YogeLiu/doc_processer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: optimize padding and formula fallback

Browse Source
This commit is contained in:
liuyuanchuang
2026-03-10 09:54:54 +08:00
parent cff14904bf
commit f8173f7c0a
2 changed files with 13 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
app/services/image_processor.py
View File

@@ -104,7 +104,8 @@ class ImageProcessor:
"""Add whitespace padding around the image. """Add whitespace padding around the image.
Adds padding equal to padding_ratio * max(height, width) on each side. Adds padding equal to padding_ratio * max(height, width) on each side.
This expands the image by approximately 30% total (15% on each side). For small images (height < 80 or width < 500), uses reduced padding_ratio 0.2.
This expands the image by approximately 30% total (15% on each side) for normal images.
Args: Args:
image: Input image as numpy array in BGR format. image: Input image as numpy array in BGR format.
@@ -113,7 +114,9 @@ class ImageProcessor:
Padded image as numpy array. Padded image as numpy array.
""" """
height, width = image.shape[:2] height, width = image.shape[:2]
padding = int(max(height, width) * self.padding_ratio) # Use smaller padding ratio for small images to preserve detail
padding_ratio = 0.2 if height < 80 or width < 500 else self.padding_ratio
padding = int(max(height, width) * padding_ratio)
# Add white padding on all sides # Add white padding on all sides
padded_image = cv2.copyMakeBorder( padded_image = cv2.copyMakeBorder(

10
app/services/ocr_service.py
View File

@@ -890,8 +890,14 @@ class GLMOCREndToEndService(OCRServiceBase):
# 3. OCR: per-region (parallel) or full-image fallback # 3. OCR: per-region (parallel) or full-image fallback
if not layout_info.regions: if not layout_info.regions:
raw_content = self._call_vllm(padded, _DEFAULT_PROMPT) # No layout detected → assume it's a formula, use formula recognition
markdown_content = self._formatter._clean_content(raw_content) logger.info("No layout regions detected, treating image as formula")
raw_content = self._call_vllm(padded, _TASK_PROMPTS["formula"])
# Format as display formula markdown
formatted_content = raw_content.strip()
if not (formatted_content.startswith("$$") and formatted_content.endswith("$$")):
formatted_content = f"$$\n{formatted_content}\n$$"
markdown_content = formatted_content
else: else:
# Build task list for non-figure regions # Build task list for non-figure regions
tasks = [] tasks = []