mirror of https://github.com/PaddlePaddle/PaddleOCR.git synced 2026-08-29 04:12:51 +08:00

Files

T

History

Hinotobi f520ef4b65 fix: use safe YAML loaders for config parsing (#17930 )

* fix: use safe yaml loaders for config parsing

* test: restore YAML stubs after tests

---------

Co-authored-by: Tao Luo <luotao02@baidu.com>
Co-authored-by: Lin Manhui <mhlin425@whu.edu.cn>

2026-05-14 16:13:17 +08:00

PaddleOCR_DBNet

fix: use safe YAML loaders for config parsing (#17930 )

2026-05-14 16:13:17 +08:00

analysis.py

add pre-commit workflow (#11973 )

2024-04-21 21:46:20 +08:00

readme.md

benchmark

2022-02-21 19:40:26 +08:00

run_benchmark_det.sh

update common pre-commit configs and commit the results of running pre-commit run -a (#12516 )

2024-05-29 15:26:09 +08:00

run_det.sh

update common pre-commit configs and commit the results of running pre-commit run -a (#12516 )

2024-05-29 15:26:09 +08:00

readme.md

PaddleOCR DB/EAST/PSE 算法训练benchmark测试

PaddleOCR/benchmark目录下的文件用于获取并分析训练日志。训练采用icdar2015数据集，包括1000张训练图像和500张测试图像。模型配置采用resnet18_vd作为backbone，分别训练batch_size=8和batch_size=16的情况。

运行训练benchmark

benchmark/run_det.sh 中包含了三个过程：

安装依赖
下载数据
执行训练
日志分析获取IPS

在执行训练部分，会执行单机单卡（默认0号卡）单机多卡训练，并分别执行batch_size=8和batch_size=16的情况。所以执行完后，每种模型会得到4个日志文件。

run_det.sh 执行方式如下:

# cd PaddleOCR/
bash benchmark/run_det.sh

以DB为例，将得到四个日志文件，如下：

det_res18_db_v2.0_sp_bs16_fp32_1
det_res18_db_v2.0_sp_bs8_fp32_1
det_res18_db_v2.0_mp_bs16_fp32_1
det_res18_db_v2.0_mp_bs8_fp32_1