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

fix: single dollar sysmpol

Browse Source
This commit is contained in:
yoge
2026-03-12 23:20:14 +08:00
parent ff82021467
commit aee1a1bf3b
3 changed files with 7 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
app/core/config.py
View File

@@ -50,9 +50,7 @@ class Settings(BaseSettings):
max_tokens: int = 4096 max_tokens: int = 4096
# Model Paths # Model Paths
pp_doclayout_model_dir: str | None = ( pp_doclayout_model_dir: str | None = "/home/yoge/.cache/modelscope/hub/models/PaddlePaddle/PP-DocLayoutV3"
"/home/yoge/.cache/modelscope/hub/models/PaddlePaddle/PP-DocLayoutV3"
)
# Image Processing # Image Processing
max_image_size_mb: int = 10 max_image_size_mb: int = 10

2
app/services/glm_postprocess.py
View File

@@ -265,7 +265,7 @@ class GLMResultFormatter:
# Formula wrapping # Formula wrapping
if label == "formula": if label == "formula":
content = content.strip() content = content.strip()
for s, e in [("$$", "$$"), (r"\[", r"\]"), (r"\(", r"\)")]: for s, e in [("$$", "$$"), (r"\[", r"\]"), (r"\(", r"\)"), ("$", "$")]:
if content.startswith(s): if content.startswith(s):
content = content[len(s) :].strip() content = content[len(s) :].strip()
if content.endswith(e): if content.endswith(e):

9
app/services/ocr_service.py
View File

@@ -781,11 +781,11 @@ class MineruOCRService(OCRServiceBase):
# Task-specific prompts (from GLM-OCR SDK config.yaml) # Task-specific prompts (from GLM-OCR SDK config.yaml)
_TASK_PROMPTS: dict[str, str] = { _TASK_PROMPTS: dict[str, str] = {
"text": "Text Recognition. If the content is a formula, please ouput display latex code, else output text", "text": "Text Recognition. If the content is a formula, please output display latex code, else output text",
"formula": "Formula Recognition:", "formula": "Formula Recognition:",
"table": "Table Recognition:", "table": "Table Recognition:",
} }
_DEFAULT_PROMPT = "Text Recognition. If the content is a formula, please ouput display latex code, else output text" _DEFAULT_PROMPT = "Text Recognition. If the content is a formula, please output display latex code, else output text"
class GLMOCREndToEndService(OCRServiceBase): class GLMOCREndToEndService(OCRServiceBase):
@@ -868,7 +868,8 @@ class GLMOCREndToEndService(OCRServiceBase):
""" """
# 1. Layout detection # 1. Layout detection
img_h, img_w = image.shape[:2] img_h, img_w = image.shape[:2]
layout_info = self.layout_detector.detect(image) padded_image = self.image_processor.add_padding(image)
layout_info = self.layout_detector.detect(padded_image)
# Sort regions in reading order: top-to-bottom, left-to-right # Sort regions in reading order: top-to-bottom, left-to-right
layout_info.regions.sort(key=lambda r: (r.bbox[1], r.bbox[0])) layout_info.regions.sort(key=lambda r: (r.bbox[1], r.bbox[0]))
@@ -890,7 +891,7 @@ class GLMOCREndToEndService(OCRServiceBase):
if region.type == "figure": if region.type == "figure":
continue continue
x1, y1, x2, y2 = (int(c) for c in region.bbox) x1, y1, x2, y2 = (int(c) for c in region.bbox)
cropped = image[y1:y2, x1:x2] cropped = padded_image[y1:y2, x1:x2]
if cropped.size == 0 or cropped.shape[0] < 10 or cropped.shape[1] < 10: if cropped.size == 0 or cropped.shape[0] < 10 or cropped.shape[1] < 10:
logger.warning( logger.warning(
"Skipping region idx=%d (label=%s): crop too small %s", "Skipping region idx=%d (label=%s): crop too small %s",