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

optimize: formula is recognize text

Browse Source
This commit is contained in:
yoge
2026-03-12 22:30:27 +08:00
parent 11e9ed780d
commit ff82021467
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
app/services/layout_detector.py
View File

@@ -148,7 +148,7 @@ class LayoutDetector:
) )
) )
mixed_recognition = any(region.type == "text" and region.score > 0.3 for region in regions) mixed_recognition = any(region.type == "text" and region.score > 0.85 for region in regions)
return LayoutInfo(regions=regions, MixedRecognition=mixed_recognition) return LayoutInfo(regions=regions, MixedRecognition=mixed_recognition)

6
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 latex code, else output text", "text": "Text Recognition. If the content is a formula, please ouput 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 latex code, else output text" _DEFAULT_PROMPT = "Text Recognition. If the content is a formula, please ouput display latex code, else output text"
class GLMOCREndToEndService(OCRServiceBase): class GLMOCREndToEndService(OCRServiceBase):
@@ -874,7 +874,7 @@ class GLMOCREndToEndService(OCRServiceBase):
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]))
# 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 or (len(layout_info.regions) == 1 and not layout_info.MixedRecognition):
# No layout detected → assume it's a formula, use formula recognition # No layout detected → assume it's a formula, use formula recognition
logger.info("No layout regions detected, treating image as formula") logger.info("No layout regions detected, treating image as formula")
raw_content = self._call_vllm(image, _TASK_PROMPTS["formula"]) raw_content = self._call_vllm(image, _TASK_PROMPTS["formula"])