opensource/PaddleOCR
3
0
Fork 0
mirror of https://github.com/PaddlePaddle/PaddleOCR.git synced 2026-08-30 17:23:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix crop_margin crash with extreme aspect ratio images (#17686)

Browse Source 
Add bounds checking to crop_margin in GoTImgDecode, UniMERNetImgDecode,
and UniMERNetResize to handle edge cases:

- Return original image when cv2.findNonZero returns None (no text found)
- Return original image when bounding rect has zero width or height
- Return original image when cropped result would have aspect ratio > 200,
  which causes ValueError in downstream image processing

Fixes PaddlePaddle#17354

Co-authored-by: Lin Manhui <mhlin425@whu.edu.cn>
This commit is contained in:
Harikrishna KP
2026-02-28 08:55:55 +05:30
committed by GitHub
parent b6a561da6c
commit 925e2ed400
1 changed files with 21 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ppocr/data/imaug/unimernet_aug.py
+21
View File
@@ -495,7 +495,14 @@ class GoTImgDecode:
data = (data - min_val) / (max_val - min_val) * 255
gray = 255 * (data < 200).astype(np.uint8)
coords = cv2.findNonZero(gray) # Find all non-zero points (text)
if coords is None:
return img
a, b, w, h = cv2.boundingRect(coords) # Find minimum spanning bounding box
if w == 0 or h == 0:
return img
# Avoid extreme aspect ratios that cause errors in downstream processing
if max(w, h) / min(w, h) > 200:
return img
return img.crop((a, b, w + a, h + b))
def get_dimensions(self, img):
@@ -595,7 +602,14 @@ class UniMERNetImgDecode:
data = (data - min_val) / (max_val - min_val) * 255
gray = 255 * (data < 200).astype(np.uint8)
coords = cv2.findNonZero(gray) # Find all non-zero points (text)
if coords is None:
return img
a, b, w, h = cv2.boundingRect(coords) # Find minimum spanning bounding box
if w == 0 or h == 0:
return img
# Avoid extreme aspect ratios that cause errors in downstream processing
if max(w, h) / min(w, h) > 200:
return img
return img.crop((a, b, w + a, h + b))
def get_dimensions(self, img):
@@ -720,7 +734,14 @@ class UniMERNetResize:
gray = 255 * (data < 200).astype(np.uint8)
coords = cv2.findNonZero(gray) # Find all non-zero points (text)
if coords is None:
return img
a, b, w, h = cv2.boundingRect(coords) # Find minimum spanning bounding box
if w == 0 or h == 0:
return img
# Avoid extreme aspect ratios that cause errors in downstream processing
if max(w, h) / min(w, h) > 200:
return img
return img.crop((a, b, w + a, h + b))
def get_dimensions(self, img):