kai/invoice-master-poc-v2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

WIP

Browse Source
This commit is contained in:
Yaojia Wang
2026-02-11 23:40:38 +01:00
parent f1a7bfe6b7
commit ad5ed46b4c
117 changed files with 5741 additions and 7669 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
packages/backend/backend/pipeline/field_extractor.py
View File

@@ -40,6 +40,7 @@ from .normalizers import (
EnhancedAmountNormalizer,
EnhancedDateNormalizer,
)
from .value_selector import ValueSelector
@dataclass
@@ -169,13 +170,21 @@ class FieldExtractor:
overlap_ratio = overlap_area / token_area if token_area > 0 else 0
matching_tokens.append((token, overlap_ratio))
# Sort by overlap ratio and combine text
# Sort by overlap ratio
matching_tokens.sort(key=lambda x: -x[1])
raw_text = ' '.join(t[0].text for t in matching_tokens)
# Get field name
field_name = CLASS_TO_FIELD.get(detection.class_name, detection.class_name)
# Convert to OCRTokens for value selection, then filter
from shared.ocr.paddle_ocr import OCRToken
pdf_ocr_tokens = [
OCRToken(text=t[0].text, bbox=t[0].bbox, confidence=1.0)
for t in matching_tokens
]
value_tokens = ValueSelector.select_value_tokens(pdf_ocr_tokens, field_name)
raw_text = ' '.join(t.text for t in value_tokens)
# Normalize and validate
normalized_value, is_valid, validation_error = self._normalize_and_validate(
field_name, raw_text
@@ -223,13 +232,14 @@ class FieldExtractor:
# Run OCR on region
ocr_tokens = self.ocr_engine.extract_from_image(region)
# Combine all OCR text
raw_text = ' '.join(t.text for t in ocr_tokens)
ocr_confidence = sum(t.confidence for t in ocr_tokens) / len(ocr_tokens) if ocr_tokens else 0.0
# Get field name
field_name = CLASS_TO_FIELD.get(detection.class_name, detection.class_name)
# Select value tokens (filter out label text)
value_tokens = ValueSelector.select_value_tokens(ocr_tokens, field_name)
raw_text = ' '.join(t.text for t in value_tokens)
ocr_confidence = sum(t.confidence for t in ocr_tokens) / len(ocr_tokens) if ocr_tokens else 0.0
# Normalize and validate
normalized_value, is_valid, validation_error = self._normalize_and_validate(
field_name, raw_text