysI+jZ&JugLuz*fwb?#mS5h1elDt0hIU1
zG(_ytwB7Q>XFh>hS|p-VF-i#2<a)$<MFASGSKs3mN2?hn54M#2AV})wB5+h0>bXYH
z(*LZM;Sr8f7Dv6#chjjjrLw`>b`waq&Bqt_d(3^pwsV0>$!@Xd8efl*MFK5_V&P{G
zPz!+8VrfQcb8}h{w%5s<sTAg>-CQ*cW)`euaHOH4p)HcqWP2TygM{jEyG6)37-2d5
z9y}UmKqDXVB9GBU@Qat9UeSKiB>?pbKF{~7yU<=iD9s451G#vxodAJJmZNbrOlPiY
zhO6Wr2T}mrw)ddv96)e`-GT6+z-%1~=YYW5+QLsD_fzP4$%W(mt#LODxCl&p=oc!a
zx0Iw%wLv$6mKGvmN%raeF`lIaGWg!gtE=M&t0|B{k6%Gc3%^Lh&DIa8>2%PZ1Kv0_
zpu3PxC3}tyRmUcF*Avu*6gKubkqC{O{s=-mhaCZ(yF>!64B(n45paz#<CN#gijDkg
z^*|*})SW;nIoDu~3l5!F%=d+IYX(wgsk93h^(|G=DPcdCfT*Td79D<>fcEHKWXd__
zQNcB1qI2%QMwp9%%@0_G%QliYA#rY2C}d<{_<b9dJW$DOK>YZys|t|tCJhY@oRjL9
z#+6)Qa`+?^T&0x{^$?CBD(pK%i+Zxb;$YO{aV0ekRuIa)Id7Tw3xaF)61+g|w>vnt
zf<I3EuJ0$**D$mFpKe(U?|8tUDlnCm)VuAB%6qTt2?z4@DSN|W*Ve<yEfyNep?^z7
z(2nZgzn8X)+Pwhgm(bM2g^uLlWCF+O@3}cSo-Fg#PI%W-BeL{Y;H6C69!}r%Bm+Go
zZ{NO6kShl2$=t$%g23Kz%qC3(>6t&Fp`osO6M-r@0+SPtDf9J~rn*!{<F5z^H1yko
z53909p~ZQ{DI}sRQsUyUdT6$S;@2m_y>}VDg?97_Qy9V6q?=Uh%(jbB!*<rz>)=@c
z!xxCcx3aVQ<&*^QFlr}}%7_Zopa9SU0SA*T9~Bh^CjZjT&Omm*^SiUxq(~xnojrBM
zloS#@LpT|pIB3NZJ*-3Z`25&WJX9y-NtMsPL!^IQLm-fe>SwWs>^nyD9aBZcur@#A
zh_ZAeuR#X1#AYKb?O!PcF%hWri9PbgR@S3nDNalSa`NK0Ep@wNS>wOEyk5{JiZL?m
zPay`-I<QJo-!5OWuX&RB-)!bV>sNHHu2Ahe{%J==M@M?F4nv4sd$7iGUD-@k$ofGQ
z$#{dcrtaU_Uoph~O#Q@J^9`QN@lPXPlMfCKzE)S4ftU2|&I^NFBxdFdJQAk5`sC=A
zyLMReV5<nI1GVGbfh2GJP>ta~``qvQ94OHFu+5$+ph(o;*74N$;TC8c2deXQ<<y;I
zAH=v$$J%awEMJ4i`W2Y3V0t{Q`ifdx({4X6kAiu=Wv?*0`Mi%9h-(s$`yxF)4Y+S!
ztbQKZj--Ee)=Mr7(+w!(sF|PU&%KOAl(Phy)QhUrW^qS~VIkIh^i<z_|M%fu#mFPo
zh)#GD;3!F8_jY8xSqyUPsZS6Y5<A4?emoy-&JV&bCkkp{N(>NC>0^Qp^t6L&a9sE1
zLMs<X#C<`KF|}-*6e3YE9`eIk4G2FbABVQG+1o=4&SHpMa7xS#2|?O+)nOl+B8-VX
z@}xzXx`u`V=uyv9O3M>{qJhpVV6#JDs#C^8fBkMSANPeCI8Y{gU#7V4gBHVzb;n2S
znabr6=~p@8*JRr?N%C~LB5`0*Eo@x-8@cJoHCp@|BR5N&IWU1=w=h&D8m5vnhFmnR
zk-Vw^jDw&{gk(5MiwdYjOqXS|oNQg>s#9sq=?BaVk|G<_U2pAujM+loAdsz5pVx^%
zT&@0r!Qd~`W?$ULhh`226ZNKNiYy-Of=1&h{<36+IgfdC1)7AOzFNDeD0CtX9UYK?
z>M9U9I~_F<1~V~~KvA9&Yl>7%YD&vc5;`HTVV%|2X94+hVs`vRnI>5t4pDbJHj>~?
zvCNuRzHhQ4zuI0xzO!L^%wd+53hCM8UF^|vN3_ehXnN2F70w`n9$&GBG@lW1=Bt1G
z^w<b=pOp0o;9{p1Zq5Zum5F(5BY2Bu4P2D7*%fV+6&0h2#2KMI<6yZ5V|&I80}G37
zS*P=VQgHieG~C>{L2j{u=XNTKfmITPY>0Bhk5f=qnJt)~pC1qikJR_+8&pXQ+Ep_F
z{vK{NwdHXCuqzsW*xLSgO7Uo^g=>tpq;V&bKG%5qt2(qa*y$ehHR11O0iZ~`LME#4
zy3jypvMW^sWb!5pp0?wHj#W@s{0b5M4BQJr&}3!%;9XDXEyhMi_4G8;Ealq{I2GvF
z4@!hd<4vUKL_okaGUeW46FX4oZ0t=ISyd8oe3Z%%BL!2$*KgmVhQarY3zG1uQWXkP
zqFloNqjZ4szf)6F?w%xt?H}iDc{;U`#X;8yXk?Djas!`;@(iEc?}n0E216F{KocSw
zh&pop(e`EXf8Y}Ej|b)r|3Yjy(UH&a0X;v~!PH#(q07$%Bm0aU6%<S}u(NOOO>P3p
z(c_E)yq(R@H-51gwex9v!XO0d+Wb9rsOZ5Y;S)Pu>4>-~JYM_^HCkK(#l;0sgw}$^
z1r*fGcX;zwR#qZYLTL?fF=FNje%?uaZ?vAGsj~j6>jn`Ffn;C_U5jF78@G;J1C?f*
zD?nlaWSXqt_A&>`PM;H(yiEWexb9Dd_Vh@AS+W;QQnzQln3*EJ0>WOdn2Tkb?hwXC
z8bn0vFJHdQb@~c|yT<i!CKe>mx4IuIl&R$Gfn!CmPC)^>hk*S945g7AjCdusv3Hs_
zHwW(>%yvd}+W}b(UUyL70Iq*og~(l^#S4lzY;by6|Gw+uLMpe99FGe9SeZCsBN`hU
zoqWL!q;!3<45n~2Qi=O!1Lw7Wv#-=aPt72GJVw8YY?bHnf&C(<-%C3meEuL&qQ>*&
z!y!36ytqH2q1oXm&ZvJTrl)^451IHUZ)J%3i>yBE@NPb-DTp^4jgyvr70{3Xnfzcx
zDD(cD{C>U;y@1<tj3n3Z&Kf|SeJ9IppsIiv{Hzfou1TQEOz^Um6c^(<w4c8)?T_0A
zP{<}=luY_!h?Md~NEwws7`aY{l8E?Z0Antv-NHNQX8}NEFGykEU+M5}e@_IOWT1_D
z02G7H-2k@wS^V*>jJk-0so_n0KgrWma})7mf)pqapvqLzP)h*vlp3=^oNwR1L)!%a
z?$TSwV&wnm*1dI^1NI4cBn)HlABBJ%@@TnDFjtX@X1$o^$$(G;Qu&|vapeX0A4Kc!
zi_Ibu@!135RA@5oyX(Ph&rC(zM(pnv7)|p^pN{;!ymtNz;gcnM*1J(k)N8TXH^7&%
zv0*MZYgKLx=(~o$@@vn7M>@cY`;`y-h>{XctK4Ha&Ygw4C|}48&_?II3Cffp`{h>H
zf+&ekDcUyz?}X<;abEm3(Dm;0V!n5ok}m>sU34<e2%wd?_<=;eHrMww;<l{QVzS7i
zYwo-yXTmZJZ8EWy?L!H`jr#>ErwIk!(W%2<{<?%?*?(g@;P~`+>N~(~!ry=JO(hfd
zavYpy^lAT$soR7<>jo7%Q_G<Dy9o!0CAi|8TMS(LzJ*hqvqH(@wU5$bmzgbq5%+zh
z)Q}&{5A4`~xA0Ptj**3p?o3&w5+sCAGeAPO8<b-ykR5%=$I+~;Z7I(Tmu3j+=4Ego
zlD!390&WLVO;zSB3vC%;y|VM2zFhUyZ9Fs~Niw5UuC{+Hbd#hwvA^rE^!L>vcG@h!
z^Q4(?z7T<+Yl$E<G?)#TeUTSP8WQ(qU}BS|BZ(3q*!1j6(ol#DmFC$@cJ)cxhf5)_
zNZv;z5%tUN17Fr2tsi67<z@}!v^v+=ixx4-6c~`}%Yot=40l0pwUg7m{%d#~ltmrO
zj}aK<N~A#jpad|hijPq#!eB;!$MlD-?l0)NXDok*I)3>vs0=LKq@G2M7GUplA_ZC)
z$<<EZTDvSFE(=G%lNL4Z6oivSX&Fz331|dJheT`85&Z)I>oCC9oeW>~JAp%gD{J+E
zZq~@#&LF<Dlny$&r)bzNH7AZ{Dd)C=sPi0rW^KX^U1Vq}CCD$v$L(sl@=E-qs4xiH
zzlp#9)+kT8jVaRrEkO7b?@sv<0@c-JMAgaaBCKK#c3auNcWuj~=;Fc^j!C`)4rn<>
zMkXcz??4k$DXWrC@1dnUjW+MWxd}D`%0#&a>Pj0M8_(PAbkjdXgmCPMasaohI@Eag
z4sU!PfDz}QO3G=`uIbltT&HHY$9h{;^yxi-pr7t$yZZ8~rKqqM)L(DwRN;*H7%i}{
zD<fEY-xvJv5c2^cPbmM-@oLk;_H+;nS$|gCWUb^PQ9k5<v;ashC&WhVhzzS+r=B+F
zS<rGL@P5pI$u*JO|NWgFVvZ-E$9VaABG8F@py4gC9+g_H*4l4XRtiHkvFi(a9XPf&
zS0;eZ!zroEfhMAh685ndu-oQ_^L5tLh@qjS1|@2K%_8Wt&k)*7w{4{{zQ?yA+HXlk
z!|>G9S7N)`?@)m{6M89U^PjuAAJ2^mi#A8g`d@K1XdErn20uOCXOWCJEh|^$4U75{
zb?%x7VnBh~IKdNRP`CsdL=wPY3ASyV?~SjCiJ=G@z-P9K)v0SGs`(N%jh0peHzxCh
zB*1k3;rVX%c3;(hDD2-MmL9eTn|yy?Qqa4eIV-+2sX}Z$>PduL;_J^sOA?OrC7PQe
z^KF)xP0Q}Yl40oUV5KJEbEMEdv9i0;0C*mBP{<bQOCU~*QN3AIEvO;2N}y^ulYEM8
z{;H;4GB}^WjIdCQaPvV%V_)uHswU?6C4(Hja+eH$k8Agx8}VG11ADrG4>s1q6+b~5
z?SO?I0~_bJ!X+U~aV)iW8KsRSD1tO9JhpS)VBi=`Ruxh)F`>lYHkT{N6xiPIh?5DK
zfl7GO?K6}?f2OxIFT{Rc%E`%zBp_W;0u^@F$OS)nOavDnp#kzX(UFB|XxL-@x+@w5
z=;vpg4j4DTu0T@nPus$QqKmK}vH>G?kBcyF{=2mAeogqfH$&F$ajQB0r93NTPbZiz
zj1@07$(P5|Mv3{avt8zzov5j0(M^Az6=d^$LP$4oMgee8+@)_ksAfE{)AwBK7u|wV
zV6-`cVLUMfZDF><?3VgFji$pMqSpEZE+@@7PzTN;&eFy5DU2g{xND7n1i9aL`@e{S
z>|0tCqnE4CKYf|_qk%jju`OTu-%s-m?=Q=!=_q55U<gfD3(}o2L1rM<(PC4)El;$0
z+1W$Ie29FJLXsXu4ubs*eYfRpl(`&QLJ+Gy<jsBsR}CxDS#2dAla)RVCgGa^pp@Ob
z7Xf{Ee1ixOJ5b;;vd&3&H^$IIB8gBSSB54>6O;NS>tAh2i#B6D|6kv{&#Q)m&B>|~
za^gK4Em_4O<>@|ghHY18Q+dzbo$;8LL&MQTT=p!xd7rTXwYt&EO<Z2S5=+gRa<SR@
z=FtS>K);Sia@sm?EJuOwvVu!u_4X|*>!>T^vD9oGj*l4HVBB$=7({K3Qre^-R)cL2
zPA#{uZh=RiH4{CJHK8-y84vc2hX-wWNHIl@7?Q5fY$!}s4tMbO@M?j@`<EaJeXRjy
zGm%rCE*Um4pTchXos7wLARZq-O674#a0-uv?i`9dXJ$8qHAGD(O<mk=FQZhltn=?H
zjKH*`xem0&p(Ld&M3*Udm`H65@rfw&tR=&=GDm5#)`1_mK^424W_x0-x$c~y`*@Kl
zl2~T)^915_(RPiAs4WWlR`tI{_FcZxDG=U^E9x;|0;wMPcTeZXN$TN_uQ7lK#h7(7
zb!MOnppt%1l^B69YJ4GmywZ^cel$N(OP8d6tN*lm*Y$P80EhKjS4`>CP8vS8EkMAb
zg_%Ico(#71<k8n~PM`_W07SimhN&<>zSe*^oSm^8Vt*BvRBf=RbCl!i3K)UG`}mzs
zBGNs;Ho&q)hYSa}4tapF@6FXx0}g~E6F-$SQE%Rs0CWKQJ@EC%k(zG&j07jD<lEgX
z8ML4P0B2H!<b(Za`p+L&ndN`rmUjfSye3$^fC@W4g%d*Gz!D9nBWEDP_}%u|0qpaH
zToy`7dfYrbA9V?2<46jco99O0hocr>H41u>mN!!c`K)yZ+Ey<^YmA{tP{>f=ai3Z_
z45hSS*Yp3K*2w<5{--;Rvn2n%$-z*+<-rL3B~7yA)BTk+L?YDFqL_7(fi2S=C4ES7
zvySX>z6hO&BLq4|)Q<}gvuL{K!+ohAg-J<E->`|V)Y8{&`9zEPxJ8+w&4BHAagz*-
z+|w9Ie&e%W%6`7M`jjSy&<EQaWM+cHs??HZ9(q8khG61YpgWs(%>tUXSIkCkIBpiW
zPSFS*dq831gGT9iwn{%6X*r2TB(a&3ZTP9rso+Q!Z#=l#VEcJ&Z0)QyTTyFp2m-Gd
zPj$z>VQO=1ej|d;*I)>RHuM0!D#`5bly$oEB?w@-&g#U9X#Ozb{hEbgUoU9a`>kb`
z(-cPT6><p57v>|K2!berCPWi7gYClNo_^jodt$hUjF9Lde2shw*`&G?^9rM;FU%H(
z;_+m{dn+A+U^_4#+@LKz1(kHeY@2C|_Tp&xyHy%`p;A)Bfon)25c8sj9tmo-{Fmy+
z(w@0J*&{95fgJpFcyOsjdAE633#Q>A#)2BmKF53*{c$`27!neS*f(rVKG%Xm&kvVh
zyDL2H7U4+qgJAkGF6mJMQ}W?3t%-WRpWmprrlsY>#Y1t!l{RL8BZv2dGxx=k1~;%Y
zXq?y0k3yU`-*^6W@Vg`yOH`nRk#=&T+@oli`(1$8Ou3w*AvatIj_?(uD*{uQJh`<t
z9f&ehs<GH}Cyc>E3tA?>seqxPA|YQsc^WOUPx_)mTPk!=!KRi4jXxw`Ep*lG8NM>G
zn;wUMZ*jJ`=RneF{~6P!ZL!qlf5=)|Ey<sXl@BLMBaQ!|W1rb)cX724JltT|&mr!*
zpjBvd-=3Q+QsQ$H61|&zY{yS)v6*F4Xmc%fDZ|KzN1c*yxWLw+npo!#CDJE7KHTub
zc{tcFBlMkOxchhf^6MN%Lxs&NZT~YeU>QUEqxU@W02?XMNNuR~aBEX-IGSU8%h2^d
zC-~=ImPyl5Sc!Tl{T#RAkC5V~oqT?ZRJNa~GA|r-2^eDW=uD~?4}uzMS;9>uWor%|
zuM^XP3C`|C+Xqsj=>W!HJa6KxAy<-Q#>pDc15u7E+1QW>qA;G)wRFA<`CVftI8tKP
zj}Ua&Y6Yt?=b>Ptk8*XostOQyw{Q#?$GtU?*7twAFNKaAd)MUxPeFgY_1*R1kJ-+v
zmroBvhv*_E$cBo{SQ?s>MS=j@s6EKJ?zys26VGP<qd<e=6PqRi2BgU{Z*~CZL?<Wg
zQP|+g#sydVCIod~^1I$(fdR{;5t6{hRJ^Vu46s{NRJj+7DOSTMDgO)_aNcq^CX9M9
ztQMW?z)Ei$p4E-YnG{00Z9k4@c%T-k@@MwCdGmQ45t0g&$k8O3`NWccoAIesB4g|S
z0f8^35&4@A3GEhVlzgG1k@Qa(2NJpWHi|$xpnsiO89H$*K~l{E)#(t$#RL-jAS5AY
z13Zmucc`ng2=)P|_ivjZ00fXwR7A_U``3r@o4$xUaEea@w^_GArAhOXxLGXGqqfzu
z#U*7ZjaX(jBFaYzfS3#9yg>hfo2ZA)p=oFsYqy9_CX@j{J)II<=|mZKuiV|lL9=E6
zfOrCP9+0n`t%-vUw+TH^+1>u;vlx}1Q5J19>9uO{7Kksi7<~=wf+2Q`Kzi^KwMq^f
zf0;UDt)hMm>JR#lHPm>n#v~g!HK1Pw)MB1bYpVVlk8%Da&j`vq8$j2#91Q~l14Z(n
z<K-xCvO<Z7G$9-)O9K%3?H+Y9Ok2aWjyC}TL0^cD*?p$sr<N*3Q&Vc$c(M|Q)y}_%
zv+Q(m)E8q;+07q5pDYnEPlsT&gEmp<2m#zgeJw3eSqEn(`nb>~x`qNcsrvy;6O0uE
zMAo!bZFY;b7c6)75uNnTm4wt1-$6t5B4mHHiyc_p@W+L(g+Ona8RsBaHTGAVR8^j{
zMzhd`j~Zv5zyE#lf7BAGk3=?p-ui3j!qF<-a8ziQxOUeQ+1^>i`;2z3)y{2;Qk9$`
zYg`pc*gzhB;V$P*<%f&eu8lZP7tg~AF6>GFsgJw4%P_7-OGJUjXA=eJWCD2aZ|A0T
zl4i;ImD0xyv8H;oL{@KhcyE?c6bW_skpqg9`x#kS-rh>WQjR*X)3A{Yi_1mBY-X6q
zN3TAwOG9hbjh+vw0Q8*k#sp4<N`73V<J>jC)wYHv?YL%RId%GEN&MPKxYjm&W}=dE
zpJp-uHI)Q8vVlg}>$>o(DU#piq9rKp-YilkZdoftI~R(Fo=;(>vilRY$t}Dzk#KlP
z1{Z@m2iC^(QQPRG93-)8YP}{)-MgFc_n@N<6ZT8RDR%yRizRO^D%MotpV2SpILC~I
zR!TiF9Xb$6@U{1KrZO#a4dr}&g%5e{*V@t8!Bg-C03C58qB3<9q|EwKa~6zBc`{}>
z4Yodb8gd9f$_BstBi}}@K9l^U|4o8oD%=IM9*}MpN)Mp6hcax@1gufBhk7<EK<ir^
z0u-{Lpz;xn4TJ)=Iu#c1k^`<sbJ*6p=&#>K{E!UGhbvM-_sIj@<KTr_SiXe%c?Xl}
zxvUa0%SM%z@d)r6FVBB7z1s~fPE48FQij;&`agM|G;d<8fPQn^#hr@!@jjHwUr6YL
zSWT=8tChgw25RiQuHLrHRCJkGI%J!Z0KJa<?pTg)^F+)oa7V*r_=qnylY@Hfl~vd<
zn)>66_TPHHE8E-c{`TZ|1{X-U08q(5z>SGoCI%)SJmbgCul3XZlq!wPKg)cRrJ%dT
zAmi(fX73yrX756m2-zxVc>;1BGN^9Y189@u?$1xD(W#_E|9-vQ@|8HUT#&-pCmB8T
zsrUd@N<pDN{S7X~Pup~-u^7{myzK87%;Pa)wP5s^vA%R;%F$^$B^eeg2SBUh(Mh>+
zfVu%x@^~w=`upsgKW!e^NMyZnSzDf(!XbA=qpY!OBSlG2p*PKQRkW&`M;|o9(GN;`
zT_Tj^j#C21G{mISWI_55=@9lS310>peO`E<q(bb4FP!<b5t9{n_=yTEeJq)bmK*D@
z9L*H16yxZC+D%!tg~kxkt_LOwE0NEG(&s(r<Lu9fv~M*_4dhn+gq>0%r>9JQ=Gy7g
zwb*h7X1$ziu_3R1G{>HDcl4DUA{=*?w4iCg*CTih*XMjRC!(T&ma+k?x@#NJV?}j!
zuZG5r1;mp7>51#pBs1P!9mi{s>3?t9L==If#<AkE4omzZ`xhj|WWw`(eYK&i)()wJ
zR$@+vGZ1ghLSHQ)SxMvCNf=P>I0|x-QP!fJA3hO}Un+~DMK;)mq*;k(zt^BHF8s8~
z(_V^RmLq_%Z8<3*ra)YdSZ;>1?)zn8NQ)wJpiqiXLvlZ%0U4br%f|ZEvgsm(lE+Rk
z29HKtyQPYl8AA0(i;aV-BxZQH5ie$F+@c;3R+NxA_>VSid#}eYH<na16acu_3)SSM
ze}*Cn+ZOfneD8JNy#f<37;Ipn{Qe@u&h%{U2awn$>pOI$0$E2O;1M=F$)K{4OCbP4
z2X6BHJ@^y-y{eUou9R9aN)83Td>*ijK+IC1*N6(_D_#EnLNC-sWmYlI`6fztruBdT
zOoC1|cLbCnFEUOGjc<ZX3@9skwAv*amocK{^78dth7n84Aw|06^Gho4Zi6JIFP~ea
zjO-r^J6{F`fVbmsRZ&vCRE1Kd%pi>@t}(vg<3_<VA*dLy&Tf$hyhjK|g<QH4BfdCa
z@;ClRS_S1LG4mSihXZM2q_gpveSPqv>Y|)KeYWk7h#q-u_WT4d<navu_h=@Vi#{8B
zmAsN(QP4fI5>n)*3{Y|NoVG7EZdp^Sfb15-HsZa8@B`)eYCh;71?{_Wf5^=Tx_^x1
zdD8>P67Wu6K;*zrg?I>e1&WxBK6xPPXzMqxOCGP(>tfaG73j7C{X#&!k(h#FYlm&-
zlf)UA_pf<(tj+USwR);V5=GAjH5(G_SAvH!zY_;T0skYUwk)YG30zn4k8Rhrdw+Pb
zBqk;%0xBz;S3}-(v6@`bCXTu`xN9q%A1GEsjjAS;<qOTJ^0?y`zguIKTF!7@A7Eq&
zm`)Z|2zD@gQQ{R>&;WL@^V2RLR;+w+jfMjFHdKgtVTq8#Fi3|^77W{OUzzGJPfUmf
zCeu-7yBB9oZ)hs8mY}1&F1J%kvC_9MoZC2d809s2=cgbS<c@nmj%6U=qht_wo2o>2
z+gv3wmPi|7d=Z*wx^$y?F2ug#z)-TMr>&BouV_sKHZwc|H?cz=q;$~3+?tX`S^Sa^
z?meT@*|!-T_n_023paZZTry{mvo?257Zy+W%TTmA9EdxhTo*6`VHG|a#P0-J$Ei=j
z+OB9wOJC=@vNA(8I>oZ%D|G3kJM#+SiZ*FIg@Djv`#h4_pZCEtHZDgY;7Z$Phs)%b
zcDU}W<pU$G+Ct3j6=@rc$73wSNtNce+%R7fMR#2##J3!)AZTcdUBC7tViu3X_Rgta
zynHg96EHNm%*^kefo9wqns=Xotk85G#q^Y(g;ST^Js=P=UT#jm5r{X1;Ntg4yt-i-
z|G2dDU0hC(Zc#8j#rYuQH|Tu)U<QZ?fB()97rHNbaPmUIjp(%B@-m7Rws7N*xi3Hd
zFSyb4(=u+F6X^#P9cGHgp5uC6jgnhO_GseieL~HK<nCU8P^MBQi9%<58bM%KP|#cC
z!A#4XUqIQw??)0x=1{B|jv@0HqHD~{&26F!(0~Ynd~8Nz9sw#yXteWgK0d^J-GYlm
zc;;s``vc1A`;ZM9w)Qoupl<1b34OyLQ{86!u#nAwu<oIWxp%UrF2Xi5RmL=u<zM$+
zNKAbHOyqm#);kJ$<EVydFiTSU`EU|txNdK`Iy-=%Sf%#!IGGT;;&y;v0UQ&`oX*s`
z+Oj$VBiy(+MU7ODQoXt~O68=+dPt&K?k8y-9A4y!41<pfHzp}N*moZII*>u%&@{Pz
z8RSTfpqRs{C65$P5aoU%Bz7tioKVLKkWvqZk~6m~CM8sBXG}z<n2Kbp_52hmlvukm
zH9g#$BvAWZ^bXEi5J3ROZV1CCfhussX$?cj^!J{Qd$Z@67&D2YhJJmxW?yV?Z7gl!
zG>*QW85bGjf_K0!4t^LyWD3_$J5Y?A&x(8*d|U8Gs;W+o@fV={K%0EzFd3&=1m#xG
zEo;cy`>s2DD5L13yroQ*_y)TGw7E*jNi*ds(c$7YIz39T=+=b*tfD_0uW8lKyUo2+
zqrCYk(U~!MP|eWrWd_3_V6`}%Hl*m3zkn$i$^!z?5<s)T@yt*H?KFniUFJRC<~ezu
zc3hK*V>tp9=g%#ts^v{!9@-QcK2F~I@-=q@SMc0Gf1nGH<bcc-x$=u&N!M@&womF3
zGcXT8c^M#WX|Vncl@-B23z|DmFK;&E`}3X+76&MhW9r^4-AIR%?Yp*;?vvRW8P5I!
ze2m|%IpE;{eDrtkPcY!mLa;z7^dvaaA-3~D2|)liRM*udf(+W<$2_<|{{G)Y%ke-e
zf@Y=9yvyc`uv6m?O2!|aT3%&{96E;(xU!#)HNH3MjqH`loa;CvW2C8mzGbiQ@-9L7
z<c#0LgY+9M@9E`xBaz=J#4>sFQ*o!SBm%0fZTZIDTli-JW@)m;#q!0!?+B{D%e);#
zR$9{c)4vEvnak05*hR%g&|qbd6nsysHv1cqJmQABv%(XOHl8yA0$3bt0J8pFPFLQ~
z<N9%l+fqcB$t!?-hVdX?jM%0tzp_L2fOxLm4|lcs{17UO0Vk7)kHbsLJU;>;1_7nj
zch{$?5JpMP9M0ouJ99c)NsD!OoZ>X_tpiA9j}jR?3?5$qGvsm|#_Gy;AeBC)LFT(x
zmY?5Vk&Sff53)dCz$wcA&I@qw<*ufUH7uH+nRKm4tR#k<#S(XoB~r_N{r=q)dR_}{
zyXxudg8{wxeYH{ZsICU$esgZ(0*18D06CHyVm|Bq4l;w%t8JzZs)3sQo!3N3de7d3
z`aiRsOUy1p)6bSpjkboE%Zj)GD@F6<J$8|cIcpDN+ZQVuJC+l@ST#veytS(;^lO(r
zFERl~G(VPM$^?td@yG+MzUmGRfc?dQK!tgS@mj$pTJf69cTT$_c)09CWWp2{e-A^s
z^Dr`84NKru&BtF%ec3*!3R#EtZN6-;jNXPet#~n$z#vi5zRnVpKk3{Je^@)Y-`6(6
zz6w`o`Lv*cgkX{tPikYaS;}O8cfh;J@!>gf@Pp4m2-iZax8iqMadFNkjEO|UJb>c*
zM?g6ous#BXq)NX^$-_;cED*vJmXxG69`bNno)|j$UxD#sCMLciiNVf9(E2yIfW~5I
z-fuNeONd4gu!poQdHm-6y0^lWD^%)h7qb7p>xvb;(|*!wvQcL~+-JRVLC3|ZKBmWZ
zs%pCWoPqLfbeH45<=7WD*FP=8rg50Mr-sy!1NnbKq)<hzXj9q5ba8m=s&G-6t#J{^
z;u7Y1#KGFbl5&NeR{DqQfRfQ}lZ(~10ksmr08{akPu+m4n9N(#d6SjzwlHZD*5uZK
zQ)DRpH+JFnmvZ_zR_q1=x^lJ+Y5Dko6E+#%`7~3xNVp?;v1jm+U{j(MSttoII7#hv
z<`X&%2S?5*BxKn0A)Vz#0pqd8kF4kw^ND-~6*579eJJx}Ni!4A{==qHg0SF8=<^Lp
zQ1j1ZZ~vKn9_oJb-CJ$qg`^;tt1lqp>6u}<r;!PdQnt=~33A0@r2XfH#IH6Rf~uw3
z%ltx3TU&uHnG*S}(TpYFsHaKiz}ZjT15q22%&%eH6t9Apra<GHS{gwj_Y?b9b&Md6
zx_F^ZS(^Zo!0KWSffn0GwE!8}Ol(FtPA$l`6sx2g2X^WL3&Eq)CIS(!N+d{$=Xco$
zf3JTh#K{EvDT9-L)chmjuSqDgMyKk%F<aeOsSjH`a-(-GzE&<M*l405umX7c3>hzQ
z2BAknkU(ReGSF<Ne1;5SX#6f%f(lDcanqAk2lCcMLNUcMZ6|HQ!pKHEVU$8E00S;5
z?N@c4UI6-T!zSOf?Q&42y4_GN>7T`$b>YvKAo*+-3sR-Xnb<9poB+3q*?k@3TX;$d
zN4{9=<jZdKTt>RMWr`|KQ1BL-uMXmcWVG;3qL`4hztq_huU^&^OEEbj>~-_zDX&sn
z%R?N#bA~eiO8611Tl<$)o7F!H?7YigN#Mx?TV^$3Q5doG0#7MZHyj@+jr_OA=OCju
z97VBiGR0zW#aMiBDW~f?1j62~WnjJr#|=&-0$>fIP=v-WGx(gSI!}KuEDQlx%N!uI
zB&<nb)9mwJDVM|Oqh1Q(zuAw8Y_GRv<oUu;BvY6ICC?(d`=<(0W!ADp-0Ph()80H5
zNS36Lek*U-O>#u-Vdn|d4hvBd?8|jrKESM*w5m?s*JS4iCrOK)fVIr!NVL=Ep$fFq
zZ_+R{zw^AQNi^ePrP099L%0bXTEzTsQoX;6(4d=ut|^9+)D{OuPu%!u3(ow4lE_L3
zp%UCa%Dhg?w_JSJF1wTEnzA7Qd7}T&Iv<Z6>epsfxqNOMfh_^Z#L55>YU%$hs3K_d
z(TC4(O^npv>li<c&G2GBYP*WRIKNmQ1fcw@<}?YaHQ@>WyEC2zu`{2jq1<6<9ks?F
z-Fc3z@xESpWBCcm$mf=Jtm6#MB|dDOKGt~eE4FQArHn2djTu}#?QlAW-NIXy*Svis
zrTvGOXZ&#y&!ocNO|4Oy#<O}snvA5A69>@on_p}V0@+~U)pZo4F2wf3i*Ehlw8>gi
z*&vF1hi#aeQEtN_McvyU-r3&kh_<Ws(f4FG*uZiYeMpy#A?N58@EtzE#Glrzt@V?M
z->Y=jt*+DTVO`4{X`ucxY;|1^LSr+YEaYAILqDXCk07AE_w)FM*hUIl3gTRPc*Kyd
zT^trk#2!Jij7lOsg+v@V;pd^Ob2=eYJ4(mq^8t>5<=xjpj%_3f$4%w>Odyp=X3v%a
zgavlNkOBqYiU(c*xZTr5AI|v`L{x=tHKD<JGI4~~7sc|?t&(o-54*3R2H^aJw0&e{
z{rs${D7KPcK8Ia?19Qp!I3f3}fUL2LSbs@8`^H1gUo(uxZ`F23DGr#fL`#9{X)H(Z
zaw;!(cj<CWI>d8CChqk}&3fvzJ?6E8yRozNA^Pxp77K9S#|StfOMK;P%0y<BGVRB-
zURWu2fpyvSebzL^sIi(XoKu-Xj~|RB2=Q*?>#xxUa^>~RRhInErz^+^AT&X^c-vGr
zN~}?1rV=;Z_i2NV?;NO^&6Z_C{!W-pnvD|4RcL6QA6nT9|2>n;cDq6t<J-h@9Uzl6
zB^0-oR#y+QsmcL@z>(NXR@vFD9j|1GLdos&2tr-Ee%tQ^eR^E8Nm^@RSHT#tw^si@
z5*;=@<<lc-)TG<NSb`PhxE`Y3j}Rp_C8Hq!J&W=TQBM(phiRggRThgdvRxaefVKEI
zzcCt_Mnh*!pF+BV0%zf)>?jX{e9A6PGpd~;C{GbDMf}sqU18W3)*5`T4og$<Hjw?y
zj=YjaPhZYD*~ze&w%p-O^xs!TcU-3ZF{r+c<`w95A(AsTziN|@mg~tERVM2#KM%ED
zf%584GZjYt%}$$V2Fg8*`HkGscC_jXH&bo%FtRji3KYeoK>jDCFnCkDnb?E%I`A-w
zjX|OS+2=R=uTWY%f|LX~LYuB2v!K!<B0YWL2&$5#NC{L_fdnOj4Rc=xEFiTWSTXd*
zp38^3^AY+VEx>3;TElsrE>2_(dFtr>X_dw7im>a+bn0EJ#W$W=wS`fsA#U!*45vRY
zNy;6Hoz*_DWJ)PWuUYG531O*JNZ}|@*ky&h>@16YG2ArKDc!?JmB~<=-;IS5XP2zP
z*TCat7A$E+DfbFbeXfAoOt01_JtU7tS4Yalg}O3Vwmyq8RUxxdYlix;^`>z2i4)lc
zCgYagnQG>(z&mMq$}=BKO1vouvfgOplzgI_p%CrB-Rpk&!cb3ER@}$=gF6}=JA#C`
z%{3%gg}G)}W7@MunK}6-vCDMn=s^!uMmw*oeN`BbK&nKReA>f{w$;a6pq~EuFZKw-
zaF!P24)VMbw>r04(5*M+y*%xV|8GxW=+EzB3egla7BoRkD|%7nGQZ@9WfYNENrg)0
z+q&3BRxK`^%!NYj=+9FksjeD-y5k7l-6Y@WIX}96_^D9rB+L%cMi}rSvj-z#)D!#*
z;Mx<DmyaAAH`e`2`X&N!Mz93%FJ2ZGT9N}%8pDOqlr-rNg&Na-j1au@dQ(9HG)^Nq
zsd<^m$W)jDkDd9|Ng!heIE5dKgbq5f0D2|2ev`elHEDdLxvtf|*k8lo&|Sv5&b|w`
zif>$|`?xUUPbY{%=5iGxk6xN{wFn_Jo3GoFFQQJ+Nrb=>0x<!p*&pgL0w{ZBP0BRP
z;9m7ZVtiFsKK-v#tt27j=K0i}fiKD=yVLYypxqp!xl&A1DT;ioBYK|KU>sQeEe+T-
zFhqU-(l<0fYTS?Yd{!;*Vup!;Q^|cVpnxNMq`s;%DV)snEYF2E?X`*YJ+J9_>6^pB
zg9W>#cb?RKauP5I>+5b<FJ8-2GBO^I02Ti2MS9oBVf0f^`>l)QuSmx0K}6LI5qWwg
z($xOQB3TZ%g_N7u%HG5Wd;EHP&7}RW<DG^=&ViGwdu&;d&+9M4pT@=J?U~bB|7Sm7
zzK{XRZ$K_QfWR$^M7ey|j3WNBLCaj<7f{*|M5F<#nb$9^8t$aynd}MVIZgn1@fGP!
z2G_Vq@T(6#i9o28(P*_6Jk2M+xI|wvxJ^7`7#b2I;j*=dO_%E_@KP4D-sKttP$P3K
z-blQw&QJkRDbPL7%xFVVmUzFvMIOf_>;A7rO}3ww6Oim^GSyA3(M?`!NZwcpDX#bb
z{Ud7R6V;^|?phacdwH^>_@O-!kjx)xW4x3Y6EcL&#!q~2CLb~T#!yk{1I0cn%G=na
z>2g)CpXUijPK416t$*IU(ve7VMRnwYWx|?EV_Yo8Ot5;%z3BdC?Rzj34)nz}vwSkt
zn~DWd6LC;gXjf-#NZV=aIA3zexaWhVYTx~MNX`rM^|mj`GC3pL>4w;?!2!4@*Fj>Y
zy6#F+GUEbXTMU*e**_=4OpjJ>Ie^Iw`WsBs(!*mH|A~4Ql@8_YbWtI%1hKv(FRog~
z35;GLy!>hIBO%9;6vhc}Iri)LM#mNYT`r$J2EdzRE%f2I>sMKg+}v~yet5dsRGk2S
zTa&%ztq=Q1FoA%6p9{s&Z}wG3T;RVbY_7ot$}LuKaZ4-InD>0GNI37{;P~RrfVY6A
zFv)XCb;qpn3FWaV#o&L}(=K=su3~KjR%P_~pqqfzxhHJv-?bq$#2&8fz)&HSRF0?5
z>tH&Uu+DH?<Q5Q-G<t{H&!gcr8{QqdU?(U)XWf8db^AQ$eQ7oP?QZ3oE30v4Wj9;k
zk0z<_DuM9V2#T)^NfgQj0W^I}S%+L-gie)}E|fmL@CrYVyAP`#jNVzkHZrP;LYNNm
z2)C-~Err1sM=1Cf5XAJQ@NOr&{#wt5=QN&wcELMO0FmZ=JMB$ZjPq&7m7dF}J^5_*
z>gx4<)&WPYEs?pyivShNp!-vAteB4<eJGPR(P~%J_F~F2lc3=fAg5E~5A6ygXL)U3
zQAptxSv)N>YQ;akSI(5?(r2FjUY8b7L7S2P`XS$D!=Jz3IOm8^RntCmxc#EW6b`Ch
zZ?8A%joJ?9YMecrYt6enc32u6N@d<xoyGG^_Zy#xL`b?IYOK|vo@;Z!R2AtAq{h(y
zwGoO|Qz{`JI*q7hp1^53zk!eEMJ;GFH`;#QvE`{kKR&<fbu_OV^hbx7=kfYzLAJEc
zxJz62+fa?oy^+8?A@J|hC-b{$6{VO}WE8KS7QO46VO{h%I9K+d9WeXBZ<*5*j43pd
zOB%R_t4GytJPXO;`_Aai-NP7DJ92cq8CzO)_wJFb5X0pMQG5y$O+;!`Dl>-VdC=}$
zlTFxuD0k^mILC;~RVG>6=fCAKl6*gM%dG{xuT44%s%%QHPD}uimNMh`*F-ka4OKyt
zx`X=ijR--Xtp^4VGMc3*PS5_wh);5KN2Z;@#*8;VTUk9=Gnu@<s;G!AOg|2e`V-*L
z!J`qd<S!mWJ!*9(y)P6j6BL|B&ih@UZvQhvV%vM6TsKvbxY_mlFZpy+%RxxwUJ>lR
z`-;@^{;}nQ7cmR`QdODd!wJNcJK^wwSV>0je)yiQG%j5ylV6sB@F+r4M;;wb1<J4A
zqS|Ktx@tFflXyQ2L1Va~VYFCh{!jm&N7ZFXLsN}&#0iK}Ir#9YP-ca1`P>Y{1uUaV
znA5`{7EEl_Mc<faEBGGN0GQL$dr0=urQvtM;xQ)=PajAYkzN_B9XsFKJlxMjHSL)@
z0YaI-$!RPfYN`HucB!t|R_H{t^*EGqhUo+4fgcSkx%WrOJzf+e#gv(cquRW10~gOY
zj<6wpsjAu)P0#3f|MZx%uLha|y*j0D#wcBsw^=-!)<9x|<#KX7b;DMKY23~ou|xiq
zi4j^3q9r*a|Km};r|Y9YF`iK8xjf9%V$O%nx~2~NOzTe$J<YTw+7i%@9%i&yHv?1r
zIg)sV*GR~+-<2mc5~%kJ2OP8AcQV_rg}Z*Z%~q{*<=vu#d_s8RcEP^G51+-Gr5hX_
zrq2PtvDtCz%l^DB`JBt8o%dAtjh}-@rHV@l{YxFsmgq82gJX*R<-#y*K^5LgU$SlH
zec?fscHZ^K{>%)FMzH-P06s9?q~;G#V>80JK>l?vhXDTFcKHGhnlHIumS+Jb^M@IS
zN9_g{%xfYx_-_6FR}O7Ck{F<90F9u=_y?SJrOx2}>a{4aY}H{c(OU|?lg`>-cdn@Q
zS-)(8dND8hqS<6*mIu3@$whyL`=4I(o5>GIyKh%NFp1utKQ6UhkwF9C$R8)koqj9x
z*18R&c^pcRd#no3zJ+xLXrnx4)5--RSg3|TsJ>2}6-V?dwvEpsXW~G61J2kRzpKSq
z>8B%-dg@(t?&@_IJb0WT%$%Z4A1=OZ$GNMh^!u}UEaTV|d&gt2#{6=4=yPyfcAGt)
z^?#fN5q5<s#!nAxw$2-o^hxGLQaO#enVCc@w_~3#3NU>mz@yaiXs<)d7sE0vF|nQy
zRms8(v0oknOgdE898T&^D1k0VO!<04e$8M2{_4H|zxU8MAQIodldb~2TkmQxRsr6(
z$5kmAXKAVfjL%uTqP;+NI}H5=oYfwy_fMou3M{EOc0?~b&q9A>up6~!NJkUEp8de&
z*<4vx72Y%0JmFAdu*mFbaWv9RJ0{7<%oIod0U~&p^jZXW=OA`SfU76*%<UJGi=O!;
z)$+A*YbQ%|3>sO7_^O=^V#1K8|Bjqz_U$HR$rf-P@%f*>^8W!9z)#AOk0QP@=+=5~
z$gjo_hozz1(@0tP>1wrS9*c2^of7)7{wFFpZ2E#@?FISjeIQQXOc>Ry)eF`4IO9?H
z!#bU|jN>OueE_u2%-o+O9d;U{4&4ERD8{(Ry^vyh#TrP4lo^b%Y-;bu^gDyk=5iI_
z2PAdRmnbS6RU3atOmkh$S?=AuYp<a?DNB>}>ykzL-P_7VzN@*SmaB{J=Y-BBMG{Ku
z4JZdOa-GvN;Xc})9&eu4g|9)p`SoRvgI~0!Dj($*k28iQQAoN~>iJFBONN)glKwN}
z3lN#+y2U@m^v^tk*#FE7Vumg0I=TdEH^X{|5Ao3Yb)oyE-+UjOcZeK*-J(*m%;?s!
zlrg<DVu%yI(m8or>41udE7=j?x@Y28G?9Z04J0W}=B|WvMdT!y4Rl!C1^Maoe4ORp
zRN}9F{f1iUa+Z|$cvMEH@3h1r>pYXWTevL*5$Z~{Tg-#ce}7^MY$Qs@+-oMcQ?yn@
zeOK+5<KmGVxhr*nyLpRp)}#tZHj=xf3W$q+sm$7I9hQe_cHnK*@sF?kTa-aKiAL(z
z^RZgfMU`SNJ{cLk6d3(~3CLq2_qIOAiF}+T`wR?arHpiRveQ9DghzZK4+`rU+9s6i
zByumIVOa5xN7m42Yt#6r>~BX+o*4(NJH@!FFjg+whCY+TI@fbWCQ3@9CF$)PAziR~
z@+nDB`$_M7!t8!4_iUlo0PB1G-c{dQ#Whm;JXah*r&5Zk<^Kw`SI4vSE4P4MrIC8k
zd<<EHx-Ye~YX60@LJJdwev1vL?pC1C;nu3WYCma&X+N$^KaYRDiT60g^0-hkjrk!+
zyc3Q|Krh;UY6QgcG4)|KRV^=Qi)g9fwMaswx9*>o9nKTueL@Nuuslkyj!TZN=Nes*
zG{g*3#UfU`Pv$LQ77Vntbb;<>^(kxh=a(<`W%;Y8O~;QzqDzx9ES9=qNx`)b-AH~}
z&|DRO`kN?|@)-5?!<jHk^UVroH_C~(B~qQ~XllNvg;On+lyF0MIIjG#26XqI`k$23
zjo_-NW)Vx87>FVU{pL4sHsf{x_3-jjC82hk!zW(2XcvMKVWX0SpCSMGcu#`BYPyZ0
zn#F{J{&Jn-ac`e2yp~&6X#Y69E2k)qFX@B#wKBA;;$fls@wIMl)9gUAiLFg9YtvE>
zG#mwb3xd?`hxAGhk1b2g%f{-J!J8XCz|Bf341%CP5c$}?(|*~qi)7>3EB(r(NV<MX
zKnRSXEm+|!x<k<O`Tn$vJbO27+&32*f&1-SMsBXXT}icHzm|?oCa~LJ3g2iFuo(>W
zgrY&S0QJF}6A%m^!==pql40bK36#cmsr~<o)&kWWsDuEP3PJPgaY@Xll;yV1^_Jsg
z&=U~UU_eiua1v)^XgFX7FTZ@20rG@`Y3w@r(-X!HyAI6CWuC~bywP(jTJy$5Q`5{R
z2+zYl`7=s~5oY$Ts>nq@G&~m?2?gCG;IkloR})-c>n~aKY7#{6f<?_hJ`OOUB|~}B
ze~Ny0-N8^NGl)rn(&X-^E6{F79l&q;KkfR*O|-r|$o9qjD6;osUHcR#MFE|K^^fCn
zh?G~P*QAmIQqP-$Kks@AQ`YPpyt%^3TC=%rsy0A60kl1W*E73Fm!L;KtI@J+br+iH
zU!V8L0u7Jj2l<m$1DK{tS(45FzEdY~6Y~fSoj#oy{`gZei)N=81>`pz4xrpv<lYrT
z`DZ~BG^}3E%y1xadYtNU+I}qTP4Ds{25v+|68}qH|6ZVQ54_GTFK2}2eTY@hZrVAt
zp`QwapXrkSvoAEm7kaOHoPPvnCwUM?KLGIy#6z|D7K*le9`1kRO>JtF_8SI7NyKCF
zo#cUF$&Xzx<vyK?4<!WOOv?I)X$008G@W3F)KnPW8;U%vW5T~ogd{9FJj3PPO_w$A
zmu4+IIUF#SfM$><p03-huCZ@g-@`iV;fbXwWg>oA7&L?tJNVE8n{t~>#9V3hV*572
zu-4>_5ytQdm4Tv7=w%2N1-H>O<HUbo4q8#c;nPteKWnaA6DZRY-MD+q&2+!olM0~@
zG(a_QCd0g)agAx!{)dO{$9CxjT#@J8%ov7)kC-CeGslabN0DwRj{*yo{H~?V2Mf%D
zsPbA`Z%>>tg&(+>^f`E{d&|G8%E(2h*Vrwh9nG~RM-Do%dEamE#r5gOJ8?^|pkS62
z);ib7r6!}u=9XUto<7WG4+mpDMA#>q+J;v7`mCWrY*dA6v2%LANxoOV$4SL!(Qiqi
zjuXuG>yov;n;u*H_B#Am`I(yT?U$i+q3a$lkA1$2fnMDPAK1o2>UP@SUxxmQGlyIt
z?`NKzN)DhGqk2}Q6xH!aX&s1TFw!G1#9tEZvf@Mz;Xiu4C@`oYK7SG4tEieLpX;;1
z$RqUR!V<<c&JSw)Fn?smq<&Ak=z-)5>H6L=LPHOQ9G`n19uoW25#LoInDG6L(3->f
z+fJ?HJ>JwyqvgbSDx{*99Yl5dM)Rb|{Y9~i3^Bf;87oSO`C08Xk4`?J!&l{4T=7ZT
z7!=I;sUoc~R%WI{!g0=Te&nNb5g$A&r0!L)$hP1)%$#^e%<<vz>RC2NG*!0;-46FG
zh(7zAM}t)I7npeGCS($i4H`bzS*yk0-6B%z)7rYuU#>i!d=6iFuc0wIat5vq4sARV
z_c@a-3FF};0~@~@cUB_fj@_H0xQz+IoWtzsw8+UhN|{}rYx^Wh`1ae4=v&}(iq(i(
zS?R!2Ok0cj!7p2i2X&u>ro3}X4iaEdVB<R#I=MS|55I29Dj<{7TK5xHx?#iHVpAr&
z(KTfwMpH#x`cRS)LCY|iH;dG5bVb-ZV|Ks6u8Bv7!s~ROK>A*nQCc^Rp(LTUQkoz|
z#~b2cxzg^L(OOCz&%Zm@W{Nb#UyMcKd$TTdyOj(AChIL;D!yV_%X$A{lwRZ<rQ#YI
zYR*ui)8*%zfi~m3l4;4w%^iA`+e$h7arq#VsJO9N;6eR1T<~P=T$a2f0jFYnpcljR
z4`uk8UF>z3L0B>R3}Ig#oa(6ZY*)0teM=}kuMkG}oh;cp<lDwKo7KSj+;*(q-*5gi
zNl~uj!%)8{FARHj=v1HLBk_{^g*D9`M3#m_I1B?zpX-D3q3e+RcCqVG-|twsYiw?g
zFr3Vb0kjPP6Jfe#uICHN|5#H#bTMzGIiI)eZLbNRKE71pTSoxDLa^Qt@OEXrk3dao
zcE4OfvN^dqnirb2uH--d`=>%=(-kY|v(Yn~ilOgosHDx=xvtUOCf7VXxbBN{DAuvd
zv*V*f+oQY4<Gw01I1<QIp!7XR=>+uu<rC%ukGY}Kz=e6X^Fld@qyt@Q{}m(KgrHb7
z7bF((Rf)@g9WL8t$>(ykef^~a(3`Qx?S{Q7$Cp6v`-ecb7Z-oN-SsTU*0P0TbbW^w
z<f7+eVSCr(Jr2CE($F~bHzFb{OhX|2XS`54`lkETCn3`>J~g~*BjcoYBu_LCoQ!IC
z(gVu0KLg&F`In?c6KufOeVW;d>VLhh{hL6XW?4eHfLl5Hki%6j^H;zd8&U5_Eam*-
zkN*z1Zu)pTe?f0p@0ClVol!J}N+5CqKt7d`$2ua|AVCLeJBYN~scP%w=e#m`LLbmO
zHI!4w(tH7ON@79c_c@5i^C}j#xK@^|VBl7}<u*rjrZWYfS*PO;z5N>cr?NeF=D|Wy
z2f`~_mjm1Sk5;>nOh<iQ{#%poe+VXjFa$-%Nm}1AHs?&+T;jdIN02-aNWGKkJoP>8
zzn|Cvn#Pd*eX{okPKcVCKMHmVqgRmSR-T_Q9S10f`8Oh{&x{&8qa{6ac{|Ui68+vQ
zP;MLB%_r!LI+PHlpS7V^_-K}?xf9Gf_7X(7u-aM-Hp)F{{b$XT=z&3cS^g*gTqE50
zS4j$-O)#sZPfcP#C37|xiQ8Z7?B_2oEg*(mr^^XmR@jLKjI^9Hp4z0IeD4iEzx>et
zs7SbZu3+45zOp4$)pEtu*C+4}B4PF8`osuLtc$I7s&*&G=4*LPDXKeMXb;M`l9t3@
znfI^HaHv@bqAyI=HC>$#S|uo)#!T&S*V8ri!<}d)6hgF9O6kVYz71Uzkw1q_X=S_5
zd}?=mS1PrOnBH;B3DkLg$*Oi6ld`i#oUz!D*w%NMKfQXv3dnzQW6E#$4~4^vQR+!c
zEA~}x``#OFuHF-2VYcQQH(O6X!JHGm<mAiq8Izdx{a9j8lm0PxENzn37<}fmf|UGk
zaIkMX9HO+stBi+dc6HqcKB^(OW#b+ljykTY`YEw8#R;BUrLpNNn0v{=E`ILl9_mel
zKkHLmh3OtF%NdZw6zz%kb7tRd-!Y4Eny!jZjyfOoM5YywNLU@x35iET$T@$Q*=<?t
znLR(cTvQXVw6!HcUiChNY7)mqno@<p*7Sg-P3F#kp;$dzw3=7?CDUTR@YaLq&;O(A
zEui99)@Wfukip$uLU4CyaCdk2;KAKp0t9yt?(XguG-!|z+zFPybMHC#fA^mEdJV&3
zF;m@D^>y_pyY{x0A!CcwfVYNT@(xi-^gt9!sTIyNKiV0%6OCy>7=h+~n^FZNyVyJ)
zN>Lj8Uc876@3{>hzTy0#he$MtqkqnJior5%1sK#`E~mLm5N*iM1>AjpA%|did~Z7n
z%dBlG9z}>@Nn?Fp(4lxCMH4N79D5IHYcj(|QnfQpAzPmccy%RVU{4>Xwv>^x?(a7U
z_1k!Okf?0R^5~Duw@c4YDcL%Axt4UW0JcT|Wg%=`?GPN-`aJaOxq|C7ig;rH>g#h6
zOoE8O_ug?yc{NI*RFo)^v>0pMa3!vaerbi75nQOQl#<S>#e#3z50<e?R(ingwi_1?
zvG+ITlf`C)8!;hf0EoESjs&VqPXF+RDSq%kkKDUaRV=Nbt*0N*$bak5zC+cxWbb+k
zH<Hu)r}UX>zUU_JKr7*2z`VrVWC_o}uE5=W*Hm*kDjFXm{yqsH$Vdg0Aka(^Wo(j{
zZY(NVJ;|Ae6D0S)%X57j^?&gfLg@XjXty_pq<D32_{zuFv&&lV!~bjKJ>{RHvws#<
zXK-1faaIdu{9wJ+(++bn;QSzP5nb@06`rT8+WQc{y6X%-%jJ)U3gmL@01pnEQEO#$
zp|u@pyM7bu-LmZVxcjrs)26YhLRG$Ie?#qU^Tz0BDKp!a17)3ECl(yVGWS&5Y3^%R
zz08^l3F)sn6j1g=_LHmomsLJf-~aGWp5ZU)xid_UY*%aB6dv&NB|m);z&_u(6m2Vm
zG3TQvc^%j)T|YwpT3DxmaQdM}l9_XJ|Loj1oSex~Id)zLdRMr&2~b%g(lm_Jz~@T#
zw!3?>@LBUQK}=b=Gi$JjlfmaHFtxe1LBh}yp~FCh<gg#Dc=ru8<@O47;73NB?Zim<
z#}(d{MvCpQV~0OKc_-nAH-DU2U|#_cOQiQ0kuNU7Si<0QA#y0S{PK{qj$~`J-L|di
z=0uIa7Bdz3v3oB%g-89g>y7jChrQC_PG5$SAsJNLG|y@BS7U`oB|#|XdTT?|A7pU*
zQh0Hm?@?`-Eo`gx2fm(b1<zNpNaB7W^DF)J?Umm0njeV<52#y3(u4vzjVF%>#%lSR
z)@RK`$k%~zKL^KPkC&!pz4dN~M==kUdBe?EzeD}T_c1S&;btY%L289Ay%{1x9JfPU
zv?B5y92{R_S%yDa5f6=-inJzMa{nKhkED=rsvXI~iN+<+gcPR1_jpqE3p+priEYfa
zuchZ&9-u<)@O(U)>U^pCyi;tsPxj>@d+U`5$nA)f;tt>r08#{nyhFR$=70=B<<qZr
zrRG~b_o*iOG$9l~e>+$|^9vGb0yPjlcz3?`FEg_(IM@$_C{gDcXBSrwP_<05yJ4aC
zL;BpGIpp8qf_c5V5|*i%c)8!NUOY(Tczz(LNJ#DyOdqc^Frpm*(lCiSSCb2w>6;z#
zG(Dfgi9I6f5&y98DN<4y^*x93|DM!9=DShyu`h??8LFr#V6>%b$KZc%xV77Cj2!zF
zJqKBl4tS<}+7gLhP{$7YajnWnlZRK}RX+!6B_uIq=89_tqsIZ2q(Xq6dbPz8g^9lh
z%Vq)#h;7>$=V-q({f<{u)lB_T)5aQ8h4Rru7MtPVLR7QP`damQV9P%gppqiYdN1m{
z<ty|iGc-$FBEQ(i{-$EBe!uO|)8xAZ5H|+7!>Hb`dw!w`O(iZCP~i<})Y6@&d%CF*
z^mq%O<LSqyzG1pwv7hyEjmGIr{OjzXj~N4?a7tg{vj*a3BGUn^iZ?p7fGE5+0KO=Z
zCC+`r<Lw_@7a-92y?iRO^;FhyPt$K-bIrMFDUVXqz!yuB&=*r(9pIS<0v2(aApm8?
z0Z@Ju`21#@Dle^(5bL@2Ba%w<Wn1u|5tb(QJW;kViL0ZZ7`^Un^0azXqQM#^umuDn
zL1Z3)3L2`8GT%gVHXnC6H#CO7dv4GEwU$q-!zWxV_#8hAw(G;0!)X8eeWUVnNMqaB
zogTke7iz5WiF(9jv#&ehm#eyhGXOUV64aMIdb%I#<^*`Ux)b+HuK}it+}%w!YBI{p
z!Lgr#EC!Iv1JV#ELL2!~Iv{_^GyjV>IU85}KxL1Gsv4C~T*=>)Uxsel`_%T~&ix!v
zkF0g~iUJtdZEQvP#D2`It<JW9YfbMN>4_bXROp(|;SC-Ygf~xXEp14e8i3T8;m&)Z
zUvQo|s;Te(s8*b}@4#e%{0U;3jGfXflR34b@Kuc=bi=7f&pOBM9N@qgKTMA=@d4Lv
z84=y)U&|)2aFW;Nc3cER;+_n6lT$!I2Imt~C#%&;VT1FE@s;!G)3Rog#G?yOuZ3S0
z_qwk(TiR+2kSl1R_8W_$u0cGc@jAAx66u)du%aOfg)8wQ2KK(<pjMlooF}}b1#Ic5
zHhCX2ire2fa_*JG=j#=2p<wd(73F5|Ys(9H97?8GDz6i=R{o6zfGa8nySLjD5OiD(
zE9h}F+g*0M>8qM~7HV5h*cuREDVE$s6mOPwZRQyij_3BJv74@cGes8B@cwpwpRcUb
z_C{nRg2LkI<<$dD!#DyH4Sn&12!ta3o{&BYn%<z^sHk6sgHrmzaw-q;{JS?vw{3Jx
z$XJ%DPE3;2)cGXJeki)LWA5rLQFqCyKM4^N!6uKvdXb65;3LvYNu(Rehnb6~#3!WN
ztzHf1*_@s8aOv8Ttqf;LP`xqxZB)&`t<^Ygn|tLl^QEBP2R7`fCT2+j%Z=-Ri-ygo
zolpDGhM5WfIxeDy`owKG6glSJcCB_!t0|P2nh`(<?j>jBK+-r)J06UcvQ)eLA#kp@
zevuIDdSxR7^cJ#Q=cTo5d486e)^gIvWjWccUJ)t^y|4>CUit=llns13MLKWKDVF`b
z9V=p4uO7*et~Xe4>3OoDdHFzbtbN}gIjhYMSst7;+)D2yhERzRje7WQb11Y_354mU
zzLQO^%A>wjki56)?Jrf*L9fp(a>nZSu<_)v9SB>SxaShlnl|r&BJ|paQcfPJ`UCP@
zfXU7)r=!`QNcF30eOASp%Umc<{o{>l`Y2VI+*JH|TP{hpe)J;Vdd~Vd5GZAvAQlgh
zI~`xod>V85NxOIKWqjRGR#8&%#C-@hU_8H^t~Fnv@?&lakKTHKnHG2=jB)z*)a9M`
zQO|!0Uu=uD#N|WRKBt*I&;H7H9+O|R5Ctyeo3hhF-|2JwA8B&DO)$24j``@dok3f_
zHwA7OT=`7X>f@d9(v{Nuq)!nEa(yzq&tC;qZ0}CL5j;x31W8p0@7Zm26VX65)2=>(
z#e29y-Qwnx)zm|4=va5s*V=-Sj)`~1kd8f{%pLMadChc8DFV?s-%2nD1f3eCRZJDI
zKctVGuM4j+lZ2q^y-ag_10+!GR%!#omj%Wtj!C`)M80n!%TvtpvK1@`U7o+V`)2>~
zJ)FKvREEss{n12V?77k`>l{LeKif-on-X_u4<QkFz@T*i(k@8+ZNIEZ>5IYzW;SL^
zz@p9^2tH(D5s)cKIpvNrP`NHNs~3x(^7?P74)mTzb6;pYb6PY423@k<^PmWF;G6x(
zmlroLuaS<|?)-Lze!nNBpA@>Qiw||#EqD;}|LI>sKcOMdVY7_(^d6z|&Q#G3L_jik
zS72?rP?GDQHPX`aTVL>TE*^5Wg+oO;B;n>b!?NW=pxhiA7T?`jNgA%{;pPS|8Ylc~
z?#wF^P}|`2ZU#bvsoXz_V;*1&+FEy|(_+bPCxSt=QydAv#_pP^%rkLWnl;E*(?k4T
ztOJfRG*460j91cj$dyAh=ZM=K3N(|+<c$Hr_CkD|>p{pmR2VNHar)GG?(X2V>TigF
z{;$Iij=5f1V1?{s!%XCio=hJ%UPZlGYn~XMiYzr*{uL>-&Yls>opVxP0n~9YU%#w9
zBMO;j_BQ;i-Aadt@18{C=)ip<wH4Pr+`VX|R4>HbR~<OKO@yFhVM;2CFNPV{y5Rgs
z@8r4iUJ2U<FV{kSv`5kM7*5}KqDOv5S%Pm|I*VNU5yJ@1WwHGjS9xpsk3%kw31ekc
zc<s&QUS$K@4cMNorMN#uanlLV#~dH?OS)a!B{^ks!j!D9t&z&DOtRJZ&M86R5($}E
zK{0L2@1c3Uk%wMy`U6>4$eOgjBxU*vZi_3mar|V@f4gzeC27&uGvpb*$a+P%^DIHM
z{ES$+l`99mc1{%RhGTEj;Ha1$>qo3{>&Y@M+@cX*$4`^rOjVYqW3o2VtFu*s_r{E~
ztaK8l`C^JAUsp`Ght&Cs;uNj|jCQuwNsh{!ym6yFIze~G<Xt9rOi!yy;rs2bE6a3o
zD6g`FU&!_MLFSGh(V6@E{#3OuSeQtj_<roj;%w_h5S@|{y`=c+CBi*_CDv(>@l_^t
z39zlE0ou3t>oA4VV2MT)72z){NyN!ecq(>Ilgg5XTjZ?(7NmBGl}+GJ!iage)@6hf
zd(seUv(Dk1xY@xwaHHcRDOy^^-s5o%EP!SFWrm;Sl+V~7_Cv^_?dLe%uP46l@#*Xi
z8u-vAEE#W9vPUJaf*QA*`ex)Kaofl+chC{f{YisXs}cGkl4&-gs&^dEPf*TkXXtTt
zTaur8z4vmQQIq9QTpkeE(uzU1=2(x5CkSy<mldg`B9Rd_SVl@tp8_uPX_J?gA+ew0
z(oCno38Wd62%^z$^P$o14zO_hQO;5VHmWNDdxg6Pt*J@T%*rU;dT%qJb3fOr6M7D!
z5w_dBn^p%AN)r3wrjNHhS2+x~r7`mgrtv!P*nM@L7C5GF--!8zZ~%0g9a52zV{=i@
zXHI;Nw~+S9OgpZY!|m&cZZytzg`XNS$6N+J4I!N-A*9Yu?7&jk_@c;^sjQlw`NZ)3
zx8Pnf{<p`1?{#dumvnQfRM=5<mvrjoOl{GyRYiZb@J~UGi03Q;=z#OxKb-o1=tFrX
z!lbTZ#PX06xdwh0VFHhrIe_??54FEVIRnUzGwJpL<^8_lSb_m?1>+kD{%L~ODo-SJ
zH?8A>7EcU&o7Nu%2<6ADNunHrkG7qcu{bR~>u$%yvDZw~pVPkB&cYCk32}+NhIU&W
zq@i}dEN7`MzvHB8M}TL*@Yglo-$)2&T#EpSc>7WJQxx~}kethV3l&2xWt3@M6MF9L
zut%5dPoTUZM8KJL7_t98{C<}ol$s#JVfxeK`SO-xd>#-bL~sO1QZb65hSdmNCRx6D
ze7_aBJmMEm+{qg9An}{wuDnfzLV>R>X~w|l`?&u{4Qf|6YGg?$3a0^kAMW2r4QK@o
z(!#$HO34RVY0il8x^L4#hzGe58HZv7*>zgz@UoyyV%lpTBG+=+`p$PqwsYD~XjvfR
z%5b_XTnHsss6&5X1Z6SaC7Z2d^7Fz7tP;*tWZUE6P_-?0ELqd+h0|7osX<s%EKV7Z
zg{kTgj{_fU-F}yS6;7t`*IxCtHW^$s`AJ^OhwI973xu>a?-7cI(1eDX%xhxNT*J!l
zh9NNsmB^Rr+=e*OTc1_8tnVu)_2FMVTy7Wias?WO_r)Kdggxpwrv(fG2l1y?mMXD`
zkb{dLdbej2r*4TWX;llV{=PE@FC?!0nWV^`_Z_-pet$g7_OQGJZ$kqLq9L(QxUvUg
z-?%^LbUX8RIPj^z!Gft_&sI?A*(#-#1;V5-R5D2*O$^kt`NfFw6!n)+zBOf6R3xZ4
z528sbrIH!jq?&`y>zXB`!jp}!zeN)k4<Frw6-tg?5jVgrf}6$k6GoDU4Z@DUa8q%0
z!mXxOf*(yx3WD9RB9ypnp&(=LA<2+D4U{xf4g50{S`egSP!d0bhhzPjU^9=LnZqEl
zM%kg=v=a}>ER)>aY`;^P?k?}V@7?h<(%(#_QTNEZ+1%|CMyp0RB&dW7fe%Q=H>KZ%
zOGX|v`-D)j#dLJAk-F$yUH98j3hxT#s~Z-k$C64amZu=o5|>sCx24WkHTC;Ma4;dN
zaI<zTV!78^wNu^Zl&%T8mOV?e$<_@!1_p*)Ol-c5<OnUcyc{d%g85;by`20h%yt<C
zilvVa|4ry5^9NLODlt;m<k{_ktd|a^t;n35BI+nkO;!D6pmg-Yf4iuPYbx=Ra2zkH
zoj^q>8z7h^^n2LB6c+&?13vbP8tPbAJp#a|{zyC9jTxE-CDk^a`=GH$m2!0*$^Q|C
z?lPbm0z&yY9Y9%(r{Eb#%=h3|vS0JBpiJ8gIO=A(h`c)#0L1qqBjEs8c0cnuIo{Hu
z=|C6~h-__eQ+^dlRG&U@9_rY#3>(=$_)bqw);eqnKE4R**a95U8%wLW3$z#bCbf0$
zG!xDdy&-@Mh5McXfHN{!p8@KjIMX~OB_$9ZPwQLL4-X!E#6ZAi3h1M<&G0ucD1umO
zhcC&(oL%1?n1CD~I+hXBm+%O3*9HMtfj}j5mfzBt&qMrH{FH&0ar(bKVMj+}ZVzdh
zu|L9Y08OoQk7x5pqKK9usv0BqtSIff02R%XyqUL|BnQs})8<KtpU@{M2_fWjnnkR%
zzfAOtM`p){1d*bGguXqXz9)2p5C2ss)rn}h>Zlo8L6vNQBqfjdi0E>+_3Z-o*|5d;
z%CNi`t_FmN)+~Qkzj_z_`>q7#((v_Bb@tb}(qGT8kl%)+A+1~9{?oGApLccs?Q#5H
z-jTfsG4FyY6!7M*dYfQ_JL(-8F9JUA5HgraZGUmGNH(qzlGk63k&mhCY(Qs#`H6LG
zf3c5^oPvO??FaJAVPj(xtvQ)Hg5gMM!j6~xnP#R1Md5MCz{d9U1M`#X4epRwpa$_F
zgFfD(nv7MD=uMdy7*U(kgC30ENjE1$7BXO3P@Im4?dcnj3co>f0bj}D!43+yRX5^Y
zW0#1(DmwweoRCSVx|vCl=r5B+);0VvF%ln1c&ArLg#q3+XAgf50OZcmjC@Vkh)+nO
zKg89kPx?Z6miA^O=t|kjYu{gPesYI@bn{Q>T!l7|okk!OVW1#fYz7}^fw8p=6Ww4r
zLuI5p-PMLk&1&rj8hYS@l4GUyYZHnt_wPO4u*QX7<<-;21smC!1Z}FJoc4cn_RGt%
zNs*M#MG;^2I)}Zd@K{=38@Y7E)P_|}Pe7RpO2q>NGz4qBr=V}V1W2i?CHyc%Q$ObS
z=#8#hATB&v{P@_BKVxqdU&XLtf@q2;N#gnEM^1OxasC8Hw3bAzighBLD$pa0BlTT$
zs>FU#j0Klb?7sbH{mnJ@>FWDeQ8p>Zs-}`^+hmY*-E|rPV-~11%as-!p&{_KY|XKt
zi2@Bwstlq@)Y(a~?Bo+^rngns0YGiGCUd+81}+u}-1&H!b<-YSo>gR=rnL{7BL@jm
zHnc1WO%Xr$+@4avkF-~p%Zlb4avyTv{rFBD1!*eR7}=<GNF9X?O!!`?ghym<cGW5$
z6#B7Vqh3gRJ)zs+T%}$cdOAXp>w_R6>5#}@^#CTajMAnhqH1{-3g!OTJ@udjud#eb
z@nJRg^NO@znw3udObD34zxL5iAFUIrs4GqMePGE9?Rs?QOQD7hc9B_Ej3dhC`(ZPI
zoG%ihT2Ji(W#>HeI<iZ%=uvIyj#4+T5SGjAe_(E3kYS<Aq5IMu{GoQiGO-tpW{Diw
z?*-PRE9*Qgk8lH3plL}4Qn5!bHy2V&SG<J1rP84Zx+^4>Qdu-Z^f(jsdOeaC_hd}%
z;1XwR4Oq10l)fULztgcyYe!K?JR}<Ti%pFUd^q_)v=RU!t4+x-<Xh}QKNq;JM|Rs+
z2KESm5_jmeG$|_I=8}muqBUX#?~f#CcQDVTqIR-LRq5>ZB5be<HCxgY#f3y8?gf*$
ziM!8zU?#LdZ<0X3Bw0JiLWsu3uzSaw&GOiVzEv|;R8q(bqH|<SGEd3-?yN^NK7Ji3
z%y}cKt)O*)g8`UV6j*sVPR=U}foCX0X{Eg`c7a<`mO2!UQkVSQWU<k4L6O)a*%LO?
zfY-z?qeH9ytXXa`xa(**G5aw@C1?VkVh$DL;l?%xQ{v?nJ&U&fr@)QemOi0wOt!MP
zf;en<Bx`n4uf6M_ull&l$^@gv0D<l*y1vy_T!EotA?w#oU3Yc?zEBLO$rn<g6h`(X
zby#}uR{e#2c5zD5YXyZ3RK0Ps<bJ+zj37U(pX)~-%XYo9k8zr~F~LSYwZBhXd2j3^
z(O}-QjT*P-SdVoL$JJHa@?sFtE)3Ah<{Uol;}cL3g`#p3c6fW^5fBV02iJQXIr$kW
z;#C4-4!Iu{2%wg9U8^tqNjXUAORAj;B;2r`9y{iA7FwE{mZ6kPcepecMoiFTX0P3F
zF^bW2wzCW4$#NH4EhfWb41-?s2rforU93-kbl`8O<Kj%5cmHst`gcm{Imy@zn#I;d
zlYKEb&-^_ldaU>@QoMs7;+E>lS&o19IMuU?i05twb!AE=$d-Ixga-o08H+B^g=!CP
z2|hCZ?mDD(xNrSUClrg;TP*_Tl`tcPHwL7Q-&?AG2m2Q<{rURh>3X7DB6En(5?)lT
z#mDKw<iQoZ##z^^GeWfLznwc6LiVE6;J0QONbm_egi2sxNo#CHMu0aV1(^qfGww|%
zW)&}(CX}(?!n<cnlKg0o$mhT`jsy~pR5XrU^+{vQSUwaruYMcpth0YiqV4--J2EX=
z{zYi-kyiw1QgD{Ssi@xtwY%V=e7tQ;;Z)!Eu8~x%HsTT4z)}gDK=-J2<d76b6GM+F
z(&}ka{9xbQ?_x@2S_IQ><YiJ(1-)NI<QF9P$ilB&qzQ?OpQW<jnautcy?2`?O+g5W
zynguh>4?8kAKF~T!7B`Sp2x7LH7go~Fw<6X)F-fvsN^cV-;@5?MhupA&1We=3^7#;
zERXBwxmao%r#K%1)j=EQFA=&s^UnGUnps?%NTnR~t>R~!Cr{6Qp=<ZVq2WiL;D*Ky
zR9THM8IKl*;1N=?GfyAZ2}Tu}Qx!K~QeS?h?GI$hYv3L+UFL;d3TwAq)#E9HIS&OP
zh(V-48#<A}?DG;#Ytkmuv8{bADeu308&Mke0X?kw%)36WH~Eqmx;gCOx<8}ush;HE
z!x9>q&Ie=#+r4hu#w~m$7Pk}{k>iW(_yw&|$GBvr<v9M&dA(nQuPj<d+Gb%lgQu-2
zer5RCO}mTu;Eglv>NS&9G@il|7-S>4S)DPcbAO<n75_m;GbHt^ekUt41lmT)K;N@o
zFT{%j>pqy@fjQYj7tBM%TZD^}(!fy$8RLSBxSN|YC-7I9KW~sZ@hiRWpb|VDiTvAL
zHe*Ro_9$6op|{li4U8Oca1A?-gCkmmNyAT~=$ueu{3CC$WyAWvvxvB0QSHc9zZ2^q
zuG6}JVF(U7?s=Xn-7b!$KsSg@6R5#|w{&r65sR0EfV^Z&lP>YE<-Gq3HY`dx7Kv1+
zBgIEwg#b(^$!_Jq6{GmPP`lH8>~hbvH6yuNYYY>_z}{q-t$XUKbfs$u<^>pA{AiKD
z>^jWsvNRURNU^*NjZh?8gp!YC^awRNV(Y?@E!&lOpWM={3qm?=Qbfqct%L=kA*m!m
zSRgJ*%56xIqzM+Y6w_EqsYplUbIcNKD~7tRDAYPiRAuN;t244bgwDl>?otXDT7qOw
zqM9J~Etq1}r;{H8Hf_pwaAai@Ne4$NqY@(ImkQhZM)3Iaa<qYJ8%C-TVz_}E#FVaX
zeXY}pzG!J^>UrX6&QBlZDN+hXvGj*)pXjIiJ1e8Ph?U_k@W!cQ5VmMipb@x^$JBnP
z;`#o7^yKm&zq^E6KIG}2@GxkcybiR(<nEkF1i>e&f+Ekoozuy>#BPaFkU3priOjzs
zC*?DBp^!x{@rGa~($kF96FLYGi;S^ra}6L!+uI}{O0|uut-f7(%NdgGQmt;GKAh(q
ziiuu>SnN`!mv8pt%dA9v?D=(Fr#ej+ZVZc6d7XA_Qyw4wci&pU6ANxgo(#8!GUd0$
zc3G=VZ}d*i!?}D8^Bi3wz_SjR_MwFi3_4bNAgrM^tyHzq$=%|D3OcoF9XD*m6U6W*
z_;aVZvl$-AoWsSE4>MTc?M_LYB)_kPqX?t?OWs8ZT9#<(!tjxiiNTP8XI3@@UxQ4e
z*6?p0wiU~+<Lc>^sT~t6J<A$E%@IpFQJ0X%g<bhKeO|L%9U90XBk0LGGPF^@Vi|0J
z80l-ujT}h`QqP;rPR@UOqz1~k<aJrN`)!N8!fzm~`FVEQ2uugy1U)g{PX=0?v4`|A
z<IOVbiP8%57Z&x&Cc^qTf+P0PS#~EvK`hXMvLUwpuc&Dc?h3mS->m43E~w3J^iZC}
zIN}7Y0!7Jalp;4|r2ZTpUw5|2nyj2fIdb=A-HCU^Mx&uXJIWrY_&Y@4BO~8K0i8o!
z+c}-kDeHGmOS?X`?+}_j(Y2+DK$L+>QRAWEcIU!0M58!lqMaG;q<yV<kS+mDFo%et
zjI%PT8k~DAmq=}x9U84@zarzB10dz~c{*{d^tXM>*wk=sFR7C;lii3SI8uz!BBTa6
z%+yX2tk{Vj$0AX+4~zSC!|S@=DVyaO+sED@!J$6!CW#UU&f7>qe@87)wuJtw^(99Y
zzi8;K%;BP2tb1MqVss&EUygmrmZa2=13YuANARKc=$07+5bthj6{$Q((MsVhN;S1u
zRR!y3Igw(ePos7x?u6)Y@}wFGXF+gNp!#ut<LJDGFd%cnH3e{Ii(~|W$s)xV=91#d
z0Q}|TCz(0!q^Dfja1LrRD&`z(O~v>Uz7jkd2#1JxHkQL)8!2pJagI?@CK4WzdI^uU
z_;P%NU#Vp3w^=sDf^yprJ<`ry0pt+!36`bVF$5b`)1Mg%bA3kr4x>C*Y0F=lykb0u
zlGVuW*P~ZZyw2xjM)a(5_$MHZv86R=pjf9D=xJ~CD63-2=j!aIzT6XA-rr{?u+Gj8
z13)V^_0h4fy||(!Y&NPRPb*6m3&Vo&*L*kni8#sX(n1wPZ+^B%{TlhX<jgYTGR(Wd
z*^&``#lwI#&6k|4WjiZF(FEFxU*dnRicHN_w@+YYycCN=MK?v74_+c7RG+tg_b(Tq
zGl{u~kz|ZV4l__v{gJdpa$&sLITU&^TIlRuX9ml0-yE0h=*J)y(iVw=`HW+f=d7(9
zoK$7ppUiyOBA!G`8a)A2l$}?z)F4v}i~g;qu)wbosVDGnsszv8YTID0mSI`^<t2ZC
z0bn!xLjbHeBgKednIA;Q)L$@-=SUoxHw^)$!5(RkWDtVno+xc|h^n)l)r%XQ>_-hV
zisZbE8boN{=<Tve3SFjW66xu~3__Zh_1c8;?gpDW1yeWeh=bq08D1u(+|*Vx$O{>6
z<wZr~?rh{{O2p4flG7DUG^ZwG84f_)oUgCrc#Dc1f+QBN#0{-N`?y5T$D2T=vt8A<
z247XEQvQxAgPA{mh{CQ`6VWlT5ltj;iCr4j8;-1R1)YIX5Cae|i|LaIU&2%H($h56
zN19iXL@)T^dI@G;I1W@<i7R4_<6pk<;@D7t)qPBml|@_R;{9+jv3>f%j>sjLgC;hU
zccfDrOgb7YW1Dtsv(2w!QwjTpQb0FppXhNZzo>L!tQUdTQ=$|xDlLp1e{5%uP)0hd
zrAocT1%`%W(^Jkc=`ad>s6}9B8iFvOdDlNJn(79F;mEA7<MB2<iujh-&W0Pd8O|>z
zlcdDeieW5_ThTK3YmN$2^-m|5kmbeUd^eopJVqigrW6g>wH8{Ovn&%tE5QU6N<6aG
zW{=zzkDy2Js3%iKCz&e!b3We<)&UNRBEg(&v$_sr5)=1(uuzY+!^TqK?R~s*G4iMz
zX<W1i3f5QO^vR&u+*vZKuOiseYLya}<SWQtv*fU_p>dwye7J>{d8WgT@n1KPp>yHU
zUgJR8MisLbWCiuJXvqbvIwr{RjThU&Ds-P9=cKgMYfUl1s#fyan$#SmKmQ8n4c?@g
z3si+6M?MHC1mU%l^75%t9KAK=RfdIo5|wxofruUyLjb=MrbFEfNKiK4H-$OGjiK33
zaC?b1qTWi4io$<qKG9K8L@t+x;A8+obMAy54P2i(5wa$SJy)*p)d|ETUq$`5Tigln
zq?DionZ`skzsUW*45Azp9Dn`!_md&3gEJZi#wR&Bn5SOA%^M!+a2;V)o7?~-mEsGi
zf4Dj~60^%9#A+zvg_nu1WW^hHq|J=;gnvN)Ie7Tt7y#S0a|g+MS;yk$O6~}!PD%>Y
zV*dJr>Uw+C-;sX?mf&_%(i<j`OdNZ>fKO%EtlM=qRqw3}8b;wr#?#N$<FNdhfx!Vg
zbQ2+6TcP`>F!c$Gui7}lAeD0wlqN2RM?s17xG7yINM0|;=@k#O0;Ow(AD?($-$-#;
z-A6Z-mXDkeszaE=z4#4vZ#)4Yb_)wHKJWW8bAZ+Z;MBjItWMJJXt|zLC4hLib+6gE
z;?ftkmvPpn&{x{sem5=)qGI;IOAc4?OollAZo?S0YeLK5N*p=#dEAjMg9k67vV=Zj
z=2snmd7tsRkIGb&N>7Y8`{B4&nAMfsuf*Ljd#4QVActeF!wgMCgN!>_qQ~MY7gWk&
z&ecBXx>GO__x+~h2EG=TSA5CKz-^<Sl3<9p)`*oq%B`t)H$fMp>!6iyhD#_ueq4UD
zn831L7Z%8mUy6@fDfB>FGRwuqAmgu!?L-u3y~BkP%hgJQ;f0~rm$HXyw&S&skT)>b
zV73t|;l_wuBE2;o-{?73i<ya?7Oy*f6~?L%oM7(WCm5IM+*&FAfMEZ>S1kx2*xnDE
z_ppK0HX!v~Y3(|>qVsg!__U~y=Q=71ZItzP2etou*U+#JzsmSsAD`A|B+i)d6pn!<
z>*+S5>k@-k$l5x=O#K{eCVdH)jlgjaY29!`h_Lf@qlXt}8t$hInBnJzPWiC)jZV(f
z9Q(HAgDQOd24lqdt!Q5qDDxSm^e2X!F2YIR^55^B)nk5=UFq=V%U0lied*{K;_7}y
z$?|zlq^4I{(cS|S3ZZR9---VF;{W%(zvoY9p!3?lM|2Lvb+PSbvk-ed!1;CWSjYS@
z3hn=ecnrz2o%NPRNd}rF1rZ<EeAr1u9u)?P#imHYfFZVE61`<6%}M)IT6QXXdwk;d
zHx?iqJ{Fa`%H>A*k^H46YtgG59a#*9W29n!#`E>$`c)jY?i;6j@2hiNGf{)Xxn+~_
z(mjP4kwt{mde5}|LPiZg2onq!g00q-;rSiGv6-Zs^PRu<#Hr8oj2}s8K?Us<{dlwl
zNScfKV*6M6zn@!wK_dS1>v?`RZ`FQFfFt(J@cWS;nxyqv<8VdO!Z!TSRd;&OF#Yg>
zk)AQ&wc@<aB$a|9u+-|~^pyByZvcc7?89qr$mje;d@_eHpAmv*6IUH3<#ED$q8C36
zzQNbQCD~6pg>Grb&&KaIE<8`u2A|bnX({%&G6|;ImNZ-p*2`YQ#>P@1ZVTSwH*ws2
zi6SS+`eY9MA{gVIN8RpmECI)`V-<5~E)$vcZwvjO_X~tR5m6gAUY?|c2kr@@twrJ@
z*G0F;0oWYzuy1N$>W~B$cclaans=haB};y@gU_0b1`RzgM%?GQ5*Aa<a|Csx-gtPf
z{$zWrvRnIxu5PjqF@#xe;h*hBDtS9cj|M{0tzI!84}+Tw*GEniZHQWD=e#_GWb8c8
zA$YtT(44E`=6q1bxraxFnQvVm=qWqg&kEK9ySy#axLB}l_6d4CVGYO~Jy^mwjQ`i9
z{<=$q5P)ZUaUDokxhxCb<CLe=d2MW@_QQ)r3>)|P#UcbZ)8#Zp5DL+}Zf9)KU}cte
zgkz`j3;S=5i<kokFBa<JyB0Px;bd*y9y2^J@$;_~gJk+BkMfToG6M}BmyH^{=#%1a
zgoYBDXEkT|QzKtv{HYa4L!n4Um24S2HMXbNX{WpDg{It=4Pu*^{Wo#l%xAqKTCbc_
zv&~g({vQTI?l_yE6Pj$M!phbaUw7%bhxg5C{SeqdFJx}#0~s%2AyN(qk6YsLexk;5
zw%viOD_Z<Lve_{*@loBadyC~z?@Pt>m^7F)@FGD9R*4i2DuP{V78Pp|WhD>kVGibl
z;kw52?0nSnS?U>#y-#Dy)7{GoNtahm1f=o>G+Mi-jfWJ43?hp3D4!Vf8*q}#^rpIP
z70=;|>2#Zo`f1n9=KE3wh3rq4Q-@}nRXR;J)lCN(@s|Xqa?va$zcrystB(Ut7q(B;
z7xB-8ihP!5iEiCL!_h^Y@T4ZYvq&&^Xh!Qp)2((Pb`T8^i!`kr3n0j<cSa=b8<rC^
zti~L<jt>vR5D*Z^m8xcx>{RoenYg$F#}3>>Tm+r_3gYk|)k%evoR-;2o-9x!pJ(HN
zbp4i{X`>3L|Lf&!wgg^e?ai;*La-m{9`pO5vsUu`gzesGl`S{!3y~mrj#-QpJ_G^{
z1xl_GJB}vl{x~#r?oudGKpdIq*Yy(x-_A#Kt=sd|E$R`X%?|Kf(8qSF0al{WgVu*G
zB0;97LGBv5=wsV`98UWW7Sd@MG4xY*G6EIW7WQk3X{96~29xM20_vFJ%K!D+U(rTT
zrh|N<Dh*zw(bZ*-k)*f?Y?ry@1z0LjNnx1ZL`a3;zAOcAl1ykaJRi_5De8!~x0^DI
zyr_mfSvHmHPZuwKG4{n1OK+MJL0r^k19ToLLQr+|%p>L;dSJA9)H;8E$^W0RD9DPc
zj4XTS;Y3=pg}(YhzJ!t!MAj9@VvD|miH@v6c1vjMG0U55EP%?J5O74NrKiP4CsQ2{
zSx$`Ktk<Jj`j62ahcjgmV+fcA3EIAS_fC(5n?jsi8xtX^85Ke_`RV=HVMHbQa#p4A
z8=fW6N^ram4%s2@acHFMagq4g=l955bCMxLeaBBU*z*59I^xR2pvu^?{n+<ulqy5G
zXwr!~=ptgc@K99v3E~@juUgSIni^zMWB5fRdf}{DN|VSn0D^$#M0-Ueq&j;6;~!Q~
z&$T4B7zG6v@V?mnIv`*$d0L(xT~!QQ7*`BEy^|q>L(2XA#@W+1riTr8p0i3(yxtJg
z%3ZY7&?-;i95Oh~Bvq4twCUWbg`tdKCs&XvtV1U|x}*7~BfQbif=DdXEdpB=yRAmq
z(&<U6Aws)-#)7dMy&38lDRoQbzpMZ-ihW$}beBPz|FYSCO&bd85|zpXbZuV%l*uF-
zBhTRes}y}})PAiRg`Q{}W_gn+y9PPpYtMe3><G=dEN}l+24P7PaXBv5|C*Yag}oR9
z1`<`0SOB#)Mnb*W9OU|uqHS%#5+iw}vkAA#W8aQt2b`XFSIhNHh^_Mbf3$#U479+6
zI*1Ax0SZC|3!70RRsQ-3D9-R4j~EmeXa8I(NPD*)Y38*4`Pe}BA73vrS-!wP`(=;T
zrYa`LxAHMHqSH{Udd8n5xg|Q{d5hdRIKoFqj8QNrRP%p_2e1U(lXs<S=MXcxYj+L3
zu%<Eb(1K}OZ}~|)>|XDg0t^l1S^oEg;>yXetQ}E0EW;>o-5Kot$Sw|^AE$cnk(z&h
zkb7p@mP6xlR}cNiJqesE?c|+`Egvzy>9e!n3rer(^|2&`A>oO;_h!j|FDw6=qI*2|
zO4tY%3A9MbCjqoyI7OJD|D4AtXR$Neq9vktRYtGgmYc8ly@fDO0;~?};FwD;(-_5Q
z7Qs`&RFVI^0Wifq0~`&-tK8o(&D0noce>QR<Vx@2^$G;D|6fm%yTb937oqpYhu{Jg
zt&QjbrI+cA15Tnp@;7+Ed3*8>IEVQF!5N@QEW0DNr`28k&+cTApU~P2;Tkp%NX~3n
zY=Js|Y9SQ=^JCPaXSR<b|6^l+{%qS)QkMN^H?+Og7XM-O;>t_JSWW+&?Bpg%afW}k
zZGNi!|7u!?Y)NOWS^oh66-*iu5RtWw{@0B<J{El6O(Lsnd%1~WdNQ)Q=?xLX5&_a*
zzRexM1$JSN?$Z5dvgn4he?Vs0B*QhcNHfuYa)db4`>pd=SCT1Imv4iZ&8zfL+oIUg
z?}-ly3+EzmO_}_T9(4ci?*1ZJyb5c+nL$`6GK|tJ<&+tx<L}<_umV(xW`N3<FnVJQ
zi?1touN|i;JN}#-5m|^IA|pBX;l-wHGt}R`PF4k}YyJJ*zuN~ax=E6vNka-0IA^uk
zLtH`}vd2<!r1;n8%pCw}r@R@j+Pb>FKdi6E?eUYW!?SKg!)bm<4CexF`V{28I^e%g
zfw{jZ3^W$O5{3%7aMY+6y_<D>{o=0g9wb}#*Mz@E0WNxwC)_3O<>~${(9(39;Lx@m
zC&lnekNfoP_Yd+fRe#%tvrT?OCzeX5a4bvAh^bgei8at6LI1zb#sjc3!jT6>0Br!O
z;$22fZEcYewE(42WmA`FdQ&PY<fixkVG0T?>^LLvXlT`~yYPO$@BMS9rsQK-Vk8O;
zCT%h#LjN|7RSIAtzaIj7o#m+hNNBCv4Fg-Qj0}E_YhG1a3Pb}S0t~c#dtKlpY46Af
zaTWJy{=VD-YC17svYhu*G_T^^JA>(JI_vLI8OoL5dCN8aJ)p^5$_0%9p-tAZQGI|P
z;=zBAKI236X$<OIfJFif(y~=ltNYKQll(m-N{4Q5+3$4up%+%L_ugmbewH0cU45>Z
z0r-j(jf!i&&0mADN?<;vBQ<UOX$&AO>dw6MkW*LxBgEx#?s?y_wSfRES$>mM$majC
z%u?)%+RB7_BaU8RYmCD1cNY1D8_(bO=UYA7#ub0{XnY!RxtL>kyX%~2#Neg?CwqJT
z|9_s`bJ5yVp`Hss+;4g_0jI4#cARFxkA?7Z`uko@-e4KNv4)z^>af=Zgzo|U$(J?l
z|6IWM+qYPNskk7Y3jjM3sqEM9FwTP%+W+eP{tsLWUO=_eP!<r_onB+z{A&P#VN|eJ
zThw782gjh<+ULey0P<rH0Liet(BBp&DS}oFoDSgh*=bx*0N7nPIXTNJenNd#_>a-I
z2`MYvhBn$ySg{uX1dm)?<;NKn6^}Q;d2>2{U0}WD_(QuRpiJb)FOaRkq4Ia9F!$et
z0kblQJK)MYggaoy+J`#;2*%>ZkGvf6UjApAfsu${fdoif`~37yv%XKV%`DP`{?m69
zW_CNP&}8w-Y?rP)w>9fhGV%zJ{d&&-t}(fOk4LzsWycxK_SP9NXgqV`YX7-`&m`GZ
zY(p`bRL%v+pa8T@3u?OCd=YUnvPrgq<eNxRBvdk;2+vc<NMYEtMj~M#U$Awk6@|7i
zlPk59k0{*UjeebZ%AfT4SYzdicUKq^UG&wKwEY#TaPRXj+<bAN#OHG!qa}**+8)G_
z-nz{NH|l|k1f#>5dMFdaBO)<T)PR8e%v_J(ZQ@$-Uip)eD~s3*WMX!;sI)=C%(H9Q
zR&`KW?rBCe18*Eiqa*|>Q@<%D0Z60kJDuHUoy>0<verF?N9%Br@BN;Qf<w6-rz5fY
zlKl$h4^y5HL<a|v9ic+3$dxmixKT(yyVcR}fb8oddoS9wK3#)4_`eLOlpL(v9rY-r
z9(Ki)Mz?--)9*>u!1JdYAJ?wd@svwG5L~xsn)#G|{%K@!yy8MwynUGHnxKyTo<1|Q
zlZ1Zr^UeE>sf6K%#BcF<3H(&#5X59gb=GIDNsXD~%DVz-@J!_;Afm*y^G@j$@m|>q
z<y36RfP~}kUWiD^WOZ+yT#HJ2hmkKxQ(+`2rvmJ<L=Z`l2Mf5axs@}f?cDL3i*$!8
zS1FZ;4~x61&*PoX`i+gCT1{YRu0vRbP<ul<r*1ct6Xv%dvfse!`#F5?bhw&Fw}wz*
zK!P-eYF!N7|0tFI{c3jd&?%5?{R@Oiwd$|-;?G3b2lw<g1b+1e1*!FKi=*o#HzmOz
zu%gUu-716U{E5aPPDI4VsvdlK?;H(s!VoMnrJrd-vfPY`rp<wWYPX8dv&jh;AjTXe
z#kM8H<O_C1MoYZer!;cR4ZCNLZx3WQh(lj0`H;g3dKsLAAe&x4t|}PVU2<#3$-Yhd
zaabH379kIA#WbyclW2zvLVXJ(8xb(rgw|vi!a98<Kg-8mM>+RlE})HZYEwbe<<JY`
ziYG(=P8HJ}p^H07-gY-ttR_1IIr~FR>JIK=-L+c}%HFHjgRI|<zJ>fFtNGAgh1t{k
zF)B@%Vrs}dvpZO)f@^#%K-$##>eB{aR<WnDkox>nyG}iGjPx~3bg8Cw>XQr&nDWNg
z0cx_2Z-#^pOSZU4zqx$>l4FdFJX19xZ7_MyIYY2UZ|#h^OX@2hn+e+8+cuQ(4TJPH
zgzd%;jOYQNC0x{8B6LWq30-<UL*~|@kNP~;bIQ&2DJHd>k=f*7l=4?9j?0m^Mw#9%
zxwPig1J(T>{TTS*boV>NV?e98>Ew{RAN$Qo5;1@Rh870hCa;)ul_;J@WU|O^!U+TA
zK8M-Bhe*eSO{JPjoOdhL-!hae%#{<;s9?zy4|-84H)cr3NVH+czECMTf-@S6u-&C8
zSEWgb#cZ>!Izw(&4qpXiGH8paLw>dPim^F*jJd=ZVCvwAZZm^FbenfHcZqQ`NZ1gK
z6c9m6uH#G{t7v)EP>(l{UafRU5qk173}90-p6Yh`$2*{rI7X}<WPDg?Ayj|TIFNA~
zIlrr>c6!QHRd`e1)Jj<)ZRg)m{F@$;i`rJIE#;bz2Z#6|P6tv8wPB_6hp#@vi@4&8
zHT`aF>mDj>FR7~dk-O<<p?JH<ae+6Bt~~Xn{J#@1->#71s#F?QOOwUI(g%bPz99Je
zqt^@NLFb0nwu^-&Raj6IA0L%PqpqyqVHyc?f?(`Mn`}Q^26*0Lq!#>4{)CU!lx)9J
z;uIqblkL~@s~ObB!V^}OZ)&6;!!}Y%Xnb^jWqAO~9dNwh>~wH4I<eTc7a|!wbPerF
zaN5-#x)sn+wjmm?WgTxDorN&P?p%;_O2J<Wd`_n#(G#Ua7DFGizrJAm)U_#PEuARX
zB^pVkq9H|J4(slBhD48h)Q`UjO;Ay3{&`%DO08NkdnMO4myuE)KL~<6jpL{kXW@>)
z;MkZw9g#RExrEJst8xs#ysgsW^Q$M1{7x;3Zf3WXvI0CzoCyMb{K98(Ci7x?xEtL=
zp^0VRqplNnO@{9qzZ|4qf6tE5opEoeC|C?OdC(Jwqt!xzyO%YzymBV8#wwg#vsaID
z#F$yFrd)JvW}Yuwpq}yLMieb#7^>Xbs#9O19N0LyJV_EcWC<{Frheqiu8Y~_4Be_T
zg&EW~l;j4WJZ3;pZC_RLdqPc%Goaz()ngc=HCDzkrCP})@x3egvggvnAls}$gG|30
zIchX*AD31rr)>zh)Xl~;DMrbGn%}k??DpbbELU}fKStj08dS;(xf70~wTzAQtkj${
zQr-=>h4$$mG06UGW6mACrNQoh)YCQJ7fX-Ms#l9|{-T^u5@3DS=Voqv)NsWWgjpAz
z>`Gx2fe@sD-n6W&#M8EVC%tyC;nUM<Uf{foWJM*M9Mxz*fNCH0lK|gqx3e4Rdi%a|
zjF++s&)||7^$JPi;M|nOtbTE#I`LEZAh}L@Re3O*C{B^q;BcrSmiTJUX-<5f7u}0e
zSy3My!>|ug_$Tv?_D|5-t?&7GZW&sfERGsZ=OG+o+Y|S@*wyD~<R`<9c0l=Z-eC>I
zqLCRZj5bQz?@c~-=n%%0yw}}z%#Xz+YV_b!j48Y;my-=2D&qKgwH8W*3Uf@cM$hoM
zn-dPR#2<O&eZ|GwgSo8eASpD_Ku=mQ22Ak9GZzGBk^cbbB4LQum>3#cbWL0?%8G!K
z&nD`KW=Vza`mR0b*U-4K^VkO0DNq3GTx`T3fi5EyptYgVUFAd^qJPo%>)PnPXDVKN
z?`5YA(=(Dev<PzCy9w6a8eyPOKN!)yP+&siv7NWB78TnBk}K+iBCZBZetR<qFh`!e
zscIaH!Y-_&JX2QaMBp0nk}}dydOf(+tC>H+m_+!M$@b23jC=&s>&eDz7)QAx-UWP|
zS8L?EPu-%UXy5_Y0jHdgv$8o~0A|S(I0bn<&|77}TPINme>Q)EG)0-j+i1C11nZ<b
zDk)jSQZ<qi>XF=cfo8Jp2Yr?UmbF|(Wy-F_b~x1GtHag~rHw7ucLp%NAG9^PD?h7c
z;T(wW7-QCKE~uiHmjwjKCSoUUUVbX3{&2)bO8=t1l<FzAbVq``e}?|a`n$%6$$mG+
z(0j7E%x(tv>Om5uEW1MKFc6DYy5!d?go*3cU36oL@)MY~&Z$Td`gaSYsg#`_sv7WY
zzT;qBLismirmzHu%DU$Ij0jr1YOdL;ruD5XWi)-5C?SGBStH|ymdJd)z>yQ*LHlMR
z44UH_UdMN=7Oy<D%*I*@@*g3!!<&B$rWjks7i14Ln~T_PH?(dC7Gz~isw&_0JD!gT
z-2@;Z5Wl`-?S-i#v7G>~tvKi6ymu)s=VF20mR&631qzCWADCz{l|6Ac8ew!a6qy+|
zp-j%`?QUdELyMb9C7os2rL?~U+)Aw<K7Ye-klmXXZj7g@uS&*R9{5yKghihS)wS(0
z*u9>0;iibSzh|;K<aJ@kfKnY!{82UzJRZeE#nhkOZS~`&E1Zjwal>VQkrG$T`DPk$
zJrS}IoO}n$x`ippfA05$rBxN?Q#2C2v)4n{7{@kOZ{1144pswcd?P&;#WnLE?+8m6
zi~y~0FdN^e)6TCm{NgI;8?D8IbLGJ<dz<y<2kJNQ=iQd{w;#bwQGJ~fHq%++degL{
zLEKIBYUX3ShMk+9Zw{ggo#*Rd7=r3QN)6m4YKx8we^^CTd&XQes@K6j7e&u$me83p
zJg2l1-$jP`8oG$D8`t^lJx(M2jZu7}V51%PTMNA5xR<g@R7uaZH$X>g#?7q|q!C#G
zCy{%bGOxG^=5$0B`xHtPPHrX>PHMY&CNXdJg()#$3{6pl@J%ohjlLnOsI7}kXXljI
zk4Y~?K6&fWN9t1k=)^RKapoX5-?KheD&%9?(IexjUWk?z@Okq^R{`~UUGR`7M0?;4
z!z{vm($W{T=mjTX!k0C2leE%c!fBx=<~I9H{Etx``t0ZjEN35smhkTp##xl+=}-iI
znWquc-^@Ji7IqnY=P;+cerlpDLnP;YvLg85q99TDy;vP1DwjZZnRknT$(mtvzMTmK
zD7uv1X)7I+!z|FGoM1}#x<J%HTRWOWaEDYXjUg6<D@_q8%Lk-+lhdrGAO;2u9RA?!
zi#81dut@>>*flT(6noU|JA(OQ4kay{6(w%<1yzmzo0+AKs=uoYpQx<2@BqsZOOs3Q
zrk+4s#`Hbyh2mannX-_x%Zus*p!jWHorXT0XqK>5I|=>eLEK9}KfOd#mnLQ)+SmWx
zQc>}`vgs`rf=j_NOyu9-)8a!#9+JRMU)l6>OkrGgIx~SQ8k2|$N#%W)yd{L4rFErf
z<gE8>hwYTor$dIg%&?kRwGU!7WTOiSV8bFi*n-GtE+$rY?H-6Yt~tEOs%(m&EKNe~
zPqpRm#NPC-7~k<#NK8IgZoAue?_fj)=T#80l$>kZE_8c$X6<j{nLchEwH&hQ*j@;t
znhvhDIfQ{+v(IOkLIRq?s;`ec=tkizVrJd^@(ofSm)(fm^ih7A&|3cL%=!dsw6Exy
z)&{373z*KcbSW;g4oA3D`wM1Gyo&iYtxk{R>k*7pO3!G~1cM%GKhRB5M<!`fNZKZD
zO2COp(~O(MXnnU@$1AKT$(*iDkySYSTIZb^1*CtBtA=#eTN(yt7F^6-d|h8&^*^ip
zF7mRoTIXy*DfhKsPaQLx7g{JLp+1%4Q7FGxSHB$rLgvh=2o_f2_$N!d(vMy%D9L^$
zog}H&oSDV|&r!?loyU2fRwZuG`9*$VwIpi$qT)U#J}^FBa>slTK`ut9Ys~}dVcUwG
zC`p3=jcHZ}jYJG?ME_*48u&s9PK*KsqllnfW*2&{(xJJQg}dO!WwCkp*-rtgGH`n4
zsc;b|`Ocx+kayb1H@Yh;I@p4ZH3zgAd_N}CSP`;1sj7}$FobKhzW5rB>cVFOqLV4)
zCUi;eu3aqeGLj+5U`oO={z+h?xsW1Qc*6I*VXu-v8&!{FZHy4WE;vo&V=8wPaPY+}
z?bU7acT)UEBXxxM5!CqxLA#4)+vm7hQkZj=rRn=si>+7p)Oa{jFaSt;v{jAs?wL&>
zx14uV?9o5u#5}l%%G1euy|Ja7>gmaH6XJ0h8LF+~FN$Ib%F0mYq4C{(+aY=p;#uD=
z_!j9``(T#eYAJiLxjmP5)-mmIC}OT@Y89DvTyl=Bvqs=^DB+MlyW%4==}mV@7~Pia
z%hcJ(C0Xp}@Tc%j`<kAnZ;O3r<|KZl$;M2Fb}rtTUWtuV$x|;j(gm}3KvDv&P8d@%
za~Yrb+|Oo&BruhE)GKMps;JW6e99*X46zm1bbU4=S*E9RJ=QWoCm2vpt;r@6onIIa
zg+dY2n!>pJ|Iu_7Ky@_D8pb8Kgy8OefZ*=V2LuTOcXxMpcXtmOf@^R+5ZoPtyW3s<
zs+%eb$rk6ZGt>R{JG1??>gX(<U5gHjg%wrr`2NQQJVJVy8^;7m5xPk+U##0kAz_av
zgfL4iX^*1#Q;g!bqWXs<BPAa{prW2|D_8N<MpL&GCzp;vOK3VSuSD#U3lgHSZ~z>1
zGGR3Z^GuurLba62ADF_ix_G$whRfeMQ9mtK9vx_RUH6-#8`TY><3(!SbfXp5%^*x_
zEprm+JJC#Y5)InCha-<oe|_x8p$awjk8)TqmyCVmPqV-;uX?nyGT?W0N6R8WDY|sT
z>d?_@p+FnUb_!-L15xV~d48J?Wma(Z?(tF*s6_B(Mwbb$QIMRhR9vXa_}w|2e&833
zUY=8A<%9FHau!cY#>Y3DLt!CpKuhYzf!R1@RpeB~E)q1AK^geSsMzYwQvI8B>UZuW
zjad7Hs2V6xj130|G!l;P(;rC|>OLOGD9mny8)G#bs+GWD(qLGprf>nN-QP1$V`RD4
z(4ukV(f>UQIZX$2SIUN?Jc%oI`om5bC$YX`Y{8Rxe*#R0O7GqlN>t=2z5pxEhzS#s
zkWd}|Hk-AWhtG(lO(g&oUg1A10DrzG0<Q?Yp%QbPtgVRrQ(xKGh_MvcmmFPSaPe*F
zvMUHTf=m>-Vhs-mcK`p+0w5A8EU@B(%NT5uvdM=hq2|B;tVmbjs+%wM9iZA`xCsoW
zgciT~Ho4UfXNE!slet7K8C+TOZ~41j@Zw<v^C^6BKr?zB+%o5H2OD+?aooM8+GG66
z%F#&_ou@V0{;91Wu*8Nd#cv~O!6MJICR^2xsha{I_iXX)pU`gGW%ihkrpQUl(U}l7
za{u%@{KmJK8@e%zuhVM=!tj1$(brEobE}Gd*(Z+%0h6WKWDvK~5c?B`C~@)A$>Qb@
zSy2)J<Yh|;2iuzP(-4THPcP#;KNVU}2R6%yiYm#~sDtdt)lyV6aVgM4N+C6BH$8D^
zLIPA#C@-CtrVWQ?(j|U6!E6M44aS|pW!%r2L-UmVl$34YU_kxTrlpWDqr+0?OuYyk
zrw&m}9{hp_SA29zrIR&yTxB0j54SMoPp8nAm@;xu<7LpAkdB-^Sh^tjquY3PsvOHG
z?~vB!uHuCmEMPd<mfEyOLVK)%Ak%Za*vhU$wYPjwI=S2Mz?|d78^?KW-sI~>P~lfq
z_c_6Uk8TP!E<L=g1T{%5Z|>57<=&`9!*<El`ZzVm8`t5`W8#4D+KC86njjp7A$8{X
zs}DEaq5ogpXscu^$#)e?(YJ28!eCs{6IO|xdkpK4dwml$Mfvp^0x%pZ1k%d=AN}_b
zmo?FJFLxB8niv_n1=0RUN+s4Tg6F^apNB%WK%skAOJ3IPqXdrh9aoPHeQN8b68rQ+
zwlMb8o5dP-J$yZP$IeRfQvDPX{fCrb0!F4|GRu;&BfyDFT!4&%)Pg6V)K;l<rYIzW
zJd7UV)OqVG?&8^Rbi*+hJ90btLo6&IHV}@O0{Ml4LVs4Wu6=wd?A}-tepDv;Ig1~@
zw>1D)N=QR{k!D3xRoAoIeVjt4PDStf{^#!?-gYSCOd?Yne+jtbLzzpyLWD|427>}c
zJ}6c#h40DFpMfHrDVP*#T@uvd8<@oF8&2-wNXbnvKdyUtYQL)gx1T=#4v`i4gj?ON
zuC4N4=4@;xe|-R@DTFiCgpYjgmI3P^E45E5TuJIw2`oL;{m?0P<!F7HFgAmXv3co#
zYTvQ1s69wDMbtOpP6-J+;8C-4`>VaqBeH!%r+%|$%mx?8He+FG4kHP@&<39KMprzI
z*w2(eDu<OhrrOS{%zaJ>2TR%=Gm?3F%O^-sIno&GHDiB<O~;hMMsSa2hvDzaL1F>s
zxbo8_rUJHOf8JLrPopV{Q|a*yFx>VshFJb_=36rqh7ro2jv6<e<&8GX42(n3v`!us
zk7y9C6%KszJ;qKjx)_OxVNRddhSU=kNa=}9gC~(nqFqhj@Qyp;RRB~<zHepw>sSsU
zfK!SYM@jUUvA%ZYdc^04B8Pl_C8N!;m-32Q;4e-A)fjN@=qPU@@*<BV?fz{%jNuK?
zf8e^u!r$!rf{c1+2BDOi7YBvs9)pB@zC^mb^cGhYqwwamYnOIoR<eIqw+*BieCRwc
zMK4Jhh6V=w=6+gq<U#J}&p(}<zYo{hcAygx7+vw)k<A@9A<f>VO9NXhr527E`zE@v
zlyZKDaRj3J>}eBfSb*@Vq;}SkIv7zlz*?^9!<7pU#E?S3p!lql<;xf9A5P|sF_M(M
z?PFKH>0{iq#99fXt$}cgTIILe9lk50SfdJZMu1d}OLx)FVYPwAN+QG<tdVWq%0dlP
zz{;cd`Wwcq@!!a?qsckE-h5F)NJP^_UdZ>b5ppxf;W+U;fitmmr9oFRS9W8LQQ|;B
zrP&^JxZ;VVmT;E@bP62;)=9&dUT)VDxwug5?{co;5f6ffa^;mC#b*h1MT7<GDdy5a
z>C!hN5yr#{e9q3wc8(HD44tIno}5K_QMALt-!7+`{xj+2bGJ`ys#U>ejfYB!Y`)h=
zp6@=%&nii?P3~4b(d{eMzj$Q>AE)IXsW{1gZyPN==Y=#4B4M1SPk64ipbI^O8)b_r
zfaYS1+jVJb3MOKt^K4(unkwdz(^?ld&mec-q{R3gTY6^Gd=4!=1tg_3GtOs^i3Mdj
zxTDzE(Ru8=>KeJ*sei%$iuJ$$jiqD8qJYE5!*|gJMaft#HK*}(V>t2#J~k1)&JmQ~
z1t#qB{3l+_%+=qw!ze%Z`lAkWq(Cdk8DY$XZio}g<}Q(W9iMDGZswh0;W=O!gnhbv
z(mLH-=4HVe`HO-vvryo5*#$%jLUJno8H1hRI%qNH8ICGZL*EebIX}P^#w`J+DiynI
z4Rp>aw|a-bE!_rXdZ8r$w4TpLJ(m31^h?|wmvn1tz2hX3)<inT%K?hfrOLhj+h-qW
zb6KgZV`QG$5t&Li(86aVFUQ!V-g?>2!P@?WCK}IZhu}pK>1#K5XF>EInRcm_U#>nk
zV?J4$VQ_pq-W8W%GC}%Nex7V7UpTJqdoOFqde|uw>UJ3@EeoD`gfoKKC>j1qN(5~F
zRAkA@mKxCa7*V;7v}|xF<Yl7f*Bo2d@n4-(hQm|kd;<le&LE!AQO}17(pxw={Zg`+
zlcM~oFG*^edsXY+*nFB?iKF`#c`j9W!W0tEh~G4>pRd;OrxTupq6Wv*K(aAm!ad!m
z=+(Jiq2f<NnrRP~%(Q4&*~iX}%nKeKKapWlT>sCkyBs8DJbRxL1j+;|+2I_QyPQvu
zASk#t{s%<Ny21T|1KWi$QV^z5ev}Mh6c5-CTc)C3UNJ8XcIltt*>2*1sJQt`YFlN}
zP-)0yGcL%>QF=)ZuYEC*P1C(}DcveZ0Yh8UTRpcBCX<^YQ<`*B0duEq3PS<&pb*LV
zFS$p`I3_v?)tr-=PUO@T6J%UVhaR%bLkc0=hL>W=IhT%@i4||2ild7{|DLnKpyrEK
z`br8J^`t7dINtb>5q=M>QQhQ}t7^JJMz*UzIw1j6PKdH*K(AmahcgjQoXKwf>3%GV
zTFPMKycYQg?$=`#>QKC|3Ew3BD{C_L!KF65LK<qTd?>TO<?x--z52?Wivl&pW?&>K
zqPSbk@L2~%ph4=<tA<6b{CaXjffF$`LTB#$2B|1sF|&=tm6b<C6Q&sQYSWSki_Rq3
zQ?T)sPyzZh7R<EuN?2|6c;)nBDEc8sjjv1E+$Gu^4B^PUro{t8{wb?(L(Do#0cd_l
z__@v_64CTxVKTa7iEtUV$vFa$c?YF}<1<C+9lDOuPS7O~Dia5tR1zVxBLPk0zYioM
z8U+SLu=_9&Y?E>Fakx2h9QFAN3i_6kC&A>61tm3BZ8d!{M37=%9Q7K~NS0ueh_th;
zgTJN>>|R?!bi+euG#+MG;rF-iCSw-Z5LhRdAS3|i-KgNdxG-uY=YS;Yq2j|qsbkEq
zW^^VNzXR%tx~e05<#iV1&B7|GFuH!FXM==|U5oP;v8YmPM*}ecsx330ILZN$ZM;cg
z@wsg#wZ!K*No!kMHvE_;k9bh25<EY@Z$@{@4&TAkIkI0_F*@^g{f+{1pmc~dCEJA2
z>MM-h<7VFbej)m_DzX_>hs`m!8E&;2p}jn++BHTXWs%iTw0UWjQ2kDiE}=BE!OH7s
zlbSIhkH<fI)8Jl=e^-P<`tiw%8FYcYQN8`iOH?LK=PE}a7yR#5vI@ByR7qCd@!3V*
zmI(b*2!v1_OIhKV*AGpaTKw1YW>O&;X=+E)O%;$~!<&3!WX0J=wby7|P{^@s+F`L%
z-e};e4Hx_9*iO9Eumxiou`|E*L_&~{plu@^ua%z(>9~H}{cg*)cJwcf#~N;^*-*R-
zx$bKr?tj7cHuhZw5uqp*Dub~WeS+<IHJ@?08f}J#c@5L7g>+6>)vm^9zpH!G1e>cM
zXmITaXJ7esyhvmV^43|g-5_*AJCa(d1$j;_%y@Dr6*C5T8XP=-zw&9auI6iJ<;R&b
zKeJPbx$}-*Lzk=&l;J&~FZJAsCJRQjGh}wF3?K@nh12*ngZKHjqRRhDDuiBe=d6+c
zCR6ZiK!1mBgtTs1F&jEESx<Q`YZf(?AKP+XKb3F?73zGamTuZ%7dn+_<2N2A)l2F1
z)eBtq!9`0E1TCCd7<=`_)pR!EbGZXqdEK;<=Cz<5k#NY3l93H#Z=CgK-$Fd2@q(eD
zL2J+O8AZOBm>85JPw^XKtHGNsPK;vQD$k&~x(hiS34lUzxxN$KyU0<BlvtR7k6Pj~
zvq&mS*(ROUJ{9lRw^cqUI)u$Aj?833O#Nw_@;qy_euNs~TNC-zbU()u^E$IGzak-A
zQZS1O%?zuY;Q~fD<<?Lnm2CeI$5K;kn&sh@16g|fna3MAyf<D&{gdbculHRw2)BQK
zYv9vQ{{TW|37k-C&rUq9r#4Hcb3J8HjH;xmI7a{;J7KK8s3PMR1nQ|KrX9=Zk%bNY
zIzvQGA1QR+Bo$`$B@J5XPZrUga)$IemqrdXhB#)?RX#kkx=-k5|8^=<BmYyEA77W2
z#uc~1L*^uSho(qNE;*ZmBjgV1qf}~mt-}3|pAJ>+U!An@-TpeY9NI7i<pG2uS&Cup
z?`h4Qes6`JJ}Yz5vOohGhk`d+K1b(KNgF`ShBPIvtS=z*u^lSSTMRj|T0<s8rGwi1
zdwqIm(nmy>Bc2A|p*BiA+#3(~ppywMM$8)@$P~6sG{W4a&k9D?xQ+N)ht-GO(k}A~
z4xmq|$dFLXI_V=+u<YEC-lOL<qQCW3#V89f&D)#Wm47=p=!OwqT?9_t%$Nt#JcYJZ
znw!BnJY%#r>+RJ2BvZrfvRk2~MOo06t4KKggCEuBGB>)M<AtZ~%=2lH3uR}NY#M7N
z)XwK>2O*K4R>5WoCc5g`A0}ER>GbmtxxA1KwgCF1XJ^6x&JBWf#k*uxMiVQ8UQs)^
zqwg8r(#ABXrL{JMm6J~|i55F1i-pyfp*3`FQ*TaC50Dzt*ABo!HwysNd*R~9?-s0~
zlIr)~rnmjM)*UwPQ3@?fJ2_tog?m+<P&)PYe)CK7>tmQ;Z?oPV)1$EU;k&GC_gg!3
zl(8;(FlK$z=&eAk^bc=3I76f~y}L+3VTKr9$(3BJ;$&rQPBhMkO;Ae+X^wDy0U4Xz
zb-HRi{8O72Q5H9~COtV*0IFEbu6xul98Moh(ATm92O~KICQR$AZRL5arMGI*hjjAa
z@yW|&zv^Tnn_8{V!kFRS+4e~|+r7Rnexa!$$0i(U5=8#XR4C3<Rxk|XWMuc_$=l-}
z*UQgjuO--gapDv(bkyK(XZG=bPfUesJR8&cdBT}Vw!0hr5d1KGzLxX7;5c-rMy1xj
zC<xBqI5@9reg8dXqw=Ra6Gd6cO}xA%(+Oh`<9VA(Uz?|wQFA<=W4rHUy+y~gm0nF?
zM15}{LHU)DykN5em~s{?jq>s<P5rodF_N!Zvjbax=2A=<A)_7EEE!ax$17<(YgKxW
zf0%C1G-(&6_qvS&*aKvwMjQJXh4>y{WH<xa1{2nl#J;P31e|zkNh*}sULJD-useq=
z-A~7^#bn2#L2)rkyzoo~u$0hMn)oIE^AfnUQe`(*e<r_x!EmV_>}`m^LuJqA{ynR$
z1>ln9Hg~_Cdhc-gQn0-CrlHJN(hk-T7GSCzHTxvZWGI%=t*LRQq7b~V)kz}c4273u
zN&KwFZRM?PRl#aPCWy^_-C<8B&IMK8?fCRO3#%<SpAO#X;kp|Rt!ftw1r><9(j}%F
zV^}d2T!;U{rD=2+jYg??dVtS(qMIKPH6|F3iPyp5m6Q6ea+G3a9$jDo^s40<WpqjI
z^NB*xq@8vecy6q02yYwnp)(=$b^G_-GP-w{_NPj2!2=IRC61DMx?9g+x=7oQ5$U@+
z=uaw}8}y|cr<_>6N^F6K4V{&TUWaAaLP55jhu$5-0to;cm+g9h{Z6XzAoc|>%ZLAf
zL?|uO=bKs%Tf9e{!_j0zZt)Wy<v7LfP__9Wr_jZ5^d0+WsnxGRLhq|gt2=i<H754<
zphIgYetBlQ;##198;Lz;GaRVu4<Z!w#yubea2Q6sIi=W{OV_Jfa%yVhrq{Riqvvxj
zYeEMm5Vp}WJ}NcuD;_B<d?JlDD<sX?3-)<z`ASyHGVdoS@NX_T^W%<_n^`yxsWnL7
zTx1Ozd8{giC=iG*bG3Ly*-iXxJf2j+2LLCFf}y7j(L9Gp#_(o%m<{?Y_A9(p83kvA
z20S#3S@`x$X2=d=md=2A#RNSD#QgHxQ*TEnJ;nwbzBP%-L6|p*93OATUAw`RyNDzi
zV#-Wk^nG5hSNDaQA$t+jU`oMSI$LGZzt6mJBIps$`-@><lS1xjL-;#>k>J@fr2k~{
zub3$K(|$a-(;L;zqx0qIaAS&yS{^@VWO=4Aj3#KFkSEUQYH6ffpRQ2CMJ?h}>Yqe6
ztUf$$IKv4&RW4i}`MruU(2cck+~Z98w(O|YdVZFa$JCW~ZK1b%#Gy&ewjkecjT^bH
z*|km>LLj>cs9Nt;&AY@!CFG7TeEU_7VVG}$-vw~lot{sdoyM4&?LR(Gd|kLu6}nZ0
z;D5DHQW}bYtZ&$a^zl9<jaBVWwTS?c&IR|FVt%&B$kIz&X({0jB~tmNos=YMGG@y`
zSM{g+c_6O@1b=BNK&YVJTSubf3L&5lKbsU2Gc9s-2Qe8JZW_wPPAOmKi_Au`jB=!T
zspt3feyrwCD=4*AmA0XkIK&5v3q?jQTvV7`WLw;z7P~)we<r#7>V|qk_P1%=Z8W=V
zTrKqg$%|*w{;jdKb)?JZ8E60WRjbD%cB}JTBHR!OJwY&$lW9ZPjt_TPchq6DKJ<L7
z#nW60Zy7(bZ;5qgT6b_OMX9MmcocI-fmFKq6J%QVQ6QP2MRZX;BuOcueXh@hf$qJL
zP!Gku^Vc}`K@=?NZKJTyti`v3Qj{Xc!d~cxIfY|}&<>dba+S$qY1s{;qr5DweF-A7
z@y;VoMKS0!@{&fr+bJ7kL_Kdhk*zw#8+B>x7M!fuieVetILvWWqetBxd$0GGSu0NI
zvA-r@IQcS1v5tGdT{n;Xe*zDYxx+vnh$$y<8@m|LCleX}eOJ1fr_;XZ%5ck8H!u6+
zt}-)es>wK5OegmRO6@DTq|yn$;pVTrH%5}jsH%%Wr$Ib3`^$kZ_fX0p4k&Gy=%NB-
zaZ8O`I58`m4GTS=+d1H@T@2UyFiF=d(Kj^-H8meIV$n!*I>|XUgp2j-aw&mGyXPi5
zXb=oZJVi0diK<R-AP6-P7dS@uus9?*sL;IXoj2(ueG>rmREs))nLO{ujrx^}%S+EC
za~#>Kl9=;}pegRV-a|ro`nJUqM*mJ7hXEE>U_}Nw`S=TUQ^O+zsQR+8_RmlNSw|r0
zA8XEIl(C{$3RJEv?&EM%!Sce;OZV~K39qNx1(ykH&Z0_&pTDgY%Qy@{M!%}0agB^|
zxInJVH^|4CAUK9lJa*iezn<JWL*^Vu@Wu05nTzPw_&<N!j2ZiBG0+SS<J{jr`x{KV
z^PO^{V;sGHbt7sEtE?<nk_%l3nh<z}&{+L}jEy0ma<uNuYr>XsbBp)a$<;mHv{alM
z2GJ>w=<hOTJ@^ROqdY>q{KrHiPxjxazlI=9Lb81q>si`1UvYoo<oK9H4`~9`>R(Hx
zOw)le9QVZ3xeopoU$^czp2^y3*L=n$S{}BeJXUvz(~xyQm}}+*Xyr8UPl(?ZT%v)`
zyhQSv=Bn>~evjaOT#r72b^qF9-)FDdCj>CSHu%fVk$dCg&mMd4_wf3|b;%_CM01ON
zA?i5q{m((hgej7Z*df6vOue`3<6hdmYvcra9mUwzkI4cE)m*uA0mB2V{ye`Oa1e&?
z*})sroH+?5v_rCYe-VZ^h~tc}r?;jr!hj=C>=e$$<Wh36*JbOi3_<zP=AXChoH<iX
zrMXYgeZow(7(YjWfpKrR**t9aNfBmJVR5`K$ki-yMvgt_mfXlOF#{;6aGpFONXhu1
zq$?>DW>|^ifF#)I)ME=na?v$_7xV>PWBSp5sQC78q1h31(W@EjX`3-?je?!M=IF5-
z35yx#Y2`FOtfR%sv4en8kT;AeCdLK`6dC~xKN==XN5}*xEB!P<kVRF$@Q?6OHSE1%
z<Ut+~&kKk(GLmjPY*!~q$R#>~hJQ=ur(yO@FPaZJ#+cGef0YwK+80aks>^~7(6exk
z|8YcV|0FDX4M`*LkX<<#-4V{8gV?!=Tyks3_eAXj;j-#+kmG$Qr&nl^xnAcv|J}^t
zto6Y0Zpq9E__)s5c;x5gjC^;;oIm*_{#I0kt9hLnXxOl>iBdR~y3$l+^^2ljPZS2a
zI?SA+*VA{)Y`q6^Qee<YljJEH_Y+E31rL#!!X_(ImbfGW1-YtL*QDP|ieFs8AeVj>
zWvV!oDXf(ZeHG1_aT7~Lo>Osq0^`1?lG#?>x0i=F!7Q4SOSe#E`7xni3z^DV72VxV
zSO<h?0#(a~Yu|AQfYw?M+V}gWPdAqxgs{JEUia9TqexI#vwCR+9lnA)Z|Jl~1)|TE
zPk&T(0O&+U=4g<R?f}21wWKWU<J0?rfH&T}*lZ*mNkDeuS|qGpIL^;3R*LGq8#6l}
z0=@2Q^tA4WlI-VyVN_yS{Tz7bl2e>7y(_+H?dPw&8_$2fT!h=XJKq(uIt^I&0c^;Z
zGVeVry**f#H^@oPn<5@<m78T<MwLBJy-WD>?X&jteOMNENQi?Rul+xQa4oM4VGpxq
zf&(RKx}$hW9z+<Mc0zgC+8?Z2r!6cJr*DWFJDYY9{m&sJPpGMZ<8cuRn<72jJ!!}t
zXGG_l=yX;2*Z$vv`EBb25y|FJgJrOuXqoqRML*21`GETo?tebTl~&8Hp}*C%s;FEw
z>4*=BFrjN?75<tDI(?nQi;nQV17!PbRG$1EGRS%SOEAqr_?kqvrYDOWOJ)Fk7bsNd
zrKlPDB80^o5`+Ot0P8)shDd+byb&ajYzs#Kh`JQb>#po9eve~GDDDJ7U`HPBJSX3-
zGg$%0m&q)@EEeC(C@(ho)6j%SfL?%j4!v>jvk`Vrq})(;|L1pR84>>#zV-@iE_}ac
z;JtHwpSq}+RB|J;8jZ|}H~4vdMha8M_i?YK-e}qShFD}k?B56Qs`l-u)|0{n<OqLw
zWz5U8YhJx)?-YM*+Y(z<+r^tB!HFugg{ZVy2mmr;83h|*FkYB_V!|)B(EQR@?7BO<
zo0Z)EV6|NQL3#$;NLyrOa_^njJo5<x0{CuSQcS5-r5OOrHH<L>=&Z>eUb$wdpwg&{
zGgzo1OYs8&$s&w<+tmp#p$lE$%a4c!5zaid?eLpou*dXQhN=y0+oHp9b@SC_oFAd|
zG!j*1b#T1UZH3YvD9!`9l=ES6yQpmR7{=M*({#yisBQ0b>xS~rzqay=41!H?Oc4Yn
zXeB@Yo=@uF%ozvTllv5rf=VzNCVQQeo_Z{Bq5WRW&bM~*+5;TCPHvysxz+|5^d4o2
zFT<t&o0BnbdE3u=?57cgoRYG|2@Cqt`A;I4F;g7t<7Xn#2_JZ$3L=B*z4y*-BK|!(
zrL9bWrT{dN9m4HX^LuUXICH6m?om*goyl5%Qm^TL^Ax;^=-vWGpX{S4JAuxSP4oKT
z`Rn@mZN*U}0vsbZRnIEl<E!7JMh5`Qfk^iP0iX^e<h-|=vWj1e>dTTw$Rw}-qz)yY
zBP1VAo8Li7trz6cJ)QUKax<>0F5pgiv+s*mGaEGC*IbvoG(PMUTq5P+^#(TTX9z@^
zheSx!zXLMLB5@8mE2=<-`Ie=N2zgJ7QZ2Ws1QiO!m$68BPmH-Y3UP9wbKH9EfZ>+;
zy-&bxeFUIE!j3;5$gy-LhY7uGRp>fT1WK7NxO|{}yswh-PR@rO5Awh>z#ClyQ<V<0
z8cHpc{#$&%?0O&3=zdabOD88Jx!HujemQ^C6;;EEnR11!Pc>(}Gsk&bKXeT3(R-WI
z`_Pj5(ObYY+YUrxCdHLv5MAxd%nh?J=Pb?UAtqBV|Cv>5rH|qhxg8y0kpm^692-cU
ze1A4+o6A5y>k&hvLc)-g9EtX@oWm67{lJB#f!=#*_cp<DvApmmCHOefz4JkG`<U^i
zS+?74$2B{#d}sQ=V>kRj+v$BgusfX4HT%2g+d%KNuF7p_JTebipp)4J%ErY+$HM^2
zJ`mftfRBP26adfz-+Iz8R(`zQk)U#c(msk1{_WM;Fx~UgoqSrqzx3-t)g$jzuZBmD
zD(2UWfZsnGmjVnuiFeQRJs25=Ujh?Ukd-1ls#khicwQfJHecnl1HW0Hr@liNF_s;8
zFA2L)?^xR2*o?!-^>AP@CJ9Ee11<Uf;&YEaCFI`xg|g-f(MyNYl!JHYGKAwVUdt-l
zI3alpSg~!70jMCUcCYhxdm?cj<36;OFc_2TTswcu1Y*#<H=#e$(sqY+pEQb_t@(AA
zJ6OT*xmnZ7<0dI@)$c8=`zi|JWo)4f2bE~Mrq(Cs1BkimFhulnRQpDM{xlV)LhQ2G
zb;&$*j`%Wv{`Tqq$E`Vp@6GSpoG`X~pIzcye~9!|nbg}3BVYGoZfS7-iahju9ok7y
zdNdL0*j7<T@em401${;t#l0}ACX$WzGz8fD?D}EP@9|UawNfq^t>ftHOPQXh0NDF#
z=b{HW>OSGXcgXEBGtU<V-1XA__(t-$;jvTuVwUUKn}kcS`CsL4@N=x+Xk6$1Z|^a*
zj_`x`MTQ5v?iVNi=dA7kXkX~vpKD`nts~dD&p&jpGZ!wH()^I#SLBj>4rA=RDK?x(
zs9tSn_|A8Ugx;?^PE1&K03_8vtcm|W3-C|wp%%d6=`)e|fW00@@?K;<cvZQ#Kg}Ha
z-rK^29^Cx;KJpE~@#_V@{9!rK4K?tguWg>KH!9BT-TJhQfyUzmLuf3EQ{5m^{Bz6n
z*IycpQMTeT#g|)Xw2!^n{{M`|NrkMW*X`)Koj*m~lA)kS$H?eM;837eSmKvz{tYBG
zM4qsg%u$`6P|cNj_f1C~)oMq#X!x(D<bIx~jnLz2kHs$o7<ioYFq5J57wkS2DZU%*
za8QKnKntGHZfTYs^=>)NFCELul6ymBojXdmNer<~3X(?zy_fAT&%;5OWbDy641dH7
zG|$NAC&Z@AaINrSjtJS9opqwB`Rtc?W+g@CtsJ6{8__w6XI(IRRjEMaoG>8qAA)Jg
zp3Jy1Tbnv^L{{d=Dhn`Xi|O7(Tm%ktY8sanO1_mW*J8vs=*(^Z$f<W^?22)M8P;84
zxz?Oj?HXZPw%Q1|-(RjDYtEmT|HmyoNYcm(Yu3ATwQV~rg++VDrQq7IXXpCi2Cd}1
zjn(?EFe*`*XS`NJW{ykV;C8;bBL&EV4kGDohiwF{ckFrao2_6U=7Zep|Dr*wt}8Y#
zBP4GlqCVI%7}$N-QFKJu9lKO4GUt8u1g}(qSnxQa1m!ledmejvx+Av>dP)l}{4WJ)
z@S8vYl0IxR<Rzg&#X=F;JZtZs1j4j7<P)|nm70H&b4*S5ohQKL^wzyfOrz{;uL#!>
zgl8WN?Vz<XFPEgqq-Kgm|Iwe7`f4CfWLY@@N{TST;q(57yM*BjJE&vw+nEQu>W+t)
z6`!B;VD%!nrZ<M#x+!3!Je1XFJbk7L|Bo*{56o2AS&&Ejx8>2Tz$+?~<2oPsMWM=y
z81!}=D`dOJ9IjgvMwdqZq6NRVbMEtwYu3SwN3dVld-b$=FrtM`epZ(~QS+L!QRTrP
z`Z6lRACit+rk4w}cUBM|4*zvxnbR1`0+!(}`0Yl?eYO>?k7sDb_kLniAgPl4ftjee
zn1qP`?;z#*^<SYidmBqWVu0zn&k1F!&<AwR9r*|`OpR-MtD&!%?ho5vZ1?AnZOIpt
z&-9H6>q8nm?P!RND4KPjEr4~}4QANJyYt#8BPNuj3PE37T-*en=US-;|KOo6acnTD
zPsQz#)PjtbFw9X24Ia+gcv83RI6c1vk@ixMIi6Pg^_NWw<=MU8*&SfSxr^sPag)AG
zr}zz;2|X{pl|`E0E)m%7*2noUeyF3RW;;Dup2I}@>-pyHppVHxHn4YK*A)U@b5~_S
z|4h~p9wSDECAtyT@yKMiwV$Q>&E*`zX@+Ul>zs`EDn85}sDmh`cghZ*+<ebZ*BkV;
zo-&&|Pi=hksK~tms@CJ`oW^xS+QbaC9?+r~;s;=TIO_~y<voZ4Es#YGqd#C*yKUfC
zA~dhvq6eCdcSF<q0U0eO2g3YO<@+2E3}I;eDMHY&O$Of|;7n&qaz*<%iHg?#Bv=Q9
zoM{_*7vvI7G)O&teEz<rPW)C5Ps7zv(r9a<3*m&}C+#o_^>C~A?$Wh$bzl!()7x@e
zzww>qJLf09U-avD6MA?2_-2xOCpyFkIj(ch+fW_@k&3R>ng6kn=AyAox20|)t=pMl
z<9++f#iUyod)eu_)}XWC_0IBVW6&a2qPjll^9TYxB(8zH?MmRtt6G}M_hfqpQkIvj
zc`TT}l+DJ>w1wit_Xy!I2o_|I%d_eK%rE{c8z&`se;9haBzeDdE-ty*)X?>yI``Um
zwx7IAOzQ|X<KMx1r3*%P|9FPc@9l3|SF#*)tMQ5z|4*bnpNy=IW1aJDmbYydz;Fr@
zdI<XA2!!Rg?w7Id@gbqNOW$lbI;_EuuFV@S6B`g0GeWO39dEa_2RvYp$v;{QAjuDM
z#1%pJB|)~=S?dq`Uf2&)5MxWqcUKDMuMBi2)~4>!ChyK+VxLB^xg))6-qC;hNqvN-
zMvXFK)SPEw>3=n09lw4P>i$Y8WeR}=)ig53x+^_^J*x|y-~&f}Dyg+&L2BJ~s9?8!
zR8_nGQF>JAz0#Ax*ml3V5(9twbw0d6xRl^k!p9qUxwX;T;Uo5QcmlLU`^kSgnsp-m
z2&lxaTSBG2yE~6|?+13QZ%Ox;QQZ$whhB4rR8h=V>1O<C8Xeo8|0g8pYJFYsg?t5;
zu9#nIzBo+s9M}Wa%*9lgU~*p=n$u)b7wlb5=ad~|v)%`6*L|No4{`jt+V9!#6{ptc
z)M^GTj%3x(%*iPUcsh2H_gJKTJw{@}o9@+W<0a%huk#-DS0(StyHI;JFZ&I3=Q#2J
zTwy+ka(xuSC+YgVuj+;z)R9d2;ph?58n19xv!tZ>%0?*Uw1K2e;mL<FsDDdy8^+G#
zeCBzN3BwDo-mo3dj<*91fcp0Q_k!=d=ctd+V#FV2@PR$(E9*(9-Xz0*)D}7^9_A<i
zvy|7z8z271euiDQvp*peQUL~z8#&#3%VR?-b^A!(s1JbW`)Wcx<Z1)n`+_pmWMc|+
z0_T#)1$V!{+Bxf(?eCyLO?7!OFcUlBx7A#CC)sU1$*~+9*LDwHG;^2B<y;;NnJ?|Q
zPZIlmj@_{zd;JB0>$8U9{d&7_)#~@l<Vz=*J1#e9GJoI-y!a&~O*eBanNHOma4`>8
zn(an)_kjT}wU{a0924vhdfq`N%hG54Bof9Q|Jg!}6Oo0+JYc^tNnZ%x{}7lT5ERpy
zR6%`c8O7>w@WE|85qT0|OdX{<9c1?2%`54ORPqtZMo-{a1pPLa2p5KCWo<1@e7vtJ
z*B?619{6cK85q?Kfb)yzJ;xC<nZq~p+e$-pi#621eI&JCl8iWROlaCXIWrT#aGG|y
zoW>_PD~q9Os5(gh`v{DvC^71S$x4$=!xw2=TjqlQ5Efhz!}?UTSO`&n0YGF351mqr
z#FUf)!0DtEqa}{p;l^ogeZ1j2G=9t>M?@Ie#_z>0udb%zThZ1(IyxHuWaeYHN)KFz
z2E*zt2quGJOynIOOJgycDOI2gIK4eyaN3Tb88wcyS(%-mSNZXSkxPQ!s=T;3`0rm8
z+xAmVx`?u4-jb3Mx@#tMgyBc^A3x+}WDvLP({0z<s4zWcn9P5F-)nO@HjY;vrvR~x
zj*ddmGBfX|X8Fn`Q7age3eOd}>ag&X-|eeBTfN5nS_#3BFI7-~>U!KK=rL;LiKX!B
z7Ad4ju&6T}_~g-D34I9bgvA)H84bMt2fcbYf#UMfyWeo#RD)gKSw{|!y%s(`uT7IE
zz+XwQ3>sF4E$V$MOqf)-Khg-7dehw}+rKvmUUj9CW+mzkh4S|A&L#Kv@84%<XX`9x
zs2Rr!eV*?c&8J1X{Qi(d4ej(Bf|y%2p&T6@ce5i044seX6}rNp3>CkpN+>I1#Gg1(
zd74CsS1+IcNBJV*sHJcMToIFH*OdV<TM-my(gnOYJ3Vg`T|142Vxvif1SRqWb*_DT
z17WdoaJIQ)n+<4M_Yq1Let`uU7WfPppaLN3HRX>sqIO)6Yt7P4y?jWGa0J)#XwiF(
z;M$qjuCF`0Li|5R3~wV(+cDxI|DBrhFSRf>Gb@8B@I?$CF#P#`GzKHLlw>*QLWJ@~
z>N_;GQf~I2Go5Bzok*u@9Jl1;WT0amn&XO1vj<{HzbC~!3W%g1Hzg%SSw%(E)s+)?
z7Fns@&3_e*J64PoT_3>JIqF9G+7|YhzCPW;KtTfUFbEzEPWsU1LI*wM%XXV_9h7E+
zG2-XXpBI}Zrlm-jY@C5RW7lyRicB2og}tM!r8S!Q)ymlD06=(8HJViQ=h9s=C(!zG
zC~(8bUm~+T*o7XG%RNgtjvZWkh}aT1cPp+&?x=dBFWfds4h#%5TFjVs>r%!qgOyM9
z$Vul*Dl6%Ay-vTbdtWQiB~_1ZKv}P~r4+^T9{C@rcWa;Qi1&un9Dmc$0C?KJ^y+TY
zjbaVXR$J+OH2Qfnh@%Cj-Y%s5HHHsIjYDD;Iq*$r_&ly7Xxwk6rSOD&JNR{VqEZ>Z
z{{fK5noZVtsU{%%P2pU>HxJLdlL(s?M(w8P@4rmzjYlvcLQ#oEfsO)Val`^D^Xv|~
z9txumEV>=(-M~gk2k>D1zqZMWvfNQ8uz(bB34}Y~Xgu83`YDtt=Gg0`<?y*R08`po
z=b#@;2_|1IjhRlOY@3fm!3cP6o&V*cFWd8|A{ziW0KW!6;c^3y((k%domfb-Ef0VX
zd!LcnxU(I$LcaX1*5=}MJ&^|X!lqqv9CijlZfC2><icA72tpF1I8pVkr&_=h=29jm
zC97UII5>`nC24GqyTi%1Jq?U+F9$+S@2~e`ZZa`q<*ttrG`XAmS$3(Mf`q8h7bm~K
zi9qm)^SNJhQ{Co(vJ^$iFDwiM`0q2oi&j=vE&@<^KDRSU+?yL@mi802!*lT><u=#T
zQQ+?Fp2xZW%YDf&VD`D>#)afi*vT-<-OAvzOFU@<x>2a8sSS)O32r4P!RK25J`#()
zU*KM`1%6q!<0lCT32z{++?;RZ@i^^6u$qiw_he~9!=n~AHYV40zh$m>dimA*xag#P
z(RB|101_IL1;DR_(zQOh1`Czxb^Er|$mI$8e4vZ-{N8v9{tF)+m9_19H2oOJr@K>z
zTGh>7coO5mXuL2>arWU&DT<`oQGKD7ny9L}Z`%%wtxhz0-fgMCjMdNNt=Y*p&|d<1
z3-(#5*OYK_-9Nu}0OtI^w%7~=$w=eye@VbjDiHC=#6r<g!w^G6N4kVIX1?@sD0Dq+
zBhA;|7jtL#OSY+&DgJ9yBt?(!rE^gws~i*e|Hjy5mg9oct3Q?GBFAxXLxqU|5B;5|
z?09x=ZshzIV-yC3<nU&xRt6pCI~xB5M9ndq%h5L=EfmBLdCS7d*>FtsOrbE2A(49_
zmv|BmB?&j?(7x$I=*)~vP68ASIRF1d6OW%a@1HwMXIlO>g_`iRh@9geOy<ZvF4JfH
zhY_19eN*1M#kKA(7Hd<h+lc_fe^_^PWGO2rS8tQGpy!uUZ!`pT0YIEbQ8dB10(+W~
z=X7**K#uz5d24xfbyfF+rUbDPft4?oVh`!7LXmc}EuEq#Fx6hx0Gn_EP;VmKg!{m*
zH&YZ26FGj&sBl4*6&v?a8}sAi<A-D$ZF%2-`7|*2&6-ynQ**Tzp)s<DIpSW5aHG@9
zEtAWRIWG>Yd9m4!1eg;*d|Cc&)e019yVr2$b%y^k?>ms$m5Va<1;T@iH&FEB%^|FU
zOwDmiBe34jx}5M{*vJw)>(fK@;N`K#LIBi|)63<l`CmQ~X_Q8u*ggPmP0PYEbZ|Wl
zWMDXCT+)MbX<Nv@U7x5crSXyA5%jnIgaBc>&;lO-s_o4J#PlR3!2~A(LR5CkS=#-{
zoUu?80#@Xpv$b}1*Aq=|L6X|8C;9~GV1UaASv-t5xZ{A&WjhXtWGOnL7TYym`I+R-
zMwpos{i#Yd1|D1T0_HQZ1sC8p;&Ry%BH=LV10goOR0g=wcwDyGKMv&Aq&RSldjF)+
z(XRS$unwPQ%K5s1_fM-=hWrCwFBATKEkAyc$La&siQxE$i$czDR8&<ZH3@Ui2ywmm
z3IVZ`gM%X<{*Uv-X((14#?51dmcz)V576QSm)#V()Pl{eW{6|XC1-W@2Lv9%`Y$uI
z<9QFZKee5A_&~ZE;7|ZC+$;yOK=9gKrl&--S=s;J4N&~qPA;84dCJoLqB6M#P<1(#
zrMHK@B8GOpsjL6(_VW`hKu1AAf&aR0nXhDGMi5FC<td&Z2h57bEvxyTknvt${G}<$
zYxOAmoIz5UGJ2PI0v?b;uK}j<y?#7@oxav2twKk=DTq!k9kY~-1TO$1cQ}1Li~SCr
zXeHy*xd2i&(L(fM_U(GTTX|gBc$2iTVfvt%U++(RQ&Usc)l~#C5TNU57#Kcq>;r(o
zMwL6;9pYE=ME3jl?+5R$WFLNEU`Cox7Y_K2Fn{H#WkW*k7vx?oNRTF`wjCK@c(7D-
zs%bytZL-@?q{Nh=gZOO3=d>?x+i@w`x`8j?$pP5)dX(+YN`6mR!15{bo(ClQK|vVt
z_6SWLKWQJGfX6>>R5XDbbU@`?Ve@dr0JTprAPT~lewi8J!cj;@>;DNxymGqV@Dl=R
z$}~5Yn)Z#2Lu)VPbh1cforOK1bINwAw(S-`1dkU{{2pDo+TF5#BkTc(M`(+F-duCc
zVKpBONEtaVnspN8#h=SU%IPJQTrg0%M={_7qme`^`E*uD{||34J3AYj&+<FY+}zwp
zsx*q_3H=D1RW)t0%Q{<b5k7{jxVs5b#E^-W@SF8x`1ttcKYto8)aW3S2xg-K9qv1x
z4ZG>P^kz5!9R%R9hVXcuX#qMAfD20IJMT!Euhxdv^*;0i8uMhZnT#5YjPheI#${(;
zL{wE(6|`vLpyeN@hiW*t0zpHO$AomH$w$t@f^LY=Nq!q097_zqwLap#^U-VsAUL6?
zy8Ygh4bMJs&2fTH;etMQ=XR;Fz#3H6)EuTO%T83`Mwlv+LQ}8#A#=XodHMVbq(}f9
z#VF(jOzrLKM__i#e-hI?p#pTr;QI^s{pH`~7-^7Lh@K=iWsn}>Y)ti8m_!tgvTg^}
z7A<5=lu^*(i~}Lk&>RF&9*e<0XA%NV>v*JEI6B0f8;SOUf&jn;=lGp@>BZSY;`aiI
z`ofYr<aWMJbU{yn(wARePFrwa<8xrD<$Z9F=Qmj_lVrYB2%A}ofecrVM*gp2J&gXG
zG}>2P9Vh;0jq17bw|@xpy+?CZRM#>KE*bVtI`(V3l6_4!Kiz|cHHR85-kN;=Wom#d
zhnPNe=hyj#`7JaPp93EtxkB~kF9eA#uqp&40!;JUISsl>TJn>z1fQ~gpHHsA`WXW|
z4)pZ&5W`@w*=)Ho9YM{<NCS$A(|R%f&tFHvFk%5uR?~4@4*^4YUC%=*Sy@?HMn+=*
z<lSiNm2AuaNIYj98qA0~lQA6N|1KWAKZa$Dxn17&;m3jrpj;u6Q&Uy6v}AR3h=9Zj
zz=oZG!vkg3m<*iiAiH(`?MrSzARGXDKXL}2iRlrfSk@B01;K4||3;evA}%h!J8R7j
zx?&xM&O!+gWsq@Mh2bPPhMre}jVD<V5lG+eyuGosk3$K={n;DmG%Bh?Lhp}Ch4paq
z-6RXQmCjy78)JM*Kl-S8InWoXt<2l*ptbGHF68`cj%R1*5>ir7)Ct=QAek7#mY20e
zWAKgolOS0zkd>pLtyfYVi?(zAOA?oXtI(KZTnw}<Y;ro7#QyRHD@Ky%`Q(?yN)r<i
zud_ZN41hUh)4UdHv-06f@L%*mn`csEZC$!&Z}o-!PcrGMA41|?u5SSPZaf`CAyL@a
z(UA?*kRxuJ!gG<8*@hhWu<?wH4kYB7RIxDwhIA2rU#GXZpc6v?O0a_wF|^|1VR_Pn
z^=eY5%f<Y73(-9`CT0K&$d|l;YSG|bBjS!E@-u__e?UdWid$^d{A2J3FAnsVJX>@(
zYv8U+0c92t;T@4B2F!JJbG-)xgg>I{!t|d%qLEQikapG*O22`~E*VG61Z@^-zzU4b
zCnO{~28JR)!vVy#(nUTDb1X#ider-qr6lDd$r9F+U~zpQbbiEAp!rcl`1W%?uiv4_
z*4yi|s<N_(g++zDO2ji}-scf6e_$g;aWW5hF(B`yniwppC$^qYYbJ%gRD6VSu0)w^
z=z~b6l|LNv$ZI_gOoa*D1PomDmUHx(U%{ny`N=T^Tn#SA>YTQ#(u_DoO-<@Jj<2V7
zenrX@wNgoyybkA`_l^rXZqr>pUaqIh0{~fkWMrfOX#9*!#GBmF!LQY1O&9mYuk4xM
zl1<Kp>bPyPdBax@2<p2&k+;8UYGO+jvPR#rIMrX7Z_Vz7p`FtE1{5$T>`TnORAl9E
zU_aXvh5@zS2c%usWAF20JD?!14uszER+^UD-B>=>xNk(M#mB1wg8vdo|07PXG#fy#
zqPvr&K0tKRAVEU{s-XuOjlX9`NeHE<x_O=6uNkN!tyfA-#!~5Il=@hWRa9|ON<pGf
z0m7pn<x@sH?5+ST6l4Wv;nP4yusU2A``;4ixouS!mh@awbrT`7LxMHOF(&g&Ne2gz
zSST`N2YDdSVk?wNE_GO7$1Klh5ZHAP0mS~{Yz3Cha-)pFUgUK}hz&6MM0~DO4JLAr
z&#6q>A8B~G(LyYX)5iU3gvMg6&E?aIT&L}CTCbab(=PzjO?1H}*Yk)jV;HX1+-R;+
zjn46=I~l4SK$Fof!S?|E0R&e9M0CKZzs27|2_zs7qSlN7fxc56%VuR1eAuS=;SjK}
zq**FLOk{pd0uOO00MTl(!DLKiX&#W<@-sjjq;^z{Y@KejP;IhaBKKKvMjTbEQlBnU
zB6;Ki4b5%C_99P`2IXPBOQg~SkwDM6M{GtP^QMQAWhj?gU^Zt56B>*owY9b(4{N4U
zRYkyx`^zxqTDlCQwQ|-Yg2%>^5J`Ixt{=S@Ba21b`byTbj@zzs`CYYsz@eet@%D={
zb5dBME6i}<t$yxiPXS@vvta5U8tulE49#AhN%hSk6Vk+j;j7erCL5p2HaBe=YudJD
zfdzx&{_Wmg9cTU?*@hLD5;JO>-8s8_2-#}75El5?0-08$(=L0L*w!0l+@0vQ?ja@E
zr(UfklR>jZ>vF!H!=&A@`!7w;mo9^_)MAvF!El^Z%}8s8L-6gv2+%TNpJOnt;KL<4
zANQ}}HCw%5;qYA<WA|HHQe%|9JB#^(>1K57zs3N{AP<Nw8iIT&u=PNt$a4pjjH!@A
zK{;6PLFbh}u+Kx5-v<t^F9Bon>aZ-&aD`L5$$INA;*M(ZOhZG13Ma87x&JaiY5(-;
zQ@A*Juqc_gVBXIcKZo_GZz-u>U==5Oh`tNeZ{PYCj=N3A)8#s>92j3RLrcAOEOu^T
zoY*a@`%r;^XU`oiDXPMPd-HT{iC>Q!j5VXeZ7pi6fN}>PBm!t!wb9DrENS1UdkM;b
z*~c0<89@Kt-rm9liHuZc2>S54tvbL52n#n*%)vYX5v}+Y1Ccf1zc0vJ?PqOB3NtOI
zO)C$8<NFa_hWNi}>TJL(%oGb?vK3oGfSYUp$0(i0i8?`AoLKL_|9Wl?XZCo0tQ<q*
zxfr%^<Hn}x*Osp$xC{JV_qbOmUG?CKHQBPnC+jAmz@ggjX(C#A0Blt%oBau{jd5VV
zT8W?deSoqN0w`XMzvew?bMc#Iov{PU<14UI$bqy61SsGxWvlu;-N-nUN)+@1mWBc#
zRsc&bFc1n){+CTwED_&maRt2><$MmX%pI0Y(nfwuEIJyT3jyPsVIC<?&bCFh-<aOs
zWCM6|O(fX)Ge9ZGS7}_j!wuXNYQ^kfAbvkS?8E^@7GPsNqCUR*wmBdEX>r)`Hx68K
z(b*r*5K~i2kvEr>0gcJ6dY-f4m8t{6kNjUmE>gWajGw3$0>l?xztgX8^%rWKcFT$!
z#C@zB_<I~hE@PY@TB^}{E`5wzSK|Hc)ncKB2x!Ed*yV32*Dt6@*_XzoJ-AqJWYVh(
z2xbq!_duf+`fWOaul`fB7uXPEiJCygzJR$yY6M&pK)q%TJ4NKD(r!wpG0Ue_V!h+s
zol1*1#+{9tYV)q+YHB(X&aDq`p1J`^2PivJBV_%7F!9UjK(Vr_s-m)z933G(DT!j}
z7f=<QEdP#rdh!%4nDT{nKI}RH7K4dFt}t-x7dyS0las%_9(|Z6tDeWz&$n$2jJ2$P
z90=FlcTj-SJVGC5POp#k-g!+>5t9MGN=Jt&Xyy6<T-MN#2qdFepl)3UY}KA)Xqr0@
zJY#!GVJ304rD%9V?m#oEW<bXSFJsIo^&snChEV(<gGHmbn<&Rz00ei4XhNPunmljJ
zi<dY+Gh}kwWd3d10SefX;$rc;;}xJV`mjWIhZC;aq+$tWhGL2KJb*=2Y$?Zktp1}J
zzj+zB1TaF1fh@{@HB6za0#^$wVPHT8h%Yi=1?38Ou>sY?A^R)WrjQ3TusHN_VW9l0
zMMUQ$?=#lVIbu4!?xBW7A-1>>=PH4(M_aK!h3sC}Ge-UV`7`M56|idSej3-e<NINr
z3i>B6mT7K$xVIxCJfw6YDI*iu%*Gc^JHtypruNbW)M<q9z~%vAmw8*NUH2=F!ZpXB
z%S6BHb}sZn3*GbG9D0-lEdxuh3&+*<H2`}r>WFMuYcC0+Bqhwx&L#{+#uH`4NlN60
zrx4=S?|}e9iTV6bJirO;%VX{TIK%?zgNrA9N~l`V8kyBB{WZ<h0bor``HKX>K^DVQ
zqn2L%2eUidAFfqTPWkj_%U;%{lir6GhfOVvTcG72VS&SXrMHRcMTCUYfdi#fDG%7b
z>e%7Y2^Hx%IQT2al5%jhOWB=efE^)@029RRN~kv#hvgPAE5ZG^n>4K@j_W$Q1dvCO
z8UhIuXFcolBVwXUEmWTMth)^B)01s66N4-JcJH6}*f9fafI{jI(<Id-4V2%uS;BDi
z(2nEfm%N}^aXfb!{;zm+Q00JSQBqh4r7$x*HfGf04@t+^PNEvl^8aT6#D2S%X-G!+
zMG2*vlziiV118`$knK}O0$*NUu&ML)|CWC?1g#Jx6&8|bd_#xSDhMDP68Y6R5tS=_
z5?Rm4%xntWxf`(1_od>f2^Qn4?_o4RYGlTz!oK00Y_r&GA@7B8)nO|@Ery})r&ec8
z5@)Hn5hU5p`L;nY{R@@9mC67br7zwGzDz#j58#DfJih`0-0(Y2y(aR43mq+Oz8N?d
zu&D(fHlYDC5jbxZFM~p+LvwNrT%6A9LJ!pWwyWG{o%g!PeVhM^;`RZvZ5526cKR9!
zi$qILFPrcM<KB3MQ!r2Ij-LZL2w0$p$mOy0l*I?~LerYa?B;>{hTQpwZ6_5a1Z_jU
zoC@1bF^AU}ah^UYFTAUq9u3Gih>yGpvzU;7^`K_gGc5<oEHW2s<N$#nAgNDFxgctw
z2OA|TysAx$ooE%TDJLPs*gN|bS&W)NtRVhJP{>XRb;q{~%1ePlEh1?|+$b;1<2cVa
zc<7Y1{Dvm-1O&mnI8K{oR={Hc;wH^$SIsdpG5?3%R_v|*<G>_MXRyQL8nANG4(e?P
z{*SJ&46AaBx}`x$K^p0l?vM}xK>-m2>FzG+RFE!dNdZ9#N$Hf722nx~a8nXWcYSl8
z@80|Go*(BNeK?-I-*>IK#vEhJv5Xl}xXb5Maiy&jWU*vu;+U4yPXYofiDj?k^OxYO
zzt_r~TwXbPuP~U}Fx?RMWzUjxx^+!ouhywE;uaILpm6pNU>d`2Msn^Ptt|webHf6$
zgs-BhsreKrazNcHD=S;G_3;3a+`t0=9;1-He~)w_Y2Mi$m>Yq`8}kwrk@C}@ZApNF
zSnDA-eqT}&g+Q3Koo#R;5f?`>bGce&ay;{?tg7-eMi;4~)bh>rbT^rIaTO<XfGljF
zaf`Yn)!Gvl7T%h!3IQNn6q7PNxc#dcuFerB>h^5}kjW9tbz})3`k^w0vcdz^DGxx&
z3O+vKig!G|2P#B1l@I#q)Y_2(ubi?kOH<=8BY^w?Xxe~+`JH=x!S>sQVz#bvf8^dN
zh+h`l>f`qIJyE1tkXp*|%@OLS*n<G=j+(xEo?9kx)ymed<}^5~tVXi>flY?Go?le-
zkd+wOVapLp$pd+)fp`i80|+U6L0A4AHF?kqOS7=BAhpryI#pNKD*yWH%k7~3-{Ew7
z=L^@;ulXk$T+O4RqP)-6KOm1GA5iNk`OPpgimH6}tx#qN30szxplCyqEF&+kbGx6G
zqqm2LFf<0VaBvmCvIlxgX&n-=#fFeN@kjLnfiji*7z#8k6eM-3%!pTnYV{PO&qeLN
z;~1npjHU7!;1CwB#;#Rj7IyrpNhTF6y4Lx-^79m8EkqWY^Nb>OV`H`F&r=U(er&4a
zs%x=2!||s<VV=O{J<R-B;Dm+YKbF}0<wuJyZNp=m3hOM_hlt~H<i-G=82=4LlhfV@
znc*F@wMdhNO4BYtT_+@{9&hk&{RpzFjXY^h4m``juPm}X`1dOl3I?)`qkr=m2P;I#
z?)z8e4E6;!G>5xVl2oz81h4;0d{BEFGJ5d1d%Kz7mb_Ns(f7NL`&=U)7C6ke=whv?
zOe_QGfsvRQ0hroUmkfwUI0B0sWnTOtdXUFOiGqiL_I>hiC!O2)3*Ly0<MS$#%OcHB
zR{-bk9Q%IuCx7*Yo}V*rbSzS|n-7m}I8)gA&sd(A>Sly!_wj_vb53c6F$u#&Jf(FO
zMznaw-)tkl_fIWyGw##Vj~23K%bPvCW38TYSDJVAjW0f90*6UPkxF^-OV0@*mP0yZ
zp-F9jR@h5`saY2ud{#642c_lwU$g(!>6)7p0q{yBPsCm}n15g<XuCSIH<)4}i}sl+
zUB696reS2nGy1E-ScdXFW`)diU>|M1y}!XD;(?Ffy_D0PQXpPt+@2HYbIDIMI)mGh
zbgNOq0lt2o3C^G=pAbG*bn%4sj-^M!otf;fu68dioXbr3l;HKkK2GU3`|?#iGbGwe
ztuowTY6YBqk4#%5maAR91FH^<R^keryGUCraxMCRE2JNq4AitVBi~4{B*cSyrlda)
zA1x~s?}G^){7%sFyx+FM1Pr3Vvp*$>U^Z-e@ipjGrSQ0ora-i&Z4c~cq~fyo)4rk#
zC}57460Qh&dWMNbn|HPs0ar9p%2Acv;k1oX;%jsJsMA`lnInc0-ctLWrii+1i&62f
z14_TeM1+LCKESE6x*~qZx}R8PiJSY6W$5)Ma{FK<%}6x#UCGc`a!-w~nE>(g$8Dl6
zqN&)4_Ej#p6BEtmi!VQi%CSx|b1d~;x%J<j-7z`FLimQZl3tVDP<Ooqq85&m0rBkI
z+)<z|xhifYZf1FtV-eD-Km!^$Vcsqt7xG)sWeKF5r3Ra@sVs>Q(Cl`{4&QHOVNFLO
z(1j`yCRq3t+L&$fIfKe!Eqeo;F%<l-%76Xbzt|r|A~wD1*a_WhJIuODKg2aswsUG+
zA(Qa>rxF+_4LwZ})Tk`apSGW!7jnqB^;}?sC#I#P5nDV~`wSYEYWZxPs$sUaWR+<b
z)<b|_0Rd8Qq%z&U`N)`=mp2|z$1XTjUqX5NoA!>#^Fm44K&(7uRU)w%JBQm-sr=cx
z0zrP|gA>8UPp{#%knMcL<NdL4+mE@HO8Smn6#&UQ3CXXBG4DU9m1Um4I;wByfAMGR
z_GET2yF|<BYNj|yZ+`$VKYo*Fsf$M`5c-Bm@i85EKcU*DT2?`21zrUzBATRU$D8Ek
zp~ZG`WXrS2;!p$7X}(1q+Vni1{ngJ-zcR|oxWK4`2js;gsB9WHQ?-tgrm+E<#sZt3
z=-hT122wC}Fp{Euk}a`2=RS%gC}{n6WZ@hHs>*V@(o`i!@+}M#a=hkdCB6l1rnr`t
z7EpuGqVvpTB7coSD09IVOC|g`5*UlRPiN<VDBVEAn*g5_O`j<c0C0ZlK!6aafskKa
zSoz|*G?6AFY$dpuEvUD83_o5Nx6*ddD$Z^J4UdFe4Xf@PL(NJ`R8+hA6ci0S`slxK
zDu4l@oMsTHMc=~>mgw4g$JvL0fmgDzUiD7D#)M*l0`NWf^TsSm)M<`AKugK;-%Q=`
zJJ}E<9Z_6f&U){j{u#&mw-ic1u1NK5@gAsX<x7;hxeR2s6>H&}-kSJdN=|pW;#*Xk
z!nQswi#>_jFT%>zEcp1i{SMc&L01~&i8>#6?#NNgTCR(AlKj~J$J4=$weGN>T|`fo
zHzG32wU?436O*7*hwZIq99~N?!Fv<>U%=h$oeQffyE^o&MMp(3-;@1R1l*t6>0f7O
z2DWDYFw@5yRRzLC(Y}$TI$U8+R5zZJRln-0<~lft>O#NI6$5)FG-PJcd&GzRc7AX1
zmt35pV%(1#5z&iY%*1bAK7aG|aoXC`kmL<S$$)=<4nt)3`++m!M4P66hj&OKIewSn
zvZklwZg}5hQ&rlD3@$RIK+^K?!0@{0VL>Amm7Ye;(rZG}Z=$K|K`75FjJVgf6vz$b
z!MNOdF8V^3+ctys_3PKSNl95k>X=0*n#u+JPh2S9WlMOJh)#-~z_hm6{!eP>v9U$*
zB}Scp`=^yvyT38Lxuy<%mC~&qA~yI!Z8N&>Q1|T;FF%o!B2>+NQh%%6s(|+)(-dy<
zM>pG&iTq6o5Z$V;<Here=KQ#o32(Z*^+?ZlNxQ-b|Nf&#ky#VGZ6}P-M<u_^MOJgs
z1^;J1HHA$D;rxeVear*WI2!I(Yhvgt8Z3M-e#W9~Kak9UT)lAa)mV?xN}q2D!gu?m
zv_aeOAyw}dTd%4T=S^#v-5m@}AUAw*3-}CEz#a~B4eCF|raWBlkSkW@hM6MmdL*$v
zN8BHDnS_SFtQ`Y;TV<HWl}pG*#_&-N*`856EKt|2($?4Kq|9!|z-h+&+8sN}SIMnU
zMjOFTw7KvS&Aao-jj(zN>~Beih=tDiXau^}s&<)dV7u|pfGRI_P89_WRYMt+#+*wc
zxWYtQSd2YA)^0noe%_a}Fue$lTqCtnbiSePlgRhBzuN5L?&a}PNTiZJnhqE{ydJw?
zLsrQzW5j+O<k(%Bx=;HTN~PuG<j|oZ6CoVQc*O3o^F1kOX>X9XZ~K_8l1i_PW4(+{
zy;}(0WAN=})P96`6(N>}LxEVh*pn1VY@K?iY&^HaKjR@FGCr;@O5-&iX}>z;EpI!s
zgTFK^{%Hn1jvM!4HX~338M}Oap2;I&0Heq@pL2lx)k`*M^9)v~L#6;xzsUf~2RQ|f
za}dBV!SMl5N{S<J5&HHUSRvqkK?L7;@M9aq0v^MrN)S4d-H|xdMJ#Kt3{b~Pn(-=<
z74ebY5QhIS6WHN>q^D%wXYKEY{Jb-JJCr#-yqtg|UFnM*IWYCIB0nSu5GdXX_swL|
zZ}?leh?xV&!6Ea3^O^t2@9?Y_-yZFGt-Bkw+(QI3tmXK@+POi(`n5aByB^pr>GH-^
zDj-#1;rT=L;4*H@f-<MuHdPYOa`6gzWdgXRR3?y$6_}I+Ui^7VZ@c}Y))G)6vKzlx
zR|`sPSGL}DnyG#ba+b-g2=J#%1@Y2|0nUnKsa7|4_rjlHVtqlg%}zzUUQKbhC$CVc
zjyL=ERtCjwJ9`$JzJC{as;Aiv4u%^U50f71BCQF){XCj90PgzTyI|ny(vbRc=M6TL
zof+%gage8U@tOiJeL*7je7f}OJ<Jsytap`YmHzorN*>G=0gV^pU_x!d`I5|jD)a^k
zR4X>`t*X9;&GKfAW&568DC!&z+pvhl6CT-FR?@YgWc@r5V8~l~uXXw6Ma~N;t#6%5
z%qY?ZP2cr3g<~7?H_JI<ICcetEoa}^E@X}o+t}Ek5P#_nm1CR6v{Sq5YK*XU{blxS
zTa2Le`6n~zbip<>xIj0g^)b_P>M-MGDQ@!s-ZS*Jy8#BDzL8|I+=|%Gnq%HnYV{mG
zuvnz^+v0a%iz@Y55M1wl)Zd+Lx~Ov$M0TZm-@jsE8+*1?)WdZ}tmKRE4@zR{ySF}*
z2n5=x3BM7GdRQP3%>F9UZt-cBHuLB1>5t!oZkdoj$@-(qNTa-)ZD!7K_vm%kb`wt=
z7XR!6LHzvZ21myFEiZ1k_C56a9lGRKAkGt(_gjL83h_2D=+%x116sXZ$8!$~1>xj|
zYtY^SCg)mMek|%Y3G}@`eQx5$WAtKNfhOo;_xurtSY6kinsi=SS<GIPhQkyl@o3+p
zYN|xij7Tp-<xkT?i#sp1%N7^+-%gL?j4OpQSZ6B~TAE@=)rCtw@#$s!Qc!IokGNkO
zQesF-Y6kubC~a`qqgD*)_}7664lq1&@faeZ?I}QBunfKZ%v<O9OhbdQ;sfj)jw7wt
zkVa7D0h{9%L3rcU*`^ObU>5V|a|L>hJX>L=6c(12$YG}<=1BzsgI43(vs2`_3J|(h
zP{>r?3b0Uq{!HQ)(u7X6ys}a#hFzZexhBjscoQLJR6UPH>$808E+;2=-~a+3aTUGp
zSi(sDWn;n|3G=yK$^HXFPOT!kE#6HwhZM2gC9&#oYUbQ<_ME3Al^ZRZoWFZ&NWX>I
z3AZ+hKF$5q<oV&4=~VjT{<8p4%~OImABm3>17jMrU8N94z)-$27Of<1CroF}D7@uL
zuZYb_ER*j$XOZw_6jdwmDRce%>@&jijXt`dEacdbix`fMN|xB04>`V33m2=R*aw8)
zQHt$2CrGGEqz=l5^}QbCJ$+2f(wd;=pJo^TMcFQrJ`I?h4Xs*g(5rm*ERZY~5143@
zYeg=N;*vwjXzm9-btomx<cYGw=-ywSZ@tH@UuOYmkj8JTiX!X_w4-8&-=B>O#Ud9*
zi;w$cxgqEA!RAzj_1HUEj?L`g9TLqW7bp}+%6GwkA0K#L*$E+J5AsCn@O!egvA+Hw
z6qiSXsoc+wxHt3w=;%P}*aL@K7w|sAwNpmXMJ(s|bT%Wl84p3P6to_(=nBK<P$*4E
zZoP2Eb+dL^UcB!~b(Ho^(-HB~-?cVapZnco?Pj9wH6X8j_J>7LD++soFw-FZfo-Ob
z)^dGyj^s0-i-isrx<10iL)Ez|b8Miz+me2}oCNpW*bo<Wo>zf=3!)X!0eeVRftlBH
zdwv)Qb)8mUVYciS-vT!WUzQ(6$UO+tc`)^Upx4n$Wp&2Sc`3DuspEj(dW5Doj!vy~
z^3$%+_3xuuXKN5Sk<+ax7tobkgBbz%BCI}FWC21+=QqdFHgJ6{X@%&Hgy@9lXIP^$
z*OEYSF4xUE%~m{xsR?iv4uQ|B0q{O*`aLh6#i~>JQi{XI8-CJYnG3u=Ul4TsDTD-v
zl+B1~FmjE-|3F29GbHy8rL$-V9i}tG(XDz(@gkccGmxK+1Uln>OBgNMz>}v{W_{C5
z^zi9dvCG@(Mg1Kw(Oqd^2K%l`;bUVZz1N+l9EVILE8T~)ktMbaqgE7H&aH=J1&Z;~
z?_bh0^az7V3LH#@7{*{=9M4w>R{TvC7k3x;Tb>-YSrr9{YK*rB7^xze-ZO9yc3Qt0
zCCrwnH3zfLJHICpq|wTX%5(LHB>RL90v|EIL772U;L6IMVP*~qXT&Le4A-m+C^;=-
zpra+GJh3B0n^LVtGn`6EjP86%{?h6Tf_NnTqnV~?8c)u_dzYVFgJzqPpP+3d!om16
z*@!FUdyw!_`xaM<>}ak?GKYDxkt^pvS-zVnGxkofi1ws0W*-jm&en;qrknqnYplUb
z+$a}Kg-`rulnvrO<NJUCRYTGFe1JZeUXLme!0P$R7Jm0eD7xF0>ae_^keMsCw!7rp
z0VSHQG~I^;qTY3~Otkelr<b7ts@})hK?3VR#-=Nt(Y@aXm*n4Zf<XovFb!SZNK0L$
za}}r^^hdh*yYo?AzR+oK4E?yc@PGR;7K%73?akMJz7<CMYrkHa0{dT`)33WLj^FKm
zBTZ_JH@<c^HqgHLm?30K3{oV(X4?z{j~`lGI+ZA(g1#cq$dtmRcIzS4{(4MDtPvbJ
z&#-YAu}zZOzM5f=;=NJ)VY<#S^VrG$Msrc4f$zFb+{iYnXSL4axu^LTlLT;6s?9q?
z;|C0rM$%;(nk%=Rv3y_D(gZ%2s4gC=QYP<e#DEHM|HIwuyCiy&UJPCAudEs0utW)V
zyWf6IkYHgF8Ty=_s6$AV!<M~Gt+o%L_v3IioUg9Ux$L|Ar}XfRj2rz|DI2~(9DjLC
z7TR4#F%$OEI9z9lb?exA85=x;Y!9ZsRFec)#J-nGP~!V6q}t_q2c+O0WL3~MzM$S=
z-WriQHb722l|vJpT(q;U!o3@`ew|Cpu0=vpzKL{dLVuggh+0S%QXEMvEQ^%w1RSRM
zPANTFNid&<neM<D;<ozMx*CuJ43N{VOcG`-N)W#j*p^?^kUK#sSE0CJ37K|S^>c2x
zYSO370Hb@Wxr_o&Ep?dz;kBRt+WwDD1CtKFY7!<cnG4jL!O?a&cLWloY{uvtKislU
zFS1q15Q+qx_2l{fXs%4IzH_Vaia4}`kK1(&WqvSg!5-wt3kjzSOiy|u5s7p>d{<X2
z9OB>owIO+zi|akWUyfs5iwVOixTL{jd2LOZG!6`@G;fR?3RxW{{3@^5-j|+69J|2k
z4^E>&7ByJJzZ=(bmVoSE-VVCHhrBn6gjFS>iX%lmWu#mWJjC&mdl3hpkPom!`IY3+
zv^feKf3M+7?Ckc@)x22$mTY#Dv`Qxi^lgL~L9^F3tex__wzeD-KK2Fc_JvU}3HZP_
z7vJqBd6pQwZm;WGwxsm3{O>}HtPP3b3tzgUKu_8!Fa(W#(&s4^xxGJ^wbo+XWTasa
z+Rp9UxA{yvF+dXmNho&9xMEX+Nih!NWW>CK(`lj3k-Gkur~c!YFDs|g4BT!2vx&T-
zYk75b?}xTA>inEi&{5}Pw79hWB-azkU3ui&LqgX$nJe#(%{2&NNb9G~L^ivOB(Xz_
zin17ycwO({vu|bf0<}yOE8?={sz?&A2hvtQv#b!Aa<L5w@P#+!EA8OuGOA`tO;$<L
za*E1(en)b*9Uz9G;V@D^BkIrngIy$09C$O7$t;5vUI%*Oqf7`Z!@ob^tx%v2-r%qB
zBELdFK_3u2SvWsg3NN?@S!oDN#X&;=v&;sghmcd^>l7t743gG-46uLgU=o(+EeE~A
zA-Z|WeE-Oa_(}ZTee&3cD}ov`?)gn^d3mV0qimv{B>*WN;-}(r@Kxn$o>$=LvU#`{
zAK8;k@mwv8C(7RV{S{DHU+~)3QdHd9{Eep)TN|g}LxUTpyd_7FR`Ci`__!yRfk-0~
za>TOn?W5{bgP--zqhLf%FWd7-`j!er_Hw!@1@e@ff1Tb$^8u_Y#3LiDWHlVz+$sJ1
zz=+;+^$%XCAekV(oQN4Bb`xVHi_SZ}I0=x>_o$^P56Srzfr^ebR`BTF1%AIYKH_w?
zZWuTKMWAmoipHBf?6L5vk_FnoGbiaVA^G4>U<*k}38`q=y9ovx@BzW<i!ZW8x;MlG
zRFm+jMV>x?&N>wfCKPCM;Y0>I@?c}xW0(7TzxGc3ft{pPok04WU{VEaBUM1wzx_j&
z>1AsCqolKs)-MLeR-}81g01hNy$s$OJcb?N0`L$#Nn+YC)B&9S5wWQiit1FA(7m?~
zRUT+JnrS=|WAe~tnS(>oIeo1n-jDFIq<=Ij__#yn{<VkmA<?FLKtL<rPRXd@1x=*9
z#kHWT^p{T3Q3H>SP`pnjH>1tFn|MYTCT~@-f<H|BSpKOtUs3lZPAUa+_P*Rp9WL-Z
zNLje=!akakA+@Vk<NFB*B`Z~t3`%*3)2vPvJ5}WUx;xiBL(#Cy4NPahw#NPK=2v>o
z&&JGId52m~bX+(a3Gr02B~qTUV=%LfLzil2HJUTL9eCV)nYgsBK#Fij#An1Pso8<`
z7Uqll01X6xZ!<vnn>iBKX-xkC*B25nG!m3N#!EQJ1QP*o>jAOXLV(wRkyJLf3$xZr
zG@8hi$9S^c1!z>B)k*a^-mk!X8G_Voh3&Dt_ZETwKwI=`KfeRSz1hb1Vo)PozCB`d
z1%+X-s{L?Mzwvh%b;EYs`N!hoJb>p&apHcW?K~KOQ2_sGQr_0q9Z6zSTSZQa!zFZi
z;Qa%gNvT~Ui;ae0P*_2dT6N+CXie0U<M!*}EOGZgQx$YzSp%aL097-jzZ}rGZoShT
zq!$>#<(uOo+nmrVOmc$`V2}v`G^wQV7{UlnA+Vk(EoU302ChA%G+=?0N&lTCi`CBj
zF)(&xA4S7s^*kX*WghB|rKROijudt9!u8eh^71mD?KftSVczB&@R@bv<oK*oIyL;p
z>X;QR>1jfozm`Ekx{Hxt1_jAak3r{YXK#;m3zctHgDMZ|c2n*G^JH4Q%vCz<3rKe2
z*86O7ak7h?4FMumwrx<*3Z|1~a*Z!Hv^kD5MleW^dHeR1X3``GC~RexOg19(pDzBJ
zb!PbNObCXoC9iA=KZS88m7~Ub$IN^G8bX?yBsMlTZD*=k?niB7_;2&Wd4jZf8Aa(?
zF!l!SE%zgbiL|v9;dprYT_-(w0EevY5I`rzJx{JoQRkwjrVOO-^!6VM|5&_0Ckw_z
z$_sYymvl05o5I}zP+Kif+(9Kf!IJ@H4wyM3|6PS%wRC!NQ+gKfLvw~iPR59oa?COS
zA!0vD6cpbe*{HL#vt%O{Y<&;+%8#X`OcZznH_3%j3q)3KKj{NT3YrG6@`Pt(WXR|8
zl282@6DCpJzx!`YitPtq;$=<#_{Z;)<i}%UWB1u<de0(V#eTdcep-qJ3Ny$ErqF}f
zWb}dsjWbg*x{8;JE17hSEz1Qhqi7exA<&=d>h{AOiYwj`7O`WOm|~X)6PApM21zde
zp!mr9_%z}<l%RS}c@$S1PqRtq5)OJG$9tC_aJ|Z6tc3dFMR^c<ndG1my~=i<Ilp^V
z=SL;^zIMzZ(9!^nL0tqp!pTk-p;X_GF-t+K!Rn6{6=|d3y;#)i50tgjIEK84q^aLx
z1X6o@EjbUL&~&XWZ3pFU2Gr*y>%gU>_sl|U7pdk?sG1`GPM<FrV71$TnBef@=wE4(
zPl1EWog<NM5~=w6pNdIGA}dWhrJrkS!vup8&wab<Ng-8u+bj1fM)u*8f}aSZ1GKOA
z1va`c+K}+0RJ+D)<+0D5^~g-HGrA0LeE7Qk!cg?J2Wg}{*h~F)zF@-<fSe|&w!}rs
zCZ9B|p9oo#Q~MiE&D7eHLxZ&pI}STRMA_21v3w9ZF*zucuzf6H`M5a}N0N5&RW1=U
znG(4#l`T}pz56Zag~v5ZUEjQ|`n;VEhw+g%EHyEjRFV0${r=rcmUnXD1e&DI38V7U
zgu||`yCoVu(}#a&wSTe=R_oMQ;ltPmMuHo7B=^@pypyGh{=Z&;+{e^lZFP2-U4=NT
z4n91vCW);iKeG@R@Gt}pZdvl){*d5`9^*UK9-w_50(&~pRDp%kvtP#*4TFBlp3`cU
z8FG7XqcC!nB1-IYf^{ao+Xc+lec4^rHGGfCrl;sR9iJ^`Bp8GDj57o;kP#0(l%g?@
zy1|v9pFpqkFW(DiUJ2`e(a`92M=p&U$RhT1K9C!@pd$HxXxA`@Gey;^W*>aIpOW}I
zBnAz;q-076iRsP38V0Pq+G88Ah2v3+sQfhFy#|jnl5O7F+PX|POW>MEu^9gjoLGj&
zJNBd0b*)oHH2EE^387>%Jzyxtqn;ah{aV*3;x+75E_1!!zP^+}P8Gp8OJIhHpM6H}
z)=Uk|Vbc}a79I3{_sE7<{hm)iGc5b;T1|2n191DFCN{aZfd;_m78e)4S>T<NP&d2=
z>{NI=m`m@o(nu0GlfMRJWEf~ol|#3<MNUNChx<Af`6#+5OIlTDg5FqM)8T=D0rQVY
zUa!oc5$W2|AUlH-1xU#ecq>Xzp2W?C17<)L6EPQhKydHDc*LQgXI9bjVqJrS1@}F3
zYxh0fXU~{ObdmBtIJ8pQ)p%3ALGiQ9!|)b}fC(d{XD$5m9e9J1$8+v`G>mP9i2Dvp
zOju^0o(DZY-(d}08yqRJ^o(O8B`|UjV31*tx(i~@duFf%z5aCFW?o%WW62Pk@j@%w
z0tNnkMGxiu$-@o*TSrE*pPI^*n?3*9OqFL}=w2FNQC?(yWv&%CQBTnNCZ?&qU07Cz
zH}x8HLtm_KuVL9dgLww95+M&mXmL}qo+_bV-i}x-z_EyQFYQMiKUSU)Qn6ghqczsn
z4&Q77hndO*Gg)*$>)-JFjBXD@T~{6*WwYPpkz`ei$D32}@J9*;oV@_P^mK3K25ZJk
zOL2;(Xj5;uS8-!UQ(rnt?Fhp=#mbYaiLlU6^WWfv$($r@%P|*sGk$d1>L_4bTU&ef
z{5g{PKvDv#nIb)|H2=ryMHTA?cyDlYb!9~zY|CIJ=55(QTVz@18cdXW@xt%}Y+6vk
zs0KawE~UXQ1}zHkAjDn{KRveYH^<#!=jJZ_oS)S`8o*yzUE5MZ9IPYBddG)9J~}$O
z_~S<ySVP+6kl1i<=3qnVoCt@gM8kId<;`0!v_+W;9>$shr-6c87lI>V=-Ly=!J)V%
zqMNTR_%!+0FT>t4vN4sn*VTzM{vK~>JO=XJ{OrZib8YxO91GNM=qnvFP3Ye&Pk0D{
z9vy47j7=yf&v?uK3m;5~qKLi0(fz&E{g6V1TdPwk1&6esEhX|T?X<?*%m?|JmVZ&I
z)7~igM+T=W86j&Lz(oMEY6=RN(D?!L@UzZw5X`YUE#6hhxmxIMi>>-#p9jV7VOt-p
zGk}?Zh=*Cy3_8!Z+eZgdI9(0X;BHzNXE>Q^zrwr^PCleV-RE!ZluolJH<Ix{0Wi50
zn39od`6ew*5jgFcD)VrQQ;IMKOTyQ41x2=hGU9A0DE^2YeYxlRfQO%xQ@J{VrjG#z
zbus37QDKhLYOM$kj+-Apel!8IpG{n>+VPeZxE;$k2aN1rjX#I?jC_f(^iV)1P@M9B
zuK%%?7U}<b+Tc=B*v%Yw@8*M}{B(cq1M|DkEMtI@M@bLCd`tpA1sGXKwE(E{7k_`I
zfLR<y3Un$>URw0rs57A_LRy!=V6DljK!_Qw-=b}oAV2MNNI_q?gO+&No6370F)*yp
zURt&oW^9akQf0`edS@!iGHVz>C<c)qzueS&{dYHN5|sjxuN-Q|4$Ll^=;p}RoSlK#
zs4Duad$~W!lHuK5@^l_b=5#ePy0D9;OY{U3FqYo|-gpg=J$9ns5LAf%usy^yv?U<F
z4*soQ$YK0YSiTRlxMXks&bMa5iBK7E?l~|x2u%lO%Yz!|ueeuy<8I>=rG}!XS`}6b
ze2ebwZtrP0^_K>OwGlA)=!pAyNF0E9=|$SRGz8myP8zs>HOX9Il^$i1Z_M3t;8v%|
zJG)~R|2}Z)C9wx$8l6SI$8}w$75$epa>(Z?iOz<R>cu~kCEfgRyBe!uou(JXDmExF
zRKiiFU16>e(({N;nVTLtJO=b1F$_~6)nNc5;xNezj&7v93q|ci8Ry2;tpsiy6y(cu
za&n6LMYg>I#oPAV`^e9#2rr~b5!N9qKxw962Xp!nh6^Gsj<x%q2&K*5Yy<&?QLet~
zM1s-VrC`(dhc@3U9mnvs<1+2Mv^+(2vim-6%AQG-|1{Kk@Vs{0E-&})y^pqfQ4f-u
zf`}eX9K6?s0%!1bi%;MyJxF5z-@l#oFg-KN{SyDGe6zvErl}WH9bE4e`n4?gIPDo{
z-56S_3~0;y+gL#pe51Lh#JrcZf+Ju1N>3JDNmnL!6gxkkK~+S}U;0uKc03+dp(G;j
zBP$5asagbfuQUH>GnfDJFocOQeV9*uY-B_W#~&Qw$Px0B-83@_X%`R1H|$?bP+~*3
z($KXsMRuF@o%5UO!a%YA&lMg+!LZFsgm-0eD$6j+PC?NTm?8!X9M(%F=x`Mc<%57B
zk%o8&28MlMK<48A%zI-@LFR%<Y5;gNgu8bTbAUV!b85@g!O47ax)1)}ND{Wh+}wD_
z*({K@lGD=6b;*EQ;3&JH{s}qBB<WpQ<`1Y^*naZ8Qu{|d(l4w^#ZM$Lw9XI)CH@5;
z18<R5Z&ZbuCV+hdIgO~A1lxem34t*)bdM|=>5-k`FIqqJ5=(Gcw*8RZdv>^qTG3fj
zSvgkPIWE<ISDnS7dHxa`4b?x4&>>#exmhG^mH*g5l*!qnfq?=~gFKc94<qvXBoUtX
z7QrYMg#O(Z_@Egn%g_7cX61o2jos_YC9!WgCEK+o)Ws=A2!9yBNW6^JT<?U|0)#dI
z{#~v#MPOZoyoIAuNwRIof53&(2JQy+=g)y91n(`e>I8(B2%Z49IVPlfRI2mY<`#JV
zAT1OZy!+q?1EM~*MY?@~_go^<YjXQ|5NibKuz($m-QtCu;6W~WA?JA-Il)sH0X{7<
z;U-O45&AB=IEAj#;((BqssVcroDEEUaLx83WY;Ip_=dUWgd9b60uJIk$i?uV)qISG
zdf0V5pkLSo6i@Jn^~TeUf^^NV`mDMPntcn>($Tf9?PEc3RH8IGA;*Hcmt6a)i0ag`
zj^8IeNlc7IQ1G)MTS4HF;CKUVbkNALvn{GtH^vbhAMa_V@~G^01CatE{CT&Zo5Jkk
z%*+f2X<a^f@yE_AAkqt8EBv`a`L*Ykq}52(Z7w1$^-TP&ELJ#L7>^DYL0po_fY^<?
z)S@c4ACJL2`vsoo4XjtePb60n`}Sab$w{E3lJ*x1x>%y~f%)(TL=@-p=?^<wtyI0g
za*0MaK3(N*HX*s$i~Zc|WyrBIgRlC{6X>t4KIWgGJAl<4i38c7yGG96c}qS4-P*&$
z<HTPYOekP8>xm|Z;SLfc#yJcG>K?U%ZU^S_9`I2}GIG{B_C4%VPz=Y2Hk=Tm(I}mm
zSfT&X7Ey9gug6jGPq%xmdvsmu<M&G~U*pf{RRKLYe`SX0YBh{g(uu2k5j5*l9|icn
zU%vZU%9*X;!aNd<ch#J5#Z$7EJ8pCEZq^p8gfPF{2!`*Ynq6Q&i-4?w`K^7>Qr&?>
zf$=YvQ7NGDkh~j!g$MA=m;HXgGz?4&_QRI*0gY8->lIZdoR!#+TLC!%NN*(4N`f5L
zMG8p3v9C{6<hJA>_w}4h7d}S`H8iDG(6(j;9L-!n;u6+>adP14vV>uMj_o!H=@P?x
z2#U<e9KA3;*Hrwlup85HAs@zHvMK**vI%Blj4qEhO^A8M`tHu6_4dxSe~b?L_oa!o
z_XCU1;z@F~^8{KoNr`Sz!z>H7#Vbn?)FP2e2_0siYbAo9g2(?gD2*z{!Jx-K66dV4
z-TqdH3)~c74B$e00W_d7udD>#{B9AlB$n989aE;l$P%X|dR#Ij2bK{eQ8v;s*XSN8
z&-mGXDo$`zVw25~iCOeM<I8+jLHTEjzr|n~JIq?mHMpu|il{#3a0ICwxavInqGUH0
zVe;ICa7!QRXwd~<{E2ODp<SDpXr3mV9pZ?bq{|7|=3t%U!2PYRcn6cYPjf;Do?<T`
z`{8ee%76|Y>k<j8GJ$)DH92%>9+@Nz#a$T;p%*+#JfFcZ2QY`gsFErTXoE)6XnDrH
z=4Ox0UU1_=GZvdHaob&fs2+3(U^3a;nTT+ZLr&}6me$tFk^FD|+<uay^5@R09KDR9
z{+xSu6?wo9YUt?T+;0!OTs}GTm!1(Ff~`F`2<Zrr*6@tA5WH<%qD)QYB=J3k7ud|y
zVf79(8Y1pIT%58~t$yyi|LXmYjX+I(d{>#BgCno5j*OgK^dADHskk6|WJWO~t|JIl
zK0dF(l4NKAt?%Jf=wJB}@*hu1F8u-K?xk<|ljOd%Fi~Xg2@N3Wi|o*qq+NsftT&jq
zpremVrp^+uya^V!9F>D8`?nyM#7iAflyC0Aki8rDTF@Uol4BqwkM7~ale;P#U@N+Z
ziJ+B7dvg2X#~b(FoIKCQ;!o`yQ3JymNY3%FH^HU1)E5ttIfzsRX5w9oCyo`))7{Fv
z1<F%S^Mb#dr@A;RS`rDRPwo#;tv8JT4XnsVaUuKU@LCvNIM@InlwhWp+ku^n-X!A>
zPq)K1?QE?*+uAcOGN|1Dnl^cm<8g2OHIjgNRenTPRgN7Zi!^o22OJyP@@l;uL=Im7
zTl4gP&9hn5W=4}!Q-jQw(4@mq^oMuv=ot#LlB+*EW<t>iYu!`vBZQ<3shM=NaB*3g
zN_iTDg<ydg!K0cXgqoN*3_KNNR)BK5`S+8*G_(d^ulwek6=ywDFmu|SM;ebL;{Eoq
zFx2;0B~=qd8M=w^32m!FnK5pL)_vY2OfkI|_TvYOPMlr%kFN19H=Bt9{bDp>o!df=
zwaLjM^uq<p`mW#|RC>T@vQNYre=Y|3MXMKQ@#1TG?tk5%(9XGU|2np~T3p>~oq`>{
zCc1VU_@mUQ$0vLW;9@u0-=SW_W$lOOg>DZ{rXSyqXsxXd;*|Kj=oNTge|m7ePbqzn
zN62rl?-m|gbgM<X?2fW7wfV_R?xpVhUTXbLbNhKSw9+0ZlE{En;D**7srh~*K;M8Y
z)L(e&kUQAkAAj*Re&<cRzXcA>*8rzM$E%+emsH#SYxi77b|>rGQeYf62E0|lY|^`t
zlK))eGY=|<n#PUJ6Z`tg7HF86(uBzvh^R?li9Qn5z5AlZzx?R#gmz9}b@!>;%!leB
z`k1Mv@3D{jv~9nREa-c>t(>Y|WgeH#efV}O?YK2-;;+>l@7lAK+R;Auu>p<ew9Q-#
zODM<?Qm^D2zc!m!d<SRy1EP2!@MHdLvL1;s_mgB!le+f3fSDr*=-WTqb3a5T!$Tlf
zIx?U#_w;QprLR}(UP=W<{@h8A@r8njs3_7`c><GPBf?V#U8}3CnVFeMmuFknGJ|j^
zNlbf5U!RWd|MP0kBtnCbul6snbVV=zL0T+r;MH1C4q?zq*TsR_wGc8T5IfU=(k924
zPE6bcPa3=azFmj7L>jA?LTbVCSx+J-IpJQ-g?=2akDuSCBujUKKL$LOc+c3hXRN(_
z{0DrU8n4_E6nN5F^ldvIue_+DdY&*q*^e!ABJ0$ADH3z5ZL_&88d7~I?J?<^DBjR_
zX3ih5&L6iO@AiP21&N}IA#@k?C;*_BZg5J^GzjBNS$w|Bwx5{K^q7e>=9Vi122cBi
z8&cUqZm53^O9aSYx54wMtn);~(MR-toC{~-<R1~QtE=-~e2E=q-v3s@r=tn1$ftX0
z?J5h${#)jBOR>%u;TX9E%}ehn@hkYF{LIZmmh}6=dwcg!X(HaN<{g_WZjvlQG^98m
zLEIFi<S^glcmIM(zEdRH8p-Sb8}|a9eUFO@ulE<+1_)MUrKFkEL%J-$Ig*;ku$y~*
znoAh<O)&RVrMzjQgkQrDt|{>#MTQipPV5Xp<Z=EBsZHHcq;YB)ehLqS*=@d$#%_;<
zXl1s2MyZbkskJH~H-XlB5mhq<tOQPCK{E`=_#5J9)wt*6#kp))^L|z}l*rgVb*-Ou
zE6{^uQwYfATr9J>FU5-PfmbU7rf^R`-i4w`3z?B5hEuLEU`}}Hu_sb3=d4TFx{|7o
za?+XYwKElTnrr?rlOzWsEr3NAgfzrrnakVY1NaJ(rS1z{J+Fw~4o1hWr}#zIj3eS?
z3>?;9<Z!B9I+|wP<-n&tZ5z0m!5;=Ko2y+4R`wQ`Ue4_`BbSl^spC08d5YDwL-=%x
z@(n=JpVr8n3&!wYTrONo!4R&<e;;naUl8no_FH*g$G<?cF6=t0udhGtz?SB-_>p|e
z!Kc$$qSU?anNN(axDKu8o+~momnKVDS7do8KdXv`5u!sc9&8DEPf4*?;^1byT~c-S
zCn1+U?Nsz1j)SnMm|H&A?hn@jtq8N(C-vlCleg`>Zdu$&l{syf`7+gZM!JJq_8a2x
z2*97d?y}G^J$8mnjDS~ySyX%G45|RcRbKp-&KS0GaCUY-*)g5p2@~Ccw9L5nv(k2|
z$#%eo4I3q;SmCQjM@M7RXKm;I%tbPHTh^qU8^Xz>F(GQt{bYXo(&*|W{yN<`a6j|D
zKka`Rn1?0W_^7CsSct&9l9s=Y&P(->DjVJr2S<CR@%5qcS`=qF3dM3$ZR%*y)qWR6
z*10=40%t;{%O{ea9Jl>7E!>;#`T1w)m4y;>946r$;q{*O6{dFJwnEEhP$2o@pOC%`
z0W@E0)=|Ri9m1uoB?u#0g-i@I3Zt~RV1aLe^)`Q1Y@aqk8q9x}3?RcPcOl%6Bp9Yj
z?GtPF`_%#yfjBwL0#Y9Z?(cHz!WZ3|Hu4NiUfOf&Ep?+KOq8>K)g46E%=E0Zg|*P!
zPF%TU$l&L&OJkOgwo^e?#q!~xLqrSjX}}B{r!!|fr6pax%IUpRzvjQ%ZyRg3lm`2R
z4{|w6o@$0zuh2i0u&1Bjukh~S3>#o0?5W1%&E+s*G|oNx{9zYrnBQ1PYP+^{^QlkJ
z?CgA4#Ljv9^?Ar;$;H3DwzLJ_<CQ<-1uZKOP6@F@%K_Vdbkx+;7m$v~7*25K=Pe?L
z=S~Y;PdoWxe5nN+^RCfBKDjHTKm504Aeh?KC5KLj@?Dhh7FA-yl6Ncvdot#2dCzj=
zoYIxe+9`>ZR&^{V9<Y!24N9dCE{1%2TTAmjJBX>>D8*<b<)-F`=a!=KI2kOC9aK0!
z(I5P|cQs1bp1agXyQu6HL~3u=A89SUR7~&Q;>=fl!%CV0bHQbh8#$&q7kK_RWC8&W
zBALz2O-M4Vc;~3!R8EaoID>QMf^Jf*^qQ#BQd1Tuf<N?Qz`v97aiLpEB*&%!+sMhI
zKGZ7)&o6kC-34f<=7`77b&L)W<mP3CoYj=8(U>)LmiH(RnJN2eCEYMG87CG_ibRo4
zIXBpRds%CeqIU`kH|uQm!jA<g1g5O)9h`es_a@oRBq!tUE%<$DB<NYRe5g6NpMw>e
z5Z6as2}89ce`Y!lUJie%5t|>e9i3m)x?U9S@+99fRWhs@T5nvYOmpuS<jLi_*^wo!
zH^+5iMon(y+P~~<l}VsSa$VHryY4ad{%^|Tz^L-64tL`N$(-~2wJH_ke>J`~p|3q#
z&T3`Eq<Bc`r#8{3c0xb5|7v@|bbx~8E8!-pu7)l6TVcUVIAMh3{X+ct<*zj3ZZG3N
zbv>CUVY~r9Em0iK#f6A#q?zxO_P0+E61V#Y^i1$9M{)A7M4+mU=dq9NC$6;<QFBEr
zggZ;>)AV4Q;)j#U))o*jmG;d&pzgT)4!`xjb;OLvx1Q!+y{v77AdzIb?RpKRLH+|F
zF|V`qS|MYcuT;@I!DZuq;XP*0Cdq@{xdX<t8nz|;gNt;Rs25IppO)7->L;hZ3!B#`
z>hm9;5VFU3o#_;a+p10L?Gi`0i+Z}cAocZ>+@EcBKS!A-0BhkIz2bNL)t`9pQNL^M
zZSTH&VN1?d;f<GZepPk9_dZB7J^5xCJCEI6Mxy!WjDU{n)y;Bo`KNMI`eoU-UW5_u
zrtdqi?K7eoZlAVGpK&CouK928Kh;Yh>Qjx3>Ud3Vc&jF`(*G)l*F;55ZjgmsI-A_c
zU6|a!pP|o@Fq#`h#9`136}M_kN62B~gjP6*Cws8^t+VaH(39<?;m*Tp6Pv`24bnJW
zZ=&fsTwi-JEAB}Ns_pc{PGR0g0%wHOjINDe7fm|)`<M<)Jlga=W14KS3e_M%{Zj=p
zQH4{4a@(4Z!JgDJ-FCw^Tk99L>E6E&E6_R$yjA10nM}+@7K9hB?`ooBMVjY_ipg}C
z+=#&1c{%?tes0kH6?R=+!|AZ|Dl&r{d~*el_TOIn&pLMbiVqlNg!$kYTz&uW&B@8-
z(qoEvI)5Bfmi(YY{k7Z|fv&|fS{##J@z@_1qKky@KILi5@P>+qU=X9~D)N^(cLrVW
z%M7mmru}pO&IpdHC;5Gve4^LSMIE-s@=`@xetY;`=%q2%qh1gkT8i?^oT}Vj3(Y-=
ztlu@iW=|E}I?s9MUMAld@VA!t5hwOE3m=CwZ`8Yn1$C9bbGdGpEhEKwvJ=kOfBC<?
zlTwoC{dn_zWYhhuAFJlI^scsLq9#}#N?{1}fCKYlbYWL%?#f{iS>H*UZyj3l^ZR4l
z?w(V7Zq%gF{cD4p$!|@4E`Lq_`g^!ncXSrpzkTxS*j`-DvEK2gWcqn)kabEG0^vw^
zp9(#e^599AM5aSEzwrmt02j}*_s=#P>?~jJylBXBeH2o^ON`r>KF%Pd^DBn1UGEh}
z99o|q<ytAq<i8gtLio<tjk(lQK4PA(o?CTWKX2RcEm+Q!S4<e!wB;yFyvY7_6yT%Z
zyYtGv+?3$6b#_nuRks>m^y*3*&z&xr!1LMDKIfqGWkIz%_pLzx1j|ofaD;B=7g*Xc
zV5w=~e|vX=;*xtWv!MA-pY8qFN_xIQx2l@^UAJ((a$9@Xf-6w1XTsvdEV3{izPV^F
zZgw=<>I*mshT(qD6H8kBEI$6yD)#rIj-KFIiLeIS%0VLyU)~vJKQ7MjgMOi<TTWiu
zZoWSsnW<u|D&EKY+be+^N1|jUB1KSSDJvlRI4AAxcp|&hk8%$yjS4+W?J%3*cM+)e
zdTocb=Wr<R5;h*s&k~W~7`aTk0NvxZk}~=z$#Bqre!QHrW!tq}melCDiTXcOBG0hk
zs(9hI{FUG6YA<bnnN5ZgW!o=<RWUo4CeQljOuJ4u`&yChjaGz1423ZlnfjX<a~r;-
zT{(VMiG2Cr$nbEl1rI_K7?6ubUb-P@VQsl&j`On4apq@KRJBr!P4o2?_G$J)qeldW
zHfhe$XX~A{>uuw`sUQz*<T&U8^gToM^|#Q=b-VmdK3pH<jvw}^PZUr!tQTcQmErb?
zcWKVdeX`1uMW?WKWp`htpvi|DdwQqJvO<(J|6um;&u@>9bKJxi6vBz4M9mZCCJ5gq
z*HICUD5qzoUPvmC2*k=~IJG0Sry;w;nto?|@h_^bBu+<}EfLd4_O{%bb>FP`LK*9(
zo{XVg!wus<8mD=c9d888?@hx^uh;*)%!+?LDn{4(*}MB|;W8VU%4=$FZcf28f9!lk
zjy!m!j~hp;OqaTA?wxUGxMWRIw&d<tlN<FTzAm_qN*#}dE_Kx{T(=!BUGP&)xD3X(
zWB@Ji+P7v_H-#N|T;G5UlimrJ?pCQM3cbDzxRy9_{Kjo%)r;*MAd2+#U9bMSrskeC
zN9jR)N$+W_RUk1V`odY^$3H?xEf9$8EL|Z6p|5O&=F!@{tXrLRYui_=w9?19c*f^@
zh7OK&D_T>1d(XHeM1D!kcOIF=Rle&nudAQI2wc$)v}B)Zs;%vTC=AIM7S)NYM3qku
zF~)<>yrs&7Xxzp)B#ly5r$a)Eda$~NaCziegmJjI4xZaSXUdE5sgJM}QNUs3Tgcx>
zbtUU2rY$PlXxgUreXSTAVZxGnu9GX?X_iSB@Jj}+mb&BuBg8$h?uc&6VFbjF+{?O~
zX6Q-zA84L)LC?Ocx-@nwB8bMlTjk8%9fA|mOJyO%Xwk@RH$7ZwSga`^G8~U!vLt;~
zKSU#@i~is{c{LNu6ytz^ylI4iM1S%`)mArgrh^i8vF7KPKD%!G%i*&THErLM;pI<T
zIz{a7gRz+Fl=(#l9QYhYX(g-7KbA4$#_`*abn-F5?Wf$yWO)iRMN6oi4!EOaC#rD$
z<0qU5XS=9_n~2Yk_IUB0kYS7IjZucOM(rd&GG}(7fS~j%+=XoFtN)h^fciOblcCPY
z=erVizZDK^Y_60|V*UMS<q&ObUp$<wO`gT9CF3R653W;ep&dgyLyM=8AzKn*Y_w7X
z*Smf}UoqXMw<O)kb*4MQa%;vh>;A@^TW>lX=?61J5JkCuWu_kDdl)%(np5P8u`s@P
z`j)QsMsr&uo$KU>tS_dG2WRp#Xd_;kI*ZrnO|^Z3Y!X#dXEsxgj%Tg(_auFzN?+cv
zX8%#FH4}N4fw>^1_G5Va<!r@HfBYrO@mv5eMYQh<G+M0TBBLwFOC~3=9z?uxq_*eI
zmKi8>JqX|$T`qIkSFKn>;yM&0)!s|-@AR9Ye@ut6(H=1}^!%*6>nk3Ud1E}u7s(;k
zj9y~@J`&m>RIfQ&rs~Y1FA}IXC<46A)o^kaN%0vTCQ5qKNyGGRZB@JW6IJ6U`%beW
zZAVpGi`s4e+vS(GG4cf~Bsn`6<*olLg;eDzHjX;?bAz-Tp0<uBEZvK{6TK@{^S08X
zh3WG9e8fAW`~YQ|upDuBE353kry`&9gQnfp#rG4YS#S$qAXqW+aayabL~vx^j+?lQ
z6CxbT2Hgi3-8tiLm!d{8)OGi}c{*UK&{J7z!X=BuCaVbZ1)MLzwuperU$KQ;j^?}T
zL9*zEe~8E}b;+Bm=3d6+l`wX;WSW=#j`j+=6;D?X4wrK!M*9rg8wm<!6Ee8`aIZiZ
zZu^oW+hpUjOkUN2)_^7bg^s250IulIt<$TPQ>`(ZTOL($eu-1{0lOj@wP6tJM=?Pm
zeXuPPYH$8_ly`0y^XjzyWKEsm!{tTi9d&7HdmQ41ztQ!(aY0vcI##GG%;8ttm(2@H
z$opoJaxa)Y$+A7RuNTkxdy&Z~5HYjGU&Q}RgZYTa|95oGXl~%D+UcI5u_gcZDZ<fs
zIgWQetafKcEr2T6+UV$o5Cz=TTqt?mGpeY`*NZTaqd!9PhxGPA=$yl=?>9OiCMPGy
zu>V>R`^F0s^uL|EM4KHIQ(XKx$os${iV-90mFm@PULuv`+ob~Qnu|JgI__2xc9%oO
z^H@SJO4tc_)p;ITk7*9O`XbzT(~>0Z9b?C0SIUfzwm!HI*W~m9b#<fd>_-JbZZu9f
zA>C?iMSvq-!mfL_(1Ctm73-K1y~CZJBk@5-yyFJ)vE}IpUu@p?JT2Bd>O3acL}2`k
zQb2#llZx&vD|i2~;-syA(ZEKiCCasI*~PM48M}$2#5LOA45m;6L~#qn%EhE){w8zP
z{9?k`q>c_fh4ZQP_1+!1picu|5Z->Fx?U7(`rn6kt>QvG%8b$9795EQP7vwvW%g{*
zGSIUtegANm^I@Ro@W)*Ff;#G~%beTsi>T@`SR-~Kh^)g~)35Hvze<+6cQ;zxewr|j
zp}TvZC*Z;W9skZ+UFF?Lc7A-akY#qUiI<*P!xGm^)PXpt1n7dV8~H47Rz@=K@Xr;B
z`I@j>VDwnu)RE*SyZw*k<n{3t!$O8I&Bka<L`pTAAwkP^f6xi~`5Coh#1FYpohtL@
z+}PUUEj_%A;bIpKqW-(W_s>x(VUlV9ybzhXNq2bfJRUsS#Xd!38+Sw~$TuD!l)Z;b
zjfVA_BLyGO`Ie#M=O}p5@S~cWz2HyJt2^Q}_L2MPG-~Me3zOaW>>KjlvuO202dDBr
z4N8Sy|43wyr{!q{)m=t*5ajI%zyxd}uRI%C(-&KNdT9GD;E&^H!9IWKKJjq=2MfHF
z&-oteaMk$?Jj?0y54>f`q-erEilZ{Xoq9OLKvi>A*&*Jyy2=pe{0shpRf5~n?M82A
zwp|WnmX5LU+DR3+h$tcYm{(=p<h!2?a#hlTDvX*a!UsmKlX>!rVVSi!soQ?LhZM^(
zoDAwd?=3#0@_E?xum>s1RaCN4M^JS#VyaO_ANiJ`C2+>0m72HL&`T-#z7m!=lOyMg
z)2Y@nH`mAfeysC%rY`gT3n#}y=?C+EbwgN9uC)6-XziW^*Ooj(R+%KahDW?xO>u{Z
zau|!cye+A~xjXA@Bd-k_^_{HY>#a<o!i|nSY?<Qq76{t^R6rq0EnrZ!gV+wjpA|VG
zm1xsh{OMuwRRC9Rb#v}c{<fdddGB?`Yj=mY__PQ3HMPU0{xHewM2;1%Jx{*npy+NC
zJi2i1G0A&EoZGtOY%}Ze{?}r~_Nkuan*CbkPR<xMPwUproXah1cZj;|-Dl6)ZaVQx
zFgiI7+MyGx4#=AQcY;QYR->78v2T2K(~$h+xpl+F&oo&Z@oKTKbh)_-7lNweM@~b{
zfCJ9XBxYx4<<{quVyASdE!Z77$>ghy#4R)eScNc?n1xmIxhT)DLVw_x|2xkau1)PR
z`;w<S_<YL!QcJCigvFRAJJBuM^LArE1H~J~R}6QJ?kq}`v3z)kV)XmWy45G)YwN4Q
zx5*v(A+=a>BYBQXXR#f|0slM+n#IEOwO-+5(CA5vH@@K)9ZY`nF6FT$_GIwJbhr{m
z)+RsCVY$7=%e!xH<(7RsrYB%6Y*Uja)X&)o9y@fWn)-H*IZkAEBQNi`kMH!epa-9l
z0Lg|$j&jLB6$#y$&L_rx6?HzdQ@X2W<0q8&evAbWt<Q8;#TCk~;~Y>};oy%I(Da${
z^*7MeU~x0NNFP&B^Rg`5;Ss+fWc7m}AQq3EAtnBH8FnXH%u<8koIUSibca{N@I>t<
zcA}g7(38w>Xm4Z)g9|Ntt?=a0+xku8vNL}9>)e+6gPZ-8ib$a|3)kjN-XG1;jMvZ2
z7!2e(Y`cc2`Cc}dO{OW0Jb(1~fuvmY6@{SnJNHT(apzCJ*XA!Cw_%oL?=f8;vCpM<
zEf@!F({24}40zCPN^MXdJKoV*?piLArK;Y*`d;c5dc%0X;e0Z-%gBMwJBix6_inuG
zVYIRvwh|ON%=O){Of!vP`ID6vv@mEkp~M@3x=XuV@C>&qLZ@kj$;>uGN8pjL%h61g
z?-rco8q6ubswN=}yz2M8`S4}UpV$g6wj|B{99ns{dPbKUQ#;r}JJ{2oCt>QXxwN{b
zCDx`iT0NMdgYarGs7F!yf-Zg73VGw1`Mg)lK_PFZP|(r5THH8=AO)xULUL9W=>rsM
z^Zs)lD=*By>W4TlPid^U<&GZUyc`}LHk}RFt{S|Q@pPTkwJ$&&?XyuUeL#*jS-#{k
z(v-u|WwoLXrNWar1%-mKNFzvK+jbalfiqV1NpwPvpSI%F4;CY3?wEVx-ZMfqI{c4h
zcQEKIyEAyEx9b0>zKEb($f#+XM+6LI2CNG&^nH-Iv^{HEvz%(nI*i4C5FucdpmiI+
zzY2qtEIN+bXm=xz{Pz2zuQ<^o6G8+GM`&+H{hXRkWBq{>SzLky=S93mV1aF?Ce>BL
z<fk6o{V;i~?rt-ck2p2I{?q<@EKe@Nm8+>^*x4Iyh3RI4%Fxy}0*2$tUl%_q<YphT
zDLSRL)z-d_z#FM9kEX2UZNI`=-9H`E6peG<Z3tY%4n$cqsfs9xK1MZXILSHG9VEQm
zA^e7UGVAQ0qIS`*X#O2_@q6{86XI9Q`M-OeK5KMgi+_F;x}A7Z$hf6CPkAKquUJe7
zQ6IFi-5zFDF!#~lt<roJQ;oyl%yBtL8L}&F5NrFvolL^git0OOrmEc2KW^#mH%+HB
zw;gb#TrpLcr9!f;MQRRO&0hB2U>w8V;ZSxd;Fo(Oe{0$S;YB!%5d0tN-twu<E_~F*
zi@RHKcPnneg1fsr6nA%bcXx*Z#ogU04n>Q*p1ga{{&4pB{R8JyW^zB1Jef&W)-CJ0
z8eki9ZV~c*p^+SIdxqDu2|RXU@wCOu1_sR<HRw`90!nDO5-o9%I>ErR*~z}Wz!cN-
zJD=r#H!F=MOei}Egp;@VgKy~AChm%W7T#C@2=g+(I@odrCia`jCEaKp(&MsDW$Y<w
zBj;cLR8>zTmW=L4J5dob2^KNlF<FTe;QA5Plup;E`c&etXl8bD0zM3nideKOhpd|g
zDfsY5sZS*1C*iY~H@E@h=iiopFZ~p=dFjC;(wR9QA}|@OFacXXy!oFR@jal^v)Li6
zjh8X_Oy4QPQ{iaamx`cx8DcYV9h1z$l5*-^ym>Ad^J+8sa@bY_GDE@Qzs<B-=$q{k
za-Rek_}!r}h34mQue-MzBGPM9cQ@YD(OFpC$Q}p!x!x~L6c_8aP9SHQC)Y)Ph`!3h
z3V;l(Djn}h$BWb6Sz5MJ)yAisyWYc`HuoP~D_jL}J*%2+boUH@@v)Fi!n8XSUM^)k
zZ{77da1TL!lIURwwBW=5v~ZL+hmy_OeKA<+YsW_|oa@)st)Nyn7N^U<8k-$R{aWE9
z_r9|{7D92^+t+_HB1=UNvI=8Tiz-zE#e?w>tNS50P`idR!VX5`ntO;bYI3n&4I5WT
znHn<OZ+vD5$pmZ|2{&(TiQ^=;dkj!vJ-)m9*>xLvd!sd9QR&$h@39j6>WUAuvgs#d
zz@DhF(t74Ocf(L|c1GbaA6sdnMPlXsWo|FA>Y3B(qF0b9ek=vscwbf{u`gaKoXjS;
zM3=7WrAX=>#2XvQg&jO!F_&9S#49AtyeaHdV*8z_DYfSS7~ZkJyxIWjwiEG4_OP0!
za?u{pfOyD5Q`N1lobUW*Ic|+U+B0!cx11AS!}HK`i*rO*+`sHj+xVEc{YTjK-3NDJ
zvq7DX?=P<j<5@({>!&WDj>q3SO{a%0XNHlfo6&fa+F|X<a|5rVlJ|4tMnhj;qU8+n
z!yFTD@9Z)j?=|~Nt<+qX?YBKm0PJAnvFv8uIm?|x;U=$Xa472BoDoZgoy-n&gX1&d
z<f8wf*9K<p-(!Kc%tpoFUjjzw1ZywY@jBU!BOUzDyTLiGD_X&jjOhfr1_l<5nio1L
zn*g*<Vzl6PRpRu+xfmQl2_{GphHfP<ldv0QDXk!S<GaSnv`$zTK|umC*2$@g0*{>d
z#N?t-_>_^!2Ida!n4@prvxs5(E>#nqtx{27dbYDxxdcsAwXG4XmfyNzxN?k~x8|R~
zQ~RbeGwkw|*s=HNnwphm0j0{IGptH-0{0w$EuZyAyhNP1>obUJ+<-vthpI>W_<Iqr
zsbB>>OKVhbkA|ekx>&)HB^BE)rplpn3Kbh(lGG*^K5)Q}kuHXvktRR3qzr*|!A>A$
zsTF~hV;LTC-Ux>WK19?q?omCwFUm*<>04)$wP}4YB6IPM5h6Eg$O+yL)UR2dt&jKf
zC%-_2?_amHI|sorcfjz3fGD7d#{6Hy1!#1OQ@(hSgFoE&6;?b9=emxaZa$HkSxI|&
zQKf#NsKzhS3>C28*<MTRDRh?+M1=d-rVY3SK;+s0h5p-M^Yki~FgzECIAIL0tZi!M
z4ZMJ-N;~lz;(d$B$sWA{^_YDduC+OH8s|a?y~NT?`k%sGac(EkD=J&$7E}*aYHoAP
zeQz|HD=Rvp$bBn$ZoU#bcB7WuMQBo-1$3I5B@A%^u1JDVvTUy%WrgRwqI;C+CK;Sa
zcP}{yIy#jKDd?(NCCZA_!N3h?Hbs0BO$Kbh5@@o2>ezXT7{bkyU<6l|EiFw;y($XD
zh(%P3pm|XYinb>iNe%vfo84xUTfS<65gV6XW;y{<0<ny>f+@pc77m>22cwRTz`v|P
z@B9fr@$xkgc!=BQnWb=N6IcP!_A^9~6ToFC-hj>yD_Vavr_GZP!`cOil$OFvr)8fU
zK{q9Z8lrJCS2B=ksKF=Y+`;lvRc4OVMVIfhA!)B@t-ShqZ##FrP;LNkV`9~+3y*|5
z3eNM*WdD$Kq+ZneTZoL{oh_Ym!QGAbI5WqTk&eb%UW9yG&U5;X&iE1aepkymsQxW+
z4SfBV8F+v&W>>1a@f4%LHVwzX-vUgrE8hlC{X1sx&9gHs^guGX*Z5@cFJG;gI{dI*
z@N-};$Q3^tF1>Om&XGI%hq<yag4DpNp-Vr7#~09|XIT4`(L7Uh-r4V2iVq2bZt@I_
zjEn$wP~SqJRlRQ&i>P=%S)zU>6|vv86HV%oHU3Il#;w7<u)Zff9X7V>suxe1l*VGM
z{9`&{8%4?IJdenZ{P}n2D328equ{))aixqW9+8r=`}?s-^9qvdPz@|-GOXA<rdWyc
zIqt<c&;~!pmBVcWl+bX+;WnY7nb82UDsSjgC$=M==M@=|`8n*#@iTl{f+6?@rzelk
zR6~`ouy6gqSXq#Vh68FlB;%SawE9mGOV>v9dpQXJdz>_}Rw^U%M<hk)uDJ^+Y)mmf
z(lK!C8mwl{Veob<6aA{gi-AYq7Qdk%i?9RZptMDkaE$q-MlmcsX5+hKO!{CHgc#p^
z6}5=*)0@q2kF{oHUYvt3f3c~5`<Fz{L6jH-H_l^Sn0~<}=X_m)uzsJ~M3|pW=!ot8
zLPu(ja`cm-QbWWTW`QdZskwj^VXj&J)X0BRGka2p5>OnA2EO*}C-fIM5@@Gn4Ei4`
zx*;=U8GlK?YE&jzCOs*2DpStYP9>@k{!KicbvI1-j>vxZY@aRO8=5`5XW{)brY+X&
zGU(Mb;OCka0Cw~A4z$qROJ>v#+cFqN_y`HcqrJ}HVn-KqA3MrT8S-WZt7BWPh<-PA
z-8Szw5)=|$<eK_m{gYLU0)nY+huIX}RoxM9YXAM1r{S)pEMW+4v(sdVE2L{*L7HI#
zl0phX*<ujs*JU#0*AhSoKC-jUXW$`K6Ul9w*m9)Q)$?Hv^Mtp$jcJeJA;!vK?+DD;
z?mTfX4vBEdE#ch7X>HEE^+-5>*Gwx<QW1kXC%hZx=gk`~M(^w6cbUrg_u4&i3J4?#
zQu;NyXy2Dv$Xbi0U;_0DeavO!A~4lbp;-qn7re7EP**y3$h1ted@<d4)2eV48;P+@
z=zEk@NE_0)G(gson?vx+4ADq%)bvXA)%KYPV}DBM2i&Q*CH{#(cG!G!YA*%>-^eV)
zjkoU)TN4%2mXqkA{@J(FtAPfjou{>jhxWLVkCR1KS6^#UMq1ld$2IXi!W=-iaWJ^a
z^`^6}$-&M2&&|#gO$QV-Y=O3Ot(`x872kT1Rvh2k|EpTVX0VJPf+Lg{-5dTHCvG&!
zWR%P63cbp}i&mn1vSk|wHx*@YwSZ6}*!*TO=ZESNSZXvs+SJV&&9ZBYh=^u-^5ay+
z)XZ(6zXB{5R&U3|DPJxz<z;2Du67u)zYK3e`!z*?ShV1rcV%Vupq)ve%ql_OY7-k+
z8e>wn+p?;C32Ci9N}*Vp1D?^CV=(df?*n_)|0}+@Ne-51AV#S)uGj{VCbHg4;oGF(
zrCAb+8T;c@<w=JJ9bQx2Fop}kDz1-(iQCF;rQcXfE31x;)@Wi*n2MM;Px1s<H29(l
zj|Y@bR}bK(`I%5=q03~Zd6bu3v()4vN)TVs2JB>ll!Y@R9W4i+{9;Meq0hLH_nUvr
zV7HT|`Mje3x)8-L-m$B@=$gzV@|pe4ESC;aFw#D{dmi>^%{_^x89G>{?~fb7sicn1
z+(S8Pw%(}2Bl~ZPHROluA@Ole5uJ7M%Qxm!*g&Kfh6YU@_Dcn``xlF3VZ<(zoi1S1
z+`&GN3p!DJ5iY)$TGrC>x;1<15>NB*Gf4{uixCw?o>JU%IDG2!x|?0E?dP=Bp(fAY
z%U3#o6KhmFU1U`(sq(#*V*cF(rnKe!8~d5?p#QUN%N5eBrHysoGwsa>Y$}J;BfcNq
z?Q~F$2R3BSfx}Cl5rQlTg*bgei<Weoe8@X&S*oi7)09WAMG*g79H5G=T+xgsIfo1o
z?y?BWmB@mhqi@rquWQYnS%5sq<}i{}rc;lyjmLbF$qRXV*8@ixX&uf{jygJh!x&x~
zUaByoSO1h>Gh<<!e(t0pVD&R}BUW*;UMqX<-t#u+-|p&uCm@z8H4oo7g<qL2GT$Hl
zNHZ5myJnATZPmIjt~8uNR7PRE`cQ;l@SX*&<dx2r;JDd7!Nuz66oxg08nJ@iWNB>K
z!LmkkoaKG(Z&HpD^9==!!Vw|_E3jvv)*P0i${%7jPOr;BLP^^Vzo<1ak(xnzk}yHc
zALSW)&OdBUgN?KCq~kS&c#XP8anrZuiL{g14UR2SC&*11wlY<mz77|!fLH6@e)>F~
zDR@>Gkgi^KBxz_t^Gik}y@N}J%;D;<S4~r^7Z*u%02G=wpM*P~((9-QDgd7mTZKfp
z`f{=cTB&1%l&2h^DZ3prCKZ){NDJAiK;7(nX3p%9#fa?Zkaj|)Y7E_&F^4j!G7^us
zPn)}@nxv?l|EgG%X)geMBErkzYLW7XiRAmeqlX(T)IHP^F5yqX9Xb*ux@kOe*R=4(
zWF=0exux&St+PqctTd$h`4t)EmQV)VOC=fKYr?X$JyZT0*Z4+fgm_9x0>})7;grT4
z2{vPs=m5i@CAm{+s*lGql@-CITBq~2T99e37u8wB^rXUf8}EYW>#gIldOb9sQB#K+
z7chzHCqkXh#49G}P^{C?qAco_7IE{mSMh}88|S4>;fH2!)fn<?VyUbdDo`EW;4<kU
zj2KD65HXb^W*^dzqLQS_v*+MFzrF8^qWY%DfjC4QyZE6RXs9go!)VNlcWaF!F3h%P
zDiU|_TmaK<<Ut*r*_`_qUoiUZx#BxcLIdHplk!ueGEXyY(J1g}ST>F$AFz@bc7Yr*
zn?k?6dXoP^nSVZzNFuivB84P1w(b<#e{GaeCE@8Ujz^=Xt<IyFf;{U}JL$K?%PX!n
zk)+0U_)|?CDL)6EZT{ysnRK!8AVUzCrJyP77qY^1_ol<dE6Y(t$7T)RuR7InR!XEa
zwecjoNYY|}K33WV<d29LllX5jDXhid3c?L0R#Qg4#*q8C0))R9=h1Os8wKTAPJi@u
zV$Gpwgb|A-hkdYAB9j5u_xco*s%o6!CNTmg@WfoL)!d4Z4ovlwDQ-~pS@_@Q=W@ey
zK6+Chi~r=-fPPO<N{rbH%=GG7g);u><a?BKZk@+)MC4e-&GaPrr9QYhxN4q!?IjEN
zp~)s3acPUw+^&<zsK3eZa8h%*AE4k4tsTuIsPMOz?8l6qSUcz>=}svEE1FPnUD!Oi
zf^U_@Zji2~*T89gB{e8z)XSfgQGs&qc1qSZXHq}|nWxpynDaLg*nfu~m&e?guhR>+
zSc(3POkXV)>KtP!HgeO@g=71w1!se)6M*@~LBt>2#{|UMU*izXS>g)F1BZVu6THoM
zU&t$E+FLGnr(pSrbm-YY6iYmer1eEHau?6q^pOvv3%P^3C2wQO#!B)!Q74WBLWmB3
zCBL)goODyL{Pt=3H?h1#n4<d`^gn0HeQQoJkJr}XQ>%sj_*8PZ6gw<h<=D+Kso&E7
z)I`sxk@1jM<90-4*Ecpub+>iVth}_?a=Ij{899nlXI?~d*;U(VTPij-Qn;+hAi0DV
zyR^|gX&fmjftuyi@CT^uyuU_{;tKVB<Of&DHnd#E;hf3de*SI#9!`p41=n&jbiO3j
zdFQId%*DX-UM5k0vWrio12`R@05|0jWi~jN4`2)q6xc@LIa5=tL>sG?j2F2f8|Gx7
z>cLn0iIYOI8!NN=L*Jsj74G#DJ=Q2VT62(_r6k;*BQwl65qMrN5vHPvA0}af4CG=k
zV;|?h#};<sBx7)0AFDb2NJ7og&n3%nFdG@nkt6uR<HQKZ!i;aDqRxn!OslslRn{tV
zi6}v=A-x`DfNK}fuaD{CHoTziEy{bn-eH$XXFgRLB>cfPwlL7x%GOVn>RVPUbBUBp
zq+_LTZkCg7#Tu7(S72?TFF^6sOj!WMK+0P?aN-?sQW67mL-UQW3}0FSMmcR0p|K;0
z&M;9qe(vx-iLft~2wR{=>N~BHqvwoXS}!HVw+^j|-3bHRFpo(7*fDk_TIwtDhM)BF
z%*kI`7)kAwqST9MZ`Rte09S?RX=-k2t97fiH0lQ3-Xf>Jjriz=P0F&NOf04rDI39s
z4CSowyUGk#gMO;Slzhd9?DP8F3@y#dOyV3s+!(_Is$kbX{sysdE2h$B!cg2y%ohBV
zUmn5FuLX4^H)Lkyk3%Hs)J{js%PFazRql7JK_T4B__J3_{kR(VLQO_i0@Jg{r=`DC
zzPV)*wOkyIAB;BL4V_ra@=eAx1E-{JKZk}(N%<IO^2e;XkDUT{pQCb}_*3!!Ou}g_
zw@V-GS*|;~_>`b1=2HJ-|2-`~X%^b=GW#$U=OF7+u4`=n%pn$h1`m~?Do-2)%oIR|
zTvPSytx<$FDT_okzDmOy->$n5sS#>FCo2ZX8g!0`_YLfNR$P4N9aBR%g(xYnfenkk
zTk#vS#}_n6yiskwqf5T99On(9o-6kTM#QD@Ta2kOB_NB9UDA%9!&bln?m#EBqvQp%
z85+#3n`;dzPQ}*4w3wzcwO;I$Fo&0IU)l4P8<<qp5%9{dI}U>om2AWp8boViF{2_y
zFMd&OT&!J~r267{%dr*w%iy@;jNJaXOY!FH!l8UvC)0X#s%rV}&uq3Ixe}8D8M$$I
z!h#sv8d^A2(1+VkCN4F<OITS$h_1I7g;yj^3UoE9V68p%y=bJwT|S(vn|SF-Y6@^m
zE})ylup(-*4r2*PRoZ=xw7!bW6iJe%i^}Rx#4n~BZ5`#a7RL++?#vxEMf~3S`Wf2K
zgQI?cHy+OFsvSyNg;gd=4$qZAifg161DQHeRWyvjkwP{p0yC(or7pcjIavO<v|-Ui
zP<7#<GG!B>WU*xe%lJkiha=1G07IS;cQm<#cr;G;EvBa*A7k@xX^gB++V`B4JXs=t
z3JpNz(~xvh%zAPbhZ#2F?rjnUj??#Y53_UEEe9vAlF>OZvY=5U@!gb@XR8IZ$?jHT
zPr+a4Ft)SyCf<Jx7&!_`1jKj}A>PE2O{%3Hi2+Odoy5cCm=A*O|Az}utJxPBXectB
zM5UHnmznw8AIJ!WAJwtJLK)$%brEPeB3t)tgmqv+Q%!<FM;|7O&%#7$TE#+x3MjeN
zPjo8J4{uuKNB4doT=$mmkn@+^sgX>^Dz_3g+pqYu{DW2VS=Pl%ZE!Wv%+WT@C?ZRy
zdd*Z-fs&qjc{siK2Yt0@9~Iz_^53l?+~dX0V!%Kc82*V}F8&2Q;ID!gLIGP!5W#%K
zUo1Q~o)bf?JdWdO(|i+(Qap0lMG+XYeTs?6;z1kEnBORRyj2&aYWOYF345({-TlP0
zMdHN2u<M4GGpf|+s^xC}9MfDBX`(J_xxD*i{nPs1@=?l=DD=1Lj5t#ZLkOzj_a$bA
z?r5%obS5-yLNkmjk*sprkW;~!eP<i;WE9FwLr1?!km@B9K|?B`DK?ANuW2QJ^vU6)
zNdrL3Q~p4brY5=z&Fi?SR>~xoYlBW{;!+~ypo(h>>B(vV?N0N%ZG}6}SXzLB{qGtf
zQq~3zi@;8oZ<pHUjtQL02<!<ujfkMm@E^=f*CT&MbIgO~O4ku(Mvc^5S!tro6nL95
z0h*Q=G#2q%-e5~6jdlqBkcGzYcQcYru^%jcNLuW)))#x8n1emaVpXFJi3xRIn$~f=
zN<sw{EaNjvKsY5Lb`0hFFU=Urs;7rGO@W|{!{@Y0V<x3g6*yV|JtD*eU1OamEN-*3
z^&q(Ls}*pSOw^}ix10iNyu`g}klX9Ep*!=r$vMA%Z88$XSeur|gexATPe<1R$F_o3
zNh4)2)AF;A!CtQ%Zr`Yv67Wy;a|yx@duno6L4``L3#4sozu<7|P_Dd?ASHBh^cvka
zRT*8@<h2@zO3@QYz#Qodr$geddt|w_Kiv-{x^gnKNX`C+e$YQ;`<Hv`bpDX3mFKyh
zh02B)P@U`?IzOJ@*ec;<YBMicKepZEf{IOYUhWl+!nJK8mk5r>KSvH@&W0V-3U#el
z1$%_rxMtfa_(7+{Aa90{eX5y^$qGb?Kh+gisSO#ChNGSF$q>9pZ>0z}3d>A;Mx}zz
zHnx<bVe6tp4p#8G+vU$)V;gloQkfrBqqEX_d8}-{m_RNQ&_sv$jiV5>K#p8M@1m1y
zHNWKD$ZmYGxi-Rb$~6uI5&}zgzbQ6gK>yJQUlB1M%ZYdzzU}4`LqVFJmr*A_FJcpy
zhJ0VRvr|bJXo`7)XwS{pcnqdcNWZt%O#4?X;)0dW&}(8j8+X0seU2pPxZo71mg%-3
zX}L-mXLI^Yyu#uY<tUCfbc@gMeMf(_g13b(HQfBzq<NKF%hXlyU4rgyxC!~L0W+3^
zrWppl*o{_Z(k%IVm2)X%WbCRd;+AXjRFcp7ziTb_m#(I$5*rTBztI+6PJxr@NEunx
ztyk@(;lo^qHYws)B<}y<`7nuON5aNgjGS_w?-7!3hX1XFm;hz8tJd`6eok&>(SuEv
z;;oDOvjhE@Eq<!an4VGAsM4iMmnv8OzaMJ&L8CqY*K_~-3N6nnmFoZd^8XcTx&>to
z|DRX-FJyL9DPQ~leG^b<VnPK}_4yw|spX*7<Np|<R{(8+|HqJe-q}*s|Gv-v>*oLe
z`hkeuJKYPKykG?^dpH?rDlXcJ%5Vfk!CW*;)_m9EF(kkHr6$kx5QcM|`h3G=w^h3)
z5hyxJl-8!!2gsfC3glrEc%eZcryj1Vi-1)}EfZq~abx!Hi+BgA{%o|qN6%-6#8r!d
znFB-LhldbB^zEDp3n=j(D?dLK<tF;A&l4%NM(e*!@j@DyaEXG2%2{9@;TUP`tG2KA
zc(BbsZ?b;n3!7{gl~ww%e^-rq)T|95#^aN9gZ`xG`^VpcAtV3_T#aS!H#Y&xkay*e
zIRq;o+A^eaJRHzCKjkM9ghcsU1zYzk7o%i<{TMlV{@#tRCx63T2Iz}X3`jL5+rfZe
zUI$|kGB_$RaDS<B+NZ#)fr4Vmi2O5?<n6Sve6a>Iz+$jsoP#X3PD@&A;^oB)5eS2d
zXxaQVs>UOt+f@fYMkt#vXmcRGe4oiGjL;tw?Bl0?ZYJPyc+5>sG@l2d6!l}!T1g=f
z<0?B$zrSwG=c)kJHOsVK5tJt`Q4mmo5ibStcE!AgL&A~@43?+!u&9W4Qt-!sKkOzU
z6E9!haK=0ty=X)q{daDIsyn0f;K5U?ARA5x*5=Ja)`FKEWfMCN{VH9qfM^8DO+UQL
zmLq)SbOTSiRawK$f^D3c_$u-*SC)4XI_dB}#&=uf7$ox@WQtfPQNok(f7F1Jkgb_s
zqPC_k-LPE3T+*9*bj+BGy|tPg{U{MnFW;#K2EoBJA9C~Z&?Td{z9I=Uy|$XzX4B%Z
zgxEbkLhS~jxEwX3)VNX?Dz>h$cft7I!#^mdhcpa@G1Z1}eKzozqihI`yaOANH2b0Q
zb1sCPD!73&1!jW-rxaV*gHrg6#lb2x$wZ=M?;BNwki|VZaVHb)vrW-}>bN6N;X|^3
zC4~FJw1+cZ1UBhdD=a~_7@js@Xz>D2U|=Q&g(#SN4#QGBPgl<l*Cq=g*vzqdbLyp3
z!}t<9@(&8%>kn#+Q8l3+5T-iR>ZWC0jA?$xwiVjyT5L;7)=9Ib@4d<iJ(C-~2!J4w
z$vEDR)k+H4iF{8&L}rv)=JO4L-7^*9r#M15_Iw*ww=V8!q%tpRZfrGc-o|cmT`rL=
z1otGw$#sNg%XKzNCs%Qf^yOl|bn@!7<Vo`&%E@U44~{<!pm^?nTuLL7rp8l~etOkY
zc=gmm@I$Z3MPo)0wA3|0G779iB}DRsS0=J3fHlqO>Hli_4nt80Fnj45jj1u?>7FD8
zL(M>PxC_WCP8!*X47(zMq1+u73tw>xIWP%HwtC-8AM-p71Afm!syk>^rp*TewRlWc
z#9rH-c43FR`BI0ASRuzkZUhrs@(|@yNB`dW(T%N|@eG=eWf7$Oi<rMy<Na8pQDk9~
zCkpO!1GT)S7#LomS2k|irNi|eE9=6TI#q&Y<++W{EOp&1kNWI?<ENXBCR?ZCMkq5G
zbP^k#_Ef9Vp@>^qItS-Oe4Q%sFH+B!O4d>1EAs8mRjx0OB07^K0Gvt`gQ(~(uE&|2
z@lO}rM?`4jdj>by6h!i1Wcb!EPiJ%M7)%PUrNWGus)x?8?0U^TzF(yVV7|82mNi1k
zW<r>ZO2ZkWzG9-L5FT<TFe2&<`NU87p|NOQP<X3nLXr&Lts(~TZ<Msbh1|!~hmq~w
z`dP*?u>(^T8Qe06`euFtahDU0#A^AS0~l(H&RZE5q>BAllIgA0(Bo+S@WJ%9Vmk*D
zlH)H}zbviqpQo}Xd;O#1n;%cf3uj9($0Cu3kT{R18w+q&Ofd3@j-yUhl!bAcxi~~%
zu+TfD_>F)%d?qhxmcvgG9ua~$Zb+PLcBVd*)<vx@5NyrG-4Lbf2S$;mB!Y$%n#eBS
z>ceSG%&l64W<Odg*VLV|deiH<kW-Xa%G)ZB3^QN3Sf$JLkxiITQd5rA=P_IzG<hq2
zBCeG^o$*GDFEe`r3x9~NL~BAn<P=ifCt9YI%tJnTmhk~~!+NTMH?RQSLY8i-n(^H_
zJnDnk$3yK>gDyUX2#aol=gFKSBK)SA^=+}#ABj(TM@>HAVW4qr!IauU9g`|<>#+FH
z1~1GeEtiCiLE&pjc4Bs9dRF0?hVD&n@I(rkn;{M_7^nqVF}&xP^pKx3ukmdda?;VN
z*-uyg_Li5)3?A!NpoYAuhwZLAI7lH5T82>n;LQzvK?BBsKCG`$^D~E-)%#Sb9Q!-}
z$_6V2Jv4n=R<#X-ry^Tu__?i4*K83D6{>_2<Ik)9jb{xx-EHkSSRT`E2wO-S><x~J
z<Ah@V*mJxCdJ8L0Bt;(F|5ZxV-o|c9CpXEG%{Fr<dQm`MAjjp9Y?ERFWW+9-cy=7c
z@XqHnCR6y}yMz*Mano+|-8;cqbtoiK*O~9s67n>|a8^(}iOp<CU&UH?t{goe3^#kd
zQir7`2S;H)wyQcSs}%Vrb{oO)^a*-7X%v26scO8-xp3=jo+6HQt}g5gG=#<`tpc+S
z;oL+GDo4r7%q=99Z4z!qVi~jMnR7026KtmYDav5PkQ)JnkC*jUtyOLSxXB+m|JIdu
z`C(?aUf3!<s!~~l-<flo{PzGb{`;aMp1#~QYPEeCb>~~;EkA<QXHuzN6YM4B>Vi8z
z*W*6KttBe^kA)}7^Fi=Lb3&GqnBw{x^m$G!+-7=b^TD|Ym>{ANcbhxT%W9c(FbwZ2
zGtxgS<2|82$!LZUXMA=J5kI}0FX$zrLMBRrFXvE_tR@f7AP3!xWc8tgI9U^zQ@^+9
z#be>P*IyZ?jl^mS{JdXEeYo^Wa*xlAt92qy43>+9uHS{EOZ%t2?J?j)J9@<b3Y<sH
z_oBpE*Cjq%=LdLdmZwGd;9TqP!VP&J`l~JbQ<5Ene_^+yw&S!bsa;{9fq!CM)9rF!
z7gxQmCpW*PCWl?P#K@)}q1=kGuCx=W@GJbe>Si{3Q}5Gk7DAhcSt2`&q8d{Hl#@y2
zRbFxKqA0~yx-|yS<`9w6|AdzibqRsGHZGr;nTSfrV@Jc3OyE$V^T8JBvz}?=@JL;X
zH{_%WO-&Zby0^hSP;mCW9lDmrqCk{*J9*!T_b%QxfGk~fyzR)0YZZU2wZuhoMOU;_
znC=n8R6mJDemJPrpzxr8;@U{7yUM7<v*y)gf8^3{H9T>ft8g!5Fe{yVKIPU=qu}2{
zoMBl1h+Gs9WUhqQEp7Ti{ELDPg6hc(EuD^((S2I;({?;Un@0rEawY&F#@Ff3rPfcg
zhbRtQ&#cIQjs>uf2(Z!w|EhWX=D5AIdZ_8Dg_j~$0@G?hq&^htMpY;E+@`%n2+Ni7
zsyc5L+R191*(*~f+@naU+%aV0$O*sw3=GI``xh!D98wnN;<r$dLrDc2W&UD#y%6zM
zZJlN{u0Map9{0zFeS^idu-yEKogi7okj|B9j$r1m!33x{#bFrWrt0TSyW_ZS@2E|F
zPAV!Za5YReV14T2IM|o!MAf?Rt}MTkJxYuSA%1pHduS$e!hJ?SdokIS42-WjV~!pZ
z0{?`&cJ8AOoLuO|0lSj+r?-GN`y2P%sNRA75DrJ;sgC=`EUx3lwAFwDGwtgrRSQaH
zQAUkg!0p1C>BNC!g+FBK)KU$Y@b+A~5@gS%(|=osc~If(hlqyppQLhUrzQcLRhuQb
zwIUKrtxtycku+P|u^gY(<<UDrY}1b5%{(+2EYfqdUg$Fp+tYv@iw^l}tZ(&TzoTZB
z5b7`1I8p${7xk1pXV3S1tzBWbsTpPOGc+sTE^J4Xdi!|8pQ7}n;d9^hQjJW1wP$0N
zTupP=ZLm2hNBPr}3Dsh2nerg~rul*$A_*S=YE-JqLk-dhV?T40&y-QSi6s$S_??KQ
zg71kW08m#VO)7z8yz~-?P1@mpg)dpdHb%-^5`?aDOFm8sz@%h}KXX~+)xA3C<ZiRP
z9pRcUT$T&2C>eX!$2Ax4t7D2H9TA5=A`YFB%Khuq)(uP@X%FEPFH2Le0#ZbamC{_G
zQjjxDXxV<zy@|E4GSS2GZRW?!)YeSa&D@;^?N4qkpLlrnY1x14$%`wcynM~EqDr9V
zZM~(!9=_$Dytb0f21gzwvpeSB?J(x{iP>msrZ~4*_f8!N;itrvL28Z;uhwRoF(rQ7
zs61PBDod>*2DgXo!FVR!6|ic30;`v+_lNbUm-xhRjEms^x%1TZKc;WDpuspwhp4g8
zKDEAl2nEIE`ciQN>NM(*lj~RMo7VcW*&d0b7{iNZvBVb>5h4k*DdfZ=^3bfpaFl>~
ztT6;ZsiDpQ*4K@I4NO-Gqoms8;z{>3ehyfHESuXiTrc9G67-P9A*0KVP%l{L7+iDn
z&@&0Ow7EiEX__X%TLHsX^3hL*7=j|dTRj!>CloRs?e!Sf!*7KJP~&*QMq_gmlt{+3
zY51w_UolP_B^FF$my^vk;pS5*`rwGWF($;uR0apoLk*O93On8e>AJTJIk9qHlcrP%
zYpU>L^-?3EGXlgLG%QGKlW4wa8Axgq;~%uNT|z@W>BAlBj<h1|7B+a10&oQqr0R6>
zX{B-6R<d^C5r)|s8)lb`Jj$Bq8J+bU>?hf*FI853Q|_wm%Y}z)nm~%=Rd#K^@-fFz
zBs@Xv1V_X}ZUW<mmMI@bnKq|qu#xh{-;`WOH-=te3<-09i1i=pGyTz|h7GPR{OvTA
z)QJYK4Cf+T5AN`ly9GcqNgZ31&E=BT`V=f^lFasyqY^w>ZR+@9>m;NYtnt;`AogLs
z<%eBD6o{`y3x}gC6ar+%-L6^>#L;Yz9QN<=p7ut?>;Tv3?Xp{U-i_O?GD1h_11v3@
zwKHtO+c4_xU8+0}K|(&F(VimM<PmOL2m(yr9LxjX%+Az`_}9!y@*X!bC)j6_Yd1At
zMfh)FRaIlf?6FZ}@sL8FmOUywoKOx8Omv2#-zVYbhf+R0FO+Z_ckKwZl*1avhsT>+
z>uldz{+w@Puy-!zE;<Bnf?-!q%WT_cN@p=D<TVaRG8O01nUC9A<>v6@{f0IN{v!E%
zm$tPJ(EfG7q*+$}dV-Qvlb=>w+vIqWnMNDbxNpE~2Waz_*#DHdbj(=&x7aiBtWy?R
zkS6Y))7lz_n&xl#E7H@+`sfIN?@q@<V<;PZg%v+16>v#2#{cQjOZ26Up!ba_64`_V
z>ROQL+&g1>A6l1T-^-K8=D@4!<_>ja77&sD#=&X1`OLQx*W0WqHjK6Zs)r-wkkvkw
zQMigQ(&t0`iyS>2<(PL%ta;so;pLZHXM$H0zraXIw%*7N4qm*29b?_<)8btPzCm>4
zlgw%$t}J`7ANb6G5873>dO<e8L377;y{PaCvMzG;#NEZBHs>@Y`PUZ1wTSoB*X(xY
z=%$kD+ouk|{feEElp-O4#S~^BK^Q_zDu`_?RZAE)nOk=+a`E@m-VmyA7kq_AK5Ue3
zC=`T~2sE@QDqK(qD<TySj|q$@ECuNzkHytxUDcDM>(llzYRFpzywA)8b$+)3NFR81
zS*50CQ@i5;7p@8)JfNvCv#1wHQ{ndIWBFH){ky*B4_Wjbz0!NLpgwoC+HO1WDm~t-
z0naUE^~o|T*NWo@8$F`il~-QtaA`r?yUxO@wA(#~kniwTXXj&^<*KcWE9zJDa@A**
z6{C;p^e^2(^+E>lN3KV+jJ<xJ#7R&PxdOi~-Z{W?XDQd6rH!NHl+git<0`Wg8`IB1
zzz{506Ikcb_G^?z1Wk6urV4K~WqMnUOqwW~a+q4qjUd#ZCHxRnSW?HCy_`W@0B#rs
zFlYroS>O)|Y{UUBDHNMG_{j6?0te#^nrJ}^ncQMg4l0C+>`z_VhcfR<L%7rtylA$#
zIpJ}5DwOqPbTa32UIuJN)wJvhYSTJOf;XLgN4(lpoZx>C?x7jzq_o(fL#FwjT8ST!
zQouwhjwGmH3AAKXI4y=ml2Tv@2aQOUdc**%72=HAlU<cUGvgH_wO;T2q_>=E>iTB<
zh|g?-V(G7ONU%F7ftM_So#=ZtkyhJX@nPSIo^`<8Und7ObDM{yw12G<1wZr@4160}
zAD_1DZU6bb>epX2x|6#G9ve>N|GFKKmum8?%*tmg$GMmctiE+y_FI`+Ur~!q9Z>7*
zLVR>pOp@Y!Zsoey7dwnjdS_4Xn%exxo7~L(Qdibd=y9BQ!hE^$=~efZ*Jt{wedKj;
zpMFZ>ey7^9x{`8RxiU$;A#lg>Do=|R;Q}XB<}%+Z-N|_g#0p=9LwCw8h+-lF<MFl1
zI$}%OPEi5hg3QkH8TY|mLN466YAp+{O=ZkF@Jc`Ru5oKWeq;3T=1h{=vB7?8614lL
zBf0Zx{dHMgbZgT3R#00R-$Ms9zmu&uMAI3vf072J+$al4E9WLSp3NE1VcjonUpM<s
z=Q=%BkJ$!=l-g2)KQ1{WOz?um_D*JkBssTNR^*mLv<aj%es5Nn(oK~c2NuWdddGi7
zsmaZy^mO~pJn6p>)9)~M2tG2|FlyV>v~nMh{rh<7VXOy8ub)4;lPOQ^ze>~AwC1$!
zP`|sbbBVaap;~y?N)VEx#q_a_c@)>o`KFN;&F)+k3NQG*<d5@Z^@pN@Zx^~oJ72D@
z?{SJo?9Fs6<8|-atly~h&2W~-PsKf;AOCsid(is$B0+HK?ZSYA5J7*_Ko23+4m=IB
zz;Q87inwIN8c+n;U|@Z47I^#2C#HFW#f)pkrT^o_foFi^^IXlZKw97RKkyM6pb>Kn
z@kEWXk^Ei|Hg15$bSMH%)5e?jFA1#U%mbTlt!@W8E#-gM;Uv}b$$B+!dwVJ8nZlF-
zT$Oj`ROu3{jmf#cIpm0deQqYqCVzAARB7m3EkLgG@Q`8BwrhaK-=Bm}bOt&G+e|jK
zy?N_F><>~{301L#9L^j@Cvwq0hkN6Fq}TAJ5vZF!s-}`kUXk(u7XE-zT?xrj1`k@J
z;b;YnAsb+Ae;N#W7Tbz;3~M)4#~+$Z0xX*ylTX51wwGokw&N7%IO$SRO!4iLOs(>}
z9@0`VX<!mXh>@mj&rFAYnEcMS>e==-2TGtQtvj;WQE5tvl{6}pc>cufWNdBgf(F12
zf!y#DbXY<hHgF}2J1#nOVC7YIJO1uU9?B9?nL7vZglQB?8Wkpp;vv!=(nx6r?VH>M
z{|k#McZM<v>E#Svo%hp^Mdm9;G)@^~(l-~)Xk<(RuDW8YU9h(MA=u@<Sm&$Jc99aR
zh{*(QY1$+T+WQNT`RIA;$}ikGjszB-8nCT!z)gh+h~vMN0wLXWPtGWSiqa3-bAM;x
zj|qWEsNzbj*ta*?PY-!{#8-slpU1p3Up00piVQ8zTZ!k-+TBUBc%gY*f8|~UR-uh>
zArzU?qb-!G51jRG)p@t$T?6GXLKG<?MXF&<(1VgzLsO-3=;ox*j8%Gv+s3i8vIv{y
zzgXn<cB-4-XG`b8*YEYfw=05|9}*JKLrITHL{)bJD8+e&zDW_0ps6mv7b3&KMn*O=
zk=CFzYN4Bgog16xijgEpDJx^*2a6N}ZK3tbZSeOe%b3Iw<ZKaVN|!j75)&}*L@3D7
z(FvncLcl@?_*w>rZX}-2O>JQr)|ODy$PnWwyhp8)P^BKqc#Lh5o_02Ps@fcu<8t4<
zYyChNu#hgH^hvXl6mfr<BzTZ_Ueg5Dc#@=;+i=YT#^L}-7WG6~Xx%rX8ic8%z`Vlz
zvB0cf^HVA}7Ck#Ois1Ri-7g;fL6e@PQkXUun$cEdf=ycriZmmEg1~Gf<BYg`ge0Y2
zrgbG{^2pdB;VbV-JV2k9Oz77NZ0an}%T>oVTu?!ViSw4Sj>f0Js@)6cohP4zgWwLt
zFk$-a-OH11`jOrEd5zuE@5H)_pi4X{QvydXB7^or$GPX1{$+W^*0`do;z_<W+sF0I
zl)uZx{bsPl;D+@4j?-9(4K@*G<OGLqAzZY1CZC!0uGX=Bt!A&Ec|yk<ABoCUn%+lm
z8}IAKHW?dN9fwz({_!urXDEC9{%!vdVm6RDI>wnLv-#r7yxK-cGM7g%0%_1J(~C|=
z#xtU70O+K21eQx?ZF56t2A_>6RMH~HCSCcJyTSDGE4VezA0(`JY!W~koqTVvv%s&g
zOZ>jR_MvmRY;do2BI~Db9mq4Iw*Pv_VId}toazJWd_L03tc>!<R)ajZUY@6pLwtN5
ziJo3G>z6+|W@aak9&2=;4!Ke{1a#}@t#gT9_!oY7)U?UgZqH60n^^}>-`?uyKUbh6
z_F_;V2PVVNzke>S!N)3LSpBfX*)Jha%x0O#1|c2J+w}51e(V#{B5Pz%&I36Vk100I
zkZb>Kvyf0go?Xw!c;5Q})lC?0TRnL~lj?}8CDcJS=b2YK0iUy76|Ioj`Uf{-M{Yv8
z?8Q%Y-1iki*|I2$nr*@5=Un&wgsNKd;iAMT@5@L1xE6PMfy81{sFEIqbmLJClpOJ7
znC0A{xwR=3O{lhrAET}JlWz*!E0vU>LzB#gk<ORjIEA*5e9)imA3ZguZ<t|8?IDq&
zxK`#cA1B|0Cj<)A)CATMuLBvC7B-zn^?uE7FHG1MOI)AM<^bF^@GzjHXira1sj8|a
z(RLS7Eb@HQBtNeiz>utQLAY!oFligQRCps?Y<Uz~$_Ei)GnrwB%aD9bhn_u2Skg6+
zo{?uS1S6qCL<-h+pIVO+%6-T42Us)?GgX`k$WkIl0i8lRPDl;Hui35-08l<6|6>|T
z*(6=sq9xfe<Y_e&tdbRSO>AoGaO)=xDRhMsd(lw2!D5ign)juk=h93c{hwKYQ$nVn
zP|`*dVI=H+(x*R$CUEQ8-#5j#Ksr;UM%=^!(wZI1_9%*_T~qQSr>N+l82PcptPGYZ
zs!<5v2W*w-WZ@Bn&>1e5JT7G(Af~vgu7pYnW7#bX(kX#hp}8}Uc$rQ_5E#+P{x&p3
zbSN$vZDXau6o*8XlV`F2s&rusS?muqU?%u*6~S&9uN<MfQEh2SH675p*W>*1(%i5z
z1jzNYZHO0zlz|G77^N@q*XM?sED-qHn{h3PqEO#PcK<Dxb6H%o)(F2!)o4>UWe6|a
zRgf$nN)%lCB~l<(<&1qWkQIjDEw)N^yJ20C&PWk8J#oOuOc>LWBdnxc78p*#S+*co
zfkpNpgF-+cc5i+t+Uw&ewxoFZ47uHS6msWcW;IX{7#xvXTvr#k)}LlSZ4xgOj0%}4
z=u8Dx*~q=O*3uXaqIg9=KDrq=#1>Nha8u2d(y_Gvo!odJ;S2(}E`TH)(osLrL&>-V
zJGE#-$I!a*CtyORawb+*nNtSAc6zM2z(8<VDo!=>IY<<tK9F-hE+MFnn-S4&Mf>k;
z{wxTyC5yYXn0or9%go=pTd5CI`X?d%{OOPIan8Zu_e^~#Ph5@g*3PQ%5^_%DN{X$>
z(^7Br10>sY{YN|0sA#5lm>DfHA~J|T%06BIrOwc%A=n5{P9hB&Sx%v<P(-Xiz32Un
zw@69($k<zI-Osd=h2VVWYbRetg?Rd6)TPLTsAd{W7yvMwn-FU;SUE}<ieiQD{tiij
zMwG4$Lewx+JUWMX%poKv0akZ8Fmi|+kkVfvLnlU7nE|hDB?iM9zsP3f_JN8bt|}h@
zhC#P2r99{Ng&XP$YOvNAK&$TKYAF{$P4s-MNJjk8WMB&p3w|v)ap;&|O9!%#*8eaR
zOQhBqUnKxj<~8jWA|7QG@fEH*YgpNxfD0ZBm>7Zg+&=N5?6)<8z-U|R@GP(w`lk%q
zKI=p-A*Y+$=w~Jpmy$p}zkj_*rOJ_hw{9JzWzVqn*kZGcV3_&H{(Z%7`<wZ;a`px~
zp80P9;PJvWcmG{GEb{$HaM0(DV6J6LM_0(=OxZWCTTRPw(Mz-4k(3=J`_Efv>pyf`
zjOs)+l1X7ip_k9@;ZjU+ZZ&5;9jMC(>*T?2U+r}IeQ@<CzC%7i0}!szj9T!7E}zf^
zVZm#|DjP?0AwD~blWlW2c+2Urh0LAR%om0AOKF?<Rn^!lN(kd82&rS6nZG))7v^!S
z69N3&%th!vBo<Iyw-pR+&+e3G{Debw(9Kca{*N-KL1MdF%42u}(hgd_y<G~W>EzY%
zrk$cun8#4~A}ETDq_cwlw84a`O{9u^A<pCL-xpe>sZ@0T5s_sUL@iYj#WDBV;_@gz
z(0%<1Nb}vySnX(STW%}$kktXV>A~{A*j=~KU^SbF5S3f1D(dRdkLTsHcYciS%DH>}
zA!#GRrN@2ssgyYr7YJ1O%su({cg{a0f-W0kw^D$C0@wKbG|$-h(_`~rSC5gv_bgZb
z-xr<L@lY;!hw^YVqksG+`9VBNBtgH4al|dp<EcV3(y>*fFh2W^*y-~hju-Qi?52Nc
zM&NM=Hih7?+%PCg{(Sz{`A=O05_s~1{kH!_Pj#0=i-|lsNnl_O(j!h~um+1bDPqf3
zG-zTo8VrHdlek>AxNmWrIH6}Bh@X#;^gTy1;p1<<MV32l339k@qJ@F##(vL#95j&-
zNz5`<w%sb;*=km3(`9;HFFxq6#3%_7bD+B@LOcZd9_6~&&)rCeECP=DsGD{!p#@=D
zz_<F!+pMDBj;@$QG;>+#Q|(it=L*SUkpxqdMY+W--lXkchZ(1`Twwy~63fC~S%kTR
z<tPK;Pc8L;75HE$mU`nND&>%d=ztP}#!AY$EJaV3_(<qN>uVU2W2u9bt<itOAV#s3
z{?Z8hVE%Ce0GZY@+%Znr2Ae^%PiB_+<{ye^9RG!52of7Uy_~Nq9U0c~wg;rmYTZ%k
zarVpFsNhJ8)i<<JxcVC1X=eYINqr?(9xodAh{KsWe(fFH!Gn0g;ZPiXsgmXhUP7)1
zI%VK^G=c#GS+#RZVoDN-uni%rsC)jX<39)@FUlw=pv>@qJjudYWgv-I-dasD=D)s-
zp9ky50ceJ#uC&VM47cX=2(4ty+Hx*M4kW`v%E7x0E5-~F!hw<Vb+Q!ReOB1~T={a<
z=@=Fqejo=Hgdxb^849_i9j(4?lv2`Qb|3*lwB|t^v49L9AYQzBX}mJJ1;#$H1ERgb
z%8ZJFfZCE-tr-YOyju{AT%^1%njzqPJ#s|>Ob|tb=@CrEmsS!>a=Cdfj&HJFt|Fte
z0`af^lUIS2{5(So;}YSOA%V;pE7C@a5JbE6XLinK6>OuPLS^8QroUVdTFDZICw@5y
z2CLEU#;2tOVc4lb`tXuQCJJC@gS?6|Hf5nP;(%2SkQO?K`?Mcd+-#DnD9j6rwgRjc
zs}j<#iSInWUnn3c_CsHyak~P=1PNALVST@af)43$?`4)`m_%En3%(Box`*jo7V6?+
zuY0SoBZap53q^-2w?djDd=_v?aWpJW)Z(H$yV~=}84`mkJ)cbRzV{g(#dSFEFHMUq
z^nOj|#R5df7W|EBqr%I}8?hMBpt@hC<2McRl@BIS%4jwXoxi=r3yp#-tTEH4<^ehm
zzf(avuiG(`S*I!1QC5(k%WEMem44N6QSQfs=+xnSNNZw5E9|b2KwH3n;A}9g3Wm`z
za#ppan*p5dnuAnjevq<|q7F**=bfE9tLIKPZMumr-Fkx+yH!@xBhpt5Ql7l)EF8wV
z#r|gquc@;chLYm(y3Y<|;DUtLacWI0LyO-c9{%iK`gx9$I1Z+}!$kzmyIcA(ryz_V
z0jjG(eQ$Y&IY(vd5DW^SZ0<@|(=8UU*Vt41`_@zrPcn%4LS9ND=i9+Olk-|xSX-N0
zp6}NQ8-oW@OQ@8~LzBdLuhprN)v6YY6Np8~<dkEwr_mJiv?x371p6EM&SG~Sut|+S
zZrY71N&2M>nU9@HB}=o+6y0w+^yS>i8)4B<B&`i)9;}k1m{EA{CWeE|i5dyYY(MwK
zsN{2`6b+1<$18A(W0cWEtQV&At>#?8v67Yh=a(0U3pyo~@yJY2r3k|~<Rep<Xkd^?
zLa3!#IVZtCE`!P6N(}@X_H!m<wx11I&V!qcbsm48SPB@v^vB!pMjO1+!3L8^BnYR~
zI_cfQ=L@{q{PDS)7JE=KqHxMIc3<x$%*+C@^b9_yK&;&10C32tTkKxJ&qExciGz5d
znaAKZ=8ddQOiC|;Ov1Qh10UjJ!nc3F4Srr^@o3yGBlJxCWy}U?DUd)Ss)PTu=pbJ7
z5ol{67EX+(vq=>_-B7F6h_|e;$kP6h_jTBa<Ni0>;LW`L)-%(NG6;UGAMe2%#ZoGA
zKE-SP_vVyKS~KQjC51Qkr#EpzVH115r<((b6S5LMls(B9n(YcCf1ae-NQXAL^P~IT
z)w#-Fxn7GTcm)AOMIgzANeK2cE_5kL>vB`8@-sPmv76t=kZ^*tj@eTVd%Vs1?^>$n
z=Ev^(r{F^w%5#Ba8mvOAzCEOUWzkA9;U`6G(u#sH)!pfwkJ~`Z#7Fqo2klktF-D;_
zV~&T<hUt?c&OSD5{w~z(7{F&ivG<+%WzU<vJl}z=f9%FCyM<PV4-W{mN^S7+<UdQG
z)Y=K3?ipZ;MC~%2wcs`RCgA=aP*z42$9qN-7C$A`(DGyWH-`w#(Vp45UU#Q8)(Eb|
zMN@U}<doaPbI=Rocfp)_lTq_@bP~o{{s)f#M5o98r<0#qqRX0LzPs!jJ19*f#i_OB
zzgGoD%H`iE|5KM3DH?Rw=a`_fm3N-GM!$4Xhtv>b=B@t14=HH=o8BEqEm{Jg#S>8+
zk+lC<95!-X?qDnzmLxkPUug;jO8SHld_J7Z3Y-B5pZsTznkJbJ45)~ub_vbEvlr|I
z-In&=H*tEW3P*`YZjL|ccRp9J2peW8iiDeaOqO{rio<Vud*89jh}<|vzv`Vkj33Q>
zJZ?E)9<LvFmA|P@_*^s$J};#A45d^`I)6qULpXw1@9cX4kWoc9=QbUeP4xfCeO^v8
zzefB&FDy`@-j90To+mVV20#WIp8Gj~R<{*euirkr;Ol&R6Ue95^ZL)BcX-&QglkpU
zK6xydscppJ<%^VD6q*1Gp*TE*d+)18@o49w^$5)uH|<xa=1KZfa96F$B+nDB1CfU7
zn)(Fa?z{VXo!<*}R03^k31WaMOfBFfkR)ZzDU`?^SJXn1enC48q}Mo~cINpP!`=vU
zc1DhlQQR`X6Wj=Ag`MeO8#GxA-eTY-4W8Ow9&wwl?0D7PTdqo{1TR`dLk8sU*3T;Y
z<NxK$V6Z(7vOU7XJ{o-9;_&i;Fy^b>cQd~meLSg4m8H(Q&UkkNAe(#)(ol$eQ2)cm
ziNgB%&u(3C+tL`}IxXfHpi6YJ)clo)AyS?He`q?(peo-cjMLpA-3`(q-Q66JROu3s
z?vxhk?(UFAKw1Rpl$4f|ly2Vp{Ab=7{oruU$UL$4?zPwMPh?K0IGDbMI+PkkEJtCt
z!2_@>--^nVFgQ3kjAp&UX0Mc=G1LiZPUS++QA!Nm&N*so=DRNZtT>YlsFBgj(j^Ta
zcNlUGMJ`>br3^GYAr<sCM|2BcIXd~92o3s_dQkK%Zt6G~Fz6sX{oJP1S}CDm^LGYO
zYNvAiN(guOTniS=3-M=K1#)KuI{E0+30Y+~9A^L&s(0Gx+sqsnGF%oW=C#5AWHKHt
z0&lwFL}~(8$1Wk?2$Ca9`M9aMRi*Q3UBKU#xLUH~R837cVELp2E3FLK@>?Y%)cmA<
z+;Kh+fN14THroUtj`j-(Jzu^Z4`0YwCoT~Xh9;wzm7LlHiR0`oC2QZ#JD3`ClrK&b
z>|!X};sir-Ui}43v%p5Le`Z0!5R<7*E9kEZ7FM@(2Qu9pY<8b*o==#fkbE7KEmdD7
zV%sQCc<Mjrd`b};h~5k}kAtNMcDC=OTX0XG!MZZylkkERVZT4y{nXdX9*^kzm40$^
zhJx2;*2{4DU0LT$+{xO?xo>-Wd!1ltqG~$VN2Mf2Lac~vteVs)ly#8z<To}Zxs;6N
zQ3Xb4@LEG+*+w&XBT`1v;PZKRHNI#+J=~Z_Q%O+TvDehpP&OC&X-erED`b{7xi`%$
zt?p20N~QeWnloM*0GQ-=r<oc;CamlZHN|tMS>FOi)@hb!W@r1evvxNT`fY>9-M+f!
z!)bJzX3H@vW#8q0qDE7iNsqP6f$4PXn5~@r*T#$C)fkH{v)Ny7|HEZz$p}W1qfbcU
zN+l+MME8q5T#b#R@jyTlHuY?j@*?dYk)hR#xTHRl<j-S*t2y&=)1=aTU=bdZimqGy
z04f6$u3^26oCs&<DJ?Wx2FhAcpt=+em*cAF($bR8l$p@3je~>b#oko5+q4p&fWQls
z6Tw)pWZg)=-b{{@bR;$w3w#i&Oe{KfKcWej8QyL1p|kP#ndV3(!|nWb;XTTsSfKyN
zj9_N~>*nx#w2s1K{WOjoX-?B6S9jV7<+%lg<}(b&7J0%P3WB?-*e9jd^D7bk4jZkH
z;wSpQM+?I^o#qUzwaOkkes#2xoT8xYlKURkBZFOWRCq&l*b;q7oG&d#TkEsEF6tKq
zrHWr*A}AJ8UnJ!6bHBX3AKB#}@ugdDHon{LWYzui1v3bt1W6K2dG6QC8h-6Ezes?+
z&JuEUq@KmIBm8I&gYfOxa8TuQYCN$*J_GGYOqxhuCGoGnQKOp+*3$uitMWPT>7wkA
z&#$aHdlMV`_jl{3%Qr1#d`;n<KWXIBe~`BJIWcP;w#|z}7j{GO)aMd`;&?x*YHPOR
zGa?#*Aj4zv6DJ&lL|RXesAwQs1}*PE!)lJ_dBu1cE>7$pF-(N22ts`Np`o7t3Gy4K
zUc~<K!mt6Fa{ccBlT6aE65p3`{bdbxpvnYYIdYWisLtRtHBLOi=<7QIDcVY|eBFPB
z7M=@(t&DyqOpx<EF>I&r-$!iF1pWiao;I<_)$_@y{eE80gYsj6(9Ak>?a9JWmg~a`
zQX~$zGYU-~bUI5)6<o#E`|IoTL5U|^76y>w)AC{o)hQ0WL<B4D=g<FE)w&FjSl8_y
zn{#0SP}mNL1lFLq;uG+xIt7d*07TFzZZ1l4-jXMr|Eb;LfcWP+HwEY9C!awqy+daR
zN8U;HckM4^`yoQrr?N;e9OugRWIdAI9cA>qMj{zC#_Wi-S?3gzBR%&YfBH$uQJY|Y
z(@00@K^@HN2Z+WDQA@N-wdH?-ffx9OM#B0~P;KWGshF5pSf}+$yZd3s`FLwTjZ3-?
zdve7UzLOG!ok{G%?)Soqt)t^^%w^WiS(^IB{o1p*NE|6*(4^34WmGx2fSFzwKo|pb
zOboC1?GRjykl%u0tk4~?d^E9C+Z&Fnmv(d2rgPN>Pm8m{>D8B-e^dpkeW94x!tn5S
zTU(KFlw)tZAH`gEN4I$M?g10A=qnvtVD#K(v=qX}vTcSEMO7ApjO*%jd`#gHsRQbj
zP2|RK9kRshDCuec{K1U#32J6fch|?a^Yw!_sO0{egHqq+j30&bM0RI7&Ms&}>k$l{
zAD=tJoxjEGBx|cP47hB}ca86ki$|19bcezOoNd{`K%;h9t{*GbgrN(llBUmQ&)@Y9
zebqJ$whOq(I~2AeP#6;?tllOgGDhhslejjZ4BQBVL;T&<k%T=a7>bHa(X$sp)tCCM
z?Py>bN>^wF%xV&_2^I0r@f(_(VQca4;ATIMGk)S)c32=~syIBt@taRX<fHcM+KqB+
zeTnT%xxeuf<|@Njq`->hYcqM_<5^eyQ>T4*hf?Tln5hyQtN?>rQ`vYzEM2T%ww|%U
z&~Tav7o+Lona8_8cYcv0(dk{pscEp~NugZI^7gIYQW?f-o{(2@a<X*B)BtM0MG@5*
z$Wbphbuquk(_jAHRu)`Z%Cfb6`@q=ckrGzZ?D<lTMb39{I(Yo#lD|q2FxZ{v>eh-o
ziABq&6b@B-e`h^?t~VbHj=>%04mkXf?L0sdhR>=n6%<D~PvXAm@O4Gr+L}p!1WN{~
zAhcp=CYk4Chpcm^Zy~r((>nHr9XkXyy5`2GIUmo{Qgp=DxkeIgIE8)Pj37pJ!JpO;
zA@@!eew;=wRfBDdfw2hUlXM`bpZ$}Lp1gII&Noz6@N;5Rzf@bIX4GFrHf{}0uf&$r
z7ul4NXrfDM`{i7}TijokZS8ggC;-Rl#k;wp`CRW~M5-|3kbp?-k;D=Qzun=Je`AgT
z_$6?V1Bwq06eG^+I0ymSQ*XaaE-WY}(l<4^PcL<06i5%mQe<iQ3MHdCN#ihaD<h(m
z*w`BM6*91+5Yjdev?6SY>ba^ISx;Hf5k!!?pg5ap4s-s7+omWR7Y_A0kJcU8X;Awu
zIj*o))#LgWeS3<om<7PJtOKbQ3teW@ol$6P?CgqMZ%0Nuw*^ij@nJ`niHe*qh^1Wb
zT{4L{|1lyFPQLvF3P5~y{b41<!t#akCBe_iIwvW=OuOyboc{>->i&{qoc0!X#Z^pP
z&}6k0un%UA5(G$<bl>y0O>SK<J?H3oR@W<@$Z7e-76TWy(Zt+^k`+5bk}d@H0e<U2
zK8C3r(sn;QLP)u7Gni*pbDN0l-uaUR@zoGSGBR}Ho7<ine=0<&A(z%y1vr)I;>>Gx
zOB+pV?WVI{e2SwQ@!B_~X#+77W4Cz{ae(xJwfnOB;m0NPKu_{P{tW=Ef|s+}=#RMv
zH#Pg7*6cD7UIQ*HZ#XFeN5))nq;d2%D`V$bi)Pglq~jS0lV^2?ErTVuAbAH$Lom3^
ztd>+^OmMZPAa=*pqi!pQDWcbLJBO`m<v-JF2`u~0DZkiGwVGO$MteHzhf8hxny3t)
zP-=$AQ6u^G=k&J2o69Y#)635s(n<+h5`wAKHWM70MV7qdNvXfsZpP!f2KBoCVfMAh
zQ@(^CH`Tn`nBe}jtLq#q7{oAQXwXodV&a4kXsZK4M)r_UClm@z=NGs(FN1f9#54}h
zT3#l#&^y5;d6gNCY%(-)yp7^bjT;puNHY8z%Gly~^1f*<2uF5WM2$bO6>m#+xafR$
z=3$yP87F$h!L49T&*SCRoX{&GDQpS~3WIN*uGNhtF(rX--yp!X<)W2VsfNTlZKJ^o
zH?kCqUr4-4w+JWS@R_h&rj@V&3M{h0dy)&C@gNiGnD3(;ODH@hcOQ9geoP#1jvECI
zW?1J2HxL*cxfd_`K1lI}EpJtSMGfP}aB34POI=k$A=c5Gc=dFI7JI+<@}1WPNzZb~
zj>IX3p7G<`H!p5Fyj?T+?cSLmF1-D!{8%hYG5@ry=J)TjF})KeF+4-P*w=~92!LhL
zom6-{_6*D~DuTB?(Bv1;n~LcD@Olsl=Ggd|7y!WO08h^Z5GaAnlR03bAnrStR-yAi
zzV_?a4*;xgI~!)Js9Kg6dV^>oDVWW&!6Ts~EzSJC_TN`c!`BfniZj?%@3%N$m=xto
z=}PGNSnMb};;5bmIoobO3H)7ofhBZf@}c9O{J?RS2uwog;O@jOa+-hAZJapTWR>dL
zS`9b3&tAcAw;8o)(pTeOYf`^(#l4@XKx|Aq-OGFGzg&BMfh~FpBaH&+dE-z>=$`YI
zBnoBg^*gKua|B-TKt|*4b+|6y%U=(-2@23f=RW{>1@TM!%?Ji5A4(9Qu13463nv0>
z+1U))htv5Hy1#2tX#YA{k&-Jd+8}>>Sn3Fq-DXDyFbe138B4L?Iiy-giRaj+l)L>-
zi`(a$e|RF3)cqt4aYK4DRHo!4%v%y|vUds#^KjEHk}lvt3f80(B{7{yJqB@;-_~oq
z_y~J>*+9(m>bKKp?3<YoAFYSiM8*M;wLjCjfk1^LeY%0I@jX5IQX5+VLF};pnos<)
zcI{8%NbMF;_l*+giz$rRFr~*ey?{V0z)LsZF)iaZ3tjt`Osc@Wop0U-m_+5ZMD&Y^
zw|(ff>p$cCqv>r=FFp!*?!8_{uJiaVPuXoP4vGl00i9ocJ7e6|{(q~W0;Q1abI;#j
zRR%5gB%t^@1c-6J=X*Vq##?pE05)@;QW#unULQTnMm?G;ti8S|O;nWc^4J1!?TY0c
zdc@LlCzn#=@D)Kw1>S$!pEvI&jE_wVQL~$#F~mD95I;ZN5Z{ygp)3g<KTJ)uZC0M(
z8KNYRcRss~-`XLoQ2J(m(k^q^)p$Y=a`fMoK*2P@P*Bets4O`Wy}8&T7jzK*>mW^q
z7+eL?CBVm%_T0OEUxX!s62xR|JRrj<-rH0$gQMsnWWg$NbHx7p>xB?tn0x!IppD<$
z5#9dy-4-oqBhfze&yBM4u{-eLZP)YT$qZpv<EcuB)O;(I3&C}Zw5unqOx8o6C3}2$
z6pj?as&f`y^5YL&o<U3Nvv*OnM`_Oqz#V(Ht+x5HnB-Lq1dfKzO$?>LS>3c|3(AZ?
z9_!9;E+HvN$!k?pC!z0Y4esK6b@hvpX*?#wAHQ5919<vDqfv!5kAX@ceD;nZ5(h4+
zR8}i=l(x}dgzyX3>W!<86%OwJm-pG9KIXsK=CkEm|E;V9cqUmEsI<M7{IX3TO)j^N
zR|AGrPRrD7+R&HpbvWAa$uDA=3LP`>h3tU4eNZ^ZF{q?cmneur{~3lW;5kWR-g-G_
zWNT;lo%nkwDt<bbDO}d8YRiF(tvb9-w;H{E<wkE;A~m%Tt}93C=Fl2QYUIm;l92@q
zxhP5NWpM@avSel0FTca)Xhwun=`>Mr>qdU&pfvopZA+>JfunkR{Q1doT=ZO({PbY$
zfl;Lstw<rp?9a~x<`Vtp_k`^1?1JX44y&T$!Ur#509#do<#PvB77fjNe6?MaFS~iZ
zv0{v>%mMi0-+wnJ+m!6>?{BnO=l;_!bd6qhc-1s_adFeZ4)~*NtItuEBU&@o`AvD!
zY$<30Fx6ui+ayby?2KO>8g>)-hcaNZN)brl7-uuN4zOXyiBs8)V|l|@JlAqkU&P%w
z=~An~^%cAhB}|-~13npVVd)ZF4jM@8BncOxnVy1DFos_s_tgp~?cKH*BM16xJUm6-
zLHA2B5vID7WQYzeB54-&7dN`1bY69dFrlwrI$Kq>UCMUn5Crjw%{;1%3Fkoyg$f;M
zRH-*#Oj61`*gV~RNj1&A?4OEI4k8PKU@>ECXnOS-PZ~Lsu2zwQK!6Pa<zphfX<sNR
zx%Zz0=)OTYE*S#3Y_H8v(6!7!K0t8-2XQU{9^%s=K$r1vyJUcNs%mC~<yI-}nV=fJ
z2RTo&3P(qz3TG!l2?4uGCJUl?ggN934(nU@aa5F)G8eVY&~~n2vuQ*_Thas3%AY0S
ziuX3^EXB{?bj&VV|3PayBgf7ZKm#pGHk&~GMkKsi3`cSE#;UvhQeeq2&0Kr%W@V5&
z0GrCa&2&T<OVEb(a!N88-N{y(S&ZFLo8p&6@3ZPqX*yg*injvff$`ViC3-5yB`gE;
zkMSY}?OpncX8HmPOILK8vD$!3HUPWR$WJyK&lWO3$g!BW$08Q<;d3TG2M_z?;xwG}
z>GIlBs(GB)9No7MS$gzz#ErYTKkf?Oa5S1N@;JpJqQBtsxX{Ny%8*;}L6Zn0z%%!k
zKA!957&;!~_UH?x?RCOI;CuDTJjeX#&i$ukQZn05Me`Mrg<IxN^#UV=Op4V~Hnm%?
ziK%%*P5!<dNn>Yf4oCfVqHbr$CW%`IPtu^L6q-CdycP6+FF=z=lU|aCoy7GIMw(f)
zEHp@Upu)$m)C;6O6~BYxk(Da1GsYS~flb9;IPg9tlioT|x}CYvhNL@J1+L5e$B)Q-
zG8IG`S_wpJC!&p>jL|Qfowv=STGaV=u3;{1(aot}zHk8bB{Dc3KUO4{>`QmT<-vTT
z%@i$^b!si~DAY9P=I&mQYz8Qg&{N!^@7UZg=Qfp^S95M2@!?|WpmM%Bmg;%$*(|!C
zv2kYi98{l+&U;u-xo?|uc<$-Z(P}xp2d6Xw{4k+Ou9}*f=)>vvH!edA7$iKQkcxgN
zxer9wSrHgtGLq<Qu0M73EYegKvDDNT6Rpe{n(K9F${wfQVT~^QQbDm8Qy*w?R{!ZE
zaUrq8e|b9omHnKe(y1kolI?R=;dcU3^BLm@+Nx#eAC`KjC&ss)ac*3to*ClZDkM(R
zs7%hkJilzVg9G=)ntja2<X^k5KW96|p1i#k8<#lo3q32})YcGBsp|G^u7&TN?3Kv?
z0s>t`90EM2*uCiB!w5IfA|J{5{Czk>e@0+GGy6ubg@EU(YPB-EG<OkJblTHMZt|I~
zm^ewP<tard^!Ln=5>-DcmJ-?*$mMbdModLZ!dA2KD(@x=j)8G`CAC~`ZZ9upalYuc
zyk;E5kBYE`eN+N2f36;`1D~U~&E-wmF~}GCi<NwdY<J%U-10*~nxN@6Q2Clo>@AI>
zdKKU$8W*><WgH(L2RE<xz+*$*gyc>=1f*o}(I3Bj8JnF?t9ubE>cWUjDVP!!r@>Bc
zDdcMOF3`Voy=8f){ySEP9UM8iw4i{!{zM`}l`W8*A}BmewuiSdDBAs}p1Qohx#Zog
z_lC6IY@)XvwRj^)Ng5fE!GJyHoFXS3ynxRMt@@#z-6VD41R9xef(oYhzZ3$VUysKh
z`i_xje3i27wpA;r!iSf-Sml0eWUVDV$0g8?f%<1}u?aDi!}4sFg~U55>}AaWFQ@o~
zI4{;lvwQ7STVWDN2bS|rU})0oPTq|IqvQ91!{J1A$CC_U%c%Lm;RwbPF=x(qK2L~P
znI3f~%YicQ)=q`lGP8t0G(&H4i;W}`@pw5Qe*cmx^NuDET|eNWlZtR{;_;!a9P%ki
zJOJXXO8eO-9E)_AN#a&%)mHsw%z&Al=3kdAnpSjY+M)tQUZW3Je$gi;l>$elRntds
zk5ZT?=H4eZ;Q`(gith^u?zkfP_!UjI3|p^rk@w!!GJ6`pZEG4f4j$tRT`g>tS5ayH
zsf|iXm%Izx&i^P=S4wfZ=G--c+Ee_ZgqG@8E&v)yAyf8Y$#`xKG3E8Iqy42W7fwFQ
zR{eI^fX-_*x@UO-tMM^~9u-g$IXQUMV-{@noD%tzaQd_f!vK?-^Y#TWsmMoCf7Zl(
zY1|U^nX|n*zXm?aPLaj8c_{rQXLNLw$F_Psvpw}EZ3r8;dIIrBtj_b1LZ#@RH=$y}
z9NbtmaPc&@J2qVET5E!ltsS^DqSHc%6M1&*K#?rA^*FhOdt(2Qlz8Jp`D23#e}eub
znqiuT@E|4${e`HKiVB&u?Lc%+Pno({ffMug_>lMc&eYP<@bI{LC^s{ULAwnuL(QUP
z+FL>Mx#~V|me6%TvMMPm3K^f%O+N3ZVgudLt@O^TrO!#~XGvD-oF&ax$G_WhA(d?z
zdd2}6ps$E;70YDDUI3v_K_i>{VrgJRHY@~cy21TMD?)jx8vgU0G5CD&3Tz)8iC+Dr
z^V^f5I#VCA|Cdq}d$cUr+!7{u5HGzxO^_p`*eg;eF6u9Ar(09lwy3Sq*<glP%u<lt
z#QzF^cGA|Ch4_^RhHgr^zKx?%l-h_*2{#7=3VAVMa6YA?&E#S>h39Q}`NzQSti!ax
zSc@wD`OiAYF4}QA!ZK1)<nU;J4^-UrSgBRAU3Y|Fpdl`~!7o&x#jVM$&S(2?Uw$r~
z<;Qn<M=Hw|=ya?rVi@m9C8;})v0f7*`YjYtl{R~`7%HpZ-YCgx^;a<E^?{A)BV>*p
zy+CuTM^YEt*Xe4$F@nkVUdo^sLzTjD8vaxUZ2?(d5Cr$aBN2^XC32XW)w+IZy+8wo
zw4-yMe#(EB!%=#zhnpi0_!X!gdsS;W&Kag72goG^wuvvKP~4>`Y#A;uXyr{<2aK1~
zHyvpZ48sJI@<XN3!}#D1I{sBx^eO~-T-%q{ccGGrMAzHUD=}PRJ)arP;6Cye>pxiI
z=6u;%NYZ!+OIsE}4Kum<U@mw*J5w<cY|*ZG`%9FnsoqS94fCeB(@t2II>z+j=3n*|
zYTh9GiMce&hi^v$fITfg;>yW$^Jk`RuA!j~Gpy{7AZR^DJpo?fPL`w9_HatI{`Bwx
z+dHeW9FH>__WWYFX+b$FtdXl!-8qhZ>PYsYfta0(87tGXJ->tWG_5TTgQiFc(F<xS
zp-rb2!H@uC+(Ql(IEU<gUVfv}vF!1)UsX27?aAY>+3Lrc#0-|&n=(_DwKQ9!NqI~6
z@^Ty;iUklEM01=7f*^tjVsE+OG%Ss{13s|WFLR7U>{N~=6lNM%8Q+x{x6oEhl%WJg
zp@qhe3hDhq((ot4wA|xx_d4HW)J%wzok!ci3vMSUz<-}qNYp0R(2f7~(`U|mMxz#c
z)6?}{3^IY}-VkKo?fg%Nty{6smZen4ctJ(|P?`ol90kFp*S>$u8}itEdi`%7*x`{d
zsH<N=56z(@rbZx^K(*{IomYn8C!$z7(+3?L90xBUiq#hfgZLUkQ4;|>jW7@MYV&m`
zPg}SsskQp!NfYA17+v9XGJ*&Bg3l`>@;EW##yW`>@)LMSQp`hV?-_1zy!RZA%avZf
zCT>GyL_a~#Pt)c+#Z9qaXv*+?Zb)!kx#SjrxopoJ0*f$5y02XW4ryWLYEL$wB)Wf&
zb&HeG(N12lo|mqDtl*71@BWNOTYi`Iev467oI!A*t{`Gf?1rHG;TILxG}&(^|2^Z@
zi^{xPVT421NeltT%hAC0F!XMJOa)4BdTndU!PyLjc;UT!wD4o$*pRrNO9Zl=l)nU&
z-OsRbQHiw_lykn?vZB~P$<PT$ul8{ua;&fpRa$qIc~OrcY(@t?&BDDw%2mdMf}4j;
z9L3dFF}5>|>%OYkF)`e{R)=Z?a>>dJ;aC*I>LN$|KvOViJ3}0Z6~m=zpE6v}mb3fh
znwt0gpbP_urE7JjAqt%Vf3@?h#xnKx5B*XB7&iv`7fR69Gxkpb%u?K4@|BepsQUeI
z8X-|!VGGlC1sPW0*n8PLnCpv$>{IE<`O0ZR%oR8nx*KNHMd|_9xZZL^QNY1)pj-<c
zw~!3yhp>=i*%}c`w$UIga`bkrNC(As$GM+)DZaX3YEkSR5zLpaJJwvEzt|vYm%E{m
zLNtfu8PWu93I~#r)@EVic<C^_kj0L2(UPBx-`__=Q3&}UOOIr^x|#A6BB!hor~jH_
z)tMEyqbq>I20(L$vOR)7R49FQG}*|sOorEQ3Lwl1HJvQSO|}fzha{fGBfb0Gl-ViJ
zsm$POH8m;Ks)+~Bo$B<Ban;6HN!y^ShQIrp1ECgNk;*b|X=lpahYiX0WnHbjL^W}K
zW>Y5pi*<89uReOD;#;{h<l3J!f%hlv_aYifdo4V$orJGXpCdw-dSAfK)o*!Z`V_F3
zE4eaO^$*$?TX?hNe64u_qudFa|D4Yz`=0MFN3!VESVM)4W(`B|fqSw&etf4np0B|6
z6h|woZnyv6>?;8ihv1~BSyYYPRJHKSONO4n+N#e{9eS&E``@2J%OM6z6*l(OWhEtT
zlCB0uCApQUN!hM*8FK1`c}(9O``}p0?aHm^nAA5atG>{pz$Kpz?detPgJQuo9^$fJ
zl9pJCn)X&8tIc(HdUZ7x(v~ns8~R--?;N1u+QMXPHqN;%$4MDp<DwoHe;0N<jaI{e
z({=N3x;<1uGymci8$0*nY0D_>+9mbZYT1j`g{1}yZi*UqeS?js``k@CBEK9PkuZJ}
zxKGVEXbVjNV%FKttm<LJ=*EK2fzG2-REX3{>>png2f0zl#6{QjnTE0AkV~UhE3qiW
zBu*}^s#l9f3oAqLvStFq(yLd#%OCblx2Q)ui{3w|gqlVYW-9+p7u}YOPEssTs$oEZ
zyIOH%RWV%BV*RtYYkY>KMxbjG9<zX$Wx4gG7|S9^3NM*dF$aO>cBUwf*KY15-T{#z
zvtqT}UL-PAZd{Voi(3wEZ@dq%p9uR&rhCz15~H@+bh-WIWfO%WbY7$;w_}VUISMs8
z%%F1SsoDwDjrTHH<rw6T<ocZ{P4kt2^BT;3Sz~qAZw;J%(#T%7Wz*Jd@9!|$J)LLx
zJPDzz?rDpp#Jws0xme0kpC9s)v_6pP`XfFQ1>C{2NRLvUhJs-9sk%MeOJ)7+q->XI
zqMMOMHsZ!X)mX6&47hFN5r#D5KK8G_rG0%xAWA@sA1QY(LaXc*b<~MXRKh!PT5>r2
z;O6#HW-q9`>*H}$8Ivqai<@tI=f{w*$lLBRVDy3;iHY#w)d$+<*u}KK&rsaf?n{<;
zWsKw0aA7HRCcSWPQe6)K-?_{TPgw6Iya~u#V4x7^eu3?2Hjdo^GRM(9F=ovw$X&d5
z?G~oGplrTkV-JEkg|gDLeLPBq6RRtpjgq}~{A9fuN!ptLTf$F3{wRt8YSi+(K^07a
ztEpr4b+}LXscD8p4Nc|=wadLS*Ak<M#@~YOU_;9LQx$*VF2^KGFSMG{fu}{DTQ58C
z>6YC8VhWMMEcWk%Z`4@fvJb;-vv{EP(2c`;Uu!194X3`3se&x~<H1_dIMKu+QK{ns
zviSz5eXd<=%soPk>KUqu3>SOqtg@!EiHwJB&-9tgM)?G%7ske?k&=jtLv>o;G!w9-
zw_R;mnH^pQ-4Mnp`rUjQ6ukYDq_y-__=jn;q|qAo)5B<z8+r&Py?pRcObdyFQ^FPm
zWv=UrbXb_{1v{Z5XP|khlha&9qdt)SG$bBD6+W}H^f$yDue%F$(O#BiyP!$_i*25R
zB7*ri-$kRrVFZv1+bLDQ`d~v{$C7}nk1BuLwzhFnpXzdYLee_L`FJLU!%!)?INH4t
zH^QSHY>9&x!w1{_c<Cs#pKM;&LQ{l<%w|>`sc5$MH_G<Ihn+?1Jj>g&ZNJ9@V`OH{
zfe<9D?WMej>}I?9&1`ef<mPlidb;}BZMo)OY9U1wQ0ubRdHoOEZ`W1tUBW}$^@^ss
zT4QW4`7n5FQ5<V_yZM@}iNJthd=1eAKHO}l2NuLEXs4MiwRUI0Tpmume1hzVdFP6J
z!+^y)@^4*2v)!wP4=cidmEl`=1|`ga+TGOB(y&*{4(=c-3{-ppDi3Lm9XaEPGHIph
z(?>lshEp9nIP88rG`Ub0A&ifI%ix)`U3ep%a&ant7O=sR6@`|S%PAXvy@(Ih{JL_5
z;FTlna$_?0#h^Deh74A}xiIMt1qnO)!SXJ7|6ZggRwmQS{D@Vcv;HH<L~g~Ts0MZC
zEJ8yqk3QR$>)A^%vkqgfnNlCTGg6vJ<;ct=g%NbSBf!cL4+9seas6M!EdWy+O~w&1
zcaREQH(HxS%dkv+UOn{xp~)dxmLIme=Lr|u7^;p(Fm=->VP4e}%Dg`%q)%lsOI7jQ
ziy!y%qS)43z%*}A%F2cvkpsb0iTBDQBX!S)h1X|+loa_0R`bv|Gb?Mc)tRnj+GJ<E
zvr#c7{u%4^evN7*5_c=;@4qGgq~ye0N-6(G9GL2u_=beM$3x#kV7`hb<&VOp8{jq{
z(8>nba9OR_o_9fSquO1N3OZK#)By9;7nEq`jc4va^CEc@_;~q?pWSIU=NcCQ{^R?s
zvv$~G0+J#**$O`5U1by+Vj&k%n3xz56IDfUPmhkONQ7MIU3Y#<SzBAzN^UQCpwl$K
zZ05heo(%A3wSggR@Z4ETo2c9HVSjV9rf>Kx!_dsku2j;E%TA#v1+&+@GRZ)qgs-ny
z{GFP3i7UuSJ1j9I_mwD{vbIh{zeqKNe3>6p^1~sGx8>XV8=Gp!Q=N*wZ}8ux)T-;n
z((wX}Ayt$2PB`6N`<PUxQ##l;Fzs)K<{e)JjX7)Ze=5u;)n`ID_$E+4(mJ(x(s>~F
zOV+f*e9>^x;ad6|F2-cg`QvCwc)R`YBG~Pzc~}CaFt*#Z^{I@<)r9Mki|&>H@1$g9
zEzZzH6lO+k7KU-bX#P-OSDB(k!XWO^QlR(&k1h6Q@*fL=76GBWHw^-tZuL5F1;dMG
z4jZbxeM<o*Y649$R1-s?UA<Eq3UgJ?CbceRr*UU+j-JIh`oXg>0F#Eb6s}hvDiZ-x
zmtgE-gQdy_2AP<To7A}{_!Ow0(oAZ$p(^*);i3*J;nk|IBgntxfXA76@v{JT*(!Q)
zkdW(6o!_lXG<T?}qK1ZslnU{n^G5Or{_-wLP<A1kBQkJ8JG!h3?M9tPb2IPS*hDJ6
zA)34Usm{Wnlnw3CpPhcko=MUVoz+ir>feZf%MbZ^POFk2RCktRa$J7CHbXBJ;f<#W
z+ehXW!%a=`@#-_B-PuotayG<uzYk=2q%+iEa;!@EV*RgTeH%Q@HKz-Tzb}SD{+z5m
zN?eVLzKQy4c;DSwRrzfKCD<u_mt>=`JTi(dmR?>9!CKCB3nxUY4Ub$4;0Nx1%gPkC
zalhvh{4BE>zww!o`5z;9KCSh7k11X(HJtU{apltez4`oV8JVg}EjLp0{CR#^RaI4G
z)wlU2yBg2VS~Mj@tH8CXK{lBhdOub+p8TRvAD)P)^I`~#OIvkYL;_1XQkVBTl#MSX
z@F6UTgGh0o=58o$2sfFuNtDAr^j==~9Z1j(vK_LcN>-)AKHkGj);e-h2Hcpk4$u5?
z5t?Sn`?qzyb+e_>gO6)SRvE)=Z@pAeh9{9)^2HRh>r~TOas9S5hfrR2soEMPDFwpv
z@&0!PPB3)}Z4@m}2El|xV|r9(eaG!iR+AOg&M1jCEU&kS`2YqOr_msg1Ky7@zSoTg
zyLmk~m2Wr{a1cUG({E<HKfLEp8Upj)c@rX)G158GWY~}j0+4*OuRZL%`cxRXv}EY{
zz=dIJZ*Kv55r59TL01wn6o(vmt;TDyC`A`K6(-JG!GGHJba$Yo?0mIo?K;A4wP^<$
z1wgj?t~O^#_5Afg<TSzGk3bF~SF(aQ;WWPL&(l7$5?<o-Yh7JDChhVUItde7etzP4
zp7<h3Od3!9q=qJ?elO7p*BP75RJk?bb8&8R70Nedi!>ADH`!(|Fq(SsDOstsb`TlA
zD4?<ULlhyOJTE(g1gR4L`8VeAY3uXwJd2JR7PG#!BGpln<|rxttx>S->LnF49#$OR
z|0z5_gM&bxTrlbsE>Cb2znFKY@mIdNj0{;=g72YsZh`!ie4v&;hNz)3+a2bf#@f@P
zev=LEegj6^8!QPzu7C$Gb-|5LtSGaZDY!GOpA|FeTL}u3+yq>Gyz^}S9Tqk5O#Ml)
z%-T%}Es_Oj<?pKZr;Bxml&@Z?xvA>dHdU<w9l{7omwP^udapOzXlah_SnKt9Xi$w7
zjqv)>iv|$bI__8B_f=bgQ*<;}EQ?aiXR<pm5KDNEdfuzDy80!nhZPSAl=TWc@xa})
zOCjK})L$F;X!p)%pP}9JL|t1OAIun--JEWSKHY3Vlc!@Bhd01q(g{;uQ?s4xd;FhM
z2&&-$q+}Y7be*Y8A%Av%R#X&PyZgZasJp;9Kt~=VZTGhS4NZxK?{vR62p-TYQ6wMZ
zqK$w|?gYcxd4UJ=BP8Ima<5&S8SURJ#0+9d_bCi1`elW%n5rwRw-HqZS6<0Hm=@fU
znghe<=#$nd{5p%@uNnvP;;8Z9+|aJw{RmE&G0y>cg~T3K)LU$Q@2gw&LW?_kP9^J#
zKys<}Ylhq)?A0~p*K$JilE^d+Sppv<^ND!@F^7B(g=MN9@gUumS)?SlPI2LUKr3`N
zP<U7l>C~}F$N^~&z0pbQ>ih*wBryY>b}zw@^_L?Hl^@-QcPdx4+$EBXGwLlLYLg4C
zCop!i{jr$xJcMAZbG;d*a89d?AETgiBgrW#Q`6IMlCG}2%PkJ(ph*iF5y0N~XAjY#
z@h9cHIM(-g)d4O#C6@iAX0}%=Zumg8h0m|48RL4U=WmsG*ucIxT26b{>C<NSbud%J
zlLP22q)@A0*GVIEPy?w5lT6@r#P}%@v<PN?{~iU}4YbR)XrJ29z_WXT?%(YHdu~o1
zM1SbZ2}F(!pkTqTgJ-Uwy*(3XxmnPP88nNMnWNGLooU5<E|}k+!F~z9u9*3IU&Z*=
z617J`s(7qP!<!UBxmZWxLK2)TSw{V>GwcK1x8UH)Gc-TU+ZGR=?d{?wMg|5KBeu5+
zx*5c|epE#*F9Qq^bLVsrDeoS0+^B5LI#MZ;*5EddYjdK<b@U4ThQ{BB5E2qvP5d!m
zU0qdvJ>RKG`$=SoyNRp0wbS=9e}4rNB)hu+6l{k`+VM;{d6MVIEl3a<jQtCfV|p@*
z=E$9|Sy}Q#qt;(omxbY8wx*;crBYejOSNRCTofd!m=9jPX1X=@@|8TAlTAxbe?P3W
zr;vSUh220zkTQwNY6zjk#8a+(U6sA(ZHj+zyPLNQ%FM!!j_e!}HM4bZO8@(-?NMIf
z1vZ$4xmyJK4PO{f28I^Fzj2WhLH`;ct`G4?y0nq!?h@{Mt~h<igr+RvnSbYu+IBMS
z$Rqq&QwmkmIKBwl@^c=V700`bK>jQSIZV)$(C4TZH5dFQ=69ioz6HQU#0<`BE_?x*
z@bphbe=;4W_+rhWVg9|Hot84U+n#3EX&7hSc8Dj*tS^RX`Bc}yb%7WUPi6+SIUm1t
z3PH-)y($wrLk%KpI6`~6my3a<9`mS~u@5!%bo_Lw1<?UY4Q4t^iyD;y4o#@^K8msO
zFdNTJ*VJy*_xK+Cd|F5sAHEMM4+~X+c`*(`uF{fb_=&)_irepk1jeL!Z|N=6nB-u$
z^#V+82I=?7v5HI`Z?v)pu@f1yHFxqbjilFaF2ZPaDQeT)qT`q8C)FnXgPE`1Y6moG
z^3QiPIxx9kvW*E4z$0O8E^IO4SBV<3>Ld?pq%e^haz|(gCx_q3>s0U0RJ}T?aflp#
zFS-6%7H#mw%yqQ8;*hu$I`|#a!gt0awz_W^-aXD3hT`$@@o@z{y@GAMm{u`pcVmH&
z=3W1!i{l0pP{GtO$<qYuNoER+Ny?<J6Yoy~CvSha2_I#Ozx((u$6>S>Qc)O8LTu8P
zq#-_3DEFCI#G@LX|Gy@f>li<<^|q%wo%SNnfHwyR;W!(J+2*Sq-u=#pq(-_;@9fmd
z)N`_HTLPBv4!U>NHfry05&{PY?<+78Q3w@8JN9sO=YE0S>3UsnHs3Zq%<5W0O2Fe?
zx+~JY#U0OQPX7HS+w#TmtQfKk)d50!30xKREkhf`0!fAfhFJ4Z<;ijgy$M>(B&min
zBsWfzw4zrAM-f{hBU<?$R`{<XF1#(Nz{d;Cp$o`sZ?NEimcPHAXJ$~CMU*v0cjuv^
zhmWKL$ym`(qTnLNTqclQ(Sntlr(;N#l`(IEjD3_kOgZ)l{Zm@Le$GG&-$lZ6`H)3U
zLt{x;_?-{(*vWFYL{gkssvn#OY$V>8`bSCIA2dmf8S}E-{yvwCX?9FFJFiS`&&+G;
z-rP$Q@ilqazCS|c<6vN0%zbZ2!U(@-<CH(v-s!m31<mz2CmzsJDmvS45^E<b+VPcp
zLEusH_7;*+xMGO6gnANQ>erck?JH>RJIYWE|5c@To>J=<2Cde*{O@_dy&Q<6(g2-U
z<A+Tms3B$FZHZ8jUGh7(IDP%?Z`TRSu}pqwf#!ao+wO6M)&t7NF$&M$QlO`)l^ud5
z>tBH7vLRkm6CU6Mlz?cmS6CD0P%92p&jj;+CH3_Qptammp8k#P^+T@7v9c`kPmQ30
zrb#rn^?34+>Zv0MNNX?~$+3~IRZB<zI0aCD+7v|zJC*$L<NoljLvnKZ{p8)W$`={p
zN41ITso@~*(B$&aft7Q_OyZJ0i=}LOB5fU=R0i`l*4uGbh%FV(ddSg9*1582EL)Ug
z@CCbt98+7aEL#|nTM+fN8+rmle~^nBJGN?owWso#3_BkRZv++mJJ@ZjA#vlmjl|Gf
zk&NKV@*rR!iA0UaM(P1c{Y14v<3g9eNQF++(ehW<aiQJG5)~qFlE=<vY8A;()fm0n
z8P6F3UJDXYFDKlZHk{zyyue&SHr>;Q^NB{&??^Box^Ark9=52?4;v?VJP!4d#ewC3
zNpLsYRa5!jkz-(%d8|-4(D!~)Xt2mJCYge32~FjoW~}72G}j(j<ksu?xw(N%emk$7
z4D0fWih|P8aNwkJnO0&-7};iKV{;`q4C`8b>Mkh>`-IN~oe=}uJyhc;#b9pwh`QD@
zZSWaY*GscL+)dsa5ig5ttJQ@5lVbS1=T;a}9;E*0`!U+!n=M^uI~KQLYR`b$36IUG
z$z&$(_lPYT{gv7NpISw#w1j=BmnS^-3tV?tB%zA(mh-j76Y-{Wqn)2k4*p(73tLBq
zlvjn)RMoXvbY86&$Y?o_E?pjwaQlTWru!kCd{|X|{s=#%G)^5x4+#%Lqgn8L@S^lP
zm1;R!P<MB4w)KKVQ4)`Z;WrTdzZc*X-)BgSL0WeGA{Sko0v@Ip!_v&JD{gT@3yUdE
z<!ZkPqFrV2&EWNGL<gllNa{E>e4HvGR7Lo7?-|BCM)#T0Sy<^RPp-|VsQ@!xApVDQ
zhta<=h;B++8h*Ix6#X2qeITcs*d&~Q=F;7M**S2z&hXdRJM*Eyd8!(Djap!wfjJS<
zUP&1h{=g@{WXNxiWrxumku>PpP9fs)pL+tR;rs;yyuB-gsfFc=CoCwzw5%MyREO@~
zxr1N`BQ`m4f2Qm_gz6y(qzI>1%i%7=j5*3IuYQ@tB`B8&ZbkEs${M{khSR`dP5RyM
z%rlf)jcE_DJ8m`*-EIg^6U;*2*XbmxS4K0+EcUX>G*l+;H^V{r)11RGr)Z2*u|PfN
zZh=FJc!EFLj8H*v%SEq?RzcAAb21rajd$<W8lCcScWQ9Dtcr4GAgfgGfK7zJ&|ref
z9BobfM-Afe58r+@)&F>Vzj9}+@e&1$MiO!8qFHuVA<B8)%*VE}t#CJAgHQn5D%<F*
z7SrZ|;5>_zORtMF=sn5pe~-OBJ5!cmH{?8d<DY=%VMF!6L&r!Koa`H~s4*n2#oL`j
zzOO1obr|h+F?G_oK3My2M&XHy?ftxn^E&XBIIVv~<}vF>0mpC!NPdmh?o#hp2w&{j
zyZ*S_5R4_^v515x)bIXa-a{8BYGz}2?oH{n4jU1K0CllIPxXDbOh56H(){9N(Wlme
zmdL`N1wHJo1zcTke2<!c1CJ0Wj=FBg9C?7A%gW07bNBq^+xBFTmDNOwPYpzX1LkAg
z;Z{sw+$lO^;A4OHGM3hBIR>e9+yNRxluc&Fqx<l`X>a{IZ0jr=4P4r??WP}R)I4W<
z56#^ajNfFS^W`-nY^%`Hdz0W7p!mxK5-}L^4dXLw;HdC^B2bxBI%kQZg#-*wq9Icv
zP?{3p7Dk3L>fi<l1GtX6!X`PYLu^&kBGGFEYL*AeaML}VmS)AhH;JuV$Xyv`XrgWe
z1!78vBrCAD+>@3I2mYXVS5;Lfwdd#OzsYsnH2W>JXg@D)Xm@vcFgR;F^Yd#nFQ|dw
z5r){{)&>XZAh;y5T7?yg;n-*(1?@?xwg{X-+omEw<GdZ~cUIl)6jc{}l0~4bSyolm
zTvE-ArIHx$kl%^YCq~;Uy;PCHAs{4pluh%PE>`l&Rd97}^nELwV+F=@J}fN;^Psu1
zXEo}70|QxL%0a<y3>qt29G0Q>_pwWcc`NfhkS^Z<wQK}zgrfLa6c4w0E{gn86g*A;
z{PSvzz5+*UEp*yus@7P-=VDJwhxoD1?k+7WtP*`@j`@piMZYQIF+^tKi*LO<+kl20
zPVQjob*55E$zH?P0-S@DT23R01$hC-AklBfjmzgK&^Y4#TF7_a?Zjy?H-oi7D6&6W
zqiA0wU&5%RQ-cF-s2J1B`SBYWDi5r$`%xwg9%1e&N>P08_i)H!MU>uiT{4Mvj;)Uy
zT$w{Do<*0kq#Hyr;)7nez^3`M)cv2)vxM|+1{=$ZMA#<mCEjO625A&cX-~fZHIzxM
zQ0DTer8`w`cd%vU$^a%aGZQ){k|RvQ{&(52n_}YJ#>U13SkJ1BzQ%%3A4z2Fhimfn
z&Cql!7KLy!2n<}NeF%)I=`cOOeJU4<6HUS!4m}pevR>gMz>0Vt|GAtuww@}KBllP^
zmBXx&?lv?q7y=oV5-5KrM9+a6*}vJB6Aag_g9+<mkReDSOD;ck+RfCQE;y4z&mpLb
z1`NyKEy?o!Y6Zh?0r$&m5M$#B@bmx=jJvk+iwivz!KT6I(U&5GHz+P1CG63)Lo~QK
zY$lse4?g=RLVM^kl7oIY+ZB%0Tc=xbu&6o1RU{%CFaghZknlLlPqH&z=E!XGUv?(V
z!49GZ{AYmx37CLyZm1#hn6H{Aj@bEX-i|mcsL3<OFd^pdO_sRe&nX?RG;3R5QFTky
zJ|163Q;PT1z{8Lp{1rOh8wyX=R-7kpbbbzY9pCQq4eZB0c@lTs9Y+-IBGct`xcvJI
zgIw?wzw<H>WKiL?li!S}LpqJa0`E(v7s}?8)oK!ttlHkBN0*UShS=NXss=Y;yV8iW
z=;<q!X)~9!{u?0kcHch;Vj)$++}_@PxSa?r?dqbWLHHhyY5x3pjZB?2Cd^uiOD9iY
zYmO)-DW13cbbsQxm)sY6@`)+e$<4n)XP7}XR;3H3#wN?Y>x4{Lgyn653pl@_8h*CN
zk}ZN%k6srT8wXJfh`K}DE<v_G50|3RzfbaSfgEC7ulw$`f}CX;p5Gd$A@zdQ67<v1
zMGDZj2@Hnl%Jy2;PIRq-e9c4Ctb*LZ!5b0~#T6Bg=i(f>TQW%s1|QX19DL(i{8Txp
zz)sV*$7ucgs@{-j9cp_>MzKTR<ELrFz4Oq*p<H8@FuZBLN;LFfWqj6{Q}}bgcxeO{
zzF;n%UG$<*JS`alyJG9ThU6DJKB3k)pA<zUGW17wG0h+-$pV9D5E4;|utdrw2%nGm
z;!nRTDFY$rM6Vyxc<B3e-qMkUyT!?W%)?IMLCk6jr%xwCh`Nz2YZ4z--5Tc!L~q&^
zu}~{5OtdGac_5h9D2mB{0r!!U-<g2N=YqJH2X0%d`sd#c+a1#n-!9a6$g4EM+p8^n
zV^L`ot$CW7n;G0Z8TT_mW_cNtWpFVG599WC_7mt3f8@e|3guv6BQHZ#nUvpyhln|o
zI%+E&4BD+|AubIqf~2?6X+t`{RR}ux2r?*QQLj3-i#4yUm<|wtfS%UPB27Q{>(|D;
zc-IN=9}%Ys8fYTAsdBvq#mZ=&L>`NFsU~>UfG6el>AHJjuEuDkvrfs#G->Pme+pX1
zDV$F7k0dxky9b?2@wx8hJ_x}K=$7X3;M+jx)mnP?jEJeaV}QH|o*z5odr~PJ!M}+b
z({Fq#o%du8IFG~<mL^GlFc(S&Pyd;+;!#%Cl+Ry*uy2yE^OD(?oZwdy7t=fXHziY-
zEDoMWZl-hHGKF#_X&4p-F=I$6Pg`12?5d6<xAwVtGs2_z?Qf_|^oxx0>4q{=$(8=z
z_wARa3*{p|jiV7?aQ_S$zVWa%duL!ka?53mfY#8BuJXM)*(}fhN{Th6?*Yk8T)N`V
z6CHI*u<W1?y5@9{ZWFWoYoELfG#Q|iY19R;MP$w5F;dcpqAJoGW=R(iv<UtB*WOp!
z<Ul|eEcv?v?=&Jg@zt7N5;{jK3=P8L-KDE{c#ge#3FL!WTb;Ac9`9irEbtE+JMR@l
zK;FI`o2}3$0R>{w=zB*;rYPFs;pn-_DcWOZd%|D|m{s2sF&LnF$&(Md9W6okMex{6
zQYS;~*Sd0`%Oma%YX3`*`FkkkWYYO+JO<OrO3^mFWpT0o%yioEi-~;C{F<uOxMmTG
z?fBAoU_*gIDg9=V=6l3cn;p?=MHT-OiB{7@gxGUgcSkQxgqpHdadLqQ5fAKZMrZV*
z`9svWrx2v+lI7Lv>RPB#+-urg?noI+;;$mD@Z=pg=>)bWv-(XR0j2L5>WADww%4y;
z%fFiLG}gBGFxK*If+`d?_NN$n*z@(MswJ%A1O1%A7Zw^k7|@S3XM>?2=>HDkKR>sp
zO;FMtu1>;;$CBhW>4AZPwj2VVb_01VhY3P)+)h?0f(~lCb6Z*jpnLl^1wF>9pf-%}
zSR&C0bTN!+m)A1k7^XktY5uGCtRSo@oBsx>cH;sc=%9m&yI}ex?X3sN8kj#3+RG1#
zV$4N*d=D<^$!ZShaJ^tRlm*|2ka(bh0ml-k@ri{(K(8ncB)E0p+z`K6hfn2rJN*1G
z@qD`Zne&C(+|t3BXdDp7Cad%~JwGht;7fA>Z%(Y>CJLy9RD^Rp?iYOCZTQ~5%#>aK
zhWY{dljIL>A|R2Wb9Bepb$f~5P)Y=>2dImkytz7F#cZNRd?U6?`g9)a=(^#(8^Y<m
zKBSQDv*YIT>jM1f8FNOl{U~->$ZiKwao_p*J2v;xA@AN&RRoPml-Y6jbHFN?m;aJ0
zwyart9{{=MtyY{L<DY#ZddG!N!=NG2OdLc~rcvfuyBd3}E61IC@@<~N=agnN=N$7x
zh9_^4LM)HH{`BN5EduuMKnbAHnXCti6;*6^z2B_uXMa1nt}GcKBz&c&qWI{_S#R-7
z&&J=Z+>%AeXdsOuQx#ukeufsJ{)+jY9DKFsTof$l*wehEf3qQtqFZB#*b{JnG#F1K
z1M`NOqqwjklg%X9$K0F_l*jIJ+P(osv$KXtV^$Pvb?oI!Tg&nWG<ZB1GKI`8YV|-j
zoYi{1?V9L*zVC(xG0(T$?dAI44x01a=M0XU1z=wmFEpT&jpSfJU2*twxL^x_rxp$9
z9mJ1Wp#}%p&Xl(K5)N|jtCE&PD>f?G^j`ZF3Nj84>gc?72IpH+#i499HJa;B3#r29
z(i;AVq1SxWcz?sNhzr%dW-Q}*?u^jRZYR}TQ8#cESf}d#+EAm>3q$1dn1~YRgC*~j
zc(rKlixim<gP7%6Vos1{wX!GkGVZIA88K>(l{8ORPBoS3N)$NK24(OR<ZI>AKZDD&
zWVD200l}Grs|ec~c#%E`1Ig-Zh~n2GW|~Lsw^AKr)1Q50V^2{Q;hW8TT}3meIDyg=
zz9?!dA!meQnL!}!W3epbCjqYnu`_E{&O{4VWkj@T#G1m1uu7~&P`Q1`P;O`e>Z}WH
zlch6a6);oszMp08p}qlLVoNNMLoT1I!=ad0t3QD-1@v|~Wu-iEf0W>$fdTnVK`Wrj
zgZ>Eo9+<`WGkEQfgh{HyH?;j=PtFToOT9mhwPai-Fd)Zr8vNSDA4RLcOD30)70`f4
zQguQo2bLb5rM<;hR~`c9&jA^1F`WxC&WnG6W8KTU;!$xEVYu5F7;W|<Qt_K$u>-Nr
z`Wh^pgjSjQ7p6Q+mXYF5-_!jlg&D-V3Jd#CWtI=XtmC~!rV(QKzm;6&57r!fcQ&@1
z)cEM3QUBr@4Zvzl{!p5Ctd^ukN-txJP$&k~JM`&w{H(@cWpUrJ%8A4FdqMCbXmWA9
zh6w7J5KJxhoY-URjZ{8Q-6}u45$99W)nMvLoAa0skQ9#^-AzL=w=?BaD&AN{WI<ha
z-4h&0!@K3~kJO2$%^GWHapv#edBaMDw|iDBhOh7_t0LtkzF0k#kUXu)?t^E>qWb1S
zba;ekFUJ&l3L-9x^dm3zh5U216WSu%2C1@cL-%{bEL3)`a{FE6l74Jzbpd^_yUoUJ
zxk|o$Kzrj4_yZ|uDT-l#zH)pX&<MQK_Ur(Dv9XV@?2k>(>GI1_UkkQb=J|p?M=OYY
zxYCnA4gn{J_svE`8%X1DrJ+j%Kp^V}tA=hs6~9$lgkC%HUe<7oNMz9+d~JGeZb$PD
z=;5MZFUW7-lTLE4^^eB^N1yPGlyqeW{o*z_rVZWJVYWm+Vq{!4bDpkUuZ_4PtINZE
zej(mJZ#_nfLvgY{Y@p|Kf*tTP_=GbEd$H`)yM!=~!zbOqxRadIpqVqkhdW2ueKV3S
zmYKMc-!Pd}P5Dl4aMI=lyI&+5$1`a_;tWfT=8*8Pg3G<b+8@w|-3@k~*aFLMiVF)1
zBTN4VKHW$>{EG|x1Hx>$g$o<QBUVhRMiOCUIZ8}MKU?%XEnG6g)Zg^KE=j(5lkTbZ
zaoi)3L_cVzUG0lM5eCwLg*^OUI?j|S{V%ji)pMdw?sfF^L@>XD7a}bX91cOitlnDs
z!p)}KXuWyNs7avxv-v=D0SHDNk5|i}vJ0uKWVISi-;h+F!zL0)%`ZZ22EFF_o`xko
ziJM;RH`T=81;~2kxC7mE1&R#lMBD$^x~izGx^4{!0wN_KNSAa=cS?(ZfV6aXOLs_j
zcZhU{ba!_*(%p5|`;GD69dFPv^ab}`G3QegG$9RCj;d96!yv7I5+cOi7#!wUecN_@
zv^d}7@(FzUHu&?_uD}ZykZv1-bS+D5Jn_Q{0Yzj6#Cg~ds7N8N_p#7ew=#GR?=%Z$
z?7EY7S$=GxdDNM53ACK>^B#TN6b_keWsS?<KCvEG-JS64`npo=9!T`G2BHQx2Mf~e
z-#}TFd~YAR7}48W%PRi<Od)W877^?Zt?$~>%O}v75QfKL+a0Ifi^}~y6C`fS*-rFp
zXL~5U0`1y+u%UkdE*{D-jKrEXP-`rTanXWf^(WFpC_yLNYGP;Dg{3YyE-5oc6+t1G
z(E4c56Z)a5_f{5rT9o;^{43Hzj#s#Q!oiB~4~~xy0&0bYN|nnpQk|_xk>{`6OgOva
z6ms2(;!ZMMs=5V4p7Q(kSs3Qe-CHGkbDNu&;XW;XudJ4@{fYzYZFUHe$gVWI@{sG&
z#F4%gfpqqu{=ttZ7n{d^Js`|2{42ZecNB0X(a_TN?_K~jT|r4n%)x;T3Zj<>kNPMx
z^c^NWAxN*47<{=ed)WYV*&m*=G~M!lo}(AeV8d{Mp&k->+!RZ0t483Jn=jR)|C9Jk
zf2HK=>WXfl0;~=uEXJY}xn8wQa@<tCw4d->lKj*LQ5ANywq-@gB@9CQej9g|4^zBt
z64pU9{52QEn#l_KC-X@d0yAD-7!ow8b(1vAA0z{YlJL+v{1LDlOfr|%ylRezZ5TeB
zcj16Fjvd7y59{G%#nsfwb`4O-|BV<fpd}R?LJBjQ(_Awx^Om9)Gr($qd_YISz=cbG
zC3CSP;FqjEY+`T!OWutApWjUW6F#KA3)>FI9A}=-Snl>!-W)^QRMKyfpB}CuxOFBb
z^(y7S?zoN2!XUAdsJc3>feG2OB<8v+Oo?u`n}m|8Aj52UNM&ZFv8Mm7uSEU517g~3
z!>1Ty(M*}i`r;v0g@vjG6HK48bzloH2h^707cPI!@FaeB?z$ygBpZA?0LQ?5?v$eL
z1}NWgbUg5Mfty@2W{|=bEc4fEpae4l3MZ;OPLNZ!3>$J)2NGMRW@aYl=KlRwb-g(r
zUZ^oc7cK|kGg!?;(%MJ_9sLSYz=kBLq=W?^9zUHv05cAPwjCc!N5|*(=8?yZm58wk
zDobRd(94a1?(~EpjBhgN;wu$)nQa4|K?`{t>H1BXb)y=)GEhKJl0M1RX~w*Vb`EH4
zA7HyTFcGovBc|tdgJUt_M6~!vpSz2(EAkaLPjAC5hzM5nZtSb^7;`r{diAk+`^Bpb
zCKf7TlHiN1td`*&gy`33Qk#grtfEOh9^4WntXoJFZ`u)v#pOjKC1a$k@oOl@JI0QG
zro&gTBhvxuH(N13m3YFLP?SgnCSi0}vS9M^WHp1$Y$j3IiAAa905dSpp<-2K(QSA?
zsM4ORflB(fE@?>+bCC3+Nj0+gUe6kLEdZ=OJ}er3cYjBLu%W;3iX+V|r7xbHSDD!?
zC3&ua%!tCo2W8(@X{?$N)^RRTQQ~TLY^=@mvDfpfipp+&1T^5@zxib|nrmTABw~Lt
z-M(<CoG<<r2888n;EY}~5_tG9Jv|NjOD@wtvMKE2HD)RaH<%MAKht{$89{DUAmM&`
zQDLFZ{pDV<T1^O0-qgRI2WS@8%j3nb-=){p0rAX9u^*AYG9)Q7DK^-!Y|*l*8E`oq
zE73jer)k)2dcHecYC{hsY(MKw6-<RX_d^>>dO~Js-g;-*_)MXuuKw$K{_0MHp|>5L
za45Jp9YGCE9DR1UAv}sq+;(bR+-ZvxMFPQi*i++<NW<-{d&zk@>~OBaw-bdw6eA+V
z=L=L``ff_T(!|={su5IK%Q+3mR3gy(9IgWn*>K^K%F1X^hr|Mp3{D|m0lhg}Q&&2x
z=8Y5GyEUX&XqDfQQ7wEjo~*?}<yQw@>DYK<Tqkar_w(y0zF9>IlEH0GVc-SydX<@$
z!@*40iTjDG29R=>db}ctsaU<AsjJ`rQ^a}yFXvado;yjRTXta~qL*1|LqI30Kq?5l
zU;&W1;Ye!#M)$8M1s1y<8HR?fci@u#W*;iNYI+3JQ=3zYa%`EGu!q;Tn-MXC+qWw>
zExnLM4;08aAV&}M<3FAs?|_4xssZi~g4>S*pE|46<NRr$is~kMK80_SNxeFpj{zs<
z1rVQ>18w2waC}wF-r~`^Gg<W+|4|>wPumzD<5FnI{m^n{%_F?$6EwS56xD;zOEeZ*
zd6?hWR%8t!$~;+K*UV1LG{zh?I5s*irhE$($}q90egSH$vOI%c$G)At2?wiFzk`0`
zwQlVeeXr-1h-|$Lymuk0B=!6IMJ6T|AOSG!czGx`i(v}S&yQb~+nt{2zV*`xr|$=&
z)b^d7VaOTs{PdtGuXUEyd#jQ>-R}ci9yzvy7Z4GCl)fzIyXvjf5kdVgH12RxKQCXo
z-?55H0T4hb@VME<w*F~MZ_mUAuwHma$ud~q*?Z^rm&T}g(GoB0T7WJ@ZQ#9S2h$dy
z)9t^gn#nSlctFy8mH^BFg#xV*1Y7aKnA%r)>1%@@BWYacL%arHMmPm25DnA;K-;-W
zgI|+2%m`v41_rBv3Ilwo<$Ap+kj!7|g47O>N+D4&X&6l*WKNR@aLmk>mUL__XYcs<
z_X&Lq!LY8w{{DdGS3bL(luL;aG_$O~!*h<RmJ$ALYwh15@^0JqmO6>N3T;QEt_^hg
z1$2r!FeW}6aSP|nzhTo1->p-<)`nTg5x+?{kDmBO7Z;U+{i8ZHTHixc5rZZ~;*MYM
zW*(mU<0!@K)5xOLC(Fa{UqA1PP<K~ytXNCjxPJ~4Mrf~+69}704rt`2pjetP4uf^0
zo<D1c&&1p#{&6lZ-Y|Q>Ng?rO@ExcAhUmVT*>LvT0t`gVi=x3TKKoTyE0|H%OWnLv
zU$5_Gz9V&rTdl-Wl$E+DdvKBG?oB>B=CPxjh#2`fZs~t2ZG9=iH~n6;FbWfiPx}nj
z<+*hquEFF*2e)t*bpD8&*f>C`GglFNyYaKY-H$<iX!^mpzmNOtb|DZv>+|hF)fP7{
zuuIcutJRa|0B!!p<o(-n5Kafs@||}$i>#G`uwGRo+=Ibyh^*V7EhqX-x7pa)Ay5Yo
zk)?ax^yI_Si6<nB`S9yrEtT;wg_3kU9m!h}Fbri5R9nCTuFvve_IqV;hFI$h+Y5@D
zJrSe3eQM_oIn4|!Yw1p5LB{JWwEn&!S&pTkSPDzT)!qv}-<{4a&xX|TH7Zf(d<E*w
z)xiONW?6bF!6p|&b<y4}xiYQhxkp332DfwsWRG`sv7f1i2O8BrRcyP45>}x?$?;y1
zM}^!Co*P)EU?(PqDL2(@Sj5D-M9V*fewOP0G|DH9BVKH&)+w|u%DZZ;O9Hp(bZkvM
z!`w^o>y+HXhch%LSJM?{{SM!4jvNsDxCQ^DNij*sKaInPTm1G;XEUVH>}Iw(jiB5n
zJ?;NY!h(g<QREj;PM<;co;Z(~C@Da_F?*?tl4&8P<6mT6t{l6$cyGuNJ-z)E?RYr;
z3LPf7{XY}keOG@$eNCDT2o<tyFRnNkXQ^Mo{(dca9K@O%ZU)F|JWs2_@gc4o+m;-0
z7b<b)<<;baqDsSg-T(4<IR#Y5z&pbV#qD%Fbh_R-Y#=c&PesSMQ~J%btm*uXUT?U=
ztnQ{ioESuj)L>`ShH2>vm<@2=jje&4*zL&sYd4@%+5`myP_UHhEY;gU2mz4kS`Z{$
zQ$E%p{#}?)pOe;2Tw32Dp#3s>2s5zb0vhZ5$n_+vqK1QwTUyNEb&1Ru9og>kHF$v^
zU@8so=%PhN$8KKiH|4m7qx6$1`#mDr=X~llg={D#%kNYk&ZB&Xb1PEZX%uN+e&^@P
zyH36FJ4}2pDp<DC+;BoEK$7tb#Bqk@d^u=RQDW}SXZ2U^z0O)wBrDvG<~Ny-LN=e|
zuf_Be{y1FHw^ie4L?v`35NPN@_D<F~PZneCgJ7@yI7I3JLKmSqL1ZJOLj#Ow#AJ9S
zTRE`Nr!c+c0$jJ3?hmNMCDWsYnx@kZq!Gqd0Fi(D_RYGNzzM>A{Jh%VwoQT9B87Zk
zpFuQUfd1%Vt9s)N&yIb6k?}IL)L@4^C<o&VG_YUNxUr_U4lln0@<a2lBV35ddP#}q
z-}rPs*G6Y1qY-?_O#@ZhxKDGv?4%7K+g?adspWPi#_h0Qox+gN{F&$OA&CTjjC`^<
zcrr*<Ty%4s0_#ZdeWoH@i*~ysD>1cn3eGYq`8JF$!KRYRcrXnqW;*2=FNSMJ*WG79
z7Y9nw8VV_8au)V+Y3I)g4R#KcPMW0i;$}ytYCo=v=*k4Kr_7K{BIwJ@(SCft{c!zH
zOC;i$O|bVzXF9BMMJ5>&tB?1>a^6Yfn=>lopx9)!47LB^O1#2)_U@z_I=!#u@T$!t
z>87gBbE$<5s{%I6RR-RwuT=}xM>-CemV?BCpbwu6Ig>xCZ)sN>%geXfhu3GKnW6jp
z>&DeGnfjVqSkQZ6@Wm_G0u6!n<5}+o(DGO5_rHr(2FZi%Shm^#6gj`+1!RMOV!q5e
zFywBP)U5z<Li!J|G>>onWAp%mo8xq{0>rR4T66;~kq;iT-NAg}%ZBhlZmX0dB7f+_
zFnc9<AMKN`<54BzR2m-e#CPndX=%yA1zq27Q_6Q&UG><@4BYf>V+IA0p`?haPZFsr
zt+xHSS0j`<b|rzSwi(7H&BFm89I5VG-+;GSwYBka4i1lp>v}wS)&J811nw>W<q$%{
zApunykiY_cdoO_*@DxtX&SLK!<xZNY)N3~XWK%PHje>7{`w#=w`18s%n@V@CTnBKS
zvZLv5aasI8c-lTLuJi4<c-l3xpu4bdp|{X0c-z}3@#K@w_<A+@n3&_J&$PeX{|IV-
z_MY?)t}%QVN$XO>=tkfBCa7K-Rw-j3Olx#|8>dX?a;3yxqA6n0$urb+PGQG{c$Ae*
zBawY=Wayw;`BIXnKwVv#2^*w4*<{t<`?x3PNn?tF^}LgkoJ_^Y==41NWmGN0G(`c#
zM&qcNh4bGV*Zr&Momb+VzvohatbE`DJSx!OvzcPjs+59UlCdn2z13C$h_JI25Zaso
z#vo|Zo%K=0?`;B+Bge_0aR|oJNe3iJ>*F?~YNbA+*TtFVVCyTQ`QBbhJ?+f#1vI5T
zzwpuIt1|`M<=<39=&%8%O4L+x1tumYHo!*r`z)BP8cI<{kqYl1rch<f8G#p5TrO2~
zaG++2oxR*ti4T>xIx_U#ty*T@4+ljI<KfdNvb2`zOcD`ebK}gcmUMqmQA6>$UU%KN
zLn67|-T44o7RCYx=1>6X6OawDma27c%JJ_}bjB%w-W7()d5)>k@!&K69g0^lz*Ape
zpG3q)Yw`VL`Y}d4!^&RSSegzdg6I0adD{-)z{vF8NF5Qbt)NbX8d8Kuzm^-KmNa62
z6Jtaf12F{0+FHShF)27i&W{OsyqaHri}>bZ5>_Zb?m3l>Pfdw-?td=5+sBfy2yt%U
zBp-V|o=++1$2Mp5_lk}?wE9k>WF%CR;~Su!Ic~nvlmtmEO4dqiI9|Zh5D}_!)#>Y*
z>d2QT=`Ka{fgDqS7UX=fM3Ms?1Von@9#9*ph0TqtR@Q21AD%#`Dl8)M3NpY#{$eN{
z5F3^Aba8QUUP)89ps2M(rY~VTm@O-S$QPlNomas5nFPF-yBZ#M+f9H{|BKal{X0Ne
z_h$k;3e+x?r)0Z2?PkFf6vFRPzc+6M#Sayrr1gLQKGs;T#X7#w8O63N+yC~wj!nz`
zZ1u8qh4VMvYS!$PgaL0r7|!E*?{j1?kz**|G%t319PI#1CfvY;ya}Q(Y4itrAz_Vd
z0Wqs;eWD`Odm+WlUax>S-3vpUu_C>aItWV~Uo4*&3(>rQ_J_&s)>@?w;0|;e^^w5L
zSz>HElPW$rH`fQaN>G3|wMTedk^!qae4^Tvrqi;Rv@}vLzTI0v^jF<LN0rFy!UBpF
zn?qrOSu3|Y@b}V)7-1y^D~PPDtc(CvK^m?HXi~7#o#f4_E~jr+Jq3W6s>Rg;D4CqW
z3=62qDK$_&0SN&r;QWXHjOT=&BB-Nj9WkN!STBP&AasKqMd+L#g@x4^qmMrYv^&<x
z<P@r&lwa7_{Ya&1U4_@2hEQTevSQxLdO1SDvi9qja;4>j)_A3Awhd-y${*nV(7kzD
z6A4!@k6_h1B)V^GHMxerxv6!B{bi08fw5#@&Y$LtUD}Rws^6afuI)k53R0@`M`OJo
zk5wq8*5kRRVHYF$6#yUd58i@BSy_1}`8yB{1MUX<78e&ck;5uoWMuKovM2{G|4wHQ
zzGR-%h~%bRY-8~PUI-#{fm!?a)~*2`qR9i&R1k{}Nx;BNhGewZptnAr4?3s!0%40d
zNUr(0UkJhN!j-#L0z3A+LDh%^bpL?Xyc=<95#i6Ed@BONB-?>|1y1nhczKw7veIB`
zdb$Esn@$n*-3bbB*6Z(>V>q>N2}~T#AH0%2jplvCdq-QZJF^YkT>4I{pZ4D9y0Lmx
zCdyGEROB?g$-=sG(?|2PAQ2W`^=SS+{WW425uB(Rq{q=NB#|wnd!Gz(=(S3c*xk7y
z4dPcRd3Ej4Khs716iwrIoD`Z`Bh-`mAFd2YI34coEnA-)JD75HhLe|DRH!?fyK~7R
zL?weMKUy`Ny@qHSgb}`Q?~#LL5_rmcz3Ax)zEe<k5EiY<dy`{KQq>1kd<h0>{zc~7
z^2qzI?r-4+J;(geoFzXwEO|W0&yKBm;L#rJ8(UZvR5l_tXdo2=u@Gi<TpKFPLML4W
z|227$N{Nu)(OUxM<Qd1DsCcDF6mp$&jOVuupQNvy=n@7={9ms}NhS3oALyxo(={QK
z3I)J(;L3%d&DArHmjwa_|HiVy|CEhkcU1_z?AUT8j6r~928Qo1W|R`@YHD66usLGf
z0XlsU4*<L{RqF9F2vmtuI0xdu%?EynDjf)5?Pe7uwqZH<!y)l#5cP9&2Pm3AXMG1o
zcfqf@J+7Xgw25DzxjLOd(~O=QcxIK>raPG)41LQROLaPSIkuCR${Oz{FQ&FNbR)|9
zu^4UviKquKVXOiqatP$IuiW0iEUbE#?!k7akMSnU!b*HihVssc_%LooT%=M)@r?v;
zwv;SR(a*+DO0r2}(*=6+3wM%WV(!h5VgXohPF`MYHTm8(rCH<93O;^5&iDO_XL)j^
ze7*ig8H30iOM={S+0P|d$2>4-Kt|@%Ng?B>A<=W;OgV3Kw$TlASYHI5>;q7VzEB`=
z-j8+Gw`PH;*CG|1=6VUrZUk7G@V)se$_nLMkC1Dsh!<w^e=s>=4KMZXCdEZgv#q8E
z^bIrYe`t&PcaWKW^TtuUu?tWbzX|S$EFknub+`)ilkfVy-q~MMCt|>bj{bqsfYY8g
z#(EI8Btqi5?3-WldHSUQO$wCxah_HPLeP$CW|J-8S}QCTYma8!Xnr=M{blnur_(I|
z0S#(Z6e71%?_WGf*bKQq$Pa#ze$`$->y02nZ370yfewG924M4YoMZ=;2vF!$84lxu
za|?3kTD9C!fExj#G6TWRHW2ORljm(Gku6~8PG&OIK<+&N2Y381zN~><)7=B@7jAG#
zpo65KzQFyqL-d^$AG-ZLoYM^9LbUZf<T=?GG`W1iyKd{#=p$x|YFJ@nVFUpaVahRA
zyxr<TCR!4~qTIGnAtz`cdD$H;=82KCYajS6jfkD@M5v<iM9?;+*l54Ep~V4&VF0fr
zY<W7e;_;dfC*Zm5;b>WRYn;x5i8-+c@XmOSc-4<~AJY9(2+#wWt<VI--+*afmQ_||
z4+TUqPz#UtMUg?KLP1h0Azwj~2{5$<LfGh1&A%)r6S+P&)ADvzf}A22>11hNHeCo$
znr{ff%_k<y_~672yw*KokWfu3%lR9-aWQNgAku?K0V;vvrGECk&!;zf(PouL-cW%N
zh9Rt#ulC>3oWe2teEVJa`cHd3IUH6?e~RdeUt}{(sOv&ah3@P@d;rtD^08#9v~*Y2
zd9++ukl*LO?eJz5A!oTp3l2n`H>B|4V&t%*1C*jpUN^A^E)5S1Z=M;jy7`-xLzX^1
zNGJ!j^?~A=Lc%c<*?n9{#h7~P4P{LA48zw&Vxz0lH<aE!#MYDS%9(TfNmO~MOg1FA
z-RvxMx`ffBmMm~zIZ58bdh^ttJRIZljY|7eW__jGVtg+g`K3TJXYh+RmIC#{56gnN
z4klnq0Fshf>w}UyK(~S-7c@zK58@S~hYd}PjjjKsgShXBzfM-H{SJs|LSC_m=Zd)Y
zFW|Uw-b-N2$`hrQC~(V{;X?6}={fD9B_OWRoO=-TqdQ&cxQLAZ#SkefpDscnn_XNS
z%CTap>o5tbU)Dwjn_eSs+bxN6o9~xg8t%j9?-?}u0Hc7Q<VWPZb*{=w{co~>qvfQ5
z)CuiY!*M<j(wspqmE%vWIgc=Tx_|=}m6df8j6t@e%0i5h#CWJv2+<)6v^y^>fcOw`
z@$e8g5!Mm)g<}wb2;eIU=Rd*6Z)>>O>ANH~nqwcQLS4OTDB*^q9zwHUh~US7KgZ*h
z9RCxhIFSve`~6I2jpW)X--_B9yQj+q<5Zccpm=nYNT3~%HVYNBWV$P)i{-?B4bD)J
zE|}c1JIYmno=2-aVGYa75O|@zNRgr%g$zi@jRg&jH>B{jT$oXSTgVT8LVj9D^~}u5
zf?mDhI&FKkOpcwn#{?$V?`9ye75bq)KJE%QP5FbvcM~{MoT|Ov7eg9Hh@J|FNkN=a
z_(ms%n&Z!*dxL|{qswv&FxMr`{n_g7M)T#l_11a#_*v7L@DpQGWJ_~sc(6nzj@(ZP
zMnN%LELz2175vY62^z1!>`E}H2$Lmgh%(SC;hF~H1?R)D5VRi^E5AYqKyYM~O_j~H
zq&Gw2GD9OPqIX;;fR}8x?b1C15OxfI4_Q1w09ieCksE_}<61MIutt`x9yB4S>uooq
zNBck=ZCZL*8VV7wGZVnHt!-MHmyKl>a9VSk<0SBy#DQvzYg?M~0;I}-!80^|0ioED
zNW(X}x4)zCZMx8@h_QF0(ziBori9JTb<)Sl#F>rT1i@h!l!LJ6!uvF(3sIZ=yg5!g
z{6lc=wjJUGajqj@U3J-XjjeZb9p_Tpk9-sS+XH%86D{~B;>35wUy{{qgf`-7!b}0g
z@=|^T+-msFV;(r!#Q0al9jx(ho-Rv=ON1>jO8^y6tn+Yv1gaT8<R-BgNTK^r14wv~
z?`EkJOGTp{Tq$cfjC44;<&uuwSSnwERtThTb%HVyO?V8Lv6y@D!NA-l+d+DOkxl{`
zzQ^65HVZ%!0625_Caspb2Fw}-Xz<?H?GNMH=ZDr8yww}<4O}(aZ@l3N*FdN^#1Fp>
zzUBim##LDa3IFN``bla^Ku87D6;9iQd@j&X(omDeC#$)pN_x5ZOhr<`YMnQ52MAFM
zeJsMEmQPIlcb7K`6g=D^Xf1I3DcquUmXK9Mfp5#{b&=t(C=1>Tl#DExn0J4DB<<?@
zH!0J~_y<^Ykn8!|fP@p^XYIgTm-ozPP!NDW)nz9M^s<o32#^k)ASjdzgvV(XgR+rO
z`{jNXnV5E5^^!LNw7-El;oKn*P70{4FfxNabnb9SQ3HGaD8wNMmWFDDE*xkvb^+F8
z2tb!q+X3oOLcWD<7s#xL1PxnB`6Gsm(TlW`M2b~N<qm-+!TW`n?2z8mWdfO{7!>k@
z8q(BC%<uhY9i*Y^xbhQOyw#1I!c&Dlzi!g?`j5nj%<;H=$ridF6;+5#aVJ@SA?NtE
zzoUtb%*d7?N1Gt~*8dcp?s_Q6n3R7XIQqkOva8p~LB_aB*&x3rT1N&OvkB#{Uv668
zwm-xIRb40ka8kDvA(52>_+3+y>(+qhGf(0%T6W-Y)xY*2s8S&WXp-s$1wb1Dw6cP#
zs!5Y;QU#N|Ac+AGYmFAQZ7RQg6#z*KsAq!i&maf`3o9UB2%#NFWH*y7l}86!P+c|e
z+z1)5>R%_j^M~sNHX(}SXkv-UM_tc=t9}2PT6Dm;q+u^Vr5<=^frtjAHDWQr)q}YM
z*|{gczyb1a`Hvmc%sLQavC+kd6*|U7?_Tf4ydj^Lz{r)UVoJcgQP^X#ZNP~-$KNPN
z`$)z#sYVli>ibqc7l9X*9=YRT^;*D-86#pHwe<yTrYo`j+pb`Y51iJSCj5r#nlSjz
z$)2tM+`zf&ay6~8e7O`ANF;Q!(j4632(!wHJWdjaWipazgIlD0goM+GsUAzn@5Tu(
zV5>^6P%0$yhX}mRnPEKele_nZ3L}Wgga)zci6a{w{$78)pKZ-=X!yVcw+ncPSJqwV
z;)y$Ei67$-O#g63R4N9*O;Wx{UF3?>pbJIOY%P5%xR}$6kXKO&48JJ0DBf*Ra(GLy
zG>W;|{MN^7;Xdy3>gQW$h3IQ?tUXM6c~~kpWE)s5AN62?U$IahA5z{2kbsIC)SwCm
zSl^UBcrGn19VS2h@Qd{S&#4?f@H<j?#^=@Hi6Id3dQp$5+UWC4#|<0)dR&COzJFQ3
zKJ*t%G)&`hYTUU6d3@t)c6}s)r=ZV_9NeC3bf5)-tIX_dkXv5w1*t%QhDW(fYq-t(
zm1gJA#B7g^44A$fqoG`d9&)#7fq!TAOmN%B@3FR)H8tBJyXI7!TH-m27}@sB@rFz!
z4#Px2f&8R>>C+;4bxixb`3&0{f0R#_0)l9JhrCZgKlD+0zCX@3(l^YC6ZAkP1^kW!
zX^+tvHaz1C%o7)qn49?w00~55`+WbowGjj*ly63vpXSPrx3UbChy(O0pDAM2$3#%`
zj7HN%J4E0*5I#K+(|nTQJ9w*HtkDj3;Cg-~gd(`L*)&mkO}K{`ArGH8cP=ysj7Amk
zgkEh0A5D92$UATF27}yt&;KT98-iM@)MOs8mf^{~N#AiQ4CL;&l>BaSL_hQHk&Jc0
zjBxInL!==42m4>uCd&=kt^~BY*GNRX!UO)qMh7%DhKION5Yr%?{u$`+O_mV$qrlrn
zjLjmdj`{vBVe*KiJPZIMReC~FnYw5#pSzi1Ozht|3;~pI2-J@nB-N3Q*w&dy4|qhJ
z*2IvZ4f2@dpnieKaI8EZ)H8htAacRa2<_lsXiRlghPj!knE^r&u?<Y{*!|aBGJimo
zD^XomH3@j{TFd1Dh`($kjfVhEMNAB?ZrL$K<OzEr&1t0{KndNDL1N&}9MTCS=z;tQ
znL@t&&nYkH#dGj~au=vD7`WA~d)n*uw!O~<MRVbdRcW#gP2MA&N<xS5n%sO?B+Ylc
z0|CD|Er)Ye`iRSCF6`@g<&24v^o!=W+_tO&;2eVL`~y^U5I+xqs`RT=0ipJEv$_j>
z3ie({kl0pG1Owg0-b5Y}B&-`Ggo+Kl|Eur>us--V_3Ht65W6J!91lng&qis5=gadY
zDDn`RejFG<@Kw;+<pSpxXduf$GS`axjV2UeeFD&6z{Ru`E&UICt{mSLCa9Jn<Q;%q
zf&4>UT|MR1_hQYz?@MZ4#e#Mv%pVE+9FoZbs%K~>r(=!k8JTq6B#0dh;`4vL+x_IQ
z|1E)Rp<cJ3uGsR_sDpJ2PBhri${({|MWUh`Y2GUn9Y@uI-Rg^0tx6B(+9K0|tvE-}
zyya6|;e!1bDusTbD9&~%4S7-%%!{HzZ?N8J6AJEb7qx4LwD7TjmTOUbJ?|nDJYHtw
z*%`&1jcl6_MYeBkP{byU)OA_8Tz^@n{zqEXejxjbhEVwb?{71V9zv@Jq^~8_DPjP5
z70aGXiOPRVj2vb*re?Pp#v$yDP*EeSNtTJQj{2H@4g}>~gGY|@^l;;ZAo|Nk0HC7;
z{*wT>mIHRtpv^fO40WwdKuB_|>lf4=vONJUKUjF+U|qg?^t@hS|6Y4geDcU=R?_U0
zVOLg4zKAn82aJPA(68Zr=m&8^OM-s1z1A)76pUth^-1zn@K5xgJ3$=le2P^=1(9L+
z&TqsFl-`)mEKG2mBVC_#ek1yRP@sOtGlLh-S5HSL-N?9`y5d@LH56DgbG+<666bdu
z<(B*K_)Vn%)D}-O%VtNrcLI!q9-8zEh6DvAhAH&!cv7`$t2Y;C46g1^MV?n>2qe~k
z7qqytHDx<I0c9u{3L-Z$3$YxOZC|rM@`$ZKGFt-xdv`Y-RyHe%Va$|-p%D4>WVVVw
z`u?ENEG$sdGWi}y(KCrD-B6(yMp{RPm3E^Ki|yCh%OLWdxs-sK8hstv%K4QQ2)qtv
zw!A70X0FsK+eMg6{gj=@wK>?jVbJvO;Q50=eujV702bB!XHE^mh)VOmlKMey<&P4L
zX;xW?c+L)E`uao7xxVJ?S#4djMR?HF%IJ2NTCmsqz3Roc>Y~<i?|RbZNJ4XtHcb~z
zRsG8EN}3KdPc~FMuT<_pHoRD9)C&CZ-;?KwHRYeS=YC@`5=+m2R6f<GpSbTjk*ZZT
zM~04$ArjNmoHZzf&2!Q>mTs;Md<>Z8-$dYV1G)`W!k_<0zTNEF;+8C>+@xnb?`7<O
z2epg(a*DcZJ`?Xf-?Ibo$uLGxK|kI`yxcyf*o_S5t*ebPczNEOJQEa6hg|)X4Ye6@
zv{027c<dCAkqQ<ka<@OObJ&~s2ytxQx9nn-Im4Z}?eg3WuRh8?uM0c|+9?`RwI;~9
z?<;cd)`Ko-w?&Iyuh@XQ=FEu8L&L%iKK&YJwZ~~Vc+u)_n$>f>)%JGXN{A^<n4``!
zp2m+qtv740Ok{Gql(qYmh0BIJEgs!^Y{I{I&$LQI7lg((WER=@6$a+8M)_=({yi~t
zNU#>%2)WS^lxp~np6X1FfboGcJgV!L*7-J++sWS*v6*PLgER7X@?>A|V|o-~G^Ok}
zi;kU<%kKX01lPQ^K=AghPT#c^cR_8#TfH2nk{XI44@Tn_4F01(>^4}}97iozzrg8|
z?quz3)5VFNy0|ZSWzUs7z)Uf`=6B*ctNq^_b6Hc&5Z;7cxoC6GprQRWRVmqE&oPpu
z(z35quQcl9G3zwE_|01;AITk2z080=_cJ-SX|3Z*u@z2FO_p)t&oyY+6N2P}(P>w5
z*p-Ksn~|Gz`iH_{jw5~t)z7CVk7=!+Ruft_u*%lT9^+bBT4&6CiSD9^l6LFZd)XE=
z7PPRV8uYQuZ@xg+9#+I3)zCFR%+$4DDH(WrW)MA85WSlFDPUgNqGq@UKAay&NN%9b
zuj09+>l=El)2qJ2!4I}v*%?~;?Aa(EQ_z*|9?G|zEi31`?>!|+-)TY5989QlJ4M2u
zmP9O6wujj8Weu)!=RK}f${q2noT2>K?EU%jusRw?tKz&4qUjPo!Pi!5s#8qc%2?J;
zNPfIk3z+)ZoPXl#alGphu5Q`jkKO!F><R>{_MD$oVs6MicF94VdZW5|wVbFb34G;A
zHxLWl#-9qexx0;c{ugnp?s;HELi*Uj_y~h4@W{0Kw7fbD{v$2)>P5q3bt^w<TOS8o
zhi=?PZ%#R0c2>6_5g`s`t^@+tdICLQ6a(uhHHqKu=-daLeT)V}!nBJwG*sz~hn-Kc
z>svv(MUN)#ATiBm&nt|vRGEr=4A(FA`K#0*H_g-EnOH4(Mo~cL5G~tLiW!Xb!RXc)
zh?+81PI3hk{;^Oiz4Q1-?GXBBa=6aLr{OSXz%Z%W0%w#<W{M^KeF?dkz$Gu}tS`Ep
z0X&GAV`if=muev`R3v-(SGX4-*(_x@K;mk(j9<O24Q$9eM;Z;b{wbD?!61xX?>KLK
zz*(co8J!B_1z0&axw&;~)XuU{-Hlg{K?^7%73_Io^^8pY(c@^=BkZPWm+x_)>~X-H
z*W=WeXsMAJkpv8I@Ju{;XgS#wA-YG~<QxgmrA3^!d<%vw9^X;?1O3b+P!qFzUW$AA
z*;(?BY_rKypRM_#l9IX|+huHyixX`Fe~Fl0SGTB&KbcYx*M@uBuTw>VD@EtBVhjx~
z&W%8ie}T2NuG<62mIAT-QZV776jmHTPTqroX7|nB%@>So#_yCpBFY!^LTK}G`W5pJ
zu6N`({$#f-jfqxsf4_m|42iPjRJos<I#{PNIX_^YvA>k%Vm#sKDV@3s&{;Fax3OIN
z5o*5-M}hg_`h(IqleGajmQebLyJ8cwH=e#qvG2S^w`G%hXrY|C;QwRrw`)3o{?A9r
zyLl$lbbX?hi`@H%poW#u>(h~w7ajiRvxo~oG$zYEKFC$#0xegNovV!M-N2&dZU>U(
z7SGCanmsIc5?WYjMeI`02X1(UGM=aXeRv&SMiYaD)Ej%0Qv8ly{_6@-e#$z7TL=o0
z%HtFs;eM!hcyu%@tb}A}ynT}lI<kZ&KUDXB{|Tv!XU!~|uas>^9-mKM_+K`(AAjB5
ztn|D*J-KUv02^E2KG{g!1ZU@qmtDB$WyCXjnfsC1-4>+)7oKYb?>R!*&*>t=s0YZB
zPGHbX9?WLvzsy1%*1i3^`eezV-7-YUe}<?a^;qS3J$gIndB=D$2zEqOl@(WyHaKxh
zO+rVxVh(P@FP5$wM$i7~MVw;4bZ8fgYYyRJ<A1^DGR*$`^W-d@yC;EA2Ob{&0%z4n
zD8A?5lNiPNwAD{$$Gl}r!{J}3uzM7&1|tpOTilot;b0tmN@I8;F+o_>??k)Q-A}uM
zHCXYCoY;T7aEpYN225*kpL|Twc?X?3Hr?gYSx5=<uaml6JWRaNS^DXy3@D9HLzyvD
zt8njWa$j#?eyHp;LimCsZZHksqxf5waUw^OW90STUD4{rQ_9&BMn)n5#m65aLGpJ4
zWldp_(TR1fPe-r=OSJPj2T|@vkqeUKa%8Zr=%vZBb*VQcC|NohpDKUv6)G%!RZ!Q~
z$1J(%Rv7b<MO-pyEzUEZHZNhPWb`eIKhb`zUELO90KnO^LVB|={o9oWqlwDlt_?4a
zZsj(QwHEZd<F&)JO$Uv7A5mJTIFmD>gKAalW5pFX;J^53NDJhNAi!rvY*a1tWUw_2
zW+F~)UTJ9l{P2C#fGVR<z|Tgu0A@a`bYHdLtHPYKC8;od7)5T^5`5iTyA;=E_EN2S
zYj1#}53xg=!(e#Ld)mMIt2@PVn|a5V8Jgw^F5=~GX3L&a1FtrPMAOZ4P}+tMBmUi3
zV7dL_f?}}>+K5GOT}E(9J{!)iqx??`@UP;AfvB!w6Lf-GWJGs;<}cL_3+zs#n_O}O
z#i#f+yU=5yq)k8UAImbH3!1k2E-HUzMmKQA3hq29KxgDYWup{)JT0zy<^3+^{nJh(
z@)Y5?X2b4FrO0}L*2}9U=P&iQ_zk&9VL`j`q%+j<ax!iH?uq!kU(PO$po<LExlQkS
zLINvVP-Vt`GU#Q;+sLx&?Vg@Smfi1{ZNqz>!Vkjcn@>;p%T7t<T(Ox1j8vAo9uUA7
z2tOue+<g#60PNI}I>rTH^m;&_z`_v%e+#rc|52#B{5K6Qp8xF%V3M;Lu~~J$S#RF0
zU9`*r7tCIQhsnx`FDic@7$*1Zk{yvOy?NW1=J@I&Sk4UyIp;IdD345%Zz*54GER?H
zpN}3P+r=nwzYHkBH)H4b7E*stvVzZJD|11Hf(*=Qz2Ir>gw#{{CL5rB$~a(qX(xJw
zE~|fff4%gp{%X5yYt(bF2yL|E*3arT-)e*CWz#eCKHUSSk2;H9Sl!QV_x0zQ%u<}y
zyhw{eg0k(#u+8MBX1<VX=C#HP{1V12^zhTIW4NtUt|7cmHnU)EQ`OfCBC)pgn%4M+
zVyT#9+OV=8Y<$$YdJK!pU{+jF3QW?Q((LL-Y?x%JXt`It18Hj#)E_+V)+9?2(gUeS
zIV_n!lPj&q?#uiexH;Wm9d35~XGJ`NSDAP5gSyS55Y11+D&Kv#LE)45cSP-rfY!Fz
zm5&;?0-gy1cZVV$X$=*$YzGIO57bwE4*aDRvoU((a(*2;5F9Vb%IHy`bbI3~?+%8i
z^ALOvQ_L0{(rj{$YQEq&Lux(Mfz+HRy(!kqf>H*D;9);m?RfYPu8QI1oV30uDorW$
zoCOIyZ^9di?7t$qZR3Rpxe0tr0B)<w9IJ&;Xs4x4&1T02+o_xX-QO=40;9kb;lls4
zCpU^xfkOtgAm97H7GL=OIS1rS6bBG`*s)``8S!6y0PhF#(nvc5UUpybsq}gZ1fEQu
zS5_ZfolkbT^25&LSX*AH5Nxgo_ukH~I&ic;@DOdjJRBM&_Vj@sc2p$enf2fb+Vl2I
z0Cw$KWu=G7-mQDE)n|6~Y1nVg#-VBCLrY=!*+>J|ry?%=(KMc$Vtv!@6$}<*^Lc?_
ze0)ZOwZB$w$ZH>h-@=9`e#OUi3*#TUY@0(8$0tFcRi={qv*VYI*Zww0v@uzd1_1*j
zWJ0l0KJS@>c^rAUDQw%5P!7axpf3I4qzmNJA-C%j{jnyGa@U1Tq<<9a94Yao&B&Hv
z-Y)r*jeFN0qVL+g<d?JQHsyPg*f_lv&+*(^=SrT04Hqf5X?9@2f(PPt_waL&zU~xM
zf;E?Ju5KIn-h-RC^F;p|vF`tgU5_S&T$Lh;%J)p%7waF^x-;IJ=j{dwnd2D6$@C27
zPcOd{+F|v|JFR-#^_`nM#kZV`AtAgc53QzHq1;~Wc%}C^L_^xuFe%I-`gZy-G=|3N
zFkNo(M!|W}{YBN*ION#y&zlb_)spm%pHl_7+Ykr3TA$kl)@<=Pd2bXN_x^Pq;c9b7
zOJ8#NI+8a-!}(~4ZbjDM^?5{C-%>Hg4*E60oTpMTeU~dah)<QS!IhX2a4`2Zz#UR3
z)_$Wh$ZW|)!O(lj&cG;^waxBS<&<#1;ENycBd^ymGI26fJ5959Xc%!sI9fUCuBSS^
zW2&pk!;YTxs^ZI=6;4{dLAkJzih>ZUaLf~Vcn*?BU!K+s5h?oA=Ek3RI=TFo8&VGm
z2j>l~7%5h(5)^cFxooEwtq*3r6d22P<Vt4Oqe`o1tg?@59IfzPewH6ac{NF`TMSLy
zQ8>9*(Z2^NIbcH%_+S6p6*Q){>#;+@s0l6TzPs2pv%i!GjRGU$Ynl$=SE{IYl__U|
z!F?K1BtkXaPnWGMd`{s^xsX%31ICn}5L3Kv=#s;%W^Qb~?e>61av2xVqKBHjXUf+s
z{r8|Y1jwgHy?)FU9|W?d&MR@RIadbrygJ=5CI)!e3Up3&22VBK*bf^YHip~a-yIqW
z44yo#oScE7AVZ4)27ef<ZwOAz4+gwJOxg2(+0H@B@j)%=cL=-=>Xja=$JYsr+B=|L
z%sG+nx+Qu!Bgzp+y|~T6cIm8Zel1I^qqkIwiu&RDMnW|)Uv$+41+^5ayv{Q{r!z>M
z4IN2}PM2cXfFtunxwxi@H*e@j`vrN85r%N0K{vL@2LN^M4Rk&4_NH!roqv0iRa&M-
zx~{O$B%WKm7*9WO`=M9im7u;Q16LWN!u9ORr#J;!wk6N1a65x$<yyqdWtm#MA{)90
zk+~^i&R5y*spIraBbrxRN;P|FwLKXn{P5dP+HtZj+P;q}l0-zJl9Y`#r_>Ag-}6%*
z^<2J9EjlY!fI)qS&7kpP;FQwyoU%H)h|zQVC@8j1=!8#PT>Nb(Qv!j-HZ160WzDFY
z3_eRTAD%Da-&XVzZZ@B|GXi~mpwZcO8t8x@(|4N80#<)%U6S&@+>Kg&_)gjZMm_Q2
zS<yF^6MzOBBtF5;!F~l*`pLX)F?5x4Ip$n^fVS()@hysmL#Q@x2xZurq<>k{Q&~3G
z8D8sitIgg3^vhYx!@v7<&N#CtfQ+JV_F=c&!mXU<bHpf8a0*przX9di;mOJ9#rIi<
zi^`mw9HYGAN%SN>8<4ZS1l<{RzA@J$i>gKsSHSc&Kaq?+%dY6H0;&AC=|;m0YyAk^
z*B*}}cD;tL&A(!qn-8QuWVHn1OF3zOXMNU;Js>mhaYg6khH11tErKS$h}MZnXx;h}
z?j;f%yddMjulVtsN%2;J=0kmwggq=NV#MmZO>W-R`(|yPUp5W6)nrlOf~1?)hw`EI
zv{YolFZO!^APCi@>#yLSq}MbC_sH+Ne&7^@=fvX0s(GNXi>QoOe|aC^n88TkZSh5;
zr7l?UZ^x8mUhV?-piEiJ99H+jQZ4VwwhaS8%wP0}!7IlI_l9Rx9mNt|JtPY)y1KQR
zeCU%~JkM7XyMvG0RzA0<RZf{biXsYk>)X)vSt~reuaD6l-{_akiJq}of9h)_-{UGr
zXg$E*{q5-JqNRa+^ia-cS@&>4aZTr6^6~HHljYZT+h*IhzaIAeai*>-Fb5s0%Hs#)
zvnP*R9N`ML*%y~8mn`uP7N4EED`Qw{<9Ewxb@|F*>51V^<#DPP!Wp8Pc6~bhX=Ay)
zWh;Eb{%p@<=KW^gv3qL2PG*=1++WR>k$7_tQ-34t8NB+c<?sVf=cGX)c1^FqQ=v9=
z?I4jhk|NqQ3P!stA!-r{RRf~obT1{HC_VLY%^eE0?w;DHJZLn-V^D2c9<#+@?#gd}
zDcE0TQASvgF1GyQM8VhKUIR2J^Ucb`@Rmb~2!4`y5r^Am-RUja$APC(ftM0HybM*l
z)3uu3>4ghzyWWGbla)@_(S+qk7JO^Vxy`a?BhNHQyPZZk)N*ncesYUKjvUnC`Bb+$
z3R+FWANC_Q7s%Ssv1)w`H(iCdt<TqT&-OO46cL8E=_fxq80$|f6~c?+`;2|fXiIAF
zg6Fcg>_%iJMMI+s4WXL1!x)oh7ui@>&*STc79<Hd79<hSX17mxS}z{t!H~LpIX4gr
zX4I@zb<Kgsz5R&f7s#mn`I2JGSTO0d2;m{A51lg*LryBmqaNA|)4$Zt?t%hKqQvxs
zboj%tfuM1X@k1HZ0Gs{;|37_h_M;@nqZvK85vO$E4k>Y*F{##5{lY!io3j5cWmB!?
zB>37c{<{`N>xGk3TJ2-5-F#z~f~B#OWWE8P6`}^UfD>{aY-%hik3zCo<fQ1V6MkvL
zSo!F~%U_~h4Ad`&%hq_Sx5!pxcOp;_MuGo+JfgC$u9xX8*q@uttGHw-<tz8*VVixd
zvc<CpEQBrF^r-@#>7ly>s~4s8Mj8{i(>Tgq8VB7@e5+UCCz~m)+a+#bvToYw9=(+d
zlIyOeXe!yrLmZ2<WFTkn^IZzmZ4}YdLATC7c!iP2HPk0I{`07au^YaJhlfjd0iHEA
z^2Z72xgKXF1N`0wUJ1O-)yo~Llu?%?ox{}4#g?PR9pX@Gtf=0<={6zGts6Y6XRq|%
zq4Hl+gD&|c(54tB`_f*g`?*}VDH<mLwsPm|I#b(l<zCA4^{fwLdI|r>-2qXd1%l4M
z8ol!|uj9V;Y+jVx;2xFMvFbmLR2m#L-M3`*NnPqxtVs5Z*(O;&qYNVEXidh}Lut&Z
zy(EP%S)L_ZyJ=G%@$mJ#`pvf777gg`_EV8q$$k9GS53#h%#;q>;6s#BccEeP4UX8=
z8Q*PX@xz2L{l)O-(9t27tEPF3?C9kYWLx!uCuAOuLA%$c?yJWSWwnR776S9<lnX18
zgX!ixv%;qwSi4%U!|LLDE&ue;<R$w$5n~@<+A0RuP{Dum>q+~RiD@iQ`vtMR9_oq1
zlBjFm4AqcV>3toP>qQO{Ww9|Ykp%t>^8nfKO_l?l#ZyT$+2%Ox16grT^tMNzQN4?f
z8>3a7+#ja;(2NDmTDK-V?Zvt2r)B|{<`}RW7iTa^RyPR2!Ew^MaK3#(x#;P-T$RsS
zWhT%S6iF>GlvN_T$L$cjg(7b$dS#&T{if;mo%IDq#$uIsU<sbN_M6#w5rLPXV|A~(
zY0psuk4v@_k84NvSjIIESFKhLbtfeXV?HBh`9aQr6MW{Gx$4-xdqUFs^>=!_9msqm
zE-TLoEm6(5g+$wO=m|JAoY#Kgg`bU^jtI0>*t;_<#d5o1B}TGzHt`Q1($!mkW;8eR
zML$z3j?}N5Rr0)UdDwQUXl)hX!4V@2++VYQ)%4J81;;KMmC{*rP-@M|7qI9gGcvyX
zjRhe*9}!Wkyzzmt=_jpPh1dm0>3h{lE%li^ZEkUOeQBw}%dR<BKL`30wkZ~G!6Vc>
z<FHFqlG6~Lo7?k6?NQSj(gsLJ6uq4MbWs!EEZn-!sivKO0E5S7lvHkx({m-xyzot*
zfwA$*RB0y7H})6vMLx$9SL1V@sQcT3k7(<IT1=-;t2akg^%-{QgO6#fJMGu?DnAb$
zx}4!6C<_n;)W2*;C`^cOe2n8QLea9f8%2FubBdeD4GaYE4Rd%z7L8wWa}AwrVooJT
ztVQePM}u@UcYIrP&*SFZt4sG*rPsN9D`APNVeA_5XsQ$8E?hk)C7<*6-fRt-;z(_&
zs*x$MnB}uhIE3?PIUtp}%{Lz&VI@dO37&T)&f`%iH2TzwTnm$zRQzS@WC~@((j277
z<K$z>eU^W2_k3w3whyXTt7k~?jIJYhaO&VS=XIw*L2%>39nD!5``7YX19mA~ZF;<E
zX0!|+x|1oLrZoPovCGM}(C|he%@3pjTpVkdW}ZON)Abb2O7a%PkBqw0OOk_y?I+Lp
zz_aaFi^#juzHMhDJjwEQ^R15MyG<NDnaeDR>HWCkJeQ{o*UcmQG)8&r^mipxc|MW$
z=5IPWH2O(;WMlg)lW}WUaGAur<Y~two!3)C7FQTyZn5JHF#Jpls-^6>L}q<>&hsY-
zQ1YYO?2CV<m~y3ec6P0}$o;`uBp1Y$3->WO(fqtoZ4qMrOn);@L)>8L`gbtId01ut
zukB;W0{&DmpRXSlXP;z0OVp&q5aCH9u1$4ckC&__vPnR#4$kZ=RSusfOSP<5#QOla
zlR&o%KzY9Wm4r7+Idk^Q`0%VWNNtkSrqh%y!2o$AZY!~S&b=63oN822>hQz?pO9Y$
zt^HRo&&PV%egPAM?f71;gcIZ%OrZdsQS?o*YYp@*1*xrXgLEIQ)ui>OjHR$XK7?Wf
z9uREaqcgmHO4UI^V;!3RF8A#nS#Rj#9y*Km;``jJ@Z4a#qHq(pV%VREecvb$cJXP1
zvAKUH^mCCC-Yb2c+=Q~9$Qh{+fc=Ky>b|x15B++iAv#eXCqReyfSe=KA^W`La&?dA
zHaO$ndsC|ayuqMTz>X?UW(<FDmOJ)xyIVOt-uuqi_?KZ|t!#0Cjr*97J9Z12iLV6J
zGwvBD#Z;ZCu$k72!wdbs_)J<DR>7AG`9kq(cYgHTxr}`&uNzKpL->`bD>P(rd{y=^
z9UPv--v;>NuPZH+KZl6w4&BCi=}(!Nn1~j(@`RCXa9lZyAjnksY1=;#Fvku3VS?@9
z9R7Qzqu4SK?E5w4Tga}ldsoAg-R}rdQ(0&1K7HGSGflY}1=2U#_-TO$4)e#^@PnM;
z_C+YWub2yDkJe1GPkyotjCICRJf<6kpkp>R+b;dB*@ON{pLD>Z`lf_)!nPqO(o~f}
zC{hS*7tJrnCJd>@(c@p@hUmBMK!qhXm=zb>FR;T~3nBv!gjnCN(qrLlpr@?xCZ$QF
zb-7q-)IJ2*XpZ%~@)w&7N?L(G6&^tx?@DNHyCCyCvLKGcyf<y7SKqCfu_w***K}>d
zJhNMADyXS5LoA#8=7X?sKUwYKiG+FDLKjg{ya@9?AnWz(DU1AXggj1<HjA}(2CMXm
zx*xuchgZQ7q+*7a>#JO-X(e2%g$`LfE>vDMrpBN$g{26F7%IEAa_!ju*w(wKvH!U%
zozk?%3}2{&)05yON?h|!Ndac+GyP^`v}w$(vQslUlpOtj7#aVd_|xNO*7xgDyg}Kx
z=+}w5*+`9rRey9y4C$q(XSff*DtRf)o84V)n(3|lQ01RKB(Y<o;Y8N@0x$5Y@LLrp
zZ;{2D^ql&;3n~e`Xr1REmOQ?)gmZl<wAbcy>~A+kA>|>u8?kJO%#Su#lD|PaeX)8r
z!iL>0y@~IZ%XL#pm-q2EIiq!{!g+@Azf97-t*QJ8^#xz@(?+L;0|9Y;xQt<qXSzek
z^xnU(lrpMk{KQ|9s)vG9r5Iy<ulkr`|Awc2+Q6qax}WA}?xq<*WZq6jTmzUs4jG^i
znGKX%#5FN8iGPu8I}+4l=G+@88d~%GY<;74)f)SW-(F_Y&(5dk8!lfU%C3}Qd+rB^
zmTM#GLEX?UwGnjrr7@9ka(_-SH|s4}O_gWiEA6^}IWuj=@xtJ$V=5)O^fRKd2tDn(
z!aeaHR3Y4bMCM?D9sPz}&t*4-t&@N7-e0yz@uX{uqLuzV0Y3Tt+OJ#1L77{_@^9#a
z52Of%!qx_h?aViX=EBOB8wo`vvA>flEvrVANlcY;ZhO#Ksk5+W0ukl4Sv<{%>rf3=
z>E0H{b1scAGKQAirDA)pehFck4kXPNqDIDpc}sBdoGhV&F&!fDTdDj{y(6#eZ$$)A
z?lXOi8qh_iJH+mPSLbz_$VVoLMnsZRFZFq^J~)Kol<ce;WgnJ+X#M+g>rfg>TI<qQ
zen%^Nk&mF{r@xMdT;B%2r*F>H2<>K%V6VguODT_>BO0t_`TE0$9Fs81f-p;U!UA24
z{f-$sg#l6A03BAy8;{akiS5Jat=A$`xDRmF#+`EAZIkRLB#pFj#`1&S6^fqVionpW
zEcc@OppE=D6*D@%I4^!LciRP5x+j5L92=(PjnpO%A_FlDb(|JOnw4-En(5?dtn|q{
z?a}VpZOvUt$$tIDuY_d*SX|VuIvFN>;|_cxDmUAIWl`fW8Vl=Wii=$)(Hi0>73ke{
zMPVz^QZu?~Y=n~)&`<L<%Iq3y<uPQo-Gz9*B8*h=ziS+csREFUzVPXm#=V_M(MF9y
zJ-u|eYG$jBhEtHG7{JWM|KT$m3b*zmbHM82tICnRNnLH#Ild}T-E*gguAjZxN=^D>
zyJ;r(XL)Ff{WbUQ91&?a3;*%eN2oTF5AeRCeIKn!B^Wl1uQ};aDCLN^VQFM)de?b(
z@_CJ3|6Ncoio=4nE*(PpB0h2w$FIh5_^PV`TSY3Em)e~#;_jaACv6CV{3fq>+W#?k
zmO*g@YokVky95vJ?he6S7I$~|;7+jM4#5HhcL)+7xG(PRt{WssxRbBy`*F^#`~OyL
z6>Qba?9}vh_uKM5kEWj*rWB{K@b27UAP07yq(%%SMiG@gt?B_u9%IV*Ph=WJ0qf(A
zP$@<WZ|#Wa!-$PqLZA{q!_j+SO}8uKhS9oi?ljBd6V7ydFEKKui%}lpi6{^8c=m@Z
zl39nz8Z`v9!Z_;>0lc_im5VMbf+4To)?jLvCyj}zk6NytC26jr(C-x~>0v`pD%W`P
z(Va}4txT=3Vy2(mr7#B{ir}n{^!>iwy<XEfFSEy=;X+afhOb#&{34@geaPGr^00Fh
zgOlCjJ&lH^@02ts<xUQoPtBCXRg^S}&pJ}Xlwv&d@M&P4SqlYHJ@olyiXiA$5!<m%
zZu*&yY_b(D=~|-|mO^$HF|)q1B(i{jV?HnT@la$p!Y+qEL%{lju%$G2YMFbj(bgV3
zx3oJxxAcHK*41{_91btmYo+gug>|Y=S*4{`*h9%FqBnSF01u&7qCCtfKDipb#}ze`
z(;$Z1DyoQz4$(-b5yo$7W(LcMZ=R93J)$A4D=ql0BMxAGIsp}GINF)mwP{CfDB4Y=
zLH{f4WC=sc?m)Z8^FhlAav;54T_xsEn;hD*oAT%jV{3{s)JV{<R87ALRwRf7X*1b_
z@~BbCBR^2yh#nf-wo*@n#+>teeA3x%;|42#*n4-d2V2cTCEJ=Lmd<Q<PJQ~E=2IjZ
zDTB&>-bqIS!5*`|?~!Zw#K+v(@P+5nG^aRn$1wbiWN4>&9r!B}h6IQ^md=(=V}J(O
zk|leLCiMruNRteR)kE@F0`DjR(=CEE$^3+L@s0_kf4<I++W$%0HQvLOlwJTC{Ji`A
z3yqo+j-EDu2BvBML&a=M_gEXH<ATzq9Y{N{Mo0Qkitlq2qa)Do1ty!}mn+CPod^l<
ze5InLc{G2c7U`4kOXh$-XhurYYy2{mjV;oR$e)BKs1ogg$vHydDIx{m^H9of+s_VW
zr?n>>Rv%<hIICb(A<couO04AcBu$!0yGe2VsKz1_HW!US3P!=Sk^Sb8Q0szqr2sx_
zbTXk_(@-cOx!J^-D|0M@#L5Yj5yV-v5DV@$vr?+_@CJTK)1)0hGNSKOQ9$~jkFP~-
z^HuSkpb6#e@C|a{7S3J-BdR~CfV6pK-!xK7tEOUFH-yuMnLaKaK6Z}lfHEl^BCq83
z<sm<B@0|G}N<6PjS)h;L$l*|uJYCx)UN(8b*py^=cOJeh>zV~wGjq7T4N0SiUe=ai
z2tpKN^&&uPZKL@M9)oPRg#QNh`A9%e*pfgn>`cCfw1D4S>2Z4LcyPAPRV1OQUTa7u
zyGbLLeLs)RYOoz7%YJ0o%4MU;u$fst0`gwK3M+7UV?cMa>J)_ofy*<<h!~h+DR*P1
z(-nPJCOOjEnT9D3BWo?H_o_TNr}$#QVFp5pr9q`FniddQvMB|oaSd=ZidO>GkXz1_
z30&A!3<t$#vR1T4wTDjDI7XQukdjXDlz-><<$<){zTnP}5-yIJ4|p3|u`IFL4qaJB
zNO+ON#u<@@4~>5KPV(cWsU2-;TGj9v3wL8!PPVnnkHgJ6nFQr3pzS>se_6q{-Hpkp
zZ|e97ODlL)aWe<^gOOLckGn>LmRF<7Y9_Wy>npDlr{>3wqUsce!|XIIO(z0AR10EY
zcG`kI?H#Rnx|6-9@1Krl=3?(ma!OmqcM%FhorYhFYuKIJiI-22IQdSh$EzS}ryg!*
zILq|J2;Zs>-f7JGk{Z=6^VYHZc-KB}RWs4teq&Y4K6PYyc3MN=t9y`IvhT+41W$Xv
zp7O5#5!zZ12xz+S79ID>sH3Q`Y`9ZUz}ZCM$i&t%<|QQ15mV@AuYI3q?S57exwhCb
zjx6nZQ>x~xGVSwhxxt`SUMjqOTgMnN{b9`p=TpX0&xmz^XYC{v@mW2!<R59I?$uJM
zfDN7TG?t@=&cxQPr~AP+{{XQ~m(#jT2t%FNC}!}kfU%Rw#Zz;jQ=D<W;S}Bc()i2V
zVMRi<^kbaBPAAzs7S7|en>m_cKRfDPw$(~vYu3Br<P5^*xV1kHdGBhi^%t%cUji;j
zjo0%6p%-9QRZ9BLDf`><%NH1NlZyU6^_!1)H`UozVdr0_aq+0YuLQdG@nkScfiDLH
zTB*7C);Ur8%m0Ppl2$@}JpwX{BeBef`BrI)db^EK`SkY-f?Af<7#XRLNH|Y<Us||p
z*DRiuQ1_mSych8>z2Z2!UN#VooFP)V>OL|gqQmoTNAz^ov#sNX{tCO^lrILOa;=BD
z|88DyztZ9goF~W@+$heqr;ml*gw~Ocq~ka{EBRd;m%&9hK(q!U6Bt<`;IK1uIy)5h
z<@Lfhk~AHWN7jFG&chE92y(d#Gg-}t*=_aLY(yn%5Ljf7M>{;=O-TMRHjgAX)DooJ
z@_uCKNDnpkUU-cfg4*z^fw$2GM%nXY?Npxzc8R~PqwV>TE8;b0gyD74$i>+<PHS0%
z<C~7EqaBtMo>4Q7nwVTa^3<wgw}L0=%n730zPf>u2##xE(b+$q=}B~wz8&r6xWI=p
zi^Y{9+7%?58R|IR2~V`VHz^;dvf8%Yi><l`l{}g#%?oNZ!;#?XuWbGacB0X5b<)#!
zq~KX<F(aTer3!aLHQLvoGG9<h>oFjJD=rjDKIY`x<yL{MIZ(AOIL>JDIT{l8q54}x
z&}?QHj0;;BKQ#`D=(I}70DoQOQkV=8G7_el6JQ%UBt5e67_jq)!E|5`+w-No`q#k^
z^ZSaXu^_CcrV};`+-VBnFaIdasMHtLio52EKg&|K>*YkJ51V&0lZZ}p>un<Z@KRh8
z(XOb9GwKL~W&LL6@78XO_aTv|fg81dc}T`?k09mqU=we$VGXqVLIbWPmanth!S@Mm
zsa+gIeUJ7^*Pou5x~lm7ZDhL5!qo=|YpzM}6Elky@S1Y=hTmP!aXY8${qJ4?X$Rku
z(n5X|U1$HU4FT=OmN%fsc?7w!XrWg-E^zg)$5~-b@b%!oXIiwU%(7TzuQl-wekPk=
z*e^VU*xyejxSHQjpLm#?jD0=w7&0_iIRpV!G+|?haQNB)B6IODLu3oKLgb4YDBJl)
z_8t^Jw<ASyW<RxYGu7+Y2t+WFqeHQXTI(60X0xiF8_vEG^5j_R(kz{jv<jos_oZfv
zyz9<(A9YB|4O9gFy|zk?@vK{2Q=65pW>|;c$MW17n2lO2>Bw<xXp_tg>m041+i@~5
z*liqm?8QXg8%pEnBo__4nx81c6|0^d{;tP58Prb=052*pB@p}=sP8Cbl$NC&O;q0p
z#0hAvp|;JHD(|i>X?e#Vw+CmOBN|vw5)R#k&x3ph1o4@<BJEX0z36KUnR6R`upPQQ
z_L`?lDZ0`Hj=US|OJrzR9UgzkNymYJl6p*O^k-vHy(TiHxymQRzK8_`pQ~HG`D{24
zIREO`Y@{zyZX44>how`NKLS%D{<yV^Vc>3!TQ_s4#4Tz#RPg3cC%XS>@4Xj1T;fEv
z2#wJna<lBZ^YmeyQZ;j*`qvOh?GAAm)x9xulQHSF7)VSw525b)Q7U3J89Mj*Ttw>n
z0}~l}EqljqP{=pc@q|;Uv&!v6v!(B?Dnlu}NL-jEMOt#|7hGGOddvu-ZXCZgA<pUw
z?wNAMekkBe{IeggVPO&H+#$@A`?n}};yCrvhSrqdtlf-YiZ2bxjIdf~kTI09zRl^?
zKbbTw`&xCrf9jD-Lt&q<K-=0|_DZQLH|3FsA5vns#EuFJ^dmTO?iVH~4du6U2N5}c
z*YMt5M2)k@hpGA-Oy!a$Wthj1TzySH4!Jy!-<rZvEu?cg?ROy@j+@%9e5cR5HVG8I
zq7-594_LH(L7`rhp$gXpm{EiGf`(QCS%p>vOBhXsJc+QO7OZM3dhk%60fyYQhP8))
zPh(Sp(b3ll=4TSr{x;-8B%l)o-tqtVfQ5l*Ckl87|M{^Y`=0^Ye;pjpl^CP{Uq=F8
zhq8yISpM^}|4!VS0JgaN=Vkw$>vs5Wocdo^M(yW^08RgY-$s;Ko-Z(R`@arm|NsBC
z`!b)P;+sC>B+B^;t6(iiZ9R~X)vjzAU|IN1BG8l~U5WGT372qF6i4S#mBNW59ys8y
z5kp^z2prJ~qhE<sj&Uo0y1Lm+1xcLt3LHgqfP*`0*ZhLxlQCf`6%>$;NH+Zw5)ufa
zptYp6>R$Z>WEEoaxB|Dh5~#oeQgB!wwcNmyM{60XbKEAmd0dqpUPe+8c^(^!1BY@V
z`ggWPwYjhYzCRIeFd;Jm&YM@sl>C|b*TSn_$_!X9)?1bO2`ljK;Aq4>{zWJMJP(Kf
zbKU2MAmfLUsXC+FgTDm;AaN68Vz{hwKM#q*F|clx$8i`w_bM_P(;~0Ve{Iu!;{alp
zbNt+$2dnzB5y;4Ym5slZ?#(J-15yBr^DQc4sWvb(^Jp)x|Fd}Q8C#wDvvbSYTx-M2
zr3lu=GWYnRTk8}do$6!I-U{7Q^i*h)u;s6hzmY~F<~cBENvropw*C2G;I5N2UN`+i
zHwsE%mC0GCWlnqGD#@W$c8;KbE!Vi|4-0xl3+63a*bi$>B^rVeLSo!~QcrsvdYc3a
zGz(P4O)^h*O`aL#0`pHfTJ~b^O#IK6Bk%<X`mxSLDfzWOGllUNn|QQ$j-Z{uFMXo?
z9&Q6O328cpA^fZ%KDNB@hZ9WSC-@PYKp2xtoy+LGoE_KHq(}USaNdCCPz|PV<%GwW
zMmE)I-qr)DfXftQQ^;Hi^7zN|L@K%=epEEgsb-j``%F9{4d<fYYgdh>2ZL=QkteNA
zK?zfsq=u>=6O-J5P~Su3cV4f*T*XV|?IV$@MVT_8fD4)At1=L@6~INI8VQC8C-+zW
zT=E$}Oo&f)r;LRi<)0LnaVzKFUv(J-Vl^{M3zl)L=}<pRHG%Sna99CBLxRvaaat~N
zL-MNrQHof~ycdfxE8MS3GfxHmlXVxhxpJRg0J4EhSX%gkpggj>Be9bZ=*^%pz+}xE
zSeo}<<v8#5d3YUPs`Ezs%Z%t0;FRi=s|(8>BKRv)fT8w=Vw(1t1pY{b(aGSz-q8e(
zZ|dkGewCXN1g0wJ^A#qf3&daQ=t>p6ntN2LO|KV1$a9q&b{uN`ncr_WLFLw9vh(S;
zPgwxn>JMYC=*x8+Vw*UmYfK|cOEmgW^hCO3Ii=sIk6t=aDF-&omkW9?POajHkMB!X
zV8)kL_wHBauS`Up&n}!De0-mU8XCCr(N4P)I+%oqsD#J*4R<IV66B-=1z}!=`7Ak;
zV@9NC=h8bRgh`q1T`kf1sZ(?tFvL}dN3b&6M#$cOLNgghLGK|ECi$(ie6-&5sZZx4
z*NF9Hbi~U3Z^zuQL71(qD*X@swO{F;i1P8BY&Tx**)j{@(L+$EB?1U;DMpa{Cd*L|
zx+zG*YR_zzL_-f2915oA=%RA3=b}GB9|_pEl24x@GHweFbp&A~jf;N+C$onho|rC8
z@XIJ9-NNC1mf!LTne`&w%NiBnLJcM0qWB$BoI|)owH=$(+=fRotox&dkXuu=Lu=i-
zi`86vt4>X8$gNVPwjQ&Dq*7w5F00Hg$OWrjaHs)Mmf|<cQ8r!a9C4HeNP^yc(NAP2
zZ}^!V1*u!`(H#s-fb{l4g9U&bhT~(zkOh0KAL-XYkfwD+vfPi_gIeKVI+(vPMyQd1
z+~V7u89)c$_kTH<t1fvB5MbR{!e7C1$jmm&NF6^1;bZ37E+jNH;C8NdoZaP^5m?X2
zXq`Oo5`qP$!d5hjka#8uSEVYxeWQBn-oh5mJzg2e^$xlaQATtqwiHTmN8^=H!C~&&
zijH0C+u8BM*U<oy(muj{fDsi}LOL4rF^fggI_?lZ64La-kNL{B<dZYYEKz)JSP%kn
zA|%iEdNuO?dpZ|_ykE_KBu{kGNqNeU#7(vNHJ05mkG0q_-anLidMJ^3yOO>sg9?4!
z%Js<Y89I?2%Wd+bHmSqpPEp#G(T~#+{UExF9{Do80JOkmcV{CM`AW5_HNI@dgEn)K
zO>^djz$m|UBex+Vlx%)tYehN>H^iqEtoE}cLqXI;erk5|kH?j9YS-eXV+XB^0%p6>
ziX`q^k{Gv#I^Ef>^^~a4J~GY=xUxf2eHxvO35H<#-p>hnR5>^t)<01mrztqFsBr>@
zxf9fm+^=%BzwM_*$C(nig_RX&_-Roy1Ywc8^GF~{O&)X4ME=qocBhv}UW(s$cn*@O
z>4gUdst;sL2m)j~n1Zr5L5GH4zI)Hi$_VHS9Sgq0o#%Me#NBgSr`KiCmGy9|El(Ma
zCXPBK5ukx@uh0SGrC%_B2cZ(jgJxX5!uC)uS>(76`Rn9u`=xX^4pPP+K|v>l-|BK=
zX2~FmlwYQ5WV?J7H@u=Si+qUPCf7NkDOhXd%@LznDu{geB_W;>DLclth;Xfii&4w$
zgK`C03WK3VY{LBEns<W3Hh%Yg>%O!CZmh-PT1ZUKBvH`=Z?oS#w{ed+pFA&RPwVW?
zepu9`j#g^HL12)_2r8N;7Rpsr*ecPalY4NJg|5K8Pbk<h!dE8gV)&waQvejNRsGE6
ze5G5`N9ZOiI%<~I@BC2AKFHY<trm7^h*bZn4C6a|9@K!`pCAU?CbKtGkMB3k>Ui&Z
zXXdC7jmP=7jmUQi^r-sfh=8_Y#4^qNDynjqS^D*0&-&_o6$UC<j(lRg79X*m$&l@X
zs*)#Nxwc~N-U@aZ#C-dU(#x#ND<#|kJ6VVg`mT3JHglt;CaxFH2)`u0tQ=`SH@PKx
z6fA84VHtEt*E;0WrwArC{$C$ZUnX*U+!9Q#1AF{+O(<hsoCWaam(_m42&$LV8q$iJ
zgfqDt*=h5shVs_(Jxf>$Ved6(m6>Oj988X{m_>#exDhz-MBDRD)(H;yG;O!x66Df&
z5t(m`)-T;9ak(kA1yR{+(XE4pZmUC9&u@MU5a(m>Deb^Qz7{=5jNt(Ly3Vh*#K#nl
zVD<=3I)34%iz4QTfG>6=vDBqVKy<BY%&{zPrAjfXj#!?rKwi<4Bix54Jt$iV#Zlo<
zH-*QQb}>A+Ujh<3QsHi^mOmn#vxXn$u7?x%bh}3KYnj5IHL2RZq70;s{~f?d<*Zh~
zjpf(9xLC4arBoc-4yV%joEuG5MagrjX0R;V70Wo796k5P&>NoC3WhMj#_+s;g41D5
zw{EG49nc(L{Q%`~ML%^9`-`D`QT}OQjSwS;hu@z!v*@+)^2C{XQ4y`rqF9c^ue^TJ
z;p|<>)^x2e0}?vSX!tRKu4_&9Wls*+80L3&=f}iVDTQF%$yNK|9uvV006an@;t3Zf
z?{HQuW4*d;a(I_8ZtgrP)!#z+RaM%N^@!EE%Dlg{SbMR6caY#z%{<s}B~0?>gFH4~
z2|Kenh!&2qm*d+8>s{%IB1k9^3=c;>B;kjtVMU+1z{{5ZKyq00^pwLyAK54Rx;VqZ
z$@v(XoV?vnA{|uo8x^wGDn|2-K1R3n+r+>~&+B;)jRM*bV%Ql;X02i0g$cyW@r&a4
z#58Y0b3+-tQ8P#5m4BI?xhoSoL^m&79*MDBqY;kUxzX7pQ;zcY4v9SF@1z9Ijm;JI
z?G#O}LNJZTP`$M<l4h3qfDbe_YT!-65I?0rs76GHjw_f7*-iw!()clA+C|jEU8ae$
z8}V-moIYnke)L^BY}5ksLcB_{b9>O3*V__r)KSvM^t3r<7^*N2BdkdM?fAN#meq-}
z(*tzz+d*;1AzOv%hd52v4T6LUzg)>*JiRC?N3>-eTB>axR-GL9dw23uQWxV4$`)lf
zU&A>hz81Jod>W0r8o%9OkvmEo5a{Q3TF<$EjklwG)WT~X3k(=n0XJA0TH|S!!et-a
zrrz>sK#{9%D^#0~x<>B$+9f|bbySSXjo;Z|$;S35ty{0$k_9WC*=pRlj~HBjFS_$q
z<EvvU7@vjJw0nNPg6Pq?wy0aHjyP=Mzj}Nx7RjPsg+|v~7t(;NJ}5K)oqwKtE0qx#
zRVl~RLQ3)wzaPX0P_8zBkF?M6?0@adb>y?Gy;l29Wo%{R7;Co%HeYa!qL}gJUw4Z!
zsoqzxhWi~FNJ0zLG~4a{PR=2Lj`p@vax`f~rSK$i%^voRNQT8umfB5%T{PT*;ak#5
zw5ddHY^jdXrL&rC9qxGye%U>*TfSvsexaD8n+c1-Ku2$_+=OaEyUfdd<(uNBB;!ke
zz5EV&Ioi50sh{$Fr|kh{+%R|Rh>*cV&7Tw4R_m@Z{0qu5f!~sP;b@SyW8(cE4bdrt
zw})VT$C*m#QX3JT3r0lTDuF3gQmyGR2fzf_EERHWCh(b^dG>4VhV~{A9I01I>eup2
zhm!P|W5ewEKw`x2b8EwP>|`n0M}LH`=?_jX8{|FX9ZLpy7k@Ado(e7rHuaP^0}!ML
zFvn!5<Q=Vp0{I3l;%DA2zTl7>Pj+A<jtt5oyc)JV&W4&j>tM5W9%q$VK^FQo1$$Y_
zV0vQ;ZUAPPry8e8M}F3rlQm+fo4xwUr9A1bUV0YY)AW>{D47mSmRFz8wv>DhW}R(U
z=Q)B-1PNSe#uY?RRdA0RTs&qMPg0f6_Uu7Ooy@Zb^XX>br%b7CfONCMr}qSVx+Ees
z=U**Tkc%G3V-@fQ2e?X+OhlR($2E^oODHet%;Kkg6EPCFg{EU6YFY)qH4S__^c?#*
zCI~GRl%(63W*}A(D(zC^6oQ!{n(5f944f#r91LItb0tf>GTkQO0ScAMkK}YjoSh|x
zuTFQ?YKHV%U5s$)7R%_mZd#gfV=Sd=0O8e!84rO6wMwLnVfG>a+np4m2v~kpA4Np3
z#k$*&m?pOQlQgbb&aL6t*pg%PPxkFH)KlTu!rF{2CE^|VFt*_XQoV71X$5Z0#d!XY
zhFIDfs|2R1)^PS;)Lv`0o>>0Tng+bmFedhX7ATSt*?eA6k4lkNTZvX<r13dX=ool=
zZr?gm?+!UzClMgvnO#`*Q@m+J;xk&BjLLWbcBk~WS0DUb(r|Cam@_7XFnTl>Sf<1&
zDaBWx=#b8}$DEJ*xS3O?9vjbNK|EU~#b^#WbD4`Hs&<ejC_Qjqx*n#}4`jg~5!^(Q
zb^5ZFQ2i4W&m^B-L%$K6loGfaM~2^oZ0n#_1yW*Re>yq7J73?m=C#UJ)1extw55&v
zJnfbREvTKQ)$|ap$UhptLQ%wK#mYqk8DBtVbN;rF*07eI`?8)P@H3Nc!Oi{0Fm)K*
zFgJryl5gk5rgeu58qN2dFXxdB&WiRaI7T|y4cNbf)=cfY<x8bL$?A}L!kppXk&byk
zNtdeCBw9q&erCAjudNFuD8djo$-AD^OLXxV$LV={d+K=;ae0cW+mSdHcgMO9;mG)Q
zmAmJkyrxslkJA|6lxS^o<xl=d^p^rhFJXjam%+UCzV`5l7Mp5UqzL{!8vCgnn+P01
zZa)`E;;f5acedf-!E``9@sTa<ULUPup%r4@zGYx>ipOxPUK8r-eNOoOGLN3+duBar
zngRW&)9f4Wh?OE;GU{ziSvc)&wnAKgDX(8_jk|_{<67U7_5JE;mJbalKBuMEGI@vj
z<U1(c2P!l-_+%TVbMatGwFYDd<7aprd(br4`sNArtNz_rO_~1lS5a&9@2cw1k#bU`
z0a$|vLe1!LJ^S`s>@P->B<FWc7?pS~h0Pbc4>|0^L_29FFH3J+i4gkdIbU!o#4sg}
zWYaYng1`rJ;9&xE#VqyAds~b@J&pLpoO_+dfAMJ@^aQMJ@q2s5XqJA+E?^sfxX4I!
zO0&nitgI{GjZxBx{2u6RYMKaHk2d+tNwTzRck0!;Wq6-U6oAxcOfKZAr{HViRo&r%
zN@i)+_$NquSRB)2=*%xTKtEl)7OxcUPMs+>tu5h~Vbg5E9&)`kYrG=zK^mX*y6O)9
z`cQ31qyvPz2U}=O!AlJM)xLX$XRe0T73uL=>>-!L79(dFf?Od5t-KpYs#w9<SS`@F
zv+9^y=<xS%@Lkt$@e!J3q|?`wR>VRYmV>WEZNAE#J<k9356X--GbCCNoxDqwY;*rD
z{iIdc&v8TF>mx-uYDY+Z`Y8#i^mo)$^&Mx?;x}D~yr$NC1?ryniDy*Fl2H^%x*nS2
zFU?M5gB2M<^X5RGX+_G=en_8CUiy+W%ymjZdeoX%y406Eq`av5rFc9?9=CUB&p~8o
z7xC*z#CCR;ElA@!U!E>6A|-g+y{qAdnlH2&Zgtt6mHt*~CGmmj-4+g=oExP>$(6xC
zJe71HsJ}Gb^R_50gBiR&n+x~h&|Xe2P7TJ35{>jzZ)ZZuc4d)PW7ya>X%lwk^=u4X
znI9K9i?~riN>d}r_gQ$Iff`2L!DW&YiJY9a0~@+;nlq@CY9-{|j<6q&C48M?6~LHJ
zzXTb(uO2TF13Xzxmh&FIn@u?T<^N$xHsuB1%KSd>aAvJp=A6lR?O#RnK6_qpE;oo;
znPu1_`_Y(mA}_^Y=D}=6zuO{v-0@b>#g~{`RSEgKb(H7tqUGh&icSqDCovL5O^r!j
zjU3A)=$wR^1du2P@epiWgZ`38sl!PX*yQ=9&A5F;6e<r97Y_?@RKRa7nfnTsz-~-!
z>8v{4^Z-MK1&*|}u;r+sZzrm{eGe8|Hfz8l;}4M60MRjddfG`d%U~3<1EzbIJ_XI)
z)p6dYY=s_CP$jxtddhrBbfv$TiaC9ffBnHiXhUU7T~JCT+Qe**$+m;V4G%bHJTm{Q
zs96|aqHADUN?0t5MI_A~RL1c9sFFg-L&`x3+8FFE+8O;tts(=1`XvKZX$j_g_Z3j7
zW)Nv?v6OXl?u!nm{;sy)lOLvf{N01AHM2Y}jx<M^EY7!EM#m~E@M=I)KXh3r(s71*
zvRj=MWf8)1&PRjY_s**v-5R9oNrkVf)lO*3(B)Bs&ew6+D#lDDnG&z7`Qpbg9|uzt
zB3#bye#OI07_;6}Gn$~or>BdFHo&E(O>@$=#LmlU<`sT<pg^^H3z6s2v?O|ZoA#F2
zA}jF-jp`N-nYh#H_V;4#9^=!5{yd;~{kTplC!#q>4L>bYJXZ<X4bN+^s<9Yecpt92
z^Hlsx)K-S;ehCv;#)IN>nQ;6tGAZY$p2iV0sjdOt934>R973QRg_Se^J&2pADI87<
zD&_Q>QaUwm_B^y2#g1OVWAznrxM#f(Yql6Tw6UXHHS4tF2wG0j2db;Dq0NEnw#0r}
zr{$8LNeIenhs&Q>C6*~3ox!=;?L^xaeoXgWkrYZ6Q7G!KWn3La9g~wDE(1*k9jW(G
z!Fri?jk>dZ*l5_7$KAhyghY=Qwe$8)0pa;Y;=7bnp(l;56|4(ud1E%A>a^Xh;Fq4z
z96FlwT$0OLP@YhC9U1^{*Eu_+C8xS$1ZMU&&k;fU3Xd_RoZaseNQ!8aH5u_lA+=MS
z5t0OXq!C|Gb9GfibM_?z#a_QM_4LEPH~f_ZE$PjOS-;%T?Q8RkaU|8{bOlb_-(ix^
zXSiGYTQ*lP`Y=ggyvD>eJQC3cxznSJRHUfSWJp9rG@O|sw26ubR*|Y34Zir7A?jaa
z6Ym=tiggdGE!7hzpy17Q$SF80HSLhjJ_q)Vu0Ojze)Vj&gTm4^z%MZ;uE$@Ff+)L3
zv3%0`k#n`!oiP3Rbz95^_PK1DyIMU=P$Q?z*SIiW<k4D|aFl#y*vdpSm3N&7y=D+(
z;kaP2)*Z!`*d>yx`uh9hlPgzpTd4BY!Cq<-%JA7F_hpZ9PzRwu*)x_&J>eu!_&GE2
z*4*FzU;i$9aVwtZHfUI8S2P4jg$<<S80QBaa%iSCSidsr>nk8q*g84E+YcqR#Qx18
zyZ*<q{k67v-}=)}Op~HVAMSn*$j46xxdW)j!>upgMV&Z|IeY*jT2&R}n{1+UIXPmt
zyj>3VFo#{vCD-eYnl$M$cTr!*rU{K^UnFBYz#iUz76X8&fJG#bNB^<9ivIU)R4Dku
z3M(Rpg2io}<EpT9I)MY(jLMT0t2{<?qV~%w@f0*gg(m1<+9qQ**osopYbZffA@onv
zH*`;aj8aX>7T-@@-KWBjJ{NWizDdstbKGK=YFcEgN~S{a0)y8C_9^ye#6|9I!xj~p
z=9&rzjSfQuzeUOQ5ppx<ZJ9IpV@KjCnl4|`Fu`?bNxH=nVfVA2K@6N{@idaeItoQn
zO4g*~QwksLxo1yA+aXEF>1~i&%g~ue<RePdSqa?XEbhVGEI$x(rCyXAODU*`L9FeA
zjH$g|vagtEHkV@wo@F3|x;#Z3wo!Sb^`2sLIWJ&_6t}KQL#BZmOE*T2gdva=IFrUR
zG_g<HE%jncwj%8RjNx+qLva@lOVawjJB3w4EBIwU*y^ugQZ{bx0%H=&n7p~AT@(mu
z)t<7+r{hdt=$>kRC`+vrA4Picn;lCt2S60k*ViZD+$V2dY;JB=V{9=6NouRemlm<r
zs4+Z%F4u-yH!tD~_$jV3@W2$+AQiih$tk`J7;#^?C3^nw$c+hlyQ)cg<|%)G;iSkC
zv7s?=LW-v;3ad9Qk&tQNQz#2f*PLS9NDd^@l_+vaD5ifquafR^*-NNivTx}a|I@!k
z#DS${F+(ZgF9XshQK-ZQQCFlV@qb}D)3aIRXLtgkO4ajl4Jf4iabt4!YE;k{vd&K9
zk<q)P!IsRhi<OT2fw4hXbQ@?;Vx#=I!O!QJAkV4LkwPMgS{lSA;8dftG7=U#SR3?%
zug8e*8(|$0TFN<j={vtay{bJIKbt^F$Pp82^l-5g)|MR#MavTUkxJSS0Z-HQE5<$%
z9g{yKhgC%`^ZFXh_>dOqtRIWiznuCFg(qhE`K@c5Os%6@)Q&x1CAs8mB1QSC&H7&2
z{(kDWsNr!VIeK;E&#3Q4+S*!+=H73aEItpIazriLHvW+k*Yje8ij<uoN2$j6=zgk`
z{~%#ZaS`O8GxIvgW^}yBvY7<z&KY~2ow~*|mt8q3co|Og803z5a9j&#^`z6Qd}aT;
zsFi-hsjR;7{vi@d@%4A{U%!R7RfP^-%!DG)Bt5kYvqc$Uuq2|p7x{tRvlfV%HkdUz
zTDO>^c{b5$O8eNyH1y+V^pBQ8FJ;EkI|XG%6VGMciT{8m94>?hL`q#78rMIo-u$|>
zb3@PG&a9zi63He33fFS89S#5*v;o8x|G4RZ%evm{OxyYAjZ6>#zP&w}3T|w5YLDxw
z^JtcGYRUcRCog>>+|29@W{LEe9@Eor^mYV0Fi}Aqxl7Hz=k2MANA5?v&Bihi@^U7;
zGWQ+cV>Ae%Nl7g36A=a3O2+P|6o`)D&nSqEp-!|Wl}Yn=LH+MufE*=(mh$5Aq?N!U
zG<<hA3ZKU>Y+b2K;lDMu_WN2NGL7pH2;hDr#9A{ED1RIPXK6lr@yxQ6g=v;7>|5t9
zhOBn-m=~O?UJlrI38w|qqe#~|or^O0jH5XZ<;&pA+Yf<v0y%%t@KYYp`bt>;F3o*}
zXgojwk~63Myxc!+-lvpB%t0&}0=m({JnpG=o1@$7y$7gi!Rdb{*B@rmm!WqG6cL3v
zWrjX;ssPRoAn>s}t#YTOX}*M2lNGiBOH0uIP>+!kp#U%XELzUAMDg0*MbN~SVR{b2
z+11tcFP;2(KjjOFKnehL&X-UORq!HCj;|jZ_13k+s-$J1OE=#H{L><54?ISPpzP69
zy3RpkqaN!TX~xVX-o8&Nl_twiQ*y}V{AFrMmU?(K@NOgrI3A%(Mq887-_Sx3;^Xhg
zO!2GGK+?7HpAu7{==B^J@Z&VAR9vZJzrF#wf%1{Bb?o>IT4=wq8)b|VaM)&Ous|2d
z?%}cKm3D_Ob;v50Op}}ungkZ|bBuuU4Cr7vN%4<~E_h%`5pid9yk|-O6bmuz%g@t$
z-;!0BWZCHZtNl8mxlEOgy91>Uk*c=rVj~PnpooZwsDP`tg(n#gVpShI92k;pFGN2)
z?@O4tWP>WC%D5uZ<S3%e<xNaJL`pzYun5kvR>Y5`{`5mwI?ox)?n9>kt`+MhQ3~=#
z34k5s@6#{WZd&ImCVW<vJA=cFM;sb)MrI7yA{*q;dK&Re&sV?z^25j#-!t3n7f}f~
z{zg>E@xvCs%+{jEy><F5diGCX2(>qycm!Y|y$eA_;_Zeq3@xhrj@4c;#M_qak?A~y
z@F)${^0u^<N#VZ++xEr3({DOt>k$eh>pW1)@!KV3&?kJ_ouT`)_u*|h2?~|m56amb
zOM^mM2F6PKJ0U0Nc<pjK+$IP!WHgm_@t2R8{nJfMVWP3V-C7l7`8@@@ihloLILe=i
zb!1NH+_`9XMWIFgu?ntMUQSvrd_q0WjU(*3h8jHK0X1g)mGz~}AG!Pq8@ee2^rhvR
zjB*s}(hF*jH}+<3dvx<qqvk!{tvZbUM$?q>1B{V)DVonxHhD?*80NlH>`12M_~?#!
z1eE4aPfzq2z@q~p$3Q}xD{!bYhU9T=Drk{qfCQ}&i2r#Fh@LtturQK11yh?iHo4>!
z$CGI1`9){z^L-G4K2-N7#GeDE8>%!lZ%SM_H*cV5(PpnmGEhiqd=Y}h_AZ-oXkm|k
z;{3P_jc)UYPD*BuT;re^*fk8tWydbh?F2HNy2B#u-6_?VYz<XGo`-E@$Si3oYJaQo
z<}D|{gQLS%XB}+W6YaC{Sq~0u4AK!b+#zu8+~Wm59qY9Z{FLCyBd%9HoXSC-4m&e@
zcFo?`&T|;muKhc@#O^3HtAL9!V4K8<>)u|~1W_N!Aya?lX>ex!y%+qJT*Y+*A3j5D
z8&K53hFQNasy{UUg6#9&S3N@KSIxumDn=SYs0cA}pCBtXOdZjdtH`~Twb!sRycKs0
zWT6tlCbfjNe<A*r-Fs(;K~iw#EJ=@9YeiN@0Yd~;^b_KTw`Wxdc<L*7SnTQHk9YOm
zA6Jxc41o}~pbJIA`|vA(QQqTvg-7fSLfP?H`52Ba9%}F901$9~2Lf)-F#vN($Zd;>
zK*}3`1Dh~XFy;3cB>Gya3Zt@UAmf}%{}!-iMpZ|47g%x!Z_6v^_<U~?Tp?2aeRSHE
zISK|IRq%aDjZ@o=zi~6M1qmuXRR8wLowaPx0OiO-hoRp7WIrWk+R0ucTX%ZzFVig>
zgkRqx{C~1h@~E8F{$*TzFMjT{ABmgs^QuMsv@84}vs{8zCEgJ_d65x0@TOZa0v>Sy
zqu9$SrwfiqLyvsWrE&ELH3gc&VhLe{yxc@6Z0J)s=INW+3|vc2g9g1;XEGeJ6IW$x
zSi06OT&tsz?2vayM)$ADA11&6P{{`>q=Nk0NOtEQwm(Yep_$ePDBur=)oHSGXLs!O
zLLZEqJg56_H9;XEz4=$vL}Aeoz^Fl}rADw96-KQNz|0G>p3nE71_rafC4NO!CyXFp
z$Vbv}#f$G+AS3&990X{qJ5@ccOVd1=^HmLJNrGQ)253STwSu9g^Us-e7}29OVvmIK
z!HF2#EAYW)yE7K`V~0z$n)dn>?6fe+W@q9oi21nKPGba|CBXbACl8dWo8k3gdL3Dj
z{pKg_#3V*lW+G|{!HlHqO<gDuBBS0|p0CHr>1iK<WC}Hj*SvcC2eEv5w|Ct`IUzYk
zY39Af==q;^V}y}}o!9wQ_;(^i6twjVSff+3y6qzYMsE~=$~jc}2{i9q0NGu0Kea%?
zyry?-^}4}S9RXnXPBN)r0JdY`MLAuuPuO@@m_o2;SY#v}TW7iP{;cemCQ6T2x#LEw
z6EOg<FV<zg1%P9GJ=S2l3AFfh1sc8ro}qv6&fCYm+SkttE8NPE)AhXsp#(tf_p_Ic
z27%PumH>QKa&O2W-ELT>CTEx)!(+?B&P%02S-r`(={dRl*(|Uow7L6<p7M$>tk8!#
znG*P!6e*;V>)F|bYQrZ=HWm}KU%p3QWvj6fqKI~s>%X@5w{h(Kqv*%sdz<s13$w$}
zP@3cQNbuZmc^^p$)u%>$M}%+97)&XvfSs!h^b&W9${<8c=Q1#Hpr<*mYU5yzrSz$@
z*4ldC-xd54e#S?w=|p%Dnx<q)a8S8CUY4n8uS1e#m!ew~NxN|UEcQYc!Z55fB$E?s
zbBD}qTc1tCaQOj+|9xkU<Jt>pa<e)<3<bQZh~Q_d@eIm6?21SRSyk)Z<h_xXw(<xG
z#Z1Ii?NQzNJ4%4(3~T?omn=*6PlBcHMqI&RseRr`sORu!uHf%L4v&Mowa?pM)|-IC
zY#Z}aU4%Lw$Ma!3sm;sAN^eoqchjuCmGuO<Ma;Nqkc4V&f89=ZfkYABMEcl&oP9zL
zb1CI-o3`cFxM>pK(_iv-1sDI|VEj$yRu+Iva6Oy~|GK-u5aTAcK>c12BsgX-wNQ8(
zpli&+S%uk06U>zDonF<}@(m2woed)`CC|^z=Hji-3yRde=4gCO!YPMlqv~rAVc(8&
z3fy|(3%oq&-^8eupLv^I|JtzauQ~)r-ts0^g2kvm#Y6~(LPd$)ADd)x+hM=H!*#I6
zgc;fXF!ikS7hns|@)p)hrpz3<oW<@pT#s}mx&z7<&HyPP(1H`w!~^%wW_Ol2d=?3X
z(X}mK`~1PW0-V?+hHpdm$@zh<XFve?qbM>l_|~uw#7PoYZuF}QozTjZ3gK1`f&Y+}
z&>kuSR~|#H*Sa@zqbUU1Obg8K98GFGE{r3qko|na*mBkmIV1EMztyKacNy_r@Yz}9
z=KM1dWSo`p{U^)7UAHxv>Ge2o+PaU}=w)qm)rOW_aH7FRIsR<?D>L^%GAs!3fuoOu
z_Z70ek)|$_+DHB<w)Bdyx9P#pR=0)t>%G{UG+eQEQsH(u7kK|VhAQiI-_h9bga)v@
z560p#K|KOIy5(atdLLF~%o^vK-<KFY;SW7;ua2wDF=)iR^+B)p#=1|aKt%K}@%><%
zLvo}9HY!cCyA!gO;<E0ju%bUx+=6cw30dYe1;xdmt_E-P=lu8<5lby^Y0*tly>QyS
z_Y%cjQQna$(jaqLs&(5Pi)fGffrR}^oaM7AGIWD7!_Km#lQ)+p7dFrI*R~av;@Hm3
z$1QAuMia53a{oXR*^y_&UY@r<9|DdgKbS9%BtHj<)zP4mD{8{0#n4A1nrVn#4_k6?
zGT~>+qWM%{*J|2ap?_RxO)CHCMC^d6&_SEZ)#A;w$^Efr30qmMiclyLG`(~Ll}xYB
zdXv_^qk%Y{WEQUXWd5F9cX_6F^2hpE`_ZJht`0Aynl5Jvt}d}Qp=lt(2eQr{9?4sB
zE>A$?Gq-uk*riws>h~rKxK6A#Y!3q(8dijo{Gca0fF+zp_ktnhNKMS=ynXWhO(Cbu
zwnhMuQ#34ok{2K3nwszt|GX8PLM@<32lbvJtQmj41mg=wyVB=HLJ4ue`U~t@C~aSP
z2JCx-fb!k_S1E1yF<9O=whEKo<M0mc8F_mEsIvm*FP7EGh!qXm<Wwh$9?U~*<}?Yf
z5%JN*^foNiG0WeOc^TmK4#%L4%qKzeJQ9Cm;hAEn=$NqaB|(hVo35mQ)@ZcOgk;#0
z`P_28T>W1Om$tODe4O2n%H}(OB50Gk*{!8vJ~1AOK*E8_;&(IuD->~tACf~4a&8RS
zy<c-S>kWf%=Z<U17&KBO;Ic-&@>}X8;Iu?s)Gdi-@9uCvoFInmp>bbGlNvE!>i4y|
z?eYJ8+~XL6mBX(j6Nh16B_HT!kmul)WZc2t?oi01F9{R6dLlJ4GP2Yg`q&=@<vOn}
z>-%eRv{Mcq(MQZ_vpHtV@WCO(kc<=RTPV`Z%0e$7`s4=|{xBs9DMNX??H3l`snqa9
zhFScK;AIk4%hs;4)Kg%HnbCVUB)Dx^VED!Qe2)E*Sq@FNQdI%#c`)rNN$hth#?5Dz
zhuMt=TnE(gj!anB*A_0ok^;o*Db&BumJ&e~`?_4(q26)XTbuVn|LhVuWxwz4%#@BD
z096>#pgi5zw5E(xnq=?J$1UZ(#zjgCd3%r)^>p!NKsq^&9(qXEB4Qz7+cVR;#Sp>a
zMD`rH31b9#f)*>v!M+U_oLY38CYMR-Q5-^^=jStoOy)D|{pB5kU=#TsIb1|@>}Kyq
ztpmW*^BsAKd}O^FE4eAESCRjZ1j<<^G^h}5ogIR|$ZUi-B=2g=^vTVQZYqKB;`tC(
zer_VMC<B03Mgp2=zGfr`mW>U?lXJrfpg6OH|IjzL`bv8eG^I!tc_sZ!2N*?PJiWaa
zV=<-BKCL!d5hIX#V*Z2KL2iQR3WG5ubT3kyc}*w=XWO>Gg4zZ_ll8hg)UTXQSnA2L
z1%5d+iaPwQ?OsrvV-4S`J%@9MhKbOZC?`2JaUmmAiPuZ*5FngKwf*pq#hv$l3eNf<
zA!WEo<Xud?pUk?+(S+H3f1+u8(OaQ6t!lnf8ISyKtA*RcNBWX-7mr2#>Lg2X^@u+y
zE^)l>7Fw`{G$!u_(yY1n?&2g^P$;1)NKh2{y;GWN32bjmez<P7&$XK!vPfZB4$*wT
zVmJTIekjU^uorsq)fq-C6Z>u=7cn8*k9+RB>Jgx{+?n1%IN0hu>-PUMfAlxB^**fb
zrN)V|ZntiY7v>4$a=0h(D3`=zg5#e#9&pCa%+1ZI!7XiWf|DpD2>_9WtaA1UGy=LO
z5dVj;zt*f3fXSJ06kg+$_@CsWg_%C8jdSbX>*dQ7_R8o&*FMaM7nNC1fJ$t}C<0X&
zGI3W^ciTzx4bFb`y$A$#*+hF!zjYqvKWFhc;7cHa1&&Anqg2E@>v@1GK6$><geAdl
zF@##CA`(Q`I<L|$r|z#z5BMwn{~fQ6%MtkWd*gC10w+P%VI0s;{>cpf^PaWqHwch`
zI{j(w<(N*x7Jj+@W(as);(-zQ&ED|b($eR&lz)KWcAk$H9o#V1L-YA>HNFcQftIr7
zd6?(iN#}6`e^r=%8lmbYh?w90fSq1}YrdZw5l57#N@hK>ht6NAqlu+{YkzL};X36M
zY}y}(KCj<o&r|TSPp8*nkEgyL3*+zaPYC#tB>`JvSJzj-nCS-O^_(o#eIQ<PG+qAK
ze;Y{P@RPe>EnI@~<@a&ogXph&<u%~()?3)4k&W&2%O+DJO$mz>+<Z=yq*OCro0fkK
z#l<Ea%HboAz-uv&Cv`^6J|~)XSEW&J8<*VCF>;GUbBO8aohat8TpkhWOQiD_%l;^X
z&u)jVFNd|Hz0udg%7TsRdK8G!bX{u#RKlN-^Wg20R8@%)O3cv3iH@l+#0m_+97%5j
zU0(eW!Y=*5M2E|w!{!hKas&Z}!&#t;0iEt!JYZ4@O?_Th1nO@CF@k4lU**CXA+Niq
zljI#~(5E~rX9KS*qktLde~+8Z_|py;7HPg4!YBJ~u`r?<TBeVs4agcXb}2zR?^GN5
zLZ!}6CQ($7n~tcbu|F{-h~RTpz`f@ZLtDNO|Fuu|cHkQ*rQ02YWe^{#a)I;8cgkP)
zhcl^SW=%;=OL`<qr|!)z5xVH3X*@nb73OR>ALY;|WoQQTdQmJJ`)X1L7QIvK!k(vt
zx=$yinNaRV-le$7j7g>Ms*63;`Ha3Qx))N%QoeN)E8zRENGr(e9Hj_Xv;=L82LRN{
zkC->}slgPAEJKJx38ceVvLwlLX5mNpzW|f+w0>$C!(ee$2>}4sH)wSn^*ZnO2YiIg
zpGmz5(%oe$G|e}E`u;QZ)Ir$_Y%|6fxf2vtc_2;mQ=p%J1xs(llg<H=s4T}#Ig)6v
z=FM7k-b+=QEwemmphT^rq`A3ms~S{6Zq;n?r&rBadm*YBtja{K*s|zbEU0vN1ZbaU
zOBEplX47{F|CBHPfcXDxZbag}K;w6WG;sY#qDWj;7GYy!BP11%M9A9_T~JzG%Eg7P
zG$k`6IVB+h9iLfOyg}D7Ns@c0-gbr(pUoIZ!ud1#vfRhPX3$#QYr_&^0GtkZvT#FE
zRRkGqR2JwE!y|Uj8eEZvjWM!B0mmPdQZ^^Nh0^89J07SrfJ)|YZH9>T(vw16q+c_;
zu`x;Nv)(HUV1ZsXVWNc*@msNeeH>;zFHOT`AM#&~Kn|ct>f1#Py-ysq;sR30!_!r~
zr{%&8*y%^aX~HUNQ&t(}L8e3TkhY7bHQlyP&~^)VidEXxj!^vlfk|9{n^T!C&UMpZ
zMa0pot@VbABT41p0EQz82(ht}E$cdnxMf+OwCEuJkGoz21dbiS*jq25>nK?Yeps{L
zXpVVYjZz{e8g>;!s(8AhNYmbHwAWq%caRk+UEo@z0qbCs5j3c>$TB%?3FHu`w(p(u
z)r(cg$Dkldgb%i@c%!D<D=2~rfBg7SWP%Z>(QfE`&FtINce%@#f^SlBocC@{A=?)v
zquzdybGJGS6Q2o%o_VZ{*?7KpcF+i*yKfZ|1k{CCZN!mQjVAI%O7tcgrCfrgnNLnj
zYBVi>AI}UlFs&xJoo9Vp$cY|pg5;as6r)pPRv=z1DcF^w0?b&&lT+Pj_MACYF%h<b
z5l-b2dw8bKv%S&{Uk2)z;<e}}nwWuY)W-F8zc*ri`BK;p5X0?oU$_aB{t}#i^UR2x
zS?T^@i^H;ZjT$|E^qT%mGG$AYy?*JkcJcSRS#oHiq2+FFFH=*JgR^0$K&#bvKQvn3
zQNX(yFwtsTvRX;luZx)2vVdx>|A^%39o5j5cslSdiF?oR+D|D5$q#}*I0FlobL@J`
z(W<&hoX~L5xaA>^k}%)9?`&(kRNUs2mu?$+#1iE+p<`%e7{?#K$zbDgaT+L!U(%1-
zl#gN{!>KC{y}e!3-E3mygFU8%kHjoKJ;vz0Y>#~~mH*gx-%a)}x*tj)i%Uw<`+^E-
zAtxqyFe^4r5Idk>U0V|X46J%UYXQ{Honv{~qoXx?o-Ff{T%5|~(Rl5qN(-OoBzZY>
zLhYKX!)W=>W|nvq(m|O9)sTnuXW9#LxsUiS>i6cO;`l!$TobPfNDEZ<r33FgX&yX%
zn!?qZFmZ__dKlnK7RjmP7v^K(tf%rPt+w7+0ehSb!(het{4oHWIGhDcMCg(QJ(|_F
zJt_PJe1vB4JM$W<2Y@Zy<~*xm{&|L+qUww)oX*+)f4p<CRdts4=PN)6ebGGU-I=)c
za>y6|fiIt7_``Vp4nk@X<S#$<s(Kjec*Xaw^bdcZ6SFw)bpOqyK?z?EBYb+uzS;pe
zu@8{o;SL~7{KDiMt(nHeAB8x2&wak9Wxj-W*7+Rjy*$)yoWT9L;!S{J{imQ@FwZpy
zH9fBMnuG*a%+(fWmhh9V<;aRn6mhUuIMemn(wIh~NbkR;R;Q({(~|x!`|z!g7cPe>
ze-tj2=;lS7VZ4Az(`&HCl<C=GjpKS-GEG&dhgc0T+|7|h7l5-%Tf^Uv70@No{Vl^X
ze3mGNd*rL46&Rteac;g5q7yRL$CUxO>VaXmIosyy>~Yh5S<QL+_CTf#!&#KY^L=6}
z=&`_a9YS0mYq9y)=jDm7XJ@4UGbNnx?Z8|B<o3hOz+R%7GK)*#nSoJ(sP#PAgPnip
z7peq-{JugnRH1!L88Qxzf<->|BjHv&*S`fOVyPzuKSLNBENWIefD9rO?8WbutzWIq
zT8mL~H+;@oD<p4ZHgb4By}y*F$D?8|e?bk~FEsdez=e5MWyE}r=)}<ABVC-lP!!|;
z3(VDW8u8V^kXg^-ImVS)887aCM_!m2Z7SqiV~E#Cw-RAxi9K(R!|f+SLb~;)6bgg5
zC@iKl>S-ZZ1P;P$rTgzgFo$X4k>uSDLaB%E-Q+<)(5kywy|Vo!9!+9EfNLnLxt7uh
za~P~z@#Dv-aw=>+w1gBro-aN@JfR01CYE$cM#f(Ps+c{PR9a=T*%Enh@F3l?1k4W*
zkj18@1FSs$>mi_YY780EkI&8=*d1ySOlIdB-4BQag)%YyT{4-Q2Ha*~(LQPCh*yM(
zS?9u{nHlIClz67k#Nm6*Y-}i(rnvu38>P<<Q44}%SD7OC2&@e|mGTyk_<BR7DbW02
z=c?P3|E`tk`Gr!i6AT>y!ouR2rh#Ib5_vF0kBTSaL1v&S1;yC6{PQzHX9;-F0`6GR
zA%e@cy#$hYLb0HLpi8}s2FuYD5=?)<am{}_LUjUQ&FcWg3b9_s0p3Y4U=WkTKHA~2
zo*;<lXdl*w0z0?uyaNM2u*-kA>QD!7U$z3xjnd=m<HHAB%JT%2Aoll><R?npf*;kz
z5|ad8KwBAwy3rIbqPV?jH?VxVU+_UJNH`n;qn3uODIQoGn_gM@KQw(~TxNZ=cD8NX
z=497oyQU{klQo%>ZQHgvnUihX#@W5+J>UA>^<j7Kz5i>i3+oD9NCU1c3)yoE9E8)Z
zLPbJ+lv!~7!~am5)ZvWh7*w%`I#V7IXH7~b%<QbcBpL2z92K0?uuCf@4;2>2QpP3`
ztTMxKlDA^5ChiZB7l<z;Jl336MIT9m2^Oxfbn|eD8D>G80<r#524ful7`#%ghat?+
z2V^s0I`L6e;jjy&Myn%QyQjhDbI<gOZc!48uem#w{dc+CzXucPw286=EEX&la-?Cr
zM?Iun)tK%=+bO!NO%<%Qx;@7ga4RIbXun6jD)c_cK;EGmP$^@~Fkk8fvkU5RY+@-z
zfGA%<$soa}{ba);fUwGa+KPnH<_4_>#f&Oc`%VT#aMM0MC`B*;w0}`Wkjb-}$FWWm
zTBdr-`$Bq(pcfwq_l6s}EDl>H>o%jUy)L}^^g<nwLtMwPd))o4_o?<zyeVcvjWow7
zGgjQ+XZEhEHP6R36#_vk2w`wXcOaXG##pP_3{w_|Sv5Ux7Os|#%zpl)9iKVq`bOQ^
zMqyDOo*snWF_a#)4C0o-cCXudee=!!T+CjDo<s&t0dS-Yfnhnj_|`-aIr2!!r-jtw
zVYb#*=#`=j0PGqigs7yNDm01?!aT>sW^{ar@*m<sA^kh`h6My17GDmzSXnZPXP1EJ
zQWeyFhg504>ns+tKMy2`)oIXc*8;WV69u>vOH-nf>eE#@qj;V-6TVfH7X%(^H<doG
zNPSJDKO<1pH6!U6p!_h=0zkHfO@`26d^?17CxA$-+aEwCL{p8|tIOzoLVc0^p?UWk
zB#?@elb;`eNw0Ags^|mf+a*Zd$!{@D(sdQYY`2FhMJBEWP1DK(L#2>5Zm+5+{OmkW
z+8>^AT}AG;7b}Is=R_k*6A`eX4>pKHb}W1qkyE8KEOja=#5|B>vr3DlTf0SfYQHG0
z8xY6ZhQzX(uJqSAaRL&;UVrn4{c$8%>%}%xS+=WSn4N~f#uC@)2d6J*);vU#R+%T$
zZ9WjOm(AQ3LJ<}t%648+-?DQv_Po#R9a@_-*8QsA^`Q7E&VVUBFf=KU{PjWA`+eg@
zh<^okWR>A$^<XVqRof0+*G(Zas+<JF{I5s*jR;&l=K;(&VAC&o)Q=y9(s>~uypqoE
z)m%`tm>Sk__7PXzi|w1w3LnpT;cgeH<~#=}CO35ue!BgCOsYO@8vi{vhCbv(V^>qg
zHk0H3$9(9~zCu+a*QCMzK^z8CF<ugvH*KIi9A~~<nVvtuM=ILtA-02KfOw|e8x!!y
zzwxJ5a}s>l6))54hN*aA;OeigR3P+=<^C`Fx_LlF@dC;v$U@*Zf0T}n<=UPDA^vDl
zT4Nw4iomW1Gy=~P=luM<p|SBgvD$lC#sWUW9?iEgg+VNM?-PGs@DwKwD?pMPL<cRj
z&)w$|#uxHRj&`W7(VGIpy{ub7>R+3;AQ{zyGL!w0!KTm+l8$d2WX5uGvJv#qMjr`n
z?)O2=Wc|L!t~UwhT@s$x3gp}V@~$b{k5el)GN#wb=BLAIuX)se7(iaIQ&q=~$J-6|
zXQ$MVv2k@XXihy`(BFNm(@Lkp5@)|{pCZJ}aICB4DNnXfy^X!ou8IN+kiv#q8rqhR
zHp5pi*vH=%UH5+vbHoe5jGCZr){K-|zG2d^O$|9wG6k>2^;Z9@NmV3x^_bH=I+LeK
z=j^M0epea7f*<SD&r)@iztj2gd4!o(>l1VzhbY{bfr@R?l*n!O`Tkh%Eyt~)AuL@H
zn#y7fcIo}tZbAHQ8kAETvcj&;N@ZqdCXRXuh*h?N+WBaV0g;b`AcasZRwx9Lw#HN0
z?Sgpb{!a@)tvCmX0x%8-Inj+tK$8$StH9659b26o0^l+5j~gDgOxQd?UM)dP=q}Hr
zgO#P_h={*cV_JZmWm=u*^Ud1s4nV=vgM5u$n3s{51#Tn^7U~9sUI1tRmEKi1E+YwP
zOIIR;<KpDBw`g*<REq%@ltGEErL4l|YdI1}37}g48e0PLsTA*Se1okTMXl{Yigv7y
z)|~@68acD08CJ<hz7=AWW|PQ(rWO;>7djZ$^^p7qC;|`%v)QFu6W-Uedi=O5cJs;q
zo^SRUw0E>DRAhm{$`{2SKpNO>RvJ9JMPY!P0K;0~f<4>El9|o-hGYNJJ`kC->dFL;
zgz5SDAQXQf+4n0TfXqA=hP35^k2-W|eLWJz-=Q^wxI7^aV01E~ecpC?(`95wCb&1h
zo10S;5)ys`n1}C~e2t(>a3m@M<fgXyS+!6cZ<Xe?5L?U*tIPG~w_hPVCJ2m-jJ%F!
ze?w3+Ja0yL^O-y^fneie;7Fp7dAiO19LF`!h^6k0r^qcX{`0knyGvI7ht<yC(C>m)
z{^f%8oR$i9=b~rehG}LU3Zy+@Z4i-|F!Z%5`Gq3dL85WX2<05wX0?4r<h|2kNP8K#
zQxn|xNBPkzG#FpQv5k$*H=qv(u(lr!#(jaX9gm%t`#`Xi84x)Qf*|BcAqoS9a3lpx
zreFDVVt<gx=@trZbCeevsp4Bs9~z2(mUo((p3{l`6LrfdvJ|WWFx&m=ZXm{umxhqS
zS*}oHh)S}qa;5*lfw(s2uZl7&lSIjzfLchng(?(5iheE}q@Mz(kz=q%S`lx_{Kf_n
zplDN^r1JJM(OlGQk?@7KBl0?fZEM2<)IzxI4ivNI8DeE&zdXNtJWotC*8vfwd~JNJ
zeT@NCl<PPqINaU@=Gp85n+`GK?LY>lR)-raJU@1==RO<~-w{b3Pu@-;DSaYb;s((|
zrLL5Db36-xKPittEx8wDWn#GFhFe#6a&3O*M+*%Csj8v7DJ@z(Fv)ZEg9qKu^g)Q1
z@T&cdqWi=%ew+~FW5Nc{)Lw8r<ZQowhJQb|7u&S8I`3CgacUf8_wfo_vI?seEx?59
zt23P)^kj$;wYyw@1ikur?9LvSd!&lp0U}BmEz{TljEwK0fvVgL!HTif9~E}rftW4o
z`1lwY(p8v}X_Lpk(m4SL8URWiV2^&ZWZxbbS$UuFlj&)o6>zJIdADjOE=9caHW>xj
zqCa7T-WFgG>DiB$qxB=>>fXAr!P5AV_#?=c32{Z@VLE#@=OoC@VL|{SuAvDKAN!TI
zjsS$R_Xa>ajt7<!cwY4);Iah*X?T1=VbXi!xznaFpxQdBFC)>~qR#E6!p9(Ja0Tje
z7_e|;KjuhxaRi+LH<hwn9c5DaR7|h5lV|oVbCb#&vAAqE27PBhkHY=~!c&3xy`=ea
z&0Aoxd<+Gwrk?$PwJQ3Gskk$a8mW*Aw>WLaKMxio*lJDd&Gi5jvYLjq+nO*mrTCZz
zCiX_tSu+r#7F#UtsSebjzgGR0#T|*|Ox<ADn>F|Q6W<=L^4EUvPUH}p>=o=PZoMXZ
zRf`a<Ky{b%$xx+Ri8Hvn?@AJ-zEl6|e%1(_dOH6_Obz&k%Bx$##0QKRM&Z75iY-he
zJ0<1jg-G*6=W5Y7Q95T|@2UhAn2$BBJVuz#79~4Q$&x}W%%!lzD&bVW)JnlVzB}`h
zM&P#_knR!HeWYDe1^Z*1i(t006HCXiQzoYtzTF(Mx8AU8ilM$7iD*L;OaLkJz`N!3
zbGa_4{?i*usQW{cfc+VVBN%y#Cw<EMqP}9SCHSkQrO6$R&}=3hgC{{4bVu10&Vv8E
zw+6{@{@_I*7R)$bZLa^d&i9q}2y`^sEfwLBj$C!#@4tcPc?gh*9$#D>^uaUl-Dbaj
z!{sx79OWj#W%AQBGlebCgeGOV?~f!s0VG*_X7ixB&ht26#~-@t{M@L%;oL&k*?#6Z
zST~2i)|__55JGOS7N$PHeO%X@d++>*dENu_pZ`5F`DJ@2pW`=Q-{sf4Z^|#`*43Fj
z5%Gr-dtIs|Jj+B9z0uXW_Xl@wKn+flvB4Gj%f2cZ*K8;YEkt`XZ#t1@^F97lTREg+
z(W(D^Zxu}F$+B|DAD@I(CeR5&(;ci1HnSnvhRs}5SsA(A8$NbvGBKKXce$AjBoj7u
zys@1_6TyhOEzDjhl4s+<gi?+co8T=z1eSoxUDO5kVRuhLQ-5}2lb%!%``;O#cWQ#y
z6I?;I@n&@dTlnitEJ>uaeaMjn?&XCu?F`J{)?Q;p(0E`7?hH47FYvk&o+&FQ)0929
zDfg2=L*u;9(U8`-zq^|TYNlML3@z`LFStVw8HL-40I4Y30iar&5(ofG2hofBxz*#B
z;c?Nb=f0oVTCD*>^YVC^9UujW(0HO8&3dB}uSj)tb#HB;r(uUXDMiWHLG1D+BS%w#
zN_k+UeZ|`bL=rkm;sAXPI1fNb<1g`$vi@FoUX0afj>o`&$NDxF7>dpdOk3;sNx>ZD
zdF5e`K2KgL4){kYmwGO=)LA^7LXy_jagx;-vb0jK_rf9Zi7<&t93XK(C_IC(L1AUW
zsJ6tmPBcDu6s25JPUFE>PP7PZAqF(DXHBx_s6&$8=g0f)Uz1k71Av4#UTLs&Z!h`u
zBL01pn}NFsWcSvb17l%{a{kwWsv--hJyz|qw3ApppRS;6Q7fGT##eR*!g>E`>IQA*
z649(+Ut19|ZS$OzZ|(9N03l)fUwBLW5fEi%ut8PXi1=ZeTM0{w!)b*9RLX=gQ`tf^
zdjeG^e%>nls`e}(&aBNvF@{hbC`&iM_snk622MX)R(_|6R;G@ag_xWEJp9i5t;7_p
z*Xs|Jb%QhBK36|ZfBkK?>2n5VhW$!AkHs8H-JG1NBr0`=ou}C5(Hl8}ENqcHKRhV}
z@#|8no(GVEr{9(k6$SZ4<n90kaWWhR<oE9;Wk9ROWq7Mw0Z{ZrC2ZA5gZQ<lgNr>^
zD`TsT1~cq6-%tIjV9}`nP&SW%M3e3zD|9cyU|LCSSo#NhxB_ArOd$oE<Kpg%;rQNe
zwjj1O2VH!_G?=>iXodamJP^<-U$|3X4upDCJKT&=&jMd9i$-Ub@ePEziF^lmV}||x
z4JLD5Bq=-Y3LYDVByE|xShxxTE?)xER6!=cZ5BNhXdaK*QwZvFN<2lVZU{s?(&uNa
z6tU@G1TKzs=uS%3#+R;{E-VFzgdl-Xp^a$7v6k_yl|C3cyK0Mx!`7_W+EY~+nWAFW
zD~Vt7Dx98iE2X~~2sM!ymHgzR==JZ}=`|zmXfP9xR7W9NtQ!`&K^KOR&`me#vOYgB
z)q!uodb|4bsX#CUAO>L<7Z;uxY225*8`n{;N|hS!fMUS`w@;4g8L2qK0S@HSZHQkw
z4vVRE4=oVI^L2HzNb4tXtB}2~x|L*E0clQKTiXq*<^CwAlQZM}R?6@cH+2Roh`_Sl
z>oT8NHe_JQtDYm|V#GVw&kc>r1RNCxyi85-Kb+CqQT*OxQbNCz!2cwW&8m}O=8{p5
z#UIl-9tif=PYKFtYHC)2{aZN^t`}uJNc((I1nQA&l4(P8^WpmDrMD|}e6-+bz}-b~
z>u%I$t=R+~kjnwj(gntM6ZCq|COp`4z{sr1KuJPUbt{NFCkP@Q=|v%qs0dqaX%qV6
zG4uI0d*8MDl@LU?`Ijkhw*OqJ@vGEruKO_nC-@eN%=X_f%2&e?jEFzzI3?5{gl&*Q
zyeRl0g9pM*a`^|b$oQ9$gc0aGk2U`g{o|wTLbZ=)*UU<=w|6S@2LALyv>!NBOMh$}
zDKtUP!bJLVmnr<ZzCOlI%7Y*HEyM<Ce{ikkuH^4#oBTq<?hYe{P30@1huf0E_4S6v
z1{S-mo^Q&n?OqW)cS;MD`eK_8=Lst-4$RRg-R<u#T_5+=j@JX&x5rbSPz4m_8nk-|
zD7|Iah6U!UTnfC$6g5e@K*k`zK`8L<V*p><_*edsqR-1L9?l_i(%r};;dq1#p+{fc
z@BE?A=_t`!I1hospM&YBSj?%%TH;|agiAPi52WO5Nb!H{|F$4zc%Q7cUMw!MI=%Th
zxiWzZFQ;s7dSh=sPrC8|LWETJ^GHwI<;Ud~d-C^}dlTR=^?bg?uCsc!>bfgrj&Uo@
z0~ywW$M!k@0kG8p3u!8)3Du#ZepUd7cTqmy?c;;Qvjs93g*TNTGsJ4t1M;Qu2SR^e
zs}iL*6{SuhWsc#-vOr~>BbZ2_OZz;y#?kxn)ZJp_=6opof7FP!Z;oNp6sRdBxYd*2
zS+!kMLy{?joVJWISL?#VAsuM~<8*}SQZBi>`>a`$?fT%1=~yky3$oSJ>s>%F;rI(;
z`cj=GJGkfdPWeFB5|m1AB;{<dXHAoL&ICM<c-MRJ;kfCGJMoxd_e~jn%k3n8Dge>|
z><lspfCTQ;^dp%7|JIt<<2vwc*8ewxM&$EmU)6em_i;`98LQ}dqS_@Z$>K9f@Ucz&
z;l~U>Yz(?h5yt(&_CVO}H83D#0K?gTQ-i2KL^s4hLZ<?>l;hvf|9iIAMdhS%l^qu}
zeuE;B4-H}2f%|4sVSK*CGuYy0@^<}EEb*ntRv?9Fy_BiNiN}@iZUTLNjq1e6k2Kxx
z3;MdEjOQ*_R_g^X=~5ah%D@OE+1lUtbb)qI?beG`2P4$P(8IlfG+`-hrf@Pl6gGh<
z0Ism8M1I{yTfmw`qaqEovOzpit-vtQ0;Dt7EI6#%H18YrH(1V2^W8dlGuaz5J<hDG
z0Q=W4!eIyhqoK64VKlC!3e&q0lJ{~JK~tsq>&=?1LK8TjA`+t0NY?TQ4ubuF-veU}
zX@BO`UsncgLtv3Yp**<}hZbQE+h{ef51m}n@V8b}TI|ZeXZ!gOSLg$#!TtuNlymK~
z#|y|&iXZ+SqYfGLV;l&M+wxHqsa+B+vQDc*>~6-!mzl+&way~5R}t2V4dAuZ7<o?(
zoH~F^l`R#<*F=qG!AUp%CtfQq0TxI~D-L5>Yq2A4{1d7Vg*wct@w<xME6nOVm2)<G
zniXdy>vD}Wq%y-u9i0p*Sy<;{VgrU8tjczjGERYmuhOr*QW9G6XAdJ<iVObvId_eL
zP*Dm0GH3@BFihCIr4bdV)UM7@9fwV>Z%Ci@3}dMzDzau1X#;=;ClR~=2rU=0&IGUW
zi$rO{_T1UwDXOoJ2O14tj!I4-)$VcK&mx|BH&HP!_|n<g`TFb@p6B<BIC7Rqk${$@
z5e7$n?5n=?^i1pZ9L6L@A78%;ZKkPho~$$mps@P|_22oM>!Oe2U;EvsVQ3+UnC3>1
zNliLTU}^^GDNHs<3wY8_y8cSb8)IsZ0B-JSRXcG1!!A|VJ9Qbb@O}L`=6ORpPoGv-
z=jlf<FKi%;NHHY;+K;z6pFRP>-f`WBFU5H9yeu|i*;)#!5tuZnyI~S*yI1e?%w2z*
znOh@#UREUTsx2`|f`st~xIfR+$AQjyW5dAzaQ$l%#@6V_sw*KqQ$Y5AX%>Z}E=V-R
zRe?zmk}4=qP_6xIHRcN%FukHemq>#ltMI8uGoyw+1W9d-pjM6~oS-tjvVvU6IM_1K
zjw15wUt8@2b`wnQn+FnuGr~Rz>*-)O1?>+3f3zmZCx5i#a2+Gp8)P&@(FIW6du%!e
z{b6jCIo2Amn*2_PD9Qb3NV4R*I+!x|*D#vmq^>fk#2D=qu3eKguS*6c*pieBQ+gJ%
zyfV^T%p19+>i?GOO#hQ4$p_n?Ly>?a@_M5MnRI*faNnyP3Jsq`zlcT_=(8-&9PE7D
zh>i%?+6szdnnD7QXkkOnec<O`ox6|U)433t%bISTVQl_$ol!*$Z#CmE@d-y;nTtb%
z;r{$bwVxX0KHlbtL>QiB{{B+<?Qsph<FNDc$7++!c*nKQLiOgbTsfDDO!Mttyy8t^
zwPzHo$sh|*sy3c4X0_^nL4p8nse}4O?7d6}UH0dWx%=wsh9m&hmups=EmUZKc_+SR
zMF1GF1Mq)<ZGhGLeO&4J0iSJud&PH2hQ5?1>KoX2s>$~I&Z41`ENHwlLG33?f%HDa
zam2k~!va|%1$&k6h#OV^UzuHZ<85OQ1VX+9``tl%z}Myg@F1*Kb2LSYjz3qYvnM`g
z8KQVy$Xu8yy-xb&Q^_}N?w*M1=c|9|k%pQ`ZJ`(4zmFk#O(@K2WkcNZc5u{Z;a)2$
zywhtuKmCf>2Kq*GVDbb>cxl{E3j%soBRhWe{N<I^{z1m{V{VI_+2Fp)!uTXK+@wfd
zE>(-lQYI9KKi;~f$r0-{?&y`^FNb{J7gp9U^5?1%5q~aqgl@mDNLTjKr8Y-XXMfN?
zPI<?~=KdfeQCSmu6<PST)&pSwsX{)2uInDiZ1!x?*gm_TCcr_O#foG9c?SKwn@zN{
z4s;&pba<yo>?{2c>ruNy*4jE8$>bAD>NrYlk|au84vyBegu7BaBm4n=S~0o2r|0Ii
zZ!L{)&wfL^=5JC@=M%0Vc0loV1ekQYyV!VKeg}jvgbm0Kqktx4qu89!Ch`mfW)W!A
zHva6<I^1EHs;)O^07`2Wy~H_qGbI8UQV^gL(#X`-s5Y1Y`V*~o7e;H1s_mI<&v5$c
zj=1~{hyN-wb!%v=hPNlK=2mJ;NHVMq)Q*2pKP)g6dn`nv_RjS954L^I*iHrLq+vXO
zE78~a4in&~^QqK4+K>8Hc4FfEhQTC3V?@DLvXKap{dp#@-Dr>hj`R)U{8n~Q)N)!$
zg{-O2pR4v&?H#C?!s=Ku>agcRwiXnYK`rmi-G|`j*;3l{KMai=??P>NYk8-x3#;|N
ztMl#006NgC<!ZgZCXQLVtyTLOzeialqdefrhdP6T<TP`4HP}!|$*7Y(=j^FjJkKCB
zY}N)TlMO>b;J}gg7P;4ZyS?ADACCwe4k`xLofh`%gBVDHOdx)!*5JM2p)x}ul1MM>
zDvd~=$Y%8ko~#T-IY!mEjT+ik{V|XOB%5Yu7=s!j5X{6evJTJt$+yIx&a`6&zk}!Q
zj2etaW`MAav7$=7k4}2P4WmVggIGFw6KE<u!)(yOlU!0(?ey7C{Lur*IOO9fjAwzx
z)h1>D>N%dmO0HM|bzIzj(H3r=?Uh6!Gp(aagW>k^e0Ur94;UFTsh>srn0=vS;-+VI
z;*1JaXsi2@Rc2;q<29}DO#lD~`X+E!Rk0O9DW-0<!<~KJDp9tu@9VUDb%&6QmKz+=
z<6{O9+x`uaq;#RWw@(FDPgT<x!RpS&qrw_QLt1Gjh>8<iq@FiC0+Bk$$`Uq!HgA<G
z`tLnhMI7nB)kuh3xgIz+W*>E?<n^w1cZi+>16kH#F|6=`H@rXADBo(pW@&3KMDv;f
z#0nUnAZ!GzX*~t?J^{^Qp~}dr$dYS}ONB=Hvn`62AHakGY~hl#hygP}IBT#&YeXa0
zBX~LRTQWA^1Y`X6_`;Q0loGy$WctX)g*Gb*`dHB#8*5~c#|`#yfUm~0h8WcPxBw?M
z8cm=@@|8tY5y!D{Gy8DVA}`S>cTd-8ut1edW_Wy!`!4VyUZ;l6V3L4BVE;E3fq)0a
zhSAH*tDw~EU!H{4sb?aUyi;7VB*+S`#%jkwp>}gicTns?tw&#8-%BXbjLe;BOesbw
zB+kInzg_L-X?NT%KrnPvryJSYrKWuB`1m*4`;I&zgy;Qk_Ft(zhqpB?l|tKD4Qdqa
z55_fz*-GHNj?|(|@Sqouz2!xCG-jo}7=)J1+n>DbPW!u6wc$a@wU{NzDep+7R>r;=
zZP}L0cug6!5y0yB?(=G-X}UZ!BPu5sj_>nRvRdKzyj*ExW30%Tx!!notf1}tzp(?2
z88$vy?S}tQ`=bc_27_C<G<x2iZykl<jR0V;mE)xAB;G)d$2o81Bc4Vw5A{Es%-q1j
z3Q}bfr(m==EeW-kQ6q+d0^j*yX}UNXs!E<Dqg<uoeIFA{UV0wPSn=9;j+-EgK2zh{
z$4Zy)>dZR+3i`TOg0L2BnbpK($9u89o@a@o7pTG|KhN2nU3Mq=rT%F-GZB=_y=*Gr
ziv4JScGSC;IJd9F>evbAUE75NhyjQBaGZ2jmou#$@qI|VL@Q!IBxrVKAsK<wdoo{s
zd~{?qqQX4H^THCvcS6I*$EVd`p*)%6dBUuSMostG)oh$=_^HwG$jXReA{k+K^Peb8
z5`(rsYj6-DGs|nPNQTGF(;HmK8l?rc4W%0WVkcGQvcS=Ngjq+ZvE^8}W#!`D*f0%-
zE)MSx4%^h2o8>9*B%b5lF~bkCN*l(8rbfG)k>VLa3mS5%sA8>gp2rTy=gaJO0y}T9
zT%M99;?rzxC`6p$xQkzcKE05hTkn_VPgmO+t{d(dJogYyHk*TggziWIcqAOd9~_y7
zWHao>D6w%tj%?w_ZZKlU?zye2SMeu<u%RupW0rSXBE9C!^705!IG#WDH=#*~os!r=
z0;O~+YCO-@Zt<FIDz>ZmFaHBdZx_bD-?#z;Gn8^Tyq@CT8~j9z*6-?~iS?9}A);*s
zN*O+bB(Yp9V%CjI9%v%oEKv{uJ-oB|`S@A3AJ+&<c?%W@U_H~FOlzo+17TgVxDXTr
zV0121`UZ=vk+EO}IZG;{bcq3R3!m_)DLPPcWrXm27^)Y)h=C)_OY4K#!<L9yO8xM?
zdx}?m-OKhDR$NSApzs3(j4m2!j69dbJL#_qWhjD7TwE#C0cB+oViB{f<vEe};|Qeo
z9dLvw-!U2zxH=lBy7e<qFbGnLRykMAdEUdvA?>yoDh64jzROlagWOjQ*F^*KXYV9t
zn0>2OuRZ^0FuOij>Y7hb^iH-F{p^A`hc(zC$1-Z3AQ<7JV0nB?4!kwjPunliAwHK)
zL31~A6?g`v`Px)+tc?$=ToHDdfM3A*uT17fBJJOV1rMD_ea)C`b3bo-WIYEds`wJd
zSurp$^j`L-{w*zto46nFLhDd-j)d9@`f-nE5AD&$KfU*l9a-P`<Mdy8zp^NPyM_Q{
z$g~8`<M=xJZT`!`O4CLDRaWL}jquDbDXS8n-{ffB0bu6~Eace<A^(E7OopONPwN%g
zcXN&_)0rtQ#T*}X)&s$t*?Ve@whnEQLMT}jV3zCfoe(*|SIwl5(P`pi;>5QUN(E?N
zy(pD^=V?$wEJ~Y4;Cn(SlxaB}st+9K$V}j;ar^0_)`&x|+>HG`Uk5_7E0@qur-M_N
z5%0@KIV(JqrE<5wpf}>r?aS#R_0uwqa*6ae7HUFvvk2YjsYc8ddjhz12a)A{m%1dt
z?ZwMP43OD?-{R8Hh2J{RHW@M`PIeS9o~+J1ZG~7!;dN;SX{K;blPOvEVBX!XZc3b3
z%kFmdx)rj*SRoEF{i{oH$kC`@E>NkoyZa}_t~e>I&berCE1<}YB^?`gh)0_-<Y}BS
zN=~shdWtABu9Qq86PMBgr*ej3o?$QfUnQh4xxA`^z`dGAl^(9mslT(aMNfdKB8aMJ
zWoXU0k}CaJoz|Z{;9Yfq0P|0y^i><g0p1Moo|pSmaxrFtSzGU0e8Kxt#Q_Rs5L?k&
z^K)yR1Ru?&S^7&y9>e`M+Cdx$IFlZWiP|A1$(SGCC5PQE_<u8O6!9!n!k*GKd~W<o
zfhJ=^&(NXQ%LjA<c8lxXPOk+sCH{OzIRSvH*gzMpQBRHmtFc;bsnyY3#(qH-FZk2|
z3M_z+K^!EZ&kME6`8f-&aSm)6OStap-R`3A8TpmqBH8cW7Wk;jr{xhW2eV2!jsg{B
z1zp)&vM;CU4H>B3L?I(*t{uCC<C*;2S^0Lh)H0$#(`~&_VOf73h-LA6l(zV<$Sp8^
zop4#f6?~um&|CX`>Z4rEUF)B+B~ORxwBbZ2;bRa1jcG`bf)nu>%EAVsQ|Olp-2@h1
zN>;ALC>^-tW!@c4tnZ8uXh!Fp<DJ*7L0itk4R_!TpTGzc@ig@w_T3C^+|OX=z(P4E
zZ^+8IpMkS+@BlHr<mn;LW8aFX<<k30lPXcj6l)qvhpT&(eCIQz#$T_yq+tVfED`|n
z-^ltBBn&;R+6Q$;uSUUqfoXiM9wM3F7NXRY$%ZlznCVPUuUWMRK*3-}Z%@||IIR8v
z&BW$(q_pUQBbN)Y>Uxo5F2&sWda*Sk^m9PA_lfltl;v@_ABV3DJUpkmc##!Vbvjj~
z7}C~bb+zUFVbz3Rt5U<{v#yvy*^n*|w{mHn{jxIXmP(cf%A}F3mMTB4U@0#rPPSr!
z#ujx+xw6e06=fkWFz4VkoNTJ}#e4SYQp;rLGFJkjk^)vcA9OJ0<$*R9O%)?cds9V`
zo>Zs1IZ4*CPZI-DR5k&H*<Ar0eW|2Q9Ynk3ISiFe$Ds8`4_2s%PGn6mM9Ik-I~b$?
z&Fex%u|}fB`P@OMzb>F|AiI>39Vv+wt1Xg=eT=6uZ{|Gixn>Y}<tPA*RGWYZ=-MMe
z<%@&^2n3bIZfs{R8$Ge4`{brvnFJAR=U~Lj+;F~7HZ^*Gv-<4g+2VSq<KtT7&Xclo
zHvnp#)~UZg9zu07J4VVe9kXzi=Ns&nXx&)IQL$!u*r+KP)?h+_F;$)$X2)6K4QtNu
zB<mAphyICaLLhRU@*F<msQI*p1xr=#YY&S=8P{|Hl*R39Y!>$&5d#0L-CzHw1?YZ~
zw?Oo9d(r-B0GT-*d(q#^en9d)Y?@Gd7M99>L{<BD9Mqm^o90im7iwhFiP0gR%Z}SG
z?@fO&V!p?KnYYW3z?VO>A4xC_Ak$|zO^yguNMT>Kf5!xoMEUi*FSmHJvlkIaY7W6@
zilT7u(xkC8m~6KA$WpbREoNiHDpKld>z_s~mmyW4XjPNg{HvDT(kqSpom*2Y@QV^=
zonxTkKzL!9lc=U!mG|BFv|j5R77HuuwVx~41<`yjMEOj}o2--nmB!(m@SD-8>eRG0
zeqUF<q(`|FE+a8?Fe{7cjGD{0Zgf?xnYr_RIh$Q%-Rwe*W1~&w{P$ATw^PmPy^{P4
ziTCceSxvX0)lJX^RDlY}OneqFUAp2Q2sEEV-Jh$jt*;mDty})^JdYW-ZcfV%3DPVp
zaS050bHrj=PvmSK9*l{<P_|e2#rV#PK-82FwkziRmOi+g3Gq*srAs2q9fmqT2S^>w
zK$Suww)5M0oHlyDV7Lf8A%B3kscG_Hg#Mo0(>}?U*xa={q1KKD$F}I5pu)V;G(6rq
zDWfhs1jBvM`LJS*%jyXM)~ng|fGHV8FapT=jxR3{eZi5o_!qs>YhR-u=qQD!HKfYd
z3_8<tp=oMRs^o3v=i`WI-2BWpxkv3k4qvg0GvaM*9Bab9hbIJ<G5>%ptu}6a3~Ua)
zC&4a)Eucwd@mx|`>dGtvP56tEcCIKid8M#qo(;#2gd;UY+(c5xM<?H_<MPuhqpHkk
zZJ4Q;>vt$npp?MWHl)f<3%IDc{_~5VhQLVk0LGqXf@YAy`Xyo`2DVdE3oB)+;vZ`r
z5-@&skeqD)Ey$^3Ep_CiQK)?f|92*`oP<dtTz2X-3RcLD;*S&?f5JC1#sk%?j7S6L
z$~jHcAPX8)9$gaTwBXA_FYK4zEm9KxFnT}^g&IuW9xWUev87$!wr2Ep{J&+(SR#^)
zjXABvDma&zzLPS@#fpjiB~w4M^7dIyu)of9jD}4<P0fBm$RBXq3{bp|AG7m5Z3XPQ
z)G_3jmxq_C7LzGg5wjDqH{10)Z|-_sdJr!(B!H+$cRjve4qa-WQ~fEb*=U|a988q0
z(fYm(-<6Jqb-6@ZbpRz9GmKHLFO)!I{x6BMBQXmVhF<RM8#k$vsCzgUbd-7=oh)Ar
zWl411x8gW8)V}9>S&Y86|Fr9~Vuv}vcx~Npb>CmsLwI6oq0pnJR+uU{zh?@j%BVgJ
z%?L{`^5Dli^j+KTBP&w;)VoE{*Y9sVlS}^F8!2q8^A33q&MQOXPoH=vtn(E_e!jeh
zS=T4-``>$8BmzDk*~%sEITV=!EriTYlni!~v6z%^A_sn}DkU!Jyt-)(ZOQIv7$3>X
zAt>s94B#f9C9uA#H92@CIy8%X;LyQAdtFx_JqucawG^T%7=x6<jV5?_yQ%Smc}mjs
zuC#G4)Il&BNd~b8$O21LRwhdg%g=YZYlnbNXK$GIkLnQu>AEZagtr=5Mq-{zLe0@g
zG<aEHpliqNIf3`nHbxM*Za~m~20gYH@0fsk&#1pB64ePNjNjGJ%7@5n0+K<6E@Be*
zyX1uzDdJp(^`An_lsOdykdTwP+65I(P7AzAK^t1_g{{|5XMu7Jeu+vH+BTFd<%<gb
zh<_Ew8$H<FcD#Tt1K2q7^7^{YZieA*EIkVs9P#;aX=ry=3r}8NWqN*a;c@-rmYLqG
zGSk5Z0sWJyJRtBM%Ml!e=@o8SgH)7Xzsea~_yPdf>;Tjopg9!9?YsbSk(0CYzvboQ
z6&~wh2YDZ#%&rZ>GdFK~-6lc55eWBe{5N8s9oW0$b6{UcV_8Wng^_xmh20kt@U$CU
z@A>S|u;BSJn;j+Wo&PVSAWW0J1673}W-O)0GWC~ME(=Tq#$HhGZ~_<#8)-A14P%f{
zHdM{VlqUY{L#nK5-Sh{H9}-EtQ-mk!TS0HRxoMnl!d?~u+CpVQ9t+eqI_wHx$jl1X
z8U9ONO{O^QxSa1j;(cHINzRS~oOKW@mI)%A@?L8%O^^(RB)wvs5)7S|^$%XOp<l_v
zZr!?JdNCzP5v(2id(E%+t|SteCh$e_0x;HS4SuqG5F(I%J0bm>!V6A&k))aAG!A|}
zlqz(`D4K@VF*J#J@UUDGN~BZ5VRYn=xTi@TH?nzr(?JC~=Q+cWh>39l<+NrxZ{m!U
zLGmaXv08GX!LuYZ475-|s8!t_i5X(hCM*WP&T*7JZINQEo<>W_1DG_5nb>QWhkNq!
znM528NYR4=O9elw+E*j_{s0XkFFNqUx&+~y%XUqnzM1+d3K=vi`LsPBTcGU5fF_As
zJ{knpns|_n68h>ctlk1nF^yFBX*KWnazm0hh>HC)_NhH~Y{S1d0A2oI{HM`dwd$0^
zNRQF<Y`9ma>xS89ikkU_hOx19^$+Lz7_Y?wMh}?__}p7*?uas%bGoY7G)FAhC0h$&
z@r<mDOde!?q)~XV(NM!GtElv3Sc)N{363jXA3>R^GLP4b69^y)6Z*0#M18xFPC#QY
z?8<n)JtU_Du<?Lca;YyF6fi^vpj-j&#u|V_>P8guj&08>wu3EF%%1Fey%+-yGTBwP
zz~|d&IvW#+u_S@<4)<FWFz?UTP5ZZN+q+X%996rQ<IC;4*-ZJplX?2*LGG@}=fiA6
z2SiY7=T%eeKIf-$jmn)cdZyy`c7Cd(#>rZZORNa^)%EzW8%=g+q3OJCOvW1tUd&?{
zhL7d|--QiaJT)?1ESDCpf}Z(r;yWvr+Ro2mAd=cG4{4`QhjLXAB(<0|EDdaNBfI9)
zo10T}@8-EbWi{K7RUfk@eq@^byBPhHj&I`{<>MER(8cbV6q<Vu7kagJIa}%lNZ#;-
zxOVRw>#y(T%O8~sG$qcbkJAP|Cyj9GMCFNd+P`Ya)snmrT%MUO)+~_eDadHF8>XBK
zu8Xlkfqv|+v4;Z`@|~vb;VSl>mb7=T(j;Deu{k8eF0g-4A@Co%cGHf%!Z#-ZQ8*&)
z*0ARxCJ=DdB1Ln+{qqf2rA8oVj)c<>IM)G?;$Ib_EbJTYr-11T2H<#^Jdq;W*&>gI
zKr9C|g}r&`(`=EAU0}-voF)NwksNV)>a=lqLEm=|gIFhP52u%7DAt@-6`zuQK8l;1
z$w)NSPc(8(WvO(Xvl^u>Vscc?tu5jpXi4G|TNu`az${{7M&N?V{&qX6J97mwpQb_F
zgn`$MvYW_<<KRnygKM-=`=Xjss^_5h%=I3AXAC7S59*=_RSfyd>(XE~e(p#3+|2+b
zQrWvi87Q+F?xK6jhnxNq>MJsmpvL=J^8S!6oJU9Cy`+Z>BEzBJ1z>n9fC6c(w;p+C
zTFe}r@wGRPMjBXe#NM#*p9FfY9+>&83I4G`J0BKz_2TKA73V_5X%tB~CRNv<w}wuG
zvv8=AQH9GgceOa{Ldz899$$W{oZx+Gw}iEP{c;3cYPxM#YNPd{N~K2rqMOnapmhVZ
zccqJ`8NdP$3Yi@YVN3~jSI%0#_KZYAUsit#XXpN~yc~~ntBOVdJni?2dXV8`HpaT(
z$!PC-AS@73@*UP$gbVGNFKD|dH<Z98+(Yb-siO0G=Lad#-~-O`pbbvz{~&9n+wbXU
zG~55ug4T`uVZ~}E9NL7{#r(D5N>}E^>?zIRf+sOlQiP;b$4!AF(M-7ud=weL#Lx%L
zfLrX%)J11R89RH4aZ6H}vEayq(6x>nj7v|N8J_2?#ej;F9iY@G6xYR1l)>r{*vcB~
zf~1$V7-{T(28_oXcud>+_hG?b4TYnjMlB@%FdzqIx16D<W3c8I3dd$1#1R7qX+Q(|
zZL0OrKy&lr@YBn@Vf_0H#e&9|qo&WKzK5n1jZ9#DcS6*^)O}pKQWl)*KP;6l0`?5N
z^aw#C%Aw+?84c|F`sQ8m$JCghPQMQ1&4}%5z}3EAfMSgO9YH@b=nkV*EwbD2_>3Ta
zKa#q;yHmn|Ob1XD8i!w>BQ^xCYxXpS851R&kckb#v+Kbu1$=}`S6+`(Sv7g`rIafh
z-E2l<R^bQ>l?#o;^9=`#isgDP%cdXh=s2L!wi|uB>CJy#=%yTh{*dW#uHe8Tg;yy=
z;oD9f%7WaI%YH`7e!nTEJ&{+-0^x$oD+{gv-g#Cx_Cr^z$sF+>$}l~q9rtH*$=js%
zs9RwjNo??Vi%bO#T;ROXnlrQKyR<PUS~xA+Bd?pB2sD>5T(L_{5}9?3^zbc190dea
zDMg=ES!s2%X~unS&B|fy`Uh-zDaEfd(u%>8wc00Vg))L;3I>l_CCY^y!mD#5sgyE}
zbIe>7HX9ZS9P;778?T>Om@(5Tzo8TDLb|#HxqdGBHYUl$Q#GOr-+YViMy%d$&Si!<
zxif0SZ%RINUmLo47?s^Ny2O7;y>bUi^FxS_)fo`Zg%fbihgd9L7JC>M7H5PGM~*T&
zUnFNm$V6(cFDjAH6CP7ehyU-p(2-G9>$V4~mfK^yq9Ytdsf87|$}|m!2Bkq0o|*(z
zn@ymzwaqZ2#crsKrKmDaF#;M|!PW*e(H?a38@1t6ALYfL+(>1{t3e2Q#(5o|$28}k
ztIPf4VrKleAu+D#{!*Hi2Q|Eov=Njtf0=k&1+X1@UTb1C$Qm&HCP^&GpWkD1E2$Sm
z7?$t3iQid1?eASv`E8M)#+<qJ#)g>NP~b7zwoHh-eJ>5}2@fXI1Gv)gZQqE1=VntJ
zJM3%4-B%R?oIzZV;R|mdfTtAMabA7Xbfbe;%er;D?X>l1H`{JEK5uK6xIL!4>y>ys
zogX=$llwC}o#(S!aBu0Z3u*7k{#9aUci*rrLLwaNI;uH)6v#Nt%d8eRY7C_SoN$}g
z8m{cd-Pa01`&5#5j}N4%Zy&D6j!)Zvh8!`x>HA#`kx31XvT!J_I)91HC`dMn=<g-w
z;}$f+an;kHVT>|PZI4iA{TOdi24S_0RfC{|OcYfC{9P9Rc~M|)GkinG%s##%XwOhH
zsz#Gj(lYdvE41r>tS;24bl6YSWc%?K=KmBih}>&E1+?h8>J$(*zZmu(tA8W>)n37B
zKh#^k60z(ex!MW>CHHFWacIvWm_~W}Z;%nS0Or&oI(KBjqegI-9-dk5;rOnPF_)~o
zWjt#a@AFv#`}*?pfl=WtdCJ(d&$YTc6<)|5b~tyB!t>5nXlSw9k<l>Q14EM1^D#L6
z>55uVOt~z`c{h^#OGE{1`S3*`iSbF4y!oeSFuK@v>jkKMByd1CJGUj7-kzkCs2;b>
z#WyVC4RsfZ>0;xs5dW*6D$VmW_v)Yj>;;tn%1v8-09IHKbgS&%uXrLp9!lw+i3FFx
zSaM{xW@`%1M<ad1H;Pq~=SxA#h^OXMZ^I;^%#v4@vw{AYo0mTK^W!E_$3rH;Y`|uW
zqCwsh$|F$#J0NGl=nOAOV}cKKhAu)nMk5d0qwBvhsA-G;=I98e4X*@B>41epL<3!<
zY!(jZAh^(;JmFB4@M?L&vUOQhP?5F4rK;vEUIz^&4X_Ii@x=;1#*!GBc*)lg%EyPX
z;cScJeVw>0k?tHmF$d3+O<X;UzRRE$#)Q_BC#c*cHD`_o(=YP(Oc3`2PnJC(KD>Eo
z?#|@y*hWN}9>M2eJAJGcc9#p$Gbhn?PKS%nR~bS6)rOl2)4+v*&pMZFjy(D-I3&)D
z_SOVHM1!$nm|@p}6<CyC?=MHeb##Ikw!kaa(Ofw&Z&>P#;65Uked;f(vkjQ#cDsE(
z;QvcAlBeJ1@_6|9)q;Itlg-&1vNbee7JiydkO?X?Ke<fEDu)TtJBJa~tGgV*+XKnC
z8RgJpbKoDnG_%LcOO<}18E8Ec8mg8J>z~)gkc_86$`)x^AD>pXkEFn!DdQ$0lHzKW
zs>S3%{zfDqfV!&u#^Y6J%mK4awJB*G!L_k4Md*`diNyP))I*7)S)nCTsIu$9y*BKz
zaZ6NzdhHSFAanUnP0-WPDmC!i!B-Our{H^u#+thZ+rGJwGbc!amA06|lI=IKV3$}^
zNMJ=JwXK;wiXjH%pm1SaPIw=vv4ib(BT+<h7>Gq{n*|Aas5HZUUUqIHQ%_tfl$tyQ
z2w-c2wkona@le?B-4v2;)G(b%kon@pR&BcpX(=xSZ`U#MzVwI(H)G7yQ;xv8Em>fl
z*XIFxURX&iH@UC@^9gg$nhm4e&0i0b=JG~^7cF^+CKq$Lr=dU;t}Nu@NxGO&Vba<l
z(sFKHkdA5ud@Jd_$hCQJ4Ed9LK6Q2S*gky(Mrrx55pa<l|8Yo6<wABsQBgdp0WRfQ
z0YqW#N)7TU$CLZt={A?;xMWT$+Zk}p-~Ylg;LxMcNv!zyEiL)rfUk(MFOm?_=59EN
z|LNmNv?aI33}uQSl_|#<RKhWTr7RSs!gR1fr7P4p4+(_T#TYISCBJ<bl5W6Mw!v?K
zp{tzPw8Lsxfh&MUxXlxWxx8g^lZjzzl&FAIH*?X|f&5cxvpj4iB=Wr+vw;hTex$DG
zQVesU&4ugEwJO{4dGhn>PxDSfb1<O#YGafUN#cC(n29^Ym5B8is7vQ`c5zX{;MKua
zqeF_o@<$u#Cv09#ity1r{Us<x!Rp;gBBQ1!wM3i@vaa9pR8LKYZ$IsA$6_F4Y;rP6
zebhrq8(F@qBeQiqp}YKWrh%01h@&+ehOkRVnfGlA;{*}ZK>y@`Cbwv2SC?ovKD#Xh
z5c2bvJ6YED)$NZk{<;90<163e3wIYOhH>VM7lk2x8qgOyl&8)PRQtsbhWbcOpN7Lg
z{1h&9XhQsoWB9zH`3wa72+LLYP`-W0CZ~=!RxjqFbvZqM?%dEbIqrQ^3vEP3{^lf9
zob}sktvIK~s}L1k?4o4Mr$y1E+ji$8P(x??P^G$LB37AFnQGQ+r!U&^YeH@X$h_a3
zlW+M^2NTqtuy(iWirzb9f=|834#$+q(~FBk|D=c~(gDl%)2kd)YalrCx9NFC#(8}v
z+rwRA*VU(r&B3p~kg;`GRg&qWW@-!Rrm@*gF^Sr?vGu?QiDu9p6hjsO_V54)CbTb5
z@c{3J(*_`8yve-u{PSg|2KJl*#(D0q9l#ISJgijALCvh)*Ms=2Pdnj&f%Y39Ljl&6
zR&{^;{<<talJdscYMVZ<Xo=$w+JRHZWc&zF(vX5}6q^Uvnbs>cQ$)?<5nibE-?u*Y
z<E%q&ZhfezL|%8OMGD!X#tw@QZ~-xsW#1N4ZA~~u;Z7pG`K$<u?jFSVx14e)u=<mV
zvp}{g3e5-KMt~&*^1!}{MLkzeV8O9D&@~gVSmx#PzP7n)G(?qr5A$Zs-)|)mNF-UR
zoB6HP$HUA8%9t*?C@7~CW0FCbVKKo>u&?za;HN_HzQ>50G7s1fVf;*QJUTMS(_wqU
z#)R&VaEZ6wKR(0BOdk}TsYbvqCIZ))A7GFdfDY&X-T3U+H|&zdQwFV?uYDD{k$3lx
zKA(@@Gk9Vt!?7D+-62SV_L!$WX^2~|cHhn^lV^`EO+NZ0GMr{$N#OI=Gd1Ci$BuoV
zt==y_{<v*@{5dk{Ok#bF%3uC@1E{whCO8+b#vP~pFxOtPKL<>c4L9OQ@$hj8g}Qc&
zzt#QZ)#VbH%<OLAuX@+N^h5@3<z%$fXTpH(H@_h+p<aQ<wSIlqk2nCF4oLj+VF7mM
ztWBXem~O8l<EqvdKBP=99`~%<?B=(a{jucwmw$=lX#yTo2Hl#pz^d4*WpSEOyKr_q
zPf~{?0|&!y1rwAxya;{KX<6U4M?B~<3~9I_I#eiHkx`C?BDy-T<9ZCgYu)N*wB-BW
zot1-%N*4*#5bjw+pjX=FnV9N2gDEgtU;1=r&u#ejIYPW)Cnjcka8%HU`h^)_P8NJU
zn>CZij(fWQEkM;I3T6S6MVi8|{b|yG=Npg)>3tVZj3z%5t0|#d`~V{ZxuX(7rTwfT
z0G&qXx<<F$W<U^FWfD&^#V8V~<m;0UEGSq$dK^la<}84>*=cC@!uN|P#RzJ`2r^!4
zvH!A4V=QQ+2V0{}4*5gEgJ_yC5SZ{N_M)9`1MXcNxk`=j_+nk#&M(!;m!hb~2sAe<
z5yhX7-=%>PO;PZ9FRn2ts*K*6kvh`}4Pc-sWcnsE`Aap*ZQz3C)43$Jz8q;q%Re1X
zE1L&*J<qNgF;?=>%<tz7ZIy5$m7*#IO;^-3uXwedp1(D;3l<h!(8CPrG@`P=51sUo
z&$@I?@$4c*u>OCSks~0CwO_LzA}uVau4eXEcz=Jt-hd~uQ{!CG`0+6$^kKh-Y=SUr
z{{661ZYuQgM14^&lC_FR9)!u+Am$KKl$VogX4&&ZK-=gii$T!Ry7-VKI(z0n-a1J`
zA7V*<mSO3Mch`43UqJ7D6Xn_wz#(I~kd9D_nA_c*{g`!c0CMs4kI*Q$egzjZfQp-f
zj1)DFU6Fsikv>)R0?wq?AXHTb%>ayh39<*yCtW}3@&d=}UGjF+vuq4Dwa(P4q#rwL
zeL5T1P@JZmFUbmYoxoZ*k>Uk;gvN%^HS)u{-B!Y%9>3`MJzQdr{dERt_0y>GEz@X>
zEwSEx?IGRJhkr&<J4%MnA7*((Dd!V-+%?t{-?7i537>?-PjZ18?feCIx46x>Q1t%?
zGc(72XA|U%&ut=L=e!i_2yB=iL-rBZ!1(FCBZKwgGEHWQg>-d%Y-<z<M(Vx`OHCS=
zkT<RHe4wtR0uv?ZLb`s+rDP#*49O<XiWPyvs2(U=>VzO-J@HpY)nc!TImvQ>7L@@b
zNOBar+m@}?BPfn061;*{x|)v3^L#qbLVH^lp;0HOLz_C*Tb7(asld|*{pDJmk5fAj
zLzahAJL|A=6R-HkX@xpXn15FmAqB89S;)PfNZnfF@MB)T+N_hgY#*YED(&y|(WS%=
z+MdQ?N?)X5L5zZ>IUE`-O`Rmf^%^XSa%<D;&?@z{T=ZGa{g%<$#C9Bt;J1|U>Iszd
z;Ry0<%S`**38B4_A+FnPctHrBaP!AjIijGwe2mJ3#M(&%eHcw>Ze`g1VCQQ;<c@5>
z{{g%z;j>dbgeGu+c4e?an;k^V&@ful0Ww^#+4k22Bhq`ck0<|^;pMnKN!7o+bFax~
z05iEBFbF_kQ9@#^lfg<}Ndw~cg+#+oJYcJ%(ar}tNb}`>TGtmim)vzX(6TN74UZuh
zdxVy6FD}^AH1s5Xe1R@aIy#zxO}v_{X?qK{J;W)ZKpnW1zWGGu#qZkoGC@dqA*CZ#
zK4|N6x_b9>GfE|F(l}}f=YZ#N)5nKEF~M2{$~DXm@WM^l*3H`HFZ|o)GIq)!fA9W$
zuxYxqTP>Nl(%Sqaba=}B*5)9w%4(V4`c7TR+uQqR(rkSYv2xG-Ntj0hPS)-maWNiy
z63sv+sCKeQAcOqTYBz%xq_ny?WHcW1TE}t++5|b6Fvbvf2|-&?NvTGo+B5-gN{Q3I
z+_oi6(fhp5?NW!KdDEL89-B#A1{3oFYc;7GX7dx+gE<M<C2*1xv*WKmE<Y|VK~ngf
z>2@FSk$bqio)I6G&542aVGYs@Fh*Lu(M7Vmb2n{<<uQey0YZQLlbMhy3aJAvA{zE^
z*c{ALD&tyiBB*d%r^M1~Vmph9G(af|Y@RYbRDCC;#br$n9i2cwAF4moM;%W|2t~Pm
z3b}F=g-#%dm8o|*?+eW~J5sC2^0l?~{AbhD-TihPt;e{ZcPv~;la9r~J{F@h+U@h6
zz0_KdottAVo!o6sXY+2jMI~v~5u-)%XFPj~naB;q?ly<TwC$|iFn_bBEYUiJ<&>q4
zgE5jp$9_cU4Lo@#!E*EUU-XWH%~lq0n(hd9*UlDQXIaj=J+>YyyR1z3YTNRJNpbxW
z-@-Vb+e%+FSbfM`loQumT%`{-ZNG$^IIVD9<Nx1+BX+cqqmP4v8%N;xAR~c}P6lk1
zx}v6-jHjR+nm-5BGPU`(T<NUWz?LkwX$*$b9_N)sLmvUTWTr?X32BNiG&mak)z24e
z8VV*bu%O&$|HKqByw`{vr%Si$QLU*hUc~`uL_lgjKVBUNFf~E0$_?McO#X+xvkHnU
z?7B2A!7XTTcXxMpC%AiXcPF?8cXto&5Zv8@ySvZHRQ)&KH*-H1Q`J;;cXiQ+)6M(t
zv-euhiseg@E)s57u>mIsv-tWzhW84%5w*H~`J|};e%@Q)SyMoSs!b%g3N>A*n1P<A
z5L=aEJX>Vn0{9Gffp$d*oOJwtI49MH4}`jb-S?g5nr?aJvgdSCyFT2>i9)GF5Dplx
zld=-cIwP@4isGbvy_yDqkRObE*8TZ${T<(VTFBYWt-<5g>|bSN)r1or+K}rBjK_DF
z6tcLJfmL+Sy2W(tYx)E^V7U;c8U(!IX#m~?P*_Bq)-NU$Q$>ph0m@R4Nk9I_kpac2
zs`mB_m(#`T_e;NSgt&D;QaOZWHkmn+$>p@_p9|tD?tcV1wKEpm;~YH|qv5Uzps&ct
zU49QKUroAo|NQ6CApE<zwLW}JkqshT>vM?M$TrW-b98>`bTTSBqjBj*O@|~MVoOza
zbbM;NGW_KAHiT;*MYs)Yo;$Y~zl>3Zs!ozwd|T?x(2@~dNxXDKkr$hP3G_kj)5XCs
zXBKJdFM>s>rPq>N?oQ@x!Tx^TuU!da^0+Zczb6+{SCpczW6!-gYwS2qz`YmWSb#+T
zF<|HvN7_Yi5LugGl|Jwxcu3bigZYI{f7vb<-UI=nbu>@U@OpvyH{MH#pfZe4B2#D(
zA~yi<1eOCp@x7k*%>6%FfT(}G2H>s!wZz6DfopQZ^&<!8!f(!qWlA-V`Kloo3)UM<
zg*aMFFl3Br<P^q|7&m?XwecXRe^3BbT_LbZPEJlPP5Mqhe$pv!A<a~gbio*0_d!;=
z7mu{5#ojCr$*sFju(3L@7ne=C?_3Y}S^?NQq5SWjE;ohX%fN?W-KODiuciSCpkS~u
zdQO+BM5KE6MxdGP&bqL46{XPhADK*V_C_?Ve>wjZCI7}5e}B4kJFuJ4`T`g{pRk6$
z0++JGSp7hg*<&|BY|kU6Cl{_{+C_@ny=*@D;Ze;Char=s#sJO0hnWSLzJ7f}K)LAJ
zbJ2>e8j$Ep-oV_GVA5*##1Y9G>dITt>vdpR(*ZU{+k4{Imlt9tO8;>~Jau2BssYK`
z1BBh2kB<+@lsM&tL#_!kmYC=r!xyLhGXj^N@TU4t<7qcPC|D(fn6SKjN<pO*g~oR1
z=e4EXJx>$YJ;&%|@!SY=I-gVs(~pQLq`0T$B2Xh*;7g*XA@W?S*Cfl=8=9xC;>Xh+
zkb^HVQ$&m5za`u}Pw52;a)-gAFS-x%RXvN8!dX;qA5*kA9ub8Nx(W(HTr`N|Tcc_u
zP$we(8n3x<7)SxSZA3oZC90^0>~0$_m>|~eO8{cSn~eVRJgD3Fa54)h!lgEM)Rg6{
z<9mVE@qp9m6H)0F^hD8P<Tz-Q-M5X=zg@Ej6H8^<wkKVGml1%*W46IcIv*{Hf|aZ^
zJqky;ZkWCRcXoDeoDdo_eiS5x-}=UG^NWNZ@VpJ$iWp<Fdk2H3jA|PZjW`SCqt>q%
z!O&QStb=R*6{KJ4s?ePZS4g*}6{ZUQkpxBK{cVi>N_u*nh1wF*xC+|g@q;qfE*HY6
zYv59kK;qHQayfyyB1x-R=sm~{jkqLVdGj?@ru9bTbBJUtmC0*}2txjcf&#<mHGzf`
zz&<jltZ33Gn@6S$R{WhlU)5YzQxhZT;lB(zxs9D5JUgDQS}|#9_<WNL*oQ=T0e|k>
z+o4;%)1f@TMWXok*rE!cAnk(_Jb_#aT)zAIbTzj^H|WU-(rliB6MWhBr;8qP@2uq^
znO)=KR<=1c*9!>g6%t)whfN^wPsmf-_T7ySr1lDviBj2V4A+wSc@5GWVFZ(<+fVuC
zhRqId0%=Hu!Re^k^fi$h=)|=0_d%eFRzTu6D^yJV;BYp;1X6YuxD>VeYpbax_azII
zxtWqJqP24;ycsWaite;(*&qZt;j0QNhH^STeb5`LwSRZ5GR)Z}CY4PB%0S$(%3$QG
z{e4rzoTN+oF`wzvqG$v@yAc(dk8dofi_s#Z!*_kTO9ma8SmWY|L)A+1$S+`<>h3(y
z8UActF)=c=12252Ns8!gu)=u=CI}ktXt`Ppl=W8@Gz!efF!6-mC7Em;)FYaUH5&t;
zFMaz}o&tGxY6*l=bwfr?GDFg{i=rcx%4*kW)EZl5GdU{Qivh@ip8BDvcC*V31v&3S
z<b~48QV>8DAG2Uat0j_1ZGPS8<V67D_?cv>Vg`T=qZfai(h2O`%_)>5kj_{!T%&#S
zAE2lQY-4?nhf{tLRNsZ4<S4pAtWIdbzzv8)2<|imm})PubR}~jRyX!!emroQ-=596
za98v{za0vU5jAX>{MJ4ulp<a2`khbij};+1Jv2QtBc%RI(}{Yl!NLcFT>JKLD%O7%
zSg{$0RzM!^@5Mlz0Cp0}4{qQa@z3%WuROH3BiIA;HyBMmTyVEoF5~yd%c+v|b+udP
zPhfF{sbT{v9}~dM1yZS5*H;GVV;dKT#|an%fyozT5$9=T1Bm&hVO^zGLnx_~k!~C#
zRSvI*V<WRHO&32L81_k(=I(ozz7~vVqth1<?d(I<grtt+m1@K4NW%EYv^n!ys|9N1
z<R9k?OSJ%JEBziSLUnH<D*GGHj0XTX9<4h;v)46?G7f}+8}7d5RIk!JR=1<I)^KZ=
zqNcl-$`phvdHHexC1%Kr3?)To3!VgRY!8dXPq>CU(l^RkE~^^(hb(y#!k7+2iO8Du
z_j5`H@`ps3Y#Ngb6rl4qUPH9mK3&1UKO0=z(TZq9HH5&}x4NW#-RQ@2{bK`UVr24V
zvzxTWu(^w_m3lnpnR;cANlRVbc(uv|j^%`;8mc`hE0O=JSA{DUzj4;CYfbb`=ZOFc
z@xN*2*$!JO82D>fSe-m4^QSJZcXlGAzMZ|1UxL&^c^Rg;AMa@i)e`WQ^mV<j2YCUt
z4zTk4m=VLQ14x=BF>C!|%le$!>N?kNztusfDd8d*@i=|)aMeE$^H}ipTOwsuYsIJY
zco~i9O1f{+lN=t34>MhGQgvOi)nH@oHNlZ1Xh9-coOm5Zz(i*VQ35aa$DVtUe1QoO
z%Oqz|8jCq`JZ-u_ECqY@;GPS6jgA~ZRb#^gO+r>pGFlBlLlz4vN7pQn3cAm3k?}4Y
zu-0;!zWAm05`$cA7oFH-98ZeDk!NobTDU0Huwt^G?ox*Y|DKnWuyD#g(3Y`by~xEp
zK5q$+%Y{QkcF$WPn1GEeJiGR%b~HMq`TR_*P_40vx1|6&UVWtLVC9@=Mp#Vm?=7FN
zjY?TOwun6mNlV~)=euTdo2R2xi1_v&%m_t0U3J1L!WMedRc>f-w2?+1Gch=Hgvj__
ztf)?&1ognRz&TB*<P>m~%8<=$sFJQ*zVB~oQ#(C=0n+7I;A`1yQu$U#eX03^gdtWk
zh+jnEz)=z^E&rs4{zEor`aG<hEpHx<DEv5Q+cN$W2XHeuVzirnDt_l3uUO~){<4R=
z_4YOA>8a%RCr9`GNkt9NQvmuHz(4%c-IK9`!_ct|Noy_Oa;!5R28uOke8yyOM6B4W
zo;2jZ*UL{jfsT7ukdNn_F4V^jtX?JiUsK5%zCqvCUV&L_hr>ujUAA(z&g@;$>w7AO
z@G}<`lhk^?cu70h!T#l-H3y?!2NFU=OPqT>*cbSSxbOjR;Z0>_Lg^R^x?~9@^g^i<
z)EWF>_`eH+OQH4m2G0$SRrvm-CPbheB4P=}w*DEa)E`xa?dCkfeBVzGF7kZU326px
zN?RruK0e~4n<h)hd^fqLKB}1Pa2MC;I^~PMtX_T!{;+1$8#cT)khbnHeDm{%Hl~#v
z`#77D;D1xv{4t=^@d<v&MLD#EM({GNe{-zxNoPEK2G`SUJC$%KM~g#bBNczTQ#Z_R
zI73M_P5wA4p}@`<YDKmmFX4;CsXY5n@c~ps^`9TRfZG@r=mgM<aECSa94mg7Hg?tW
z>4FvOpTiV_ZGTwczI>sQ3a(hNd;8AnueELHSS(K;n@>wwCK*Sns^f^KaHZy7+srn~
z3d;JF(_*Fa7nJtVjW{q;YWQ<0Iv=SN{98SmN#s|%#rb%6A(D|Mayiwt$e?`cJChK0
zB_ds@az#6y)|5i&K`%8<k#FdWC`h*Wk20`EJo*@20`M-4Jq1!j?=q;2p+OX9yD|Ce
z2d^);woE@(F1l1UV`}Oa7Gr}iJLJQW!AY2&?aYK3T!#@SgC0cT1GR(dtQ)l|q$;Pe
z(7!@1zRzjrHjznzWysEspZuBsJ(3Qc7)d*fUX3!5fgNE(l_vaquWJBjZ1S83xhsfe
zh%kik{!CECBK7K#)~wQnj;E{CR+9GsvP}9<Mq16i;qmFQspgVmF#6s7-4HjSRsrk~
zK;BcH)pDZc(nUX8@9GDIkdnTf#BZtjI>~=a+x0Za09Na{Pu<M-iU)wFkXpB@6Gkzp
z`&PiDsp5r!L1Nc15M&{l7z~V&>n=-YvKx-q+`1JsYll3$O%*IsxT-0pjFpsVv>g`+
z*6+aWI`0u>S;VYM+}qz{)<7s7ae2Bh;fTn@w!dFzOpt&3mW2Xerb->>PBR!uJoX*s
z4>n7Q!%$jD9iRmQWbr|R)8i+q<*<u5z!jKi9*f5AKwG-(KLmCwz(rdA?O>cSOD}~A
zY)eo*z_$fA?Hq84I#I`YW|=i$2h;)`jvoL>>*j9|QaxbL{4;<wS_ovNNZe~a`?doL
z*Wc&OYyWb~v{IC$X5C{a)0q;DW3m-GyO)BBU{{~Gp$&=64HcXPg+jinG5dn?02wQ7
z?d_nPC#3~D&r)-&U>@4m>~`znNzB?z+ScDH7AZEs$Mj&ft8)Z6w5<hxWW)D_x>&Eb
zKKOqdVs=`T1uoq(IBfo*gzg?5m<t<%Xq{vUp%-V1&2XbSDJm_r`o#?9Q&0sURp-9I
zqGpLmM_OusOd;7B_7_i(`da*hlt)ofei;KW2FBFM1Bvi>X+l&&1v7;u=Fosh*?6tl
ze1<!A<QbR?07Nu`x4+QXJg&40R*^R$l%t~?m^depidJqeb*m`|zajfClhkUkmJ_6v
zlFrnkU9z93)zH;%htOgG3)CeTkz#U^nV$fUr`z=0R?tvgd4^bMbEtrg7mh8YxL}ob
z-pZf4p{B!<oNIg=!?HhLnhFykFhZ1+3%|QQr{+s@QaS@V`{|S?&wG(TwYn5^n3A_3
zPCLrR?O%|$Qia<{xEc*+_tFWhpfn?qP*4D9UA>YG7-LzoXZ*urZ#S>m8J}yHe1}VE
z2CdBC^<YM$kdN96s&8^AUBQxs{YT{hy2ZNy4tlK!Vf;4ppQaob+?OVtTwcaf8xvNa
z4o(Hv)G408?fG{J<qB#Q)*kicMhg|eMpqgDg9DqDj@|`4G%%qG*TRG{4N&sy4#Z<T
zX+J-KpI_;FafDLZakPfV^?TZyaBCOqfAktbBoHqFtMGN@_d1*^1bgmdlSb&mR9gO+
zFuxYsNh$^2Z8?i7tvYWEaoiYz3(+K5_uGukzV44xfH!M@p`<!1jaMRzB@o|az38P=
z|7Em!{$1SPhskTRc?#+^rq{hwyjGODHa`+1hU*1u^T(aAKujv33Zq@p=a~LRv?Bgb
zSWY3TshXyWcnVVV)f+&GIlZ<OBu!@TH%!%Wp41|KlUsMS>6CI)W$*Jd*5?h3gW!4$
z9AC*|tl4%A<jDTcVQ;;EdNtoxNJ9e?NH6>6`0{<@0!h6h)cqr3iB`h;-Q~XNAMWjK
zuKNvq+Q`}3Rw=vqV>{M6+-b^30BF1kv~0NIuGXwAF#57pi&6Eigz{f>UA9n4WaITD
zd>E5(bm9M4d64<D$l6=e3GWPBgcmSQlPNkPMHIC$2BNXD%129srb)~t*-}}GTGsR1
z)}WKPhPB(dWW13AKIE!oyg4B*-hEH6fKF5MYgYw2`f6uNx^uvf&5V&#+6DBdx-)L(
zjg4<)(8u>)e&L`HIH!Mpe8m;QE<<P5fI<I>q}w-$sW;y7d`tOdLf`v{0hIGNnB)4x
zY3owmknVKJF1KJcqBeo(o34$|3AXReZqC2Km#3Fk>nWz&?OmeIKa_rihWg4AUc2~x
zklzQUKg~=XfZ4=}hH4sRaQ|pMf4iyn`e$f8K0Vz%F3C9ngoXd|4r#UOgW>3Z_n_2P
zio!_Uq$95h&N)uU^_ka=V!y>r5zh~=Ml&xyKeRIX$CH*vx{=<|xGnkyG?p?IrEB9f
zVHVb+fvhnPCKiISgHa>e)H9c-;c4$5_;V><A8bhNsMgUb%Z!0BW-i8|(dvLhiTs85
zsT>GqzebV!1tufl&$rT$Pe;xO!?YRsaf7`JXi%|$8V-?4>|B?M>J-FNI}}3KGP+zB
zsZg~Ov_8<Vsh{~xn>W54$o|c#;?i>gb?whKsU%lRG>W9q?TG?>ABQ@@y~}VSDdhNu
znuj0QpfDUy>QaHtkgH!!JhHmFvOwSox{!on8W->G_+zhZUN^hWv9_MVgj$;x$vl(5
ztdDChA(V2XO;;N&X}|wQCW2_vjs`d^)X85sw!B#Y=*B&oV07G9QI!4diON_~vg$Ey
zna-IYQ`WD?TZe?SJYLOts3mbI%z4w3F^pA`FE4I{h2pq!d0D2hs!+CH%)YTk1H)V6
zs%;8^&gh0o?y<V$WI~4eU-NP_G4Jo~VFYFqtI5^XVgCKGoC7su0V0c^hh~L~E{uJc
z#+Z{81mO>{{u1r$)(u)EdeF{@KvH)kuWYF5nUfxEqbrEIwp6J*!y{?3CP|03%%zbZ
zD7_GnJ1dLtv4`@{(zO6DLE!T*=OXq&*G@R)e}e}eyB+~_@x;G6(V~JCSfv5)<mscB
z#I*Ux+jE`zJV2wuUP>l~1sIj4vxQRsgs%X7Kon{h@HC`xI^@?FDvO3-_k^R6Z~}Yo
z^SQkrZZC=dzujy7T81)`SQ|(YE|uvyv6wOW$p>OZBuPpd0iN?EOQoWame-ZPk6Uc4
z5*u-kCK4=o-~WBs=>Sl`E_HvfNqlob@3Rt*#p~5F&lJx-l+9qKT3o7IoK}+Cc&29d
zn|Slh2z<=;AqN4PQMa$AV}G1V!g&G1pV90yu$>mPw4`qLhbq@Bm64BXC2+DFEe+iO
z(<CK`>0U);E2=-vF>GKF!kQyD5-AO43pk=`qC6RonGS6-Ly}UDZB3L#zqoyHLfCk_
zouDU(DazEBSC)WaAdwO00F-J?4?_e>4#|&aT{mtTtpZE;u?#B-I&vF5#%EIazR|D0
zF{;oZ#eZ!C>b6(!CX-Dz<gaLlHO2=Rgj7V&0!yD_43?7p0V&2(s5PK@30dSKh5Qp-
zY)72jkb7d!sz_NLk#WTTxz0|iA_ymYK@lxK8VCs>A^G(`Nu%$nijwTWzSHE4X)qM_
z2B8+^85<1!43_nm;{HR12hg9K={ZC7H+?WOa=b(PIF(j$V>16W+K615vECuW3lmps
z<>hre@qyMW_}{!R{K)fXAVmaVn~5A-Ta~LV<$QCPXX0(>3{Z|-af%VwvDizO7IU~l
zkzn$gRC?RN?A`+sTWlKQ@PL3H0PO{a0A8-0XDFx|7Qam%DYi`>#&B8f5S$KTqh-HA
zTcz=PEmKvp(oR{#pim4;F1;_~>y!wq>-5|l+QMKSqKV?SFjuU<(^_-Dg@SB4<5wY}
z@}95QDo9vU$Ctd2L=cJOQG@7mR?>A_M<|(q0X<7?e;V<IR|Gvlsh|6K>@f#zsa-(>
z0Vo)r8j7y%WhZ|t^>ZCo&Vl@5x)+@GI9rR=;mpalogyp(c33jfhm}$=*GGv{z4eW~
zajk-vsq^am$bZ*ZG|UzaJfi{RfA}9qk#mxMoPD=N!*lERr>$k*Zk$sn+FTAmuKtKb
zKEGG%i(7r!n-74#7>=Nz6CVwHABx?1<<RQ~@>n7&$o4DOKTc{kV~a-CK5`QM`}Tc!
z{qvLl=51}l^$gR$S?3B;0Yu+cGVxFUXHj=;u`z0T%1GK2K#0m@`vTf=naP4&G*Daz
zRpcFr#nJpes`1GIz#1V{iqDYH9r(PeFJ$IuYDnXZ|HS9_69m?P%T2=dAkRP_Daa9O
z<?bZTmOJwjeHJ%eO4yv<-n_=;G$=QtY!yj&z>MC<_D(X=HngvVlF+3H@_laNXzw)E
z^6tSTM@_14^FRe8<%bxACZZ+S`TB_Km|TV4_+v--*^c#niIaNu?Lf@;R(QMijO!E;
zV#Wd&<|c$7yf&!Gct;LMB9)W0B};hWzg~&0>+(WN)%WGfX@9x}!Ta%7l+hr{3o3<H
zPq?7_ILqblR`>h2>W&-fY_AKhB;`!^H>vqooUX4_^nT=|-Cc)z|DtJ9ov!z~{`nHU
zpMHcQej5dr7OX|KOaU986hMB%WOk={fX9?tT}>Z~L~!AV9<mlzd;avjYZc{%y#rNd
zdNF*+V^xL^dX`4Fs2yK>FITjpFu>AhCZ#z-<4;$La~EdpqJvc+Ssu+Jk^lUGD$Ikg
z{ITkSNLe|0(WyKa-1#`_IZt>#1U`c5K+1wVH#q3%msoeHC4-8j@p(q+_+8Jz>8_C5
z)ThT45GGrj7nyexn~nMEQB#XcW>+l*O~tk!>0dF4<1WD!mzq04yVuqViceK7&K)^j
zp`DvDo!0|W7Yc$phiM{&6s;Lu5iARQvak5MXl~OBk_vkaLWDASU57X}yx}U;Yan*d
zH!5BE%iGw>rnx1aWJ8;{T@QLU5omc_d1J3aX3ZxvCj+46quMdfJT~7B5;Yslni|%Q
zH>7|<$$+?_t)c?03eNRn9TBS#{qF8e@HCUO3?QTxKuD}W!o>)k;uQt^sX6FpOmGWp
zHZIPePUcI%n76yGT3#NsomLxfw7KAUoq1O-P?C+@7VC%gXhR3CdtG({A<5X*8xHbm
z%*lWpXFyAKWHCyF>bN^dg9S^&f8L4^)H0sA_r;T4nYz(o5p%%`niy0^a1b1u8Fo;N
zkBJ!Tpi2;<NtJX>QzVdhiHlh-lOm|@98D1S<13!W^7g-<bDD+X`2bI|aPzD2bSAn$
z_Nx15p%Ms<aQw<iF1rz8KQ#0zsI9eQGSnDO8+HjicCWuEhAuIvcHg9)@*JA<4<@(*
zJ)B}JKE?7je64~OCYc36#8N!q#?iAdy@qf%wTFjhAV=ryqUa@)Oh$~{Iu@T^h1#%-
zK#JwE`Wnkrc9_7%(~oJ(BY8=bm_H_;R^2WRx$C_~FU<iA6wncLZSV^<J|K`QeN`?~
zeQPL3o;BjFZJ{_Jfgj37Ff7ur^bl8H3-iD8fI~{a9IEx4u!>^Tv`k|&<^D_VH-u&P
z`-5A#taixp0oyzu`R?(KILaF7)yq?ANXC`Fdt*i3LPU&gS(pbmQ#kQS8}S>4xj-o#
zpTSl0siJ%bgarn9nGh3}ZE9hIPSZm&CIv%7QC<<@mo*#~Gg%YGcK*_PYF;TJV<}?7
zgIGDrLW?RdKt*DgkPf4m5SUp!Qo2aZDwg<fXvlN%o1|O;{0gbV>~DSJm$*NuU$!%;
zeKwH7_fScQL>T?9i4P(mP={NOu5wzg!MrK*=w!1lUV^W1h?fY_LJqCsFs_DnW36>k
zWmUav*l4xu-CKrHs`~`(0&jCMk5U6Hw0f|F(*5a#87!nKiITrsIt5e3VVT$P61H-G
z?_IIiy_?0w!SZ~i{~42Bmm`pS_#W^(RfHnuhkgCwoBhUErGtpJz+_gs@xk)sgA0gL
z@mc#=eQnHQ7ba<&zhL8hdg0Dc$OJEi9L<Mldy-q@o~NnitpWUyHI|3>q1_soNT}d2
znI<yv;7*c4Ri%6k#-#6ghX4;I;e(VFN7{#@-+h{b501YmHM#t9j??CAmu{EShrUng
z`QyGz*Zp40*@H#GhNf!rl9n&QA_X#@<8*JkuKqI*kvKf=Bu1*AN9yI-#jnq#X-U;a
z0%w`kHqSg*DC5ZK=2s>fQN_~`V$vNj8a>ZASRLY@m&FKZzO6TZV<>%U-t-Xv+(m%a
zfN!Nrom5@6acDg!rA!5^PTXL=;7LOo>iy+7G4-|o8)uC~PBK>VR)G0^nf+o-WfEmJ
z1*x}j{^<9$eC?Bs%6I?)bl1)V(ohAmdZP|P^G3n-GrFdww^%8d72|!al=nyJ)Qn}j
z_7-MifP_8&v)uK2V90R2!FC-(s{q_r{IRKz*AhSOay`V)*CmMQ{K6moSq=5k1%lIs
z*a1}mO^vIAj!<%FLXqgOT#!&A=8D(oii!sDz9%NcJBE7X=VuEsV3MNAu3@3-lDS5=
zn%gsZ6>_RpKbk{|+GfE0!mn-3>rM~b-@Z*s96+zY{*iYjK2AhF;xHsFuwWY#Qo8HI
zsMOb2g8KV$jVlm)d&}#_b-HBKmHE{t2Fi5QYwy*rV&z?gA;=<W=}770Q}=&&on$m=
zyqOWBX0XcY(6J1a@~@(6BiI-Ky4Dd$d0`@zN(mu$AV#%%D47s->Odpuug};{x7g;&
zY9wBP?%~bXOs|UEFQO1q?idV85TZS$SOk;@hDqVbBhu<A-x`V=GO^pySVn$1^JIxm
zoHUNea*vG9y5<w7TVUynss`pt?*0^oq^hL;g5DUfp|;U>nECECLD0RmdAP3ALRlEO
z<M`}rwa|80H?`UQzSe#Cm?rJ$QZ)mmjj7m1K7c3Sz0`-L3Q6tw3C<%~6k&#xhUa+P
zMIaDd70P+&ewC+NsnZ>gCeK@h#DJbZv>XXduqq}CWlf93*CDtuKF2-0XzK=^0=ytV
z!S5h_6PKcBUWeFRY!Sv5$LMDLahG=fxd*z2R*w-@b&#rrKLfY>My<Mgn<mp?2%Gn@
zb!rgPu%Iag$>&JKW}}Ub!B-84;>j8s#MI9`6-qAb>T$DsM{HSllzBj{C`Vl#L$|Nt
zYAlD@ba->1NX(xYEkS1@N9F6-QL8bzwS6?)$%}I1*}Um~)9GN?G;J%!x7y){f9A1v
zO;){x-S?@+75*xHE>K}yxLF}srP4MeY3lU)(VFAua``1&#O~|yX?h;a(Bq9o)FjcM
z^QfLF(?M-k7Dqtsi)J-^q;9)9N)k~MG7SthO46?J=aWP8rY{8$mEkvA_uccl=-G|w
zAF|Bh^}J<{6#=M5g&72oXi>79{t?v&usi@?g4ZtiC7+raM#c4d>Ped!^T^mqTlHh>
zZ`QhTOfMxnstV8j$x06S(=}h(tau52@+4{1VwhtIF7tf)H2ZFVehEMxw+^)^_}aT&
zIa23cCfs?=0JcwIF4x8KuEjy9EID0dq@-|xW4*!Olp4RFjIL`_-}@W=u}Bm0$_jQy
z%WrKn9p|?8yFgcH7bv%hiw5l97wzk;m*@`;4jvRc`1DL@n43iT3V!Ay5vwu7zsV!A
zVQj{xsl+6K*#<{~QgN2E?W`ZWB=Y%7!1)Yit%^F_fJ^+37J%M;6GhbVD0VW|fE|AD
zjImu^Tp>AIzTgI*uSPDiM`Zshi3JqL+&7tpv@-sHu+qVd<Pb~67#mdtl63_QGWX2o
zC}pf1^n4^+i1?g8*Mg_YoDZVdLrxi!9J4f1ko9NsjZu$>MA1@}_oiv{tor^Bc`iI(
zb@7n;j~36*N0wO6x4Bk4AXcdR(n$t4(J8&HBs@eg)eF8J<>2!f^OLN*7{+ui`=xof
zuhPJ#NUwdC<rG7ksw%~{no&f(g4QR!OvMvZZZ8<N%Fdxtv>ucm!+jN%%=|Olgh7pf
zw;67~jMAf?QxLy0R)C9RQMJr8e^U5~8D2{_K5LlR2fs>Qdn<P3nV0P=hGUF?b(nw7
z#R@O{i3d6cwKfh87P!kH`2E1{NGMLjsn?ueVo*{GmUvIU;4CiIuS$b2Jw58L?P8FG
zt3{g{gQ<8(Z(-d}=$nWU%b6(#NSC?t8_nvjq$<9bE&4MYt*Fqaxs90!CQ}{aD#2kz
zc2!y@qs=XSc$I<GOxbuOrGzkn|8cuY66b9S=998YcjBO|ER&k|$);5kzIP$|jH<2`
zcREubn8Mw9vUV?Vth^}<m)icRrbBxfkfyoW-16Xt20i-M+k_E|5vS$-+V7_163nsQ
zy^1Yyv7pKDaR)2OaBPw1RS|^XZ90q#z|;R@doWQ62ZwBscuT&`nF}m=g@nLJeDi_N
z0YbG>Ne`PAAD%PY82#Mrx^XpeJgwOGl9j(CZJ!<T<sj>zUggnL?|JDogML}^{AaPD
z?>vj<Je)2*aS{Q2+f~asx(##AMj#5_JnglZXpk3(zMbO1Bmkq@pAMF$h#K`M=l{5l
z&FkY1b&XkU@+n0yHsF1jUm&S?Z<3}gxaObL>>s1|=3)}owd})TDOr4Fv|)wH9MjXH
zN>;SgHkK5{e7(B24w%<EJ}Z*XS13)YP40b_B);2>oQt#WJIVjuHs#JKh9O0*Q(s=}
zzP{YE_37s~b-vXeG|GzSwsZac#X42-`Rx8B0Lk0`5+DGu_J>NX-K^^SejBwCjyls^
zmcqqT)4iUa46BnH?6ctD-_P_?A|vk-v!1A<z=IawzdWHowqG{yqQOlWl0B~^Ywxx@
zljIJQ%bK`fYLYv|bFq48AV-@ybYxeX6xho&e!ST5zRsz+96qu;TjWkPZRTa4_x2k@
zvPcxcj%LQ-=SB^f98x$AU&<DEDfs-8^YOq_W;s1Nt=Uh?Rmle6_c17oEY`{>(w(Z2
zYOdUO;eRFLYrz&UVX}q2zP4UHODNaHO!vKT>yUEiM9(iq-ltK7aaN8S`ZQ-5`AfsE
z^t)JM6(q~vE)FdpH}$nltFEm6zHiuGHG(a#b4!>-jZxHWDx6%+#Jk|0LB`I<PI=Rm
zt(sB$EV$8bUWw&+pROaV{3UJ(OByw~?JZmvQB#>F_sY>fiko$oh2cK-`KIo?(xGjQ
zP2k^j8F_mQpJnb1<FWCc{rJY!fN0cNaHry*Y&MBpeB6-^-qF0g7D+uY9~PxR78|de
zj@}t;tX~yO`xr4``P*9fVV=2}_sp_-eL2i`3}5V8@F}SRPhJ7Hlhj22FwaG>+5wqS
z@9^!4QO<wfi()Re%BuA+&4^m3@!40d@+A&(F^PdYEH9zG-88svWhn*^lrJpoJLFww
z=Qy01>&MbpBvF>V*pC<eUmC`XgWm>R16xE;)1{$jH~aS9AR5K@k=sWNWFaSKjzsq{
zpA|k&8D`@e0$mjf$KDq@<@b(ItphMp4ZS=*-Di|Dw^e`H!Q?e2Do0}UOMErK@sRyv
zrL!D_oJ;-AkZajVvBFLvP}5*=Gc9sJ^qg5i0V&V>Y(n3;m#V~BIm20B>og}dT<5Z8
z2Ev}@jHj06RqizRrnWm0US3{mH3lT~Wzp%_o6lYxgfMW&1{<^Dps0L}*>rha#0$(F
zeJ+vrT9hi>gK-F~x{33Cd~CU%?iVl{j`^nA)Y}SM<GnB@$*FhKz_6Fm;+-ryusndG
zu=S@nF$P={mAnEjYawL(q_5Jl%F8r~saV6N$#$%6m@jeTC(o~p43glUW}ZtzatYlf
z8}4IK`tCPnR79O#_xiYg&Ex<(;HVt{(Yis!m&<RHhgU&zwhbUGTruveo<=x^QQ|Zq
z>Rc;MHv!T#f%a)M5e1D(WGK$Awgm9B4g}j}Mp2@dOwIo?4<K@@(KjWCsVFpbNzSah
z!UH;lh-mDot$$3aETtqlYECxw-Lz?T2~D;_cq6VD(Gu<XGULa-W_tFW2vfj}y;zq3
z^nqB75pivSkzFroEMv}xW#b4UM(R8n4*Pa+6<l%O@oNfv1vZU2&oK7%TRms~&$pt}
zAMSc@^>A3vOy)U{ymrr%-G4jH$8@(u7Jlp$qj5flOi`R999EsJDND~q&*(9-Y{Mo-
z4(OF7Uy4obd*MY(z=w&fHbd6;jCh3AK@mJM{1%x+z%Diw&$FKJdHg`qX6t45uvyUF
z6@g_}!evD^qK}n#GC(a>cIUOVh#)trB*rL+6&19MR!+&{`O2t?br_7VraRNd6x8dV
z#pCAqhbRn$F-<m{i5U+ON&2ZbS7qdUiBB|I8YKxDkGzj2#}B6p8ka9xZd(54ppki+
zo6*{21*jR51@%~6azN)@VpNKYEdvWqRC&mbU1pc(@z0b6=s6b@Q<w3ptd&f4Rj}cR
zXjdd<h)uRfLJciT+h{XbNb#;1m4bRv9k@ml?4%_tnoU$87@|M?6J=EvYucqaiAAES
z{dEh`6|vwWHKX8UQ-he#t>I6=&Qt!J;(VOyzwfgKD`6jOfFim)hLcgZcD60WZPE1b
zt2Js*hs`XJ>Dm-IMWPlpbX*DAs03-R@~6Ow(PU`|@{3v}-oil=)T(nomy-oDkAWkb
z6=cxkY5qxjsa0i#Z!qERU2k*w<qWb|6_K#Tkk_=zVn3NkwzphKzR?%V6yZ-J4jvI)
zM7`2@lb5$g1s9wkGiF$7amNiMCql?*$f2!uaDYTb5zE`t*?4ARgNZf6tPj8YU`pvy
zTWV-2_)s>;pIh7G_twH-#_>F}xL;#juF8$fLp;pdN3!S24#Ez0>fee^M9(}iwE!tm
zGN8UVH}R}Z7;KY;=)quybkXIMnr6NnQm}x-4ZeF+Z#pjvr&l*7pfwbzHiQ#T4+D*=
zBSc%_suM7nq}q<G3xXi<c%m^UPd)&mAaG;Z0aT1+ZLrOwo8O0U$P-b3xj!P0>z=pj
zzdMN|FuA0sw^+ut>IJHOuPG|>ToE~+&l)Uvh#sqRh$7Y#28y=_9G|#=i1y`sC{|f{
zr2bQgey(jp?&Z1tC-^`9IOwkfR<4s$5{F5L?GVga))f{j#!`=SYyJKooO^|d^2EVC
zrm3BVQh{1U=!sI3*z+DGE?bsWuj_PK6#mNP8s4T;$+T@i#UST?X_+3X0puWfB&zx9
zEH!sb&aLjJet!9vmCFOzr}70)ll&BTocS2Mdp*V%Fxn-wEKH(Dpg^z!e~xRM&ZxR$
zUiGRV)c5s1-<5uS$=dgeA{wTM1+}Ga94RYS)fPe9EXcK_9@<SQE_3?jyg+5bXC`Or
z@crfc<L`zfw4cLR4XK7BTPkX5HG3H9m|5Ez)I5=_#;VYaSnRa*iFsMO`KkH?hZOpM
z8NDDFxE_lJ_O4C0z5a=a0O|20_bIKs!22Ell0JUF)h6IIy~CH1ri$TzvG^Bt&;_XL
zch-Y<V+G<dJN`mi*fyA3j<CV-mV4U>y<(fLD>GCT5^s8>TlrT_B;1nalsb#u!0Mmp
z><Cgd4HU>)%gsg@%kzfnzfS1y0?OdMV+B9%+uK__*JVRswJ0y&i4#FQPL^A6$L9+_
zb46;fHU!Nf-gWJZ3Bxz5_d@Xx!(Z?9c@*dgVdx4s$9oy~X38j{`B@+q+HccX2c*Xg
z?J~LOsw}x&wk}{t@McqLt6r|bI)eoy^g|@G*da7dF4ov(26bA4^B;)XSr1MufY@eC
z$nflUIT_&gnFN+uNk(XrZv|&Beplw&e#23^WTn0&5Xo3#nQO*^nK3rwm=3%u6k&*0
zgCytD563R}jWmIHbGAAmlyzjqZC{Dle$y0GM7G2-(B&Ft;&q22EkVrvQ(f(m^aih7
zi5Dl$qGbw>C@+$%BP||W9a^c{N-c$}e8+aH$nJ*UXmK>NjX_yY+PQ|O?fpEUzWq$T
zVV1teZ4^GDZQe_|tkSw*Z{CwxaW!U?+^RF7IDd$o=Y1WV9GAB8Y~>j%Jhy3dPAu1=
zb<{0gf{nfkN*VT?*$DIJnOQ*CRb!N88DnWa-r0P(R?)(a0|XdUh^gctF9yyq%&;9g
z2HR;s=9mq|XHeGLFZI~)-@F|ZSe2;V`FeQ^Nc?0fHqvomr^$oLyl%O%Cx6F2Id4}!
zP=9gvQ`P%#C3Tm=6py|(?%P(X{H`pr6vZUVT3q?Vm}2gT{4f%iV+n&_p`6{w(=TB?
zikJ6j^|vJjzrR;JNxKi<V&B5&!k^ocF+MJL_SAvO$S6qZ$B6ZZ9Bg~)RRNP+C%(Wn
z77L*)Ycx|~ocn?d1C$_S$hk0;EZ8De;FzC^Br@2L_9XYaSSumF&YvaMZ^OJkB&qt2
z(hP#<?6^$geUvjtT)y`R-H_8>N=>#87v-nv*B4Ao;s=KujzeX*^<bk*r8{}Z-bWZZ
z+SVuHEcnqfoMkJ8y(t@T?P<XF{yZR<iX1GT_wz;uHz+cGsrz8*wI0V9##Rtlg!&QN
zShQ$Tn!tBEIdv!Vw4ZR2dNwNdXq0)C%W&F;@Col?v>ajtRX94|(7|nNeciL1TIT_L
zegSXu+^ctsYQ4>RQF~5)RnO9mkrz}QPB*-+*T8d2+Z?ICyxv+mo(1=Ts;U;)gYLfY
z8OzqpqlPS=6tmo=Mcd?pW}#bxO&IvwAxKmUL~`i9WU%gfIM2!swNs&hwf;TJ*(aOw
zvc!)Sk08gf86`6IIpuQvDAmiu@CHFDd7nS#J$yQRAo`xHQJQOUha6(mK@%%;8+8B7
z5$lY^@`W~tDzfT0zfnrW**cuxa9TAhZj6jt1x2_g?ENwHIQT9rK!ih`|F!hc@y1&I
z^i_08Kv{a_*2OKmg!;f?SIp|26;c1wIcb#W%PvtsIc<Ms)itTGzlWP-_sT={;G}tM
z^|d=dM-DsuW_rnbxGsd7BL@cLnx~_M29p^mwGrn<P=`G<K}~}<l2Nqg2%7d*#bdN1
zf|}Aq%>*w~>Ctgps_JeVt+Tw1o?W61u+>AjC5k20S=iPVlcOUD!eozPDTiy-uG0t_
z+UIpa7&hhonlUR3?v48u#JQLCxx`b}B)kn<4p&shq)VWf?oV_<bI^2!mS<UJgjk1{
zL|*hInOb(96hpFH$O&dT&Pdc!F8!PRqdVT%aKUnoSJNauv<DJ$DYXbp+1Av>VKgss
z8D@`2B#KH$K%0z)0@)=Y8(V1o80}>{IvDhoa>)L<krqU-DAVc@x_17Ib~2W<VsAK=
zb=;;s&B5o0=+}NbJUU}nApD9bciZs7d~)*A4W&Nar}%yOY@txvI>V<kJz`*!#(JFi
zvh%<mTMaJi{8uixG+8(f8e93}kDsF7BB<k32SUGK&^XtN?nju(GSZR$gw--Eq8_w+
zXPn_g{AMd^+z|a8j-9PS#3Pf~AQdadidA}_F_rqlEN0+)_rd+3(;pMaL`%n~hj+50
z%HPt6Mj0eV7T{$=87{ObKI!Tn;mqR~g+BuZK2nX;rm21=q*)U0n}N$*zFrryulR7v
z^Old+!-~aZFJP~V_K;@vqfa{KuE*pURYren`XT60T~u$XJoEJsU9BnJY0{l*k&>4#
z6f{nR$Fho%QoB2x390~|X@<tMl0N0xf!l<-h;j69U_XvbcphdTVmYX-MVVZ*0)7=M
zkr`2&uqgwnN=IDaY3yn3=)=z-b`!V<JEHA^l3BHwq7K?9gD8{&gx!(fkHSins2*r=
zWJt2o7HYzX!PEMFQoeMf1L~t7iW%wbjjF}Os9t%bCO`9*2dy-k1?ZFgL_-#Ju_HmE
z@m%W`5z<4^L&p&ie8HM$HM34FcvmM3IH$Tyva@c4t)m%Qf0R?hS+2RR=h$~ER*1yi
z|Ji}u@B(+mEuCa3nxbhD=b)$AXDg!?k&Z`pq@2XM&1V~$WO1-v9863rL@iwo;~6V`
z^KOVoPWiH+&0e_`zJ`I_7=-szhss7fJ)>MjPhCxn$4ZKdNh9epOSojrT5X9fQ8zKu
zYl2<oM#tmQWA(2h$WFCSEM%544LvT0^JrlI5lJ+@LH_pm?&Rp#WRss`DJh^qzPI<9
zGt)W!Nq!r@*LTXwe(oE7@FJJ~bez~OEe+QsKP#mFnhPs8n7e$p&r}>uCv<q}eKTHk
zkBn+onpVmL&U1F!YP&XzS#s~kI9(^c{n|l{BBLBCh&T+X^?+;a{|kol`*f?^g7sLq
z7H-fYm%izov&`{#XfBhY^mzLsGU0l=%xG(qIp)d=JTw}it87I(#+oX^RtRx0q%R8c
zr~zd{&E(K`b#E7+cltA&@Lh$db)>S_X7^e%A!Le%*|g+O3aDTxK@3h78p8%ZNO=R&
z1x1WBA6Qw$S0jl<Jx#7tl&Nu{%g#_tt1d9Khj_W|Yu?Rq3QOVam(Huq5eRa=$nxbg
zH?)dYHNE3|`>nJKS1?EtRFPnySh2fTE-n(A)8*ZIc4$x+RU$93AU4sVt+z2DITBzX
zU}>5Jd|RLLWXG$`hX5mZDnYvQDvoKZcDMkc<is~mQXAMw7GM<deOK!a7*d=|dY7Y4
zVEKixpMI)`slX}IhdD*3K%B>f<$SEdXeKKbl2>dNH&w&S>U!_qScEs$j7wEa1VX2+
zg$^}axQC-+5M}@_n!{I;pAD<hMPQ*zO9;RGtc(iRcHG%)#;0f=R;qPb-6R+hyt!rn
zX`}q$$dBbLPDxGf%yLB25__XYd~|g2S6O+%d?n?5)>!D@4y)RnO)bcar5n$S)fA8X
zGj3|%eRFXSId4)#g&;7;znYSMad5Z()+TyRPAuL-2^Yqee{(s()p#9tC^Amhw8E^)
zsWGRGzV(q<m0HJNhv)vn*%YGUG=Y=7&pmTX$1yUtIPC!OXIDe>=wZ5j@shk4%=*Z=
z+NkL%e~s$mT#_>r7Wn3V_vAh^2Rp7=A@P#KKK{Yk(1#n7d{|8c%}kWz3u+Wqo}F>b
z-$L}#geBHs6mg9w5OQoICJ-f@?-?OQYHD&pwE|!&NOfm?OCCSTdeXHBo}1*GYcD(E
z4RF4o7T-6TEr;bZ4R)lB=TS8f|5n*Nkh|B``cm}`RSYHRj=3)S0xUC8Hb2jBu5HdB
z%ZwxpWhBK7915h9SpaoRT=f1+md&+A%G)_DN`tU~y;ePr`BafO;n^Iu(z&V95;nYM
zc_r(Y=|U*9Q5W7UcB{`c1lXNFC{es~SpgKoIFxzPFUCKw>650kG{zf$4HdfGIW~E~
z&DXNTtY{(_z*~eUvoE<ub^aZ~n)8bFxb%wsJDjBoW^s40YbAANFGl<<SOumnMqS`Q
zgp%KfgWxM#bu!$Y2@P%<VQa9gN~-TzQf9X~?{Uk@a(+T~Ykn>GokZpenU}kanc;{Z
zPdzFd9egnfa^g!NYQ&*|mco}lg-GFKQvRg6HL%lSsmtAQY74XOxMGT)N(`1GHQ`_1
zxKkNK7E}ofBa>!Uj~kOYGU>R!Z<EX+>@J>sJPuJRk1~(Go9aK;zOX6hcY{JQ8Vf$F
zu4L@5WvMT-wj8p$^aRAa3e+<ym1UOD4Fxl`Z4vA<X#8x{{Y6G2Z)C?brwFcM?+G3Q
zo7fxP-ML${{QDFJal-||uxmfN*OgrsQGjs&2)!Y-`{f2>6zhd=$x6HcD~UJplq;C%
zbh&4{G(LN1MTOP{n)-E;EJ6*z<8uBy8BfOaIi;h<YRnuwTuz4e`L{Z>80XRihpast
zuBc{I{quu!*Wf6+mf^Fh4~Q_8Ytd1XmCsRc(r3oJ8a>V1MBpFpb1#1HS*8pfIK*_0
zDh{=Hg0J97*afANisELG;8Uy1N^sCgad+eC^mpGB;`@<6=UM26Ef`%ca*L_^P3pcY
zvTW-lr)hbfmU$~AgYttC?K?S_rgg)XrkC;7CQe!L@FB?#*D$geId_BCx;)D(jFqSO
zEg^7P9k_T9r21@TRkWMA<!tyk=M*W6Tc<24X%IA+h8$mS)L9;EwyBawS4ZUi(N*(T
z#?+GH#J~@gh)1%fMqki^R;Oltz_^|#k#p<q478S`YV>c6MZk+&^Raw75!A$hjireU
zkB~V}vWW+S%ovV{OxM7_g@dM7e?yZCo^HKzbNUltSH7`XpuOENoL_-`_B_*3O$Z_E
z77Zh^sBn&NjJZm;b!xM`Jd9g(<7uG&Yw29KWVjSBq4w;o%|@CE#M5_zl>;PdA`kk$
z5h_WGE5t>Ud9s=bj#rHV)@c=c#sc0#CEMcxoh?gPK`soxK{*7rue_IjBdXTsmN)t5
zg%HD%4juDvw!SlG>-if+>6Iw*L)5%aN9wWJo*ygKW|zi}pGW40d@lat*EuHEO&W&l
zI2j7~WaTQX0(NV25-<dnf^MzJ3ze7sDz{pduJFyVd(#ZkJfp)B4tRcELlTItGmD5K
z_|aKCn_8Ag3m`1<1TJpVRwz=(9iZR&A8Saz`w5Rfzl@wxwSh1xjQboO3M3aX<g`87
z5G>4gHh#SyHx@F}7ea>kkDK8Fi8S4RJO@1df9dH4g`yN?tK6iQa^(~KWaSp>qD>`!
z0X-_024$<C$|nWVWo_P@Nsl_CkHdUjE&nk``S-BW1wXoE5yD+d`na_xtw&6{>^8s+
z{8Ea_77QP3Q#Hm9?Mm_ajc=dxzt4k^`}mnkSu!)+zOC8^#b48&;n71s=#J)&kD{CG
zoA<F~<rDv8&Hp;9Lhhs2rH)oqMNz;Z`@i~6f$Ic>sYrL>-GKj1+rMWCxCBqSbOQ+^
z_#+JWpOYU#7zH}~WM!HnbY*FCuwjrxu#$_*kYr_hb!i`6s{8wQzSRHz8*7_ARHdt8
zl4kpRTzcUA5u~Yfy11c!mvnSNKT-MLyyx|R(`250N>ODcDsvw)rRe@24mq;+--$kf
zDVh(x@r3yf<DFW$?&-sIwiKX{w3I`8NwxLqBu(}0|8mG@T}a!Me=n1iLn|R(Uz27_
zfKLjPKuP+Y=?C0Zm*d;3KN59OGYAq`Dj(wh*Z=$)3L$3;|2xsjFJFHwe#x;XIb72I
zc*gp@=bt41--n9)1nXo4``;(74=sSQ>A!z<uv+Z@fBHX8%m4kH_~gS8R`Ms{8lM{a
zqJd$_Oh;}Af@t@EsJZbXkGU0h)Qp<b3l-eB)p-lN;v)S1Wzk3sQl44z`kojdc_G==
zQR6R0P+HZSa;frMlh-{p(8v$63ZcEf73Mgp+6C8B+K~f}c#SUuX8Q{wmKlTGLoq&>
z{6PU)R=UW!HA(0&eQ72?0Sp6%TbU}z5jK#RE#h?n3{j3$5gbZRETlj0C;z%ZxF{HC
zQ15P8nG`9xY-y<?B{9lRvJ}6sKN;8-eJ-vG9f#T25W{+NM+IVsjimOyH#`tRMTHAs
zgFn2-U?KjvAqo=T?;nj;&%E<ZbP=$a_2dfpxmCaSCV45BcF7`2sAcQA#R~e>AWDd^
zs1UFPOmEpgdc#9SUtH#S*)ldZ&d<Nw->513O{n@Bmsv#k_Zdx#hK?wEgrui&2X210
zAtTblB2cR4H|7e?;!Bc3iesxbx)Uft<!h1F;sW^<XC~SG0qfbk_^Ei&Rxhp)cju$q
z{=8qEbEcY4*NYurPgb2Cx$jBVM}N(hs63}$(kZDt?uspi(Ug-7JD<1H7|TTqqrgz2
z7cd%4j!#5kSQQZ)F#7mi4z;d$DVBC2w$;_N`z~!!TM_QS_z=WUQq$JeC@fFV!+(*J
z8_a`W4&SmYqlTfh4ekCPEx<k6)>J-ok*TR_AQ?;34Q2Av<dU6<ZWU2X9Zq9?BYtC*
zBK5Y9j49iUM<o><E139U_nxte>eF<-l&bpf8dFo1MVd$=jA5`6s5RZDSCk!A3%)6B
zbGamEhrOkh%#S%wb=W{VNh-srKCjrtbYw6!H53s+y0k3Ng$?tM4EzweU&@hKBvfRT
zO%Rt%L&(BCSR!l9@KE67iLi^>Ny@><iqeL%U0si+*^SKt8ZZIC^~q$OtmrW)y}FI(
z!-}AoVhXE@Ub0|Obw_$(3Ll3M^22DbXyWM8Dapt@tp!%%qe$Y70EozFSFv}>pIy@y
zqmu`|jV(J>jxl}~mhStrR;|k}<AbM*+l%&O^TwuL1@K2uxIMm^xqy470A*6O3%-3E
zxgz5E1SSQBivG7OskW_3GQnpU=>`kDM<yeb;XHWGOs)$P=(r_ZD=k}wPaFHo_U^Hz
zuVb;XqO6{S)_m#KlhsXTHSKTf8;`B>DJhmI`}-b+j^|Y|>NIc`P<vgK(}nfMEtI)K
z#^Ia?X+J0<JVHwq4+(0%>-M2dLMXFf3^0P>NNzmU`b(qC5^)*`U6TKxKoBtCL%X0k
zz1x4cEPB-4()?&CIE5Q~)ukBL^j*JMqN(esnbgD8blhCIq3=HEC1xCmQ;d?Ljl)8*
zoVF3H!v_)hX~=;Q&|}#6hMsPW#cY6lc5-B0_J?Jx;P^oz!p^S|cedaAqF-YtjfCy=
z>YTn3zksQfwWo0KO{Xftfm<RZ?vdT=J2P35^7|<kiFDb==h>@WO<l`YjWc2uo9&wN
z65Hl$ha1&v0tGnFIZNp}B9v7h4MRKLEb>{~n`vMnw}*4tDXT7tV2eC*P7Vc0{ygeK
z4}9qRX>)WgFJ1a6kY}q~C$@wzLgC8S2wueoVB-l%w~)dk1j9W9v7*JJhK&I?nxvW<
za0euG!ar#?A%G?Lf01=oQFU!il*TP^aVJP{cXxt21ef6M?(R--hu|*3-QDdFoZ#-R
zz4?3eQ$H|xVUW$vS+%NWRn4!*p39hz(5On6Wu#I~#gGP$Hc<)qNQL$PR9uw9lA8T8
z)fPcolpZ7gL^qTM?K~tTYhI<)%gUTligeSpw~gI}lLd+9qPnx4(`UecqqMQL*Q--<
z1jRRQ&Sz+jm&6gg+j5l2S>G?qgt}jLX(GJWc|qB1J6Bm=-pyEq0f-4~NJ4&sF1qL<
zhtS^C<jT@q2c;(+B!|aFYAq#=L);Ros&95F3LEPeZO`mOw|(968izWlUHJ2<C0171
zD0S_oSgv@6(Qn+my(OS$f7x@Awk|@$==fW-05i9mc<78N?r7DwZ=^$#kxyMPLBf9B
z_0#!Q3iW4|?H{(k6Cup4q`%@*9VkhoVip$Gt}rP~ewp?b03j;*JLQx&eZ$v{DP7Ac
z{)V5uS@BIoCIhEeD{--Tw(f3YRZ|7o5QdPy^-G>Qmh;&p_wXkw&9>zq46q=_C(m~^
z@Uv&*&c?sI{Id7%t@Nb6{;c|>yOodjp=3G4Yw5ytSi%VLVZ)124;cARm!){SfC&PR
zA2uy2*%<|Rc$2^ek^E|t%=0K|LP{(nP%QXV;G=ilRoO8U{KpLpU*RIc00%!wSw&TB
zwW(p_mvW&E;6--&tz{&FGHZcW#+#tPiYa+0#vqNuew%g5xc2>$?yD}6b9-Zs#ldu-
zDAMmD-2Y~H0G&NIS}M6u1?B*f`dRsb&`Nd#g(41=5|8oQyfJBNT8lP{JrajM+2(RK
z0EUP_)grh22b#yUK9ht9b4GNjWe_vw&Q);nm{YX;-r6q5@5ZUK#Z*U^vp29q6tjbp
z>MDGr=ho748X!cu=ss*}Tj9nHds$0nJK-2TbEGMcMOHCwd;}r{Ucyl+)9fV7ie^i6
zHGf>?{<;lUf!3MB;qPgaOvd=xfBxsafjJJk{>GwN{(PV2K&M;&43smJM((69(Oi8N
zd$#}AR^WIj?3{D`?_5>p5=7V&>cdLLV5<jA)sCI-gWC~^mpO1bO!oNKq=o5>W_yuj
z$4<>gZy~%j4icB41HT~Q>FJ5n6W4l`k|}if`d`79PEcE}S4xddmXo<|^rhP0hl%Rn
zghzPeA!)6##ZGRN4VC2x2Lu~lz?j*ykB6Jx`p^8;66a9N_2<f()rJPwTx%QKx<Cpt
zJmkD^X>`>AQMvp+3%LdG@o=~9J!Eh(guL=11XBZcRx2$q#egx$c8*^ILEhKJw9BRR
zC*nxRp!xT?yKdWc4%*zqd#rkKDP6s+WFP0Z5}VbZ4zr^?4Im9Gi`AwJbzQxSUDWlK
z*}L9eeWBGRAKdm<)253-oEL{;d~{P2<@$#!_npSUf$(qfw}m*`_|7;YW1*ci!c`Ez
zzZzs@gj^#1yy?>$bcA|T&0h$_tduZaO=bw>oSg3br%s~D^dHXNOEi00JSBb6wco|>
zAEtw82YIpQ%cnN7XTJC%0^*FBa1VDc$+1$)8o@y`4FADBD~9tvy-IvA;5Yr{$B!(n
z1W|m3EBz&R)E_W?z520Ikz_^7M?;1Gz-ll^1^1@`gQ~L1y$9b98!G<nQYqAR74IXN
z|BG${di;WqL4x`R=BU~S6xcRurlsPnO5m&w1EreIq{ppz!$0@3U@`AXhwR5iGBGtk
zS|Y0TQT{nKPur>86S{6D;110ZGoah3(B<xRagw5eTHLLy6V6V^RHl<9hks%<`Qmk6
zY=)g``t<Es5^qBqMY)w5rY1>~`jVEAG%!gV(cV$>s6UqI=%O_nG{a5t>P3YyL=2hb
zm|~7N-;2xKTP6d!f8gKtusA;9fE7VZaa)@(_h;x%?a94P>=vAKa1+H~;Z7i%|43WT
znxttiPN5j`SH?TI?n6V|iN|)WGk=PfDh)6}Be0N1xB0mcvQ-MBsz<t^pvdtBR;8%)
z0DmQ(Jo^aPj&7(s#ZxFPM=1IuVl}+|E$AOu_*)<QPB_gMu~J&B?C<Ix!sQ3Br!-MY
z7Y5AowU9E>s2AEvN}XFxHrgWQz$0kTP0v^cDM$S6A(NevHh@i$S~Wb0#QAaGrlT{E
z2BRQ4EXouqnIL7XiWXIL?>MODyXOst)>vl_9SXH#R7qLJY|GFb;{ZohaXAFcTRaX=
zzmjHh+KF3DHh=u1$hpZnxV%p7&O*Au52ujF<Qnpy%w$s_d=Y_Dxly%CKkt-KkLzeS
z4R+#OcSZ+kO$oeMXBYqpOtd(uz*~fZ6X1-ftcnRliO4Ejy0gQfrHeHm){+fh8o+jx
z=kv=|Kb4@lnZ<d7C3~hM-O64J?5?rKCuZ8`+({;BtR8<3hZIxHLk{b`uvC&n8@uDa
zu?Bq0;~+;&lD+^(vDokwt;#bk6kNbWaC!0jm6~$h1WEP-t^0+f6f<6oMRXqsNr22S
zGR+cf7WKzBa<N0*E<^^3=&D-P3+=S3qv)Zh_|^JZl_rB(lOTVTRX!qfIl42i?G?H#
zCK#d{)_yNTL1w%_<ltTN444B7nc!|%&+oa)`4$=0LI#LYQBj?}Vak>fm%mxBf#CgS
zmaS;0M*QCoWQIm9j{Cq`ad`h@kRkQ71-K&3m6_iqktetm#Chk*PkEku^KazR-y_uO
z&T@zxCs^Ykicii(Q#8T<PO+NrxAD53?a$oZ1b_Q6cj_T4!D(4$a$zQ@EkgzIVC0VD
z{^0k1)A1jDj^>lnn{m{w`~bg9I5<97_FIW+;KyaA-@tD+0xa0PA4I~29B2cT*f6!b
zX~|u=cWZ|o-Rg#RKXv2w@tE%R-5YX9X+0t*bF4VV1~%MTJj`80AyPCtD4qN+o^|u{
z#aMf<{9gA~uS5wH)ms@U)8o{si?Yj?AKLtg5_=1+3y&s)aC}Aiac;`%eS=yb=iT{Z
zGFYO9|3~><87-l|@;v@lP`RL&`tUZDY1vw0>ekiip*odPbn-#_=GwHj{x~C+kVb#d
zOQy$pQUn1s>wZV2%5r+)B4Q#)jPrXxTM>y|edohWAw{<Ic!-rScu`WkW6r!?Fw($H
zvWRrFQp7rg51hJa)G9YABMJ5A*#FhYnEj{B3!k(cw7&OZ1Pm|TjbCr}qwT3m!>iJX
zh%5)0-|XToC`VHzCwDDVnXF8|yiX}-mumf~MN_Dvr2Q6$ao@wio;(K802WR8AHVq{
zFp=PJhh4KXLmI1ZU;u8$j=psLPFu6vft{tGyYzai@WCdoDyzKh#HkZU$VTAPchzwb
ziR=JOQP%@zOJEEwTM)p8ce>_GsxncI;vze}*m6ByWjmo5s&e>(lanS%5k{HDo(^Ep
zSPQHkg8|*fnJKL`XX&UZntF80)^820Lt7N|F<&0rM35{u?Elf0g*l->GQ;d2ga)S-
z8@X!XX&AZIo^#Ce)a|;15%H^8B&bD;ME!wTHy>iQzfrd{e0dW*t-Ai4{Q%2(J2fki
zy+6Mi)SBj&xX#J`!uvE~CNTfy^5<<nw!?yw#GRT<8%GNUE+&8k)^5I8tnqEEi9>-X
z$#%s%$n_H_7%kMmnQYSy0?jW;Q8S6by~otl*{nAGOna*|<BwvU=0o?dbHXXOqSodf
zM0<8Txj$R;zM1TVUB4P8ng-UV<CvFaOn!g8!-@f&cl&e4!@*<-t3lDpQy}icp00If
zejfaZ$;qofshx;aYI$09t0vPd^BN-2EjYE(4ok=v|3dL?iH9WT+1xKu8|reO(sTa-
z^cZ6<0>j!<25F6v-!>v+2o2`fa~wy(HJHK&TMl_;crgWVspu+AzvYxDeyK`P=A<(;
znqgbG*!+RWbD`odW{;wp-=6=r(l9`y{W@sa>$ds!;B?lwbiGc$S(H8w@%-<ZWV7Rp
zqN{!n|I=m{uwu8<pJg;IfD-IGjBKKdlr!m}Ah2EWlj-&%LtEZ{tk-*mQdPZL=luv7
zXY9_>5PLse_shbB3FC+!ZogDYv?iPg%vdHX%vV%KuwieudE@x}+PRJT+&7THIHILD
zHoE<5V7q67Ba9UFds6K-W_v#O18T^0+;de!#4`O+Y=%5bv_zjR1BD<B;Q(UN-hVcw
zLf}OSOrHOQt3l2kGd+_RW1d0<HT`ht6{>hIj`CNj(FJs+vU;WUV?LTn5VzJrx?Skn
z{h8n1i<_MUbx>5~f#3&UtoyULHym5jP`gwe9Wt9-f82#%d}fy?E7*{+Lhu3K0k?^+
z!ByAf3wn_f&%e(8l|SHLz#GWbJNBCF{4(iRyTYs}fh1O*|Hv3jp680Jyj;9uMmud9
zFWEWe3|0hVZSpCR#jgkl2$wP)cr~s2c<0?8z8p3y1FZ)+?E@lypv+b7VV5Oc*q$i8
zS3@AZ7f|DCg!<<&$i!}}TuycIoAVg=RyYgO?!)aP_T$Z_dt7tcZ$rfibE{kL!J`56
z3G<MFmSydu1V$93$5&2s1#jS|xDN<`{9RPY#2^Nwxur^X+HAKu1uUOR!3C)6I?s}B
zNH9`qqnBFXcoY5cr-7=??s1cD#?!i*O0|L<KY_xrc{ajO)Q+MK1My6cvNxIYoz6Lx
zA8qo-YTo<kO~ucU6VzDot2SD@R9!PS)<F?5yPC>9NG=_oGcLb-_6~2<8*y=JC~2=W
zKWW-nj>$EGT(wNA)H-tck9S5z=xC-oiUz3UXuOCL#A%<gux5NpeQc-@5KzOb1s48r
zn41Q#6INLhE?{a9ToRMegN6AE8Xuqj?%|CDx3MMi1Dw`}#(&<kY&;HeCJ7)+$`bIf
zu5y}g=y{#py`D~Y1$*wV6^IS1s_PEoGx|X?V@PB=%G+$cxV`%W1x^95Y4ITtHKO9j
zneLb!Mx><=7;R)Z$)|WF$$rCed|Ef|<LtVl-+0-0xZZep-=sq7JuvzY#wUA#F1%^)
zy<XY9%`b)?$Exg>F9Lc?m`<W5;SGiSdDPlZPp!6)#8QMMo<`td+3IOQP<JpdQs@4e
z^Sa}PVMHc4{`+sgeb$cw_9dO~Kq$Rt{9pl`XEz9^)MTpJXlWKv(7mq{bGI|BstvdZ
z9G6W}9j3UaS6?p7rhjjB5p7OyzD{o1|1}7^N@;%O09*1t|MGc!2t0P@iM~UB^=^Qm
z0AEFPe1lae&scj?i?Y^q*NIP8|Nm80jl1w)owcxy<<B!5HYa7UkmA(WpEL2ll&sqy
zas-C5tMJdCdy!jq#%Lp$uNz{*uJ$>*)2DyFDTrbV_k(zwuCosK!fG~O?>1O1Zfr6=
z#<e#-9i=6y_jItB^J8+fXlepmc;aYY^O##WhGqWS)fm_7e6m{ennsSo)Naz}95ODs
zOsgF;MP}C%L9>**{3WeUmfMYNLwqD^Vx<U<4Y`0RY{3V=0<X&7^R&$5^1|>^dG(V*
zyU63h;+R{~7&&T%C6=`Mbp~^-{mho4p)++crOFNLlJVnN?&npc`X||TzUjXr`lh+h
zSGo(h$(vrJrzmg7UaU_QqbuxhM==&#WNajkXpO@PJ_<M4ZFZ4S#yQ0TcfcQyn|%Rn
z`6=|7bvqI2lwC@TRd}M8zr&v-)0Y-d;%G1s{{ai^P4F#jzbkCqDfR`qkOz9G7J#|+
z7f4fywnuULzizf?w|94Ef38p7Oz|QYp#vKb8T;ygiA0>)f+7cJA%<`=gvG*RPinez
z05A)55rza|y6)*T&r2(Jv}X2qzBUHN?}4y{Ex8ZfAKiBzy{0>tZXTLV)=z{0{AXL;
zcTaun?Wim@GY|tUk?S`Hma?H+=67T6TqjPb1KhOYV0@BIKSY**N;OaHj6L~#?xET3
z5vA40wkF`#8F}T2k}*0K?mL0sarkeC69>^9u#}y?KU;mtBHp=5`=P*(dBG3hvMIrU
z+;~5dwt6eo_bB!G)t$RCVJOW4Kmk5pQ4jj;YblOxt@K<CXOE=NYyWL>;Z^tBum9Yu
zg38^QC-PZf2F5BG|B`yxt~L!CDb?53_L#G6_rekx-|SBoCOzIa+wE1>=QPu;t>pym
zTmM&RTYWs#0KAT(X`49@2ZrhHqNH!>q9+u2B!yVo+TQlED=7>mpiaSe5()kr$p#jH
zkeuEyfCfeRD0(-b5LQ&lYNUtVnE$0HKhiCi-TSmW&gr3v{?Kqarn>d6iI?jdH_U1D
zgb^mM?3z!J^ZWbDB&Sn<45xf<*w}{MX6M^~IS#w19ov{STO%d>NL6&IMYsxBb-;+h
zEeTjkaB(@FfbSw@%T4te9Rq{n5<7_M{BH-QzftTnhgr4@7LYV}Z!*NpdhJpx!4m8H
zr9R{#8*tBqDjPpnIo5z6-(WKi9QaWH<v4lr^MMD5_7_!aA;YhKhXsZJFJ6bKt8(>D
zgae(akXpQz5aLRMKW_4xMq%O28O-sPnf#v8f7=T&D6M&NQc?;XaqPUoDjvtwyC$?Z
zpF!Jq%6%Hs`wR3|T1M>Do=G4eE}*`S=Dt!+@!iOtw;m7@x^IC4aJqj2*f!D;{m6n@
zW_&xi)@v=F!>@e6qvQmcIq4^PFnfVl{UR$RQBLO(#LxY%IlC7j$NHP<Uc}dMywlRq
zRYuF;0t0*#1_Ovw?HSqvzNBsCpWp<Ke^FBxaW@QJ3_kBSdRzbVhil(W+uOu9Y5|NE
zxdF>WPeyv;gNhJ-zR@<L=tb}2MD@?C!!X)IWr^$J)#;MqD#p$`sJZ*?tmEy>4t9&2
z(PNhI!K?p{^SbMyGtMZM`_qU1&Vz9CwwP0Cah@lW-brtRvb|YtZdt6@oe?7}VQmO)
zp+^H+XY4m@n937KClJ^Z39aSP08P)!3aWDx@U~^(64ra?4>DorMZJIJ`<v0lzu9_b
zEAae4^iTJ~*Nlir%}uu%kup0~72b2s&v&eh@}@0HZ<q)etcI|k;x=Q=s(Fm+_D0rj
z@{NNfboc+Ha?#Pz-Fn3meLN0-^0?-=@W*#1u)wE7`$1QsGK~L{e#nC#-uStY%yC-(
z?TP)$>qwS;&FjQOnuU$L-}S2NJOV|fE>OutAxJ6F7n0ILCYZmtU6xGsl8>419pA|}
z3xmBigm1eWw|Qr#<MwtSWiTMa;LiPV@ls<vPd#iHv%O+fvy<)jT=FS7^l=YEW9#%t
z2wbDa3cmlvWaMx6IajvRbz8@NihAe>WAwes#f_oRvG0ZUzAYfy3r3Ii8rXDuwd;QH
z>NwH8*zPBKXVQD0i|GKo(sqk$yhGGGuQB7wJagcbAyktWAP`FFNXRPHWOkvRwkRvp
z`gZ(B$U^vQrS8yQ^qI&GPc8k1j?6CaCN~F%!^<qaPb6t(V+o4?s5BY&li)wCOzQ1{
zF1%r@Fer2EA!_{IPBb+%^!f0Bw3-7<MB4l6%aY?Ld1_umpPUv9`Du<@PFwlqzlcq5
z{){Vm%33NFcx()#UD(x&zL+64kRw14FsLf3whBGjhbO7^ee+CGUIYAM!QNmRtE~W8
z%X~Fu_d6q*s+!Yoo#5aXAjeuO9n5=vWiQ6fGdvzgyH6`MJ!U@o8NTwBNcDYnL9^$O
zBB%8qLAwO0ZQEUi^|CLesxo55t9M3%AktPf{i^#%)5A4y_YJQm$^j{}c-LKwcBzZn
zwu^!C<*%wtyMyV?HwNF22R|Y!L>jT^!1iVvu_94bsEx$I^-8^N!UF;~oAthji?yNJ
zKh`S^@L*!2v&w;3v_-WD=g;r!*6l{DDNq~#bfZ+cfA*xTc7N&7v+m0N<c=4{D+&PH
zAy2*8Zc4P@odcvT;NZWly`O0>U$0LVf^PD$JK;DT$kmtLFQzJIgF&o%2lgtz?{KP?
zuX^4VOs+f6v_BAfoD}@j7!?)Bl9=5$m5(B8*6jHy#G*MGc2MPXAGn9WWTi{v|F%83
zzW*CTb^>w`@R?YNRwNbf4>cR7v_TC1pf-r2QtA9K4AR+Vs&6zR1@tug@o`FO&=u&a
z9DuRHUB>yp+P-EF2e*+n7g=|d1M$P({UNBxYt3Lisomb}ZZ_?yV^jYEcar2vP>pu4
z^;-EJZU9TKit085|0=h&Nxk>fZd1T?O2>It%zEvwzKllKn1*5aLeuf8iOECCw58TQ
z4jM@1b*ScpLj9))`R3cX+tbj}d-CV4RdM5fGp7DZ09Y6~=bqvmV(x3gUwA^#mjJ|h
zofI5gjv$?{J~z6p3a|&I73G38(kR~l{ooYL3Q9}EM+9WurwpD0TsE4p`uVY~Y%{K`
zX^-Q|!R7!QbU9Tc8J)y7JbjhqoiWO&_m5@X3#RsR#}7Y{me&PfpG1);MG_yMBNd<+
zlmtbFA_Of~_#NzGnjtUmA+oOsyC~8fyyi|Bf`nKVx==>R(?6btJ|_59Dsyy#+pDJ0
z*RQhifjcpHM0J;f=i+8}G^_q$*{9~(cbxVbYBzotJ`ly|BSUj%7}eyYcQQ5UB!P{C
zqu##PYpP2MAI)g=;5!ee>mL9-`~Ctn?AGjZznke?8zZ})hhVP_@I#M8lSF=N0MTi8
zCGcK!B-Z%eRos>zrGr;0<5ezpZ8i4go9A%G0j$zQ9615Ko-`;Mp(osz$i*tY>z85w
zQNW|!cG;(XL*c!k;h-GKA-VZTF76Fz5){_ep^`w6ibv}MK2Q@m{P<Q?^?z?KMgjW|
z*dBmfN|h`f{gt;#M%d0qhN$J^{4n<X@ueR}`xjWUW<vY?5S+g7cSwLdg7HPYdi&}i
zXE(c*wY0q4aJQ6M$LC44@@H2PVPhvXMam12ckoGdd#uHLF|+f?j5X2Ge&Z|^?o(E7
zewxPpikV>Sdmpjq!}%oO-DqxOWAd=>CV^nqXcmqCfot{Zc2rlq%CMwY_d&MPxFDhn
zA-rO$4ap-cj*3+uQmD(*WN&j)9K|UogC|5Wb*)4)I_iI5CNQp0-Ideq_kbXkm(FAW
z@d-gN5|0BH5(^aM|Da@OKSs;(d-w2XII(Wy`yZBe;H%r3@Y{$d7+N+;X^-F+!qS@u
zZt_Zg!18L%e%=M>r=X_~KEj@n#v1*{y^CVo{pKpZ7-zT%d2i6O^;OE=M}kLIHkZf<
z306L%XL0Exz98<6ZrJyqMl?I$$LkG&!>+;S@2D2+Up$ubdkC1%KZBmQCJAn*u$Vq3
zF#t<a7=kqjj}(!%5F<PBZRuP@ep!l?bCf1bW7&{Zi@;bMH5ero!$h6R`0=78#o~Nc
zfqCxpy5HP27W3iqRj}<Gp}O;43#G*kymstN$lYwFlK+15qQyz?{ltvZaYh&x{PNRi
za+uI-ra(Q1G)^F}%jZQ|UH#x%?47?Iu$o-!2t^Bk<2$tMe5<lM2=PjlIAh5Ijv|!(
z8$wT;%ZdBP+r!aP4bi999y>dGo=$oI!^O2Am(D8PIC>ZY?clW!O<?)A2w@sA2K6s0
zMsQ6eYkwC_(KB!9b^A1o^IEPY1a~W)x}vaRgMD5jDD4*)mo5dnpQzzR8<%tj-KO~Q
zsgVr8D07_&Qh#3OuK5ms-LrY}KxpYum;U){ZW&$A=`Pv(244>xQB{?MGTWUlRnHZ*
zdENIV#qZ*Mq|44W<NL?E4-ShvSWWjmw(IF;*la;4@3*OY!&OQ$xT6!QcBMd+zGSwW
zodTQvtE1RBwl?9a^S^Y~)D)#SY3T{%N+~;W+$pMVn^O5?Sfj5g9&lfb#A7JJMR<64
z!UN;UzhD=jM<qr2y=<p;K$J!rnnhoz!>bI`4d;eqq8ZT}r1j&-VC?btkP4s`-wvbw
zyvmOe9$u+1qEm*^pdC5*Lf~{-r(L=%oU+jQd|Xinas)P=Oz32bFbr@KlBCJZRpKmx
ze`!SfVuk*DrCdihr{rqmLS5CNpU#!sQmGwJk>kI7i2pYWkPO0Z7@AwwETmD_ud<;G
z7r|oCfs;0DFxct|5LHUTY=C89`~Z$}D!`@s>JQ{P#4;%V;d4X-DJ*UghD-|GZ{YX?
zh@N7(orZb&@%O1co$MUMhU*Deqi7jL3iz7NdpsaEjNr&v%rTy&t*)%N-z}=^qVXf*
zvKWUexTaVk`^(5~x~|%M*0E5+JwP&GmFk_605~|Iiv6m<1sa1texOSwTyA(>m`GDK
zZ*~*eyF6{+m_SSA^K0zqL^KR(6>ZZ%!iMx_uxM!L>YgQJYO#9WngixUl#zpm#XQ=b
zp6vAW^uY!rYpqVGB8qfa^;%|t&$Bl}UcHL5H3J6DK{HV(caO(o*pS@mckn)*8&fN^
zCM0+%H)|~1ygf=a%p<z0ubCd}ZFje&NRpCpPN=M=ofFZTjCFYi>{~teOmHLtv;7kD
zm$r~1aK-D)F#xT#_M}vI`C!fmAflB25VDZD$uhOgabmqIfT_Redik`h0+UVP+FEUv
zaWIoexxB#JbE8G^aIWmti2ZnVSj%0T1`rnJ?rsdqa1glHb54wofKth=-)`uIAN3V5
z$=0Wy1%ZZu17)t)jD?>Hmz3X7_w_gWt{!fw4BecYPM_Acvg`(}+{?oel9X)L!0GX#
zl|^=O!&jCeeR!!#@$f&9;}j`W;eabC78vi{MO+h>6>$>DZ-m*F6}(??&1FjNskEj=
zjBVh=tkBw(=#_FfZ2H7OVMt=-1akB@nyuc{4m!rqJ`axDSCquTdw}hn>w|*WX~)eJ
zSTupE0>Cf>UiznNq>joUP-#rynq>vZq2*Hv;j2USf@>uhnneHxVCiI1Goad=Dagb1
zc1|!_wMq#6t9){my)PZyMlc29apF%L9Gzy=V3rdUdG?-dN27k>YxXO}C1>^5TgySn
z0+PUoqg5wXI0%N>VOuniIIphm*Z#g~3ZtI<#YQLK;SSIWA&tvFopDX1O{tp`Bu`_3
zD=3aE31jA;)lx{Tdpua3`YCc1(IBN-)xW$h9}5L(@4Iom!nX3B|K8mC3?FlO?Tj8A
zye=2e73Nn~#%$Zqr)GM;*X8IU8MAMa(|8q;NBTT0$3apO99oTtxTqpi_*|pB1Mc+o
z{r~aiCE0ycnuY8fI|+GMXO&+E_~jI~vlrg76!FEEpu>^orma6u3)b>hnJ+cYR_~i~
zkfG{$EUUgAXl}X~^H>4yo;$7Q85O{mqVJBcExo#qEA=M-0|CTQ78=dcva<(D2!Y=z
zcFX-S(N%@s+ZZ)<5(HuZ9qmaaGt|#Xv*>?qR@M6xcHY@D717OYaXJI=!(Yc6)jy~*
zFZ9{^innqcT*yO+rIf#gl4HZ{_JLK3!}kiKQ`UMPy-fJJQ2>_rH%^*CXb@F9h=<Y#
z%|{yMFJIEvuQ9kL3BeVDm4yu1U=&VQbCS$I66&C5gjE<CHIVhc2NwzHAg(1U!mD^(
zN)WdI@-^3%Kjh<eYS+Xmcl~D-*b0#L>d%UegKphGT7~hN_?3l*mwCOPo?&1B#RQQY
zv(xj=5b$vy#dravIir7lA*Vomq0?#z2VQEl5VXcx3(H&kADdJW7Ap{4*X<r7_!MrB
zmW-Snk}5D8g2x`2l7a=4K97i^A1z??{%9GB(ryED`qbOI_TXi7Q;MR$@*qMe^8Uhy
z_AJ(~s%@}W*wffND51=-3z7*@gKG4q0b##C8?d2MZpk=)Ipxi%sO~}~u`^v9OXhpK
zo&S7P4WoEzhaPZP`?tF#qj{awr2n`!ou`fkYXA^{S&m2O%XvE^zT*GXr%#rdTAWi<
z_rt2>jV<T^$n^k71>`5NZ525$IW#~T{|mLxM!$LBdDfz!jdPIJdYvYmRYw=zcI-C;
zk^{P>nNNfco%Z|~x`MzN?qg(eF@?OqlW~$VfQcCVp}-drP4?Mb%#O%}6E_&~lEM1T
zzv`{>MimTUgI1U{AkM4JRsq-NZ&((*(ja-xftI=QoIbG)sj@4COiHiDJ)~=mZcIEw
zAsiCP<(dd%gXENX0Ff6$F$-VQanJJUlC88s{v93tJ29M?<ZONYcUqt8dFAAq{QA=D
zd{bHj&If|E(=6vx3@vf(nb0LYpO{uBGspzp;pn}>#JK6xq`8POfZg}q$cXd?n^8mt
zMTEua+`3O^WUDub+hZ&V(Hm8S2_P=+Pv<Jk2YTR;*J5R0Pf^AixqfY?)#Cy!D~l;5
z)vcAtXO96%W!mc8Kmz9Y;bym>G<7h75F8!DQ<tp~{sqT`Fib+gcUY&fqY{&QFA=!&
zt`}RdjLb|6%?AGj7a`@FzNb<}X761|d7h)fXeqm@JpdJB2&AqZt4v`BKRp8?&?$Ko
zCtu&bdK;ptFeq3f(h$uaO{?#bHaJ6v?Kq3zySt`ATai!my&xu1DujaNAJarPgW+lk
z2EIMGVUZ<QPL=&AnVXvnOuh`GBs>TwjU~Q9c4UawXtzZ|=D$S|#`xX2xbkyHu=lXe
zsAD~E(0s;R$UvIB5b>ArZud#74b5rswAu4DuU3{a3N|~xmDSP0iBPMX6R1x;e9yD5
z5Ed)|V<8S_NQyFB)X9nU%RW2g%-llHUh2;Y07)o@At}%QBpQq&n=8N%DW8dtugPRb
z_=%DP?EK}AG-yojgU414yf3*WgYAi{Yo~ne#;7Sjm1b}(D<rh>wPn_)fk*yvA*emG
zZWWr@+13EA*yUOSNP0P>YN0v~6)|p+vL;_dFp<1he*t;4a>v$*oMydT0Vu*!a|_Ec
zy3@+WT}ST1qzx@jA#FMBmuD758}d1k(z$8<gap#>WO$k85E!jgLR_f}A2;zJ&7;Uj
zktzRro!&Bk0G=vX3z(vR!<=cZjs@70s5_V%fRFP*FYJZcMu!KTcGE-@Ectj!wUG4a
ze{V0>PeuZ;T-#BNDL(gv)3-<SfHsO-RaI5_voquC%YBi?DOn^$w2P2oKDw5+wuj%9
zVIXn7o-REl!Nmv}kO45+6TbsGT>teRt!AAew9~<qNviI!ewEBv(eVMs-)Aq<Mo%fs
zC2CC|89}mL+Zm+=zVGa%>P<Z_0m#!x;0Yt{?_D!8GS<kOQUD?)D?oUbe&=`ew{UZo
zaC;^F4mk$7pb>Ss3H_`2>j~BMVGuH_+tEwM{XOD-7k5xy!SU0xrrCVGnS$qCsmWMb
ziGRo!#*XmZ_g#vg-tS`ft9F9<S*y0?cqy1hcd_>|7}u(?1Lxi9BDLdbb*Bs6-9|@N
z3bOL}5GTSdV~=LmWWi!Pj+qVw?LQD95AHC7aU`Z8XuvAvDqowhU{aljz*aOP8|}=E
zHg4?f2fl^{fMWu7c4<9#OS-A#gxbI>dROq<4tTkM%@S#L%$7@q^uEpuYB^uA>eYh9
zD1sk}%)dWs>BZ}R2J-mp2+DNdX@1;gTdNl)EvK0PSjWjlb+eav5#!+hs37xjnMdR}
z!expW#`Bp9JFDn_5Pl)iXmlj!_q-JnMO7}deyiTaZhs>aN201BRLT@!<<J!i!o(f;
z6{t0O|9$ukJ4yv@mt^eLw>XkS2uhxNH;I5mDQQ^I5w4iXUj|wEZExCdKg3LL3%KD+
zT7Rblk!I)AE^9Jfk2ewDuH)A+AwOI&sbAoD445(cQ>2^0hXTP?s%B-=g36-(<v<WM
z27-#Js*<|;a8CdvKoT;$3LHdS_n_+I5t5XK2BDs#$X6v!2@gdR^oI!AV-?ug*fcaX
z{Q=Y^K&2gC0X6cyzYKHlEU?0v1jnMQpm<yRgrJJC{mzUlvXofMf<t{&z_LalL4YWW
zHz&Jlg0!^xMOfFm@Ah<Egi(HTd#u`{566s1(1Ko}PG@ws!hyx$nd_~J)Y$21R^7{c
zOEmH#qdwO)`)kChmB1hk4VJ@B9-B@_B&njf*i6^e$8+({X!&}v!H({2aT78chdb2e
zg!7N(9Zx9Y*U{Ifn^ARDv-{B1?>GdrVsQU-QVZ>kxS;!ZgPey2&8#y^h235^#dj%T
zgixp>yFiN*%k}Oa`>O-eES9qTNKSav<nRCiY<F=<?>qbvj~X?T_zfWDvY5{hS2oIa
zzA!A+KH@1$1~d&Zf9D%#3eU=xH*~&#Xu`?di=y>}v02w2MfhmOxINSiY8qOmJBGsK
zA@a>OFE1|#`(v!$r5J*#C_PrmX%C5I0nH<&6J}P~4)RrI227=^J8sb;2VL_;*eP?^
zLVn0Qa5w>s0!3|}`$)Kclpm45z7b=Kv$3%~8dr-A%d?wQRTw2w1iAsaK5z<Wv{->(
zkxw<2-~vRA=_y*d^Rvz+0X7b{f~4&#+>4~;OEnhz9^yixiNk?>V>q?=Y0Ouil!F|f
zoJ0lgUtF={Qv80has2lh#f9Z%Wuw(r8K;CS15aq2ob3aznRZ<f$~3%Py+N>MOC(hl
z$)U=T0FUc4@3f=7f8fk6V<G16a`xk?gG(YBN=}ahY2~<Y!~gWy!>nwq)M})GOq>_C
ze&#FcmTDHX{>ucg1B3{MZ%>z5wTbkP7uwA&Envv&)Lwkx@eU@<jNi<GIF1odBCdsq
zMVeGEi;hr%w9eXM!)!d*&DPPU=nqQJVIY%26rm9*Pg~_e#)n5CA(e;&+gkXc;Brw~
zi&#T6M4qZQ@mWx%&}oVg46i-kXmbI~BY-1-JcJj^I~y<uSbUe%)uk&NT3G)Sy6F&;
z<+4YU6dwI2?3+N@yh63^G@aeh(7*(QH37g8kR*7oPljO&NcD?bThj+9qL*C|zlS1^
zS!xM$gYR0R!2@-txt*QaV0j}OC#O+mFI<GU{vaq$7*1^TL7QeI5$DSst=dl7uL9Rm
z%dgbw{f`d}vysXY@YL)gR1L8ajvkRk|G{Q#t(1%`O|`t7ml3%2)j<gd*P^&3!NqG-
zp>AO9ZN8svYIJ#X8{jws)C<P}REd@qPr{CmurP-ShfKU{ji{?eV>*A#*ZYkZ>fNkA
z=8u;K|0S{YamQ%<1yVN67JCvcE5N#3MmbU#C%P`0NU#f$0@rMqMr1VnSHDJ=A@vXr
z9)8A=-cqIz&4mUqt?=tG*lp8i>1~4y5tnVoZ4$4j$MYr4_wD)e{pF$nVic$<KV9{)
zvNcg4bl<BH-C}j=y_>P9Z<tlhO3u7j-nBaVR&i8N@?yYAM$}aUQ21^y`*AJA;d*-$
zDSnyDp@g1$k;IFjQNpu~1SlUSh<_1Ap;i!j!oP6&yhXs?J2@?eRS92kjdYMR#~peH
zQ$^(1fzk;&5jo2r?C0}*lru&Y0oxVW`{Mx7=ikTJ{fsToe}?hto&4$OP9dKKr$QwL
zxIPIqRO`xh#h+yKper6Kg<y7Z1m7-~F-WcW#Ynoyep9uCCsx!IV@)RFuxkE5z^xs(
zOaS1D!cd)GTxQ@w2(D`u#N#P7#%FCvV01bSX*yj((<?kZjW@}F{zZ_du=!P^wLBTb
ze)3t&i4;!@wGv6cAd*04WE?1|mfpiFHhWvW#7ob$D+sDpTbistb8rd*KC)3CI6IZo
zpYz6DA8!{oz$_e+nE}Ns&xRUsV{^D-EL&apFu|O+n(=&MxBw2^gbKu`%l?!1hE21`
zQ(aktu~&FQsM@}|suHinZ<Z4m;Z}jf^{J^VJ(@PKBq=fNb4M2Hx*k>Os5%T*2=zT(
z-9P52Kq-X_NQ6W{^*HT&U!Uc5-s=5aTJZP)0K(@>WdC7zKSKMw+)p0A{u_3M6aMiX
z^6llJ#7tK=wcpCpTpv&wZ%=1~+Rr-@m5hd*#Rx0+>I7c{gdt&5np;P~6dtE#uo+FS
z^2J`4<T$}+YV-sxEiFsS%6?Z@#{xg(IBe6*tUvq}oS>)5<;nfy-@gM|-33R-o91Dj
z8F}gmV)5As;*ERh@4mzaeZK~HIcy8n<zn2loDY@@+Q1+{O^x-|i<X}E=R0H3+?+y|
z$DTp)m0Iop`3v)2BqfueLo`sQzC*hE4N8|=c`1#yo-Z~NVf7xnoZvF)4>NRc&XMT3
zADOH<k5@4atCjBmMOWA!`e;iAhlH_zbr{xH-j*A*_|oQDzbdU^Re#`dW3&n)%sYRi
z%Ew<gj-AZMNoBEO;nZ-7Zt*&APgNEWp%o*bqypBC<`(8;8B$55q;F4FQ-6Plo&RGG
zUcZ?wwzC$>5b@GeucMWb(RzJN%k_L%@X2ivMnQSb&?>W8xwp{X68X#Gqzs>)-8|mh
z3Z+#Ff#88Gf)f=Vs0d9(0^jeiJh~pHT(<inrfx(=W$}%i#387V3Ps2LUr5K}TD>zC
zFZ>Ov7!uHa;{yxP@vkeQv&(dK$Az4f(4qVF<wl<lAp|WJH<1iJ03&!TNYbL_3&!=L
zK9n>~<$a~n$rL?;V=&D(YI0HG|ES5Nr6F>+D#i}72lGIp29kLyLj(;VQ&$=O@(R}b
zU-D?d$-8|h1xP#&QX+{OBABgjq!EEx^&>KgNpNnJTr$|y<opQi7*S%*)*}ZMir;$3
zK)G(=@@#$557`JkaKSDQ*QDS&Z(u7WVf>I9#-KTID@#wVL10sZvFq_$P3Lu+lbu>5
zYJVuH(k;~jiX`#?yjzxZnXG(u<9=sI6mvuojXD@<coh7n2h&#Y`w)&w9+RY9BQX<b
zj;$tybJHX2s{K@$7=kVm2Ub)Do&tZ$dAJF4>RS_EJgcq9c_Pn>d4n3%Zzx<eBkt|X
z>6wOImxR1*<!uTA#@D@R6?OIb3U$Wvii)Gz0@%*y`(8<a690t?cT%o-Z2~~Da_`1}
zKRi7x?Cb=bQ(%7RYD;)|@y5i&AmXrv4`6RrLuL6_#RlUaG^3oPHJf$Ic$Foh%X5MI
z8_9D`X*L;hpYNhGgXYS1{)Cu4`$~tzM`n2E9It5IXEOzareLXH%%+q5w_hp?XWMHl
zFJ0?vSQb)|jWf(#p<c6(E%>2o%8dhs8UagVj53#QuI{C5l?1+Lg;vf4yHfPM5aBRd
z_fJDvQMy%iYba2|X9_e1!QF7v@%&hRKa>zO1c)6_!a{W#qHw*a5(Kk@)%QeSs}&M6
z3n7Zken9lqwHDpIHFE?@&*7%R8OJhVx7pwTnJ76KD2Q(eN1gUhpW~utUZ^O|h(ASD
zqmdKM-impMy1E!p^OzMBVu<*oVO0vlYd1ZKphG1XD${8%YnAXQoCVA?k8k21F=I!G
z^)ZoMmREgs{Tq;yk}{#@x>&B-;Bns$%HWFqG@`Yzv>eMh0?@6sJF&d3N;3FHGZZlY
zPMK!$Fh$LEn`qfASh{%I3V7d(YttP6DDV=GH{L`Jh1@*u4W`a6E9LVm0*~_kcVp_4
z!B>AHbiZcChb-tW7~&lZT>Mw#dhegwYTYLmct#j~{G}F;dYr>{$Xu;bL)^!QU!Ub$
zy4MA|or#st)V%Fau(w=(QCvHw{1d7Lu<{CvHLNgP-AjfhmIO>&86YKA)J}ZUfD?`*
z<X568>|9?g)$Q^o2eg}epNx9nN*%NLJ&@M({W`c6^kBY|RITso(M5g#<z{13b&=Wh
z&t<nBwFeUNg~dBzxWQ1Qvjq^y5HX3bUs8IsqNZyK=;2NS$$9-Il<+{Jf`FX%syI-`
zr867BJZydtKY3$<02+?om{m=uxUMce7;rQ2F?QZ!Qz!@w<c5u?z^*RVm-Qf&5K7VW
zquNeTyc((cs=d082+EpjUq_tBMlzl6n{ZBT!)c+}-!C4qgoxFn%EEF$>va_Ltu$Ia
z4K15p-RONp9<Stp>3k&OXXpPEQk6qy`1q$v@Sd*VG)~`a-vcfmLaxHyT-WcP&GnS@
zgX>=~`Xd0pNSlm!f4u#X<*}bG+i8Y(HYhJGHSv6RqS@w54bWjsfd6zljaNdeRgM|p
zq^aZO=j)Y5XTzLLeDD6yHSJypgvQS{mX(#!=TBN$nC^`w1Lbe@-gxQ_poB96lrG=r
z`CMOr`d113^YRz&5Xsg$LbfLAfPunh_@vw7yrE8qRq2)9#J{wiw4#HXU{v5*?Q2Bt
zOlzjQx5wr2I#e?bVYxLee6l*$u)oX7T%R;{Y`*oP!t}|Vdk`Z@n_n73*s!Ic#m^L=
zpguy%z{3o!POG6I1E7_m5bUfljD!kbcg%-P&KL#nZS)Rcx})O_gmS;)#2D>zQ!vJ3
zyNYH(qWY#f6lRh9fh}@={#U%<e|N_>lSc?*Of{P5kFK-;B!oT9U46*}G(g}lQ}Bs=
z{zsZN&y@;%PzxGmx+L`;sYwk^f^|@?M<9`FH&xY9hnc0!C=F2ZF5zA{=B;e_cd`Ea
zp&4~u@RSn*0Sa<)GMg?|q{_>W7x~>&Kh1|Egwta64vOX+r4|Ds6h5b|@tvjm>i*?u
z7&=yY7JV`Y*$UbKvvhEhA!qs-o$ELMsVQY_Xt1qCF{10>=w3qtCNsLZwt!B#hGt<Y
z18l$Zl_&9sQkwYjJw-U@ER?o<rC?=)m_a-_)g-4?j6fmD=*_Uk3`y=)diyb|G6;3v
zvA{m&IVWLBVlpXeaX4X@uCYbOsBg{kJOL~q8_wkIm*BE?9GC*6CngA+a2267O)_Qh
z)}8<pTYoCJBFFA|7z2VGahkhde~}Yd9N!!?k?o~U-l@yX^)CMAP7A&e0H1fXA=-`D
z=Lx0L&WDX8o~YgaaAgTfC5%}vGX{=XO%YZpG(0>!a|;V$Eh-mh$1QO5tmbAp#xRrH
z!@Aq^rOf--m^9HJbXXSV)<ZNh-@ildxUGJv6nW0zhEabpTg4r6mPozaimrF-57Ml5
z%<0Js772|cW2PdX)wtjEg9rlMZtv@@```YEe!9UnbwQnLbF!$SqW&>htUr|nQ!+u5
z%woTnAH^_MDOf<2uR+&V=tBdVq%bCWg+jvT>03OLVFR>AKxDvwTv_TG`W%KU<Ge_B
zWu(mYW>La1X?&-l-@HV~CZh(le7|0tn8r|%-w4#>DIs9tlZ)~U%PLR=`a28R2@jik
zfpyHvE^k0sQaC<7roun~;t3CV@_*D6VnEH_KQ!eEs?t5t*u=~>tVE)zrPJ+7Xj{o%
z=M7bV#T%!s=|~7*P?la<U1jYdty;f}i>r<hxn#$kFU+TZ`J)bvh1qhFIUbUfkfv<5
zEmWk4mO6$2TmkA?O4Z5XB#30<!@&mr5!qwhxzp`tb6oxl8fObO@4%}5?;W&ODu@DI
zWf8c_W`Y0u!E6Y?v}DoDHpz6+a7S|7tg3QV><oAsiY}T^sJTfS*JP!5!-#yj0O7l!
zwJ2n0NIVz?Xrw?HAp!*qr?T`<uQT>w6vVGQ7{J&WcW6@+2ch?+FokSdPpv`!+F$7q
zVv&ZL5c3YP(|cKj<Jj}RDc{5s6$HufP+Lcxbm*A9N9=t)Pv~D?mjJi-&3BHk+00fc
z9n%VSt?3_bkN<9oxcu$1-srz&v0Dk`ngN8uhUVta{f63A+jj8UrJr8l#&0D_3CXnT
zj)%X)OkLZO8zVoT35#A!6p(j;jm@nw+M2a4|DZUkwbG#nX~XgQAvTBSCBvv5b&!}>
zeh|urBd@4cN!Oy5*>R!s8j{7=xa!XmA{cU<in2JwpNttQfb0x3Z&TzNRK3~Yi&%f>
zLl!YQ!U<^mWD#CDbGvuVT`vCt`m5<oUcj=4l?`>WwD&%1?!Wom_l9OC8a-r^yt1-i
z%GCP`ex3oq8^mJ#`5AueZKzaX8k;B|gz1qYN=~T-*X26Iw06;_@Mhk<n-b@b|0`|L
ztj0?C5>%JhP#CR9QQGq>ueR1U$g8^B=bZcXk&^SzO`b5FUTYxGW%@i<=H^?R&2Pwd
z9ak`ulap)Ack&c+83VhciNN-w4-UU)gIgxbZDrNoRC_TCz+wkgYL7Pqqs3raOVz#l
zn(e+{&0bfrKP&Ra#zs-40f1riDQa&De!y1hbVlZMTDC8oYHc<J*!kXM0`%cP9R>-8
z)Pkm_6rcbFSN!qgJJ9ATtEecbcKb?tETOEc8<Ug(bi<TXRYQdGV%{}a><|>}=lv0M
zNNx7^6LMw2%e?ukP~>9uwWW~NN;!24oz5197Qvl(&l=Ea)f)}jM|zTLH9GT?%XI>}
zI!|!e?7^(G)s0*N?{=pbY1I{zqYfA6QK^&4)BJEH4rO$`gHZr6dg?ftmPsmFF;Yl6
zrK35Va1q+F{7YdYdXby{m?e!`gV9jSN_S27I}Pk<GvjB{s`qiMerRsD<Ay9a@YCxw
zhnp*4(EN<fY<={dwC45$NK_xF;~?j_t)Dh*CI`!Tn{p3f8rzKOYcW0e6V2hW9`2tu
zKfsas?=UJjt%;@35LzLM3}=@o$>F_DyLbV)8|WN}N8|K{VE+tnOxgT+f9L>OPSJ$?
z8Pt@y5IqeR^XAspqSB*4TYD04+dq!?tDytBg;&m1eqPq_6D!JvVEvz{1M-j`9PVT9
z%tHZ!H<{Yz7S<xvu#WH~aDo2KKm8Kzd=GdRKzDswGNn*7sr|>n#0aI;t~7;I%7=d`
z1>UY)FnFnUoel@s*XE>~f=8j04l$rWhIP1BI-c21W5o&=Kk&+Q^*jlWo{G_Hh~fS`
zYY2hB?b^Ol)j!Qac@|%Ofq?a<kh<5j%GTt?H99<XzFVDx@^1TZ&QBwkc@qHSmOsYA
zJpW`;OQb-z4@bT%<2`W!F{wX!{ZoH>jt2E7kjC*P@ZbN1BoQqxo!E0UuSy5nz4d0(
zQ&Xg!AT6tWsz|4+q5n4vunjDgOjJ(-gM(l~#Mv?8hF?*h4=;S<7zX84Dand4(7)A(
z_67vYgV(K3D1Qw0g^%4+f9*-euOx<JTH!YuY!OA_xm1G-s5Nmd$4C&R)TjUDc;O*1
zW+MH2hEie8t6WL4&}Di(YFe-#nEyQ`1h&&uu&BpfHh{l_MR7l9JDS=I=+3tS{qZc8
z%7TGO&h?3b2y|RdR|4RXO@+rHs#|U&8Kk<UGdV>Q8l|+X^sK+i^ZfflCN+zEDahrV
zf23r8kq&vZ=2iRB!hG8L!tj|g0hH%Io_Ba=><v4y%e%~Q5N=gAv4(X)%F4!*TEl>d
zVGS)U{d|80+4GJrAUwT7c;2Ny%kf*1a#1=zc<UzjgD+0--5q9a5d<<N9v#NkhMA&i
z?*q=^S{HNs`=H@3Zvll;2ngYjKXlGL+4gt!35B~>J}p%iXGx<VE!h0fb4{7(W&}~v
zB9W7U<@_NB&B5U2y-O&>=C1bXzNB1WdTNl20-=yp;L(#pTVNE~jyEkdiRC10&X>{$
zOF{6>xp>0`8<>_ZsH>Y@6vAN*Er=5Ua0n%)M*bqd!FmAi+Nt)7E*XO1&-UmqQVGY`
ztKp!T0V&FFig_4wN0q2(H7F<dVQ0YLm6r%=pzDqd&C03{!-1oHYxvMUg}}{|5l%~g
zY&0%ku;AN~g7?QCnbxiCsPva9-rZpR{Rz!A&l5j%C_o!Hod`$0LnPpWv)_$qh>={4
zN=bQISc)?}UlZ>F42_kvwPjpf*zLOc6M^<f=x2!owEQwZzXN#{NbOZd6xv@^Ued<0
zbwbVi7b47&P?X$(<Ax!1FSofqL3@`%cUP@<XAoy0q|IkJB0x0TT=PQ|5sxh%V5b5(
z%GYJvL>CTShG2bwNOF6qq#yS6*)&XXpDc+T>_4(Uz)VRv2&6htccT5VA`mD95Uf}%
zEz6JA+f#GBG<~lVi6EH(wbu7_=Y)Gdt)TvmmOLI!x=0>Eub%s?U72<rxZq;2^}K%p
zWn*HE_nu70%Q>(4SF=^g1H5=N)#kV-k!LRJjB*&sDmLYit>RfZ^5v8Q$5#qk#}zDU
zHYhY{A*!s-LWdsHGNkvN*Bh{w<1r#25(D6r9C4h`Qvf#z+}gk-a892b=lXzxU0Zs&
zH|7rD^7_{86Cd&gTUk$UI*r+gZ+zyg(@oco(0j`O=!Z^6?p1)59w|j7MB@8-qdO4H
z&&~A(!@tx!A31)$_N#Va+%)q5P87zRJmPn;1XA;+E9KbIlJ9mz$zPzO#Uip1iAno8
z68ChyN2kS=NG_8y1N;i0OW*-4tFm~OClF1$E1{{eZCNMCz|pARro`(9i=gwIt#Pnc
z%AprKCQj=X0J>EGr0aGJ=Z0bK+r6mdptfBnaE=uK9rJLupzLHZDjgUrMAg~IU=A7$
zBU~8_A%*^xRMVQaH(O~kaQW}{Y(C7*Y|6N%^8rTxJd2xv%fV2*vZ7-wbJB<Q+C)L%
z3hF@59vu7qz;CX;I8+Bi>H5Z(xZ)-4>uS@4R_Upywff26`Qt&Ptzrc)M-&|{ToR8F
zS)ejnVj`-9-#!&X+qHBbQ+|E@e3=6nF3hL?1oZaSm&eQgh{@yZWzj$f$ZH;f483)K
z4a%hcsnc-A3eC~IsElr!g_#Sh=IW4(TYw9Y!uO-DpYAhRVN%YonxsPrV)}O96Pb{Z
z-|ebQ%_B`-UI;<Wr}(SShXX%f*I_T2tmvDR1!-MKjo)6ya$#wF%{7+j+%E%Bb)hNu
zCpM5F3fvKlGw$QA@Hbj$5t;^O&27fI(7)!!d9})1@l&*d(mdeg>(@*@V0E*3WD(5f
zbZQ8d7^TGa5sx$MQbo$6qv;_dO_-t#brw)?)F>jTFH%49*;_?0)ez85din<NIFC@?
z2|86K;Zm(J0={AI{@IC&q2M4y{H_m`<QzEk4MO!@Pzs0^=}hkgZ-vNq3P&ICBx&=1
z<N5W6@P?vs`_H}3q)?!pU0w54MurSac+~~Lr^yz&c3kht6BLMC$c)_9F;S^1QI>;c
zrAL8+eEg5ecpC2#xS_NptPV5^Gzm`70>lhemRK=u!rIqWBaA5K0ECEd&@2)Y;?Z(I
zE<=Q^>n*@2@PZBId4DREE8qj<<roOifVfou{(J8F9H2xRFaI$CG#MawG(fT63Xagv
zYShMM`Y$}{Kd@<uR^;Cde}aX3j!Lq{{|`;)7+h!DcH!7|)7Wlot1%lVjn$~JjmEZZ
z+qP}9jct5)o_D@qo#{-{&ffdJ_I0jvtz!XZAHZpr_dSXLkKL6S1b#5;D3a&=x4*qb
z1~+t)$Bbs1P*XfkK$Al?0wH@kAQ_}It6#$vRS*hcNQB}u<a~hut*LUxft15<xbgAj
zeSR4NRtYWOmpozh6<a`!^@(S%q`qRjgN*z4uR!mm2~c8PuH6Y&!G@0?S>P$LBZwA<
zB90o5pgFW3OhujnK{EwB*`vV2<pS&?ip$FOXmy{YpkAArnoMP`yp4=TyFw+_Vpwbb
z2IIuk%>gWrAWDfZ(ft&4(5#vIFme%zm`M*GJR~Bh1;V?W_oLJYRYg$Lk}}~4>G-um
zv9v$9qy@;SpTHid!|=Bzod%htxXE>M6sY{)g#wr)vC;arzEW?e#li(tc_)_670RRt
z120fj`!(EWdn=#&y%FHwVP#_@E-<?NJENgwaDTc2IP8F@)!fq3XfmDa(<GnHeIN=K
z2!>y9OQB(d><7);ru7dnNOe&fpk3bm!Ry``e3hi8fIx{|UDX82=(K)(hkX8Vkn(iC
zHoy48e&Zl1&qL`PIMNPk$Eu8g4^+AK!e~GUIt#gUD%BJQH<sp7Icd^2Mz?W>*Moy3
z&?FviKp}bRc{*aTKcoX}e@nHvhkNZ*A49~JuA3?=lr1yXf!hXQ`uy>->D{eve<rEI
zetNozeEc$*^hyH%a~?KuxyZwk?C`gTY|xkMF7Lb2QW3WO*Q3uJ6<O^Vx0dGnp_a|>
zZAwb+u{O(1az-=gQc$Q2YL*}*Lg6g04s)~AF^0BN;cUN(a}lCst`q!w9zESASqN1e
zIo_2Ot}MH2>hbhDq^9==dXK%YTbD=Yi>Rf%b3c3?y0HBQ51}eM@4t_yunbWO(?iKt
z(u^)9Nx2+UQ}$U=$`@aMKp4T!Sh8*x_-<foG%iWdpTB-TDW@QY`ubdM1|o_>Z)Rn4
z;9dEI&+)zt+8rni0F<0%GeU<jQNT7`k91Tkv`4OnJbjg@i9^9C@I6p836nu(w=ts4
z5%8Il^#NPUUPR^j0#LOZ0`h$8w3w^*(-v}IQVxmB2rUC2KY&2H^KLPV1(-9APfjvq
zi|txiRd#~w_21kgdG43T-+u$H&hb>Xu+K{ZusNG9k|+G>^<?}B)Y(p8w=EZ7M~lBf
zzl`ZLOB*NeE)xkXvo{I2Lc$ubf|-}v1J{JG$-#jD{meOiw){q2xuZ$+-M*afrQ3JH
zZ-O52Y`q~{cNx;SpR*BI&^^d>MFIAvrZK!bUueQY-7YUYkINoGjS_Otl4$$MH--=U
zHGJ!@i*^^D5B8HS)`aIS!(NpA6rS0-jZSBp&lj8^74&y`gViQyZi!Xu1KY%XuXXsw
z4Qn(HT@!<aaf|1VMG^n|GRV(ospk*1BSR}S3P$k*;>URT(l`fx(<Sxc6mn=ky$&!<
z+V8MzOZI)>s&0WiJ6ltY?^tCsUbhw}*LHqMOUj8&v+rjekntMT@=pmJB&yZ@HmRF_
zA_VFIrX+1QYQP#&0Gv9&Ayr8;1qAw_1V#M8zYT`tW{}e~ECM&8XoH&Joj3|12H`wZ
zDWzGcUAbN=c}2=_Jibt1xQi=;OV?2?S%=p=BO6eH@B&fepJXLA4=`LfP!UuLt&s?|
zEduxX?W0hhu{xR^^HZM!gN%X#j)Xh(>kPXPR3%m@CRF~NC1uiZSR6zKO_j&428~$F
z4-z>vgQ`ck-fFtC#MuRO<<LoknFR&OP&q^_DO1}o65P`G>iW~-dBRDT7{Z({3_|QJ
zwY@7GSkmO71+mhVxTdsJ`3`YXUGo3ps)S)e3(!LDRaFTG_UVknT<*IXj6s{gn?(zI
zC9F0MS;xhx{!%%i1%N8Gh7>iz3gIvU7FJ-w+5!ERKs|{BnAs5W=@RhP4ky0YfL#u4
z37mHe?+Ho-cK=FJpW!t2SFR7VuKm^FCJ$#|#a^MjhQB7-Y}#2StI3_}*vQhGuq`BE
zN^D^TMgp2smkMM4lA$@;az`YYj%ZTa%e5Qa_n9B_j3DYU6>ujLihlSdlCXfOBxOWY
z`L)>QZKtGO2bbG{>hA82OcLmqY46ULjKcx`BI(I*qRwPu6vXHG*{zrH^U&+B0hBKt
zE}#V798g!^p6?{S#7rIi`}>z_t=`k;L$KrrS?H*bB2MeW=t2fe^M33Y&3vXGATAB=
z4qTuQ)hH`aKuP-{U}V@*>KCU=N%}_`CVta_>uylf?F+=#l~4)|873|up%71jutE$J
z_5~h+KR|K152q7s{q1sDS=kud=zETS>*El3Cn~MH8E`=6$WMV~WZ;`suzZ!+Tj-LE
z^-+K+2ftK)fnHY^c9%vVA02JGeS+`Bt7$Z|i1|A%B?1L1EU{Q+0sOg0Mxb8zQ-g*%
zRtDp+Mm4+F2XG`U4W^Q{y%84vG=P(NYmhj}0(6s^%}u%e7dE1=AM;B~EFO=Sr$70D
zl%~#U-fWE!Lz3a;c8|xga#k-+AK<Suv$p=P`27PMX>gQA_b<<_*Ry1z<$vDUE<Rzf
zB@Qrk<;s+p!=3s%ub0K9&H?_t`|a+IAyU+UZofp=dU~D7^@K5VOY_6_8}DwwkF&EH
zB#I7Ic_w3@!TdWxL`u>R2BVi!)Nrr-Fo4o*VDNe!e(I*X57`!|o=2y8`gT`HOUnWR
zPa{}9YW)gsX+V2qnic}lT5bv(4;I0Z<zJO`W99X9qHDh$bw2>&4D?tZrMn~Ll9W80
zU8Ijy2z=OBcRVf9S1Pyg_WXXhb>}=;+p_y*K<siE^QX9?!g#hPjz(@M4z5N5^R?<;
zzk1#ELFe_wb?XK~`8(sf?KVjF-O-Y^H~z~-m%w-YH!dfF?b9elG|WH@N10;b@H{>e
zpTR_v<L7%Ji3FV~#II}A6ih58orv3lJj9+831^f)p4ZjaS@z)%sHjL+EnSC~`S4d&
z4bd@8OQ$h>yene_QDHVXqz?X`N;LlswXYEhRMA@1!{O-Eu4_&Bco<+w=J(`Q5AFpy
z@m$Qv9lV&+gyvNV3QYQVMFZL5dcgB?zwpHK?uAu1{JWBu>gDN1>eITKsi!Tse<aYB
zz?66Vg|BpT!CVOrpR)Lem}O33Yd5yMKm;(4q5l>Xvo{0>q;i9S+JaBRwpj(&-)w(v
zbGVaHz2;ODLXm6PM_kj{kF!^-b?3ASct4zJ3ABd?_Bg`w(?CI^pl1GZ2pf3!bJwlC
z6H~7w>2Dp$0D0>3mCAa)?J?8kxf1An@cg82Ftps-u>*O>o;aj~>5)IC4<azlyQ74k
zMj{G=N6Dh~J5dKKK}gV#cfupv`xR6Bk%FW-z}MdSg5Qk_{4y-Nc%5&0aFY}v`p+F*
z#-s7n5(SUvUEWa^l^vPDF61*iiUve@peJb#o@v`6P}4WmyQh?Bk!3E4i$pdm9`A8C
zvQ+^ezYqo*HvX9!B&ei8jj!~guc*|(B`PV(;#zwkXh?{8>eB8!i7!bpRbwVak+6!_
zdTgxpXp8V+5Mb&9KOifOpT$;QZ``#3eVtpST+|$l#7`st6^MdoySxK3Zeb0&{`7FR
zmh?F~e!@HeQ^}?V&X+#%j7LrOL_tYmyy9f6FofLb<{zTNKH3bq%R88;Pd?r5m|v&r
ztMI~2N7`@e8p!^RkCN4Fdh0xcM%jk^`ppJur!T5)7s?-J2CNz&iO&Fn5~uv#{3>-w
zm}m}3r1anX{d-JoMJ4X8t@`4QTV@OD8VW#%&cHwWaX?zm9zfG=HuTFOQE`!408KU<
z16V-7t_*BePb?KF^{#xAs$wTr8FF+oVe2bzSb3P$B2vYM(}<{y%6Ib3f=z6Sm%~8>
z^Pd6GM$m1Bs-Zw4J_o&z&sp9OcoX<FEX}s2#fdhuo&IhUpFV_m*7ckJ^2QvpHCe+c
z1)(O7dfQQKI%CX@$B2<)vg?N_ErWAtw8_s|S<}nK$I-8#ou+}%EE)}!eamk%TnnYI
z8PbmJ!7(-Ntf@m!kXP^}YGe|Esrg2fso?_TfDm|_X_&Zk;`>FP@S^zNaXzoO6e=ni
zUfd!<2*e>9FIv~3mD-x{BeZ(6ZdwP+y*%WjAT~H+<0(AYK8J~)chH?r9mh8I^2sn#
z|300|3>hNb?zjZU@Noy(J6Q1dXo+%gTtKNX+menq_@d@#yEt7gZ}W$b8u#c??E|%>
zHsDb2HVa1AHJs8Uu>mcOgP_K2@xZh_{ly_kD!GfcpqQ!d&PXW6P)xIA#|#!HNMr|t
zk1~u0N}N+FtT6(t6aOugshJ4ih$*UmC*`a3c2L(JJv9a)Q*GYYEtUC<&Ge<V*IC9h
z3&KdJ;{^O3cecoo;$Yy?8-)vGV<N(P0;^Uf6<z67fCMGaa3RNvw9F<W&md9>mV7-@
z3K?EofXl(_RtCMq-2-7gJ40$<m!Orf^g}ap`a_sc^14<3rfFf<k`7*<3v;k$2pyU#
z1DT<DTdCUx#K^Iv3eXIPA}3Crdr-du9zEcAq}Q5;>FEuG=eo=Wp#`r_B#T7Nxdy3w
zMfo-S8Q@#il<*-F$zbFxY{B49*-Q9X+W1Y+g`pzM_!jdrgA*?6a_PG@>%0%+wr}rY
zJ6fWBorMi()pL+|Ne&HLUUyU9Y%lv^gB?13#0s%Wxrt-_Y_gtYKS^3%wkPx&I*Hy5
zC_aCgb8JUix)!Cpmt)FE7oT1X*swkBguO1$?3<5w$IK&^Y=?d0IIl0KjL0B}A+S9z
zJU-i;>{bSnW;ZuCjRF5Eux+O}=kd*p<V5o&wb8!yhyV1P{K4mKfv?HCPVW5JxB>Vf
zojnEEI=zDtUx9=~WF*q(PsqzV1q5?!y7)KgdWZd+;R88sdgz?Ta%29b&R2xz%?z|*
zRjIeTFx73kCAnXuPX64xH|q8Q4JftbbD2l|Zui00b>MfAH3gX4|DB#1%~!h-v|b1S
zJyZ@Lxf{LLZ0my=YV61>knO8zwhM1Iqh#9fweE8D5#pLvBMaoU$ud^6ke6_?Ho(J7
z4t0kQCq}av2*@jazyfF}<~zPvO<f#dJ*)wg(praqN+y%(|M3KXR5$#w+XKTzryoRu
z=CD5fMb`&^$J||cfqWLfw)0to$3^=Mds$vEini_3K{^0;z~gqUtqZ$Z8v&sgq=fZ+
zzTG)%+M0y<U8Y)*+uE86I5K~aQWs6PIooss6tb&>Z{A!yne$5Ld$3xd%$BKV(lqah
z1Rt?fX{A&mTcF-De_9)=<z!X)<0np#NDIo+s33tkYQl248B{1_mB3N`<m4VyeQe6Q
zBhV3(fzbDO)w=WY^=99$g<*>);irbV^9g;^^x~p83WTVlXlR})H-swt(Bj&eCcHk;
z)2dq6E`b{Sp3_ExG(+dI)%Qe;;-VLM?>72|bIjd`aMn!xtv}V#Tk&7#@Ym%37oos0
zgds_arWt@Gfx%<Gp!W-J_#B7>q$go)IEsYDBrW*8Z)Z+B)(BPt&VW6@E+82hFIVfQ
zd###-tRGlN;uzwn=uX)V5uAX~XDNj4JsO^Be>5&@=26llX{pJe(!@g#W=D*~R?9u(
z$AyuoR;mjzgu0y0d|#>d;`>A3Mui0pk{FxZFFZpb+MFXuZM#;f1Xn<^s~$&&G!FLn
zlq$9gt}POs9X3o_&%z?wH!k;Ktwz9L$5JM@4S{R+Srr5xi*6XL6cBjbow_O-fa#6)
zh2c&=i^)*jM7jmY!;%DS1M^^yrZga*0R6^UT}B5X>wtHFS@00(9vdm2Qi|GU6><TX
zF*DQCJ`IV0q0h0TDh_KyPC$!x*{W#+rY~CHbq8790`Ml77a~cMW)Q&<F9S7?b@m@X
z8|}L$3^B9bRkC5^4qzzE0gRtOM)(HEvjCL>0=&xI`9tfAi;2GK<`p%!ue26neD9BY
zD@W={#e_UTfa%9q8->KD#CiSS;($q)e7KWS2#jvsa0$MWxTdqE2r*ujy4C29%Ewdu
z7pnamOCx-};P`w#YBZ{c^@8*|W9cLlIAtq(MU1Q4Ir$ra*pfH`uRpw<BQmo{ZC>uc
zf>J_O&@7ryB4tYrt}9B!Ui1D!H0a}>aX6ldILbE!%meLD_gd|5=Q0fKm%<Q6_u1Q+
z#hcF<L^Zha+xI*dzx%aH@VW^($qyHMlkFm4FfjNO3M3VBNNOY#|1x1c-=0Y8=->mZ
zx$$Hse*-0wS+EkY`&CbQ&7}(CiB!k(pACx>CzdUo9)kPKRGL1GRCW?z64`21KY@f^
z^wg2(SJld-q^TnTf`Y^(@+A9U6BhQ@cxvUxv--6e&AD1VP;|=Eix-fiKmo%bL)(-7
zT0oo73#_Zc5WWQbkc=L~C?L5C#!C%3IX(W_#BJegn0p;t%CuyL2;vg=>-O>{?N?ZO
z8tq0QHdBbJ%jFUyvjt<QnS7l<i=c_utrRsQcOBF#RqXG7ZzBhGm5X^zIC5yFJ2`J{
zaiz}$p0_jQRXl(Z&lOnXw%*z&z^&hUbF9S2*9S3*ny1K?Y`DH|6wTmoTz$g_BPzpJ
zgc4<TjQkx5$?^I)6kb9&K#+G*1KO`16oJ<>u=EpL>N*KK)AYf%E!L{IHUqVnn!Z3(
zvL`Q!_ms{AGP~n=7MC#yB<;=fM&5cKIA);o1PFrO8qpdp<XA|xJc6aG7f>y*X+><F
z&O2%YP{?q+uI6-6Dyf8jDlO!Jz(UlLWT0tzIgL%3X*Gcs%pacopi5FxDlTz)k)`v6
zb2*!;YIV|8^U+8N^oPEpdwn$(_}DrlwH0lW5C4E*F&(GV73|ue{CG6FcTHky10plC
zm~SlSsZ>Xj!}v`bJ7hupWftiEFf`{<r_7CDHD_cw0;1{~SG<zWe<Oo_sG4gwmd`(9
zyMZ@N4X?r+DJUW(bi6<vNhwnOXk%kHS94l3bVf??paLQyqzEFwK1n)9K@Z4B#*Q7R
zKW>t^-Io7*X8ch<!K|?2-1X9T@wA`s{|45RFM_<2n6)u#bMZlM1xyjQ#ZWt5H<RPe
zjOucHR-tKOV|H6HlXIZ+AEp-*dE{JUEK^%Ap1Pt&n_6j&u>oX1O2xhs5fl3&NWj#%
zoaket>47@6t(`dzzPPPh)w!nG^9~sf?m4-S_lPGa<`nT|4&op$<BXfYLKOc9ON{L&
zk)okjdDt8ewn-7#{6@lJv!Q93Af<NT;vmJrLdwj%Xydx|;F{TMdo+pfv6Iy)!S*8|
zq{IE;y5Y?E@}O#cQukwGXt~Q<04O`(iy0y~sGsBb)n*WN1m5~8IdVh9i>^zvrb8m)
zD;lPtSHhZkx4(3S4>?}@Vi;&|Mq#gP_(P*lQj#oMG|q93t`qTO3AKHQdyVNza7m-d
z2~qH&1r70TkUo90kNVM??=AB^CSJ9WqU<#I5T`HyH_t$D9hB*)WH%9^SsF-&3}h5~
z--X%l;tkH;N4?#jD|`P!`AZh|i00Xq4wlGK(gcg?WLZqeV)pSUaAYKamK9!El4l?C
zo$sVdiAl6$vE&!9gMy0NYl5vXY($H-`g`KIsHu2#I&h$BiH|A@`j*J(7wP&9aw^(E
zC@5;%aTGpS;)cVR)&_L4CrJlzI#3W0<$EntZGHeA>Wc7V&pV$^`u%+_YbOv82T~0p
zAxlC;`qpDjiIBef;uNRXMh(6oD@-EWiis7cg*Z>FM~kD$qEeAI7uVWJlEhA*>A_I%
zqTnovnz~#`S00SLoFDmaINM%Cp1rO>f3%;Ru7VtIxSx{}S+2DrmW%k%AL)LQWt#1m
ztobjG2xsBCa`l!fbpUO~la=Z6)KnnNh*ik#{P1SE!2;X)q9sY7;|Ux@hp5UN_k7d-
zo1)&FYZ|BVo;jHzQ{c`y-@{Ia+CPdD&Zqha%S|VK=M6_}=Z&^M$nQ&9+~V3&Aw15f
z^I^txT20mzQe-I0?N%&+PoBg1wAnvRz~M$l!G7w^{c@zvaZMG_On;?L>%q-M5@#|R
znE12F_USQZo8t;6zDZEmaH@VIv`t)Ue|LO6w0Wn<A1)GjiPC<lJz*TI<|H_<hs`S-
zEJ|K+jEqNDu=-A5C|bVLJ!?S8r-A$DD>VPr)_drc3oIdc+xe;_-TaG8dTsaOFvwR@
zv|Oky|L9{;O-%4Hf;lQfA_K{KFE6ZvA{U6NS2EBVq%1H+y4b_ZJFv;jdD%$*#~$Qb
z;2QiTD<|=dm;{AV2t(lhw8b6S`K(>K40OxG-)mEpMvYjqSUb?BUWX<WECQ=1M?95!
zAd=o`4f_K&gew`F1gnQ+-TQ(-7B^p#d7VslHZ5NWFRot4OK-|kLdJ5%gMflEFfx9g
zS!uFbC!I}{Fy<9RgAT@$-=;3}rFI|C(e;6y1D)l{=wt@Z7wkkUEq43slp^Ixwebl=
z&V%^4;E`&>2L?WHNS6~Pbs4z+{g7~tVEFL_%>8`{Em20?2=2V6yqI2~UN}d|Noy@r
zP_2Ltmd;mA+nFuvWncHE0{VQf$A)kGMc@)U-Ene7UAg+m7uw~q$A)xf@Qb(V|9b)K
zH&;v*t3^Y7ZzlOefE;Vo%XvrGC#1}WUaL`Q>1RZJS4WbNKIh&1T{`bKiCYLl;7i1l
z^N90bejHVJp0?Ii_#DS&g^j>4N7s|=`|XtVB8|GmN|OyC2(s79369~i#~>8Lssj`n
zHcOcL#l`$=jT(2sKa?y&SXdaNl%(vdE5<=1D0}9RhY#J4l$Y}sXn4zw)`YkvB-aJ!
zF2AjpjxF%+!hi_2htqcd7U67-T*ja|Ile#)3vK-z4bjpDGD47|DHpOZfdRQgND9yv
zDm_Dih?AK;h2J1&EU~PK6h2YYAmFz7>X)=#kF?tLj$heD`CqX=5FJ14k2^?y>zzt+
zp(dh{G1Xm9a1}4J`r7c{!e}wHCr8&ak_}}@3{LAw=Ds(`qH~^oIl4@l@{DP%bRChd
zJYUQZ*A!zaG6ZER9E)3CG+m|w@7pfi&5hwUH3_Az>4j*HN1P2O4(>6m7O!e*jMp4L
z2;XLGwjGIfMTprMflL-&(%$sY^yyE%yLs|AJ}IyFIKFvuw6Du@hUUBD=z2GO8@#>7
zj4x61@Gh>*dE@;+@`)`kx_AAxGDodMWyo?D?JvDeH+qdq<aVe$p{)#p$;N0yBAO>A
zMLBKnw5x{@dnB5YVvHW_qMb%OVkJ(3YKBt;NoPoJWQGo&%jE<A%?&B*d7hFROm(es
z)9VGF_x_P&5p%B8pT+8eqpKx}Ws(yS_3`ridn}NN`eJqi*8A;!w(8nm;K4rK8)$OT
z+5xAF;3aH+>bkCU?hwJGD;&(BHg`|{dyU?HdayO1*45}A#?T4jV<E>A(eiYc?s>nm
z{<{w`{$Stt$DQ9Ee45>(2GI_Mh%%G1ZxhsS6k>Ip1VmEt_;A}H_dm$s-{MrKm<M-b
z;@38Q_=v)_L7aZ?<qj=RCqv35i~g;UrVkI2Rl#d7_B_x1=h@@A4SEOVEY_j5^z6hq
zOa6g}*5l~+nimbxjPYQjS)ec@S>8hn+i@1<t^`KLKqyL0h0%@_S~6cI*55cyJ|y9@
zuQW+>nv(`7#}w23bSdDUNf87qUuj_?RLk8+o+f2PXpdL#S2n_zze7}GwPkcp-5J0T
z)mG~fZy!k_ebD}zYC+}O%SA>`xBWeDQvPpA|4jWHwOT||741<qu|j@;i+o+RPfBbd
zPCPj=#PL<KiN2744elW{I2NHN#&PP=PY)&lDFz(~^YNCb<??dJp;#C|3opaBm~V(4
zOyR?_?s}B*x<2(zr`~6<aTzD@j_z^c?1>B_dmApR#MPMq1;&Do<0}-1e9TtoI%jg;
zX~=VuzE_#?>t|PcXa{5B*D7yPqrpi;o94`|$A~SWg8p-So)*&woeT%1;;)CLpi5on
zcbs`>k~W<?;$A0BUef65ZB8fFMysB*6Y&PS1xxJ^ltCEi&emI#^iAu+;V~QlE|&L1
zV+v@QmacM@Y>LQOyU3@4J8nH+wmoayx*h!bdy$DC;;ECPBgMJnMCTEunQ~09QE0cd
z5-78VZ_1Fj<P|gd_2s7jLe>z1G$<bYHpkkbIAd5y(i2e+75Fb{O|1d2A$JL;tDH1}
zVZfCzycfm03i*W0d(SxnO`|3#gZs!qGX+DW829B{+kL_MRYKJ<Q(;is#g+&jhq@$L
zQLF)|F~@w5A^i7_X@z7oQ@V*+oSxlL{=5d77}Z&DGqyt6Obqg1)Ne6LQXbcOfdU`X
z6Q}DHQwn2LSDS{?#_vy!zwqu-f>V05kfaj^S%8BX$?I;(!9X>UTz$x_+^Ut@Zks{w
zB)5t;7y~>{JdWeR^Nk<OH|$$$@)9skx4&#$47yKwKgE6Y0yZ^?<m+en;v{se)alEn
ztXD07#JKc!m}cX~5|JfgM}Wlik4ZRjKo->yanbk-*UVnmQK$QDb81ja3)g2Lb+f~r
zDNv~TlQ2kk?j1iwu=1S`S6?<zed(<enJtzmKoj~vF<2BzED*x1{b<SiZRrWHsY`^%
zJhUJ2X#=!1bouH%kU8$Kd~L^Nxx?5X*uY@a=Ht?%8*gtH<Vl>)_q!M1h_^vU9n1?d
zU=*|*v9NK0--IY<d6l<mj(Gs!eDjX`oi=<B-I+H0GU21yp&0YhVe7ZYODs}}v+=z3
z7sQ+1H}ozSt|wYsql0r(L{j^hTQ@YDz>g9e_q-wVk1sv&3rjeJjN~Sh=J(ezc!U11
zFl%38ty-9dV*cli6h-;d!LUMqo)z#TS)Fd)IEH_%cD*tA`PRBrvrKllnAPigq>IoE
zd~>dJ9GxhvPvSeD>RDSO9;MqOee1aAE~i$<S$OX5hcP2rUL9xti>67tjnc^i1d<h8
zT0P=ah4X+Iv{{Wyz;8VP6$FoiENkP}Wy9bqThV(Ux9@%X;@vO*J|4xShC?PMp>eC4
z|CK}DH0S;C!{eL4lcB)4Rn`aHh4;scd1_%CrQ>9c=>yWH$J^H*gNrzel{#Yp9W^;C
z>&S27$nJ2(ln|pJ9)-fP4jr^-`&5+E(ybYNBvx+Yk*kX3ty30(zHY_*3%6?gPQ{+7
zJcfUKBvT9&ID2WiRTMf>WAI}-tD~IZsUyJ&LOPZTBx-xeXyaejZ-`^`ac<>k`M*kw
zDVbrtxzRw0_e(*iu?z?d3QjZ`nI+M1;)+Wpq}aqY<(K0-<hlmMu&sK7ms<Ri8FCif
zY<ZS`&&hIJ!gpQ5)6wzvKi_Hgue}2yQp>%lkH?3nv#oclbc^4M_S@aSN>$AZ*P{J7
zsA&MTxXSV+h3#2G(Z%cW&(`GUdbGRKa|?icUKl@|SN(q2(A@*9PkS{(1cr<_m%tmI
zba%1{?(p)OFEU6L)UT~aL}Adjl`)xFkBKYWNnj5%a4eadAB@g2L7$hvfJhBE2Nk6u
z3rTjaL2?b^HMacuF}!7FBJDgZTTS`N*Iv3lXazsW?l@r*%DA&)rdfk&YXXfw-rq1(
zMySN^d4CH8E?p-?lL*!Cy6~1Vfk6hj>4(g6-R*#u?(~oIFA{*a@2;LC)6%5IP#_QH
z)lhh8+xEjf@A<l5;G1HDz^`!7NFy|4H8Cq%k(lu=8d`R+3(9T}eKxc2rSAc_aJJ9W
znMvH3^Q`j^OJ+j~c$hTC%zh|V4+dlegx9gDSL(-=8sV7~0(lQ@l{**62wVY1K;;@A
z4y?a=PRUm|RGEKfgPB0$C6%pFZ^}w{=N$UHZ;1^+7jnNnwpO-W_QL0wb5ErkA#^>m
z2=ryWtz_ALygm3oAe57bi(?pYK727%4lVey;6{+X_x<C|dpnQtjS8qy0bylV<NQ9Z
zIA^8RXnX(9)IgYIP_V8!*k!7kPSQ~pl+%#&<#x{XByF=FyYmv#>3m%hCqli;1kq7v
zZr}YP_hxEHoROHxVbtNVhVWUR|0+ZtAhzF!h#E<aV`8RdwuE47dj(4IM*J$x-Byps
zf7XoJ(vU_=_-mFLWR=ElKQ*}#45P;XU`JH<Pk6WO=RMr}CGrOfIy&-t-Jcv<*MWwm
z8uu3@(Etlw2R6u{eS0>q&IhINV{~vFru5c#vPNEebda{rE0DZG6NQAwR^p)UMEFKj
zgws_|j)(J=zGV)CvCsil6JEG-{BjXdoQSV*OHmk^*1@3VoQ%JAbXa=-q_#O->`&@A
zJDun_TtR+np5<CaOYXpw&UsX=iau*BHyt%(2qdx8H8pqbvtIB3It?;O?;n9z{muKv
zCdDN~?U;r^d!FQ4hjgd~HKgXRw98~3kEBhExZe!QcUqHlq&2uQod=B90W(YLd`xi7
zcE_`9%uC*)`#&vYFMpuTMj??LK<Gk+1WW!fwiR9BM^)5-jz7Z`Cdy$B4jFd*)|MFM
zd1{zplbbBMH^kd+^Nq<hqgxcwg#Y@IU&)N<MG=<?qUt)R%5|fRi3Fq4Pcxn^3?)@m
zf7{vn!TBj;>j__$&ZKei6ZV;IpCNM`7rD<Cvp_B(bH5TM5&Cz2vhLqH@Vy^lvmNK1
z78fELvFumM%F2J$i>;IJR+~+*<a0^r_4m)y4e~oAGJi314K;Z>y`*Hp{ZE&Hu#_y_
zG<3F@p$Ee5KfyR-zJaKeFTvD!Hx^6wZ-_R#dxRgHF9Zv)p1*L@6Tj_R_5cWyfkZ3p
z-Mr|2$IWo7-LatnJX?d0{<V%lXl*9Ftvz^AWd--C!<%mg)*F)HcKTCyHcBIh$ew>i
zJnqVLjhFZy^hb$JZB~2lY&=K@duN+X7`36rgsy?ahY?#H-Iph$Q|~B(4_p+G@qQOx
zxsw{>^{a*|6p=pTB>q3nA4MPiYF!>YQ2KAr4cgvsckTPFNuTtmbgr4kRSsFT_-UxV
z2d|g4Q(f)tTG}d*Y^2NvizU0~e?|7I=0*6i)(g)au|w{$2XaC$)pFOzqP{Z0b#-~r
zbM(-y5SZhYRAVgqb6b-d(v(<)yn$?F^~G}rKQd(sJzU(<6(lHfb6;?OBQyUKR_UOP
zVT>Ov6*V?FiPQAGCn!UbIvAfYrKR4az=3-kuRr0^gR#vNP3hXI<%9@{8PyCAmMaY3
z>CTI~o)$sADv|f_`{ZHT)t0;vI2hhWP-cx+z0ZETicM+{Abgu4oCrZCa;Z-2<(lMs
zvFvq`F{aN81W$KeHJ=X)v+Oz|03PNM0FX7@^^rq3dW#;$7PSAhpR3np+!G)V1GltZ
z;GKf(2K;MM<rbMQr=~Xmb~qF;D}48CmSG5x;^tssRzbWZBawYc-tiSLkuKTJp03ZH
z(kMrv@$6KUo=fA&@sC|GzptW(a>;@zPGZ|n#ZsNhNoSN0{2}=eqK}heoQCAtyT+wE
z)#$mJEC1_G9n`${7pfw)%J$5uG8st#&Vn}DNT#OLcT7^tRR;p#atQ^LM{T;l@5{dj
z0QR+N;I3K;(DFeD+)tSpTK3WIdaI*jtCz#6pj%dTcSGG4hQ5AR4l%85tdU2JatmXl
znBABBZAelR&49WOB?q2IQEMaQZ!KLzvC{8H`0s&cEHP}#6ws+K0uKlUKDd6iJjp0Q
z4sxYJ6qwU`2Pi5lt3dA2of(Vdn#pHV1Lr|6!l-|;rX)rEGubFVkvM{?am=qN6*cV<
z$0Y5^Ezh00Q{dJ`!Uv`8xS(RW*`AV1r49_>_HH#ngpWPmW56z9!j%6eMXvL-_2lOy
z@OlS6blzA=cc(g@v|LTt0D5}-8E_t6BD>-JNAUxeGD*&x)XDt^YP?Z@2v`0viSwyS
z)VI!QO>GSVT0jfHS6uG&bT?5=nXIw^;z2lC+bnx@LES-d@;fS3uJluGljOkVsDf42
z-#+s5pZ>)6ikG`x1Uu@ThU2YtVdzBn!1P!oB+$ZFy1XaM|9p7r4f_Vx{oObx-RW6;
z^K$UyX*Q{|zaZ(S)ERhDh8NV<!(!7;2XKYrDA_{d_Q8s)`Y!c|152!=DGPY2&bE-d
zlbkn3lDclmWhA1(0Akk{ErwPfvw*Ib`7QMCoo#aBbpAVn9gRw5=x_6sS?=Qo>|e6T
zpQeqRwIbHZ*CV2XF#`9(nmmzUMlrw^t2CxClgx#i=o+jt?}z=<=XK+_U3-TJUDIDK
zI<n%yD0~T*Ft}E1#`oO|Xp!Zp=nW%SN$<9=4WkAO7}INaFl3n9pSRjmzimAOcekcO
zX#76)+|7p^B!p7XHs@0lbtLoDps$<1MVhSVSihG`VFWDyy|Iosi<Mx9W=%w*t8ucX
z2P$rk8!*Aq?6oj|m?(zp^IBj2ti}mVAFk{$<-`N4uCC60(?>7P7tYPeCJ{$W_Bb^p
zoa;pOY|OCFK+r!?zRwq7_GjHvu3=l3Fe|XQCqaqh+zX%kM4qj5xO3>f9jnzCORA`7
z+pb_*uDj6WGf@XvU(c|{5H|md0LYYEP}^a;?Fm;d;SGt>-#n4hII^J8%mxk9Ytp?p
zx!T`VwnV~X=%+lORE$ebm{c^FdG~(2pUK+nO6xu|17G^tm6SiN)0(UfT&LU#3PM~2
zdQG@wWB;<Q;RquP1%e(SpnU!LG;0Bq2l+WSz%z8kyY1M*S%f(HkDZ2o)&0Qbw?7pB
zY^adqI>7k(ew}0}lM4%5Lrx4kVHVz;lWM-k>%=3hwjf{Ze#^<*Dmugr!8G69-TQH$
z^%4E6&&`kavB0?#81TC=qVz&T>9MVM9zOm?Jq~~8_cQIaV;5UE&`8bSn*k=DvDa?`
z4>zo4TlN5zc|LH2Z#!@3%l9K2srg<o!`7+RSfe7)751w+uLn_|I!5siwnh~smOoKq
z6=W?OnBAJ2SQKc1jws-PsOeGh-6-JvwtfY$?)B9{r5yrbMq(}<@_wLn;=Cr*O!?ii
zkbl)87UZM+W6-rqGqCK?$i3NbSdR>Ret=lq5)50JujzRMY2UpqkE0Xd(KpZ+8nXTj
z>H-Ea-gakH1Ka99@?}d_a170tO7Ghp7klM(lOzjnL$CAlZ@ixcczH80h6u<4uZ$lz
zQ6FF+DmA{&@6R8j0HJZY*_QNVDQlFkt!u<pi{pUA6h%4u*YC=0KdY$&N-s<JDBf$B
zY^J8EwyNrf9`M(_qOSgrn`*3dzUyJqPm=y;In1PUXWZN*=N(SgZ36KVbq$!L-4{<9
z6Ta8lTOyFcezLfN`ZSezhv6H-6i?)ePy-0w#?2@ew#XVUjKtr8dAaHuLM208c!eOS
zgI?>9{H-VU#jx_se^l%KUe))6Ftwi9%IGT{m&=7WWO!BC;N~I+v!!=g!P|J8-C->1
zryqC{J}RHLT3|DYTetqBPzq+{M-kpX+xh_a*n!vG9%q@_4`-PK*<P1Da$Gn(W-0&`
zCL|HlaP)mD!u=y%-hi(-+$k}L$@MSS?Ll&W>R%~#Ck!DvKLy`rHfb~pNglTv!WVez
zt4q?TV`s_r%R-Ku_lK^m3?Ap}(<1?Fn(wtS6lNsRMmSHXOcaMlM<qjulj<G_2-=_w
z$Q4P02zDeg$TZLmT2k$1D7**19K$a!4c${{gYeAml{wgMI4eDNRNA)(zV*8ckRRmB
z{82PxRGE1?>&Gz<=G|Mjcj;^4*iNUoHt>nDVbwR09b&9jkqnM8wdP0mX|h!GPEBNE
z`-N*9SUibI4EghUG>Nv(N&(g26fB7eN=Zul{*~6{-Z-hWrS1U%>R0kAYlKvx2I2$-
z6OIM-s9y#4&u+-6R-8D`N>K<x6|Ztx2S`O-e@Zx1KCifW<%%ocTDqEq1F>%{sww(h
zsJ|+}Ent~!d+IVLAgA0KGfi2q9>yn*cG3;0p?xdrbTG@JQt(qq4zdk?EU_Hz<-#-U
z_EsEO+@;vE_83+e;_)tZJZj2FAjFa~u&dLB^cR2G(=K@!LrrGj%Y;kThAm=&n3J>y
z86m>qC89=?jOXm{ot*_ML@_EyZx4J6!%(HPzbnq6l9H7wC>gyhW=`$*p@k#ZY>S!}
zEqDBNo`xqX@((Umya{ys#AkS?PdJeNX%KvCdy>bA$Y9xSQxySAa&_``cj^P}6Y-k`
zv@gXNwR)9RlQW4*sqGBKz&GmjSL(F6oLT&w%1$1$0#Of@OTAGDrrx&GDh=Tg`4_LD
zl}KKc?m&|_z@(b~Ox-ULtXN<sO@AukGFnUIA}<lri^H~pxzin6>o~=^hhR+J0+t46
zFioq9fjL3)**38qm-R|E$#vX%2+#v?45>oFEY45?#t#$=i<}q_s+oMAVv86Psv$kL
zwg+ljqxFR4evoTiG=>Cqv{^k|r-ss2s1fHLMuL#*-9vjVZ6=CjJ|_h#j=GSNpwfOX
zeHwEaJIY{K>UfAG<*1~~yYbXct*P!ZIjGQ@H~QD$m{}YpCD8cfFud~mO*Q#<h@t&G
z3%?`xuA2lLK=Lj2?AK?sw$rJAQS%;BNyhF_FlMQ^S8UXVhCQY*=+_VP8<>HyfE%4X
z;J0=~&^yK@2%=sAA`iTZ*6<8U{hElrPK$(KN107uzOBwl1uV7S->nWEsxWv?7HB`G
zOSf?rITV1xrM>iCcmOoy`r0v9!npi+FKf-G!V*VPSKh6f+@3$#u&L-UR^gnDfz({h
zVdIUi8?K2f7hERjnmIb;ahUbb{3dz0it_Skh(E{9UQBKv#5${I7;mDT?9I+B;1M8E
z%p~H2<5W<a{@UV@v8IR-p~g^W@ey%=pz0&cj(Z@+kE4RM@9|yiY`wJ+CdaVy3^y@$
zApxqUq@s9sC3=->Us;H)FpBGt`)1$!+sjJx)sJ$>i#wh(oAUjz<<il}p_o$U*9-4U
z=>O=MpL-DixT7;}1(f0psHoKd{_syp1>jpv0#=7sZ)0<An~J^Nj^&JLnECPa4~7=T
zm=IPN62u_XveF%^m$Y-MbC*@QAgKJFK&E^#2_f=7R^Y@Wl7fdw4R-_|X!&g0N|C-~
z!2~ifd<{qs9i9Cb|02;X0@n7Dv!g>a+~8Ij)UfXPnz6t8#Ru;NjdBl4lUA%Fk0Pah
zqv5|Cy4Wjh>cA%r>gBpW-x%vd=I{4l8)`A-Q*Qc-UO*izD@uYORgoIS6l*Yl{7;3E
zTl=R_jgpuc3}_kzYM<Hs<%j2S+`8isLwW;H2N2@@JdXW1B=6N4R|LlW)KgGE?x|~G
zP|n48D01;ZL|wvvQ(fNtsxfuIlJ$XS`(kCuYe>5F=2vzph`+ElhzZ&C3g0E8ZV4ml
zUt;6=vKEhpOlBHAYQSAIw-TW0E+ZZQ4vkc8!nmK`KsiH4$ApK4P#S7CUp?*OVOhW1
zfp7FAN-Wp(1-(-hIg{+!O+l4rg$+sFkg5V`pEUy$&8Zz_l_23FUO-ZvmoF=&Ic_av
zjbxnd$BMOMmPn=znwLIKVRHaaZPn+*sjjaNh(c7k2d60@inRZYB7y*G!B&hXJRMV5
zJpYAhH*W|xuV)tmLS4D!)S$eN>1y6DSQ;vh*`AdAb=kCq1pq46&J(jbil8i5%SA|)
zNsh-R@+;~wDHhB4)>Rrp`nhF(iQD+?(JQX6Fo9Z^<S$gMB#c0I@ud~43x`UyT9px0
z8zQOvdE*I}6pJ#kz^#A!-Z7*MY3}w(<Og9x`Di?tB#A7`<hur&@qE9sC7AbBj+T0p
zHNKc4P_CIQ2DFAt>W0UlfPH>LLlbO8Nzza;poWf5Uhj#=eJJ`n&k^kY@ngC~_I{v-
z+hg7F=`mP#Jdz9QBhGriEn-8>PC1`cKA94@6xwKtj;2Bg1wFFi$~x2CnB25EI{@n8
zj!MfacL50Ozt>tMNaBC@Bg_@3sjVgosvbKxx_S?hI_X`;U;B2VfzN=FRJ`#@gK?Yl
zMIqKyA${5%S5)jvYybKx8(joj$z4>|2Q>dp%djlRrYQa(LTH0%e9bA1%Tn9sg5+!v
z$|JCjwiP7TBX7!*=SK)8o`sSw0;@v{89_}$<%gBU$3cLisZ&S#@)S@KBznOlPPx<9
z;y8a-d_xm9V_-LnUh{|oXgWRFWN0u#FCc{R#>*qIIG}lwLnv&%Yy?p?;?>s0Xd@Ml
zvP@(l+n)X`tE#0ohGZ0o_Q5fEm-BABW_%k;ISGNF#UkB_H>}jCBK?Q=D(JnKk^vO#
z0Y*&5BTDEsmD$B`#%M?GDqjK>51@<uL0YV$=Qrm7eU=k%H=Htq*+B|HrFdbBU6c;g
z0!{BmB!LP-X-^1`!Cl>{Q`PEaRvg7JUG&@9(p;9T_og_rnr0?BjJ{NmP=GMqqw$GA
z#!R<1q_p6HVoY@+Cmju=Dqj3}R^ay(Fe?P56|guCii0z*v0y>IYC00p^h}dAJhwf*
z%tl6+2CcDy?<=;j5;%E~28LjJz35+*w^3m8gD3yEm-E+`x>@DAiN7=a@{=x%OEwI5
zP4Ca!W=D}NQ;$WV^!PyT2qYtrCn+ozg9Y0&kx}{KT9zV~2CJVLpx%*WnXpozDgcdB
z&W5mL4Z@3vs+>G7J-YTU_WEpHrwJc~B^Uxh;2K?EIO??@8BFtajYhF}`c)LKPNy*E
zqj-%q^_YhNEOj<fXORpSBJOPnAw=?3*p67-2vwxwM6o=xT{W^u7GkAA&)5#mv3^?N
zQjKHFFNSzg@K%~Et~>ED?~3c6?R@Y;$$Z}Td0dfOT`AYUaeA8W&0i5^%dGE{_RPaN
zAI@x1Jf^o@!-C~dl%yoUOzD6RBv7LEanR`9nn2-Ft;q$j&CcDoy*A$_AEsFm0UI<E
zlep9AvXBab%l*KaD!H22*K!qcL5)Op#jeCFmvXl*d8Wu9ENvsMQc;5my-XYLJs~(-
z&OItwU)dl&=MA2{6;KDIND!njmdG~e&4rQEhU@Q4rKTFny&}jL2W^!u{7{IRA!NrW
zF7KFX*Gb+_Nz1~tbzirN2{EbGC@HjQ)XmAJDVl}D$0sU4(E?GSl%`n;Btg5>S-PCE
zC5IG`?<*1*E%=I|l=N|piBt*$P%+e%m?bR11O1rDM<{!j>lRgy#Z+Iu90mpVeRB&k
zpq3Gem*~j8*0L9yv6_NfB6khiEq++CEIuqb;dVD22?y-Q=xc8dPY)DFIzY}V0tcbK
z6*w^Qmc|nbREG!~*Z2NrGe<tJ7k`h*F%Rg`>z?gJvlvCYa<eU06Boav3Bx>D&M3Qo
z{ll#gOuU$>-?<ddM3xmiV<<MYq|KH6v(%Ozz88n=Z+1(yJ?kD5mWVM#90Hcm@vwpv
znb@IrpkDmWUMN3JE1WH3AFI=>KS)KPB|7h}DEJczQ1LXScf{A^Oo733u_~sB5k9mR
z`EQolQY-G6ooW3U7EHY;WL_{oghrZ54jdv+9R2CujEjo$|GfZIXno!f58hLgZBFJM
z3{YS(&h0N5T-~}xk|Lh$v5JPyK0HpALtIJTYql>-)>E)NQ}-FIr}tGGlb;1wpMAP`
zUX&2R)*PO27?s&{Gz8wSw(l>xZwCQ^xTGp7j-6U_FpVZhI%_qvQrK8}bi}N|7Hj(;
zXip8!@3;bQq+P*Ztp$B*bdB^DqYd^+37=r#;o-|Q>Y#jv2))w*^-eDKMxHr$S^KR}
z!ocL!I9$|{ZR!vn=&j<G@O5L&#Y3pwhi^Fpv$gt-g*XWIY*I>x32Je<SEE{<nDT|D
z5*`lv!m)fBD$FqdOB8ibA8go<fN1pUK~BUU1Jie1g9m4_u6hK>`6w!@>@UAp0iyjf
zX^ZGm)X>IrWvN7ksi{V6DwHj5`@O87-OR8aqo7Y~Nkarb!oD7^8UV&B+zyGSr>Ffj
z*k@2(l8~L23o{TuamwmT7HCzXyt0;K{9eybQ$ZIrV{FCcbs0CcdKW%Iz>w>$<vGJX
zvS4DINV(qzfjboF>*k@?T3s($C#ycRE$kj|tw9T)uoZdZ_r*AuRpXnoQ^PWddhGDG
zj~$-h#Ej+WBX=~AloTZFz7auG!to)Pz7wz~R1OImenUgJl*V3OFT!0aA?u=P=@ejv
zXgYIt0eU5<8eoat(QgJLkX1;e(*dY~^wnVerE1UBzaKdVza+VmbV^e->=v8TpkjsS
z%&gxsFG-Ps9q%T?Q0p1jI!S0wCuUgrPLr2Gq%g`Bp*S}FnUEyc5pah(IwM5xj|^1F
z9VMB2S&_QMmEqvMlaSJI&gW%#arqIgaR=vFWEHe&fux^#;W-usLLNZ!(B~MLh8uqS
zHq0ajqK(LkvXC!k;fuBi_h+8y^hikFo%c?|A5c|Ov!|Dsq!GVw*B<Fks&#oG=Pfff
zc_Dt4<4+k_7Bok-xKi|IJ+kD9a=&PYR#xw<y^#Z9Pq#yx>UtLICvnas)(dywryEG+
z!o2`ZpolHa{s}o_=Z?`9tpq}=Zt;Z;(OLmDf)Pn`z<yA}bT$}jto~TLY-!C1!b}*b
zoJu$tWA@p6i7YYFHV801)0ADDCksuk7B>zN-lHBx)8%j>QPHr~`|tu;EYC2NvD~O^
zqsj7*R6$@eBgrqZ`1$;bt!PLk$tkQRYgDtrzS@&IFgslg#V6L(u9Ily>8ah6Iq@M@
z(R0zb(Yr@odtPh0TM~EOD$DLpPhC>dU_kiqzF?zsN=3lLWSc6kC!^5H4Di3H2@%n)
z*@zI~&A}B)<Y#9W7UrLNd1HJ>7}q;|5G|6$z}Pr((r`ygx+v>&T${d0G^q?5oA}_o
zVff)yFyg2H>G<#xg4UBC5P@PaE!b&80oBB(%CuX%|0)Wuy+$ajph@jIvakebm>Wa)
z89X3Vqkb{Ev%duMYV(v}9C0lM^p$K%C|JpCB}zsT$YL<iwHVfA>|!V&!{^A<hB2cG
zqcv<=F^fO)s`IyOa3{OAJz_-%Z38$rpw+m9FOLgYi%ocoaC6V)e=!E?*KqkXU@7DL
z66nPU@`NTJ!OK})#SuZT8}y=PEujeP&x4N=j495`3^PKL_ELPATDPsQ^%<m)g@Bgr
z{bFb)?5{470SfR-)4O}keCHPZw`;`620;|#==*jQ!l;xCHd-BrB%KsuezQnc4A-wm
zk{bkXDy!P3P#aetv@@64?Xs>2W$HnQO8MCok4kx#_wRcCoy?eG_J}WFu2IUY*Sa&L
zO()kh6DD%KbL;cLM=J3oBEJh_6=wuD#wmm#rl3-}Uf?Q=?Mwlj=W=*|uKvuqvtbrQ
zg0>e?Nu>@dbAR6hSDHM!(Z_iL5sxxI<*Q82gE3SJZ~hfMRr4$mIY=7hPs~)$b~tC>
z*tu~&v-fxBub;oc6%MYRCH;$Ujw3{DWHKGPlQEL53KWGeH@CPgm{A~D$^FfJ%AHT0
ziPO_9cH!cgmtmk)?W?=IAqEx!G!jTypqs_SfxmSD1mgIhLlVA6+GQ%28uJ}8xA^bC
z%Syr*%wGX$`ht4qp^DRc{XIB6F$dH$+s^@O=r^V@V0|DxV3YCa$Feq!jiEE_Fz6GU
z34RSaeZS05z$m!&@Li&dPs@6Lr0z5-43@sg2`wzJW0V}%9$R@nvI<$YPNcM`TtDl$
z&o4S27+dF3A;YGRS1x(7Z!F8>TQ^luH#4rlIx7ra@An*zf=@7EY`pgcv|C=kQ-bh}
zZ{3m|tK5bUq?~;5oo+z05@03DhoaT=C>c`wbrq<%=p1DTZBCd>$3umoA=liYR7Xn2
z75U&9MTzaHR5m0O%(N?$7;Gnu7@lW(vRKjnG{~WUNHfX%f_F#XdJV^a-$Yo1heQ?X
z_sIjxrDEj2;_$xVprfa+X>J|{;E8sjuHVNKtU%>4W_lrcLA|FpnrV(<qev<jn5|MV
zqLrEtF_)%s5+vOwEPmR{IEg5cX;f1yMW$)P#uABH-UswTr_ZEj?1NkUDV_T&l0E!q
zb`-JE;a^CqLx5R*EER;*p55&CzC<JuHjx;97g5ED<vj*n{uFC9ma$#pMd%LV+1i5)
z2^jy9Um3qN4kk})McC}m*c(`tA1t0OWTSAS?z;POkNTLnzJ-9QBt~ewyPrKt{9qX?
z{9ZBlf}6sBQFkIE-Zo323L00c?TTSeufawQ#cm#$Jd(Sss?D!ewkvtY<bVHMHO{Vr
z%vU|=M6DYFl?K)RFT!T;g8#u51gMbooh+*~=9ZH#CO|skzu~QHzLrXG7#l~Iz1;5&
zYCi4M5TdwRP3nwKMmt%mItrz&6clv<+E||e{O0KEE3du1y(UIX0qUCPuN5WYu_by2
zhT7`t>5<k!8~_9XzjcMbW!q1|U%#nu1RFE1!1ET);xhK&N!YaMMhm=UhJX@axm@iF
z1R0n(IKHc>jPf^PrTt)@ev#md3>P;f;Hl06E(YNPF~Ekd^8@vsuRZCc()G(x=9A1U
zC#o5h#3O5Abg8sn$gr)UsVu3nZA7yN^F{?xFg*;aHq&^RW=-!(ciF^)FH?dWp21ya
zeOzaEOXFWsY(Ya?xq7gzRXLrqHPzJ9#apLYhdVy*mJniUQa&;|Kej~kiM>7c37HB>
zr(8xXzjxWQ_l9II!kgBXtZ3Crt`LfN$h+)t(3D?UbK!JSz}+Acid--wiw}P0SFZVt
z3NO@Y(SgO^J+?2f7Ss!9xBsFH5#V_d!hH&q)JZ5Cm=z}uTP6b>+4QTM=yTJ~|Bt)3
zj*7Bv+ehh=Zlt6UP#6j6l2lS0N_yxN32BfnK~RuJQ0eaOM!F;<6@;O?+1KFneDC|M
z-&*^>z1H4))*5svGjq>%U)Nd3ah}JenRV1ShfL;*q4Fzcy@$^ENMZ3KkQj!Y(wP$a
z9G(i3CdY6xEDytqTBPoeA6*x!iwqRTait%9!1NME4s6GKC3G(hDGZeu&iYX`L8&;q
z_{3QxHtV_%O!!!!uZZU4QtgjdE_TXUNWE6l)V8=A99Ml{`eloYO=F_>T7^GfIlm%;
zL6#s=F;c<aMbth-Sl{mD$o7|h-1;b`RFX(L(V}jM42q@Qb0saF{4qD)1BIil<FFuc
zu{p#950KV@NfiIbUtkkoxOq1m2~%7m;hQT)xUH1IU8P-Qwli+RopXMC`+2`@Xo`LI
zDmQe{nA3NkJ*&xF^fdHRKE@f%do5wFz!Vpfr}@@!vefVM_n==B=}g}`rs0zZRZ~q5
zFQ{mfpLzr=Qcr~ZRQo7ns8L7!{AJ<723s{;wNWK8F=n9}%Fxd{BmHu{D6jM3TLm9l
zsbxrwIcEvvdRkj5YqMsj+xkZc42_}^$Fc3t$Y}2R27{t`@E!@o+ht^oNQ3MdLypX6
zN`B1u+Ob_MRXlnZZCWTmBGydDbw6ijHTlav4Gx1^QtXJO%o2yv3!?8;49}mvZ+DPa
zS|*$Fc{3FqDaWK@P#;s>uxFQ=zHlg@Ib(1k0^u#v<t`%hfBG$$Ni>95BvHoZK@fGG
zvjJ(9U0vwG;5^1~FHY6Pgtu7O`8i8Yfw7Ne4CoS2A$yL~_Z*`GEtWt~Bc=;omd2FD
zo4p{l2QCq0$vJDT+idnN+)e}%<D3xdE~(R9ZWvlF&MGe>YGRy9gYG#_Qyn|bMlm)|
zXqVjZ`e-e~`Il<s!SI*v4;Vv{O=`w2`?hyaCl=0?l_F(0-H4clY{RDmSQvYk{b8FP
zMU$n05txE%@qAF(-;Ef&u~FP9iK7Gwu_U(j!^g#`_ooeXoLXG(;KrFAp%zn9k+OA(
zR&{f6BxhS}y&Elq%|He+Y<13sM)G7crE>N~Fx_a`!jT(1dS{7=(Q?_R{odiXGX=hW
z(7Wx4b8?LqNX-}tiF8ucF=8)uL)oaO;raUdA-ro7rY0Qte4+Sm@Vj1Jfvx&5oKaKy
z30>JMK0F1^Il6K($(`0Wwj=x2*@HGs<hd`?AWj6doAAk%^&lUa$(4IwDXBiroJ}>U
zQ_-MuHyz>x|1S9c(l+yaqhnJe<!}efj<{=H;}@Ey$?NvQs8`o9jujWvk4DjRV4DrW
zT`mKky_U1$$2Q+co3KmMsQQ2ZELSzQf0wf}kC<jW^cnRZ6%GGDY&{;zFa1FZvCz?;
z;@~Sm)_`2-D;mq^`JFP{^E9NC(DkQdRIE>i=6vd;BhNiLe(^)1czA3W{KL}r{P8h3
zTUg|uhH>;45DT(9mJ{QYp2k=VrUv50n~yt(`T8-k`Qe8U-hG-+Ck>HXeOp4x-XEJ~
zRafP}qV$XS+W|);gz+7iouy?>+)3oMBr_U$@+tJ*R+UUnR;(8%ZFH!SZetVQ$mDRK
zUfb#~S0y~C`g3l+gcs^bE<gCbzGr|G;3J<^MO$9P7kMj;t}m+Eh<rbm3i+daQ;Uw$
z>uf0WLY&n&iczZ^r!24=DTL!;!h`49QE_^2z`-vi?m#w<E#xMeWN0{9^A(t#gA20r
z<SR@xRj05Q4s5jEH9}6cnYm2s3V49UKvekNbXkP=C#`eidRdjzH*$zEAxZrd((?}J
zul3Ot?I$LBSwHMq3EZ!z@SM(*-e+T*ut^CLkr3QN-W9wfo;#&srIKa-*5m&ERQiR4
zk*UtzV~TTalNaL&e@d{QH$4;(R;tja$;`;$z~I@}@zr}}`}G%JET;Q%MkQOMN?VaP
z_lGQ#Q3WL{uB6FiClB#M=de(5<@ZVtO&}j^AVG=KsuUd?#jRm}x6LLBq53eryCKn1
z`o4@Wu;rdHVf>byHHYwo8{zUf^VXj|l8nTC_qG(1DE6d0?ob|!wFrae;rt|?`Ad3=
zol>)2Jtab>3@h!I@n2WNv@F8~pN~uClx}zialVyN>dRA4=Xw$=sL}ao%x&nN%9Xy)
zXB|aRtvH(cy5ou016r*<1;mpx&LQ-lQa{BJzq}g0UiXXJF2t@cvLc?uX2sUTp*ave
zh<w0&x58XS*<Fk-z2c{<dkx`hw?SItQ?=~7Z1@47+03X!+sm9)h-Mw@&4*SsjW71T
zqgS-qkRVbw(MLxMe9x~{&x~^~>?4jvmavxOqO*I*48vr;<KHpg(9!Xs*>aF)3=7P{
zdw>zr`6yM~zBmrqT2AmO4cUsLDLWfUBMZ9JoCZC)GTJ@otD1rD^W#Twa}P6N!^f<}
z0W<-|7D;F2kD}h&sj4Pe*XgoKjb)+T&=ka9?To-&zGN92m8TmkQKoCJJS`lin=+cg
z4cS8B_`!-<vTt#9;A&f%pshh#U{&~v#DiA)-2=?pbKhleKG;O;l%cqnL9YDpaP|H+
zF(jItuUGwe>M$^n={v5OG#yJ8c5~*Pc-_P~!?9~6F-;0a^S%pP>#XFjcdV(cc(U&C
z8AZHnxSpYBc9cMmpfTFMKXX7rf=Q{a#y~3hBy-Ryox!eC(kPP0+z&mNQUzUwq04=a
z18d1y;X}fG))2I8lULYZ%H<N@i=*N*n3z518d>=P(s|2Z!dE#Qn@z$Ofl><x$AVb?
zEKx6f)KWTio>+^<X?$s~U6OpJt9w6pRB7aZsh9biOZDYTreQt>sC40e0BMkUuci=w
zepKA2pvBTb<7anRAqIM1^EhlX#`%TFzv-z{p)zK!3Ciw@vz2dtkD7^}pCgaYEc5q4
zPuhxnLCT>aD3unPox<+9EI$Ifo6jmzaVaE<dx4Ix(p8wW^2gCKA+eeDF%^XM`FKSH
z?}4Nwq3!6`czDydg@e|W=WRkECn2hnLt%>r?U>Pll2FYKQplO-HaoK=1qDT7x=2`p
zj)^#hTmdBMMYv9%eJ!amr#q)YHNIqI>1;eCm=I+ftCNh(=#dCt$LG-ijmYx#m4x|_
z<H}G%t4E%ib*AC?6>N$Rh+5AWWYuG0)$YFimD8P;_d1{1OF*M(E<>4|F7N&yoCopG
zFpCa@p1AO7!08^+9~Bp-Kh^N|zQb@#@*9L$M9y*_X6-&m;@+2)!4<@8!r*<tKx)zS
zvk&)3FqmCdwLC<REmOB$hqwgKbb$2en^`fzH|YwL{T?g1u%hlT><98mXtKt~NwZ$`
z-^-9RyRizpGt`vKO0y$yalhio+!^MXab@OUv|uV=D(fiu@Nm<SYw^_&TNffKl3a*T
zE?NDw`V-SgsS;c)Qp#YRN4}Lt9M?j4y|zPSAA@Fe(uIs=hP9}4<sN8kycj!>e_dv?
z`s*Y}o(<*vt7dcn5mQ#&cV1~m$lYkF!GWdZ;P;sxA(W41IsAyc%yF^(BSFJHV^=6d
z`a?^i0+;f27zSLk+2@I9c#}rTm*X@_rnti0xXyS@n<H=3W6BaMu7sS{e3_Py@gMn0
zI27}{nUz^z6BDF;4;$dyaZOFUm(hnic+Otge=oH4;feHmDb8+3ac?W~^0cREaABqG
zPXiB_i%Gok<HR22mkNQO?KF*PA;{`x81En*a;VOeI1uZ~a>%{X7A$6hAfbk10B+rr
zMebdaK~)=JEBNz^XMC?@|1FXeG)zhX5*w?Q5KEZJnmFUh%c5X)JV~nV%A*!do#G^z
z-4yV)K$@5#NI|$w{3|JI!SYv8P8;0&xenU4e;O)g7VJORV(SO<D?I)BdCIs=J4&PZ
z10|_VD1_K6-UH3n*h;#iD0n0(QQ4p~_a;JHQ09?r!l77snl5L`Mf)-D)_K$HHjyK{
z0G#M6cXTMMr9w<|b8TEvOeHqBf7Y++p7E&p$@!j`2Mr#secL`BT>;B7U%Cn2WYt#l
z@iRqSb=~WtkzZ5R;q9qNIm*gm{o?Ot_*r^ix4zR#w-D;E8j>n*pHNjIa2E;g-VYK}
zQLC{pu++)o6H5IVYqqipUH&i!V^N|o{J?LvV(P`t%B^<ug%n5ONhGbc?eAQQr-B)3
zD(OQ->~9l~CuVJ>$UYJ&qNQLaiBh<4u*s%zInLr|n4J^q4i#7}yf(9F{kWqkaV=ie
z^XRO+aPI~2#pj^`hWI15x}{tPB98)bZA<B*Ebgar%1?wOK4bICu9qmDRTggktfABR
z$Ran+Q>us;U>zDAlZD;!LoMEV+w!&PV7Et?p5uf`P|nzXBYEIkYIY_D>~9Ys33?+A
z(Qm&Kjd#IhBS<vHhmDnpX046}x1q6C(JuqmQNoqbTQsGafAtY3M+<}*thHlON0WCi
zCl$wE397%Jeu%Q=E(FG;eZDYf<ecP1(PYDqjY`qUVcL0^PEt?>C3;S{`C`N!tI)PU
zs=$NIojz2)=Tz&-PlZT#YQe}4+mB%%K1zE~o)RiBsK=u-<S5o_KiLwy&r(p?7p%(w
zA#*aZ;>o6?Rvso-Z~h{LH?}TAa~HyFn-)abtJsu62pLqNba3~E=-`wp+x}jAD&zfq
z#w(un7yd)CXmt|fN_vfWmaNdgVI<W?KQF=$+|yqN)TU`aMDx=dKGK&8WmwIO(81VK
zm&)?TAEUo<a`gz3{Ghb85Qf`}e2o79?or1r(IK=w@+Bk_NonJ07oTPv3DaGa)Tg>G
zTj4Cgvn!vR2*z9P*OW7>D%(opb`X}>BYi(6**dazBxR&2e-glVRQ&DFp{za|E3rOn
zf5lFA@e!%9{$^`+Z?k_!C$@iYjeHOKYm`v(7P6c;lkVd`j+;EnkvfB!sZY=SR*OyU
zQ5;vIGrL0dJ`6r~3a1)ZoG%=gcUByC-R{T}j3U7yuJ>n@UG3^UwD?s!<TzhVr^<0_
zX4I$&vypp~WGe2K*qHp$Q~db$+uhQGD(1+8@j*%)8%1$r;_sh?YeN}`k`+SnklOQ@
zgC`b}Xc$&!7-XS6WEK^fY?&H5Oe6v8Y<@<>Uom9N%1!!k7EHn*s~hj9CH1L?$<4-L
z%S$A>h1^;vbX%y5^4;2SN5)092`)Q7=NTsX^yuLQ>U@H-OW}vlJl>4D1(V)mX}z1;
z^9@UR*mZ_7K)X;b8Zzou=-2g?3BRLzfO3syQdjyZbJ%#{B^OT)hQ9AB_L|W}@6X3f
zarBYzo?_Mo^2ZalXMFQ}O=vUc7M|In8*q~Le)OZBfl1ezBRb_L%Qm<D`2>|~M%L6l
z-TVt;-Y-frRD&j8VfI#J&8(<##&-xUe&Hx)Zq}ADGs;*|zN1LY$DnCFhu7R2IY!rs
zw3-ey9E&o~@)uftD)XU{itXfGqHRUz_pdOIE3(P0lGoC!uX@X;v4=E5U}`&$g2Zj-
zLHdN8oDjL9fdu1Uyw8h1!&zbWnbfR(W9plmM=Rw$cRlAbJvcUnjZ^j)_4cY+y<Cxm
zCRW}l2X~$_<<QQ{DbtDev_vOqX%o$l#sWq^e^UHJvwZEFNFzAmcdQ{%^{13Z<E@=z
zf11OCWz9s#F+8y+cbUwwwHXs!y1k=SO_}8cIeL6Lk<)F^Y3z#c?#bLOKqkV>^-~+3
zcwOv|b?jV-s?=ky&&$wHtUv7jp;~D;ZOJ_{<}h-IXVsiB$@{`9(}au<TTGMM8Rapg
z_>CYI$Uc-=BqRH#opHpA68L^#^Kf92nb7V34rjGtn`c%H*$DiC|Kp9Fsdg5$6=Ee3
zm$y}S@Ic3vN$#)%MLK07G&Qj^k!roQ9UWz>Y%yUg=Oq6dllFrP$4nzLSMmZ<=Lwft
z=tbb<Mn~&rZJgNN^Q@UrD6j2)%Ag7h5gG&fm5;Ih$#e`y%8UyPtB?I%*r+nQ5hReQ
z%tZr<v*-@%AJ@YZn1X!gu?xoY?V$pts5?>Hf-Oo-Am2W}7~bkXJT|k|PZYk`cqcgR
zdg5--t(!cQ7=Ywt|1(d)61m3h^q^3GA;Z-*R^dfK$vi`GXPJw|-K@A~=U>lbvzvf$
zQeX98)c~*5(Ry|NZxxSt&1kxikoWw-*S^QwqVq*{O(!?-i>o5z^!rZt-jl?BkdjoB
z0=0?E&i;D=yeJ&kc~e!*le+3J?Q9%gdh0b(MCjUW*;OtFdXy^U>e`U06Uw~c!pFrX
zHlajCQy84~R9@J*f^Ag}iiZTsQ13A}rF*AryW>#<?Wv}zwS%Bl-#}D4+80~DPGWt^
zw%uVJN<$cd=g(BKcaCdU<p;KTL~12!BAP6ZNy>Kq7(V0u^4N#x$K6xzChAK<oZ_{-
z;|0`#&L&=A8;hd$sktzvr&l{4(WA%IuZ5`+_j<OB2FgMybEe+db8{uYJ}e8D(+!=T
zNHx>h-o%YaV79m1oZ8Rcuz7|TPKl;mtA4a=UZz^Oij3egGyHtR8W#GWM^8$&+B2E|
zdaQ~D=`#M;<I&pvablwXJYs}YGvfdA`q0qt?l_hI^?Di8L5&=i|9S&Pr!?sY|9L$T
zQQ`rq_5TYlVavFSvf+eO_U#EB(i+WRt>IqG?Nx2+H=JxaSix`^&@$OGXSbFc*}@r(
z&L6vx$_5hhbb%g{t^~t9s=rInlSX=#M_^<)N;>o8$rFTfdWR!m3zu;r&!D`#T@}{>
zPt~L>BvM%^D0IsyRRwcLOfi;+pTBc>m{3+-jT9Zcj#)9&=<iK1rT?O4LvW60@InH^
zQ6EWTbvP=ilk(!M6Sfs^?B2fPF^4*7B}FJdMPv|R&jF2}hN@WH!JtTVOiax2NwpLH
zj@YC%d(iFm^)$(4Owu`no{E*mZckO6cxPeSeNl%y8|GZ$x4*EdOVe$a6SpRmsuI%M
zt0)Hm@cDzfsm3VqreuD(x%Wr;Si&;}4y~nm{@%jr$L!-X0pXd8Km>XPkWHC~oMM@g
zv+B#8J97~4*t6YV7O9dwCT`z_Eno~in5|>S8Rh$ymv`*GMxzs73Ag`y>wD%^$>!-u
zyCG_ZEW)Cq-^a&EC12XU5vI+{JCCDHzkRFuuVd@)&8Q@)pY!YMHX0PuKDLasYR_+V
zvE^rxx?Kv>g-ozqHYkN${giZc7#p1O)Vevki2pJ@;im|7-=U(SGPbtHf#J5dx7%50
zs*}3yU^BdDRum~K8Cz%n@pofShakSwzNm$p8|9AJppYL`sty$T&$gdd4ru=Nftg{*
zX<3jpY6bA!V7OLERq!c+=IZ=BYu^tMh-(e)X_5jf{VT=hjsr}|)Atqag9k4jAWlKq
zMbg4(0K>#)3rFDa^<Yt+b(h8Mwab_;7{9q9|L>C#WBB0z_VC>-2|oUp3qbV$n}hiO
zv@Mw3QtPK&iG)0+gf$KOiW*1O4pNr4QXc5Qltz8Uw<P9=eZ42R*M?>P-j>^+zot5e
z#&VA^7PH=V-`sI?z%qVNLguD?b2f$a@?xmD`bPYJ-?tz3LAqTDj*Dk4As5kjyLyXG
zSJhs)Zc;bbiC5i+U^2_~#ecuOj1NQoZ0?cxVl>i%Zv&+pd!X7ey#q%4$qqNB55}<X
z)zbg@`?!^nkZk0WAi}0<wl}ts?x~Q~`mThX?cS}wNBMRef_=y~^X`NOGsQSki0lv2
zf!_)0ok)m!Gqnv(P5I5W$Xl`&bC?p^cTBp3#u36c38Wh&qe$$V36lTyeY~e0id0Ht
zIox503?bA+=Ef|-Vo15;MeC+`W4DHUGurIyegnChD#l`9ulQfi!SKg+QEnxZFK&uM
z<g{d?`K0{$kT06{obR>QV&P5Y6Vm_sm>9!1^}IeMWYRUnHD$gx*)0y6i(+Dn-#iz8
z{MVh3k@36H&Hi5=r)Nd_{{t>4nKuJSa1f$Q>K0UPIdxGMR8qoRPp^R;Uw(DF7kutV
zLZZ*9FU4r2Kj(3zIBnTEEl~2VlsZLavd=6`2Q;U@Ir8+_b-6k&y3%UGgyFiU!6I8d
z3Bc$upc}v=EL@jO0KHu)({zLv`aD9qS)Pu{-v7D_9&nQE|N7YC_H&NqlouhEyVXUI
zU}i-y7s1;%Z4LMVq$;~vIht-}Md{+Hb|HCE*+Y*;*I8gE>v7G^Z|1LEs@SZf^78&T
z3vUTzvYX6S*=<%Y_|Q0Z5PAL*{JWWtU7T3QXJT4a3R4@!-M;1C8p3&d9zm+hLJ<+Q
zu*^`PGprDcwj=i8!-vRs@6_X2CUt>F>3>vE=`5ToFrJi>V*8{dbz}!_C4y6q(VYGh
zyM9?PD~}^Q^E0shYeFy>4DpG!;M1e&if^}gPm{GukZwTG(k6-zPs{%N5M}pmv{6=^
z@G}lwq$z<j(9a1!r`|7<y-i5Ic#26`BP%Nl;K+hej)O5~R8~nSJbZi={oVwwkz0Qr
zpA?Vn?4}0kN~mIL@d}iUtyg5gpiqRs518w>wFmoapr<JY64h*8iC=B1vkh3rb#0o6
z1n%aX++NV;JB+CW+6IE}0G8H$7H<|vs4(`ij7xEhbd3gRG(BKTb2Aa?n)OtTWmZyA
zL3y9G{skp*y4sPZI@+T77@jo+7u%?9edQ?6#ALx4?q6C8*{)rqLwFll8A6b{lE!6R
z-SaZxpEZ|tdNh95%U&LCiC0t^XGUeFQzcvdcjXJF;6$=VLx+M<VVRa8HbVlf<m1f|
zjVD0P=r<r}BS7=IpAW&BI@jwj3);(*2wYf-*O#m1&u$I=SlFEMP3<DtX4jsZhKr0{
zmJ9B-piWm<k?6!jfP)5D<P5&>-{b3U3+}rI*MN4MQ~*Q5iI3dKUsqQ>GtsF#kW>!t
z7VP@+49OPEi2-67qg$>*7n|k1Wsxy4L}1qiJ4Fdb&0=G!MRU$U1<eY7F1hWdip_u8
zvgYX6b}l*c5KD0aE480V(HL6nbCmzfT?GoQCky{4a-booOyZPIM)7(j6AlSx4=<UF
zW$rD8s^4D2MNw;I8Ag2thVKr!Y%37fSU>XkXx0lRXihhIiy+#A+|xc}S=Fa$#xjFl
z^-d-GN!pquZORsV#a*wthNdr8IJ<EE@K+XV=LfEZ{0r+~Km<P`Ol<AxQMefTdk<vj
zW>fA|V%+I!Jv$sA%N@8{<qy)ib@tq)lY>~#FywFeku^wVf4N6F;UmHUAoqb}sxyfJ
z8`$|X{ZoelE&ic?-{`kzRCILaV4-;1N@$gzS&GlntPO5%I~)>9*Qc(&@hxANhQAn?
zU*+a}F*|bJ7yzT|FSDHzh97w-DJzd`xz=8v!xhIWu9TcEQ&qCTAN9`bqsK>??0i0I
zf9=BKfm$`2H{Z8^Sl89w{>!mzK*r>t@LLS}$IT(hLpnAfK+2fiB9>LZ5D^M&>FL2z
z!@REw1X^?|EQun3ZmyeK0@#osg5&D1dQx!aG*~I6a-auphv+ALbk%^j`cP;zPHiT;
z$9KQ)lw(mCDUI7&PQGt80%^Bk$Fj}?E3G@_Q}7Y^5onhbK?oG;H2a7>o_mEle<6_4
zf6yRSyjl~_Z72>~4z#Aa-CZ$1%e8`ZMK}bK+ZuJ0Zi8r3ht)>k@k9SqDqKg5Cxu(H
z)hj75YwK`k#hIw0Z_hrV@Q8~K#y@yDv*ld>)0NZ!>n*dQd7y5<+wViXuN|~9G)W@y
zFciw`dtEK)%i9a47y7qtVe&BYWEe;PYq`(A8?(8d?TfMvIU(fR8ct~3f5jjs<m)+)
z$l7^7;^UL4o=x4;2Smp#yshxe;X~|yF0?|6MXJ15!Oq?{b@dDp2lh1C$uoQhsH?Sf
zqFhqe3#e@NnAOxDk3NmGwh?)Hyp{_vbK$r#;HTa|sHjobX=|(fw^dgyxYCuVd2#ye
z-1cCV`5U1t#tV8qiGt~_MGunz7<%%FCUj!E9=k|RUj8iG_nZxgh__uV045tkQyTnZ
zZeAX(gxjzbX64(o|5?Opct80+yY;U(_31p$Qz@i9b2tBe8?t_P2}sqg7m2qo>1Pid
zF3L8GG`6(F@&o5}|HSv^MDnK&sq|;|aQSy6wsOY<KBBXOKE6W+1e>LsmYWNUHudH9
zQ-fW12Dcx2a*50gExMD+j->4@3U(n5BslT2Lyc>o?WB|22f_{d3z2J1mm_))8n!~6
z1|1jQVphT5^27g01AB_Ljxxx3U{Y3yb&YA9hP^wUomfkn<x_SX1Y75iYdsAZ>;OI>
z__1Ov$OxcNbtu_jt!wHGD=syzqj@nDJsRI1!_a2FKEEVOC);}W2j<kIM}eP%7DY9G
z@s3I=Pt8_qWv@9sd~>(t3|}}EUnhPk#ipB>&h>{el9G}V7+<nV#Y9ISxQLX=&e&f@
zwNQC$+e1aG8ac+3^x4~;k@KW)Pt`GcfbsAUi9KG^l&7jXDPShj(_4&W%6?CN9L=nV
zgmADoKYmaGEBn=|Z+KX({zfp0zH+N|cdfMTD=XnlD*o42wS1j7dDF*6p?s%Jt@gRo
z3nQD|^%`v7i+GdXJM7l3dV*Vuw@}8#)c>f;cDfB2lcoh&i`t{fmpZkt&1M^qfIwrs
z#MPLDT{XPXVyP{t`IX!KTQ9f|w=Kr0yjvb2K^Ayxk&1teIL!WB_pqLyYItL58ER16
zz1$h07#oO+FZdbH8i)$EgOih!7v32`QEu8x3{t=u^`LUpueA5WGsWgxPw-q_w(_3k
zt&`de*5>6=(V7Lo2;Y!fZ*f%2qzW6mW@{ci{((m>O9bvmG6R}NBOUPd;G6fwae@y4
zGs<`b1mwUe7%CN#Sfi0uf1hP_am9b}HxP``pXGc~nEH<H_CQU~;$2t$)OpW;dbxPj
zIwSxW{ASgcDH{Po5WACMV4GeLnWZ@`#+q>@HTw;!Nn4mE6mYbTFTLxGt;5O_7VOK9
z9d+x=V{EH{wOG)rlFZ{05NJT5)PMMlq4n{f2<+sifBj<aebok%&BNr5@Jzq!`&1`?
z?#v=7na+%asbQ)qvrjmG<xhnww;bl|h|+35`93YsQbrj`z<gCy*)uMsMbjNf*droD
zAWtie1p=B<w>^{X`W^A`Or!Y*cS%}n{%&CQ>#H;owE}bJfm@H&v%;2b=<&ix2?=%&
zNnDcD5~HZzs1=nQM|4u#7HK8_oOix`G03`{9`z&3p$&Ob($`$nJe-u-y)|jK!wc6#
z-7oG#Q3jq|!RKfL_$myCH@Ijh^OPva)VTAK$hceAY1jS7eg713T#K-iZeuIg3ZpSG
zK4IbTcOku=VsaFN7V~XqAm$Pt=XmXfrA-~<;;Fs%8RrF5QJ61qh&Kg~a3Uzy!kjI!
zlAIo8y$5Ta64!<Bg-5h9{8bOMYJ0TrjxEg;!<1Ubaltb_nris-xbn%8yQa3h9>YdT
zpRvqn={l=!YKf_*E^_a11D(SoCfrCUSHvndqs@lLfNp`4pbfGHW}tdxCHbfEHRC>i
z2F&&|Qc!hZRt(L|!7EMj{wI@2^h8lZo}36%zGoyI<XGj6*YeyrGRVhUV=*ied2QFN
z;NAHw+>m|BcXLmh>l=4i>{oAET#T`%sLB@^ZL&leH@@VvzK!#kewBNN;t55NnXPdL
z>-1vjAYN34A29a&5XnFE$ySdw((^kPz2CYN`J|s{!*`+gT%#Qtzx-WAQ+nPSIVepx
zrAEjj<r-pHqElj=fX8$WIS$^6WE=d2kq{%)Nog|CMuTWIX|ku`d6_+Zt~7>eYj&X^
zv}oKR9Kgy1(3zfy!LX-l&c=2wSAS?`Xz#o!t$7?vOXN|dE1ypZ{*IC-&7cq{F>yWX
zE5ol%+Ahd?X(JCKD=eO|s}p35f@QJ;wMC`IbUs=X8=h7|ts}B99q4dj`1#zquR7nA
zdjdWsneFWk`!2>Xo~i|fs=nu$JQn$Y(skJWh47wRiyAcThpgZ|RNQ#;dmgjXvSU9?
z;ltnMKKjvOWuAXRi^W_>$67j{g`8er<Q-qR6vkmj^H$M|^IzcBTtmj?6)w&X1>Ovp
z^!7#aJ>5f7l3?Y%a>GNJtn*1QgKMG6461#k%j|aeTKBiU7vuecH@p}-W?vL@qa?J<
zFYjv~uUFdA6ImVT3o|J)(Vxr4(&*?%36c%jpOmFG6E<BS-?Fu!V@t}uGKMEHTStE+
z)@DBEV;5pfb-x~jRN9F!1Pc7=D~l8znl;K7+Fs=yxN3DNPO2<UTC{%kfE$qzNpOsN
zjAOMC@=OwULmX+QXtk`KOH@G~hGAPFn@OodaigOMJMYg=nc1YR@lK-HpcJhS?0AUr
z-P#*>FBUf?q=xRLjTbpwExM8X9ppJrKTGVlQa`uJC}_bP^^qrsLf<IhMPc)q*V$OV
zA~oWjD!$Phb=Kw|8lr?)nnxxn@VJ=p3-%GBy5lQ8YNX$R&(gk=n@h~Q=Lq>Yu;S%G
zZw6hn?i$Y~N;>^cgmfW@q5JutO_pB%W&1Oa&Buk%ck41_cVa)HD?Vk*li$Lt5ADNL
zfcc-CmVvF|U+Nr7%JrR|s-1KUt})%pcwYsW)l7CQhH+!fl&TB#nG);uOW7@5E%!8%
z2ayR$NlAN6tuVUHvb&cy$15o|0h}-;>PP(PsaLcW3(CG9jj3agrY6Gmo=Z?1J;`HV
zU9zQCUveyBGzx7$OwteVEsy@2Q?8P0sAKB552iQny^~Q?(uo&@<@$He*2!V%NOgCT
zLc`t!=i%ahqxBj=-H4d!D`Qq9HmaMq+6d%Rp1l~5bq}zC>)OoLS?(^^nD5JdIa4YR
zcI5rWs66IcX!1F=VPtMwMJ$%|YUSSh3(1*vwbSqhuF%m?j?=og18pMQ4TL6lx>n%R
zTb{qtySJbJEhv{oTL~xmxqdV(@6!CT;mUw!Lhyug)QCs+^FCQz@xo3&j`Y3OSPG$1
zN9TGCZHxOJI@FA{g7zmQ-H?o_fU>V2vPc{0K-z_@p%b4#Si)fB_(E1HhojH_Qy-<5
zdPS<9Yc{RQ*wABiW3xL-qyvBcSR1efy5e@@S8J+Z`q|ZxL4VZH_mQCstQ+~iEofmT
zW6i_#N*JwP!ANd%d}V9Gd+xZK@Hg<rFQLmrDxwp+rl8Q{`N=|tayS)UcM^Z3cRu~>
zbcp`yeM8=puby@K!ZBvOu`CUTEZj>aE78UE1xBC8@*G^N$2V2-R(;ABS_!Rp%Vk@e
z^<QIWgYLvH3)IqA7F*kIEb#KUUNjkqOrs8}ya}tlu6oiryKtE0kq-64;xf*5ndFZ6
z_fRU0#h4bRLI#W1={Umu{n&7t@En!=7N1G~hBLCG?H7dXxh!g|3jGvi+K_Q(W82gA
ziXo9Dnx#*)E`Itw-^11@d(3!t+|>K`XVuxg5MZ$lCcm)b=pl;3!NI{Bl?@z*++MZ}
zl+gs`ZMiyYScG2pRN%v0oEu03x?^d@{H|+Onuc@DWjukByJ?G@eSOth(yUgPYc#A;
z?BS8`Wjk2kCrcLhnLC^F&ss@yy`#Gg4HA<%P#->0b1PGm+Sr(3$nQS0n|H=UXz%KX
z`Xb*fa1O1!S-76`*vc`26KzeFFU7w+9nTi;q`%pDgBa03;>Q#vY~A%k^1EaY8On>d
zh@7SYNHnt0QmxY~bv}5!poI1I3BFhbt)aHF)5EjxEJ%8<TG0_*zY(j^ug~uo@GxEC
z$iB6zubXo&sjNrV!rF2S9)&U9$&d)2N2c7pI_P<(0NXvPgS}|NZ>3}V2Lw5W44g`>
z>4~&Clz8T_<Lp1`K5t>t`Q)ziiQO-^@7Q~waJ{*i>sdI*n){GYn4t2Eoy6sC`z4s1
zCeqk&XV{fR`G^&pfv_8*(Wsc0*QVQF+D+T;kA4R@5D{Nj3N3vXvS=k`mFVm8_$LHn
zAG1Y&->>&Ba!+$Q7<haDy6poE(pFNKv;&-?x8Ki<oJ!H>d2ofKZm;2Fpco>@=l06~
zIVILGpDRB?&a{m+bwis~+yyb4fr!21v0HSj-(M7BH7fi%Pi>GmC1?>o@+&ov+k0=p
zZOD`R*A@LGQV9LN4DQw4HcEeNZJV6XM8U&X*PE$3CjD10-2l}dwZ$qswF`cI0mnka
z#h@3R{Bd3iF@oU>?OTcNL{hw3|Muw-s0P}SLK@ueJ9y5{*8ItgsM+u1JnLW+Z$*ql
z0^PU-EeqFUvNTuJ^3~n*SZGSbl^3XPYxN7g)bqO-U2@&7@OpNo-s5cnIyWOuUm#RD
zc*pQm+Q8%dPU+;~n&~NMzb9~<+u<oU%G5WIsmcDgmXFDA3`zzD#h@(oWo#V@q!MC=
z%P6)@J9OlO^h_LLgJ^k~Uq8LUZdF@F2dd}$&J>zDi^Sa>NfRASbYeBe?HC4*gPixn
zQF5LX)^FEEa2C~M8tmk8%((B!s+T$M`GC4nCLLUBVg)PJDUgtoo_;NH$cvQ$O|#hG
z9VF|gJx%wPu+Bh2Gi$>g07GB{VGE`%%C7EiAuy|d*8ONe=yCaGvd+;!kBrG!6eKdz
z&SGo@uWLffM3a2mb62Y9tOZlM!d4=sIJA${L_UWDlyg87(%aihHLLlzNZLa~PV*6&
z;v88SOV4Bd8zP6*^O<hu5fvT{mhGb4Fss?<j#`(?F?$yO`s~{cN(<cZ#zTCcfPS^I
z*!#hxf2VYUkoU5cs!;3)RnagK)l=oBy>|{Ei0?WrLp>Y)ZTM@z_mmaU^KW=LePd)|
z@;kEK?)b5w9U(51EL6|wEgenSvC;g&_8qrvr}TbQ0?`(bU(^7B3^y>`z6MM%p00D@
zd|72{4CX%_SuD3SltPYY9VD8(e;%d0UhchG*?t3?u!tx>s$FD!b)^4}{?;DEh-G3|
z<>gsZaANcvq--_Ph4ITcO3(NA9Qgle0a1MyWt!)*!=p!h0Da@nXKk54R_D1Q3Xpy4
z6!j$i)o$eO{cy${(PGgxJ;c<MCNS}HR58|L$prR!pZOu<>G=;JbK5%PbIhB`zDzab
zh5>Xpk{s~?(6%<p<uexu9r)9cmj}G7bvn@RgXw_IboaFCAr?jHcr!b~Nv;yy_h$`1
zA4<+|#V})+M*MN*jVwo_P!DVwA9oG+L~TD<<>m7&Xv%ZqfH^yV6deMYFy0K3i`C!o
zO&vlWj>8eq&x!S_f1rAQ>gJ;OtZS;6`6)@uWb?Y6H#L^m_2|&SVK!}NM|hkIp85o>
zUJ<?S)z5`XJd1P}a#g+h_!U(_<S|zDu?vBg<eNo;)5C7@;Cv3-8?pXR4}W`87zj2R
zXbP#RQjQhpp4(thYZUvUA2o?x5%!gscjC-@T%XOA`iMRI^w11+ekgGn4lc7vB!U#t
z@9KCq8c}ZMo80+jJxReEpDng!m2y6}W-NMz)K9#-2roSf^yu&mVnK=#{tE<~s)F>D
zj#iuW#Z^#5{m+MH>X;g!eWje{0|f{KO>x;$c4_29+6!LR7vcN)o2}SMeA`cZ<hCr>
z<*7<43u==oP6aQk@pDj%C*YGPMvUE8FX>Hz+-O41cGo#L3`7M>p9AngyO_rX7;~jl
zf2>8qS7_MW^b>Q~)q+1dBC~}O6!k=bUG@6SX^3pu4gzq6*suC8ZFvZ7<QN2gzEGT_
z*U)yC{vD<tct_2Y>WSi0D$&8t4gl_wO}9s@L0x36mqX`|6;}t6wcfNGdDuPHKDfHA
zn)J+(U{Tt@vGD|R7Wr?kD!nd0q%%KV_E63rC6SHKZ+3ZLSidY$bGhGtc74*+{if(b
zr8%OHq@ma*(WlTK`|_uRPbfkT717j`v1IJe4VhfNsKn}eANlnX{;9cT_cG?X4oa6?
zCv3Dq&w%<2s(+ITKTU2vqZS9!ASYAAu1Ciy$p-EHcd^pGVQzP;cZ;X0dL3(^vgsLm
zP0@wq*B2v6qjw2nrP)7*xw%!~*XoJ+lw{Hy1g*b%Sds4I^{*+E=eM!q&(H)FhW&yT
z3{^AzMte=}_M{eUwv?7j_XF)&f)t<^CTR)-^JJ+{N1}aXJ_>?^+~sgxy#{JBU?6S2
zxw9B@fu4?GY0c~3Pl5caAh1R!Ki|-IQEu14A2xxBXF%C0`%$Smzs+Rf?*t2)eyEnd
z?8gpLs1Byq<gV|{4hwJB2|umc08pJ=ALP_pFu@ae-C^ZV@OJj`yg@YCX7Q0(VW-gJ
zl?xu^mTKe#><n5dZ77+IW-(JffW~oaLuvjfyI^+_z6t*0l0D;62~v+Nk&5x$TmHeL
z$L_e3XuEWuI=s&M4VQbe>7j;SDK>_?<l=GMHpZe&ZcqspmwN9v-8AmRd_u1!5QP63
zqGZnBh*(W=`u?&akfPzx%<C8&ygsdlSR=SA>xv1-XZAbG?B<;vNdkN*H4gA6UI$4w
z(<R2Ji2f=d#R5Uh=0~D_Y+t~Q8)@Pnk=VyO%KVMl=)2^(+2<BDBX)Rem=`?B!Euf^
zDmjhp&&I*|u?v_p_UB79D~dFaZR(udBVi@q?R#Wn8Sy~^<L6MQ@P8r;UEP76rq$w@
zHKBP*I-ggZ(AW~YS%H+3%jz}p5CB*3l-lz*<{1I8*}2np>a$)|d)AL+h_z2-6HVso
z*Df`Yt*4s=m+1)ZjtLT?2pz_j)&mB&^c*zMqMl9<`NXq_1N&O|vidccYg+DcWbIE7
zNnexF@QFw#1beYRoH<<RWYcLzEr*KfhqI2*8YmpWi<{FB1L@ajvKG9dtLRMyp_y1D
zi(~0^34&2Nr6%Y@Ze6@R`~IL>t5O)=^2#E#j0-zsIYi=|NCJ%dM{=9UH+c7sDNj;N
zXhCQ9Z6*hMy+%hI+i5Jw`|NA7@^LG&ZsETkz26!5_Ws=$u?T6mh9%V-N`jheN{MH*
zf3$596SRBp3<{>6G_IkflnTPJ4yURf5!NO;_Peh&25U7elY#J=)3IvXZU5?|OC7f*
z+Q*A`$78Dg6!gM{J0TD?$AbZC027U=`ZBDg-^c@fZUR3Gv5VRUxc*Mt%z3$0EQ%4&
zN2oVPQMz5{9-&LE2ByZ8{)t%A%@wr}7!Zcg0Qv#&5`^{wm<Q^M*#ZWE`U<X@SuE)_
zx|Wq=pu8}{-aB+p9F8t)i%jm0`!_=W2+7(SwVwQh0&0g%dECnP_ZuD-dG)r*MrfDY
z<IT5hCEKq__?|xTdEKvkG#=68I6l(c`hZf{<X<kp{_lRsByo7kNQSYkpGu0+g%-yA
zn-eiTkKQ%6-e(4FVD#B;9NRk+6SzVq<qF(Bhue4W*=DsHHVFxd-pv-A^6B%k+=svl
z7Ei%`pN3c6xnu4s%1^87o>p#uba%4YaHx~QFwzA-gyEeDF@-ugOBZHv>_>{`(@xPP
zY*kB8jZdC;y|M1hWCvj?G+)!{bSdOuUDY><!)+gEbp$@<iCg>VSBei~s&kjy;QDF*
zDU*Fe=mK>G^R_~5TD0%N_?xDQs}?vL!NbKTd#x*xc2n7}{USE7DI9sU35YPJoVkuZ
zeYbu}4%=u({q!)a+KyNQL(diqIlRqjtgw22%fUC<0muaFId+m2=-0=WB}3xYHk&Q6
zi^X;tJg|$9py(2{ds!asvYq8oVs-O|BpgBK+*=_!8NN@$9Di^S9vP-nN`xK^Nu+;?
z5)KDnq7)QE(~f^J{cu<^K{9YZ?w=j;592?IDr}b}g3f;U4TMqC>MZhWz5mCgcJsjw
zlU{vWXSDukrkKMKK1zbAs`IOZ!N|5nNS+&x{AvYA-0=qszJAf^O!@0#MaV-|g-SH2
zc!u*>PcH;WHh#|rx&;Pp3?VdaeG?=!Cj}P!N5#IWf>HMa*r+ykdsThP(%1;>LO@Z|
zWeO_sK4QhKONbwC7DPB>UkFBwCIBo1;B@4-5Gn|kBt9@>gFa20`l+NLVCzX)Om(_;
z)k3y@=3ZYAtcIzIHqhBwDv8ApkAD$~%t#H0K5;_@)msD*yfAP=D{a|Lcgrp4nIrn#
zi~&`|dOP9{d+|c1OX_#4wwKP*1DQ#6G8T1Ty;^Sg!Ua@k(#9&tZ~f#!qR+ix2$z1D
zA0~axr>jp7U@Kb^zX4gubRrL$;<1r?xksNml$!i_AGnpk(o%+^n?V8{u~HP9CGQ<?
z?(CG;IGFUWv39(=Z;%#;7OtzV-)i|WxkZ;&H5)$plWjezyNtDO8(YR6x}4A%vAQ?6
zCi)>l{J8Y^^6^hc%$5F=;-N2jugL3Wvngq53%mO-kkA?un_7k|3UyFhHA}`|c=VY6
zF{(M@IU*@}c^lD_$=e<{41_OVH9Z_ob3YuBnT#>h-)@4d6gBKWEx?vhSBm+~tlszS
zlUEc%N;WQ^ZTE!{o;ooJuFO>ose776kjLec{>j`xYY3gcWBhN<E}@ObX@fw_Y^EsU
zcghlgUmt<Ex0NL-B%6+b0DV%&=X`8-#-0nI<e^4-^h46`>SQ4ubknNGrx9{gS?D+L
z{>orHU|}c|q~6;dcU2Mn{q?-EY%xmt&F*eHfkwO;HJ@yp-4uFqwNWJ4u|$30KnaY-
zf-9z!6e`QkSx_p2;Q>X@pgUS>Wa}?;y)-)Te$3~8i(;5==UQ;0mBjbXGvE2dl|{cY
zT=8_;Cq$D>^kMjJ6Eoqt4Gx71X8h2_(sRR{b7LK&Yk#6D@rt0UW5bS)q=1?|`NxL~
zB&T3Ywk1_^>IgybGZv5E!G-p7RZ(z!ZjPGU5qSjyT?lx&gSz-tLnTnnhza#0C8kTw
zIKh~nH(Kkbue9<8N<=Ie(oz-xb79*PZ?6+3tiO5d`NLP|XRfU8yI+{}(@%vTDWMC(
zurk_=;5%Q>nlSp3eI|NWrc1y5TdW-~5CbsyyeB~L!ZVL6^SHuRNLO&@DdUUvI+H5l
zyYm|4uZB`j2ote1JTi^2%%n=<8~Tu=v0b!D@;vzDq2mlHrnCb_XEsIG<ec%*<$&VF
z6|A8u{%xzudh+W0>gMLCVTE?5MJo++XCb_E>DkcC#8K&OXu0UQaA81g4TW}sq;-g7
zl1^!p3(*{{Q;jKZl=`h_{~Sk}J}2Cxk<;s*E76;4TV9aBPegvsrMv(9O>pX2CdKn<
zkJJYRt-&503V))NaUZ@%o|Fyj`yedOJp^niSO1JkmHEZMNzq?^pebO_wNOaPl2WtM
z&#6ShqkrU|+bPE9J-SlFTP*y7!U1dTxXQ)lg~j(xjj^G;<vQu;<>_MGLJKj!+aYne
zBJaL(=8Gm{w3wIQJOH#RFj-l{Ve==~&1S#*;((7Vwv48Ha^e2@XVwXw&H2jgyLMUH
z4dTLUrP!$Vzv*Bk(nygjlSkF%3)lS_;?J-kvvj@M4W?A$ZnnFWj0GV-bU%M}PV?9{
znY*Dm5^Mt!&Chsw{-Te6fBp|Qv@Dv6M&d`o`nWRzzi-uo`gmkNZ&w9ek{>P~hZxB!
zqdrbwtnM&-Vo~|{BJpafK@&=*7z@~vHj@QQAYEJoPYW<Q95L&AOQGX~iZC`z>F0>q
z1BCN6(*ZSmz`9^WJVR(*DFI)v+Y{iUfPXUYDSmS{RK4R6B@sAm-%O=LDQ#qG`n7QZ
z_+DCh%R;-U!?MXL8wdXAWIF;^7#<!TAU`OP%DU}X6c|p@gxsQn-}ONm48%y!KO!%v
z%-BqJ7fwwS<_%BE`WRj1O>0nL*kr$>eq82twtw6>a4_$=I+Q*HL<eT<>_E4Cw)j!*
z)=ZD@KAUe_^86tyoyQ*r#B&}9U{<9U=WVFZ(rhkTN97Auh0yyXIC#TfA_yP=q~w7>
zTI(0C?5i9W27xHWbd?=-U<=?=P*PE~-i=s~HcGMi2l+RpA0y?Hd`XBG%!72WHk3`m
z(gE!R0Gi_S=yps04EFF1|DF&kS%6j9Esfh2-%r8VAqbYfC8s*l9PCB=(%3uz%5@JF
zf%KK@?4s?_#sna+V!gE~stvXP(~P&82En{X`-VL^fM1NQgA02L2^QI|2@oApL~vgm
z2O)fX<M!o%O*3ohU%cw*1W#eAib9ZLD(ILJI$UbV&Kc5N>~hGco}{=ctOPC%5<ZLq
z*4hjKK?C~Jjw8>XTS{6??H5Kr$Z+$I;#|*Bc7LNBtzrJF7~dlcX{^KkPyv~{4R(|;
zke8!a%I7>b);hSZZaOLyy_g<a1yfKL`#6hjxsQ3fc58>cqX<zP-1}ell;QeOfZ3D{
z8_?oq82tt1U@g@w+RWV?p?L7Qy3FB>X{Q}*K!)Q}iRDOU6}hzpoR=q9p3WUOs?v3e
z$krM2tI}Gj)W0O@i-cCm{Poeho8ynVN{m`j5E;_xsUQ;*)77%AaKYV8fk$17DV{G9
zCK@K^2Dc}zg};VpTGlrK_Dyt#4wDT)4O5&)740mh3ajV)z_mXpYC7L=l^7H<0Mm(4
zcL4<^=jU4U)oe3>tJ_P=@$x}15bh6Z>`oDxp-W3kR9ZX&Yh%vdEQ+H(<ZdrYr@RD|
zHOL?8^&l`~pNp+-K!KUP$2oLRb&;<#t8{w|SIMWoLa!9bhCJsFPTMG&B%vc&+G93a
zZ=JFXVIxffr=Gqgbo=uM^f!C-G`^R6TYOjvMQo7}Ts%BPP=o|rZS-D!^F=)+tY)*U
zzvg@+AEAtZ3}{b$%4gp##M=>4QwSNa$?k;(bDelf*i0{xn(7BjR=F3@&hoS)F7d+*
z0sx*ayW$nK)(uX3yhId)_Le)3gtX%RsUrOP<WM4LC5L<N8r<bf5m!k*wUgC%;MJAc
z>8Qtj`SQ#o(9ns+NP}m9t`zc;x2fd*pQIQcS#?os8M$~fP(%RmS#hM2M#G7o_tC_b
zXSC2NrDl;{FJdfy9yA{qSNh9EOaw!TP@0;Ze{FiR9-!M&ni|gdQOkb9et=~oN!!+`
zILECGm*m2YbSgM!Fm}SvG9(p{U0J+hwbiKpo0ZVK7$O6^#^7RR3do79?UqqAf8HL)
zp~EGWm8F@=02Bd1i<yNB<;F+#B>lSm=oES_AOqVwu5E4Ur)*DsGmp?Du00%mc2>A>
z{$P~vx5Le`12{zlUW=%_;+RJj?;^OH5<uOoSnSgLZs#KDyVe=ZtoC9MCk#-k3YN<S
z=5f{DTwmw_6<jlDtT3j}_5|eL@Wwa*ZpokYDtC6vq=1DB%Z^P`;|CwpAvm>t5mc??
z+uPOmHvt<$R{d4~gX%?$$$jg|?fv;`En)#J3W08~PBJZ|SHN&W0KcZA18q<UQ9J{@
z^fD?fCV@`l@&Ul~CU+W`(GEagSk!I<?h3z9Fbexh&7L_|o()5UN3+zBJ>E=Fo%EEk
zVr{}JI>09B{_rhhG|}_2#T$kj0m@6fL!uw<Japssb-spY7@~HRt|Zwswa0-0*q`+F
zb@DiNLwrv|8g@Immups{Rsu;>T7;%B^nny9kc>`X6I)j-FlekRj6fJsV4H?Ytj1WY
z?0@#1AHjuwy?;VUO}$dmj=v&tb0Fb<T%JC#6he=$glQHqD+Vfgvx|AT4jV4bww`x*
z+lyAO%H_$#PZtA#QXeYPT<N-iX0cv(<H>ygHD^3YZD`(i$JV#0*K>ORf_U`2A~IH8
zgoE3Q*7cR=#uK0t39P7}U=*O8eRX{F1<P8v1mjL-DxgA^arGXgAZ#Q;na0l8yb#bJ
zNBPXarr!#E;)g&sc*A-942mB!l~~KY;-fVma5vSF!+I+im6(*os2E!%QgC(&cJBmv
zuv*B=C=We(swDUGquNz>aVFA=$>sA@TSppz+%$?ZfnP?oFEJYbBXvutsqK{(GZ?CM
zwQdLHa}NW@W6sm5e(tqHUeypED;J<gTWWGCiLf`wbPWuY+7*`SBHN9LbdD2Ht-TBa
z?@+|G9r>K|w*ediv6}Z)-Erh;9MR@I?d!J$iW-k~5t^<D(qa@iWGcez%qoIQXyjbl
zNW=EkBjYd1yXggN?oz{U3&3Zq$wFtpJ|Nh5Ak_fpwC<-pwci8XiX2)h3E|15&%a4T
zcNdvbu0FEGPzXJ7L<8HI_uuUd0qnI!s8@iET@Zp~nNY_L@XqD8m~@MNsQ4|q{Y3IK
ztw4cFVZL8MZ`@UczjLOjivT6XZ8;2V5s$8%cdVyYb>(GT?Vdxx1|ZddG5#|!G$7}v
zfuvn-uIxDng+hRnMIabvpY$im!`#!x)}3dH3iy<P8kU3a#Vlgl-Dg^TLBS|gzr5`-
zoT`T^`Ghqf^G~3Z$f1?^5g2fIr-xMm7-vxKsH{jTDqa=EXcxN0n-gKuIqj*0C6lj4
zWGNhvpADKGtSVn_!*Bcujjk_=$iLIC-&A4xv+#(;wd(eR^ja3_dY}6y76EXWo0^)+
zJy>j9n{ygM&kD#zfInj@B6vC6zv_=?Y|em=l-B(9aW${7aGxSuY6p0twIG%QFpxx8
zmMtVMs0M`q&@j98m}1I%%f=7}>b!bxOF@xQQDwk!0UFt?FotFJi}9yeM9l{x%|oDQ
zps*Y$+;G6tZB6x0Tf#-cs&!-Qa<AutLC{nKTmK_3C`7c$A3@Sgaoo5sxTanEnx(bn
zXv%P*sCL3xq0@2B)f@c4P*C0*M>BjD&t(kCN)GZU0xRd4?A9XL05^_W5%~2qHgL#m
zl~>y)Yz53={v(Sh|9;*or~_>+I!#_eP^cr=r4aDdK=5=Qfi=fcr5Y()x^&aNL7aAc
z;Xj{+Yc@h`2Y$4%hks5Vao;J8S&4c_UAvlOqkS|=#w*+bOy0(uIDRc90a-JWI=mtw
z&7rwB^<*f26Wo>Wl$O>3K9X@+M4$A1gdA?hszEQ0#ctG_rerpWS~_g)sfR3+O3}|3
z8r*q(_b7Z@nW9BeT0S*j{;mqJ*<G$`y$FCi@E<+N9$OL`a>pNXA9MmkK0p+>e>UWs
z2K?aw5<*xHL~$UZ(kU~=9`e0Da0hise5DXm<)3?HpBf_(HQ{!A5!>u+aav1d5g-xn
zHu@or@2Z35PKD%O#(R&r#Vh4()_c}nE68@Oe2*TrsWlTi4aQ_yiRtp>MkK?cgo1Z`
z);vDr7vhWVN|tfiJV^sFS!^JvZ@mIbkLqHf4KDEqEZVFyoJ`vW#1cTsyqfm@k3dP~
zT|RMf6`Jk>phZ8XDv=6!pRBB`Ablc5M^rR_0aXKHr0i7JyPR4@XhvRMDTs>05qzo|
zM)U-03mP(@9fcrx(f~>wf<gJqecsiPpXN$>Qb*wOP@w2)(B%iu&vnf+kG(&+xev{A
z#)5T~Yt~Y1s;j^3MgOI6zj5!rxlFgM*~)IX@V(*(8bOcc>Gn+Vg??d2M&TdqPRiG8
z9jCu;$9r7zYDwB>=fo%aOSJgi&yAK2Z?gC|b6k0{=-!Eg6Z~?1pM3r>>oM~GRoIt@
zL-oIZ+n18Pk9}8&LX$0fg(&2kB|?^rvNPG4NY+G3)<IIpS|<BWB@9zB_I(@1Zj8Z%
z-+R>edVbI6`Qve2T*tX)&biKc-|u^Q-LLz;KLR+UtY+p!g2w&>kc3lr94~3z+%P<a
zZ)JgdRihPH<ywB~X=b^2{c8BH-?91QkqG5#T0{QmLoX&fjO^22)6;&d#R|rEI)Ku^
zMMjs-)(?h8gO+69yuOP>gY}Gk%iwxm+Pox<EzQ!^vc^jU2s$P1ts8!!ha4v6_AP<P
z1sybiywe!u8ys@%RRc*Un=WN6U^j~Ex8)mVQtX1ku%rJt=K%jR0DP?z+|8?Z6J)ht
zli%{FYi~J^+0ZWV&QXR+<Na{1h>n|P{>jramPwAH^as7gAj_er%Xr`|ezm<F5c~mY
zALN?spl~em*<ilh=|8rh{BX%bIDnkL(>$C#Y7GY0gMYE{cZa&TsL~4t2&0t`l@D>{
zhecUeJ|W9W#^vuom~7uz3<?vfb|xP`?$sczRF%1A3zaAUfT^QYDeHbcOh%+XH>L6x
z-ALt^1*<o-|JBy|hW<ZRd`l>Rl>_`p8LQ7Rxo~GQ%&PbNYYQKge8)%cS>2R!?(+KF
z5W8`l0$J{4x_1F?csWmNYl5tJU*H}Z2lGD6xCjK(%nYbLb@FJUGy3enXCZGO9$sE5
z+!I&up5qe}$zQPCaI*A4s}nyvvQYM3UM}%_WRHYnzCGtlPfKgzZkO-iZzty<Kt)|$
z>lFqB7$BK=j-B4Qlh3hmD@fd7OusV@5vRrb^=ZR>z?Cd&RvcJdl*~$E`}XbI9k8xh
z;%5vTx+>ff{Xaf%Yg72E6@z76r#p=`5bZ3FMW;TUrMeo-Ha=C?Zpd1x8vbj=sul2}
zX%Po|3*HdAM$5Z3A5=2Dm$XK|R*qtwWoRVh!VH<hwKfDJ{=mU(duRBV%rg>-S{)t$
z@-<-0lfC%|oJkHbuQ+p@L$Ioo(>!RkY@i-qot8HL2<_qB@+dz)kYTxMXm`JK9%K?c
z%KuRVj?YL^IGA>|P`6w*3pk3PnxIz=kMb+AV+c3QRNQO<$&W^75cWf=$f0o$!;D>B
zjB8vKG^^~jv_*vr<{UXaKpsX0swYO)?ZGEk?|f~&;`uRPdxB$h|J?V6bDxsh`C{U~
znumjo`bD)XW@a)rb;(I!R8_T|NM5@bicf&r6^J*`1Z(~qIgv5nk6hQ+9Or7*NeRu(
z_!?xrzute|nOd^Tv>O*?j#RJ}L(g;Ru;1}`b^Sn0(we5!hVR0OckaA2?z1$7VO3f-
zpC9apUGicrFme^CdXryfN{GEh%~f$f{rLl(3R9IwbGZAr&*%tt29IPF+&fB(&a|0G
z@>kyT=p;3SwXdRjdr@oC(Y2HC1>yn7yMB+av-+?B^_Y75g{Y3YEcVfWo?u<yr&O&F
z8n(Jkn*r~HP7iL^L2G$BG21;EjtU1-Lup9Yda+81uD8NWp%+`)<<JoohcVE?yMKv;
zb94}OgW|xklEm3GIVPZW#TmHwDZlPBpo^;>gCb0o!)eVPFkP<8y}Bt<Tt5KJUo_y<
ziLdy4J|Zb!s?pYSX2?*o60|qGA=~(6sjq;Ci7s9@&o7}wm|+(X$2&NT)mn#<&MHCM
zl0tjm^%&Y;_yc6hy*3_1uCO(uAVj{DJ$IGU=}g~<)-W}+iZ6~6wD?oT)&M&?`l58W
zDji$=b8LDj8TY-LN|uQ`@te5zwU^=u`26W_v+_#gk2v&z-rR3ji8`QcKMZRqx%|m6
zZD6H*Z|o0FuWK`VL~wx@pX2mljuS0UQ9aF*0oV0U827lh;!+oWyDxf??)G@*?~KqW
zdu91VbART^tQ7T?bKUG{qO&(|DdS(w;33x|neU-j#P@l(37va2IxRDEc405@8<(uS
z^_@4|iAD7^?P1P2p(!lzjR=_w7ubW$oQoujJ{Bx#`QBjl>-Nx6h}{HGeIhmxbZS#W
z&!W;HtDE;iHg$S)G>x;d)(rtkdP1SeSEHE?AED~9Y5-rZ{i8ZGul9yK(^aFa%Q~Do
z;IdoqWkX}@*PHkziTZtyW{kedF^%FqG%_!C1MfW^&xBX6sA%cfxUr$02omG4t$rkV
zq|+BXp8v8B2fkeer9Ex%ZFVz%Uj1)DvvoliJJ!+$%1YHN1-WC_Yd@hoAMN2ApD#Ng
zpzlj)Pv!`%UUSX&Ok19ar?*Jqx!IR28F#kVwf>3R7nKclFXb&6v5i^Rd3dnVUvqyd
ze{}PqOPP`rL%3h<;IcjRn#r|vbFoxFv%W9YJJtW;Hm7C6Ycs|snQqDDy3N_mn)8)x
zG}?FJa&d3{J`dQ**}1zv7iIuyt%B^JI1mzhm0S?xrlEY8;Q$4#G{Gp=5Za-|A>W9a
z1>c_a{A4@-w{uXEFhc<F=E=V!!X=C|6*8IJd$g>3l(U=MvP_cRcc%JIitXFqoFct@
z?|3+5Vvzb%h>4pydYcq#ZSpczM>Po|`TaP^s}<=FO=<AjZA!wP(Y{rhw`Sm`_&6nf
z^?^~b2lv2=CvtubVP$AHaK6-*Q-|+?@4ae)_tK-oNCWyN9ys?EMkhw(rOK4Qd9r2H
z_Wc&ay;zSI^9Yc(M#YQkpS`Tmv<^X?d@0(|9jZ1NJ=C6VoShKc$bfD^h$$r|Bg73B
zBh^o&f^H(8xC{}`Ch-yp<M9&)3&9+DwHZ0k%wGm4PFOF=4SvvE&#=8M!1|gwrzk%-
zEb(<zVstluE-k@Tu`7gIcaeVefsAv)MOjL{`}Q_`P5TWe%EiwY85KC2>9@<p1N(n6
z<VlSlr+Mzw&qFr4PW}ktHdsO%e<TlRt2jSa3j>kdOYv2ba`29fHtWs!3xBY!1Fq-z
z{iNu1DsHFUo5sRspG6EPUi`yBdQi2Eb59>qbd)F=zxQd$hsf<K5Ely@aswU07Q*Bg
zgk8)BgdzvUtrlALFtN$BCLXuuKK5*^^}PLqiO*d|4qID}=RoKF!%Blfqp3f)a+&6u
z|3k-_WXu*c#h1*C)wVx#VOxqCup+;!ceRZ&<XfAu%iRp((6)kUB<v@&Rn+yNRiag*
zvHZ4{GpA{71zhF>nRUXYPi(0bJfJE(4;5V-$X`Fa1anmo`()?nAzKc&sUEWnF6$}0
zfTpzNE27ob;*O%RHyyWm7$u}fqn{J-LxOjQk)+-r*An;SM+g}7c(zL-p~I2FqiGNN
z>(cm+O~xN5P%osY?ny3MIqcq^YIE2+Dc*eWdQfBU^M`9L!L=eabtET`Eq}aqEH%1-
zsSw8jhVVYreZD$<>jI(V7~U$o+IMCxAN!Lxu>(jX*r2EF6TnqMC1)5O!?VTX7}at_
z(LCj)ubds-)XJ_DT>AQ6DljE!KJc{0X;Y-QW17uy;D_F?I5ta6zSELqp#fojj?<}C
zQUJimz!ecp1R$osSIkNQ30_ZWejBO5@T6)*7uh60Yo01cGm_-c<J>>q%#!Awd^Ns1
zCY!$Ths5Ww8m_Cq#Q1j@qp4?VRc59s)L5EjHoIJMmY0{AqOT%g;`&rSQZE0YIt|ZI
zv79VAY}Y)jeEeMcM#-Zo(yfJ$5nG{3h$a4VsW&E=v*LMQrcJ1_g(_0j1^;>>gSj4I
z7=cPmmQpcKDQA=jAi9dxfJ~sKlCh_<-Y=n-;*7cXFb*ui<a)Yt=8wm9n7`gKe!d7=
zvUGJi8bctzF1bhKoi**kBV@V(BOq%|8Dx!YV7)!T2t@%G!y`9ejh@W{EDo1lKA9rO
z1hnM#Oqwv0Jh{kjJDPg`K%>)63$Z=2*Vtt;67^78TAE+Q=V^DUY&8kPfA@HGkrD;x
z?yjMx?Um8fO4=zx)+VYmx`O4HczxGrEuOz3_3`m?3{o3%YWyS!=Kt(`s6C5Ya_<fB
z;v4&y3Ceye-1giroy-i0@MdENwouu#JEB3_Ln-TdGMfkdSlzH_SQM5vz{x!NinoHc
z_b0Y@zUU8pRkUD^W?-t5WwawL(3GbtGfYg!P0S5%^@oOXj45=)%+}Z_3Ap%niz$*;
zvhR@)N|s%63{*36B?hbmIrhi3vixVZ-wN8_5vbla7|@f_OW?Qn_D%vdHC<gNKrehP
zg#vT`Ul0C7`=<n>+I<(^ZAo>VtB^-Zzs!Pm8lInID8t;9!atFW@$Q`-^lUh%-Yg&F
zS}ysg$3|_hwS1zPz{I6iQxi@Q6>M!?)Mf=N#b{;$mx$K@`qh(yPseO9od2Sf^Mgey
z;w2*a0e6)cTQ;VJf`U~~|C+XGF~FCFs#{+;ii+c+j161*)h16hWZWtayx@6**W;;G
z&t`r%iy|qTxvI{IArrQ==D82cWKQd<d+JdA&K^>pa*y8Cgp(~=-h`W-UZhdDjaa<*
zwZiUH;R2eMiSDUAVOSyV?X5qA?2C>iG?U9u<(aAQ5-Ccd<1EC_J72HCt2;?MAwv9S
zfx757g|?1#6M`1`x6lUvI}rrU*8NlbHKwJ87K;;2Nli@$P5`iH{F-n?g-qZiX@b6G
z50Bkz$qCWyNDy#T<_8Ra+Ol)V)a{Yl+^&nBdaAy9n&~`&R-a*}t>3PngShI&E96zy
zy+5jh-Sr*xO%3<sw8&FESFCDmW|_j#v0S#E!t1s(tUkDN$8TP~j<llVeK$^SE5fDU
zoU?LWR{B5Y78ZdAgh3VfZ)N?LrtkT~fz2!mmH8vLc~beeQZQ3u6#Sezu|jV0f{VOj
z86FIo4SE(SX;}A=2j7}bS^n*O;*+&Rr|{gv?iE%A^!ng^zM+;_S*bt<$eQq3>l6Sh
zmknEI`@**Q(U`yEexMPnp7NxLa>w-C^?Kvx3rmB^0+)bM8~GL9<87b!G-sW}5srDp
z({lDRF6E?K7rSEXukA7W@r%*pFQ1L0?_b`<2|8ZB<Vsa*%Wb&Up2Jj$fYD?pnkjG<
zPxWTBT)+-}azv>5Cw%>?v(`woC1ZBBp#(DuBZPA9R>bksLB3NyOkt1w`AyAI_wJL1
zMjnr1fw350h5`k=GX@5&g*Huvu1uhA7GfS<YL|0U&D1ejEYm#>D5c2RHWNMyg<O4(
zt+e4U3`Ne_fl&ZdNF^gW6d}OJZ95r~t+98>gwuO>!@e*=5{<uObFcWUVOfNb>q7oH
zr6;+l*Mn+kIi~RE6fQyV(gosnOzBeX$Fj2wWfSky1#2<twy^k^H>YMBTFN6zpTFoP
z9=N<qu^<?}6$Z$=ny}&RfnV~VI9)a8_nZU|GpH$QBWEBeY=c)7fTv`cLixhUQq_5I
zWP}$u*$#()+R+TFnuJE&&0CF3t4u5($JZYELW+7OJr4o{Yzg%r{5IRwOAULj6=G<t
zTwPt6=%~YDe~FB){x}AV_-V-MHM6euh6(3(kUB7CAfZ4Y_)^%y_nZXbyY?EyS~R8C
zppC!x{`N!NJT`4kAif|EnR~~ClTWPGAUfSu;)q~PI102^tyl28jG`L(Y2CN%zjbu9
z9aHHl@ECcJN!?(S-WHUqgB=?&v3sQZ2-9ktVwO8YP638YV#z~`T<?*1wMa@-BkWN?
z*@8HW?xEqX#PN~_cZ}>V;6gJ?vl~t5UG^U76}+}MMX)|tTA6p2=g3WgC0sI8d9=t%
zapT4f;3CYD;K9TX$JtF&5DRE2SbfKHCwAPoj7pd1W&)805MV&iM18M_e<fkO1pp*U
zFizYb-?1LV6=?Dqe_tWo{PtON;h`}RH*xc}-w!~LNSNUjduNB8HubkHTWvb*VQ3i%
z4dBZ)N0vSHTazhq(`&75I8pILxaBma35`%`2ine9V^93v(f(jK&NJDG$AnX!_c`!g
zBO_@MurJ=Ikr69CjX!{$pi4cwuNzW_>1UZAeZjk$Vcle3f>qGyNxI3K5H=UeF`w@9
zC1qtZB#memA&+iYvf=Y%d1a$4V$%fcaE2~IxNsr#1Ql4Dz-)-7W+E&B7^=Q?q6+ZQ
zSTuHpDRTDv3wD=s4)tf0vn5IHP|iRN6$>*<G-m!R^}DQuewRlP?-zbo2f<a#PP?p%
z3oi@S^OQP4le0*i#T)K<k8;zsEqi7ObFYlIKv2FFA@tkMp;?>tV^dRaTN{;SiZ_7%
z6(vtT35$%3bZ@jcHX3^(K5puBixjk0Lq;bqtQ}x4LUISvA<kd{-b!BY<3!`Z0sv(B
zFHv8yYnsx%29qr82ESla_Y@n*uMv`V$q>YiB_$;k8@scKcvPpS7l28<wiiuaisw$=
z7<u!R7+O)1f$f~_Vl(a2Mw{2+gWU6j@gvaOKBI`j;^7jMeS38$K5Zg@?`DQ$VEAOp
zLlCIU&pUxzeCg^kP_<eE7(0bsXlRzWJ{VV3QeIwt`W$d4nA%;Kkr<47BmMES1ajCQ
zrGGNIncgg$S7fYUw%*eYBUk&yyp;r!1za=G-5@mdAmf$hF|(Fm-0=Ia<Yrf@RFJWT
z0Zs7_7Z=#e6d;&mD_kgdStws6lEgd1Byns}^4hG^(PmOtEBLXBR1(gZQ_swP+>TZs
zE!oA6tSUVYcg>mj%3n7auv)caV@f&kNNYg}=DY_d+)lm6^dwA}0o3yAkN)RL-p`oY
zS{2*C`PY8gC1-D<$bJuqMO~#$L9KB@%3mk__JJNeOxMv+YN?5oFMa@YleA_I^jUGL
zWC*P1rj=>R8S;F^)C)Ea_V(f6HAS`CtGfItq)==)SoOp<0#Z{)q7c>w>=o!YfZTuS
zd?f+U0Sv|rj9G44uNJupIz%BfLNmIhyU7C%<dlq?C&7qYKx1&Sd^vcAfiF_Jn2_ZY
z-kRwpJ5lmfB~p4+r7@$nU9ITl6TZAYoKHL<Mj79zSZHNgivktgV9&s?z22ZOIQEve
z95+B8L6yMI2Vbz7M}g#eRRE!PS5)=q9q;#*_s_?2m&V;NS5+9OE9+X#ZSoxYSiUGK
z4yAH?L9di@NEl5a#Prz*1_vc>WV-#w^&Z)h&%N0S%>nfb5i;EXq#R#|r%J}{;w+lu
zfUB)57BcDV<U~u3t3TV&gWF4|rs}tG)4>9`hn3G5L%J&7NyaJuc({;q&f4<zLUFs0
zg{|s~k#nCjE;!d<sz#;G*JRq2WN}Gv7Zd5YnJrSxYRq^S8tP;mZE3CSwPM@nF|MJL
z&LFm(_WE86!`6iCT{PkR)X*@d(Og#-mk!hh#Udd{4J7!|ELDFQ9lIVocdDiBIhxBF
z4^DksZt#wfo-AHg8|K{n*o&=<{*cqtHl76aW4=Q>F|-b2t4$Elald|FsXywPpB^|M
z_9y%~@c-`_{|OvI)vxq9RJ5N;4tcbSApJbhU`bkjJ@VrFTH&eKl?3v^2Z6r_|2;GD
z<G$Ikg8uvG;=jA0p)%j!T2|U2(OfHoxm*>k&r=kyP$eG}`29Wj@0nmb=)(Win~iO_
zOXm5%&h$4$TS0f^A;H5aqs(F1RpMR@N)%Gxj)J36wwI8enZxM8<;1jMMhK~WpEw=V
zm_FQGG>Q>Q8>ZhO3dEr3Ao!k@<n#k<(4ISV=jOP1lLOSGQKKx?co02Vn}a=nQ5v=K
zB3<19deniky@;Hx`zO{dC;i~cLG##2=#DCMps(`!eOKttU8vsx%6742bkny3rD5WE
z6b;>pJqVB$B{@Cw+csW4!XCDu4)f9laB9TiI)O$vDA9Ld8MLK%q3C)w3k&~=LZl0n
zruv>|AEqYpx}>MG{N~-jh7l<qccLbcIau+Fc|KEIS+I3V4VvH9<K}S2fHoDP(Nt^@
zPmv;_cOP3$+1x}dkh%B@E*S6u`P(?`o^Nj>^&Kh=oARSeA5F`TFhQzp;g1=AD`CfM
zT1bT>ugY{pN9~~1l#fsDnqbEgDKs{`ADw5f&5)%M%CxG^TTVz9pcJGz<n&_Mu;X|d
z%g=gmuFLNc)h-)$EI53fB8RrH*?CuaLjpJWJ9Gb=R_J31`eClxic|Jbe+nU5+trc*
z@M>6Ah}TPl!}@w_HGO*Dx1@&C@9ZJ2K@(tW@vybw{mqM*)R=0M%GAj+C6fD;Y=5E<
z#bH))2(|KkX!Y9W!)GW?NWH+W{*=<P6E&PE;N8Uix>E}7%WP8}%d0xfGLX9zhfEu^
z6j*B<qCN&Zkt!QSY&xx$ka7h0=dMrhf$UUUKQ7DGe~sEm8GAt)n^M?>YM$xVPvXNs
zSHG(eH4(7D#@$m>JNui|EvZ>BN9>q%%fq7ewe5o}{yTTApS`JDT3n|pLmCZ=CaHg%
z%pGuVad6(N49bde%e|nEhfncL{evA#-~TWeeizr_a#z%1pUd>fD}}IS*z(oLxop&E
zMA209@rMc9IpJVFFXr89Umo=2B_t(Tl%B!XNG+Oqb(usaOyxZBCwOB8enI0u3D@Uv
z8nnL$_dks!eOF8yrfwl-!O({4@4<GV-%;&nPZE$rifjmlHLoJ-$3lBmuxSmd*RHuh
zcW$>>V#n6kiM27-9aK})E#<1%36_!ylP$&{mDlH7aO@D4TeM$1Dd`rOZIWz$`s&^H
zN3B>C=BoxgXh!SI1=&MGr&7b4n<ufNKDSo=3kji93B6mZgv)H{4p0%ZH-s0kHUDMI
zQMuH->5<0Wa*I*+^lkeoID{l<_90#<FN1IKmJQX3v!nW>Q8+AD7(O_#O%$@IFr|PU
zgAlgFOhr=3f`{j5k?ySpl@Uz4Hbu*1SyV;#Ehm)eiDt?njj1$gf4d2=Mu3O+YskW#
zeUHVhb9>NdD<CzafZoX-TEDd{zF`Cj&P>-aePyCihXxx2hTG$4qG*^ykg!<=g0vbY
znujzSIhU7@;><K?tZ~VUtSVjzSdMS+_!^@A70wE;nFZ54`CD)ULV)(U7e2hJdE2wR
zK5SkK*4xM|lQCSsY^KTFKo+V|!nq9*$nxOPsDk@-AO=~(OTO4<Y9R$=J0YesabWox
zkVF!<*~1-D7R>ed<{r>b>5ZoXBsL6z99{2LAvVC)pxCi2f<IF^rBEU123Y--@Trl^
z-y6zU9|(!h^!X4y9DHu|KuscvhhRS?OO4JPt_I`{rV!?UA(Y0c-}k0u2hayxQ|Q}<
zxH`Jum_~J*q~&ju)J<;CBL-!i(}K5;4vA2|43{fIiR*oPv8aq@Mm5vN>6?0{gIp{&
z&#skqsu1JT)v45OnBkT<H~3m+7p=rAq#z`LDKwQ{&x*p7w%-uFPp^#}^6_=Td$kr#
zuU?xWq>@Mly6S;m4z0t~HQyZCNw3}mU+h?Q&x(Cf<^`WAECQy}C`8`=zo715M@61j
z93=1jG>@JC-FGd?vU+m<4<-KCV1Ykn`v2;5_V4bT|Bv<fSMyECzq@t*e=k5fdJ+lG
U^lE#|O96iLb&Rf+U9o%eKLi6)vH$=8

diff --git a/applications/液晶屏读数识别.md b/applications/液晶屏读数识别.md
index 6b1ea24306..b85bf56daf 100644
--- a/applications/液晶屏读数识别.md
+++ b/applications/液晶屏读数识别.md
@@ -23,15 +23,15 @@
 - [7. PaddleServing部署](#7-PaddleServing部署)
 
 
-# 1. 项目背景及意义
+## 1. 项目背景及意义
 目前光学字符识别(OCR)技术在我们的生活当中被广泛使用，但是大多数模型在通用场景下的准确性还有待提高，针对于此我们借助飞桨提供的PaddleOCR套件较容易的实现了在垂类场景下的应用。
 
 该项目以国家质量基础（NQI）为准绳，充分利用大数据、云计算、物联网等高新技术，构建覆盖计量端、实验室端、数据端和硬件端的完整计量解决方案，解决传统计量校准中存在的难题，拓宽计量检测服务体系和服务领域；解决无数传接口或数传接口不统一、不公开的计量设备，以及计量设备所处的环境比较恶劣，不适合人工读取数据。通过OCR技术实现远程计量，引领计量行业向智慧计量转型和发展。
 
-# 2. 项目内容
+## 2. 项目内容
 本项目基于PaddleOCR开源套件，以PP-OCRv3检测和识别模型为基础，针对液晶屏读数识别场景进行优化。
 
-# 3. 安装环境
+## 3. 安装环境
 
 ```python
 # 首先git官方的PaddleOCR项目，安装需要的依赖
@@ -41,16 +41,16 @@ cd PaddleOCR
 pip install -r requirements.txt
 ```
 
-# 4. 文字检测
+## 4. 文字检测
 文本检测的任务是定位出输入图像中的文字区域。近年来学术界关于文本检测的研究非常丰富，一类方法将文本检测视为目标检测中的一个特定场景，基于通用目标检测算法进行改进适配，如TextBoxes[1]基于一阶段目标检测器SSD[2]算法，调整目标框使之适合极端长宽比的文本行，CTPN[3]则是基于Faster RCNN[4]架构改进而来。但是文本检测与目标检测在目标信息以及任务本身上仍存在一些区别，如文本一般长宽比较大，往往呈“条状”，文本行之间可能比较密集，弯曲文本等，因此又衍生了很多专用于文本检测的算法。本项目基于PP-OCRv3算法进行优化。
 
-## 4.1 PP-OCRv3检测算法介绍
+### 4.1 PP-OCRv3检测算法介绍
 PP-OCRv3检测模型是对PP-OCRv2中的CML（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。如下图所示，CML的核心思想结合了①传统的Teacher指导Student的标准蒸馏与 ②Students网络之间的DML互学习，可以让Students网络互学习的同时，Teacher网络予以指导。PP-OCRv3分别针对教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
 ![](https://ai-studio-static-online.cdn.bcebos.com/c306b2f028364805a55494d435ab553a76cf5ae5dd3f4649a948ea9aeaeb28b8)
 
 详细优化策略描述请参考[PP-OCRv3优化策略](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md#2)
 
-## 4.2 数据准备
+### 4.2 数据准备
 [计量设备屏幕字符检测数据集](https://aistudio.baidu.com/aistudio/datasetdetail/127845)数据来源于实际项目中各种计量设备的数显屏，以及在网上搜集的一些其他数显屏，包含训练集755张，测试集355张。
 
 ```python
@@ -83,11 +83,11 @@ def get_one_image(train):
 
 get_one_image(train)  
 ```
-![det_png](imgs/det.png)
+![det_png](https://ai-studio-static-online.cdn.bcebos.com/0639da09b774458096ae577e82b2c59e89ced6a00f55458f946997ab7472a4f8)
 
-## 4.3 模型训练
+### 4.3 模型训练
 
-### 4.3.1 预训练模型直接评估
+#### 4.3.1 预训练模型直接评估
 下载我们需要的PP-OCRv3检测预训练模型，更多选择请自行选择其他的[文字检测模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md#1-%E6%96%87%E6%9C%AC%E6%A3%80%E6%B5%8B%E6%A8%A1%E5%9E%8B)
 
 ```python
@@ -110,8 +110,8 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Globa
 |---|---------------------------|---|
 | 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
 
-### 4.3.2 预训练模型直接finetune
-#### 修改配置文件
+#### 4.3.2 预训练模型直接finetune
+##### 修改配置文件
 我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
 ```
 epoch:100
@@ -123,7 +123,7 @@ learning_rate: 0.00025
 num_workers: 0 # 如果单卡训练，建议将Train和Eval的loader部分的num_workers设置为0，否则会出现`/dev/shm insufficient`的报错
 ```
 
-#### 开始训练
+##### 开始训练
 使用我们上面修改的配置文件configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml，训练命令如下：
 
 ```python
@@ -144,7 +144,7 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Globa
 | 0 | PP-OCRv3中英文超轻量检测预训练模型直接预测 |47.5%|
 | 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
 
-### 4.3.3 基于预训练模型Finetune_student模型
+#### 4.3.3 基于预训练模型Finetune_student模型
 
 我们使用configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
 ```
@@ -177,7 +177,7 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml -o G
 | 1 | PP-OCRv3中英文超轻量检测预训练模型fintune |65.2%|
 | 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
 
-### 4.3.4 基于预训练模型Finetune_teacher模型
+#### 4.3.4 基于预训练模型Finetune_teacher模型
 
 首先需要从提供的预训练模型best_accuracy.pdparams中提取teacher参数，组合成适合dml训练的初始化模型，提取代码如下：
 
@@ -236,7 +236,7 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml -o Globa
 | 2 | PP-OCRv3中英文超轻量检测预训练模型fintune学生模型 |80.0%|
 | 3 | PP-OCRv3中英文超轻量检测预训练模型fintune教师模型 |84.8%|
 
-### 4.3.5 采用CML蒸馏进一步提升student模型精度
+#### 4.3.5 采用CML蒸馏进一步提升student模型精度
 
 需要从4.3.3和4.3.4训练得到的best_accuracy.pdparams中提取各自代表student和teacher的参数，组合成适合cml训练的初始化模型，提取代码如下：
 
@@ -298,9 +298,15 @@ python tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Globa
 | 3 | PP-OCRv3中英文超轻量检测预训练模型fintune教师模型 |84.8%|
 | 4 | 基于2和3训练好的模型fintune |82.7%|
 
-###  4.3.6 模型导出推理
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型推理。
+
+####  4.3.6 模型导出推理
 训练完成后，可以将训练模型转换成inference模型。inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
-####  4.3.6.1 模型导出
+#####  4.3.6.1 模型导出
 导出命令如下：
 
 ```python
@@ -312,7 +318,7 @@ python tools/export_model.py \
 
 ```
 
-####  4.3.6.2 模型推理
+#####  4.3.6.2 模型推理
 导出模型后，可以使用如下命令进行推理预测：
 
 ```python
@@ -320,10 +326,10 @@ python tools/export_model.py \
 python tools/infer/predict_det.py --image_dir="train_data/icdar2015/text_localization/test/1.jpg" --det_model_dir="./inference/det_ppocrv3/Student"
 ```
 
-# 5. 文字识别
+## 5. 文字识别
 文本识别的任务是识别出图像中的文字内容，一般输入来自于文本检测得到的文本框截取出的图像文字区域。文本识别一般可以根据待识别文本形状分为规则文本识别和不规则文本识别两大类。规则文本主要指印刷字体、扫描文本等，文本大致处在水平线位置；不规则文本往往不在水平位置，存在弯曲、遮挡、模糊等问题。不规则文本场景具有很大的挑战性，也是目前文本识别领域的主要研究方向。本项目基于PP-OCRv3算法进行优化。
 
-## 5.1 PP-OCRv3识别算法介绍
+### 5.1 PP-OCRv3识别算法介绍
 PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2205.00159)优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。如下图所示，PP-OCRv3采用了6个优化策略。
 ![](https://ai-studio-static-online.cdn.bcebos.com/d4f5344b5b854d50be738671598a89a45689c6704c4d481fb904dd7cf72f2a1a)
 
@@ -337,7 +343,7 @@ PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2
 
 详细优化策略描述请参考[PP-OCRv3优化策略](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/PP-OCRv3_introduction.md#3-%E8%AF%86%E5%88%AB%E4%BC%98%E5%8C%96)
 
-## 5.2 数据准备
+### 5.2 数据准备
 [计量设备屏幕字符识别数据集](https://aistudio.baidu.com/aistudio/datasetdetail/128714)数据来源于实际项目中各种计量设备的数显屏，以及在网上搜集的一些其他数显屏，包含训练集19912张，测试集4099张。
 
 ```python
@@ -368,10 +374,10 @@ def get_one_image(train):
 get_one_image(train)
 ```
 
-![rec_png](imgs/rec.png)
+![rec_png](https://ai-studio-static-online.cdn.bcebos.com/3de0d475c69746d0a184029001ef07c85fd68816d66d4beaa10e6ef60030f9b4)
 
-## 5.3 模型训练
-###  下载预训练模型
+### 5.3 模型训练
+####  下载预训练模型
 下载我们需要的PP-OCRv3识别预训练模型，更多选择请自行选择其他的[文字识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/doc/doc_ch/models_list.md#2-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E6%A8%A1%E5%9E%8B)
 
 ```python
@@ -381,7 +387,7 @@ wget -P ./pretrained_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch
 tar -xf ./pretrained_models/ch_PP-OCRv3_rec_train.tar -C pretrained_models
 ```
 
-###  修改配置文件
+####  修改配置文件
 我们使用configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml，主要修改训练轮数和学习率参相关参数，设置预训练模型路径，设置数据集路径。 另外，batch_size可根据自己机器显存大小进行调整。 具体修改如下几个地方：
 ```
   epoch_num: 100 # 训练epoch数
@@ -439,7 +445,7 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o
 |---|---------------------------|---|
 | 0 | PP-OCRv3中英文超轻量识别预训练模型直接预测 |70.4%|
 
-###  开始训练
+####  开始训练
 我们使用上面修改好的配置文件configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml，预训练模型，数据集路径，学习率，训练轮数等都已经设置完毕后，可以使用下面命令开始训练。
 
 ```python
@@ -460,9 +466,15 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o
 | 0 | PP-OCRv3中英文超轻量识别预训练模型直接预测 |70.4%|
 | 1 | PP-OCRv3中英文超轻量识别预训练模型finetune |82.2%|
 
-##  5.4 模型导出推理
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型推理。
+
+###  5.4 模型导出推理
 训练完成后，可以将训练模型转换成inference模型。inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
-###  模型导出
+####  模型导出
 导出命令如下：
 
 ```python
@@ -470,7 +482,7 @@ python tools/eval.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o
 python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml -o Global.pretrained_model="./output/ch_PP-OCR_v3_rec/best_accuracy" Global.save_inference_dir="./inference/rec_ppocrv3/"
 ```
 
-### 模型推理
+#### 模型推理
 导出模型后，可以使用如下命令进行推理预测
 
 ```python
@@ -478,7 +490,7 @@ python tools/export_model.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillatio
 python tools/infer/predict_rec.py --image_dir="train_data/ic15_data/test/1_crop_0.jpg" --rec_model_dir="./inference/rec_ppocrv3/Student"
 ```
 
-# 6. 系统串联
+## 6. 系统串联
 我们将上面训练好的检测和识别模型进行系统串联测试，命令如下：
 
 ```python
@@ -500,9 +512,9 @@ plt.imshow(img)
 plt.show()
 ```
 
-![sys_res_png](imgs/sys_res.png)
+![sys_res_png](https://ai-studio-static-online.cdn.bcebos.com/901ab741cb46441ebec510b37e63b9d8d1b7c95f63cc4e5e8757f35179ae6373)
 
-## 6.1 后处理
+### 6.1 后处理
 如果需要获取key-value信息，可以基于启发式的规则，将识别结果与关键字库进行匹配；如果匹配上了，则取该字段为key, 后面一个字段为value。
 
 ```python
@@ -523,7 +535,7 @@ def postprocess(rec_res):
 key_value = postprocess(filter_rec_res)
 ```
 
-# 7. PaddleServing部署
+## 7. PaddleServing部署
 首先需要安装PaddleServing部署相关的环境
 
 ```python
@@ -532,7 +544,7 @@ python -m pip install paddle_serving_client
 python -m pip install paddle-serving-app
 ```
 
-##  7.1 转化检测模型
+###  7.1 转化检测模型
 
 ```python
 cd deploy/pdserving/
@@ -543,7 +555,7 @@ python -m paddle_serving_client.convert --dirname ../../inference/det_ppocrv3/St
                                          --serving_client ./ppocr_det_v3_client/
 ```
 
-## 7.2 转化识别模型
+### 7.2 转化识别模型
 
 ```python
 python -m paddle_serving_client.convert --dirname ../../inference/rec_ppocrv3/Student \
@@ -554,7 +566,7 @@ python -m paddle_serving_client.convert --dirname ../../inference/rec_ppocrv3/St
 ```
 
 
-## 7.3 启动服务
+### 7.3 启动服务
 首先可以将后处理代码加入到web_service.py中，具体修改如下：
 ```
 # 代码153行后面增加下面代码
@@ -582,7 +594,7 @@ res = {"result": str(key_value)}
 python web_service.py 2>&1 >log.txt
 ```
 
-## 7.4 发送请求
+### 7.4 发送请求
 然后再开启一个新的终端，运行下面的客户端代码
 
 ```python

From 2916625a8f88f1e6d065fbdfe064ee910712c1b4 Mon Sep 17 00:00:00 2001
From: xiaoting <31891223+tink2123@users.noreply.github.com>
Date: Wed, 29 Jun 2022 15:31:48 +0800
Subject: [PATCH 080/134] =?UTF-8?q?Update=20=E5=85=89=E5=8A=9F=E7=8E=87?=
 =?UTF-8?q?=E8=AE=A1=E6=95=B0=E7=A0=81=E7=AE=A1=E5=AD=97=E7=AC=A6=E8=AF=86?=
 =?UTF-8?q?=E5=88=AB.md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 applications/光功率计数码管字符识别.md | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/applications/光功率计数码管字符识别.md b/applications/光功率计数码管字符识别.md
index d6c11259bf..2a35cb170e 100644
--- a/applications/光功率计数码管字符识别.md
+++ b/applications/光功率计数码管字符识别.md
@@ -21,12 +21,16 @@
 
 第二步，使用文本识别模型获得准确的数字和单位信息。
 
-本项目主要介绍如何完成第二步文本识别部分，包括：
+本项目主要介绍如何完成第二步文本识别部分，包括：真实评估集的建立、训练数据的合成、基于 PP-OCRv3 和 SVTR_Tiny 两个模型进行训练，以及评估和推理。
 
-* 真实评估集的建立
-* 训练数据的合成
-* 基于 PP-OCRv3 和 SVTR_Tiny 两个模型进行训练
-* 评估并推理
+本项目难点如下：
+
+- 光功率计数码管字符数据较少，难以获取。
+- 数码管中小数点占像素较少，容易漏识别。
+
+针对以上问题， 本例选用 PP-OCRv3 和 SVTR_Tiny 两个高精度模型训练，同时提供了真实数据挖掘案例和数据合成案例。基于 PP-OCRv3 模型，在构建的真实评估集上精度从 52% 提升至 72%，SVTR_Tiny 模型精度可达到 78.9%。
+
+aistudio项目链接: [光功率计数码管字符识别](https://aistudio.baidu.com/aistudio/projectdetail/4049044?contributionType=1)
 
 ## 2. PaddleOCR 快速使用
 

From 46fafa298710cf7b4555a98aa03fefa18fc16c33 Mon Sep 17 00:00:00 2001
From: an1018 <614803115@qq.com>
Date: Wed, 29 Jun 2022 17:01:47 +0800
Subject: [PATCH 081/134] update

---
 applications/PCB字符识别/PCB字符识别.md |  6 ++++-
 applications/多模态表单识别.md          | 36 ++++++++++++-------------
 2 files changed, 22 insertions(+), 20 deletions(-)

diff --git a/applications/PCB字符识别/PCB字符识别.md b/applications/PCB字符识别/PCB字符识别.md
index a5052e2897..ee13bacffd 100644
--- a/applications/PCB字符识别/PCB字符识别.md
+++ b/applications/PCB字符识别/PCB字符识别.md
@@ -206,7 +206,11 @@ Eval.dataset.transforms.DetResizeForTest:  尺寸
         limit_type: 'min'
 ```
 
-然后执行评估代码
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型评估。
 
 
 ```python
diff --git a/applications/多模态表单识别.md b/applications/多模态表单识别.md
index 27f981101a..471ca633c1 100644
--- a/applications/多模态表单识别.md
+++ b/applications/多模态表单识别.md
@@ -7,12 +7,12 @@
     - [3.2 转换为PaddleOCR检测和识别格式](#32-转换为paddleocr检测和识别格式)
   - [4 OCR](#4-ocr)
     - [4.1 文本检测](#41-文本检测)
-      - [**4.1.1 方案1：预训练模型**](#411-方案1预训练模型)
-      - [**4.1.2 方案2：XFUND数据集+fine-tune**](#412-方案2xfund数据集fine-tune)
+      - [4.1.1 方案1：预训练模型](#411-方案1预训练模型)
+      - [4.1.2 方案2：XFUND数据集+fine-tune](#412-方案2xfund数据集fine-tune)
     - [4.2 文本识别](#42-文本识别)
-      - [**4.2.1 方案1：预训练模型**](#421-方案1预训练模型)
-      - [**4.2.2 方案2：XFUND数据集+finetune**](#422-方案2xfund数据集finetune)
-      - [**4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据**](#423-方案3xfund数据集finetune真实通用识别数据)
+      - [4.2.1 方案1：预训练模型](#421-方案1预训练模型)
+      - [4.2.2 方案2：XFUND数据集+finetune](#422-方案2xfund数据集finetune)
+      - [4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据](#423-方案3xfund数据集finetune真实通用识别数据)
   - [5 文档视觉问答(DOC-VQA)](#5-文档视觉问答doc-vqa)
     - [5.1 SER](#51-ser)
       - [5.1.1 模型训练](#511-模型训练)
@@ -195,7 +195,7 @@ python trans_xfund_data.py
 -  **PP-OCRv2中英文超轻量检测预训练模型**
 -  **XFUND数据集+fine-tune**
 
-#### **4.1.1 方案1：预训练模型**
+#### 4.1.1 方案1：预训练模型
 
 **1）下载预训练模型**
 
@@ -265,7 +265,7 @@ python tools/eval.py \
 
 使用文本检测预训练模型在XFUND验证集上评估，达到77%左右，充分说明ppocr提供的预训练模型具有泛化能力。
 
-#### **4.1.2 方案2：XFUND数据集+fine-tune**
+#### 4.1.2 方案2：XFUND数据集+fine-tune
 
 PaddleOCR提供的蒸馏预训练模型包含了多个模型的参数，我们提取Student模型的参数，在XFUND数据集上进行finetune，可以参考如下代码：
 
@@ -316,7 +316,13 @@ CUDA_VISIBLE_DEVICES=0 python tools/train.py \
 <center><img src="https://ai-studio-static-online.cdn.bcebos.com/5a75137c5f924dfeb6956b5818812298cc3dc7992ac84954b4175be9adf83c77"></center>
 <center>图8 文本检测方案2-模型评估</center>
 
-使用训练好的模型进行评估，更新模型路径`Global.checkpoints`。
+使用训练好的模型进行评估，更新模型路径`Global.checkpoints`。如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+将下载或训练完成的模型放置在对应目录下即可完成模型评估
 
 
 ```python
@@ -412,7 +418,7 @@ use_gpu：是否使用GPU
 - XFUND数据集+fine-tune
 - XFUND数据集+fine-tune+真实通用识别数据
 
-#### **4.2.1 方案1：预训练模型**
+#### 4.2.1 方案1：预训练模型
 
 **1）下载预训练模型**
 
@@ -461,7 +467,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
 
 使用文本预训练模型在XFUND验证集上评估，acc达到67%左右，充分说明ppocr提供的预训练模型具有泛化能力。
 
-#### **4.2.2 方案2：XFUND数据集+finetune**
+#### 4.2.2 方案2：XFUND数据集+finetune
 
 同检测模型，我们提取Student模型的参数，在XFUND数据集上进行finetune，可以参考如下代码：
 
@@ -528,7 +534,7 @@ CUDA_VISIBLE_DEVICES=0 python tools/eval.py \
 
 使用XFUND数据集+finetune训练，在验证集上评估达到72%左右，说明 finetune会提升垂类场景效果。
 
-#### **4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据**
+#### 4.2.3 方案3：XFUND数据集+finetune+真实通用识别数据
 
 接着我们在上述`XFUND数据集+finetune`实验的基础上，添加真实通用识别数据，进一步提升识别效果。首先准备真实通用识别数据，并上传到AIStudio：
 
@@ -873,14 +879,6 @@ with open('output/re/infer_results.txt', 'r', encoding='utf-8') as fin:
 workbook.close()
 ```
 
-## 获得模型
-
-如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
-<div align="left">
-<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
-</div>
-将下载或训练完成的模型放置在对应目录下即可完成模型推理
-
 ## 更多资源
 
 - 更多深度学习知识、产业案例、面试宝典等，请参考：[awesome-DeepLearning](https://github.com/paddlepaddle/awesome-DeepLearning)

From 304162824093060b964b097c4d29012b00b5ed9a Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Wed, 29 Jun 2022 09:12:27 +0000
Subject: [PATCH 082/134] fix serving log

---
 test_tipc/test_serving_infer_cpp.sh    |  6 ++++--
 test_tipc/test_serving_infer_python.sh | 15 ++++++++-------
 2 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/test_tipc/test_serving_infer_cpp.sh b/test_tipc/test_serving_infer_cpp.sh
index 4088c66f57..6e313cd949 100644
--- a/test_tipc/test_serving_infer_cpp.sh
+++ b/test_tipc/test_serving_infer_cpp.sh
@@ -87,11 +87,13 @@ function func_serving(){
     set_image_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
     python_list=(${python_list})
     cd ${serving_dir_value}
+    unset https_proxy
+    unset http_proxy
     # cpp serving
     for gpu_id in ${gpu_value[*]}; do
         if [ ${gpu_id} = "null" ]; then
             server_log_path="${LOG_PATH}/cpp_server_cpu.log"
-            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > ${server_log_path} 2>&1 "
+            web_service_cpp_cmd="nohup ${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > ${server_log_path} 2>&1 &"
             eval $web_service_cpp_cmd
             last_status=${PIPESTATUS[0]}
             status_check $last_status "${web_service_cpp_cmd}" "${status_log}" "${model_name}"
@@ -105,7 +107,7 @@ function func_serving(){
             ps ux | grep -i ${port_value} | awk '{print $2}' | xargs kill -s 9
         else
             server_log_path="${LOG_PATH}/cpp_server_gpu.log"
-            web_service_cpp_cmd="${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > ${server_log_path} 2>&1 "
+            web_service_cpp_cmd="nohup ${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > ${server_log_path} 2>&1 &"
             eval $web_service_cpp_cmd
             sleep 5s
             _save_log_path="${LOG_PATH}/cpp_client_gpu.log"
diff --git a/test_tipc/test_serving_infer_python.sh b/test_tipc/test_serving_infer_python.sh
index 57dab6aeb5..b14517265d 100644
--- a/test_tipc/test_serving_infer_python.sh
+++ b/test_tipc/test_serving_infer_python.sh
@@ -112,7 +112,8 @@ function func_serving(){
     
     cd ${serving_dir_value}
     python=${python_list[0]}
-        
+    unset https_proxy
+    unset http_proxy
     # python serving
     for use_gpu in ${web_use_gpu_list[*]}; do
         if [ ${use_gpu} = "null" ]; then
@@ -123,19 +124,19 @@ function func_serving(){
                     if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2.0" ] || [ ${model_name} = "ch_ppocr_server_v2.0" ]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 "
+                        web_service_cmd="nohup ${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "det" ]]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} > ${server_log_path} 2>&1 "
+                        web_service_cmd="nohup ${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} > ${server_log_path} 2>&1 &"
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "rec" ]]; then
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} > ${server_log_path} 2>&1 "
+                        web_service_cmd="nohup ${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
@@ -174,19 +175,19 @@ function func_serving(){
                     if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2.0" ] || [ ${model_name} = "ch_ppocr_server_v2.0" ]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 "
+                        web_service_cmd="nohup ${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "det" ]]; then
                         set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} > ${server_log_path} 2>&1 "
+                        web_service_cmd="nohup ${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} > ${server_log_path} 2>&1 &"
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"
                     elif [[ ${model_name} =~ "rec" ]]; then
                         set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
-                        web_service_cmd="${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} > ${server_log_path} 2>&1 "
+                        web_service_cmd="nohup ${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
                         eval $web_service_cmd
                         last_status=${PIPESTATUS[0]}
                         status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}"

From 0aa00f1ce840de5d764a786276eaf12d8a11890d Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Wed, 29 Jun 2022 09:39:15 +0000
Subject: [PATCH 083/134] update case

---
 applications/液晶屏读数识别.md | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/applications/液晶屏读数识别.md b/applications/液晶屏读数识别.md
index b85bf56daf..ff2fb2cb48 100644
--- a/applications/液晶屏读数识别.md
+++ b/applications/液晶屏读数识别.md
@@ -31,6 +31,8 @@
 ## 2. 项目内容
 本项目基于PaddleOCR开源套件，以PP-OCRv3检测和识别模型为基础，针对液晶屏读数识别场景进行优化。
 
+Aistudio项目链接：[OCR液晶屏读数识别](https://aistudio.baidu.com/aistudio/projectdetail/4080130)
+
 ## 3. 安装环境
 
 ```python

From b998a95c7d8ea24af16b8406d9429885910b4900 Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 18:54:34 +0800
Subject: [PATCH 084/134] add application.md

---
 doc/doc_ch/application.md | 41 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
 create mode 100644 doc/doc_ch/application.md

diff --git a/doc/doc_ch/application.md b/doc/doc_ch/application.md
new file mode 100644
index 0000000000..5135dfac10
--- /dev/null
+++ b/doc/doc_ch/application.md
@@ -0,0 +1,41 @@
+# 场景应用
+
+PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。
+
+> 如需下载全部垂类模型，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群获取20G OCR学习大礼包（内含《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料）
+
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
+> 如果您是企业开发者且未在下述场景中找到合适的方案，可以填写[OCR应用合作调研问卷](https://paddle.wjx.cn/vj/QwF7GKw.aspx)，免费与官方团队展开不同层次的合作，包括但不限于问题抽象、确定技术方案、项目答疑、共同研发等。如果您已经使用PaddleOCR落地项目，也可以填写此问卷，与飞桨平台共同宣传推广，提升企业技术品宣。期待您的提交！
+
+## 通用
+
+| 类别                   | 亮点     | 类别       | 亮点         |
+| ---------------------- | -------- | ---------- | ------------ |
+| 高精度中文识别模型SVTR | 新增模型 | 手写体识别 | 新增字形支持 |
+
+## 制造
+
+| 类别           | 亮点                           | 类别           | 亮点                 |
+| -------------- | ------------------------------ | -------------- | -------------------- |
+| 数码管识别     | 数码管数据合成、漏识别调优     | 电表识别       | 大分辨率图像检测调优 |
+| 液晶屏读数识别 | 检测模型蒸馏、Serving部署      | PCB文字识别    | 小尺寸文本检测与识别 |
+| 包装生产日期   | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测 | 非文字形态识别       |
+
+## 金融
+
+| 类别           | 亮点                     | 类别         | 亮点                  |
+| -------------- | ------------------------ | ------------ | --------------------- |
+| 表单VQA        | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
+| 增值税发票     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
+| 印章检测与识别 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
+
+## 交通
+
+| 类别              | 亮点                           | 类别       | 亮点     |
+| ----------------- | ------------------------------ | ---------- | -------- |
+| 车牌识别          | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
+| 驾驶证/行驶证识别 | 尽请期待                       |            |          |
\ No newline at end of file

From 14dedcacf6dab4eda75927f1feff9177f961c932 Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 18:56:24 +0800
Subject: [PATCH 085/134] move application.md

---
 applications/application.md | 41 +++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
 create mode 100644 applications/application.md

diff --git a/applications/application.md b/applications/application.md
new file mode 100644
index 0000000000..5135dfac10
--- /dev/null
+++ b/applications/application.md
@@ -0,0 +1,41 @@
+# 场景应用
+
+PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。
+
+> 如需下载全部垂类模型，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群获取20G OCR学习大礼包（内含《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料）
+
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
+> 如果您是企业开发者且未在下述场景中找到合适的方案，可以填写[OCR应用合作调研问卷](https://paddle.wjx.cn/vj/QwF7GKw.aspx)，免费与官方团队展开不同层次的合作，包括但不限于问题抽象、确定技术方案、项目答疑、共同研发等。如果您已经使用PaddleOCR落地项目，也可以填写此问卷，与飞桨平台共同宣传推广，提升企业技术品宣。期待您的提交！
+
+## 通用
+
+| 类别                   | 亮点     | 类别       | 亮点         |
+| ---------------------- | -------- | ---------- | ------------ |
+| 高精度中文识别模型SVTR | 新增模型 | 手写体识别 | 新增字形支持 |
+
+## 制造
+
+| 类别           | 亮点                           | 类别           | 亮点                 |
+| -------------- | ------------------------------ | -------------- | -------------------- |
+| 数码管识别     | 数码管数据合成、漏识别调优     | 电表识别       | 大分辨率图像检测调优 |
+| 液晶屏读数识别 | 检测模型蒸馏、Serving部署      | PCB文字识别    | 小尺寸文本检测与识别 |
+| 包装生产日期   | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测 | 非文字形态识别       |
+
+## 金融
+
+| 类别           | 亮点                     | 类别         | 亮点                  |
+| -------------- | ------------------------ | ------------ | --------------------- |
+| 表单VQA        | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
+| 增值税发票     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
+| 印章检测与识别 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
+
+## 交通
+
+| 类别              | 亮点                           | 类别       | 亮点     |
+| ----------------- | ------------------------------ | ---------- | -------- |
+| 车牌识别          | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
+| 驾驶证/行驶证识别 | 尽请期待                       |            |          |
\ No newline at end of file

From 8dd60b566ffa4fc2697bf6b1c3dfcb1d890680d5 Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 18:57:26 +0800
Subject: [PATCH 086/134] update name

---
 applications/README.md | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
 create mode 100644 applications/README.md

diff --git a/applications/README.md b/applications/README.md
new file mode 100644
index 0000000000..5135dfac10
--- /dev/null
+++ b/applications/README.md
@@ -0,0 +1,41 @@
+# 场景应用
+
+PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。
+
+> 如需下载全部垂类模型，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群获取20G OCR学习大礼包（内含《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料）
+
+<div align="center">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
+> 如果您是企业开发者且未在下述场景中找到合适的方案，可以填写[OCR应用合作调研问卷](https://paddle.wjx.cn/vj/QwF7GKw.aspx)，免费与官方团队展开不同层次的合作，包括但不限于问题抽象、确定技术方案、项目答疑、共同研发等。如果您已经使用PaddleOCR落地项目，也可以填写此问卷，与飞桨平台共同宣传推广，提升企业技术品宣。期待您的提交！
+
+## 通用
+
+| 类别                   | 亮点     | 类别       | 亮点         |
+| ---------------------- | -------- | ---------- | ------------ |
+| 高精度中文识别模型SVTR | 新增模型 | 手写体识别 | 新增字形支持 |
+
+## 制造
+
+| 类别           | 亮点                           | 类别           | 亮点                 |
+| -------------- | ------------------------------ | -------------- | -------------------- |
+| 数码管识别     | 数码管数据合成、漏识别调优     | 电表识别       | 大分辨率图像检测调优 |
+| 液晶屏读数识别 | 检测模型蒸馏、Serving部署      | PCB文字识别    | 小尺寸文本检测与识别 |
+| 包装生产日期   | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测 | 非文字形态识别       |
+
+## 金融
+
+| 类别           | 亮点                     | 类别         | 亮点                  |
+| -------------- | ------------------------ | ------------ | --------------------- |
+| 表单VQA        | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
+| 增值税发票     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
+| 印章检测与识别 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
+
+## 交通
+
+| 类别              | 亮点                           | 类别       | 亮点     |
+| ----------------- | ------------------------------ | ---------- | -------- |
+| 车牌识别          | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
+| 驾驶证/行驶证识别 | 尽请期待                       |            |          |
\ No newline at end of file

From ab13b42f91527cf6224ef18e81104105c45a4a73 Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 18:58:05 +0800
Subject: [PATCH 087/134] Delete application.md

---
 applications/application.md | 41 -------------------------------------
 1 file changed, 41 deletions(-)
 delete mode 100644 applications/application.md

diff --git a/applications/application.md b/applications/application.md
deleted file mode 100644
index 5135dfac10..0000000000
--- a/applications/application.md
+++ /dev/null
@@ -1,41 +0,0 @@
-# 场景应用
-
-PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。
-
-> 如需下载全部垂类模型，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群获取20G OCR学习大礼包（内含《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料）
-
-<div align="center">
-<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
-</div>
-
-
-> 如果您是企业开发者且未在下述场景中找到合适的方案，可以填写[OCR应用合作调研问卷](https://paddle.wjx.cn/vj/QwF7GKw.aspx)，免费与官方团队展开不同层次的合作，包括但不限于问题抽象、确定技术方案、项目答疑、共同研发等。如果您已经使用PaddleOCR落地项目，也可以填写此问卷，与飞桨平台共同宣传推广，提升企业技术品宣。期待您的提交！
-
-## 通用
-
-| 类别                   | 亮点     | 类别       | 亮点         |
-| ---------------------- | -------- | ---------- | ------------ |
-| 高精度中文识别模型SVTR | 新增模型 | 手写体识别 | 新增字形支持 |
-
-## 制造
-
-| 类别           | 亮点                           | 类别           | 亮点                 |
-| -------------- | ------------------------------ | -------------- | -------------------- |
-| 数码管识别     | 数码管数据合成、漏识别调优     | 电表识别       | 大分辨率图像检测调优 |
-| 液晶屏读数识别 | 检测模型蒸馏、Serving部署      | PCB文字识别    | 小尺寸文本检测与识别 |
-| 包装生产日期   | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测 | 非文字形态识别       |
-
-## 金融
-
-| 类别           | 亮点                     | 类别         | 亮点                  |
-| -------------- | ------------------------ | ------------ | --------------------- |
-| 表单VQA        | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
-| 增值税发票     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
-| 印章检测与识别 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
-
-## 交通
-
-| 类别              | 亮点                           | 类别       | 亮点     |
-| ----------------- | ------------------------------ | ---------- | -------- |
-| 车牌识别          | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
-| 驾驶证/行驶证识别 | 尽请期待                       |            |          |
\ No newline at end of file

From 427628bb3f5e726df5f531fed78b5101b2edb93c Mon Sep 17 00:00:00 2001
From: Leif <4603009@qq.com>
Date: Wed, 29 Jun 2022 19:19:39 +0800
Subject: [PATCH 088/134] Update readme of applications

Update readme of applications
---
 applications/README.md                        |  34 +++++++++---------
 .../corpus/digital.txt                        |   0
 .../fonts/DS-DIGI.TTF                         | Bin
 .../fonts/DS-DIGIB.TTF                        | Bin
 .../光功率计数码管字符识别.md                 |   0
 5 files changed, 17 insertions(+), 17 deletions(-)
 rename applications/{ => 光功率计数码管字符识别}/corpus/digital.txt (100%)
 rename applications/{ => 光功率计数码管字符识别}/fonts/DS-DIGI.TTF (100%)
 rename applications/{ => 光功率计数码管字符识别}/fonts/DS-DIGIB.TTF (100%)
 rename applications/{ => 光功率计数码管字符识别}/光功率计数码管字符识别.md (100%)

diff --git a/applications/README.md b/applications/README.md
index 5135dfac10..eba1e205dc 100644
--- a/applications/README.md
+++ b/applications/README.md
@@ -13,29 +13,29 @@ PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR
 
 ## 通用
 
-| 类别                   | 亮点     | 类别       | 亮点         |
-| ---------------------- | -------- | ---------- | ------------ |
-| 高精度中文识别模型SVTR | 新增模型 | 手写体识别 | 新增字形支持 |
+| 类别                                              | 亮点     | 类别       | 亮点         |
+| ------------------------------------------------- | -------- | ---------- | ------------ |
+| [高精度中文识别模型SVTR](./高精度中文识别模型.md) | 新增模型 | 手写体识别 | 新增字形支持 |
 
 ## 制造
 
-| 类别           | 亮点                           | 类别           | 亮点                 |
-| -------------- | ------------------------------ | -------------- | -------------------- |
-| 数码管识别     | 数码管数据合成、漏识别调优     | 电表识别       | 大分辨率图像检测调优 |
-| 液晶屏读数识别 | 检测模型蒸馏、Serving部署      | PCB文字识别    | 小尺寸文本检测与识别 |
-| 包装生产日期   | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测 | 非文字形态识别       |
+| 类别                                                         | 亮点                           | 类别                                        | 亮点                 |
+| ------------------------------------------------------------ | ------------------------------ | ------------------------------------------- | -------------------- |
+| [数码管识别](./光功率计数码管字符识别/光功率计数码管字符识别.md) | 数码管数据合成、漏识别调优     | 电表识别                                    | 大分辨率图像检测调优 |
+| [液晶屏读数识别](./液晶屏读数识别.md)                        | 检测模型蒸馏、Serving部署      | [PCB文字识别](./PCB字符识别/PCB字符识别.md) | 小尺寸文本检测与识别 |
+| [包装生产日期](./包装生产日期识别.md)                        | 点阵字符合成、过曝过暗文字识别 | 液晶屏缺陷检测                              | 非文字字符识别       |
 
 ## 金融
 
-| 类别           | 亮点                     | 类别         | 亮点                  |
-| -------------- | ------------------------ | ------------ | --------------------- |
-| 表单VQA        | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
-| 增值税发票     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
-| 印章检测与识别 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
+| 类别                           | 亮点                     | 类别         | 亮点                  |
+| ------------------------------ | ------------------------ | ------------ | --------------------- |
+| [表单VQA](./多模态表单识别.md) | 多模态通用表单结构化提取 | 通用卡证识别 | 通用结构化提取        |
+| 增值税发票                     | 尽请期待                 | 身份证识别   | 结构化提取、图像阴影  |
+| 印章检测与识别                 | 端到端弯曲文本识别       | 合同比对     | 密集文本检测、NLP串联 |
 
 ## 交通
 
-| 类别              | 亮点                           | 类别       | 亮点     |
-| ----------------- | ------------------------------ | ---------- | -------- |
-| 车牌识别          | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
-| 驾驶证/行驶证识别 | 尽请期待                       |            |          |
\ No newline at end of file
+| 类别                            | 亮点                           | 类别       | 亮点     |
+| ------------------------------- | ------------------------------ | ---------- | -------- |
+| [车牌识别](./轻量级车牌识别.md) | 多角度图像、轻量模型、端侧部署 | 快递单识别 | 尽请期待 |
+| 驾驶证/行驶证识别               | 尽请期待                       |            |          |
\ No newline at end of file
diff --git a/applications/corpus/digital.txt b/applications/光功率计数码管字符识别/corpus/digital.txt
similarity index 100%
rename from applications/corpus/digital.txt
rename to applications/光功率计数码管字符识别/corpus/digital.txt
diff --git a/applications/fonts/DS-DIGI.TTF b/applications/光功率计数码管字符识别/fonts/DS-DIGI.TTF
similarity index 100%
rename from applications/fonts/DS-DIGI.TTF
rename to applications/光功率计数码管字符识别/fonts/DS-DIGI.TTF
diff --git a/applications/fonts/DS-DIGIB.TTF b/applications/光功率计数码管字符识别/fonts/DS-DIGIB.TTF
similarity index 100%
rename from applications/fonts/DS-DIGIB.TTF
rename to applications/光功率计数码管字符识别/fonts/DS-DIGIB.TTF
diff --git a/applications/光功率计数码管字符识别.md b/applications/光功率计数码管字符识别/光功率计数码管字符识别.md
similarity index 100%
rename from applications/光功率计数码管字符识别.md
rename to applications/光功率计数码管字符识别/光功率计数码管字符识别.md

From 894221e105bfda4bb576f2396954108a9d95c86d Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 19:29:20 +0800
Subject: [PATCH 089/134] Update README_ch.md

---
 README_ch.md | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/README_ch.md b/README_ch.md
index f837980498..e801ce561c 100755
--- a/README_ch.md
+++ b/README_ch.md
@@ -71,6 +71,8 @@ PaddleOCR旨在打造一套丰富、领先、且实用的OCR工具库，助力
 ## 《动手学OCR》电子书
 - [《动手学OCR》电子书📚](./doc/doc_ch/ocr_book.md)
 
+## 场景应用
+- PaddleOCR场景应用覆盖通用，制造、金融、交通行业的主要OCR垂类应用，在PP-OCR、PP-Structure的通用能力基础之上，以notebook的形式展示利用场景数据微调、模型优化方法、数据增广等内容，为开发者快速落地OCR应用提供示范与启发。详情可查看[README](./applications)。
 
 <a name="开源社区"></a>
 ## 开源社区

From 4959c1c2720da9cfe9ea59b6a7825ec31a196ada Mon Sep 17 00:00:00 2001
From: huangjun12 <12272008@bjtu.edu.cn>
Date: Wed, 29 Jun 2022 11:56:55 +0000
Subject: [PATCH 090/134] update url

---
 applications/包装生产日期识别.md | 174 ++++---------------------------
 1 file changed, 18 insertions(+), 156 deletions(-)

diff --git a/applications/包装生产日期识别.md b/applications/包装生产日期识别.md
index 4f74a4a148..a0fec65f31 100644
--- a/applications/包装生产日期识别.md
+++ b/applications/包装生产日期识别.md
@@ -40,6 +40,8 @@
 | 真实数据finetune | 71.33|
 | 真实+合成数据finetune | 86.99|
 
+AIStudio项目链接： [一种基于PaddleOCR的包装生产日期识别方法](https://aistudio.baidu.com/aistudio/projectdetail/4287736)
+
 ## 2. 环境搭建
 
 本任务基于Aistudio完成, 具体环境如下：
@@ -96,7 +98,6 @@ PaddleOCR
 │   └── val.list                 # 测试集数据文件列表
 |   ├── bg                       # 合成数据所需背景图像
 │   └── corpus                   # 合成数据所需语料
-│   └── rec_vit_sub_64_363_all/  # svtr_tiny高精度识别模型
 ```
 
 ## 4. 直接使用PP-OCRv3模型评估
@@ -417,160 +418,9 @@ Eval.loader.batch_size_per_card: 评估单卡batch size
 
 ### 6.1 python爬虫获取数据
 
-本节介绍如何使用python脚本爬取网络图片，这里以爬取百度图片为例。首先将图片链接都爬取下来，然后用多进程下载图片。使用到的工具有requests库和谷歌浏览器（抓包工具）。
+- 推荐使用[爬虫工具](https://github.com/Joeclinton1/google-images-download)获取无标签图片。
 
-准备工作：
-1. 获取请求头需要的的设备信息：
-(1) 使用谷歌浏览器，打开百度图片，搜索关键词【包装生产日期】
-(2) 右键，选择【检查】，打开抓包工具
-(3) Network >> XHR >> 选定网址 >> Headers >> RequestHeaders >> UserAgents
-![](https://ai-studio-static-online.cdn.bcebos.com/715fa3f1255544789ddf5699084783b6b88055835a7049e3b588f4f564e14972)
-
-2. 获取网页参数
-用类似的方式，在XHR中选择Payload获取网页参数
-![](https://ai-studio-static-online.cdn.bcebos.com/2a8b91ed854a40e589cc9a9596873192cc730c60d5754373a403f395a81ec44f)
-
-
-接下来编写代码进行图片抓取：
-
-```python
-import requests
-import json
-import tqdm
-import os
-
-img_num = 0        # 记录保存下来的图片数量
-num_per_page = 50  # 每一页展示的图片数量
-page_num = 40      # 页数
-page_id = 0        # 每一页开始索引
-
-# 图片链接保存文件路径
-list_path = '/home/aistudio/sprider'
-if not os.path.exists(list_path):
-    os.mkdir(list_path)
-f_list = open(os.path.join(list_path, 'file.list'), 'w')
-
-
-img_dict = {}   # 图片链接字典，去重
-
-# 搜索关键词
-key_word = ['包装产品生产日期', '包装日期', '生产日期', '保质期', '纸箱生产日期', '商品日期', '黑色生产日期', '黑底生产日期']
-
-for key in key_word:
-    bar = tqdm.tqdm(total=page_num*num_per_page)
-    for page_id in range(page_num):
-        # 使用准备好的请求头设备信息
-        header = {
-            'User-Agent':'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/101.0.4951.64 Safari/537.36'
-        }
-
-        # 使用准备好的网页参数
-        url = 'https://image.baidu.com/search/acjson?'
-        param = {
-        'tn': 'resultjson_com',
-        'logid': '8931113288182398301',
-        'ipn': 'rj',
-        'ct': '201326592',
-        'is': '',
-        'fp': 'result',
-        'fr':'',
-        'word': key,       # 搜索关键词
-        'queryWord': key,  # 搜索关键词
-        'cl': '2',
-        'lm': '-1',
-        'ie': 'utf-8',
-        'oe': 'utf-8',
-        'adpicid': '',
-        'st': '-1',
-        'z': '',
-        'ic': '0',
-        'hd': '',
-        'latest': '',
-        'copyright': '',
-        's': '',
-        'se': '',
-        'tab': '',
-        'width': '',
-        'height': '',
-        'face': '0',
-        'istype': '2',
-        'qc': '',
-        'nc': '1',
-        'expermode': '',
-        'nojc': '',
-        'isAsync': '',
-        'pn': str(page_id*num_per_page),  # 每一页开始索引
-        'rn': str(num_per_page),          # 每一页爬取的图片数量
-        'gsm': '1e',
-        '1652689395185':''
-        }
-
-        # 获取网页信息  
-        page_text = requests.get(url=url, headers=header, params=param, timeout=20)
-
-        try:
-            page_text = page_text.json()  # 网页信息转成json字典
-        except:
-            continue
-
-        info_list = page_text['data']
-
-        # 获取不到有效信息后，退出搜索
-        if info_list == [{}]:
-            print('===break===')
-            break
-
-        del info_list[-1]
-
-        img_path_list = []
-        for info in info_list:
-            img_url = info['thumbURL']   # 图片源地址
-            #bar.update(1)
-            if img_url not in img_dict:
-                img_dict[img_url] = 1
-                f_list.write(img_url + '\n')
-
-f_list.close()
-print('saved image number:', img_num+1)
-```
-
-```python
-# 使用多进程下载网页图片
-
-import requests
-from concurrent.futures import ProcessPoolExecutor
-
-file_path = '/home/aistudio/sprider/file.list'
-image_save_path = '/home/aistudio/sprider/data'
-
-if not os.path.exists(image_save_path):
-    os.mkdir(image_save_path)
-
-f = open(file_path, 'r')
-lines = f.readlines()
-
-header = {
-    'User-Agent':'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/101.0.4951.64 Safari/537.36'
-}
-
-def process(img_url, img_num):
-    try:
-        img_data = requests.get(url=img_url, headers=header, timeout=20).content
-        img_path = os.path.join(image_save_path, '{:05d}.jpg'.format(img_num))
-        with open(img_path, 'wb') as fp:
-            fp.write(img_data)
-        img_num += 1
-    except Exception as e:
-        print('connect Failed, just continue')
-
-with ProcessPoolExecutor(max_workers=10) as executer:
-    for idx, line in enumerate(lines):
-        executer.submit(process, line.strip(), idx)
-
-print('done')
-```
-
-完成以上步骤，便可得到获取的图片，目录格式如下：
+图片获取后，可按如下目录格式组织：
 
 ```txt
 sprider
@@ -591,7 +441,13 @@ sprider
 
 首先下载预训练模型，PP-OCRv3检测模型下载链接：https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar
 
-svtr-tiny识别模型，我们使用的是在内部数据上训练的高精度模型，本项目提供下载链接，完成步骤3的数据准备后，可得数据路径位于:
+svtr-tiny识别模型，我们使用的是在内部数据上训练的高精度模型。如需获取该训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+
+
+完成下载后，可将模型存储于如下位置:
 
 ```shell
 PaddleOCR
@@ -824,4 +680,10 @@ python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
 
 可以看到，相较于原始PP-OCRv3的识别精度62.99%，使用合成数据+真实数据finetune后，识别精度能提升24%。
 
-训练完成后，PP-OCRv3模型的推理部署方法可以参考repo文档： https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/deploy/README_ch.md
+如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+<div align="left">
+<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
+</div>
+将下载或训练完成的模型放置在对应目录下即可完成模型推理。
+
+模型的推理部署方法可以参考repo文档： https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/deploy/README_ch.md

From 3ae853f070e0cc64fc263afa2d6b021421ae2494 Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 20:03:34 +0800
Subject: [PATCH 091/134] =?UTF-8?q?Update=20=E5=8C=85=E8=A3=85=E7=94=9F?=
 =?UTF-8?q?=E4=BA=A7=E6=97=A5=E6=9C=9F=E8=AF=86=E5=88=AB.md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 applications/包装生产日期识别.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/applications/包装生产日期识别.md b/applications/包装生产日期识别.md
index a0fec65f31..d24bacd584 100644
--- a/applications/包装生产日期识别.md
+++ b/applications/包装生产日期识别.md
@@ -441,7 +441,7 @@ sprider
 
 首先下载预训练模型，PP-OCRv3检测模型下载链接：https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar
 
-svtr-tiny识别模型，我们使用的是在内部数据上训练的高精度模型。如需获取该训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
+如需获取svtr-tiny高精度中文识别预训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
 <div align="left">
 <img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
 </div>

From 5bf0bd870929b2f40227b802dc3b90ef6e0225e5 Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Wed, 29 Jun 2022 20:07:24 +0800
Subject: [PATCH 092/134] =?UTF-8?q?Update=20=E5=8C=85=E8=A3=85=E7=94=9F?=
 =?UTF-8?q?=E4=BA=A7=E6=97=A5=E6=9C=9F=E8=AF=86=E5=88=AB.md?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 applications/包装生产日期识别.md | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/applications/包装生产日期识别.md b/applications/包装生产日期识别.md
index d24bacd584..73c174c4f9 100644
--- a/applications/包装生产日期识别.md
+++ b/applications/包装生产日期识别.md
@@ -680,10 +680,6 @@ python tools/train.py -c configs/rec/PP-OCRv3/ch_PP-OCRv3_rec_distillation.yml \
 
 可以看到，相较于原始PP-OCRv3的识别精度62.99%，使用合成数据+真实数据finetune后，识别精度能提升24%。
 
-如需获取已训练模型，请扫码填写问卷，加入PaddleOCR官方交流群获取全部OCR垂类模型下载链接、《动手学OCR》电子书等全套OCR学习资料🎁
-<div align="left">
-<img src="https://ai-studio-static-online.cdn.bcebos.com/dd721099bd50478f9d5fb13d8dd00fad69c22d6848244fd3a1d3980d7fefc63e"  width = "150" height = "150" />
-</div>
-将下载或训练完成的模型放置在对应目录下即可完成模型推理。
+如需获取已训练模型，可以同样扫描上方二维码下载，将下载或训练完成的模型放置在对应目录下即可完成模型推理。
 
 模型的推理部署方法可以参考repo文档： https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.5/deploy/README_ch.md

From 9da6038214ce90094ebb09c88f84fe73f4845092 Mon Sep 17 00:00:00 2001
From: Topdu <784990967@qq.com>
Date: Wed, 29 Jun 2022 12:08:45 +0000
Subject: [PATCH 093/134] fix BaseAug

---
 doc/doc_ch/algorithm_overview.md    | 4 ++--
 doc/doc_en/algorithm_overview_en.md | 4 ++--
 ppocr/data/imaug/__init__.py        | 6 +++---
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index eb81e4cd6d..fbf0129ddd 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -86,8 +86,8 @@
 |SAR|Resnet31| 87.20% | rec_r31_sar | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
-|ABINet|Resnet45| 90.75% | rec_r45_abinet_en | [训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet | [训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
 <a name="2"></a>
 
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index 28aca7c0d1..a3fa7a2d91 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -85,8 +85,8 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 |SAR|Resnet31| 87.20% | rec_r31_sar | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
 |SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
 |SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
-|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce_en | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
-|ABINet|Resnet45| 90.75% | rec_r45_abinet_en | [trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet | [trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
 
 <a name="2"></a>
 
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
index 63dfda91f8..c958c7ae4e 100644
--- a/ppocr/data/imaug/__init__.py
+++ b/ppocr/data/imaug/__init__.py
@@ -23,9 +23,9 @@ from .random_crop_data import EastRandomCropData, RandomCropImgMask
 from .make_pse_gt import MakePseGt
 
 
-from .rec_img_aug import RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
-    SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, \
-    ABINetRecResizeImg, SVTRRecResizeImg, ABINetRecAug
+from .rec_img_aug import BaseDataAugmentation, RecAug, RecConAug, RecResizeImg, ClsResizeImg, \
+                         SRNRecResizeImg, GrayRecResizeImg, SARRecResizeImg, PRENResizeImg, \
+                         ABINetRecResizeImg, SVTRRecResizeImg, ABINetRecAug
 from .ssl_img_aug import SSLRotateResize
 from .randaugment import RandAugment
 from .copy_paste import CopyPaste

From d915416da9cca4b7af8642cdc7279f841343e9cc Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Thu, 30 Jun 2022 04:25:26 +0000
Subject: [PATCH 094/134] add db++

---
 tools/program.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tools/program.py b/tools/program.py
index 620c61f093..d86be5badb 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -574,7 +574,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR', 'DB++'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
+        'ViTSTR', 'ABINet', 'DB++'
     ]
 
     if use_xpu:

From 7a9c8af11d38bb246a9be048219d8574abb3b0de Mon Sep 17 00:00:00 2001
From: Liliya WU <liliya.wu@arm.com>
Date: Thu, 30 Jun 2022 13:39:21 +0800
Subject: [PATCH 095/134] update scripts to run in cloud-based avh

---
 deploy/avh/README.md        | 54 +++++++++++++++----------
 deploy/avh/configure_avh.sh | 79 +++++++++++++++++++++++++++++++++++++
 deploy/avh/run_demo.sh      | 53 ++++++++++++++++++++-----
 3 files changed, 155 insertions(+), 31 deletions(-)
 create mode 100755 deploy/avh/configure_avh.sh

diff --git a/deploy/avh/README.md b/deploy/avh/README.md
index 922fb4da92..a64bfd3ad6 100644
--- a/deploy/avh/README.md
+++ b/deploy/avh/README.md
@@ -15,20 +15,22 @@
 <!--- specific language governing permissions and limitations -->
 <!--- under the License. -->
 
-Running PaddleOCR text recognition model via TVM on bare metal Arm(R) Cortex(R)-M55 CPU and CMSIS-NN
-===============================================================
+Running PaddleOCR text recognition model on bare metal Arm(R) Cortex(R)-M55 CPU using Arm Virtual Hardware
+======================================================================
 
-This folder contains an example of how to use TVM to run a PaddleOCR model
-on bare metal Cortex(R)-M55 CPU and CMSIS-NN.
+This folder contains an example of how to run a PaddleOCR model on bare metal [Cortex(R)-M55 CPU](https://www.arm.com/products/silicon-ip-cpu/cortex-m/cortex-m55) using [Arm Virtual Hardware](https://www.arm.com/products/development-tools/simulation/virtual-hardware).
 
-Prerequisites
+
+Running environment and prerequisites
 -------------
-If the demo is run in the ci_cpu Docker container provided with TVM, then the following
-software will already be installed.
+Case 1: If the demo is run in Arm Virtual Hardware Amazon Machine Image(AMI) instance hosted by [AWS](https://aws.amazon.com/marketplace/pp/prodview-urbpq7yo5va7g?sr=0-1&ref_=beagle&applicationId=AWSMPContessa)/[AWS China](https://awsmarketplace.amazonaws.cn/marketplace/pp/prodview-2y7nefntbmybu), the following software will be installed through [configure_avh.sh](./configure_avh.sh) script. It will install automatically when you run the application through [run_demo.sh](./run_demo.sh) script.
+You can refer to this [guide](https://arm-software.github.io/AVH/main/examples/html/MicroSpeech.html#amilaunch) to launch an Arm Virtual Hardware AMI instance.
 
-If the demo is not run in the ci_cpu Docker container, then you will need the following:
+Case 2: If the demo is run in the [ci_cpu Docker container](https://github.com/apache/tvm/blob/main/docker/Dockerfile.ci_cpu) provided with [TVM](https://github.com/apache/tvm), then the following software will already be installed.
+
+Case 3: If the demo is not run in the ci_cpu Docker container, then you will need the following:
 - Software required to build and run the demo (These can all be installed by running 
-  https://github.com/apache/tvm/blob/main/docker/install/ubuntu_install_ethosu_driver_stack.sh .)
+  tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh.)
   - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
   - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
   - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
@@ -40,20 +42,23 @@ If the demo is not run in the ci_cpu Docker container, then you will need the fo
     pip install -r ./requirements.txt
     ```
 
-You will also need TVM which can either be:
-  - Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))
-    - When building from source, the following need to be set in config.cmake:
-      - set(USE_CMSISNN ON)
-      - set(USE_MICRO ON)
-      - set(USE_LLVM ON)
-  - Installed from TLCPack nightly(see [TLCPack](https://tlcpack.ai/))
-
+In case2 and case3:  
+    
 You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.
 For example if you've installed these in ```/opt/arm``` , then you would do the following:
 ```bash
 export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
 ```
 
+You will also need TVM which can either be:
+  - Installed from TLCPack(see [TLCPack](https://tlcpack.ai/))
+  - Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    - When building from source, the following need to be set in config.cmake:
+      - set(USE_CMSISNN ON)
+      - set(USE_MICRO ON)
+      - set(USE_LLVM ON)
+
+
 Running the demo application
 ----------------------------
 Type the following command to run the bare metal text recognition application ([src/demo_bare_metal.c](./src/demo_bare_metal.c)):
@@ -62,6 +67,12 @@ Type the following command to run the bare metal text recognition application ([
 ./run_demo.sh
 ```
 
+If you are not able to use Arm Virtual Hardware Amazon Machine Image(AMI) instance hosted by AWS/AWS China, specify argument --enable_FVP to 1 to make the application run on local Fixed Virtual Platforms (FVPs) executables.
+
+```bash
+./run_demo.sh --enable_FVP 1
+```
+
 If the Ethos(TM)-U platform and/or CMSIS have not been installed in /opt/arm/ethosu then
 the locations for these can be specified as arguments to run_demo.sh, for example:
 
@@ -70,13 +81,14 @@ the locations for these can be specified as arguments to run_demo.sh, for exampl
 --ethosu_platform_path /home/tvm-user/ethosu/core_platform
 ```
 
-This will:
+With [run_demo.sh](./run_demo.sh) to run the demo application, it will:
+- Set up running environment by installing the required prerequisites automatically if running in Arm Virtual Hardware Amazon AMI instance(not specify --enable_FVP to 1)
 - Download a PaddleOCR text recognition model
 - Use tvmc to compile the text recognition model for Cortex(R)-M55 CPU and CMSIS-NN
 - Create a C header file inputs.c containing the image data as a C array
 - Create a C header file outputs.c containing a C array where the output of inference will be stored
 - Build the demo application
-- Run the demo application on a Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software
+- Run the demo application on a Arm Virtual Hardware based on Arm(R) Corstone(TM)-300 software
 - The application will report the text on the image and the corresponding score.
 
 Using your own image
@@ -92,9 +104,9 @@ python3 ./convert_image.py path/to/image
 
 Model description
 -----------------
-In this demo, the model we use is an English recognition model based on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md). PP-OCRv3 is the third version of the PP-OCR series model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). This series of models has the following features:
+The example is built on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md) English recognition model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). PP-OCRv3 is the third version of the PP-OCR series model. This series of models has the following features:
   - PP-OCRv3: ultra-lightweight OCR system: detection (3.6M) + direction classifier (1.4M) + recognition (12M) = 17.0M
   - Support more than 80 kinds of multi-language recognition models, including English, Chinese, French, German, Arabic, Korean, Japanese and so on. For details
   - Support vertical text recognition, and long text recognition
 
-The text recognition model in PP-OCRv3 supports more than 80 languages. In the process of model development, since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current model.
\ No newline at end of file
+The text recognition model in PP-OCRv3 supports more than 80 languages. In the process of model development, since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current model.
diff --git a/deploy/avh/configure_avh.sh b/deploy/avh/configure_avh.sh
new file mode 100755
index 0000000000..8042fd81d2
--- /dev/null
+++ b/deploy/avh/configure_avh.sh
@@ -0,0 +1,79 @@
+#!/bin/bash
+# Copyright (c) 2022 Arm Limited and Contributors. All rights reserved.
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+set -e
+set -u
+set -o pipefail
+
+# Show usage
+function show_usage() {
+    cat <<EOF
+Usage: Set up running environment by installing the required prerequisites.
+-h, --help
+    Display this help message.
+EOF
+}
+
+if [ "$#" -eq 1 ] && [ "$1" == "--help" -o "$1" == "-h" ]; then
+    show_usage
+    exit 0
+elif [ "$#" -ge 1 ]; then
+        show_usage
+        exit 1
+fi
+
+echo -e "\e[36mStart setting up running environment\e[0m"
+
+# Install CMSIS
+echo -e "\e[36mStart installing CMSIS\e[0m"
+CMSIS_PATH="/opt/arm/ethosu/cmsis" 
+mkdir -p "${CMSIS_PATH}"
+
+CMSIS_SHA="977abe9849781a2e788b02282986480ff4e25ea6"
+CMSIS_SHASUM="86c88d9341439fbb78664f11f3f25bc9fda3cd7de89359324019a4d87d169939eea85b7fdbfa6ad03aa428c6b515ef2f8cd52299ce1959a5444d4ac305f934cc"
+CMSIS_URL="http://github.com/ARM-software/CMSIS_5/archive/${CMSIS_SHA}.tar.gz"
+DOWNLOAD_PATH="/tmp/${CMSIS_SHA}.tar.gz"
+
+wget ${CMSIS_URL} -O "${DOWNLOAD_PATH}"
+echo "$CMSIS_SHASUM" ${DOWNLOAD_PATH} | sha512sum -c
+tar -xf "${DOWNLOAD_PATH}" -C "${CMSIS_PATH}" --strip-components=1
+touch "${CMSIS_PATH}"/"${CMSIS_SHA}".sha
+echo -e "\e[36mCMSIS Installation SUCCESS\e[0m"
+
+# Install Arm(R) Ethos(TM)-U NPU driver stack
+echo -e "\e[36mStart installing Arm(R) Ethos(TM)-U NPU driver stack\e[0m"
+git clone "https://review.mlplatform.org/ml/ethos-u/ethos-u-core-platform" /opt/arm/ethosu/core_platform 
+cd /opt/arm/ethosu/core_platform 
+git checkout tags/"21.11"
+echo -e "\e[36mArm(R) Ethos(TM)-U Core Platform Installation SUCCESS\e[0m"
+ 
+# Install Arm(R) GNU Toolchain
+echo -e "\e[36mStart installing Arm(R) GNU Toolchain\e[0m"
+mkdir -p /opt/arm/gcc-arm-none-eabi
+export gcc_arm_url='https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2?revision=ca0cbf9c-9de2-491c-ac48-898b5bbc0443&la=en&hash=68760A8AE66026BCF99F05AC017A6A50C6FD832A'
+curl --retry 64 -sSL ${gcc_arm_url} | tar -C /opt/arm/gcc-arm-none-eabi --strip-components=1 -jx 
+export PATH=/opt/arm/gcc-arm-none-eabi/bin:$PATH 
+arm-none-eabi-gcc --version
+arm-none-eabi-g++ --version
+echo -e "\e[36mArm(R) Arm(R) GNU Toolchain Installation SUCCESS\e[0m"
+
+# Install TVM from TLCPack
+echo -e "\e[36mStart installing TVM\e[0m"
+pip install tlcpack-nightly -f https://tlcpack.ai/wheels
+echo -e "\e[36mTVM Installation SUCCESS\e[0m"
\ No newline at end of file
diff --git a/deploy/avh/run_demo.sh b/deploy/avh/run_demo.sh
index dd8e8f163d..01f950c13d 100755
--- a/deploy/avh/run_demo.sh
+++ b/deploy/avh/run_demo.sh
@@ -1,4 +1,5 @@
 #!/bin/bash
+# Copyright (c) 2022 Arm Limited and Contributors. All rights reserved.
 # Licensed to the Apache Software Foundation (ASF) under one
 # or more contributor license agreements.  See the NOTICE file
 # distributed with this work for additional information
@@ -24,18 +25,28 @@ function show_usage() {
     cat <<EOF
 Usage: run_demo.sh
 -h, --help
-    Display this help message.
+	Display this help message.
 --cmsis_path CMSIS_PATH
-    Set path to CMSIS.
+	Set path to CMSIS.
 --ethosu_platform_path ETHOSU_PLATFORM_PATH
-    Set path to Arm(R) Ethos(TM)-U core platform.
+	Set path to Arm(R) Ethos(TM)-U core platform.
 --fvp_path FVP_PATH
-   Set path to FVP.
+	Set path to FVP.
 --cmake_path
-   Set path to cmake.
+	Set path to cmake.
+--enable_FVP
+	Set 1 to run application on local Fixed Virtual Platforms (FVPs) executables.
 EOF
 }
 
+# Configure environment variables
+FVP_enable=0
+export PATH=/opt/arm/gcc-arm-none-eabi/bin:$PATH
+
+# Install python libraries
+echo -e "\e[36mInstall python libraries\e[0m"
+sudo pip install -r ./requirements.txt
+
 # Parse arguments
 while (( $# )); do
     case "$1" in
@@ -91,6 +102,18 @@ while (( $# )); do
                 exit 1
             fi
             ;;
+            
+        --enable_FVP)
+            if [ $# -gt 1 ] && [ "$2" == "1" -o "$2" == "0" ];
+            then
+                FVP_enable="$2"
+                shift 2
+            else
+                echo 'ERROR: --enable_FVP requires a right argument 1 or 0' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
 
         -*|--*)
             echo "Error: Unknown flag: $1" >&2
@@ -100,17 +123,27 @@ while (( $# )); do
     esac
 done
 
+# Choose running environment: cloud(default) or local environment
+Platform="VHT_Corstone_SSE-300_Ethos-U55"
+if [ $FVP_enable == "1" ]; then
+	Platform="FVP_Corstone_SSE-300_Ethos-U55"
+	echo -e "\e[36mRun application on local Fixed Virtual Platforms (FVPs)\e[0m"
+else
+	if [ ! -d "/opt/arm/" ]; then
+		sudo ./configure_avh.sh
+	fi
+fi
 
 # Directories
 script_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
 
-
 # Make build directory
-rm -rf build
 make cleanall
 mkdir -p build
 cd build
 
+# Get PaddlePaddle inference model
+echo -e "\e[36mDownload PaddlePaddle inference model\e[0m"
 wget https://paddleocr.bj.bcebos.com/tvm/ocr_en.tar
 tar -xf ocr_en.tar
 
@@ -144,9 +177,9 @@ cd ${script_dir}
 echo ${script_dir}
 make
 
-# Run demo executable on the FVP
-FVP_Corstone_SSE-300_Ethos-U55 -C cpu0.CFGDTCMSZ=15 \
+# Run demo executable on the AVH
+$Platform -C cpu0.CFGDTCMSZ=15 \
 -C cpu0.CFGITCMSZ=15 -C mps3_board.uart0.out_file=\"-\" -C mps3_board.uart0.shutdown_tag=\"EXITTHESIM\" \
 -C mps3_board.visualisation.disable-visualisation=1 -C mps3_board.telnetterminal0.start_telnet=0 \
 -C mps3_board.telnetterminal1.start_telnet=0 -C mps3_board.telnetterminal2.start_telnet=0 -C mps3_board.telnetterminal5.start_telnet=0 \
-./build/demo
\ No newline at end of file
+./build/demo --stat

From e13ec733a628006cf7295652600bbf8c1604755f Mon Sep 17 00:00:00 2001
From: littletomatodonkey <dazhiningsibuqu@163.com>
Date: Thu, 30 Jun 2022 15:23:31 +0800
Subject: [PATCH 096/134] unify kie and ser for vqa data format (#6704)

* unify kie and ser for vqa data format

* fix config and label ops

* fix doc

* add distort bbox
---
 configs/kie/kie_unet_sdmgr.yml           |  4 +--
 configs/vqa/ser/layoutlm.yml             |  6 ++--
 configs/vqa/ser/layoutlmv2.yml           |  6 ++--
 configs/vqa/ser/layoutxlm.yml            |  6 ++--
 ppocr/data/imaug/label_ops.py            | 46 +++++++++++++++---------
 ppocr/data/imaug/vqa/__init__.py         |  7 +++-
 ppocr/data/imaug/vqa/augment.py          | 37 +++++++++++++++++++
 ppocr/utils/utility.py                   | 25 ++++++-------
 ppocr/utils/visual.py                    | 28 +++++++++++----
 ppstructure/docs/kie.md                  |  2 +-
 ppstructure/docs/kie_en.md               |  2 +-
 ppstructure/vqa/README.md                |  4 +--
 ppstructure/vqa/README_ch.md             |  4 +--
 ppstructure/vqa/labels/labels_ser.txt    |  3 --
 ppstructure/vqa/tools/trans_xfun_data.py | 18 ++++------
 tools/infer_kie.py                       | 14 ++++----
 tools/infer_vqa_token_ser.py             | 30 ++++++++++++----
 tools/program.py                         |  4 +--
 18 files changed, 161 insertions(+), 85 deletions(-)
 create mode 100644 ppocr/data/imaug/vqa/augment.py
 delete mode 100644 ppstructure/vqa/labels/labels_ser.txt

diff --git a/configs/kie/kie_unet_sdmgr.yml b/configs/kie/kie_unet_sdmgr.yml
index a6968aaa3a..da2e4fda50 100644
--- a/configs/kie/kie_unet_sdmgr.yml
+++ b/configs/kie/kie_unet_sdmgr.yml
@@ -17,7 +17,7 @@ Global:
   checkpoints:
   save_inference_dir:
   use_visualdl: False
-  class_path: ./train_data/wildreceipt/class_list.txt
+  class_path: &class_path ./train_data/wildreceipt/class_list.txt
   infer_img: ./train_data/wildreceipt/1.txt
   save_res_path: ./output/sdmgr_kie/predicts_kie.txt
   img_scale: [ 1024, 512 ]
@@ -72,6 +72,7 @@ Train:
           order: 'hwc'
       - KieLabelEncode: # Class handling label
           character_dict_path: ./train_data/wildreceipt/dict.txt
+          class_path: *class_path
       - KieResize:
       - ToCHWImage:
       - KeepKeys:
@@ -88,7 +89,6 @@ Eval:
     data_dir: ./train_data/wildreceipt
     label_file_list:
       - ./train_data/wildreceipt/wildreceipt_test.txt
-      # - /paddle/data/PaddleOCR/train_data/wildreceipt/1.txt
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
diff --git a/configs/vqa/ser/layoutlm.yml b/configs/vqa/ser/layoutlm.yml
index 87131170c9..47ab093e1f 100644
--- a/configs/vqa/ser/layoutlm.yml
+++ b/configs/vqa/ser/layoutlm.yml
@@ -43,7 +43,7 @@ Optimizer:
     
 PostProcess:
   name: VQASerTokenLayoutLMPostProcess
-  class_path: &class_path ppstructure/vqa/labels/labels_ser.txt
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
 
 Metric:
   name: VQASerTokenMetric
@@ -54,7 +54,7 @@ Train:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_train/image
     label_file_list: 
-      - train_data/XFUND/zh_train/xfun_normalize_train.json
+      - train_data/XFUND/zh_train/train.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
@@ -89,7 +89,7 @@ Eval:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_val/image
     label_file_list:
-      - train_data/XFUND/zh_val/xfun_normalize_val.json
+      - train_data/XFUND/zh_val/val.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
diff --git a/configs/vqa/ser/layoutlmv2.yml b/configs/vqa/ser/layoutlmv2.yml
index 33406252b3..d6a9c03e5e 100644
--- a/configs/vqa/ser/layoutlmv2.yml
+++ b/configs/vqa/ser/layoutlmv2.yml
@@ -44,7 +44,7 @@ Optimizer:
     
 PostProcess:
   name: VQASerTokenLayoutLMPostProcess
-  class_path: &class_path ppstructure/vqa/labels/labels_ser.txt
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
 
 Metric:
   name: VQASerTokenMetric
@@ -55,7 +55,7 @@ Train:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_train/image
     label_file_list: 
-      - train_data/XFUND/zh_train/xfun_normalize_train.json
+      - train_data/XFUND/zh_train/train.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
@@ -90,7 +90,7 @@ Eval:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_val/image
     label_file_list:
-      - train_data/XFUND/zh_val/xfun_normalize_val.json
+      - train_data/XFUND/zh_val/val.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
diff --git a/configs/vqa/ser/layoutxlm.yml b/configs/vqa/ser/layoutxlm.yml
index eb1cca5a21..3686989ccf 100644
--- a/configs/vqa/ser/layoutxlm.yml
+++ b/configs/vqa/ser/layoutxlm.yml
@@ -11,7 +11,7 @@ Global:
   save_inference_dir:
   use_visualdl: False
   seed: 2022
-  infer_img: doc/vqa/input/zh_val_42.jpg
+  infer_img: ppstructure/docs/vqa/input/zh_val_42.jpg
   save_res_path: ./output/ser
 
 Architecture:
@@ -54,7 +54,7 @@ Train:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_train/image
     label_file_list: 
-      - train_data/XFUND/zh_train/xfun_normalize_train.json
+      - train_data/XFUND/zh_train/train.json
     ratio_list: [ 1.0 ]
     transforms:
       - DecodeImage: # load image
@@ -90,7 +90,7 @@ Eval:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_val/image
     label_file_list:
-      - train_data/XFUND/zh_val/xfun_normalize_val.json
+      - train_data/XFUND/zh_val/val.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 312d6dc9ad..c95b326224 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -259,15 +259,26 @@ class E2ELabelEncodeTrain(object):
 
 
 class KieLabelEncode(object):
-    def __init__(self, character_dict_path, norm=10, directed=False, **kwargs):
+    def __init__(self,
+                 character_dict_path,
+                 class_path,
+                 norm=10,
+                 directed=False,
+                 **kwargs):
         super(KieLabelEncode, self).__init__()
         self.dict = dict({'': 0})
+        self.label2classid_map = dict()
         with open(character_dict_path, 'r', encoding='utf-8') as fr:
             idx = 1
             for line in fr:
                 char = line.strip()
                 self.dict[char] = idx
                 idx += 1
+        with open(class_path, "r") as fin:
+            lines = fin.readlines()
+            for idx, line in enumerate(lines):
+                line = line.strip("\n")
+                self.label2classid_map[line] = idx
         self.norm = norm
         self.directed = directed
 
@@ -408,7 +419,7 @@ class KieLabelEncode(object):
             text_ind = [self.dict[c] for c in text if c in self.dict]
             text_inds.append(text_ind)
             if 'label' in ann.keys():
-                labels.append(ann['label'])
+                labels.append(self.label2classid_map[ann['label']])
             elif 'key_cls' in ann.keys():
                 labels.append(ann['key_cls'])
             else:
@@ -876,15 +887,16 @@ class VQATokenLabelEncode(object):
         for info in ocr_info:
             if train_re:
                 # for re
-                if len(info["text"]) == 0:
+                if len(info["transcription"]) == 0:
                     empty_entity.add(info["id"])
                     continue
                 id2label[info["id"]] = info["label"]
                 relations.extend([tuple(sorted(l)) for l in info["linking"]])
             # smooth_box
+            info["bbox"] = self.trans_poly_to_bbox(info["points"])
             bbox = self._smooth_box(info["bbox"], height, width)
 
-            text = info["text"]
+            text = info["transcription"]
             encode_res = self.tokenizer.encode(
                 text, pad_to_max_seq_len=False, return_attention_mask=True)
 
@@ -900,7 +912,7 @@ class VQATokenLabelEncode(object):
                 label = info['label']
                 gt_label = self._parse_label(label, encode_res)
 
-            # construct entities for re
+# construct entities for re
             if train_re:
                 if gt_label[0] != self.label2id_map["O"]:
                     entity_id_to_index_map[info["id"]] = len(entities)
@@ -944,29 +956,29 @@ class VQATokenLabelEncode(object):
             data['entity_id_to_index_map'] = entity_id_to_index_map
         return data
 
-    def _load_ocr_info(self, data):
-        def trans_poly_to_bbox(poly):
-            x1 = np.min([p[0] for p in poly])
-            x2 = np.max([p[0] for p in poly])
-            y1 = np.min([p[1] for p in poly])
-            y2 = np.max([p[1] for p in poly])
-            return [x1, y1, x2, y2]
+    def trans_poly_to_bbox(self, poly):
+        x1 = np.min([p[0] for p in poly])
+        x2 = np.max([p[0] for p in poly])
+        y1 = np.min([p[1] for p in poly])
+        y2 = np.max([p[1] for p in poly])
+        return [x1, y1, x2, y2]
 
+    def _load_ocr_info(self, data):
         if self.infer_mode:
             ocr_result = self.ocr_engine.ocr(data['image'], cls=False)
             ocr_info = []
             for res in ocr_result:
                 ocr_info.append({
-                    "text": res[1][0],
-                    "bbox": trans_poly_to_bbox(res[0]),
-                    "poly": res[0],
+                    "transcription": res[1][0],
+                    "bbox": self.trans_poly_to_bbox(res[0]),
+                    "points": res[0],
                 })
             return ocr_info
         else:
             info = data['label']
             # read text info
             info_dict = json.loads(info)
-            return info_dict["ocr_info"]
+            return info_dict
 
     def _smooth_box(self, bbox, height, width):
         bbox[0] = int(bbox[0] * 1000.0 / width)
@@ -977,7 +989,7 @@ class VQATokenLabelEncode(object):
 
     def _parse_label(self, label, encode_res):
         gt_label = []
-        if label.lower() == "other":
+        if label.lower() in ["other", "others", "ignore"]:
             gt_label.extend([0] * len(encode_res["input_ids"]))
         else:
             gt_label.append(self.label2id_map[("b-" + label).upper()])
diff --git a/ppocr/data/imaug/vqa/__init__.py b/ppocr/data/imaug/vqa/__init__.py
index a5025e7985..bde1751155 100644
--- a/ppocr/data/imaug/vqa/__init__.py
+++ b/ppocr/data/imaug/vqa/__init__.py
@@ -13,7 +13,12 @@
 # limitations under the License.
 
 from .token import VQATokenPad, VQASerTokenChunk, VQAReTokenChunk, VQAReTokenRelation
+from .augment import DistortBBox
 
 __all__ = [
-    'VQATokenPad', 'VQASerTokenChunk', 'VQAReTokenChunk', 'VQAReTokenRelation'
+    'VQATokenPad',
+    'VQASerTokenChunk',
+    'VQAReTokenChunk',
+    'VQAReTokenRelation',
+    'DistortBBox',
 ]
diff --git a/ppocr/data/imaug/vqa/augment.py b/ppocr/data/imaug/vqa/augment.py
new file mode 100644
index 0000000000..fcdc9685e9
--- /dev/null
+++ b/ppocr/data/imaug/vqa/augment.py
@@ -0,0 +1,37 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import numpy as np
+import random
+
+
+class DistortBBox:
+    def __init__(self, prob=0.5, max_scale=1, **kwargs):
+        """Random distort bbox
+        """
+        self.prob = prob
+        self.max_scale = max_scale
+
+    def __call__(self, data):
+        if random.random() > self.prob:
+            return data
+        bbox = np.array(data['bbox'])
+        rnd_scale = (np.random.rand(*bbox.shape) - 0.5) * 2 * self.max_scale
+        bbox = np.round(bbox + rnd_scale).astype(bbox.dtype)
+        data['bbox'] = np.clip(data['bbox'], 0, 1000)
+        data['bbox'] = bbox.tolist()
+        sys.stdout.flush()
+        return data
diff --git a/ppocr/utils/utility.py b/ppocr/utils/utility.py
index 4a25ff8b2f..b881fcab20 100755
--- a/ppocr/utils/utility.py
+++ b/ppocr/utils/utility.py
@@ -91,18 +91,19 @@ def check_and_read_gif(img_path):
 def load_vqa_bio_label_maps(label_map_path):
     with open(label_map_path, "r", encoding='utf-8') as fin:
         lines = fin.readlines()
-    lines = [line.strip() for line in lines]
-    if "O" not in lines:
-        lines.insert(0, "O")
-    labels = []
-    for line in lines:
-        if line == "O":
-            labels.append("O")
-        else:
-            labels.append("B-" + line)
-            labels.append("I-" + line)
-    label2id_map = {label: idx for idx, label in enumerate(labels)}
-    id2label_map = {idx: label for idx, label in enumerate(labels)}
+    old_lines = [line.strip() for line in lines]
+    lines = ["O"]
+    for line in old_lines:
+        # "O" has already been in lines
+        if line.upper() in ["OTHER", "OTHERS", "IGNORE"]:
+            continue
+        lines.append(line)
+    labels = ["O"]
+    for line in lines[1:]:
+        labels.append("B-" + line)
+        labels.append("I-" + line)
+    label2id_map = {label.upper(): idx for idx, label in enumerate(labels)}
+    id2label_map = {idx: label.upper() for idx, label in enumerate(labels)}
     return label2id_map, id2label_map
 
 
diff --git a/ppocr/utils/visual.py b/ppocr/utils/visual.py
index 7a8c1674a7..235eb572a3 100644
--- a/ppocr/utils/visual.py
+++ b/ppocr/utils/visual.py
@@ -19,7 +19,7 @@ from PIL import Image, ImageDraw, ImageFont
 def draw_ser_results(image,
                      ocr_results,
                      font_path="doc/fonts/simfang.ttf",
-                     font_size=18):
+                     font_size=14):
     np.random.seed(2021)
     color = (np.random.permutation(range(255)),
              np.random.permutation(range(255)),
@@ -40,9 +40,15 @@ def draw_ser_results(image,
         if ocr_info["pred_id"] not in color_map:
             continue
         color = color_map[ocr_info["pred_id"]]
-        text = "{}: {}".format(ocr_info["pred"], ocr_info["text"])
+        text = "{}: {}".format(ocr_info["pred"], ocr_info["transcription"])
 
-        draw_box_txt(ocr_info["bbox"], text, draw, font, font_size, color)
+        if "bbox" in ocr_info:
+            # draw with ocr engine
+            bbox = ocr_info["bbox"]
+        else:
+            # draw with ocr groundtruth
+            bbox = trans_poly_to_bbox(ocr_info["points"])
+        draw_box_txt(bbox, text, draw, font, font_size, color)
 
     img_new = Image.blend(image, img_new, 0.5)
     return np.array(img_new)
@@ -62,6 +68,14 @@ def draw_box_txt(bbox, text, draw, font, font_size, color):
     draw.text((bbox[0][0] + 1, start_y), text, fill=(255, 255, 255), font=font)
 
 
+def trans_poly_to_bbox(poly):
+    x1 = np.min([p[0] for p in poly])
+    x2 = np.max([p[0] for p in poly])
+    y1 = np.min([p[1] for p in poly])
+    y2 = np.max([p[1] for p in poly])
+    return [x1, y1, x2, y2]
+
+
 def draw_re_results(image,
                     result,
                     font_path="doc/fonts/simfang.ttf",
@@ -80,10 +94,10 @@ def draw_re_results(image,
     color_line = (0, 255, 0)
 
     for ocr_info_head, ocr_info_tail in result:
-        draw_box_txt(ocr_info_head["bbox"], ocr_info_head["text"], draw, font,
-                     font_size, color_head)
-        draw_box_txt(ocr_info_tail["bbox"], ocr_info_tail["text"], draw, font,
-                     font_size, color_tail)
+        draw_box_txt(ocr_info_head["bbox"], ocr_info_head["transcription"],
+                     draw, font, font_size, color_head)
+        draw_box_txt(ocr_info_tail["bbox"], ocr_info_tail["transcription"],
+                     draw, font, font_size, color_tail)
 
         center_head = (
             (ocr_info_head['bbox'][0] + ocr_info_head['bbox'][2]) // 2,
diff --git a/ppstructure/docs/kie.md b/ppstructure/docs/kie.md
index 35498b3347..315dd9f7ba 100644
--- a/ppstructure/docs/kie.md
+++ b/ppstructure/docs/kie.md
@@ -16,7 +16,7 @@ SDMGR是一个关键信息提取算法，将每个检测到的文本区域分类
 训练和测试的数据采用wildreceipt数据集，通过如下指令下载数据集：
 
 ```
-wget https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/wildreceipt.tar && tar xf wildreceipt.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/wildreceipt.tar && tar xf wildreceipt.tar
 ```
 
 执行预测：
diff --git a/ppstructure/docs/kie_en.md b/ppstructure/docs/kie_en.md
index 1fe38b0b39..7b3752223d 100644
--- a/ppstructure/docs/kie_en.md
+++ b/ppstructure/docs/kie_en.md
@@ -15,7 +15,7 @@ This section provides a tutorial example on how to quickly use, train, and evalu
 [Wildreceipt dataset](https://paperswithcode.com/dataset/wildreceipt) is used for this tutorial. It contains 1765 photos, with 25 classes, and 50000 text boxes, which can be downloaded by wget:
 
 ```shell
-wget https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/wildreceipt.tar && tar xf wildreceipt.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/wildreceipt.tar && tar xf wildreceipt.tar
 ```
 
 Download the pretrained model and predict the result:
diff --git a/ppstructure/vqa/README.md b/ppstructure/vqa/README.md
index e3a10671dd..711ffa3138 100644
--- a/ppstructure/vqa/README.md
+++ b/ppstructure/vqa/README.md
@@ -125,13 +125,13 @@ If you want to experience the prediction process directly, you can download the
 
 * Download the processed dataset
 
-The download address of the processed XFUND Chinese dataset: [https://paddleocr.bj.bcebos.com/dataset/XFUND.tar](https://paddleocr.bj.bcebos.com/dataset/XFUND.tar).
+The download address of the processed XFUND Chinese dataset: [link](https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar).
 
 
 Download and unzip the dataset, and place the dataset in the current directory after unzipping.
 
 ```shell
-wget https://paddleocr.bj.bcebos.com/dataset/XFUND.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar
 ````
 
 * Convert the dataset
diff --git a/ppstructure/vqa/README_ch.md b/ppstructure/vqa/README_ch.md
index b677dc07bc..297ba64f82 100644
--- a/ppstructure/vqa/README_ch.md
+++ b/ppstructure/vqa/README_ch.md
@@ -122,13 +122,13 @@ python3 -m pip install -r ppstructure/vqa/requirements.txt
 
 * 下载处理好的数据集
 
-处理好的XFUND中文数据集下载地址：[https://paddleocr.bj.bcebos.com/dataset/XFUND.tar](https://paddleocr.bj.bcebos.com/dataset/XFUND.tar)。
+处理好的XFUND中文数据集下载地址：[链接](https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar)。
 
 
 下载并解压该数据集，解压后将数据集放置在当前目录下。
 
 ```shell
-wget https://paddleocr.bj.bcebos.com/dataset/XFUND.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar
 ```
 
 * 转换数据集
diff --git a/ppstructure/vqa/labels/labels_ser.txt b/ppstructure/vqa/labels/labels_ser.txt
deleted file mode 100644
index 508e481124..0000000000
--- a/ppstructure/vqa/labels/labels_ser.txt
+++ /dev/null
@@ -1,3 +0,0 @@
-QUESTION
-ANSWER
-HEADER
diff --git a/ppstructure/vqa/tools/trans_xfun_data.py b/ppstructure/vqa/tools/trans_xfun_data.py
index 93ec98163c..11d221bea4 100644
--- a/ppstructure/vqa/tools/trans_xfun_data.py
+++ b/ppstructure/vqa/tools/trans_xfun_data.py
@@ -21,26 +21,22 @@ def transfer_xfun_data(json_path=None, output_file=None):
 
     json_info = json.loads(lines[0])
     documents = json_info["documents"]
-    label_info = {}
     with open(output_file, "w", encoding='utf-8') as fout:
         for idx, document in enumerate(documents):
+            label_info = []
             img_info = document["img"]
             document = document["document"]
             image_path = img_info["fname"]
 
-            label_info["height"] = img_info["height"]
-            label_info["width"] = img_info["width"]
-
-            label_info["ocr_info"] = []
-
             for doc in document:
-                label_info["ocr_info"].append({
-                    "text": doc["text"],
+                x1, y1, x2, y2 = doc["box"]
+                points = [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+                label_info.append({
+                    "transcription": doc["text"],
                     "label": doc["label"],
-                    "bbox": doc["box"],
+                    "points": points,
                     "id": doc["id"],
-                    "linking": doc["linking"],
-                    "words": doc["words"]
+                    "linking": doc["linking"]
                 })
 
             fout.write(image_path + "\t" + json.dumps(
diff --git a/tools/infer_kie.py b/tools/infer_kie.py
index 0cb0b8702c..346e2e0aee 100755
--- a/tools/infer_kie.py
+++ b/tools/infer_kie.py
@@ -39,13 +39,12 @@ import time
 
 
 def read_class_list(filepath):
-    dict = {}
+    ret = {}
     with open(filepath, "r") as f:
         lines = f.readlines()
-        for line in lines:
-            key, value = line.split(" ")
-            dict[key] = value.rstrip()
-    return dict
+        for idx, line in enumerate(lines):
+            ret[idx] = line.strip("\n")
+    return ret
 
 
 def draw_kie_result(batch, node, idx_to_cls, count):
@@ -71,7 +70,7 @@ def draw_kie_result(batch, node, idx_to_cls, count):
         x_min = int(min([point[0] for point in new_box]))
         y_min = int(min([point[1] for point in new_box]))
 
-        pred_label = str(node_pred_label[i])
+        pred_label = node_pred_label[i]
         if pred_label in idx_to_cls:
             pred_label = idx_to_cls[pred_label]
         pred_score = '{:.2f}'.format(node_pred_score[i])
@@ -109,8 +108,7 @@ def main():
     save_res_path = config['Global']['save_res_path']
     class_path = config['Global']['class_path']
     idx_to_cls = read_class_list(class_path)
-    if not os.path.exists(os.path.dirname(save_res_path)):
-        os.makedirs(os.path.dirname(save_res_path))
+    os.makedirs(os.path.dirname(save_res_path), exist_ok=True)
 
     model.eval()
 
diff --git a/tools/infer_vqa_token_ser.py b/tools/infer_vqa_token_ser.py
index 83ed72b392..39ada64a99 100755
--- a/tools/infer_vqa_token_ser.py
+++ b/tools/infer_vqa_token_ser.py
@@ -86,15 +86,16 @@ class SerPredictor(object):
                 ]
 
             transforms.append(op)
-        global_config['infer_mode'] = True
+        if config["Global"].get("infer_mode", None) is None:
+            global_config['infer_mode'] = True
         self.ops = create_operators(config['Eval']['dataset']['transforms'],
                                     global_config)
         self.model.eval()
 
-    def __call__(self, img_path):
-        with open(img_path, 'rb') as f:
+    def __call__(self, data):
+        with open(data["img_path"], 'rb') as f:
             img = f.read()
-            data = {'image': img}
+        data["image"] = img
         batch = transform(data, self.ops)
         batch = to_tensor(batch)
         preds = self.model(batch)
@@ -112,20 +113,35 @@ if __name__ == '__main__':
 
     ser_engine = SerPredictor(config)
 
-    infer_imgs = get_image_file_list(config['Global']['infer_img'])
+    if config["Global"].get("infer_mode", None) is False:
+        data_dir = config['Eval']['dataset']['data_dir']
+        with open(config['Global']['infer_img'], "rb") as f:
+            infer_imgs = f.readlines()
+    else:
+        infer_imgs = get_image_file_list(config['Global']['infer_img'])
+
     with open(
             os.path.join(config['Global']['save_res_path'],
                          "infer_results.txt"),
             "w",
             encoding='utf-8') as fout:
-        for idx, img_path in enumerate(infer_imgs):
+        for idx, info in enumerate(infer_imgs):
+            if config["Global"].get("infer_mode", None) is False:
+                data_line = info.decode('utf-8')
+                substr = data_line.strip("\n").split("\t")
+                img_path = os.path.join(data_dir, substr[0])
+                data = {'img_path': img_path, 'label': substr[1]}
+            else:
+                img_path = info
+                data = {'img_path': img_path}
+
             save_img_path = os.path.join(
                 config['Global']['save_res_path'],
                 os.path.splitext(os.path.basename(img_path))[0] + "_ser.jpg")
             logger.info("process: [{}/{}], save result to {}".format(
                 idx, len(infer_imgs), save_img_path))
 
-            result, _ = ser_engine(img_path)
+            result, _ = ser_engine(data)
             result = result[0]
             fout.write(img_path + "\t" + json.dumps(
                 {
diff --git a/tools/program.py b/tools/program.py
index aa3ba82c44..f598feb7e8 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -576,8 +576,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
-        'ViTSTR', 'ABINet'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'LayoutLMv2', 'PREN', 'FCE',
+        'SVTR', 'ViTSTR', 'ABINet'
     ]
 
     if use_xpu:

From c26e7aee61b74f7e20e6fe62b5942d753b1440f7 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Thu, 30 Jun 2022 08:57:25 +0000
Subject: [PATCH 097/134] add db++

---
 tools/program.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tools/program.py b/tools/program.py
index 620c61f093..e8e72fdd2d 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -574,7 +574,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR', 'DB++'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'LayoutLMv2', 'PREN', 'FCE',
+        'SVTR', 'ViTSTR', 'ABINet', 'DB++'
     ]
 
     if use_xpu:

From 872d73daf5b8bcb4fd26afb194796fa42257b473 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Thu, 30 Jun 2022 08:59:05 +0000
Subject: [PATCH 098/134] add db++

---
 tools/program.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/program.py b/tools/program.py
index d86be5badb..e8e72fdd2d 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -574,8 +574,8 @@ def preprocess(is_train=False):
     assert alg in [
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
-        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'PREN', 'FCE', 'SVTR',
-        'ViTSTR', 'ABINet', 'DB++'
+        'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'LayoutLMv2', 'PREN', 'FCE',
+        'SVTR', 'ViTSTR', 'ABINet', 'DB++'
     ]
 
     if use_xpu:

From b1ee778fb367b960c7dd13a15cf6b90518ccc3c0 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Thu, 30 Jun 2022 09:10:23 +0000
Subject: [PATCH 099/134] rm proxy

---
 test_tipc/test_serving_infer_cpp.sh    | 3 +--
 test_tipc/test_serving_infer_python.sh | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/test_tipc/test_serving_infer_cpp.sh b/test_tipc/test_serving_infer_cpp.sh
index 6e313cd949..0be6a45adf 100644
--- a/test_tipc/test_serving_infer_cpp.sh
+++ b/test_tipc/test_serving_infer_cpp.sh
@@ -87,8 +87,7 @@ function func_serving(){
     set_image_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
     python_list=(${python_list})
     cd ${serving_dir_value}
-    unset https_proxy
-    unset http_proxy
+    
     # cpp serving
     for gpu_id in ${gpu_value[*]}; do
         if [ ${gpu_id} = "null" ]; then
diff --git a/test_tipc/test_serving_infer_python.sh b/test_tipc/test_serving_infer_python.sh
index b14517265d..4ccccc06e2 100644
--- a/test_tipc/test_serving_infer_python.sh
+++ b/test_tipc/test_serving_infer_python.sh
@@ -112,8 +112,7 @@ function func_serving(){
     
     cd ${serving_dir_value}
     python=${python_list[0]}
-    unset https_proxy
-    unset http_proxy
+    
     # python serving
     for use_gpu in ${web_use_gpu_list[*]}; do
         if [ ${use_gpu} = "null" ]; then

From d4400fbf8cd577e11371b8deeda84e9913abd9f2 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Fri, 1 Jul 2022 02:40:42 +0000
Subject: [PATCH 100/134] update log

---
 test_tipc/test_ptq_inference_python.sh   | 2 +-
 test_tipc/test_train_inference_python.sh | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/test_tipc/test_ptq_inference_python.sh b/test_tipc/test_ptq_inference_python.sh
index c1aa3daa6c..288e609896 100644
--- a/test_tipc/test_ptq_inference_python.sh
+++ b/test_tipc/test_ptq_inference_python.sh
@@ -139,7 +139,7 @@ if [ ${MODE} = "whole_infer" ]; then
             save_infer_dir="${infer_model}_klquant"
             set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
             set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
-            export_log_path="${LOG_PATH}/_export_${Count}.log"
+            export_log_path="${LOG_PATH}_export_${Count}.log"
             export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
             echo ${infer_run_exports[Count]} 
             echo $export_cmd
diff --git a/test_tipc/test_train_inference_python.sh b/test_tipc/test_train_inference_python.sh
index fa68cb2632..907efcec90 100644
--- a/test_tipc/test_train_inference_python.sh
+++ b/test_tipc/test_train_inference_python.sh
@@ -193,7 +193,7 @@ if [ ${MODE} = "whole_infer" ]; then
             save_infer_dir="${infer_model}"
             set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
             set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
-            export_log_path="${LOG_PATH}/_export_${Count}.log"
+            export_log_path="${LOG_PATH}_export_${Count}.log"
             export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
             echo ${infer_run_exports[Count]} 
             echo $export_cmd
@@ -295,6 +295,7 @@ else
                 fi
                 # run train
                 eval $cmd
+                eval "cat ${save_log}/train.log >> ${save_log}.log"
                 status_check $? "${cmd}" "${status_log}" "${model_name}"
 
                 set_eval_pretrain=$(func_set_params "${pretrain_model_key}" "${save_log}/${train_model_name}")

From 8d46a1fbbe33d37fc858c53afd0e9fcd9cc185fa Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Fri, 1 Jul 2022 08:52:08 +0000
Subject: [PATCH 101/134] add dygraph2static support of layoutlm series SER
 model

---
 configs/vqa/re/layoutlmv2.yml                 |  14 +-
 configs/vqa/re/layoutxlm.yml                  |  12 +-
 configs/vqa/ser/layoutlm.yml                  |   6 +-
 configs/vqa/ser/layoutlmv2.yml                |   6 +-
 configs/vqa/ser/layoutxlm.yml                 |   6 +-
 ppocr/data/imaug/label_ops.py                 |   2 +-
 ppocr/losses/vqa_token_layoutlm_loss.py       |   4 +-
 ppocr/modeling/backbones/vqa_layoutlm.py      |  50 +++---
 .../vqa_token_ser_layoutlm_postprocess.py     |   9 +-
 ppstructure/docs/models_list.md               |  21 +--
 ppstructure/docs/models_list_en.md            |  21 +--
 ppstructure/utility.py                        |   8 +
 ppstructure/vqa/README.md                     |  53 ++++--
 ppstructure/vqa/README_ch.md                  |  39 ++++-
 ppstructure/vqa/predict_vqa_token_ser.py      | 161 ++++++++++++++++++
 ppstructure/vqa/requirements.txt              |   5 +-
 tools/export_model.py                         |  18 +-
 tools/infer/utility.py                        |  11 +-
 tools/infer_vqa_token_ser.py                  |  18 +-
 tools/infer_vqa_token_ser_re.py               |  16 +-
 20 files changed, 368 insertions(+), 112 deletions(-)
 create mode 100644 ppstructure/vqa/predict_vqa_token_ser.py

diff --git a/configs/vqa/re/layoutlmv2.yml b/configs/vqa/re/layoutlmv2.yml
index 2fa5fd1165..737dbf6b60 100644
--- a/configs/vqa/re/layoutlmv2.yml
+++ b/configs/vqa/re/layoutlmv2.yml
@@ -11,7 +11,7 @@ Global:
   save_inference_dir:
   use_visualdl: False
   seed: 2048
-  infer_img: doc/vqa/input/zh_val_21.jpg
+  infer_img: ppstructure/docs/vqa/input/zh_val_21.jpg
   save_res_path: ./output/re/
 
 Architecture:
@@ -21,7 +21,7 @@ Architecture:
   Backbone:
     name: LayoutLMv2ForRe
     pretrained: True
-    checkpoints: 
+    checkpoints:
 
 Loss:
   name: LossFromOutput
@@ -52,7 +52,7 @@ Train:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_train/image
     label_file_list: 
-      - train_data/XFUND/zh_train/xfun_normalize_train.json
+      - train_data/XFUND/zh_train/train.json
     ratio_list: [ 1.0 ]
     transforms:
       - DecodeImage: # load image
@@ -61,7 +61,7 @@ Train:
       - VQATokenLabelEncode: # Class handling label
           contains_re: True
           algorithm: *algorithm
-          class_path: &class_path ppstructure/vqa/labels/labels_ser.txt
+          class_path: &class_path train_data/XFUND/class_list_xfun.txt
       - VQATokenPad:
           max_seq_len: &max_seq_len 512
           return_attention_mask: True
@@ -77,7 +77,7 @@ Train:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'bbox', 'image', 'attention_mask', 'token_type_ids','entities', 'relations'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids','image', 'entities', 'relations'] # dataloader will return list in this order
   loader:
     shuffle: True
     drop_last: False
@@ -90,7 +90,7 @@ Eval:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_val/image
     label_file_list:
-      - train_data/XFUND/zh_val/xfun_normalize_val.json
+      - train_data/XFUND/zh_val/val.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
@@ -114,7 +114,7 @@ Eval:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'bbox', 'image', 'attention_mask', 'token_type_ids','entities', 'relations'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image','entities', 'relations'] # dataloader will return list in this order
   loader:
     shuffle: False
     drop_last: False
diff --git a/configs/vqa/re/layoutxlm.yml b/configs/vqa/re/layoutxlm.yml
index ff16120ac1..d8585bb725 100644
--- a/configs/vqa/re/layoutxlm.yml
+++ b/configs/vqa/re/layoutxlm.yml
@@ -11,7 +11,7 @@ Global:
   save_inference_dir:
   use_visualdl: False
   seed: 2022
-  infer_img: doc/vqa/input/zh_val_21.jpg
+  infer_img: ppstructure/docs/vqa/input/zh_val_21.jpg
   save_res_path: ./output/re/
 
 Architecture:
@@ -52,7 +52,7 @@ Train:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_train/image
     label_file_list: 
-      - train_data/XFUND/zh_train/xfun_normalize_train.json
+      - train_data/XFUND/zh_train/train.json
     ratio_list: [ 1.0 ]
     transforms:
       - DecodeImage: # load image
@@ -61,7 +61,7 @@ Train:
       - VQATokenLabelEncode: # Class handling label
           contains_re: True
           algorithm: *algorithm
-          class_path: &class_path ppstructure/vqa/labels/labels_ser.txt
+          class_path: &class_path train_data/XFUND/class_list_xfun.txt
       - VQATokenPad:
           max_seq_len: &max_seq_len 512
           return_attention_mask: True
@@ -77,7 +77,7 @@ Train:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'bbox', 'image', 'attention_mask', 'token_type_ids','entities', 'relations'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox','attention_mask', 'token_type_ids', 'image', 'entities', 'relations'] # dataloader will return list in this order
   loader:
     shuffle: True
     drop_last: False
@@ -90,7 +90,7 @@ Eval:
     name: SimpleDataSet
     data_dir: train_data/XFUND/zh_val/image
     label_file_list:
-      - train_data/XFUND/zh_val/xfun_normalize_val.json
+      - train_data/XFUND/zh_val/val.json
     transforms:
       - DecodeImage: # load image
           img_mode: RGB
@@ -114,7 +114,7 @@ Eval:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'bbox', 'image', 'attention_mask', 'token_type_ids','entities', 'relations'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'entities', 'relations'] # dataloader will return list in this order
   loader:
     shuffle: False
     drop_last: False
diff --git a/configs/vqa/ser/layoutlm.yml b/configs/vqa/ser/layoutlm.yml
index 47ab093e1f..53e114defd 100644
--- a/configs/vqa/ser/layoutlm.yml
+++ b/configs/vqa/ser/layoutlm.yml
@@ -11,7 +11,7 @@ Global:
   save_inference_dir:
   use_visualdl: False
   seed: 2022
-  infer_img: doc/vqa/input/zh_val_0.jpg
+  infer_img: ppstructure/docs/vqa/input/zh_val_42.jpg
   save_res_path: ./output/ser/
 
 Architecture:
@@ -77,7 +77,7 @@ Train:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids','labels', 'bbox', 'image', 'attention_mask', 'token_type_ids'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
   loader:
     shuffle: True
     drop_last: False
@@ -112,7 +112,7 @@ Eval:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'labels', 'bbox', 'image', 'attention_mask', 'token_type_ids'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
   loader:
     shuffle: False
     drop_last: False
diff --git a/configs/vqa/ser/layoutlmv2.yml b/configs/vqa/ser/layoutlmv2.yml
index d6a9c03e5e..e48c746956 100644
--- a/configs/vqa/ser/layoutlmv2.yml
+++ b/configs/vqa/ser/layoutlmv2.yml
@@ -11,7 +11,7 @@ Global:
   save_inference_dir:
   use_visualdl: False
   seed: 2022
-  infer_img: doc/vqa/input/zh_val_0.jpg
+  infer_img: ppstructure/docs/vqa/input/zh_val_42.jpg
   save_res_path: ./output/ser/
 
 Architecture:
@@ -78,7 +78,7 @@ Train:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids','labels', 'bbox', 'image', 'attention_mask', 'token_type_ids'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
   loader:
     shuffle: True
     drop_last: False
@@ -113,7 +113,7 @@ Eval:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'labels', 'bbox', 'image', 'attention_mask', 'token_type_ids'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
   loader:
     shuffle: False
     drop_last: False
diff --git a/configs/vqa/ser/layoutxlm.yml b/configs/vqa/ser/layoutxlm.yml
index 3686989ccf..fa9df192af 100644
--- a/configs/vqa/ser/layoutxlm.yml
+++ b/configs/vqa/ser/layoutxlm.yml
@@ -43,7 +43,7 @@ Optimizer:
     
 PostProcess:
   name: VQASerTokenLayoutLMPostProcess
-  class_path: &class_path ppstructure/vqa/labels/labels_ser.txt
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
 
 Metric:
   name: VQASerTokenMetric
@@ -78,7 +78,7 @@ Train:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids','labels', 'bbox', 'image', 'attention_mask', 'token_type_ids'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
   loader:
     shuffle: True
     drop_last: False
@@ -113,7 +113,7 @@ Eval:
           order: 'hwc'
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: [ 'input_ids', 'labels', 'bbox', 'image', 'attention_mask', 'token_type_ids'] # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
   loader:
     shuffle: False
     drop_last: False
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index c95b326224..0723e97ae7 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -912,7 +912,7 @@ class VQATokenLabelEncode(object):
                 label = info['label']
                 gt_label = self._parse_label(label, encode_res)
 
-# construct entities for re
+            # construct entities for re
             if train_re:
                 if gt_label[0] != self.label2id_map["O"]:
                     entity_id_to_index_map[info["id"]] = len(entities)
diff --git a/ppocr/losses/vqa_token_layoutlm_loss.py b/ppocr/losses/vqa_token_layoutlm_loss.py
index 244893d97d..f9cd463473 100755
--- a/ppocr/losses/vqa_token_layoutlm_loss.py
+++ b/ppocr/losses/vqa_token_layoutlm_loss.py
@@ -27,8 +27,8 @@ class VQASerTokenLayoutLMLoss(nn.Layer):
         self.ignore_index = self.loss_class.ignore_index
 
     def forward(self, predicts, batch):
-        labels = batch[1]
-        attention_mask = batch[4]
+        labels = batch[5]
+        attention_mask = batch[2]
         if attention_mask is not None:
             active_loss = attention_mask.reshape([-1, ]) == 1
             active_outputs = predicts.reshape(
diff --git a/ppocr/modeling/backbones/vqa_layoutlm.py b/ppocr/modeling/backbones/vqa_layoutlm.py
index ede5b7a35a..1542519651 100644
--- a/ppocr/modeling/backbones/vqa_layoutlm.py
+++ b/ppocr/modeling/backbones/vqa_layoutlm.py
@@ -74,9 +74,9 @@ class LayoutLMForSer(NLPBaseModel):
     def forward(self, x):
         x = self.model(
             input_ids=x[0],
-            bbox=x[2],
-            attention_mask=x[4],
-            token_type_ids=x[5],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
             position_ids=None,
             output_hidden_states=False)
         return x
@@ -96,13 +96,15 @@ class LayoutLMv2ForSer(NLPBaseModel):
     def forward(self, x):
         x = self.model(
             input_ids=x[0],
-            bbox=x[2],
-            image=x[3],
-            attention_mask=x[4],
-            token_type_ids=x[5],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=x[4],
             position_ids=None,
             head_mask=None,
             labels=None)
+        if not self.training:
+            return x
         return x[0]
 
 
@@ -119,14 +121,16 @@ class LayoutXLMForSer(NLPBaseModel):
 
     def forward(self, x):
         x = self.model(
-            input_ids=x[0],
-            bbox=x[2],
-            image=x[3],
-            attention_mask=x[4],
-            token_type_ids=x[5],
-            position_ids=None,
-            head_mask=None,
-            labels=None)
+                        input_ids=x[0],
+                        bbox=x[1],
+                        attention_mask=x[2],
+                        token_type_ids=x[3],
+                        image=x[4],
+                        position_ids=None,
+                        head_mask=None,
+                        labels=None)
+        if not self.training:
+            return x
         return x[0]
 
 
@@ -140,12 +144,12 @@ class LayoutLMv2ForRe(NLPBaseModel):
         x = self.model(
             input_ids=x[0],
             bbox=x[1],
-            labels=None,
-            image=x[2],
-            attention_mask=x[3],
-            token_type_ids=x[4],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=x[4],
             position_ids=None,
             head_mask=None,
+            labels=None,
             entities=x[5],
             relations=x[6])
         return x
@@ -161,12 +165,12 @@ class LayoutXLMForRe(NLPBaseModel):
         x = self.model(
             input_ids=x[0],
             bbox=x[1],
-            labels=None,
-            image=x[2],
-            attention_mask=x[3],
-            token_type_ids=x[4],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=x[4],
             position_ids=None,
             head_mask=None,
+            labels=None,
             entities=x[5],
             relations=x[6])
         return x
diff --git a/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py b/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
index 782cdea6c5..90bc52733d 100644
--- a/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
+++ b/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
@@ -41,11 +41,13 @@ class VQASerTokenLayoutLMPostProcess(object):
                 self.id2label_map_for_show[val] = key
 
     def __call__(self, preds, batch=None, *args, **kwargs):
+        if isinstance(preds, tuple):
+            preds = preds[0]
         if isinstance(preds, paddle.Tensor):
             preds = preds.numpy()
 
         if batch is not None:
-            return self._metric(preds, batch[1])
+            return self._metric(preds, batch[5])
         else:
             return self._infer(preds, **kwargs)
 
@@ -63,11 +65,10 @@ class VQASerTokenLayoutLMPostProcess(object):
                                                                           j]])
         return decode_out_list, label_decode_out_list
 
-    def _infer(self, preds, attention_masks, segment_offset_ids, ocr_infos):
+    def _infer(self, preds, segment_offset_ids, ocr_infos):
         results = []
 
-        for pred, attention_mask, segment_offset_id, ocr_info in zip(
-                preds, attention_masks, segment_offset_ids, ocr_infos):
+        for pred, segment_offset_id, ocr_info in zip(preds, segment_offset_ids, ocr_infos):
             pred = np.argmax(pred, axis=1)
             pred = [self.id2label_map[idx] for idx in pred]
 
diff --git a/ppstructure/docs/models_list.md b/ppstructure/docs/models_list.md
index c7dab999ff..71d0566bb0 100644
--- a/ppstructure/docs/models_list.md
+++ b/ppstructure/docs/models_list.md
@@ -1,11 +1,12 @@
 # PP-Structure 系列模型列表
 
-- [1. 版面分析模型](#1)
-- [2. OCR和表格识别模型](#2)
-    - [2.1 OCR](#21)
-    - [2.2 表格识别模型](#22)
-- [3. VQA模型](#3)
-- [4. KIE模型](#4)
+- [PP-Structure 系列模型列表](#pp-structure-系列模型列表)
+  - [1. 版面分析模型](#1-版面分析模型)
+  - [2. OCR和表格识别模型](#2-ocr和表格识别模型)
+    - [2.1 OCR](#21-ocr)
+    - [2.2 表格识别模型](#22-表格识别模型)
+  - [3. VQA模型](#3-vqa模型)
+  - [4. KIE模型](#4-kie模型)
 
 
 <a name="1"></a>
@@ -42,11 +43,11 @@
 
 |模型名称|模型简介|推理模型大小|下载地址|
 | --- | --- | --- | --- |
-|ser_LayoutXLM_xfun_zh|基于LayoutXLM在xfun中文数据集上训练的SER模型|1.4G|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar) |
-|re_LayoutXLM_xfun_zh|基于LayoutXLM在xfun中文数据集上训练的RE模型|1.4G|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar) |
-|ser_LayoutLMv2_xfun_zh|基于LayoutLMv2在xfun中文数据集上训练的SER模型|778M|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar) |
+|ser_LayoutXLM_xfun_zh|基于LayoutXLM在xfun中文数据集上训练的SER模型|1.4G|[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar) |
+|re_LayoutXLM_xfun_zh|基于LayoutXLM在xfun中文数据集上训练的RE模型|1.4G|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar) |
+|ser_LayoutLMv2_xfun_zh|基于LayoutLMv2在xfun中文数据集上训练的SER模型|778M|[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar) |
 |re_LayoutLMv2_xfun_zh|基于LayoutLMv2在xfun中文数据集上训练的RE模型|765M|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar) |
-|ser_LayoutLM_xfun_zh|基于LayoutLM在xfun中文数据集上训练的SER模型|430M|[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar) |
+|ser_LayoutLM_xfun_zh|基于LayoutLM在xfun中文数据集上训练的SER模型|430M|[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar) |
 
 <a name="4"></a>
 ## 4. KIE模型
diff --git a/ppstructure/docs/models_list_en.md b/ppstructure/docs/models_list_en.md
index b92c10c241..b3a0755505 100644
--- a/ppstructure/docs/models_list_en.md
+++ b/ppstructure/docs/models_list_en.md
@@ -1,11 +1,12 @@
 # PP-Structure Model list
 
-- [1. Layout Analysis](#1)
-- [2. OCR and Table Recognition](#2)
-    - [2.1 OCR](#21)
-    - [2.2 Table Recognition](#22)
-- [3. VQA](#3)
-- [4. KIE](#4)
+- [PP-Structure Model list](#pp-structure-model-list)
+  - [1. Layout Analysis](#1-layout-analysis)
+  - [2. OCR and Table Recognition](#2-ocr-and-table-recognition)
+    - [2.1 OCR](#21-ocr)
+    - [2.2 Table Recognition](#22-table-recognition)
+  - [3. VQA](#3-vqa)
+  - [4. KIE](#4-kie)
 
 
 <a name="1"></a>
@@ -42,11 +43,11 @@ If you need to use other OCR models, you can download the model in [PP-OCR model
 
 |model| description                                                    |inference model size|download|
 | --- |----------------------------------------------------------------| --- | --- |
-|ser_LayoutXLM_xfun_zh| SER model trained on xfun Chinese dataset based on LayoutXLM   |1.4G|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar) |
-|re_LayoutXLM_xfun_zh| Re model trained on xfun Chinese dataset based on LayoutXLM    |1.4G|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar) |
-|ser_LayoutLMv2_xfun_zh| SER model trained on xfun Chinese dataset based on LayoutXLMv2 |778M|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar) |
+|ser_LayoutXLM_xfun_zh| SER model trained on xfun Chinese dataset based on LayoutXLM   |1.4G|[inference model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar) |
+|re_LayoutXLM_xfun_zh| Re model trained on xfun Chinese dataset based on LayoutXLM    |1.4G|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar) |
+|ser_LayoutLMv2_xfun_zh| SER model trained on xfun Chinese dataset based on LayoutXLMv2 |778M|[inference model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar) |
 |re_LayoutLMv2_xfun_zh| Re model trained on xfun Chinese dataset based on LayoutXLMv2  |765M|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar) |
-|ser_LayoutLM_xfun_zh| SER model trained on xfun Chinese dataset based on LayoutLM    |430M|[inference model coming soon]() / [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar) |
+|ser_LayoutLM_xfun_zh| SER model trained on xfun Chinese dataset based on LayoutLM    |430M|[inference model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar) |
 
 <a name="4"></a>
 ## 4. KIE
diff --git a/ppstructure/utility.py b/ppstructure/utility.py
index 1ad902e7e6..7ad3426740 100644
--- a/ppstructure/utility.py
+++ b/ppstructure/utility.py
@@ -40,6 +40,14 @@ def init_args():
         type=ast.literal_eval,
         default=None,
         help='label map according to ppstructure/layout/README_ch.md')
+
+    # params for vqa
+    parser.add_argument("--vqa_algorithm", type=str, default='LayoutXLM')
+    parser.add_argument("--ser_model_dir", type=str)
+    parser.add_argument(
+        "--ser_dict_path",
+        type=str,
+        default="../train_data/XFUND/class_list_xfun.txt")
     # params for inference
     parser.add_argument(
         "--mode",
diff --git a/ppstructure/vqa/README.md b/ppstructure/vqa/README.md
index 711ffa3138..cfb4210050 100644
--- a/ppstructure/vqa/README.md
+++ b/ppstructure/vqa/README.md
@@ -1,19 +1,16 @@
 English | [简体中文](README_ch.md)
 
-- [Document Visual Question Answering (Doc-VQA)](#Document-Visual-Question-Answering)
-  - [1. Introduction](#1-Introduction)
+- [Document Visual Question Answering](#document-visual-question-answering)
+  - [1 Introduction](#1-introduction)
   - [2. Performance](#2-performance)
-  - [3. Effect demo](#3-Effect-demo)
+  - [3. Effect demo](#3-effect-demo)
     - [3.1 SER](#31-ser)
     - [3.2 RE](#32-re)
-  - [4. Install](#4-Install)
-    - [4.1 Installation dependencies](#41-Install-dependencies)
-    - [4.2 Install PaddleOCR](#42-Install-PaddleOCR)
-  - [5. Usage](#5-Usage)
-    - [5.1 Data and Model Preparation](#51-Data-and-Model-Preparation)
-    - [5.2 SER](#52-ser)
+  - [4. Install](#4-install)
+    - [4.1 Install dependencies](#41-install-dependencies)
     - [5.3 RE](#53-re)
-  - [6. Reference](#6-Reference-Links)
+  - [6. Reference Links](#6-reference-links)
+  - [License](#license)
 
 # Document Visual Question Answering
 
@@ -187,17 +184,17 @@ CUDA_VISIBLE_DEVICES=0 python3 tools/eval.py -c configs/vqa/ser/layoutxlm.yml -o
 ````
 Finally, `precision`, `recall`, `hmean` and other indicators will be printed
 
-* Use `OCR engine + SER` tandem prediction
+* `OCR + SER` tandem prediction based on training engine
 
-Use the following command to complete the series prediction of `OCR engine + SER`, taking the pretrained SER model as an example:
+Use the following command to complete the series prediction of `OCR engine + SER`, taking the SER model based on LayoutXLM as an example::
 
 ```shell
-CUDA_VISIBLE_DEVICES=0 python3 tools/infer_vqa_token_ser.py -c configs/vqa/ser/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/Global.infer_img=doc/vqa/input/zh_val_42.jpg
+python3.7 tools/export_model.py -c configs/vqa/ser/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/ Global.save_inference_dir=output/ser/infer
 ````
 
 Finally, the prediction result visualization image and the prediction result text file will be saved in the directory configured by the `config.Global.save_res_path` field. The prediction result text file is named `infer_results.txt`.
 
-* End-to-end evaluation of `OCR engine + SER` prediction system
+* End-to-end evaluation of `OCR + SER` prediction system
 
 First use the `tools/infer_vqa_token_ser.py` script to complete the prediction of the dataset, then use the following command to evaluate.
 
@@ -205,6 +202,24 @@ First use the `tools/infer_vqa_token_ser.py` script to complete the prediction o
 export CUDA_VISIBLE_DEVICES=0
 python3 tools/eval_with_label_end2end.py --gt_json_path XFUND/zh_val/xfun_normalize_val.json --pred_json_path output_res/infer_results.txt
 ````
+* export model
+
+Use the following command to complete the model export of the SER model, taking the SER model based on LayoutXLM as an example:
+
+```shell
+python3.7 tools/export_model.py -c configs/vqa/ser/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/ Global.save_inference_dir=output/ser/infer
+```
+The converted model will be stored in the directory specified by the `Global.save_inference_dir` field.
+
+* `OCR + SER` tandem prediction based on prediction engine
+  
+Use the following command to complete the tandem prediction of `OCR + SER` based on the prediction engine, taking the SER model based on LayoutXLM as an example:
+
+```shell
+cd ppstructure
+CUDA_VISIBLE_DEVICES=0 python3.7 vqa/predict_vqa_token_ser.py --vqa_algorithm=LayoutXLM --ser_model_dir=../output/ser/infer --ser_dict_path=../train_data/XFUND/class_list_xfun.txt --image_dir=docs/vqa/input/zh_val_42.jpg --output=output
+```
+After the prediction is successful, the visualization images and results will be saved in the directory specified by the `output` field
 
 <a name="53"></a>
 ### 5.3 RE
@@ -247,11 +262,19 @@ Finally, `precision`, `recall`, `hmean` and other indicators will be printed
 Use the following command to complete the series prediction of `OCR engine + SER + RE`, taking the pretrained SER and RE models as an example:
 ```shell
 export CUDA_VISIBLE_DEVICES=0
-python3 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/re_LayoutXLM_xfun_zh/Global.infer_img=doc/vqa/input/zh_val_21.jpg -c_ser configs/vqa/ser/layoutxlm. yml -o_ser Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/
+python3 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/re_LayoutXLM_xfun_zh/Global.infer_img=ppstructure/docs/vqa/input/zh_val_21.jpg -c_ser configs/vqa/ser/layoutxlm. yml -o_ser Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/
 ````
 
 Finally, the prediction result visualization image and the prediction result text file will be saved in the directory configured by the `config.Global.save_res_path` field. The prediction result text file is named `infer_results.txt`.
 
+* export model
+  
+cooming soon
+
+* `OCR + SER + RE` tandem prediction based on prediction engine
+
+cooming soon
+
 ## 6. Reference Links
 
 - LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, https://arxiv.org/pdf/2104.08836.pdf
diff --git a/ppstructure/vqa/README_ch.md b/ppstructure/vqa/README_ch.md
index 297ba64f82..52113b80b5 100644
--- a/ppstructure/vqa/README_ch.md
+++ b/ppstructure/vqa/README_ch.md
@@ -14,6 +14,7 @@
     - [5.2 SER](#52-ser)
     - [5.3 RE](#53-re)
   - [6. 参考链接](#6-参考链接)
+  - [License](#license)
 
 # 文档视觉问答（DOC-VQA）
 
@@ -183,16 +184,16 @@ CUDA_VISIBLE_DEVICES=0 python3 tools/eval.py -c configs/vqa/ser/layoutxlm.yml -o
 ```
 最终会打印出`precision`, `recall`, `hmean`等指标
 
-* 使用`OCR引擎 + SER`串联预测
+* 基于训练引擎的`OCR + SER`串联预测
 
-使用如下命令即可完成`OCR引擎 + SER`的串联预测, 以SER预训练模型为例:
+使用如下命令即可完成基于训练引擎的`OCR + SER`的串联预测, 以基于LayoutXLM的SER模型为例:
 ```shell
 CUDA_VISIBLE_DEVICES=0 python3 tools/infer_vqa_token_ser.py -c configs/vqa/ser/layoutxlm.yml  -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/ Global.infer_img=doc/vqa/input/zh_val_42.jpg
 ```
 
 最终会在`config.Global.save_res_path`字段所配置的目录下保存预测结果可视化图像以及预测结果文本文件，预测结果文本文件名为`infer_results.txt`。
 
-* 对`OCR引擎 + SER`预测系统进行端到端评估
+* 对`OCR + SER`预测系统进行端到端评估
 
 首先使用 `tools/infer_vqa_token_ser.py` 脚本完成数据集的预测，然后使用下面的命令进行评估。
 
@@ -200,6 +201,24 @@ CUDA_VISIBLE_DEVICES=0 python3 tools/infer_vqa_token_ser.py -c configs/vqa/ser/l
 export CUDA_VISIBLE_DEVICES=0
 python3 tools/eval_with_label_end2end.py --gt_json_path XFUND/zh_val/xfun_normalize_val.json  --pred_json_path output_res/infer_results.txt
 ```
+* 模型导出
+
+使用如下命令即可完成SER模型的模型导出, 以基于LayoutXLM的SER模型为例:
+
+```shell
+python3.7 tools/export_model.py -c configs/vqa/ser/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/ Global.save_inference_dir=output/ser/infer
+```
+转换后的模型会存放在`Global.save_inference_dir`字段指定的目录下。
+
+* 基于预测引擎的`OCR + SER`串联预测
+  
+使用如下命令即可完成基于预测引擎的`OCR + SER`的串联预测, 以基于LayoutXLM的SER模型为例:
+
+```shell
+cd ppstructure
+CUDA_VISIBLE_DEVICES=0 python3.7 vqa/predict_vqa_token_ser.py --vqa_algorithm=LayoutXLM --ser_model_dir=../output/ser/infer --ser_dict_path=../train_data/XFUND/class_list_xfun.txt --image_dir=docs/vqa/input/zh_val_42.jpg --output=output
+```
+预测成功后，可视化图片和结果会保存在`output`字段指定的目录下
 
 ### 5.3 RE
 
@@ -236,16 +255,24 @@ CUDA_VISIBLE_DEVICES=0 python3 tools/eval.py -c configs/vqa/re/layoutxlm.yml -o
 ```
 最终会打印出`precision`, `recall`, `hmean`等指标
 
-* 使用`OCR引擎 + SER + RE`串联预测
+* 基于训练引擎的`OCR + SER + RE`串联预测
 
-使用如下命令即可完成`OCR引擎 + SER + RE`的串联预测, 以预训练SER和RE模型为例：
+使用如下命令即可完成基于训练引擎的`OCR + SER + RE`串联预测, 以基于LayoutXLMSER和RE模型为例：
 ```shell
 export CUDA_VISIBLE_DEVICES=0
-python3 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/re_LayoutXLM_xfun_zh/ Global.infer_img=doc/vqa/input/zh_val_21.jpg -c_ser configs/vqa/ser/layoutxlm.yml -o_ser Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/
+python3 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Architecture.Backbone.checkpoints=pretrain/re_LayoutXLM_xfun_zh/ Global.infer_img=ppstructure/docs/vqa/input/zh_val_21.jpg -c_ser configs/vqa/ser/layoutxlm.yml -o_ser Architecture.Backbone.checkpoints=pretrain/ser_LayoutXLM_xfun_zh/
 ```
 
 最终会在`config.Global.save_res_path`字段所配置的目录下保存预测结果可视化图像以及预测结果文本文件，预测结果文本文件名为`infer_results.txt`。
 
+* 模型导出
+  
+cooming soon
+
+* 基于预测引擎的`OCR + SER + RE`串联预测
+
+cooming soon
+
 ## 6. 参考链接
 
 - LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, https://arxiv.org/pdf/2104.08836.pdf
diff --git a/ppstructure/vqa/predict_vqa_token_ser.py b/ppstructure/vqa/predict_vqa_token_ser.py
new file mode 100644
index 0000000000..f55c875713
--- /dev/null
+++ b/ppstructure/vqa/predict_vqa_token_ser.py
@@ -0,0 +1,161 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
+
+os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
+
+import cv2
+import json
+import numpy as np
+import time
+
+import tools.infer.utility as utility
+from ppocr.data import create_operators, transform
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.visual import draw_ser_results
+from ppocr.utils.utility import get_image_file_list, check_and_read_gif
+from ppstructure.utility import parse_args
+
+from paddleocr import PaddleOCR
+
+logger = get_logger()
+
+
+class SerPredictor(object):
+    def __init__(self, args):
+        self.ocr_engine = PaddleOCR(use_angle_cls=False, show_log=False)
+
+        pre_process_list = [{
+            'VQATokenLabelEncode': {
+                'algorithm': args.vqa_algorithm,
+                'class_path': args.ser_dict_path,
+                'contains_re': False,
+                'ocr_engine': self.ocr_engine
+            }
+        }, {
+            'VQATokenPad':{
+                'max_seq_len':512,
+                'return_attention_mask': True
+            }
+        }, {
+            'VQASerTokenChunk':{
+                'max_seq_len':512,
+                'return_attention_mask': True
+            }
+        }, {
+            'Resize':{
+                'size' : [224, 224]
+            }
+        }, {
+            'NormalizeImage': {
+                'std': [58.395, 57.12, 57.375],
+                'mean': [123.675, 116.28, 103.53],
+                'scale': '1',
+                'order': 'hwc'
+            }
+        }, {
+            'ToCHWImage': None
+        }, {
+            'KeepKeys': {
+                'keep_keys': [
+                    'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels',
+                    'segment_offset_id', 'ocr_info',
+                    'entities'
+                ]
+            }
+        }]
+        postprocess_params = {
+            'name': 'VQASerTokenLayoutLMPostProcess',
+            "class_path": args.ser_dict_path,
+        }
+
+        self.preprocess_op = create_operators(pre_process_list, {'infer_mode':True})
+        self.postprocess_op = build_post_process(postprocess_params)
+        self.predictor, self.input_tensor, self.output_tensors, self.config = \
+            utility.create_predictor(args, 'ser', logger)
+
+    def __call__(self, img):
+        ori_im = img.copy()
+        data = {'image': img}
+        data = transform(data, self.preprocess_op)
+        img = data[0]
+        if img is None:
+            return None, 0
+        img = np.expand_dims(img, axis=0)
+        img = img.copy()
+        starttime = time.time()
+
+        for idx in range(len(self.input_tensor)):
+            expand_input = np.expand_dims(data[idx], axis=0)
+            self.input_tensor[idx].copy_from_cpu(expand_input)
+
+        self.predictor.run()
+
+        outputs = []
+        for output_tensor in self.output_tensors:
+            output = output_tensor.copy_to_cpu()
+            outputs.append(output)
+        preds = outputs[0]
+        
+        post_result = self.postprocess_op(
+            preds,
+            segment_offset_ids=[data[6]],
+            ocr_infos=[data[7]])
+        elapse = time.time() - starttime
+        return post_result, elapse
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    ser_predictor = SerPredictor(args)
+    count = 0
+    total_time = 0
+
+    os.makedirs(args.output, exist_ok=True)
+    with open(
+            os.path.join(args.output, 'infer.txt'), mode='w',
+            encoding='utf-8') as f_w:
+        for image_file in image_file_list:
+            img, flag = check_and_read_gif(image_file)
+            if not flag:
+                img = cv2.imread(image_file)
+                img = img[:,:,::-1]
+            if img is None:
+                logger.info("error in loading image:{}".format(image_file))
+                continue
+            ser_res, elapse = ser_predictor(img)
+            ser_res = ser_res[0]
+
+            res_str = '{}\t{}\n'.format(image_file,json.dumps({"ocr_info": ser_res,}, ensure_ascii=False))
+            f_w.write(res_str)
+
+            img_res = draw_ser_results(image_file, ser_res,  font_path="../doc/fonts/simfang.ttf",)
+
+            img_save_path = os.path.join(args.output,
+                                         os.path.basename(image_file))
+            cv2.imwrite(img_save_path, img_res)
+            logger.info("save vis result to {}".format(img_save_path))
+            if count > 0:
+                total_time += elapse
+            count += 1
+            logger.info("Predict time of {}: {}".format(image_file, elapse))
+
+if __name__ == "__main__":
+    main(parse_args())
diff --git a/ppstructure/vqa/requirements.txt b/ppstructure/vqa/requirements.txt
index 0042ec0bae..fcd882274c 100644
--- a/ppstructure/vqa/requirements.txt
+++ b/ppstructure/vqa/requirements.txt
@@ -1,4 +1,7 @@
 sentencepiece
 yacs
 seqeval
-paddlenlp>=2.2.1
\ No newline at end of file
+paddlenlp>=2.2.1
+pypandoc
+attrdict
+python_docx
\ No newline at end of file
diff --git a/tools/export_model.py b/tools/export_model.py
index b10d41d5b2..752732ff7e 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -97,6 +97,22 @@ def export_single_model(model,
                 shape=[None, 1, 32, 100], dtype="float32"),
         ]
         model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] in ["LayoutLM", "LayoutLMv2", "LayoutXLM"]:
+        input_spec=[
+            paddle.static.InputSpec(
+                shape=[None, 512], dtype="int64"),  # input_ids
+            paddle.static.InputSpec(
+                shape=[None, 512, 4], dtype="int64"),  # bbox
+            paddle.static.InputSpec(
+                shape=[None, 512], dtype="int64"),  # attention_mask
+            paddle.static.InputSpec(
+                shape=[None, 512], dtype="int64"),  # token_type_ids
+            paddle.static.InputSpec(
+                shape=[None, 3, 224, 224], dtype="int64"),  # image
+        ]
+        if arch_config["algorithm"] == "LayoutLM":
+            input_spec.pop(4)
+        model = to_static(model, input_spec=[input_spec])
     else:
         infer_shape = [3, -1, -1]
         if arch_config["model_type"] == "rec":
@@ -172,7 +188,7 @@ def main():
             config["Architecture"]["Head"]["out_channels"] = char_num
 
     model = build_model(config["Architecture"])
-    load_model(config, model)
+    load_model(config, model, model_type=config['Architecture']["model_type"])
     model.eval()
 
     save_path = config["Global"]["save_inference_dir"]
diff --git a/tools/infer/utility.py b/tools/infer/utility.py
index 366212f228..aa5d2371b6 100644
--- a/tools/infer/utility.py
+++ b/tools/infer/utility.py
@@ -153,6 +153,8 @@ def create_predictor(args, mode, logger):
         model_dir = args.rec_model_dir
     elif mode == 'table':
         model_dir = args.table_model_dir
+    elif mode == 'ser':
+        model_dir = args.ser_model_dir
     else:
         model_dir = args.e2e_model_dir
 
@@ -316,8 +318,13 @@ def create_predictor(args, mode, logger):
         # create predictor
         predictor = inference.create_predictor(config)
         input_names = predictor.get_input_names()
-        for name in input_names:
-            input_tensor = predictor.get_input_handle(name)
+        if mode in ['ser','re']:
+            input_tensor = []
+            for name in input_names:
+                input_tensor.append(predictor.get_input_handle(name))
+        else:
+            for name in input_names:
+                input_tensor = predictor.get_input_handle(name)
         output_tensors = get_output_tensors(args, mode, predictor)
         return predictor, input_tensor, output_tensors, config
 
diff --git a/tools/infer_vqa_token_ser.py b/tools/infer_vqa_token_ser.py
index 39ada64a99..1e661ec2c7 100755
--- a/tools/infer_vqa_token_ser.py
+++ b/tools/infer_vqa_token_ser.py
@@ -44,6 +44,7 @@ def to_tensor(data):
     from collections import defaultdict
     data_dict = defaultdict(list)
     to_tensor_idxs = []
+    
     for idx, v in enumerate(data):
         if isinstance(v, (np.ndarray, paddle.Tensor, numbers.Number)):
             if idx not in to_tensor_idxs:
@@ -57,6 +58,7 @@ def to_tensor(data):
 class SerPredictor(object):
     def __init__(self, config):
         global_config = config['Global']
+        self.algorithm = config['Architecture']["algorithm"]
 
         # build post process
         self.post_process_class = build_post_process(config['PostProcess'],
@@ -70,7 +72,7 @@ class SerPredictor(object):
 
         from paddleocr import PaddleOCR
 
-        self.ocr_engine = PaddleOCR(use_angle_cls=False, show_log=False)
+        self.ocr_engine = PaddleOCR(use_angle_cls=False, show_log=False, use_gpu=global_config['use_gpu'])
 
         # create data ops
         transforms = []
@@ -80,8 +82,8 @@ class SerPredictor(object):
                 op[op_name]['ocr_engine'] = self.ocr_engine
             elif op_name == 'KeepKeys':
                 op[op_name]['keep_keys'] = [
-                    'input_ids', 'labels', 'bbox', 'image', 'attention_mask',
-                    'token_type_ids', 'segment_offset_id', 'ocr_info',
+                    'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels',
+                    'segment_offset_id', 'ocr_info',
                     'entities'
                 ]
 
@@ -99,9 +101,11 @@ class SerPredictor(object):
         batch = transform(data, self.ops)
         batch = to_tensor(batch)
         preds = self.model(batch)
+        if self.algorithm in ['LayoutLMv2', 'LayoutXLM']:
+            preds = preds[0]
+        
         post_result = self.post_process_class(
             preds,
-            attention_masks=batch[4],
             segment_offset_ids=batch[6],
             ocr_infos=batch[7])
         return post_result, batch
@@ -138,8 +142,6 @@ if __name__ == '__main__':
             save_img_path = os.path.join(
                 config['Global']['save_res_path'],
                 os.path.splitext(os.path.basename(img_path))[0] + "_ser.jpg")
-            logger.info("process: [{}/{}], save result to {}".format(
-                idx, len(infer_imgs), save_img_path))
 
             result, _ = ser_engine(data)
             result = result[0]
@@ -149,3 +151,7 @@ if __name__ == '__main__':
                 }, ensure_ascii=False) + "\n")
             img_res = draw_ser_results(img_path, result)
             cv2.imwrite(save_img_path, img_res)
+
+            logger.info("process: [{}/{}], save result to {}".format(
+                idx, len(infer_imgs), save_img_path))
+
diff --git a/tools/infer_vqa_token_ser_re.py b/tools/infer_vqa_token_ser_re.py
index 6210f7f3c2..d5ae634cea 100755
--- a/tools/infer_vqa_token_ser_re.py
+++ b/tools/infer_vqa_token_ser_re.py
@@ -38,7 +38,7 @@ from ppocr.utils.save_load import load_model
 from ppocr.utils.visual import draw_re_results
 from ppocr.utils.logging import get_logger
 from ppocr.utils.utility import get_image_file_list, load_vqa_bio_label_maps, print_dict
-from tools.program import ArgsParser, load_config, merge_config, check_gpu
+from tools.program import ArgsParser, load_config, merge_config
 from tools.infer_vqa_token_ser import SerPredictor
 
 
@@ -107,7 +107,7 @@ def make_input(ser_inputs, ser_results):
     # remove ocr_info segment_offset_id and label in ser input
     ser_inputs.pop(7)
     ser_inputs.pop(6)
-    ser_inputs.pop(1)
+    ser_inputs.pop(5)
     return ser_inputs, entity_idx_dict_batch
 
 
@@ -131,9 +131,7 @@ class SerRePredictor(object):
         self.model.eval()
 
     def __call__(self, img_path):
-        ser_results, ser_inputs = self.ser_engine(img_path)
-        paddle.save(ser_inputs, 'ser_inputs.npy')
-        paddle.save(ser_results, 'ser_results.npy')
+        ser_results, ser_inputs = self.ser_engine({'img_path': img_path})
         re_input, entity_idx_dict_batch = make_input(ser_inputs, ser_results)
         preds = self.model(re_input)
         post_result = self.post_process_class(
@@ -155,7 +153,6 @@ def preprocess():
 
     # check if set use_gpu=True in paddlepaddle cpu version
     use_gpu = config['Global']['use_gpu']
-    check_gpu(use_gpu)
 
     device = 'gpu:{}'.format(dist.ParallelEnv().dev_id) if use_gpu else 'cpu'
     device = paddle.set_device(device)
@@ -185,9 +182,7 @@ if __name__ == '__main__':
         for idx, img_path in enumerate(infer_imgs):
             save_img_path = os.path.join(
                 config['Global']['save_res_path'],
-                os.path.splitext(os.path.basename(img_path))[0] + "_ser.jpg")
-            logger.info("process: [{}/{}], save result to {}".format(
-                idx, len(infer_imgs), save_img_path))
+                os.path.splitext(os.path.basename(img_path))[0] + "_ser_re.jpg")
 
             result = ser_re_engine(img_path)
             result = result[0]
@@ -197,3 +192,6 @@ if __name__ == '__main__':
                 }, ensure_ascii=False) + "\n")
             img_res = draw_re_results(img_path, result)
             cv2.imwrite(save_img_path, img_res)
+            
+            logger.info("process: [{}/{}], save result to {}".format(
+                idx, len(infer_imgs), save_img_path))
\ No newline at end of file

From ce21ad8357b3da70999442e9ca492ea7e7978473 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Fri, 1 Jul 2022 08:55:04 +0000
Subject: [PATCH 102/134] update docs

---
 ppstructure/docs/models_list.md    | 13 ++++++------
 ppstructure/docs/models_list_en.md | 13 ++++++------
 ppstructure/infer.sh               |  4 ++++
 ppstructure/vqa/README.md          | 25 +++++++++++-----------
 ppstructure/vqa/README_ch.md       | 33 +++++++++++++++---------------
 5 files changed, 44 insertions(+), 44 deletions(-)
 create mode 100644 ppstructure/infer.sh

diff --git a/ppstructure/docs/models_list.md b/ppstructure/docs/models_list.md
index 71d0566bb0..dabce3a514 100644
--- a/ppstructure/docs/models_list.md
+++ b/ppstructure/docs/models_list.md
@@ -1,12 +1,11 @@
 # PP-Structure 系列模型列表
 
-- [PP-Structure 系列模型列表](#pp-structure-系列模型列表)
-  - [1. 版面分析模型](#1-版面分析模型)
-  - [2. OCR和表格识别模型](#2-ocr和表格识别模型)
-    - [2.1 OCR](#21-ocr)
-    - [2.2 表格识别模型](#22-表格识别模型)
-  - [3. VQA模型](#3-vqa模型)
-  - [4. KIE模型](#4-kie模型)
+- [1. 版面分析模型](#1-版面分析模型)
+- [2. OCR和表格识别模型](#2-ocr和表格识别模型)
+  - [2.1 OCR](#21-ocr)
+  - [2.2 表格识别模型](#22-表格识别模型)
+- [3. VQA模型](#3-vqa模型)
+- [4. KIE模型](#4-kie模型)
 
 
 <a name="1"></a>
diff --git a/ppstructure/docs/models_list_en.md b/ppstructure/docs/models_list_en.md
index b3a0755505..e133a0bb2a 100644
--- a/ppstructure/docs/models_list_en.md
+++ b/ppstructure/docs/models_list_en.md
@@ -1,12 +1,11 @@
 # PP-Structure Model list
 
-- [PP-Structure Model list](#pp-structure-model-list)
-  - [1. Layout Analysis](#1-layout-analysis)
-  - [2. OCR and Table Recognition](#2-ocr-and-table-recognition)
-    - [2.1 OCR](#21-ocr)
-    - [2.2 Table Recognition](#22-table-recognition)
-  - [3. VQA](#3-vqa)
-  - [4. KIE](#4-kie)
+- [1. Layout Analysis](#1-layout-analysis)
+- [2. OCR and Table Recognition](#2-ocr-and-table-recognition)
+  - [2.1 OCR](#21-ocr)
+  - [2.2 Table Recognition](#22-table-recognition)
+- [3. VQA](#3-vqa)
+- [4. KIE](#4-kie)
 
 
 <a name="1"></a>
diff --git a/ppstructure/infer.sh b/ppstructure/infer.sh
new file mode 100644
index 0000000000..a08cbadf09
--- /dev/null
+++ b/ppstructure/infer.sh
@@ -0,0 +1,4 @@
+python3.7 vqa/predict_vqa_token_ser.py --vqa_algorithm=LayoutXLM --ser_model_dir=../models/ser_LayoutXLM_xfun_zh/infer --ser_dict_path=../train_data/XFUND/class_list_xfun.txt --image_dir=docs/vqa/input/zh_val_42.jpg
+
+
+python3.7 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Architecture.Backbone.checkpoints=models/re_LayoutXLM_xfun_zh/ Global.infer_img=ppstructure/docs/vqa/input/zh_val_21.jpg -c_ser configs/vqa/ser/layoutxlm.yml -o_ser Architecture.Backbone.checkpoints=models/ser_LayoutXLM_xfun_zh/
\ No newline at end of file
diff --git a/ppstructure/vqa/README.md b/ppstructure/vqa/README.md
index cfb4210050..05635265b5 100644
--- a/ppstructure/vqa/README.md
+++ b/ppstructure/vqa/README.md
@@ -1,16 +1,15 @@
 English | [简体中文](README_ch.md)
 
-- [Document Visual Question Answering](#document-visual-question-answering)
-  - [1 Introduction](#1-introduction)
-  - [2. Performance](#2-performance)
-  - [3. Effect demo](#3-effect-demo)
-    - [3.1 SER](#31-ser)
-    - [3.2 RE](#32-re)
-  - [4. Install](#4-install)
-    - [4.1 Install dependencies](#41-install-dependencies)
-    - [5.3 RE](#53-re)
-  - [6. Reference Links](#6-reference-links)
-  - [License](#license)
+- [1 Introduction](#1-introduction)
+- [2. Performance](#2-performance)
+- [3. Effect demo](#3-effect-demo)
+  - [3.1 SER](#31-ser)
+  - [3.2 RE](#32-re)
+- [4. Install](#4-install)
+  - [4.1 Install dependencies](#41-install-dependencies)
+  - [5.3 RE](#53-re)
+- [6. Reference Links](#6-reference-links)
+- [License](#license)
 
 # Document Visual Question Answering
 
@@ -212,7 +211,7 @@ python3.7 tools/export_model.py -c configs/vqa/ser/layoutxlm.yml -o Architecture
 The converted model will be stored in the directory specified by the `Global.save_inference_dir` field.
 
 * `OCR + SER` tandem prediction based on prediction engine
-  
+
 Use the following command to complete the tandem prediction of `OCR + SER` based on the prediction engine, taking the SER model based on LayoutXLM as an example:
 
 ```shell
@@ -268,7 +267,7 @@ python3 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Archi
 Finally, the prediction result visualization image and the prediction result text file will be saved in the directory configured by the `config.Global.save_res_path` field. The prediction result text file is named `infer_results.txt`.
 
 * export model
-  
+
 cooming soon
 
 * `OCR + SER + RE` tandem prediction based on prediction engine
diff --git a/ppstructure/vqa/README_ch.md b/ppstructure/vqa/README_ch.md
index 52113b80b5..b421a82d3a 100644
--- a/ppstructure/vqa/README_ch.md
+++ b/ppstructure/vqa/README_ch.md
@@ -1,20 +1,19 @@
 [English](README.md) | 简体中文
 
-- [文档视觉问答（DOC-VQA）](#文档视觉问答doc-vqa)
-  - [1. 简介](#1-简介)
-  - [2. 性能](#2-性能)
-  - [3. 效果演示](#3-效果演示)
-    - [3.1 SER](#31-ser)
-    - [3.2 RE](#32-re)
-  - [4. 安装](#4-安装)
-    - [4.1 安装依赖](#41-安装依赖)
-    - [4.2 安装PaddleOCR（包含 PP-OCR 和 VQA）](#42-安装paddleocr包含-pp-ocr-和-vqa)
-  - [5. 使用](#5-使用)
-    - [5.1 数据和预训练模型准备](#51-数据和预训练模型准备)
-    - [5.2 SER](#52-ser)
-    - [5.3 RE](#53-re)
-  - [6. 参考链接](#6-参考链接)
-  - [License](#license)
+- [1. 简介](#1-简介)
+- [2. 性能](#2-性能)
+- [3. 效果演示](#3-效果演示)
+  - [3.1 SER](#31-ser)
+  - [3.2 RE](#32-re)
+- [4. 安装](#4-安装)
+  - [4.1 安装依赖](#41-安装依赖)
+  - [4.2 安装PaddleOCR（包含 PP-OCR 和 VQA）](#42-安装paddleocr包含-pp-ocr-和-vqa)
+- [5. 使用](#5-使用)
+  - [5.1 数据和预训练模型准备](#51-数据和预训练模型准备)
+  - [5.2 SER](#52-ser)
+  - [5.3 RE](#53-re)
+- [6. 参考链接](#6-参考链接)
+- [License](#license)
 
 # 文档视觉问答（DOC-VQA）
 
@@ -211,7 +210,7 @@ python3.7 tools/export_model.py -c configs/vqa/ser/layoutxlm.yml -o Architecture
 转换后的模型会存放在`Global.save_inference_dir`字段指定的目录下。
 
 * 基于预测引擎的`OCR + SER`串联预测
-  
+
 使用如下命令即可完成基于预测引擎的`OCR + SER`的串联预测, 以基于LayoutXLM的SER模型为例:
 
 ```shell
@@ -266,7 +265,7 @@ python3 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Archi
 最终会在`config.Global.save_res_path`字段所配置的目录下保存预测结果可视化图像以及预测结果文本文件，预测结果文本文件名为`infer_results.txt`。
 
 * 模型导出
-  
+
 cooming soon
 
 * 基于预测引擎的`OCR + SER + RE`串联预测

From 907ebdfe05e6734ff2957e576c28668841ffd272 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Fri, 1 Jul 2022 09:17:55 +0000
Subject: [PATCH 103/134] add db++ tipc

---
 test_tipc/prepare.sh | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 2c9bd2901b..32df8e7865 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -59,6 +59,9 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
         cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
     fi
+    if [[ ${model_name} =~ "det_r50_db++" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams --no-check-certificate
+    fi
     cd ./pretrain_models/ && tar xf det_mv3_db_v2.0_train.tar && cd ../
     rm -rf ./train_data/icdar2015
     rm -rf ./train_data/ic15_data

From 68b379e5c62b22fee0ee556c5de4ab6b910664f7 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Fri, 1 Jul 2022 09:37:18 +0000
Subject: [PATCH 104/134] add db++ tipc

---
 .../det_r50_db++/train_infer_python.txt       | 59 +++++++++++++++++++
 1 file changed, 59 insertions(+)
 create mode 100644 test_tipc/configs/det_r50_db++/train_infer_python.txt

diff --git a/test_tipc/configs/det_r50_db++/train_infer_python.txt b/test_tipc/configs/det_r50_db++/train_infer_python.txt
new file mode 100644
index 0000000000..bcf393a52b
--- /dev/null
+++ b/test_tipc/configs/det_r50_db++/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:det_r50_db++
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/det_r50_db++_ic15.yml -o Global.pretrained_model=./pretrain_models/ResNet50_dcn_asf_synthtext_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/det_r50_db++_ic15.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/det_r50_db++_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/det_r50_db++_ic15.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py --det_algorithm="DB++"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8|16
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096

From dc7bfe8a8441167cf6303879800eee7e160e2679 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Fri, 1 Jul 2022 09:42:27 +0000
Subject: [PATCH 105/134] fix

---
 ppstructure/infer.sh                     |  4 ---
 ppstructure/vqa/predict_vqa_token_ser.py | 42 ++++++++++++++----------
 2 files changed, 25 insertions(+), 21 deletions(-)
 delete mode 100644 ppstructure/infer.sh

diff --git a/ppstructure/infer.sh b/ppstructure/infer.sh
deleted file mode 100644
index a08cbadf09..0000000000
--- a/ppstructure/infer.sh
+++ /dev/null
@@ -1,4 +0,0 @@
-python3.7 vqa/predict_vqa_token_ser.py --vqa_algorithm=LayoutXLM --ser_model_dir=../models/ser_LayoutXLM_xfun_zh/infer --ser_dict_path=../train_data/XFUND/class_list_xfun.txt --image_dir=docs/vqa/input/zh_val_42.jpg
-
-
-python3.7 tools/infer_vqa_token_ser_re.py -c configs/vqa/re/layoutxlm.yml -o Architecture.Backbone.checkpoints=models/re_LayoutXLM_xfun_zh/ Global.infer_img=ppstructure/docs/vqa/input/zh_val_21.jpg -c_ser configs/vqa/ser/layoutxlm.yml -o_ser Architecture.Backbone.checkpoints=models/ser_LayoutXLM_xfun_zh/
\ No newline at end of file
diff --git a/ppstructure/vqa/predict_vqa_token_ser.py b/ppstructure/vqa/predict_vqa_token_ser.py
index f55c875713..de0bbfe72d 100644
--- a/ppstructure/vqa/predict_vqa_token_ser.py
+++ b/ppstructure/vqa/predict_vqa_token_ser.py
@@ -16,7 +16,7 @@ import sys
 
 __dir__ = os.path.dirname(os.path.abspath(__file__))
 sys.path.append(__dir__)
-sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, '../..')))
 
 os.environ["FLAGS_allocator_strategy"] = 'auto_growth'
 
@@ -50,18 +50,18 @@ class SerPredictor(object):
                 'ocr_engine': self.ocr_engine
             }
         }, {
-            'VQATokenPad':{
-                'max_seq_len':512,
+            'VQATokenPad': {
+                'max_seq_len': 512,
                 'return_attention_mask': True
             }
         }, {
-            'VQASerTokenChunk':{
-                'max_seq_len':512,
+            'VQASerTokenChunk': {
+                'max_seq_len': 512,
                 'return_attention_mask': True
             }
         }, {
-            'Resize':{
-                'size' : [224, 224]
+            'Resize': {
+                'size': [224, 224]
             }
         }, {
             'NormalizeImage': {
@@ -75,8 +75,8 @@ class SerPredictor(object):
         }, {
             'KeepKeys': {
                 'keep_keys': [
-                    'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels',
-                    'segment_offset_id', 'ocr_info',
+                    'input_ids', 'bbox', 'attention_mask', 'token_type_ids',
+                    'image', 'labels', 'segment_offset_id', 'ocr_info',
                     'entities'
                 ]
             }
@@ -86,7 +86,8 @@ class SerPredictor(object):
             "class_path": args.ser_dict_path,
         }
 
-        self.preprocess_op = create_operators(pre_process_list, {'infer_mode':True})
+        self.preprocess_op = create_operators(pre_process_list,
+                                              {'infer_mode': True})
         self.postprocess_op = build_post_process(postprocess_params)
         self.predictor, self.input_tensor, self.output_tensors, self.config = \
             utility.create_predictor(args, 'ser', logger)
@@ -113,11 +114,9 @@ class SerPredictor(object):
             output = output_tensor.copy_to_cpu()
             outputs.append(output)
         preds = outputs[0]
-        
+
         post_result = self.postprocess_op(
-            preds,
-            segment_offset_ids=[data[6]],
-            ocr_infos=[data[7]])
+            preds, segment_offset_ids=[data[6]], ocr_infos=[data[7]])
         elapse = time.time() - starttime
         return post_result, elapse
 
@@ -136,17 +135,25 @@ def main(args):
             img, flag = check_and_read_gif(image_file)
             if not flag:
                 img = cv2.imread(image_file)
-                img = img[:,:,::-1]
+                img = img[:, :, ::-1]
             if img is None:
                 logger.info("error in loading image:{}".format(image_file))
                 continue
             ser_res, elapse = ser_predictor(img)
             ser_res = ser_res[0]
 
-            res_str = '{}\t{}\n'.format(image_file,json.dumps({"ocr_info": ser_res,}, ensure_ascii=False))
+            res_str = '{}\t{}\n'.format(
+                image_file,
+                json.dumps(
+                    {
+                        "ocr_info": ser_res,
+                    }, ensure_ascii=False))
             f_w.write(res_str)
 
-            img_res = draw_ser_results(image_file, ser_res,  font_path="../doc/fonts/simfang.ttf",)
+            img_res = draw_ser_results(
+                image_file,
+                ser_res,
+                font_path="../doc/fonts/simfang.ttf", )
 
             img_save_path = os.path.join(args.output,
                                          os.path.basename(image_file))
@@ -157,5 +164,6 @@ def main(args):
             count += 1
             logger.info("Predict time of {}: {}".format(image_file, elapse))
 
+
 if __name__ == "__main__":
     main(parse_args())

From 807dd106361ce97c0f0de59fda6ede5d15413c9d Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Fri, 1 Jul 2022 09:59:11 +0000
Subject: [PATCH 106/134] pre-commit

---
 ppocr/modeling/backbones/vqa_layoutlm.py       | 16 ++++++++--------
 .../vqa_token_ser_layoutlm_postprocess.py      |  3 ++-
 ppstructure/utility.py                         |  3 +--
 tools/export_model.py                          |  3 ++-
 tools/infer/utility.py                         |  2 +-
 tools/infer_vqa_token_ser.py                   | 18 +++++++++---------
 tools/infer_vqa_token_ser_re.py                |  4 ++--
 7 files changed, 25 insertions(+), 24 deletions(-)

diff --git a/ppocr/modeling/backbones/vqa_layoutlm.py b/ppocr/modeling/backbones/vqa_layoutlm.py
index 1542519651..2fd1b1b2a7 100644
--- a/ppocr/modeling/backbones/vqa_layoutlm.py
+++ b/ppocr/modeling/backbones/vqa_layoutlm.py
@@ -121,14 +121,14 @@ class LayoutXLMForSer(NLPBaseModel):
 
     def forward(self, x):
         x = self.model(
-                        input_ids=x[0],
-                        bbox=x[1],
-                        attention_mask=x[2],
-                        token_type_ids=x[3],
-                        image=x[4],
-                        position_ids=None,
-                        head_mask=None,
-                        labels=None)
+            input_ids=x[0],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=x[4],
+            position_ids=None,
+            head_mask=None,
+            labels=None)
         if not self.training:
             return x
         return x[0]
diff --git a/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py b/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
index 90bc52733d..8a6669f71f 100644
--- a/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
+++ b/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
@@ -68,7 +68,8 @@ class VQASerTokenLayoutLMPostProcess(object):
     def _infer(self, preds, segment_offset_ids, ocr_infos):
         results = []
 
-        for pred, segment_offset_id, ocr_info in zip(preds, segment_offset_ids, ocr_infos):
+        for pred, segment_offset_id, ocr_info in zip(preds, segment_offset_ids,
+                                                     ocr_infos):
             pred = np.argmax(pred, axis=1)
             pred = [self.id2label_map[idx] for idx in pred]
 
diff --git a/ppstructure/utility.py b/ppstructure/utility.py
index 7ad3426740..4ae56099b8 100644
--- a/ppstructure/utility.py
+++ b/ppstructure/utility.py
@@ -40,7 +40,6 @@ def init_args():
         type=ast.literal_eval,
         default=None,
         help='label map according to ppstructure/layout/README_ch.md')
-
     # params for vqa
     parser.add_argument("--vqa_algorithm", type=str, default='LayoutXLM')
     parser.add_argument("--ser_model_dir", type=str)
@@ -73,7 +72,7 @@ def init_args():
         "--recovery",
         type=bool,
         default=False,
-        help='Whether to enable layout of recovery')        
+        help='Whether to enable layout of recovery')
     return parser
 
 
diff --git a/tools/export_model.py b/tools/export_model.py
index 752732ff7e..503951a8a6 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -97,8 +97,9 @@ def export_single_model(model,
                 shape=[None, 1, 32, 100], dtype="float32"),
         ]
         model = to_static(model, input_spec=other_shape)
+
     elif arch_config["algorithm"] in ["LayoutLM", "LayoutLMv2", "LayoutXLM"]:
-        input_spec=[
+        input_spec = [
             paddle.static.InputSpec(
                 shape=[None, 512], dtype="int64"),  # input_ids
             paddle.static.InputSpec(
diff --git a/tools/infer/utility.py b/tools/infer/utility.py
index aa5d2371b6..7eb77dec74 100644
--- a/tools/infer/utility.py
+++ b/tools/infer/utility.py
@@ -318,7 +318,7 @@ def create_predictor(args, mode, logger):
         # create predictor
         predictor = inference.create_predictor(config)
         input_names = predictor.get_input_names()
-        if mode in ['ser','re']:
+        if mode in ['ser', 're']:
             input_tensor = []
             for name in input_names:
                 input_tensor.append(predictor.get_input_handle(name))
diff --git a/tools/infer_vqa_token_ser.py b/tools/infer_vqa_token_ser.py
index 1e661ec2c7..0173a554ca 100755
--- a/tools/infer_vqa_token_ser.py
+++ b/tools/infer_vqa_token_ser.py
@@ -44,7 +44,7 @@ def to_tensor(data):
     from collections import defaultdict
     data_dict = defaultdict(list)
     to_tensor_idxs = []
-    
+
     for idx, v in enumerate(data):
         if isinstance(v, (np.ndarray, paddle.Tensor, numbers.Number)):
             if idx not in to_tensor_idxs:
@@ -72,7 +72,10 @@ class SerPredictor(object):
 
         from paddleocr import PaddleOCR
 
-        self.ocr_engine = PaddleOCR(use_angle_cls=False, show_log=False, use_gpu=global_config['use_gpu'])
+        self.ocr_engine = PaddleOCR(
+            use_angle_cls=False,
+            show_log=False,
+            use_gpu=global_config['use_gpu'])
 
         # create data ops
         transforms = []
@@ -82,8 +85,8 @@ class SerPredictor(object):
                 op[op_name]['ocr_engine'] = self.ocr_engine
             elif op_name == 'KeepKeys':
                 op[op_name]['keep_keys'] = [
-                    'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels',
-                    'segment_offset_id', 'ocr_info',
+                    'input_ids', 'bbox', 'attention_mask', 'token_type_ids',
+                    'image', 'labels', 'segment_offset_id', 'ocr_info',
                     'entities'
                 ]
 
@@ -103,11 +106,9 @@ class SerPredictor(object):
         preds = self.model(batch)
         if self.algorithm in ['LayoutLMv2', 'LayoutXLM']:
             preds = preds[0]
-        
+
         post_result = self.post_process_class(
-            preds,
-            segment_offset_ids=batch[6],
-            ocr_infos=batch[7])
+            preds, segment_offset_ids=batch[6], ocr_infos=batch[7])
         return post_result, batch
 
 
@@ -154,4 +155,3 @@ if __name__ == '__main__':
 
             logger.info("process: [{}/{}], save result to {}".format(
                 idx, len(infer_imgs), save_img_path))
-
diff --git a/tools/infer_vqa_token_ser_re.py b/tools/infer_vqa_token_ser_re.py
index d5ae634cea..20ab1fe176 100755
--- a/tools/infer_vqa_token_ser_re.py
+++ b/tools/infer_vqa_token_ser_re.py
@@ -192,6 +192,6 @@ if __name__ == '__main__':
                 }, ensure_ascii=False) + "\n")
             img_res = draw_re_results(img_path, result)
             cv2.imwrite(save_img_path, img_res)
-            
+
             logger.info("process: [{}/{}], save result to {}".format(
-                idx, len(infer_imgs), save_img_path))
\ No newline at end of file
+                idx, len(infer_imgs), save_img_path))

From 8d84144029ac920f3f9b037bec10fc2646031b13 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Fri, 1 Jul 2022 10:00:10 +0000
Subject: [PATCH 107/134] pre-commit

---
 tools/export_model.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/tools/export_model.py b/tools/export_model.py
index 503951a8a6..65573cf46a 100755
--- a/tools/export_model.py
+++ b/tools/export_model.py
@@ -97,7 +97,6 @@ def export_single_model(model,
                 shape=[None, 1, 32, 100], dtype="float32"),
         ]
         model = to_static(model, input_spec=other_shape)
-
     elif arch_config["algorithm"] in ["LayoutLM", "LayoutLMv2", "LayoutXLM"]:
         input_spec = [
             paddle.static.InputSpec(

From 7c4d0d1e9154ebaf44e1ed6284411728db92bce7 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Fri, 1 Jul 2022 18:05:39 +0800
Subject: [PATCH 108/134] fix kl js loss

---
 .../det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml   |  2 +-
 ppocr/losses/basic_loss.py                    | 19 +++++++++++++------
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml b/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml
index e1831f2639..e85127a2ee 100644
--- a/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml
+++ b/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_dml.yml
@@ -60,7 +60,7 @@ Loss:
       - ["Student", "Student2"]
       maps_name: "thrink_maps"
       weight: 1.0
-      act: "softmax"
+      # act: None
       model_name_pairs: ["Student", "Student2"]
       key: maps
   - DistillationDBLoss:
diff --git a/ppocr/losses/basic_loss.py b/ppocr/losses/basic_loss.py
index 2df96ea264..74490791c2 100644
--- a/ppocr/losses/basic_loss.py
+++ b/ppocr/losses/basic_loss.py
@@ -57,17 +57,24 @@ class CELoss(nn.Layer):
 class KLJSLoss(object):
     def __init__(self, mode='kl'):
         assert mode in ['kl', 'js', 'KL', 'JS'
-                        ], "mode can only be one of ['kl', 'js', 'KL', 'JS']"
+                        ], "mode can only be one of ['kl', 'KL', 'js', 'JS']"
         self.mode = mode
 
     def __call__(self, p1, p2, reduction="mean"):
 
-        loss = paddle.multiply(p2, paddle.log((p2 + 1e-5) / (p1 + 1e-5) + 1e-5))
-
-        if self.mode.lower() == "js":
+        if self.mode.lower() == 'kl':
+            loss = paddle.multiply(p2, paddle.log((p2 + 1e-5) / (p1 + 1e-5) + 1e-5))
             loss += paddle.multiply(
-                p1, paddle.log((p1 + 1e-5) / (p2 + 1e-5) + 1e-5))
+                    p1, paddle.log((p1 + 1e-5) / (p2 + 1e-5) + 1e-5))
             loss *= 0.5
+        elif self.mode.lower() == "js":
+            loss = paddle.multiply(p2, paddle.log((2*p2 + 1e-5) / (p1 + p2 + 1e-5) + 1e-5))
+            loss += paddle.multiply(
+                    p1, paddle.log((2*p1 + 1e-5) / (p1 + p2 + 1e-5) + 1e-5))
+            loss *= 0.5
+        else:
+            raise ValueError("The mode.lower() if KLJSLoss should be one of ['kl', 'js']")
+            
         if reduction == "mean":
             loss = paddle.mean(loss, axis=[1, 2])
         elif reduction == "none" or reduction is None:
@@ -95,7 +102,7 @@ class DMLLoss(nn.Layer):
             self.act = None
 
         self.use_log = use_log
-        self.jskl_loss = KLJSLoss(mode="js")
+        self.jskl_loss = KLJSLoss(mode="kl")
 
     def _kldiv(self, x, target):
         eps = 1.0e-10

From 1861d5dda3b19d0b7e636e5df0c227c23887cea1 Mon Sep 17 00:00:00 2001
From: tink2123 <y_tink@163.com>
Date: Fri, 1 Jul 2022 18:22:46 +0800
Subject: [PATCH 109/134] fix inference doc

---
 doc/doc_ch/inference_ppocr.md    | 20 +++++++++++++++++++-
 doc/doc_en/inference_ppocr_en.md | 25 ++++++++++++++++++++++---
 2 files changed, 41 insertions(+), 4 deletions(-)

diff --git a/doc/doc_ch/inference_ppocr.md b/doc/doc_ch/inference_ppocr.md
index 472c0003f9..a9f6e6300a 100644
--- a/doc/doc_ch/inference_ppocr.md
+++ b/doc/doc_ch/inference_ppocr.md
@@ -78,9 +78,27 @@ python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg"
 Predicts of ./doc/imgs_words/ch/word_4.jpg:('实力活力', 0.9956803321838379)
 ```
 
+<a name="英文模型的推理"></a>
+
+### 2.2 英文模型的推理
+
+英文识别模型推理，可以执行如下命令， 注意修改字典路径：
+
+```
+# 下载英文数字识别模型：
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar
+tar xf en_PP-OCRv3_det_infer.tar
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./en_PP-OCRv3_det_infer/" --rec_char_dict_path="ppocr/utils/en_dict.txt"
+```
+
+执行命令后，上图的预测结果为：
+
+```
+Predicts of ./doc/imgs_words/en/word_1.png: ('JOINT', 0.998160719871521)
+```
 <a name="多语言模型的推理"></a>
 
-### 2.2 多语言模型的推理
+### 2.3 多语言模型的推理
 
 如果您需要预测的是其他语言模型，可以在[此链接](./models_list.md#%E5%A4%9A%E8%AF%AD%E8%A8%80%E8%AF%86%E5%88%AB%E6%A8%A1%E5%9E%8B)中找到对应语言的inference模型，在使用inference模型预测时，需要通过`--rec_char_dict_path`指定使用的字典路径, 同时为了得到正确的可视化结果，需要通过 `--vis_font_path` 指定可视化的字体路径，`doc/fonts/` 路径下有默认提供的小语种字体，例如韩文识别：
 ```
diff --git a/doc/doc_en/inference_ppocr_en.md b/doc/doc_en/inference_ppocr_en.md
index 935f92f514..97c176527f 100755
--- a/doc/doc_en/inference_ppocr_en.md
+++ b/doc/doc_en/inference_ppocr_en.md
@@ -8,7 +8,8 @@ This article introduces the use of the Python inference engine for the PP-OCR mo
   - [Text Detection Model Inference](#text-detection-model-inference)
   - [Text Recognition Model Inference](#text-recognition-model-inference)
     - [1. Lightweight Chinese Recognition Model Inference](#1-lightweight-chinese-recognition-model-inference)
-    - [2. Multilingual Model Inference](#2-multilingual-model-inference)
+    - [2. English Recognition Model Inference](#2-english-recognition-model-inference)
+    - [3. Multilingual Model Inference](#3-multilingual-model-inference)
   - [Angle Classification Model Inference](#angle-classification-model-inference)
   - [Text Detection Angle Classification and Recognition Inference Concatenation](#text-detection-angle-classification-and-recognition-inference-concatenation)
 
@@ -76,10 +77,28 @@ After executing the command, the prediction results (recognized text and score)
 ```bash
 Predicts of ./doc/imgs_words_en/word_10.png:('PAIN', 0.988671)
 ```
+<a name="2-english-recognition-model-inference"></a>
+### 2. English Recognition Model Inference
 
-<a name="MULTILINGUAL_MODEL_INFERENCE"></a>
+For English recognition model inference, you can execute the following commands,you need to specify the dictionary path used by `--rec_char_dict_path`:
 
-### 2. Multilingual Model Inference
+```
+# download en model：
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar
+tar xf en_PP-OCRv3_det_infer.tar
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./en_PP-OCRv3_det_infer/" --rec_char_dict_path="ppocr/utils/en_dict.txt"
+```
+
+After executing the command, the prediction result of the above figure is:
+
+```
+Predicts of ./doc/imgs_words/en/word_1.png: ('JOINT', 0.998160719871521)
+```
+
+
+<a name="3-multilingual-model-inference"></a>
+
+### 3. Multilingual Model Inference
 If you need to predict [other language models](./models_list_en.md#Multilingual), when using inference model prediction, you need to specify the dictionary path used by `--rec_char_dict_path`. At the same time, in order to get the correct visualization results,
 You need to specify the visual font path through `--vis_font_path`. There are small language fonts provided by default under the `doc/fonts` path, such as Korean recognition:
 

From 7233ed6c01a356ca253e8bedc030276024640ea2 Mon Sep 17 00:00:00 2001
From: tink2123 <y_tink@163.com>
Date: Fri, 1 Jul 2022 18:25:29 +0800
Subject: [PATCH 110/134] fix inference doc

---
 doc/doc_ch/inference_ppocr.md    | 9 ++++++---
 doc/doc_en/inference_ppocr_en.md | 3 +++
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/doc/doc_ch/inference_ppocr.md b/doc/doc_ch/inference_ppocr.md
index a9f6e6300a..622ac995d3 100644
--- a/doc/doc_ch/inference_ppocr.md
+++ b/doc/doc_ch/inference_ppocr.md
@@ -7,7 +7,8 @@
   - [1. 文本检测模型推理](#1-文本检测模型推理)
   - [2. 文本识别模型推理](#2-文本识别模型推理)
     - [2.1 超轻量中文识别模型推理](#21-超轻量中文识别模型推理)
-    - [2.2 多语言模型的推理](#22-多语言模型的推理)
+    - [2.2 英文识别模型推理](#22-英文识别模型推理)
+    - [2.3 多语言模型的推理](#23-多语言模型的推理)
   - [3. 方向分类模型推理](#3-方向分类模型推理)
   - [4. 文本检测、方向分类和文字识别串联推理](#4-文本检测方向分类和文字识别串联推理)
 
@@ -78,9 +79,9 @@ python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg"
 Predicts of ./doc/imgs_words/ch/word_4.jpg:('实力活力', 0.9956803321838379)
 ```
 
-<a name="英文模型的推理"></a>
+<a name="英文识别模型推理"></a>
 
-### 2.2 英文模型的推理
+### 2.2 英文识别模型推理
 
 英文识别模型推理，可以执行如下命令， 注意修改字典路径：
 
@@ -91,6 +92,8 @@ tar xf en_PP-OCRv3_det_infer.tar
 python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./en_PP-OCRv3_det_infer/" --rec_char_dict_path="ppocr/utils/en_dict.txt"
 ```
 
+![](../imgs_words/en/word_1.png)
+
 执行命令后，上图的预测结果为：
 
 ```
diff --git a/doc/doc_en/inference_ppocr_en.md b/doc/doc_en/inference_ppocr_en.md
index 97c176527f..0f57b0ba6b 100755
--- a/doc/doc_en/inference_ppocr_en.md
+++ b/doc/doc_en/inference_ppocr_en.md
@@ -89,6 +89,9 @@ tar xf en_PP-OCRv3_det_infer.tar
 python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./en_PP-OCRv3_det_infer/" --rec_char_dict_path="ppocr/utils/en_dict.txt"
 ```
 
+![](../imgs_words/en/word_1.png)
+
+
 After executing the command, the prediction result of the above figure is:
 
 ```

From 24712d933c3b3669356a812d177c62a271bc17a4 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Fri, 1 Jul 2022 18:28:02 +0800
Subject: [PATCH 111/134] fix

---
 configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml
index f3ad966488..8b160f6353 100644
--- a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml
@@ -65,7 +65,7 @@ Loss:
       - ["Student", "Teacher"]
       maps_name: "thrink_maps"
       weight: 1.0
-      act: "softmax"
+      # act: None
       model_name_pairs: ["Student", "Teacher"]
       key: maps
   - DistillationDBLoss:

From b541cc12be890212adcd20d5cd99b5767c7fb0f0 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 4 Jul 2022 04:26:54 +0000
Subject: [PATCH 112/134] fix shape unmatch error

---
 ppocr/postprocess/pse_postprocess/pse_postprocess.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/ppocr/postprocess/pse_postprocess/pse_postprocess.py b/ppocr/postprocess/pse_postprocess/pse_postprocess.py
index 34f1b8c9b5..962f3efe92 100755
--- a/ppocr/postprocess/pse_postprocess/pse_postprocess.py
+++ b/ppocr/postprocess/pse_postprocess/pse_postprocess.py
@@ -58,6 +58,8 @@ class PSEPostProcess(object):
 
         kernels = (pred > self.thresh).astype('float32')
         text_mask = kernels[:, 0, :, :]
+        text_mask = paddle.unsqueeze(text_mask, axis=1)
+
         kernels[:, 0:, :, :] = kernels[:, 0:, :, :] * text_mask
 
         score = score.numpy()

From a3c96e6fadcc45ba5a3a9e31579c15300ff55c56 Mon Sep 17 00:00:00 2001
From: andyjpaddle <jiangkaitao@baidu.com>
Date: Mon, 4 Jul 2022 07:08:45 +0000
Subject: [PATCH 113/134] fix east r50 hang

---
 test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt | 4 ++--
 test_tipc/prepare.sh                                          | 4 ++++
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt
index 8477a4fa74..24e4d760c3 100644
--- a/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt
+++ b/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt
@@ -1,13 +1,13 @@
 ===========================train_params===========================
 model_name:det_r50_vd_east_v2_0
 python:python3.7
-gpu_list:0
+gpu_list:0|0,1
 Global.use_gpu:True|True
 Global.auto_cast:fp32
 Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
 Global.save_model_dir:./output/
 Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
-Global.pretrained_model:null
+Global.pretrained_model:./pretrain_models/det_r50_vd_east_v2.0_train/best_accuracy
 train_model_name:latest
 train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
 null:null
diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 2c9bd2901b..073e1c6d1b 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -120,6 +120,10 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar --no-check-certificate
         cd ./pretrain_models/ && tar xf det_mv3_east_v2.0_train.tar && cd ../
     fi
+    if [ ${model_name} == "det_r50_vd_east_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_r50_vd_east_v2.0_train.tar && cd ../
+    fi
 
 elif [ ${MODE} = "whole_train_whole_infer" ];then
     wget -nc -P  ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x0_5_pretrained.pdparams --no-check-certificate

From 8d9cfade49c43dd3c4cdf7fe49fd7da6260944c7 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 4 Jul 2022 08:40:32 +0000
Subject: [PATCH 114/134] add polygons

---
 ppocr/metrics/eval_det_iou.py       | 12 +----
 ppocr/postprocess/db_postprocess.py | 69 ++++++++++++++++++++++++++---
 tools/infer_det.py                  |  6 +--
 3 files changed, 68 insertions(+), 19 deletions(-)

diff --git a/ppocr/metrics/eval_det_iou.py b/ppocr/metrics/eval_det_iou.py
index bc05e7df7d..c144886b3f 100644
--- a/ppocr/metrics/eval_det_iou.py
+++ b/ppocr/metrics/eval_det_iou.py
@@ -83,14 +83,10 @@ class DetectionIoUEvaluator(object):
 
         evaluationLog = ""
 
-        # print(len(gt))
         for n in range(len(gt)):
             points = gt[n]['points']
-            # transcription = gt[n]['text']
             dontCare = gt[n]['ignore']
-            #             points = Polygon(points)
-            #             points = points.buffer(0)
-            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+            if not Polygon(points).is_valid:
                 continue
 
             gtPol = points
@@ -105,9 +101,7 @@ class DetectionIoUEvaluator(object):
 
         for n in range(len(pred)):
             points = pred[n]['points']
-            #             points = Polygon(points)
-            #             points = points.buffer(0)
-            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+            if not Polygon(points).is_valid:
                 continue
 
             detPol = points
@@ -191,8 +185,6 @@ class DetectionIoUEvaluator(object):
         methodHmean = 0 if methodRecall + methodPrecision == 0 else 2 * \
                                                                     methodRecall * methodPrecision / (
                                                                             methodRecall + methodPrecision)
-        # print(methodRecall, methodPrecision, methodHmean)
-        # sys.exit(-1)
         methodMetrics = {
             'precision': methodPrecision,
             'recall': methodRecall,
diff --git a/ppocr/postprocess/db_postprocess.py b/ppocr/postprocess/db_postprocess.py
index 27b428ef2e..c61534d6db 100755
--- a/ppocr/postprocess/db_postprocess.py
+++ b/ppocr/postprocess/db_postprocess.py
@@ -38,6 +38,7 @@ class DBPostProcess(object):
                  unclip_ratio=2.0,
                  use_dilation=False,
                  score_mode="fast",
+                 use_polygon=False,
                  **kwargs):
         self.thresh = thresh
         self.box_thresh = box_thresh
@@ -45,6 +46,7 @@ class DBPostProcess(object):
         self.unclip_ratio = unclip_ratio
         self.min_size = 3
         self.score_mode = score_mode
+        self.use_polygon = use_polygon
         assert score_mode in [
             "slow", "fast"
         ], "Score mode must be in [slow, fast] but got: {}".format(score_mode)
@@ -52,6 +54,56 @@ class DBPostProcess(object):
         self.dilation_kernel = None if not use_dilation else np.array(
             [[1, 1], [1, 1]])
 
+    def polygons_from_bitmap(self, pred, _bitmap, dest_width, dest_height):
+        '''
+        _bitmap: single map with shape (1, H, W),
+            whose values are binarized as {0, 1}
+        '''
+
+        bitmap = _bitmap
+        height, width = bitmap.shape
+
+        boxes = []
+        scores = []
+
+        contours, _ = cv2.findContours((bitmap * 255).astype(np.uint8),
+                                       cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
+
+        for contour in contours[:self.max_candidates]:
+            epsilon = 0.002 * cv2.arcLength(contour, True)
+            approx = cv2.approxPolyDP(contour, epsilon, True)
+            points = approx.reshape((-1, 2))
+            # print(points)
+            if points.shape[0] < 4:
+                continue
+
+            score = self.box_score_fast(pred, points.reshape(-1, 2))
+            if self.box_thresh > score:
+                continue
+
+            if points.shape[0] > 2:
+                box = self.unclip(points, self.unclip_ratio)
+                if len(box) > 1:
+                    continue
+            else:
+                continue
+            box = box.reshape(-1, 2)
+            # print(box)
+
+            _, sside = self.get_mini_boxes(box.reshape((-1, 1, 2)))
+            if sside < self.min_size + 2:
+                continue
+
+            box = np.array(box)
+            box[:, 0] = np.clip(
+                np.round(box[:, 0] / width * dest_width), 0, dest_width)
+            box[:, 1] = np.clip(
+                np.round(box[:, 1] / height * dest_height), 0, dest_height)
+            boxes.append(box.tolist())
+            scores.append(score)
+        # print(boxes)
+        return boxes, scores
+
     def boxes_from_bitmap(self, pred, _bitmap, dest_width, dest_height):
         '''
         _bitmap: single map with shape (1, H, W),
@@ -85,7 +137,7 @@ class DBPostProcess(object):
             if self.box_thresh > score:
                 continue
 
-            box = self.unclip(points).reshape(-1, 1, 2)
+            box = self.unclip(points, self.unclip_ratio).reshape(-1, 1, 2)
             box, sside = self.get_mini_boxes(box)
             if sside < self.min_size + 2:
                 continue
@@ -99,8 +151,7 @@ class DBPostProcess(object):
             scores.append(score)
         return np.array(boxes, dtype=np.int16), scores
 
-    def unclip(self, box):
-        unclip_ratio = self.unclip_ratio
+    def unclip(self, box, unclip_ratio):
         poly = Polygon(box)
         distance = poly.area * unclip_ratio / poly.length
         offset = pyclipper.PyclipperOffset()
@@ -185,8 +236,12 @@ class DBPostProcess(object):
                     self.dilation_kernel)
             else:
                 mask = segmentation[batch_index]
-            boxes, scores = self.boxes_from_bitmap(pred[batch_index], mask,
-                                                   src_w, src_h)
+            if self.use_polygon:
+                boxes, scores = self.polygons_from_bitmap(pred[batch_index],
+                                                          mask, src_w, src_h)
+            else:
+                boxes, scores = self.boxes_from_bitmap(pred[batch_index], mask,
+                                                       src_w, src_h)
 
             boxes_batch.append({'points': boxes})
         return boxes_batch
@@ -202,6 +257,7 @@ class DistillationDBPostProcess(object):
                  unclip_ratio=1.5,
                  use_dilation=False,
                  score_mode="fast",
+                 use_polygon=False,
                  **kwargs):
         self.model_name = model_name
         self.key = key
@@ -211,7 +267,8 @@ class DistillationDBPostProcess(object):
             max_candidates=max_candidates,
             unclip_ratio=unclip_ratio,
             use_dilation=use_dilation,
-            score_mode=score_mode)
+            score_mode=score_mode,
+            use_polygon=use_polygon)
 
     def __call__(self, predicts, shape_list):
         results = {}
diff --git a/tools/infer_det.py b/tools/infer_det.py
index 1acecedf3e..df34652389 100755
--- a/tools/infer_det.py
+++ b/tools/infer_det.py
@@ -44,7 +44,7 @@ def draw_det_res(dt_boxes, config, img, img_name, save_path):
         import cv2
         src_im = img
         for box in dt_boxes:
-            box = box.astype(np.int32).reshape((-1, 1, 2))
+            box = np.array(box).astype(np.int32).reshape((-1, 1, 2))
             cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
         if not os.path.exists(save_path):
             os.makedirs(save_path)
@@ -106,7 +106,7 @@ def main():
                     dt_boxes_list = []
                     for box in boxes:
                         tmp_json = {"transcription": ""}
-                        tmp_json['points'] = box.tolist()
+                        tmp_json['points'] = list(box)
                         dt_boxes_list.append(tmp_json)
                     det_box_json[k] = dt_boxes_list
                     save_det_path = os.path.dirname(config['Global'][
@@ -118,7 +118,7 @@ def main():
                 # write result
                 for box in boxes:
                     tmp_json = {"transcription": ""}
-                    tmp_json['points'] = box.tolist()
+                    tmp_json['points'] = list(box)
                     dt_boxes_json.append(tmp_json)
                 save_det_path = os.path.dirname(config['Global'][
                     'save_res_path']) + "/det_results/"

From 72f93cf977ffed232e85c1736ba4355c7afb3fd7 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Mon, 4 Jul 2022 08:49:34 +0000
Subject: [PATCH 115/134] add polygons

---
 ppocr/postprocess/db_postprocess.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ppocr/postprocess/db_postprocess.py b/ppocr/postprocess/db_postprocess.py
index c61534d6db..e1041b00f9 100755
--- a/ppocr/postprocess/db_postprocess.py
+++ b/ppocr/postprocess/db_postprocess.py
@@ -236,7 +236,7 @@ class DBPostProcess(object):
                     self.dilation_kernel)
             else:
                 mask = segmentation[batch_index]
-            if self.use_polygon:
+            if self.use_polygon is True:
                 boxes, scores = self.polygons_from_bitmap(pred[batch_index],
                                                           mask, src_w, src_h)
             else:

From d1b31bf85cd6e23c8a09758ea6ca03938209ed69 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Mon, 4 Jul 2022 10:22:25 +0000
Subject: [PATCH 116/134] add ref

---
 configs/table/table_master.yml                |  64 +++++-----
 doc/doc_ch/algorithm_det_fcenet.md            |  22 ++--
 doc/doc_ch/algorithm_overview.md              |  16 ++-
 doc/doc_ch/algorithm_table_master.md          | 114 ++++++++++++++++++
 doc/doc_en/algorithm_overview_en.md           |  13 ++
 doc/doc_en/algorithm_table_master_en.md       | 112 +++++++++++++++++
 ppocr/data/imaug/label_ops.py                 |   9 ++
 ppocr/losses/table_master_loss.py             |   5 +
 .../modeling/backbones/table_master_resnet.py |   4 +
 ppocr/modeling/heads/table_master_head.py     |   5 +
 ppocr/postprocess/__init__.py                 |   2 +-
 ppstructure/docs/models_list.md               |   2 +-
 12 files changed, 319 insertions(+), 49 deletions(-)
 create mode 100644 doc/doc_ch/algorithm_table_master.md
 create mode 100644 doc/doc_en/algorithm_table_master_en.md

diff --git a/configs/table/table_master.yml b/configs/table/table_master.yml
index 9dfc0e2746..cfd98346c4 100755
--- a/configs/table/table_master.yml
+++ b/configs/table/table_master.yml
@@ -2,21 +2,19 @@ Global:
   use_gpu: true
   epoch_num: 17
   log_smooth_window: 20
-  print_batch_step: 5
+  print_batch_step: 100
   save_model_dir: ./output/table_master/
   save_epoch_step: 17
-  # evaluation is run every 400 iterations after the 0th iteration
-  eval_batch_step: [0, 400]
-  cal_metric_during_train: True
-  pretrained_model:
-  checkpoints:
-  save_inference_dir:
-  use_visualdl: False
+  eval_batch_step: [0,  6259]
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: 
+  save_inference_dir: output/table_master/infer
+  use_visualdl: false
   infer_img: ppstructure/docs/table/table.jpg
-  save_res_path: output/table_master
-  # for data or label process
+  save_res_path: ./output/table_master
   character_dict_path: ppocr/utils/dict/table_master_structure_dict.txt
-  infer_mode: False
+  infer_mode: false
   max_text_length: 500
   process_total_num: 0
   process_cut_num: 0
@@ -33,8 +31,8 @@ Optimizer:
     gamma: 0.1
     warmup_epoch: 0.02
   regularizer:
-    name: 'L2'
-    factor: 0.00000
+    name: L2
+    factor: 0.0
 
 Architecture:
   model_type: table
@@ -67,15 +65,15 @@ PostProcess:
 Metric:
   name: TableMetric
   main_indicator: acc
-  compute_bbox_metric: true # cost many time, set False for training
+  compute_bbox_metric: False
 
 Train:
   dataset:
     name: PubTabDataSet
-    data_dir: /home/zhoujun20/table/PubTabNe/pubtabnet/train/
-    label_file_list: [/home/zhoujun20/table/PubTabNe/pubtabnet/PubTabNet_2.0.0_train.jsonl]
+    data_dir: train_data/table/pubtabnet/train/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl]
     transforms:
-      - DecodeImage: # load image
+      - DecodeImage:
           img_mode: BGR
           channel_first: False
       - TableMasterLabelEncode:
@@ -88,20 +86,20 @@ Train:
       - PaddingTableImage:
           size: [480, 480]
       - TableBoxEncode:
-          use_xywh: true
+          use_xywh: True
       - NormalizeImage:
           scale: 1./255.
           mean: [0.5, 0.5, 0.5]
           std: [0.5, 0.5, 0.5]
-          order: 'hwc'
-      - ToCHWImage:
+          order: hwc
+      - ToCHWImage: null
       - KeepKeys:
-          keep_keys: ['image', 'structure', 'bboxes', 'bbox_masks','shape']
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
   loader:
     shuffle: True
-    batch_size_per_card: 8
+    batch_size_per_card: 10
     drop_last: True
-    num_workers: 1
+    num_workers: 8
 
 Eval:
   dataset:
@@ -109,7 +107,7 @@ Eval:
     data_dir: /home/zhoujun20/table/PubTabNe/pubtabnet/val/
     label_file_list: [/home/zhoujun20/table/PubTabNe/pubtabnet/val_500.jsonl]
     transforms:
-      - DecodeImage: # load image
+      - DecodeImage:
           img_mode: BGR
           channel_first: False
       - TableMasterLabelEncode:
@@ -120,19 +118,19 @@ Eval:
           max_len: 480
           resize_bboxes: True
       - PaddingTableImage:
-          size: [ 480, 480 ]
+          size: [480, 480]
       - TableBoxEncode:
-          use_xywh: true
+          use_xywh: True
       - NormalizeImage:
           scale: 1./255.
-          mean: [ 0.5, 0.5, 0.5 ]
-          std: [ 0.5, 0.5, 0.5 ]
-          order: 'hwc'
-      - ToCHWImage:
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
       - KeepKeys:
-          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks','shape' ]
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
   loader:
     shuffle: False
     drop_last: False
-    batch_size_per_card: 2
-    num_workers: 8
+    batch_size_per_card: 10
+    num_workers: 8
\ No newline at end of file
diff --git a/doc/doc_ch/algorithm_det_fcenet.md b/doc/doc_ch/algorithm_det_fcenet.md
index bd2e734204..a70caa29fb 100644
--- a/doc/doc_ch/algorithm_det_fcenet.md
+++ b/doc/doc_ch/algorithm_det_fcenet.md
@@ -1,17 +1,15 @@
 # FCENet
 
-- [1. 算法简介](#1)
-- [2. 环境配置](#2)
-- [3. 模型训练、评估、预测](#3)
-    - [3.1 训练](#3-1)
-    - [3.2 评估](#3-2)
-    - [3.3 预测](#3-3)
-- [4. 推理部署](#4)
-    - [4.1 Python推理](#4-1)
-    - [4.2 C++推理](#4-2)
-    - [4.3 Serving服务化部署](#4-3)
-    - [4.4 更多推理部署](#4-4)
-- [5. FAQ](#5)
+- [1. 算法简介](#1-算法简介)
+- [2. 环境配置](#2-环境配置)
+- [3. 模型训练、评估、预测](#3-模型训练评估预测)
+- [4. 推理部署](#4-推理部署)
+  - [4.1 Python推理](#41-python推理)
+  - [4.2 C++推理](#42-c推理)
+  - [4.3 Serving服务化部署](#43-serving服务化部署)
+  - [4.4 更多推理部署](#44-更多推理部署)
+- [5. FAQ](#5-faq)
+- [引用](#引用)
 
 <a name="1"></a>
 ## 1. 算法简介
diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index ef96f6ec12..84af5fdcce 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -1,9 +1,10 @@
 # OCR算法
 
 - [1. 两阶段算法](#1-两阶段算法)
-    - [1.1 文本检测算法](#11-文本检测算法)
-    - [1.2 文本识别算法](#12-文本识别算法)
+  - [1.1 文本检测算法](#11-文本检测算法)
+  - [1.2 文本识别算法](#12-文本识别算法)
 - [2. 端到端算法](#2-端到端算法)
+- [3. 表格识别算法](#3-表格识别算法)
 
 
 本文给出了PaddleOCR已支持的OCR算法列表，以及每个算法在**英文公开数据集**上的模型和指标，主要用于算法简介和算法性能对比，更多包括中文在内的其他数据集上的模型请参考[PP-OCR v2.0 系列模型下载](./models_list.md)。
@@ -96,3 +97,14 @@
 
 已支持的端到端OCR算法列表（戳链接获取使用教程）：
 - [x]  [PGNet](./algorithm_e2e_pgnet.md)
+
+## 3. 表格识别算法
+
+已支持的表格识别算法列表（戳链接获取使用教程）：
+- [x]  [TableMaster](./algorithm_table_master.md)
+
+在PubTabNet表格识别公开数据集上，算法效果如下：
+
+|模型|骨干网络|配置文件|acc|下载链接|
+|---|---|---|---|---|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[训练模型]|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
diff --git a/doc/doc_ch/algorithm_table_master.md b/doc/doc_ch/algorithm_table_master.md
new file mode 100644
index 0000000000..6bfd0f36fc
--- /dev/null
+++ b/doc/doc_ch/algorithm_table_master.md
@@ -0,0 +1,114 @@
+# 表格识别算法-TableMASTER
+
+- [1. 算法简介](#1-算法简介)
+- [2. 环境配置](#2-环境配置)
+- [3. 模型训练、评估、预测](#3-模型训练评估预测)
+- [4. 推理部署](#4-推理部署)
+  - [4.1 Python推理](#41-python推理)
+  - [4.2 C++推理部署](#42-c推理部署)
+  - [4.3 Serving服务化部署](#43-serving服务化部署)
+  - [4.4 更多推理部署](#44-更多推理部署)
+- [5. FAQ](#5-faq)
+- [引用](#引用)
+
+<a name="1"></a>
+## 1. 算法简介
+
+论文信息：
+> [TableMaster: PINGAN-VCGROUP’S SOLUTION FOR ICDAR 2021 COMPETITION ON SCIENTIFIC LITERATURE PARSING TASK B: TABLE RECOGNITION TO HTML](https://arxiv.org/pdf/2105.01848.pdf)
+> Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong
+> 2021
+
+在PubTabNet表格识别公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|acc|下载链接|
+| --- | --- | --- | --- | --- |
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+
+
+<a name="2"></a>
+## 2. 环境配置
+请先参考[《运行环境准备》](./environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](./clone.md)克隆项目代码。
+
+
+<a name="3"></a>
+## 3. 模型训练、评估、预测
+
+上述TableMaster模型使用PubTabNet表格识别公开数据集训练得到，数据集下载可参考 [table_datasets](./dataset/table_datasets.md)。
+
+数据下载完成后，请参考[文本识别教程](./recognition.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的模型只需要**更换配置文件**即可。
+
+<a name="4"></a>
+## 4. 推理部署
+
+<a name="4-1"></a>
+### 4.1 Python推理
+首先将训练得到best模型，转换成inference model。以基于TableResNetExtra骨干网络，在PubTabNet数据集训练的模型为例([模型下载地址](https://paddleocr.bj.bcebos.com/contribution/table_master.tar))，可以使用如下命令进行转换：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/table/table_master.yml -o Global.pretrained_model=output/table_master/best_accuracy Global.save_inference_dir=./inference/table_master
+```
+
+**注意：**
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
+
+转换成功后，在目录下有三个文件：
+```
+/inference/table_master/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+
+执行如下命令进行模型推理：
+
+```shell
+cd ppstructure/
+python3.7 table/predict_structure.py --table_model_dir=../output/table_master/table_structure_tablemaster_infer/ --table_algorithm=TableMaster --table_char_dict_path=../ppocr/utils/dict/table_master_structure_dict.txt --table_max_len=480 --image_dir=docs/table/table.jpg
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='docs/table'。
+```
+
+执行命令后，上面图像的预测结果（结构信息和表格中每个单元格的坐标）会打印到屏幕上，同时会保存单元格坐标的可视化结果。示例如下：
+结果如下：
+```shell
+[2022/06/16 13:06:54] ppocr INFO: result: ['<html>', '<body>', '<table>', '<thead>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</thead>', '<tbody>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</tbody>', '</table>', '</body>', '</html>'], [[72.17591094970703, 10.759100914001465, 60.29658508300781, 16.6805362701416], [161.85562133789062, 10.884308815002441, 14.9495210647583, 16.727018356323242], [277.79876708984375, 29.54340362548828, 31.490320205688477, 18.143272399902344],
+...
+[336.11724853515625, 280.3601989746094, 39.456939697265625, 18.121286392211914]]
+[2022/06/16 13:06:54] ppocr INFO: save vis result to ./output/table.jpg
+[2022/06/16 13:06:54] ppocr INFO: Predict time of docs/table/table.jpg: 17.36806297302246
+```
+
+**注意**：
+
+- TableMaster在推理时比较慢，建议使用GPU进行使用。
+
+<a name="4-2"></a>
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持TableMaster，所以暂未支持
+
+<a name="4-3"></a>
+### 4.3 Serving服务化部署
+
+暂不支持
+
+<a name="4-4"></a>
+### 4.4 更多推理部署
+
+暂不支持
+
+<a name="5"></a>
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{ye2021pingan,
+  title={PingAn-VCGroup's Solution for ICDAR 2021 Competition on Scientific Literature Parsing Task B: Table Recognition to HTML},
+  author={Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong},
+  journal={arXiv preprint arXiv:2105.01848},
+  year={2021}
+}
+```
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index bc96cdf235..cd277c74a3 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -4,6 +4,7 @@
   * [1.1 Text Detection Algorithms](#11)
   * [1.2 Text Recognition Algorithms](#12)
 - [2. End-to-end Algorithms](#2)
+- [3. Table Recognition Algorithms](#3)
 
 
 This tutorial lists the OCR algorithms supported by PaddleOCR, as well as the models and metrics of each algorithm on **English public datasets**. It is mainly used for algorithm introduction and algorithm performance comparison. For more models on other datasets including Chinese, please refer to [PP-OCR v2.0 models list](./models_list_en.md).
@@ -95,3 +96,15 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 
 Supported end-to-end algorithms (Click the link to get the tutorial):
 - [x]  [PGNet](./algorithm_e2e_pgnet_en.md)
+
+<a name="3"></a>
+## 3. Table Recognition Algorithms
+
+Supported table recognition algorithms (Click the link to get the tutorial):
+- [x]  [TableMaster](./algorithm_table_master_en.md)
+
+On the PubTabNet dataset, the algorithm result is as follows:
+
+|Model|Backbone|Config|Acc|Download link|
+|---|---|---|---|---|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[训练模型]|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
diff --git a/doc/doc_en/algorithm_table_master_en.md b/doc/doc_en/algorithm_table_master_en.md
new file mode 100644
index 0000000000..a557a60901
--- /dev/null
+++ b/doc/doc_en/algorithm_table_master_en.md
@@ -0,0 +1,112 @@
+# Torm Recognition Algorithm-TableMASTER
+
+- [1. Introduction](#1-introduction)
+- [2. Environment](#2-environment)
+- [3. Model Training / Evaluation / Prediction](#3-model-training--evaluation--prediction)
+- [4. Inference and Deployment](#4-inference-and-deployment)
+  - [4.1 Python Inference](#41-python-inference)
+  - [4.2 C++ Inference](#42-c-inference)
+  - [4.3 Serving](#43-serving)
+  - [4.4 More](#44-more)
+- [5. FAQ](#5-faq)
+- [Citation](#citation)
+
+<a name="1"></a>
+## 1. Introduction
+
+Paper:
+> [TableMaster: PINGAN-VCGROUP’S SOLUTION FOR ICDAR 2021 COMPETITION ON SCIENTIFIC LITERATURE PARSING TASK B: TABLE RECOGNITION TO HTML](https://arxiv.org/pdf/2105.01848.pdf)
+> Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong
+> 2021
+
+
+On the PubTabNet table recognition public data set, the algorithm reproduction acc is as follows:
+
+|Model|Backbone|Cnnfig|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[train model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+
+
+<a name="2"></a>
+## 2. Environment
+Please refer to ["Environment Preparation"](./environment_en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](./clone_en.md) to clone the project code.
+
+
+<a name="3"></a>
+## 3. Model Training / Evaluation / Prediction
+
+The above TableMaster model is trained using the PubTabNet table recognition public dataset. For the download of the dataset, please refer to [table_datasets](./dataset/table_datasets_en.md).
+
+After the data download is complete, please refer to [Text Recognition Training Tutorial](./recognition_en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different models.
+
+
+<a name="4"></a>
+## 4. Inference and Deployment
+
+<a name="4-1"></a>
+### 4.1 Python Inference
+
+First, convert the model saved in the TableMaster table recognition training process into an inference model. Taking the model based on the TableResNetExtra backbone network and trained on the PubTabNet dataset as example ([model download link](https://paddleocr.bj.bcebos.com/contribution/table_master.tar)), you can use the following command to convert:
+
+
+```shell
+python3 tools/export_model.py -c configs/table/table_master.yml -o Global.pretrained_model=output/table_master/best_accuracy Global.save_inference_dir=./inference/table_master
+```
+
+**Note: **
+- If you trained the model on your own dataset and adjusted the dictionary file, please pay attention to whether the `character_dict_path` in the modified configuration file is the correct dictionary file
+
+
+Execute the following command for model inference:
+
+```shell
+cd ppstructure/
+# When predicting all images in a folder, you can modify image_dir to a folder, such as --image_dir='docs/table'.
+python3.7 table/predict_structure.py --table_model_dir=../output/table_master/table_structure_tablemaster_infer/ --table_algorithm=TableMaster --table_char_dict_path=../ppocr/utils/dict/table_master_structure_dict.txt --table_max_len=480 --image_dir=docs/table/table.jpg
+
+```
+
+After executing the command, the prediction results of the above image (structural information and the coordinates of each cell in the table) are printed to the screen, and the visualization of the cell coordinates is also saved. An example is as follows:
+
+result：
+```shell
+[2022/06/16 13:06:54] ppocr INFO: result: ['<html>', '<body>', '<table>', '<thead>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</thead>', '<tbody>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</tbody>', '</table>', '</body>', '</html>'], [[72.17591094970703, 10.759100914001465, 60.29658508300781, 16.6805362701416], [161.85562133789062, 10.884308815002441, 14.9495210647583, 16.727018356323242], [277.79876708984375, 29.54340362548828, 31.490320205688477, 18.143272399902344],
+...
+[336.11724853515625, 280.3601989746094, 39.456939697265625, 18.121286392211914]]
+[2022/06/16 13:06:54] ppocr INFO: save vis result to ./output/table.jpg
+[2022/06/16 13:06:54] ppocr INFO: Predict time of docs/table/table.jpg: 17.36806297302246
+```
+
+**Note**:
+
+- TableMaster is relatively slow during inference, and it is recommended to use GPU for use.
+
+<a name="4-2"></a>
+### 4.2 C++ Inference
+
+Since the post-processing is not written in CPP, the TableMaster does not support CPP inference.
+
+
+<a name="4-3"></a>
+### 4.3 Serving
+
+Not supported
+
+<a name="4-4"></a>
+### 4.4 More
+
+Not supported
+
+<a name="5"></a>
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{ye2021pingan,
+  title={PingAn-VCGroup's Solution for ICDAR 2021 Competition on Scientific Literature Parsing Task B: Table Recognition to HTML},
+  author={Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong},
+  journal={arXiv preprint arXiv:2105.01848},
+  year={2021}
+}
+```
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 007927011a..96f3c98674 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -670,6 +670,10 @@ class TableLabelEncode(AttnLabelEncode):
         return data
 
     def _merge_no_span_structure(self, structure):
+        """
+        This fun code is refer from:
+        https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/table_recognition/data_preprocess.py
+        """
         new_structure = []
         i = 0
         while i < len(structure):
@@ -682,6 +686,11 @@ class TableLabelEncode(AttnLabelEncode):
         return new_structure
 
     def _replace_empty_cell_token(self, token_list, cells):
+        """
+        This fun code is refer from:
+        https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/table_recognition/data_preprocess.py
+        """
+
         bbox_idx = 0
         add_empty_bbox_token_list = []
         for token in token_list:
diff --git a/ppocr/losses/table_master_loss.py b/ppocr/losses/table_master_loss.py
index 50a773dd9a..216c4e521a 100644
--- a/ppocr/losses/table_master_loss.py
+++ b/ppocr/losses/table_master_loss.py
@@ -11,6 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This fun code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/tree/master/mmocr/models/textrecog/losses
+"""
+
 import paddle
 from paddle import nn
 
diff --git a/ppocr/modeling/backbones/table_master_resnet.py b/ppocr/modeling/backbones/table_master_resnet.py
index 82b4f37a74..f1c506ca45 100644
--- a/ppocr/modeling/backbones/table_master_resnet.py
+++ b/ppocr/modeling/backbones/table_master_resnet.py
@@ -11,6 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This fun code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/mmocr/models/textrecog/backbones/table_resnet_extra.py
+"""
 
 import paddle
 import paddle.nn as nn
diff --git a/ppocr/modeling/heads/table_master_head.py b/ppocr/modeling/heads/table_master_head.py
index 4da6e9b59f..887630a87e 100644
--- a/ppocr/modeling/heads/table_master_head.py
+++ b/ppocr/modeling/heads/table_master_head.py
@@ -11,6 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+This fun code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/mmocr/models/textrecog/decoders/master_decoder.py
+"""
+
 import copy
 import math
 import paddle
diff --git a/ppocr/postprocess/__init__.py b/ppocr/postprocess/__init__.py
index 26a23f1ea4..1d414eb2e8 100644
--- a/ppocr/postprocess/__init__.py
+++ b/ppocr/postprocess/__init__.py
@@ -26,7 +26,7 @@ from .east_postprocess import EASTPostProcess
 from .sast_postprocess import SASTPostProcess
 from .fce_postprocess import FCEPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, \
-    DistillationCTCLabelDecode, TableLabelDecode, NRTRLabelDecode, SARLabelDecode, \
+    DistillationCTCLabelDecode, NRTRLabelDecode, SARLabelDecode, \
     SEEDLabelDecode, PRENLabelDecode, ViTSTRLabelDecode, ABINetLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
diff --git a/ppstructure/docs/models_list.md b/ppstructure/docs/models_list.md
index dabce3a514..42d44009da 100644
--- a/ppstructure/docs/models_list.md
+++ b/ppstructure/docs/models_list.md
@@ -35,7 +35,7 @@
 
 |模型名称|模型简介|推理模型大小|下载地址|
 | --- | --- | --- | --- |
-|en_ppocr_mobile_v2.0_table_structure|PubLayNet数据集训练的英文表格场景的表格结构预测|18.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) |
+|en_ppocr_mobile_v2.0_table_structure|PubTabNet数据集训练的英文表格场景的表格结构预测|18.6M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) |
 
 <a name="3"></a>
 ## 3. VQA模型

From bedd89f916fb24ca349bfe0abdf593dee3aec22e Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Mon, 4 Jul 2022 20:04:57 +0800
Subject: [PATCH 117/134] fix bug

---
 deploy/lite/crnn_process.cc | 3 ++-
 deploy/lite/ocr_db_crnn.cc  | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/deploy/lite/crnn_process.cc b/deploy/lite/crnn_process.cc
index 6d5fc1504e..7cd95cddb5 100644
--- a/deploy/lite/crnn_process.cc
+++ b/deploy/lite/crnn_process.cc
@@ -34,12 +34,13 @@ cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height) {
     resize_w = imgW;
   else
     resize_w = int(ceilf(imgH * ratio));
-
+  cv::Mat resize_img;
   cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
              cv::INTER_LINEAR);
   cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
                      int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
                      {127, 127, 127});
+  return resize_img;
 }
 
 std::vector<std::string> ReadDict(std::string path) {
diff --git a/deploy/lite/ocr_db_crnn.cc b/deploy/lite/ocr_db_crnn.cc
index cb2bf7791a..fde0d07d6c 100644
--- a/deploy/lite/ocr_db_crnn.cc
+++ b/deploy/lite/ocr_db_crnn.cc
@@ -474,7 +474,7 @@ void system(char **argv){
   
     std::vector<double> rec_times;
     RunRecModel(boxes, srcimg, rec_predictor, rec_text, rec_text_score,
-                charactor_dict, cls_predictor, use_direction_classify, &rec_times);
+                charactor_dict, cls_predictor, use_direction_classify, &rec_times, rec_image_height);
   
     //// visualization
     auto img_vis = Visualization(srcimg, boxes);

From 4ec14d0e8f5539f42b3c7b527e6a5792752a23d1 Mon Sep 17 00:00:00 2001
From: Liliya Wu <94680624+Liliyaw@users.noreply.github.com>
Date: Mon, 4 Jul 2022 21:33:38 +0800
Subject: [PATCH 118/134] Update run_demo.sh

---
 deploy/avh/run_demo.sh | 1 -
 1 file changed, 1 deletion(-)

diff --git a/deploy/avh/run_demo.sh b/deploy/avh/run_demo.sh
index 01f950c13d..63de91afc8 100755
--- a/deploy/avh/run_demo.sh
+++ b/deploy/avh/run_demo.sh
@@ -1,5 +1,4 @@
 #!/bin/bash
-# Copyright (c) 2022 Arm Limited and Contributors. All rights reserved.
 # Licensed to the Apache Software Foundation (ASF) under one
 # or more contributor license agreements.  See the NOTICE file
 # distributed with this work for additional information

From adda5ddce933b9973d1c9b5296d15d437d266290 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Tue, 5 Jul 2022 02:51:07 +0000
Subject: [PATCH 119/134] fix bug

---
 doc/doc_ch/algorithm_overview.md              |  2 +-
 doc/doc_ch/algorithm_table_master.md          |  2 +-
 doc/doc_en/algorithm_overview_en.md           | 12 ++++----
 doc/doc_en/algorithm_table_master_en.md       |  4 +--
 ppocr/data/imaug/label_ops.py                 |  2 +-
 ppocr/losses/table_master_loss.py             |  2 +-
 ppocr/metrics/table_metric.py                 |  4 ---
 .../modeling/backbones/table_master_resnet.py | 28 ++++++++-----------
 ppocr/modeling/heads/table_master_head.py     |  4 +--
 ppocr/utils/visual.py                         | 13 +++++++++
 ppstructure/table/predict_structure.py        | 14 +---------
 tools/infer_table.py                          | 14 +---------
 12 files changed, 41 insertions(+), 60 deletions(-)

diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index 84af5fdcce..1efd564ce1 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -107,4 +107,4 @@
 
 |模型|骨干网络|配置文件|acc|下载链接|
 |---|---|---|---|---|
-|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[训练模型]|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar) / [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
diff --git a/doc/doc_ch/algorithm_table_master.md b/doc/doc_ch/algorithm_table_master.md
index 6bfd0f36fc..36455ed9f9 100644
--- a/doc/doc_ch/algorithm_table_master.md
+++ b/doc/doc_ch/algorithm_table_master.md
@@ -55,7 +55,7 @@ python3 tools/export_model.py -c configs/table/table_master.yml -o Global.pretra
 
 转换成功后，在目录下有三个文件：
 ```
-/inference/table_master/
+./inference/table_master/
     ├── inference.pdiparams         # 识别inference模型的参数文件
     ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
     └── inference.pdmodel           # 识别inference模型的program文件
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index cd277c74a3..8b9b3f7b0b 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -1,10 +1,10 @@
 # OCR Algorithms
 
-- [1. Two-stage Algorithms](#1)
-  * [1.1 Text Detection Algorithms](#11)
-  * [1.2 Text Recognition Algorithms](#12)
-- [2. End-to-end Algorithms](#2)
-- [3. Table Recognition Algorithms](#3)
+- [1. Two-stage Algorithms](#1-two-stage-algorithms)
+  - [1.1 Text Detection Algorithms](#11-text-detection-algorithms)
+  - [1.2 Text Recognition Algorithms](#12-text-recognition-algorithms)
+- [2. End-to-end Algorithms](#2-end-to-end-algorithms)
+- [3. Table Recognition Algorithms](#3-table-recognition-algorithms)
 
 
 This tutorial lists the OCR algorithms supported by PaddleOCR, as well as the models and metrics of each algorithm on **English public datasets**. It is mainly used for algorithm introduction and algorithm performance comparison. For more models on other datasets including Chinese, please refer to [PP-OCR v2.0 models list](./models_list_en.md).
@@ -107,4 +107,4 @@ On the PubTabNet dataset, the algorithm result is as follows:
 
 |Model|Backbone|Config|Acc|Download link|
 |---|---|---|---|---|
-|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[训练模型]|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[trained](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar) / [inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
diff --git a/doc/doc_en/algorithm_table_master_en.md b/doc/doc_en/algorithm_table_master_en.md
index a557a60901..e9249a2a05 100644
--- a/doc/doc_en/algorithm_table_master_en.md
+++ b/doc/doc_en/algorithm_table_master_en.md
@@ -1,4 +1,4 @@
-# Torm Recognition Algorithm-TableMASTER
+# Table Recognition Algorithm-TableMASTER
 
 - [1. Introduction](#1-introduction)
 - [2. Environment](#2-environment)
@@ -24,7 +24,7 @@ On the PubTabNet table recognition public data set, the algorithm reproduction a
 
 |Model|Backbone|Cnnfig|Acc|Download link|
 | --- | --- | --- | --- | --- |
-|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[train model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](../../configs/table/table_master.yml)|77.47%|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
 
 
 <a name="2"></a>
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 96f3c98674..7cc4cef46a 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -671,7 +671,7 @@ class TableLabelEncode(AttnLabelEncode):
 
     def _merge_no_span_structure(self, structure):
         """
-        This fun code is refer from:
+        This code is refer from:
         https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/table_recognition/data_preprocess.py
         """
         new_structure = []
diff --git a/ppocr/losses/table_master_loss.py b/ppocr/losses/table_master_loss.py
index 216c4e521a..dca982dbd4 100644
--- a/ppocr/losses/table_master_loss.py
+++ b/ppocr/losses/table_master_loss.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-This fun code is refer from:
+This code is refer from:
 https://github.com/JiaquanYe/TableMASTER-mmocr/tree/master/mmocr/models/textrecog/losses
 """
 
diff --git a/ppocr/metrics/table_metric.py b/ppocr/metrics/table_metric.py
index fb0075f7cb..fd2631e442 100644
--- a/ppocr/metrics/table_metric.py
+++ b/ppocr/metrics/table_metric.py
@@ -31,8 +31,6 @@ class TableStructureMetric(object):
                                              gt_structure_batch_list):
             pred_str = ''.join(pred)
             target_str = ''.join(target)
-            # pred_str = pred_str.replace('<thead>','').replace('</thead>','').replace('<tbody>','').replace('</tbody>','')
-            # target_str = target_str.replace('<thead>','').replace('</thead>','').replace('<tbody>','').replace('</tbody>','')
             if pred_str == target_str:
                 correct_num += 1
             all_num += 1
@@ -55,8 +53,6 @@ class TableStructureMetric(object):
         self.len_acc_num = 0
         self.token_nums = 0
         self.anys_dict = dict()
-        from collections import defaultdict
-        self.error_num_dict = defaultdict(int)
 
 
 class TableMetric(object):
diff --git a/ppocr/modeling/backbones/table_master_resnet.py b/ppocr/modeling/backbones/table_master_resnet.py
index f1c506ca45..dacf5ed26e 100644
--- a/ppocr/modeling/backbones/table_master_resnet.py
+++ b/ppocr/modeling/backbones/table_master_resnet.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-This fun code is refer from:
+This code is refer from:
 https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/mmocr/models/textrecog/backbones/table_resnet_extra.py
 """
 
@@ -193,47 +193,43 @@ class TableResNetExtra(nn.Layer):
 
     def forward(self, x):
         f = []
-        x = self.conv1(x)  # 1,64,480,480
+        x = self.conv1(x)
 
         x = self.bn1(x)
         x = self.relu1(x)
 
-        x = self.conv2(x)  # 1,128,480,480
+        x = self.conv2(x)
         x = self.bn2(x)
         x = self.relu2(x)
-        # (48, 160)
 
-        x = self.maxpool1(x)  # 1,64,240,240
+        x = self.maxpool1(x)
         x = self.layer1(x)
 
-        x = self.conv3(x)  # 1,256,240,240
+        x = self.conv3(x)
         x = self.bn3(x)
         x = self.relu3(x)
         f.append(x)
-        # (24, 80)
 
-        x = self.maxpool2(x)  # 1,256,120,120
+        x = self.maxpool2(x)
         x = self.layer2(x)
 
-        x = self.conv4(x)  # 1,256,120,120
+        x = self.conv4(x)
         x = self.bn4(x)
         x = self.relu4(x)
         f.append(x)
-        # (12, 40)
 
-        x = self.maxpool3(x)  # 1,256,60,60
+        x = self.maxpool3(x)
 
-        x = self.layer3(x)  # 1,512,60,60
-        x = self.conv5(x)  # 1,512,60,60
+        x = self.layer3(x)
+        x = self.conv5(x)
         x = self.bn5(x)
         x = self.relu5(x)
 
-        x = self.layer4(x)  # 1,512,60,60
-        x = self.conv6(x)  # 1,512,60,60
+        x = self.layer4(x)
+        x = self.conv6(x)
         x = self.bn6(x)
         x = self.relu6(x)
         f.append(x)
-        # (6, 40)
         return f
 
 
diff --git a/ppocr/modeling/heads/table_master_head.py b/ppocr/modeling/heads/table_master_head.py
index 887630a87e..fddbcc63fc 100644
--- a/ppocr/modeling/heads/table_master_head.py
+++ b/ppocr/modeling/heads/table_master_head.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-This fun code is refer from:
+This code is refer from:
 https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/mmocr/models/textrecog/decoders/master_decoder.py
 """
 
@@ -135,7 +135,7 @@ class TableMasterHead(nn.Layer):
         batch_size = out_enc.shape[0]
         SOS = paddle.zeros([batch_size, 1], dtype='int64') + self.SOS
         output, bbox_output = self.greedy_forward(SOS, out_enc)
-        # output = F.softmax(output)
+        output = F.softmax(output)
         return {'structure_probs': output, 'loc_preds': bbox_output}
 
     def forward(self, feat, targets=None):
diff --git a/ppocr/utils/visual.py b/ppocr/utils/visual.py
index 235eb572a3..b20222bdbd 100644
--- a/ppocr/utils/visual.py
+++ b/ppocr/utils/visual.py
@@ -110,3 +110,16 @@ def draw_re_results(image,
 
     img_new = Image.blend(image, img_new, 0.5)
     return np.array(img_new)
+
+
+def draw_rectangle(img_path, boxes, use_xywh=False):
+    img = cv2.imread(img_path)
+    img_show = img.copy()
+    for box in boxes.astype(int):
+        if use_xywh:
+            x, y, w, h = box
+            x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
+        else:
+            x1, y1, x2, y2 = box
+        cv2.rectangle(img_show, (x1, y1), (x2, y2), (255, 0, 0), 2)
+    return img_show
\ No newline at end of file
diff --git a/ppstructure/table/predict_structure.py b/ppstructure/table/predict_structure.py
index 17ec909582..00385c7665 100755
--- a/ppstructure/table/predict_structure.py
+++ b/ppstructure/table/predict_structure.py
@@ -30,6 +30,7 @@ from ppocr.data import create_operators, transform
 from ppocr.postprocess import build_post_process
 from ppocr.utils.logging import get_logger
 from ppocr.utils.utility import get_image_file_list, check_and_read_gif
+from ppocr.utils.visual import draw_rectangle
 from ppstructure.utility import parse_args
 
 logger = get_logger()
@@ -120,19 +121,6 @@ class TableStructurer(object):
         return structure_str_list, bbox_list, elapse
 
 
-def draw_rectangle(img_path, boxes, use_xywh=False):
-    img = cv2.imread(img_path)
-    img_show = img.copy()
-    for box in boxes.astype(int):
-        if use_xywh:
-            x, y, w, h = box
-            x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
-        else:
-            x1, y1, x2, y2 = box
-        cv2.rectangle(img_show, (x1, y1), (x2, y2), (255, 0, 0), 2)
-    return img_show
-
-
 def main(args):
     image_file_list = get_image_file_list(args.image_dir)
     table_structurer = TableStructurer(args)
diff --git a/tools/infer_table.py b/tools/infer_table.py
index 58e7455cbb..6c02dd8640 100644
--- a/tools/infer_table.py
+++ b/tools/infer_table.py
@@ -36,6 +36,7 @@ from ppocr.modeling.architectures import build_model
 from ppocr.postprocess import build_post_process
 from ppocr.utils.save_load import load_model
 from ppocr.utils.utility import get_image_file_list
+from ppocr.utils.visual import draw_rectangle
 import tools.program as program
 import cv2
 
@@ -111,19 +112,6 @@ def main(config, device, logger, vdl_writer):
         logger.info("success!")
 
 
-def draw_rectangle(img_path, boxes, use_xywh=False):
-    img = cv2.imread(img_path)
-    img_show = img.copy()
-    for box in boxes.astype(int):
-        if use_xywh:
-            x, y, w, h = box
-            x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
-        else:
-            x1, y1, x2, y2 = box
-        cv2.rectangle(img_show, (x1, y1), (x2, y2), (255, 0, 0), 2)
-    return img_show
-
-
 if __name__ == '__main__':
     config, device, logger, vdl_writer = program.preprocess()
     main(config, device, logger, vdl_writer)

From da6636dceebb8af3e0a39341e7beb1a8faf9c7cc Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 5 Jul 2022 03:01:10 +0000
Subject: [PATCH 120/134] add polygons

---
 ppocr/postprocess/db_postprocess.py | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/ppocr/postprocess/db_postprocess.py b/ppocr/postprocess/db_postprocess.py
index e1041b00f9..5e2553c3a0 100755
--- a/ppocr/postprocess/db_postprocess.py
+++ b/ppocr/postprocess/db_postprocess.py
@@ -73,7 +73,6 @@ class DBPostProcess(object):
             epsilon = 0.002 * cv2.arcLength(contour, True)
             approx = cv2.approxPolyDP(contour, epsilon, True)
             points = approx.reshape((-1, 2))
-            # print(points)
             if points.shape[0] < 4:
                 continue
 
@@ -88,7 +87,6 @@ class DBPostProcess(object):
             else:
                 continue
             box = box.reshape(-1, 2)
-            # print(box)
 
             _, sside = self.get_mini_boxes(box.reshape((-1, 1, 2)))
             if sside < self.min_size + 2:
@@ -101,7 +99,6 @@ class DBPostProcess(object):
                 np.round(box[:, 1] / height * dest_height), 0, dest_height)
             boxes.append(box.tolist())
             scores.append(score)
-        # print(boxes)
         return boxes, scores
 
     def boxes_from_bitmap(self, pred, _bitmap, dest_width, dest_height):

From 52d5b54569d1a10e9272c278eb98e2775ce58680 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Tue, 5 Jul 2022 03:29:43 +0000
Subject: [PATCH 121/134] add import cv2

---
 ppocr/utils/visual.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/ppocr/utils/visual.py b/ppocr/utils/visual.py
index b20222bdbd..e0fbf06abb 100644
--- a/ppocr/utils/visual.py
+++ b/ppocr/utils/visual.py
@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import cv2
 import os
 import numpy as np
 from PIL import Image, ImageDraw, ImageFont

From f39ceb193591f6b205ed4e389a48940b359424bf Mon Sep 17 00:00:00 2001
From: mmglove <gmm_email@163.com>
Date: Tue, 5 Jul 2022 15:30:14 +0800
Subject: [PATCH 122/134] fix benchmark

---
 test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt     | 2 +-
 test_tipc/configs/det_r18_vd_db_v2_0/train_infer_python.txt  | 2 +-
 test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt b/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt
index ab3aa59b60..2c8aa95344 100644
--- a/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt
+++ b/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt
@@ -54,6 +54,6 @@ random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
 ===========================train_benchmark_params==========================
 batch_size:8|16
 fp_items:fp32|fp16
-epoch:2
+epoch:15
 --profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
 flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/det_r18_vd_db_v2_0/train_infer_python.txt b/test_tipc/configs/det_r18_vd_db_v2_0/train_infer_python.txt
index 33e4dbf233..df88c0e543 100644
--- a/test_tipc/configs/det_r18_vd_db_v2_0/train_infer_python.txt
+++ b/test_tipc/configs/det_r18_vd_db_v2_0/train_infer_python.txt
@@ -54,5 +54,5 @@ random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
 ===========================train_benchmark_params==========================
 batch_size:8|16
 fp_items:fp32|fp16
-epoch:2
+epoch:15
 --profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
diff --git a/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt
index 62da89fe1c..53511e6ae2 100644
--- a/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt
+++ b/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt
@@ -54,5 +54,5 @@ random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
 ===========================train_benchmark_params==========================
 batch_size:8
 fp_items:fp32|fp16
-epoch:2
+epoch:10
 --profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile

From 96f3c6a7c51ffefbdc7fd569e731a34e35286d69 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Tue, 5 Jul 2022 08:21:40 +0000
Subject: [PATCH 123/134] add polygons

---
 tools/program.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/program.py b/tools/program.py
index e8e72fdd2d..c6422ac96c 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -575,7 +575,7 @@ def preprocess(is_train=False):
         'EAST', 'DB', 'SAST', 'Rosetta', 'CRNN', 'STARNet', 'RARE', 'SRN',
         'CLS', 'PGNet', 'Distillation', 'NRTR', 'TableAttn', 'SAR', 'PSE',
         'SEED', 'SDMGR', 'LayoutXLM', 'LayoutLM', 'LayoutLMv2', 'PREN', 'FCE',
-        'SVTR', 'ViTSTR', 'ABINet', 'DB++'
+        'SVTR', 'ViTSTR', 'ABINet', 'DB++', 'TableMaster'
     ]
 
     if use_xpu:

From 27181149a9eaa5ffc2da085fd1ad7eaff2eb68f8 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Tue, 5 Jul 2022 09:31:53 +0000
Subject: [PATCH 124/134] add tablemaster to tipc

---
 doc/doc_ch/algorithm.md                       |   3 +-
 doc/doc_ch/algorithm_overview.md              |  12 +-
 doc/doc_en/algorithm_en.md                    |   3 +-
 doc/doc_en/algorithm_overview_en.md           |  10 +-
 ppstructure/table/predict_structure.py        |   6 +-
 .../configs/en_table_structure/table_mv3.yml  |  41 +++---
 .../configs/table_master/table_master.yml     | 136 ++++++++++++++++++
 .../table_master/train_infer_python.txt       |  53 +++++++
 test_tipc/prepare.sh                          |   4 +
 test_tipc/readme.md                           |   1 +
 10 files changed, 237 insertions(+), 32 deletions(-)
 create mode 100644 test_tipc/configs/table_master/table_master.yml
 create mode 100644 test_tipc/configs/table_master/train_infer_python.txt

diff --git a/doc/doc_ch/algorithm.md b/doc/doc_ch/algorithm.md
index 3056f35d52..d50a5aa4e8 100644
--- a/doc/doc_ch/algorithm.md
+++ b/doc/doc_ch/algorithm.md
@@ -5,9 +5,10 @@ PaddleOCR将**持续新增**支持OCR领域前沿算法与模型，已支持的
 - [文本检测算法](./algorithm_overview.md#11-%E6%96%87%E6%9C%AC%E6%A3%80%E6%B5%8B%E7%AE%97%E6%B3%95)
 - [文本识别算法](./algorithm_overview.md#12-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95)
 - [端到端算法](./algorithm_overview.md#2-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95)
+- [表格识别]](./algorithm_overview.md#3-%E8%A1%A8%E6%A0%BC%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95)
 
 **欢迎广大开发者合作共建，贡献更多算法，合入有奖🎁！具体可查看[社区常规赛](https://github.com/PaddlePaddle/PaddleOCR/issues/4982)。**
 
 新增算法可参考如下教程：
 
-- [使用PaddleOCR架构添加新算法](./add_new_algorithm.md)
\ No newline at end of file
+- [使用PaddleOCR架构添加新算法](./add_new_algorithm.md)
diff --git a/doc/doc_ch/algorithm_overview.md b/doc/doc_ch/algorithm_overview.md
index 1efd564ce1..5c7adc715f 100755
--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
@@ -1,10 +1,10 @@
 # OCR算法
 
-- [1. 两阶段算法](#1-两阶段算法)
-  - [1.1 文本检测算法](#11-文本检测算法)
-  - [1.2 文本识别算法](#12-文本识别算法)
-- [2. 端到端算法](#2-端到端算法)
-- [3. 表格识别算法](#3-表格识别算法)
+- [1. 两阶段算法](#1)
+  - [1.1 文本检测算法](#11)
+  - [1.2 文本识别算法](#12)
+- [2. 端到端算法](#2)
+- [3. 表格识别算法](#3)
 
 
 本文给出了PaddleOCR已支持的OCR算法列表，以及每个算法在**英文公开数据集**上的模型和指标，主要用于算法简介和算法性能对比，更多包括中文在内的其他数据集上的模型请参考[PP-OCR v2.0 系列模型下载](./models_list.md)。
@@ -98,6 +98,8 @@
 已支持的端到端OCR算法列表（戳链接获取使用教程）：
 - [x]  [PGNet](./algorithm_e2e_pgnet.md)
 
+<a name="3"></a>
+
 ## 3. 表格识别算法
 
 已支持的表格识别算法列表（戳链接获取使用教程）：
diff --git a/doc/doc_en/algorithm_en.md b/doc/doc_en/algorithm_en.md
index fa7887eb26..c880336b4a 100644
--- a/doc/doc_en/algorithm_en.md
+++ b/doc/doc_en/algorithm_en.md
@@ -6,5 +6,6 @@ PaddleOCR will add cutting-edge OCR algorithms and models continuously. Check ou
 - [text detection algorithms](./algorithm_overview_en.md#11)
 - [text recognition algorithms](./algorithm_overview_en.md#12)
 - [end-to-end algorithms](./algorithm_overview_en.md#2)
+- [table recognition algorithms](./algorithm_overview_en.md#3)
 
-Developers are welcome to contribute more algorithms! Please refer to [add new algorithm](./add_new_algorithm_en.md) guideline.
\ No newline at end of file
+Developers are welcome to contribute more algorithms! Please refer to [add new algorithm](./add_new_algorithm_en.md) guideline.
diff --git a/doc/doc_en/algorithm_overview_en.md b/doc/doc_en/algorithm_overview_en.md
index 8b9b3f7b0b..f3c96b620c 100755
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
@@ -1,10 +1,10 @@
 # OCR Algorithms
 
-- [1. Two-stage Algorithms](#1-two-stage-algorithms)
-  - [1.1 Text Detection Algorithms](#11-text-detection-algorithms)
-  - [1.2 Text Recognition Algorithms](#12-text-recognition-algorithms)
-- [2. End-to-end Algorithms](#2-end-to-end-algorithms)
-- [3. Table Recognition Algorithms](#3-table-recognition-algorithms)
+- [1. Two-stage Algorithms](#1)
+  - [1.1 Text Detection Algorithms](#11)
+  - [1.2 Text Recognition Algorithms](#12)
+- [2. End-to-end Algorithms](#2)
+- [3. Table Recognition Algorithms](#3)
 
 
 This tutorial lists the OCR algorithms supported by PaddleOCR, as well as the models and metrics of each algorithm on **English public datasets**. It is mainly used for algorithm introduction and algorithm performance comparison. For more models on other datasets including Chinese, please refer to [PP-OCR v2.0 models list](./models_list_en.md).
diff --git a/ppstructure/table/predict_structure.py b/ppstructure/table/predict_structure.py
index 00385c7665..7a7d3169d5 100755
--- a/ppstructure/table/predict_structure.py
+++ b/ppstructure/table/predict_structure.py
@@ -118,7 +118,7 @@ class TableStructurer(object):
             '<html>', '<body>', '<table>'
         ] + structure_str_list + ['</table>', '</body>', '</html>']
         elapse = time.time() - starttime
-        return structure_str_list, bbox_list, elapse
+        return (structure_str_list, bbox_list), elapse
 
 
 def main(args):
@@ -138,8 +138,8 @@ def main(args):
             if img is None:
                 logger.info("error in loading image:{}".format(image_file))
                 continue
-            structure_str_list, bbox_list, elapse = table_structurer(img)
-
+            structure_res, elapse = table_structurer(img)
+            structure_str_list, bbox_list = structure_res
             bbox_list_str = json.dumps(bbox_list.tolist())
             logger.info("result: {}, {}".format(structure_str_list,
                                                 bbox_list_str))
diff --git a/test_tipc/configs/en_table_structure/table_mv3.yml b/test_tipc/configs/en_table_structure/table_mv3.yml
index adf326bd02..6df5a1a20b 100755
--- a/test_tipc/configs/en_table_structure/table_mv3.yml
+++ b/test_tipc/configs/en_table_structure/table_mv3.yml
@@ -1,24 +1,23 @@
 Global:
   use_gpu: true
-  epoch_num: 10
+  epoch_num: 400
   log_smooth_window: 20
   print_batch_step: 5
   save_model_dir: ./output/table_mv3/
-  save_epoch_step: 3
+  save_epoch_step: 400
   # evaluation is run every 400 iterations after the 0th iteration
   eval_batch_step: [0, 400]
   cal_metric_during_train: True
   pretrained_model:
-  checkpoints: 
+  checkpoints:
   save_inference_dir:
   use_visualdl: False
-  infer_img: doc/table/table.jpg
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: output/table_mv3
   # for data or label process
   character_dict_path: ppocr/utils/dict/table_structure_dict.txt
   character_type: en
-  max_text_length: 100
-  max_elem_length: 800
-  max_cell_num: 500
+  max_text_length: 800
   infer_mode: False
   process_total_num: 0
   process_cut_num: 0
@@ -44,11 +43,8 @@ Architecture:
   Head:
     name: TableAttentionHead
     hidden_size: 256
-    l2_decay: 0.00001
     loc_type: 2
-    max_text_length: 100
-    max_elem_length: 800
-    max_cell_num: 500
+    max_text_length: 800
 
 Loss:
   name: TableAttentionLoss
@@ -61,28 +57,34 @@ PostProcess:
 Metric:
   name: TableMetric
   main_indicator: acc
+  compute_bbox_metric: false # cost many time, set False for training
 
 Train:
   dataset:
     name: PubTabDataSet
     data_dir: ./train_data/pubtabnet/train
-    label_file_path: ./train_data/pubtabnet/train.jsonl
+    label_file_list: [./train_data/pubtabnet/train.jsonl]
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+      - TableBoxEncode:
       - ResizeTableImage:
           max_len: 488
-      - TableLabelEncode:
       - NormalizeImage:
           scale: 1./255.
           mean: [0.485, 0.456, 0.406]
           std: [0.229, 0.224, 0.225]
           order: 'hwc'
       - PaddingTableImage:
+          size: [488, 488]
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
   loader:
     shuffle: True
     batch_size_per_card: 32
@@ -93,23 +95,28 @@ Eval:
   dataset:
     name: PubTabDataSet
     data_dir: ./train_data/pubtabnet/test/
-    label_file_path: ./train_data/pubtabnet/test.jsonl
+    label_file_list: [./train_data/pubtabnet/test.jsonl]
     transforms:
       - DecodeImage: # load image
           img_mode: BGR
           channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+      - TableBoxEncode:
       - ResizeTableImage:
           max_len: 488
-      - TableLabelEncode:
       - NormalizeImage:
           scale: 1./255.
           mean: [0.485, 0.456, 0.406]
           std: [0.229, 0.224, 0.225]
           order: 'hwc'
       - PaddingTableImage:
+          size: [488, 488]
       - ToCHWImage:
       - KeepKeys:
-          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
   loader:
     shuffle: False
     drop_last: False
diff --git a/test_tipc/configs/table_master/table_master.yml b/test_tipc/configs/table_master/table_master.yml
new file mode 100644
index 0000000000..c519b5b8f4
--- /dev/null
+++ b/test_tipc/configs/table_master/table_master.yml
@@ -0,0 +1,136 @@
+Global:
+  use_gpu: true
+  epoch_num: 17
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/table_master/
+  save_epoch_step: 17
+  eval_batch_step: [0,  6259]
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: 
+  save_inference_dir: output/table_master/infer
+  use_visualdl: false
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: ./output/table_master
+  character_dict_path: ppocr/utils/dict/table_master_structure_dict.txt
+  infer_mode: false
+  max_text_length: 500
+  process_total_num: 0
+  process_cut_num: 0
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: MultiStepDecay
+    learning_rate: 0.001
+    milestones: [12, 15]
+    gamma: 0.1
+    warmup_epoch: 0.02
+  regularizer:
+    name: L2
+    factor: 0.0
+
+Architecture:
+  model_type: table
+  algorithm: TableMaster
+  Backbone:
+    name: TableResNetExtra
+    gcb_config:
+      ratio: 0.0625
+      headers: 1
+      att_scale: False
+      fusion_type: channel_add
+      layers: [False, True, True, True]
+    layers: [1,2,5,3]
+  Head:
+    name: TableMasterHead
+    hidden_size: 512
+    headers: 8
+    dropout: 0
+    d_ff: 2024
+    max_text_length: 500
+
+Loss:
+  name: TableMasterLoss
+  ignore_index: 42 # set to len of dict + 3
+
+PostProcess:
+  name: TableMasterLabelDecode
+  box_shape: pad
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ./train_data/pubtabnet/train
+    label_file_list: [./train_data/pubtabnet/train.jsonl]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: True
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          use_xywh: True
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: True
+    batch_size_per_card: 10
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ./train_data/pubtabnet/test/
+    label_file_list: [./train_data/pubtabnet/test.jsonl]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: True
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          use_xywh: True
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 10
+    num_workers: 8
\ No newline at end of file
diff --git a/test_tipc/configs/table_master/train_infer_python.txt b/test_tipc/configs/table_master/train_infer_python.txt
new file mode 100644
index 0000000000..56b8e63602
--- /dev/null
+++ b/test_tipc/configs/table_master/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:table_master
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:./pretrain_models/table_structure_tablemaster_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/table_master/table_master.yml -o Global.print_batch_step=10
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/table_master/table_master.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_structure.py  --table_char_dict_path=./ppocr/utils/dict/table_master_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --output ./output/table --table_algorithm=TableMaster --table_max_len=480 
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,480,480]}]
diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 32df8e7865..593cd872b7 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -62,6 +62,10 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
     if [[ ${model_name} =~ "det_r50_db++" ]];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams --no-check-certificate
     fi
+    if [ ${model_name} == "table_master" ];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf table_structure_tablemaster_train.tar  && cd ../
+    fi
     cd ./pretrain_models/ && tar xf det_mv3_db_v2.0_train.tar && cd ../
     rm -rf ./train_data/icdar2015
     rm -rf ./train_data/ic15_data
diff --git a/test_tipc/readme.md b/test_tipc/readme.md
index effb2f168b..1c637d76f9 100644
--- a/test_tipc/readme.md
+++ b/test_tipc/readme.md
@@ -54,6 +54,7 @@
 | NRTR   |rec_mtb_nrtr                   | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
 | SAR    |rec_r31_sar                    | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
 | PGNet  |rec_r34_vd_none_none_ctc_v2.0  | 端到端| 支持 | 多机多卡 <br> 混合精度 | - | - |
+| TableMaster  |table_structure_tablemaster_train  | 表格识别| 支持 | 多机多卡 <br> 混合精度 | - | - |
 
 
 

From 029223c6074b54ef4bed3cbebc5517deae306d90 Mon Sep 17 00:00:00 2001
From: WenmuZhou <572459439@qq.com>
Date: Tue, 5 Jul 2022 09:48:26 +0000
Subject: [PATCH 125/134] reduce epoch_num

---
 test_tipc/configs/en_table_structure/table_mv3.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/test_tipc/configs/en_table_structure/table_mv3.yml b/test_tipc/configs/en_table_structure/table_mv3.yml
index 6df5a1a20b..281038b968 100755
--- a/test_tipc/configs/en_table_structure/table_mv3.yml
+++ b/test_tipc/configs/en_table_structure/table_mv3.yml
@@ -1,10 +1,10 @@
 Global:
   use_gpu: true
-  epoch_num: 400
+  epoch_num: 10
   log_smooth_window: 20
   print_batch_step: 5
   save_model_dir: ./output/table_mv3/
-  save_epoch_step: 400
+  save_epoch_step: 3
   # evaluation is run every 400 iterations after the 0th iteration
   eval_batch_step: [0, 400]
   cal_metric_during_train: True

From 5a0108b8acb6733030a752b0291ea4b1b78c65a2 Mon Sep 17 00:00:00 2001
From: MissPenguin <lichenxia1991@163.com>
Date: Wed, 6 Jul 2022 11:17:41 +0800
Subject: [PATCH 126/134] Update README.md

---
 README.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/README.md b/README.md
index b168ecc5e5..f57672e505 100644
--- a/README.md
+++ b/README.md
@@ -72,6 +72,7 @@ PaddleOCR support a variety of cutting-edge algorithms related to OCR, and devel
 <img src="https://raw.githubusercontent.com/PaddlePaddle/PaddleOCR/dygraph/doc/joinus.PNG"  width = "200" height = "200" />
 </div>
 <a name="Supported-Chinese-model-list"></a>
+
 ## PP-OCR Series Model List（Update on September 8th）
 
 | Model introduction                                           | Model name                   | Recommended scene | Detection model                                              | Direction classifier                                         | Recognition model                                            |

From 7a99588dd8b678233eff0391aac13ebd0f7000f6 Mon Sep 17 00:00:00 2001
From: littletomatodonkey <dazhiningsibuqu@163.com>
Date: Wed, 6 Jul 2022 13:58:46 +0800
Subject: [PATCH 127/134] add more dataset yamls and fix re exceptions (#6791)

* add more dataset yamls and fix re exceptions
---
 configs/vqa/re/layoutlmv2_funsd.yml           | 125 +++++++++++++++
 ...{layoutlmv2.yml => layoutlmv2_xund_zh.yml} |   6 +-
 configs/vqa/re/layoutxlm_funsd.yml            | 129 +++++++++++++++
 .../{layoutxlm.yml => layoutxlm_xfund_zh.yml} |   0
 configs/vqa/ser/layoutlm_funsd.yml            | 124 ++++++++++++++
 configs/vqa/ser/layoutlm_sroie.yml            | 124 ++++++++++++++
 .../{layoutlm.yml => layoutlm_xfund_zh.yml}   |   6 +-
 configs/vqa/ser/layoutlmv2_funsd.yml          | 123 ++++++++++++++
 configs/vqa/ser/layoutlmv2_sroie.yml          | 123 ++++++++++++++
 ...layoutlmv2.yml => layoutlmv2_xfund_zh.yml} |   4 +-
 configs/vqa/ser/layoutxlm_funsd.yml           | 123 ++++++++++++++
 configs/vqa/ser/layoutxlm_sroie.yml           | 123 ++++++++++++++
 configs/vqa/ser/layoutxlm_wildreceipt.yml     | 123 ++++++++++++++
 .../{layoutxlm.yml => layoutxlm_xfund_zh.yml} |   4 +-
 ppocr/data/imaug/label_ops.py                 |  79 +++++++--
 ppocr/metrics/vqa_token_re_metric.py          |  33 ++--
 ppocr/modeling/backbones/vqa_layoutlm.py      |   4 +-
 ppstructure/vqa/tools/trans_funsd_label.py    | 151 ++++++++++++++++++
 18 files changed, 1366 insertions(+), 38 deletions(-)
 create mode 100644 configs/vqa/re/layoutlmv2_funsd.yml
 rename configs/vqa/re/{layoutlmv2.yml => layoutlmv2_xund_zh.yml} (95%)
 create mode 100644 configs/vqa/re/layoutxlm_funsd.yml
 rename configs/vqa/re/{layoutxlm.yml => layoutxlm_xfund_zh.yml} (100%)
 create mode 100644 configs/vqa/ser/layoutlm_funsd.yml
 create mode 100644 configs/vqa/ser/layoutlm_sroie.yml
 rename configs/vqa/ser/{layoutlm.yml => layoutlm_xfund_zh.yml} (95%)
 create mode 100644 configs/vqa/ser/layoutlmv2_funsd.yml
 create mode 100644 configs/vqa/ser/layoutlmv2_sroie.yml
 rename configs/vqa/ser/{layoutlmv2.yml => layoutlmv2_xfund_zh.yml} (96%)
 create mode 100644 configs/vqa/ser/layoutxlm_funsd.yml
 create mode 100644 configs/vqa/ser/layoutxlm_sroie.yml
 create mode 100644 configs/vqa/ser/layoutxlm_wildreceipt.yml
 rename configs/vqa/ser/{layoutxlm.yml => layoutxlm_xfund_zh.yml} (96%)
 create mode 100644 ppstructure/vqa/tools/trans_funsd_label.py

diff --git a/configs/vqa/re/layoutlmv2_funsd.yml b/configs/vqa/re/layoutlmv2_funsd.yml
new file mode 100644
index 0000000000..1c3d8f7854
--- /dev/null
+++ b/configs/vqa/re/layoutlmv2_funsd.yml
@@ -0,0 +1,125 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/re_layoutlmv2_funsd
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 57 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/FUNSD/testing_data/images/83624198.png
+  save_res_path: ./output/re_layoutlmv2_funsd/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutLMv2"
+  Transform:
+  Backbone:
+    name: LayoutLMv2ForRe
+    pretrained: True
+    checkpoints:
+
+Loss:
+  name: LossFromOutput
+  key: loss
+  reduction: mean
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10
+  lr:
+    learning_rate: 0.00005
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQAReTokenLayoutLMPostProcess
+
+Metric:
+  name: VQAReTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/training_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: &class_path train_data/FUNSD/class_list.txt
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'entities', 'relations']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+    collate_fn: ListCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/testing_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/test.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'entities', 'relations']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+    collate_fn: ListCollator
diff --git a/configs/vqa/re/layoutlmv2.yml b/configs/vqa/re/layoutlmv2_xund_zh.yml
similarity index 95%
rename from configs/vqa/re/layoutlmv2.yml
rename to configs/vqa/re/layoutlmv2_xund_zh.yml
index 737dbf6b60..986b9b5cef 100644
--- a/configs/vqa/re/layoutlmv2.yml
+++ b/configs/vqa/re/layoutlmv2_xund_zh.yml
@@ -3,16 +3,16 @@ Global:
   epoch_num: &epoch_num 200
   log_smooth_window: 10
   print_batch_step: 10
-  save_model_dir: ./output/re_layoutlmv2/
+  save_model_dir: ./output/re_layoutlmv2_xfund_zh
   save_epoch_step: 2000
   # evaluation is run every 10 iterations after the 0th iteration
-  eval_batch_step: [ 0, 19 ]
+  eval_batch_step: [ 0, 57 ]
   cal_metric_during_train: False
   save_inference_dir:
   use_visualdl: False
   seed: 2048
   infer_img: ppstructure/docs/vqa/input/zh_val_21.jpg
-  save_res_path: ./output/re/
+  save_res_path: ./output/re_layoutlmv2_xfund_zh/res/
 
 Architecture:
   model_type: vqa
diff --git a/configs/vqa/re/layoutxlm_funsd.yml b/configs/vqa/re/layoutxlm_funsd.yml
new file mode 100644
index 0000000000..af28be10d6
--- /dev/null
+++ b/configs/vqa/re/layoutxlm_funsd.yml
@@ -0,0 +1,129 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/re_layoutxlm_funsd
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 57 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/FUNSD/testing_data/images/83624198.png
+  save_res_path: ./output/re_layoutxlm_funsd/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForRe
+    pretrained: True
+    checkpoints:
+
+Loss:
+  name: LossFromOutput
+  key: loss
+  reduction: mean
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10
+  lr:
+    learning_rate: 0.00005
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQAReTokenLayoutLMPostProcess
+
+Metric:
+  name: VQAReTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/training_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/train_v4.json
+      # - ./train_data/FUNSD/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: &class_path ./train_data/FUNSD/class_list.txt
+          use_textline_bbox_info: &use_textline_bbox_info True
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'entities', 'relations']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 16
+    collate_fn: ListCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/testing_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/test_v4.json
+      # - ./train_data/FUNSD/test.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: *use_textline_bbox_info
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'entities', 'relations']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+    collate_fn: ListCollator
diff --git a/configs/vqa/re/layoutxlm.yml b/configs/vqa/re/layoutxlm_xfund_zh.yml
similarity index 100%
rename from configs/vqa/re/layoutxlm.yml
rename to configs/vqa/re/layoutxlm_xfund_zh.yml
diff --git a/configs/vqa/ser/layoutlm_funsd.yml b/configs/vqa/ser/layoutlm_funsd.yml
new file mode 100644
index 0000000000..0ef3502bc8
--- /dev/null
+++ b/configs/vqa/ser/layoutlm_funsd.yml
@@ -0,0 +1,124 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutlm_funsd
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 57 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/FUNSD/testing_data/images/83624198.png
+  save_res_path: ./output/ser_layoutlm_funsd/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutLM"
+  Transform:
+  Backbone:
+    name: LayoutLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path ./train_data/FUNSD/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/training_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/train.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: &use_textline_bbox_info True
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/FUNSD/testing_data/images/
+    label_file_list:
+      - ./train_data/FUNSD/test.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: *use_textline_bbox_info
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutlm_sroie.yml b/configs/vqa/ser/layoutlm_sroie.yml
new file mode 100644
index 0000000000..6abb1151e5
--- /dev/null
+++ b/configs/vqa/ser/layoutlm_sroie.yml
@@ -0,0 +1,124 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutlm_sroie
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 200 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/SROIE/test/X00016469670.jpg
+  save_res_path: ./output/ser_layoutlm_sroie/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutLM"
+  Transform:
+  Backbone:
+    name: LayoutLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 9
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path ./train_data/SROIE/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/SROIE/train
+    label_file_list: 
+      - ./train_data/SROIE/train.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: &use_textline_bbox_info True
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/SROIE/test
+    label_file_list: 
+      - ./train_data/SROIE/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: *use_textline_bbox_info
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutlm.yml b/configs/vqa/ser/layoutlm_xfund_zh.yml
similarity index 95%
rename from configs/vqa/ser/layoutlm.yml
rename to configs/vqa/ser/layoutlm_xfund_zh.yml
index 53e114defd..99763c1963 100644
--- a/configs/vqa/ser/layoutlm.yml
+++ b/configs/vqa/ser/layoutlm_xfund_zh.yml
@@ -3,16 +3,16 @@ Global:
   epoch_num: &epoch_num 200
   log_smooth_window: 10
   print_batch_step: 10
-  save_model_dir: ./output/ser_layoutlm/
+  save_model_dir: ./output/ser_layoutlm_xfund_zh
   save_epoch_step: 2000
   # evaluation is run every 10 iterations after the 0th iteration
-  eval_batch_step: [ 0, 19 ]
+  eval_batch_step: [ 0, 57 ]
   cal_metric_during_train: False
   save_inference_dir:
   use_visualdl: False
   seed: 2022
   infer_img: ppstructure/docs/vqa/input/zh_val_42.jpg
-  save_res_path: ./output/ser/
+  save_res_path: ./output/ser_layoutlm_xfund_zh/res/
 
 Architecture:
   model_type: vqa
diff --git a/configs/vqa/ser/layoutlmv2_funsd.yml b/configs/vqa/ser/layoutlmv2_funsd.yml
new file mode 100644
index 0000000000..438edc1aa7
--- /dev/null
+++ b/configs/vqa/ser/layoutlmv2_funsd.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutlmv2_funsd
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 100 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/FUNSD/testing_data/images/83624198.png
+  save_res_path: ./output/ser_layoutlmv2_funsd/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutLMv2"
+  Transform:
+  Backbone:
+    name: LayoutLMv2ForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+  
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path train_data/FUNSD/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/training_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/train.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/testing_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/test.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutlmv2_sroie.yml b/configs/vqa/ser/layoutlmv2_sroie.yml
new file mode 100644
index 0000000000..549beb8ec5
--- /dev/null
+++ b/configs/vqa/ser/layoutlmv2_sroie.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutlmv2_sroie
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 200 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/SROIE/test/X00016469670.jpg
+  save_res_path: ./output/ser_layoutlmv2_sroie/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutLMv2"
+  Transform:
+  Backbone:
+    name: LayoutLMv2ForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 9
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+  
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path ./train_data/SROIE/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/SROIE/train
+    label_file_list: 
+      - ./train_data/SROIE/train.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/SROIE/test
+    label_file_list: 
+      - ./train_data/SROIE/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutlmv2.yml b/configs/vqa/ser/layoutlmv2_xfund_zh.yml
similarity index 96%
rename from configs/vqa/ser/layoutlmv2.yml
rename to configs/vqa/ser/layoutlmv2_xfund_zh.yml
index e48c746956..ebdc5f3169 100644
--- a/configs/vqa/ser/layoutlmv2.yml
+++ b/configs/vqa/ser/layoutlmv2_xfund_zh.yml
@@ -3,7 +3,7 @@ Global:
   epoch_num: &epoch_num 200
   log_smooth_window: 10
   print_batch_step: 10
-  save_model_dir: ./output/ser_layoutlmv2/
+  save_model_dir: ./output/ser_layoutlmv2_xfund_zh/
   save_epoch_step: 2000
   # evaluation is run every 10 iterations after the 0th iteration
   eval_batch_step: [ 0, 19 ]
@@ -12,7 +12,7 @@ Global:
   use_visualdl: False
   seed: 2022
   infer_img: ppstructure/docs/vqa/input/zh_val_42.jpg
-  save_res_path: ./output/ser/
+  save_res_path: ./output/ser_layoutlmv2_xfund_zh/res/
 
 Architecture:
   model_type: vqa
diff --git a/configs/vqa/ser/layoutxlm_funsd.yml b/configs/vqa/ser/layoutxlm_funsd.yml
new file mode 100644
index 0000000000..be1e9d4f1e
--- /dev/null
+++ b/configs/vqa/ser/layoutxlm_funsd.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutxlm_funsd
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 57 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/FUNSD/testing_data/images/83624198.png
+  save_res_path: output/ser_layoutxlm_funsd/res/
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path ./train_data/FUNSD/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/FUNSD/training_data/images/
+    label_file_list: 
+      - ./train_data/FUNSD/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/FUNSD/testing_data/images/
+    label_file_list:
+      - ./train_data/FUNSD/test.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutxlm_sroie.yml b/configs/vqa/ser/layoutxlm_sroie.yml
new file mode 100644
index 0000000000..dd63d888d0
--- /dev/null
+++ b/configs/vqa/ser/layoutxlm_sroie.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutxlm_sroie
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 200 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data/SROIE/test/X00016469670.jpg
+  save_res_path: res_img_aug_with_gt
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 9
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path ./train_data/SROIE/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/SROIE/train
+    label_file_list: 
+      - ./train_data/SROIE/train.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/SROIE/test
+    label_file_list:
+      - ./train_data/SROIE/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutxlm_wildreceipt.yml b/configs/vqa/ser/layoutxlm_wildreceipt.yml
new file mode 100644
index 0000000000..92c0394296
--- /dev/null
+++ b/configs/vqa/ser/layoutxlm_wildreceipt.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 100
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutxlm_wildreceipt
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 200 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: train_data//wildreceipt/image_files/Image_12/10/845be0dd6f5b04866a2042abd28d558032ef2576.jpeg
+  save_res_path: ./output/ser_layoutxlm_wildreceipt/res
+
+Architecture:
+  model_type: vqa
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 51
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path ./train_data/wildreceipt/class_list.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/wildreceipt/
+    label_file_list: 
+      - ./train_data/wildreceipt/wildreceipt_train.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/wildreceipt
+    label_file_list:
+      - ./train_data/wildreceipt/wildreceipt_test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          # dataloader will return list in this order
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/vqa/ser/layoutxlm.yml b/configs/vqa/ser/layoutxlm_xfund_zh.yml
similarity index 96%
rename from configs/vqa/ser/layoutxlm.yml
rename to configs/vqa/ser/layoutxlm_xfund_zh.yml
index fa9df192af..68df7d9f03 100644
--- a/configs/vqa/ser/layoutxlm.yml
+++ b/configs/vqa/ser/layoutxlm_xfund_zh.yml
@@ -3,7 +3,7 @@ Global:
   epoch_num: &epoch_num 200
   log_smooth_window: 10
   print_batch_step: 10
-  save_model_dir: ./output/ser_layoutxlm/
+  save_model_dir: ./output/ser_layoutxlm_xfund_zh
   save_epoch_step: 2000
   # evaluation is run every 10 iterations after the 0th iteration
   eval_batch_step: [ 0, 19 ]
@@ -12,7 +12,7 @@ Global:
   use_visualdl: False
   seed: 2022
   infer_img: ppstructure/docs/vqa/input/zh_val_42.jpg
-  save_res_path: ./output/ser
+  save_res_path: ./output/ser_layoutxlm_xfund_zh/res
 
 Architecture:
   model_type: vqa
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
index 7cc4cef46a..a4087d5328 100644
--- a/ppocr/data/imaug/label_ops.py
+++ b/ppocr/data/imaug/label_ops.py
@@ -869,6 +869,7 @@ class VQATokenLabelEncode(object):
                  contains_re=False,
                  add_special_ids=False,
                  algorithm='LayoutXLM',
+                 use_textline_bbox_info=True,
                  infer_mode=False,
                  ocr_engine=None,
                  **kwargs):
@@ -897,11 +898,51 @@ class VQATokenLabelEncode(object):
         self.add_special_ids = add_special_ids
         self.infer_mode = infer_mode
         self.ocr_engine = ocr_engine
+        self.use_textline_bbox_info = use_textline_bbox_info
+
+    def split_bbox(self, bbox, text, tokenizer):
+        words = text.split()
+        token_bboxes = []
+        curr_word_idx = 0
+        x1, y1, x2, y2 = bbox
+        unit_w = (x2 - x1) / len(text)
+        for idx, word in enumerate(words):
+            curr_w = len(word) * unit_w
+            word_bbox = [x1, y1, x1 + curr_w, y2]
+            token_bboxes.extend([word_bbox] * len(tokenizer.tokenize(word)))
+            x1 += (len(word) + 1) * unit_w
+        return token_bboxes
+
+    def filter_empty_contents(self, ocr_info):
+        """
+        find out the empty texts and remove the links
+        """
+        new_ocr_info = []
+        empty_index = []
+        for idx, info in enumerate(ocr_info):
+            if len(info["transcription"]) > 0:
+                new_ocr_info.append(copy.deepcopy(info))
+            else:
+                empty_index.append(info["id"])
+
+        for idx, info in enumerate(new_ocr_info):
+            new_link = []
+            for link in info["linking"]:
+                if link[0] in empty_index or link[1] in empty_index:
+                    continue
+                new_link.append(link)
+            new_ocr_info[idx]["linking"] = new_link
+        return new_ocr_info
 
     def __call__(self, data):
         # load bbox and label info
         ocr_info = self._load_ocr_info(data)
 
+        # for re
+        train_re = self.contains_re and not self.infer_mode
+        if train_re:
+            ocr_info = self.filter_empty_contents(ocr_info)
+
         height, width, _ = data['image'].shape
 
         words_list = []
@@ -913,8 +954,6 @@ class VQATokenLabelEncode(object):
 
         entities = []
 
-        # for re
-        train_re = self.contains_re and not self.infer_mode
         if train_re:
             relations = []
             id2label = {}
@@ -924,18 +963,19 @@ class VQATokenLabelEncode(object):
         data['ocr_info'] = copy.deepcopy(ocr_info)
 
         for info in ocr_info:
+            text = info["transcription"]
+            if len(text) <= 0:
+                continue
             if train_re:
                 # for re
-                if len(info["transcription"]) == 0:
+                if len(text) == 0:
                     empty_entity.add(info["id"])
                     continue
                 id2label[info["id"]] = info["label"]
                 relations.extend([tuple(sorted(l)) for l in info["linking"]])
             # smooth_box
             info["bbox"] = self.trans_poly_to_bbox(info["points"])
-            bbox = self._smooth_box(info["bbox"], height, width)
 
-            text = info["transcription"]
             encode_res = self.tokenizer.encode(
                 text, pad_to_max_seq_len=False, return_attention_mask=True)
 
@@ -946,6 +986,19 @@ class VQATokenLabelEncode(object):
                                                                             -1]
                 encode_res["attention_mask"] = encode_res["attention_mask"][1:
                                                                             -1]
+
+            if self.use_textline_bbox_info:
+                bbox = [info["bbox"]] * len(encode_res["input_ids"])
+            else:
+                bbox = self.split_bbox(info["bbox"], info["transcription"],
+                                       self.tokenizer)
+            if len(bbox) <= 0:
+                continue
+            bbox = self._smooth_box(bbox, height, width)
+            if self.add_special_ids:
+                bbox.insert(0, [0, 0, 0, 0])
+                bbox.append([0, 0, 0, 0])
+
             # parse label
             if not self.infer_mode:
                 label = info['label']
@@ -970,7 +1023,7 @@ class VQATokenLabelEncode(object):
                 })
             input_ids_list.extend(encode_res["input_ids"])
             token_type_ids_list.extend(encode_res["token_type_ids"])
-            bbox_list.extend([bbox] * len(encode_res["input_ids"]))
+            bbox_list.extend(bbox)
             words_list.append(text)
             segment_offset_id.append(len(input_ids_list))
             if not self.infer_mode:
@@ -1019,12 +1072,14 @@ class VQATokenLabelEncode(object):
             info_dict = json.loads(info)
             return info_dict
 
-    def _smooth_box(self, bbox, height, width):
-        bbox[0] = int(bbox[0] * 1000.0 / width)
-        bbox[2] = int(bbox[2] * 1000.0 / width)
-        bbox[1] = int(bbox[1] * 1000.0 / height)
-        bbox[3] = int(bbox[3] * 1000.0 / height)
-        return bbox
+    def _smooth_box(self, bboxes, height, width):
+        bboxes = np.array(bboxes)
+        bboxes[:, 0] = bboxes[:, 0] * 1000 / width
+        bboxes[:, 2] = bboxes[:, 2] * 1000 / width
+        bboxes[:, 1] = bboxes[:, 1] * 1000 / height
+        bboxes[:, 3] = bboxes[:, 3] * 1000 / height
+        bboxes = bboxes.astype("int64").tolist()
+        return bboxes
 
     def _parse_label(self, label, encode_res):
         gt_label = []
diff --git a/ppocr/metrics/vqa_token_re_metric.py b/ppocr/metrics/vqa_token_re_metric.py
index 8a13bc0812..f84387d8be 100644
--- a/ppocr/metrics/vqa_token_re_metric.py
+++ b/ppocr/metrics/vqa_token_re_metric.py
@@ -37,23 +37,26 @@ class VQAReTokenMetric(object):
         gt_relations = []
         for b in range(len(self.relations_list)):
             rel_sent = []
-            for head, tail in zip(self.relations_list[b]["head"],
-                                  self.relations_list[b]["tail"]):
-                rel = {}
-                rel["head_id"] = head
-                rel["head"] = (self.entities_list[b]["start"][rel["head_id"]],
-                               self.entities_list[b]["end"][rel["head_id"]])
-                rel["head_type"] = self.entities_list[b]["label"][rel[
-                    "head_id"]]
+            if "head" in self.relations_list[b]:
+                for head, tail in zip(self.relations_list[b]["head"],
+                                      self.relations_list[b]["tail"]):
+                    rel = {}
+                    rel["head_id"] = head
+                    rel["head"] = (
+                        self.entities_list[b]["start"][rel["head_id"]],
+                        self.entities_list[b]["end"][rel["head_id"]])
+                    rel["head_type"] = self.entities_list[b]["label"][rel[
+                        "head_id"]]
 
-                rel["tail_id"] = tail
-                rel["tail"] = (self.entities_list[b]["start"][rel["tail_id"]],
-                               self.entities_list[b]["end"][rel["tail_id"]])
-                rel["tail_type"] = self.entities_list[b]["label"][rel[
-                    "tail_id"]]
+                    rel["tail_id"] = tail
+                    rel["tail"] = (
+                        self.entities_list[b]["start"][rel["tail_id"]],
+                        self.entities_list[b]["end"][rel["tail_id"]])
+                    rel["tail_type"] = self.entities_list[b]["label"][rel[
+                        "tail_id"]]
 
-                rel["type"] = 1
-                rel_sent.append(rel)
+                    rel["type"] = 1
+                    rel_sent.append(rel)
             gt_relations.append(rel_sent)
         re_metrics = self.re_score(
             self.pred_relations_list, gt_relations, mode="boundaries")
diff --git a/ppocr/modeling/backbones/vqa_layoutlm.py b/ppocr/modeling/backbones/vqa_layoutlm.py
index 2fd1b1b2a7..34dd9d10ea 100644
--- a/ppocr/modeling/backbones/vqa_layoutlm.py
+++ b/ppocr/modeling/backbones/vqa_layoutlm.py
@@ -43,9 +43,11 @@ class NLPBaseModel(nn.Layer):
         super(NLPBaseModel, self).__init__()
         if checkpoints is not None:
             self.model = model_class.from_pretrained(checkpoints)
+        elif isinstance(pretrained, (str, )) and os.path.exists(pretrained):
+            self.model = model_class.from_pretrained(pretrained)
         else:
             pretrained_model_name = pretrained_model_dict[base_model_class]
-            if pretrained:
+            if pretrained is True:
                 base_model = base_model_class.from_pretrained(
                     pretrained_model_name)
             else:
diff --git a/ppstructure/vqa/tools/trans_funsd_label.py b/ppstructure/vqa/tools/trans_funsd_label.py
new file mode 100644
index 0000000000..ef7d1db010
--- /dev/null
+++ b/ppstructure/vqa/tools/trans_funsd_label.py
@@ -0,0 +1,151 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+import sys
+import cv2
+import numpy as np
+from copy import deepcopy
+
+
+def trans_poly_to_bbox(poly):
+    x1 = np.min([p[0] for p in poly])
+    x2 = np.max([p[0] for p in poly])
+    y1 = np.min([p[1] for p in poly])
+    y2 = np.max([p[1] for p in poly])
+    return [x1, y1, x2, y2]
+
+
+def get_outer_poly(bbox_list):
+    x1 = min([bbox[0] for bbox in bbox_list])
+    y1 = min([bbox[1] for bbox in bbox_list])
+    x2 = max([bbox[2] for bbox in bbox_list])
+    y2 = max([bbox[3] for bbox in bbox_list])
+    return [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+
+
+def load_funsd_label(image_dir, anno_dir):
+    imgs = os.listdir(image_dir)
+    annos = os.listdir(anno_dir)
+
+    imgs = [img.replace(".png", "") for img in imgs]
+    annos = [anno.replace(".json", "") for anno in annos]
+
+    fn_info_map = dict()
+    for anno_fn in annos:
+        res = []
+        with open(os.path.join(anno_dir, anno_fn + ".json"), "r") as fin:
+            infos = json.load(fin)
+            infos = infos["form"]
+            old_id2new_id_map = dict()
+            global_new_id = 0
+            for info in infos:
+                if info["text"] is None:
+                    continue
+                words = info["words"]
+                if len(words) <= 0:
+                    continue
+                word_idx = 1
+                curr_bboxes = [words[0]["box"]]
+                curr_texts = [words[0]["text"]]
+                while word_idx < len(words):
+                    # switch to a new link
+                    if words[word_idx]["box"][0] + 10 <= words[word_idx - 1][
+                            "box"][2]:
+                        if len("".join(curr_texts[0])) > 0:
+                            res.append({
+                                "transcription": " ".join(curr_texts),
+                                "label": info["label"],
+                                "points": get_outer_poly(curr_bboxes),
+                                "linking": info["linking"],
+                                "id": global_new_id,
+                            })
+                            if info["id"] not in old_id2new_id_map:
+                                old_id2new_id_map[info["id"]] = []
+                            old_id2new_id_map[info["id"]].append(global_new_id)
+                            global_new_id += 1
+                        curr_bboxes = [words[word_idx]["box"]]
+                        curr_texts = [words[word_idx]["text"]]
+                    else:
+                        curr_bboxes.append(words[word_idx]["box"])
+                        curr_texts.append(words[word_idx]["text"])
+                    word_idx += 1
+                if len("".join(curr_texts[0])) > 0:
+                    res.append({
+                        "transcription": " ".join(curr_texts),
+                        "label": info["label"],
+                        "points": get_outer_poly(curr_bboxes),
+                        "linking": info["linking"],
+                        "id": global_new_id,
+                    })
+                    if info["id"] not in old_id2new_id_map:
+                        old_id2new_id_map[info["id"]] = []
+                    old_id2new_id_map[info["id"]].append(global_new_id)
+                    global_new_id += 1
+            res = sorted(
+                res, key=lambda r: (r["points"][0][1], r["points"][0][0]))
+            for i in range(len(res) - 1):
+                for j in range(i, 0, -1):
+                    if abs(res[j + 1]["points"][0][1] - res[j]["points"][0][1]) < 20 and \
+                            (res[j + 1]["points"][0][0] < res[j]["points"][0][0]):
+                        tmp = deepcopy(res[j])
+                        res[j] = deepcopy(res[j + 1])
+                        res[j + 1] = deepcopy(tmp)
+                    else:
+                        break
+            # re-generate unique ids
+            for idx, r in enumerate(res):
+                new_links = []
+                for link in r["linking"]:
+                    # illegal links will be removed
+                    if link[0] not in old_id2new_id_map or link[
+                            1] not in old_id2new_id_map:
+                        continue
+                    for src in old_id2new_id_map[link[0]]:
+                        for dst in old_id2new_id_map[link[1]]:
+                            new_links.append([src, dst])
+                res[idx]["linking"] = deepcopy(new_links)
+
+            fn_info_map[anno_fn] = res
+
+    return fn_info_map
+
+
+def main():
+    test_image_dir = "train_data/FUNSD/testing_data/images/"
+    test_anno_dir = "train_data/FUNSD/testing_data/annotations/"
+    test_output_dir = "train_data/FUNSD/test.json"
+
+    fn_info_map = load_funsd_label(test_image_dir, test_anno_dir)
+    with open(test_output_dir, "w") as fout:
+        for fn in fn_info_map:
+            fout.write(fn + ".png" + "\t" + json.dumps(
+                fn_info_map[fn], ensure_ascii=False) + "\n")
+
+    train_image_dir = "train_data/FUNSD/training_data/images/"
+    train_anno_dir = "train_data/FUNSD/training_data/annotations/"
+    train_output_dir = "train_data/FUNSD/train.json"
+
+    fn_info_map = load_funsd_label(train_image_dir, train_anno_dir)
+    with open(train_output_dir, "w") as fout:
+        for fn in fn_info_map:
+            fout.write(fn + ".png" + "\t" + json.dumps(
+                fn_info_map[fn], ensure_ascii=False) + "\n")
+    print("====ok====")
+    return
+
+
+if __name__ == "__main__":
+    main()

From 9a0aa1373ee6cc3acb18a0b3e4ecd9e3b5dc6ed5 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Wed, 6 Jul 2022 11:36:34 +0000
Subject: [PATCH 128/134] add tipc

---
 .../train_infer_python.txt                                      | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
 rename test_tipc/configs/{det_r50_db++ => det_r50_db_plusplus}/train_infer_python.txt (98%)

diff --git a/test_tipc/configs/det_r50_db++/train_infer_python.txt b/test_tipc/configs/det_r50_db_plusplus/train_infer_python.txt
similarity index 98%
rename from test_tipc/configs/det_r50_db++/train_infer_python.txt
rename to test_tipc/configs/det_r50_db_plusplus/train_infer_python.txt
index bcf393a52b..04a3e84585 100644
--- a/test_tipc/configs/det_r50_db++/train_infer_python.txt
+++ b/test_tipc/configs/det_r50_db_plusplus/train_infer_python.txt
@@ -1,5 +1,5 @@
 ===========================train_params===========================
-model_name:det_r50_db++
+model_name:det_r50_db_plusplus
 python:python3.7
 gpu_list:0|0,1
 Global.use_gpu:True|True

From d92fa1c6a555b925fa8ec05fc88066adb18f33e4 Mon Sep 17 00:00:00 2001
From: wangjingyeye <1025993141@qq.com>
Date: Wed, 6 Jul 2022 16:30:41 +0000
Subject: [PATCH 129/134] add tipc

---
 test_tipc/prepare.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 32df8e7865..6387795e8a 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -59,7 +59,7 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
         wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
         cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
     fi
-    if [[ ${model_name} =~ "det_r50_db++" ]];then
+    if [[ ${model_name} =~ "det_r50_db_plusplus" ]];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams --no-check-certificate
     fi
     cd ./pretrain_models/ && tar xf det_mv3_db_v2.0_train.tar && cd ../

From d6b81c6c77b1c00faaf2bb0b2d2927c1304d7c4c Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Thu, 7 Jul 2022 10:12:48 +0800
Subject: [PATCH 130/134] Update README_ch.md

---
 deploy/avh/README_ch.md | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/deploy/avh/README_ch.md b/deploy/avh/README_ch.md
index 7e6272ae12..35cc9f2b7c 100644
--- a/deploy/avh/README_ch.md
+++ b/deploy/avh/README_ch.md
@@ -14,6 +14,7 @@
 <!--- KIND, either express or implied.  See the License for the -->
 <!--- specific language governing permissions and limitations -->
 <!--- under the License. -->
+[English](README.md) | 简体中文
 
 通过TVM在 Arm(R) Cortex(R)-M55 CPU 上运行 PaddleOCR文 本能识别模型
 ===============================================================
@@ -85,9 +86,9 @@ export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/
 模型描述
 -----------------
 
-在这个demo中，我们使用的模型是基于[PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md)的英文识别模型。 PP-OCRv3是[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)发布的PP-OCR系列模型的第三个版本。 该系列模型具有以下特点：
+在这个demo中，我们使用的模型是基于[PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md)的英文识别模型。由于Arm(R) Cortex(R)-M55 CPU不支持rnn算子，我们在PP-OCRv3原始文本识别模型的基础上进行适配，最终模型大小为2.7M。
+
+PP-OCRv3是[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)发布的PP-OCR系列模型的第三个版本，该系列模型具有以下特点：
    - 超轻量级OCR系统：检测（3.6M）+方向分类器（1.4M）+识别（12M）=17.0M。
    - 支持80多种多语言识别模型，包括英文、中文、法文、德文、阿拉伯文、韩文、日文等。 
    - 支持竖排文本识别，长文本识别。
-
-PP-OCRv3 中的文本识别模型支持 80 多种语言。 在模型开发过程中，由于Arm(R) Cortex(R)-M55 CPU不支持rnn算子，我们在PP-OCRv3文本识别模型的基础上删除了不支持的算子，得到当前模型。
\ No newline at end of file

From ca7c4f5d1935243721b33b81a0922416e5dfa4bc Mon Sep 17 00:00:00 2001
From: Evezerest <50011306+Evezerest@users.noreply.github.com>
Date: Thu, 7 Jul 2022 10:15:41 +0800
Subject: [PATCH 131/134] Update README.md

---
 deploy/avh/README.md | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/deploy/avh/README.md b/deploy/avh/README.md
index a64bfd3ad6..0087103146 100644
--- a/deploy/avh/README.md
+++ b/deploy/avh/README.md
@@ -15,6 +15,8 @@
 <!--- specific language governing permissions and limitations -->
 <!--- under the License. -->
 
+English | [简体中文](README_ch.md)
+
 Running PaddleOCR text recognition model on bare metal Arm(R) Cortex(R)-M55 CPU using Arm Virtual Hardware
 ======================================================================
 
@@ -104,9 +106,11 @@ python3 ./convert_image.py path/to/image
 
 Model description
 -----------------
-The example is built on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md) English recognition model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). PP-OCRv3 is the third version of the PP-OCR series model. This series of models has the following features:
+The example is built on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/PP-OCRv3_introduction.md) English recognition model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). Since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current 2.7M English recognition model.
+
+PP-OCRv3 is the third version of the PP-OCR series model. This series of models has the following features:
   - PP-OCRv3: ultra-lightweight OCR system: detection (3.6M) + direction classifier (1.4M) + recognition (12M) = 17.0M
   - Support more than 80 kinds of multi-language recognition models, including English, Chinese, French, German, Arabic, Korean, Japanese and so on. For details
   - Support vertical text recognition, and long text recognition
 
-The text recognition model in PP-OCRv3 supports more than 80 languages. In the process of model development, since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current model.
+

From f6160082105ab364204847402236ae89fb0abe39 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Thu, 7 Jul 2022 11:38:33 +0800
Subject: [PATCH 132/134] fix benchmark train for east

---
 test_tipc/prepare.sh | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 73016ca44a..9c62dc5e2c 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -28,6 +28,10 @@ if [ ${MODE} = "benchmark_train" ];then
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar --no-check-certificate
         cd ./train_data/ && tar xf icdar2015.tar && cd ../
     fi
+    if [[ ${model_name} =~ "det_r50_vd_east_v2_0"; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_r50_vd_east_v2.0_train.tar && cd ../
+    fi
     if [[ ${model_name} =~ "det_r50_vd_east_v2_0" || ${model_name} =~ "det_r50_vd_pse_v2_0" ]];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar --no-check-certificate

From 495c4b6f0e9df0e5e8935bfe94d3a8fc57536c13 Mon Sep 17 00:00:00 2001
From: LDOUBLEV <liuvv0203@outlook.com>
Date: Thu, 7 Jul 2022 13:59:02 +0800
Subject: [PATCH 133/134] fix pretrain

---
 test_tipc/prepare.sh | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
index 9c62dc5e2c..8e1758abb8 100644
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -22,17 +22,19 @@ trainer_list=$(func_parser_value "${lines[14]}")
 
 if [ ${MODE} = "benchmark_train" ];then
     pip install -r requirements.txt
-    if [[ ${model_name} =~ "det_mv3_db_v2_0" || ${model_name} =~ "det_r50_vd_east_v2_0" || ${model_name} =~ "det_r50_vd_pse_v2_0" || ${model_name} =~ "det_r18_db_v2_0" ]];then
+    if [[ ${model_name} =~ "det_mv3_db_v2_0" || ${model_name} =~ "det_r50_vd_pse_v2_0" || ${model_name} =~ "det_r18_db_v2_0" ]];then
         rm -rf ./train_data/icdar2015
         wget -nc -P  ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams  --no-check-certificate
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar --no-check-certificate
         cd ./train_data/ && tar xf icdar2015.tar && cd ../
     fi
-    if [[ ${model_name} =~ "det_r50_vd_east_v2_0"; then
+    if [[ ${model_name} =~ "det_r50_vd_east_v2_0" ]]; then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
         cd ./pretrain_models/ && tar xf det_r50_vd_east_v2.0_train.tar && cd ../
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015.tar && cd ../
     fi
-    if [[ ${model_name} =~ "det_r50_vd_east_v2_0" || ${model_name} =~ "det_r50_vd_pse_v2_0" ]];then
+    if [[ ${model_name} =~ "det_r50_vd_pse_v2_0" ]];then
         wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
         wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar --no-check-certificate
         cd ./train_data/ && tar xf icdar2015.tar && cd ../

From e727b09709ac366e3af58109928885489b975b0d Mon Sep 17 00:00:00 2001
From: zhoujun <zjwenmu@gmail.com>
Date: Thu, 7 Jul 2022 15:40:23 +0800
Subject: [PATCH 134/134] fix config (#6823)

---
 configs/table/table_master.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/configs/table/table_master.yml b/configs/table/table_master.yml
index cfd98346c4..1e6efe32db 100755
--- a/configs/table/table_master.yml
+++ b/configs/table/table_master.yml
@@ -104,8 +104,8 @@ Train:
 Eval:
   dataset:
     name: PubTabDataSet
-    data_dir: /home/zhoujun20/table/PubTabNe/pubtabnet/val/
-    label_file_list: [/home/zhoujun20/table/PubTabNe/pubtabnet/val_500.jsonl]
+    data_dir: train_data/table/pubtabnet/train/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_val.jsonl]
     transforms:
       - DecodeImage:
           img_mode: BGR