WIP

packages/training/training/cli/analyze_labels.py

@@ -20,7 +20,7 @@ from shared.config import get_db_connection_string
 from shared.normalize import normalize_field
 from shared.matcher import FieldMatcher
 from shared.pdf import is_text_pdf, extract_text_tokens
-from training.yolo.annotation_generator import FIELD_CLASSES
+from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
 from shared.data.db import DocumentDB
 
 

packages/training/training/cli/autolabel.py

@@ -113,7 +113,7 @@ def process_single_document(args_tuple):
     # Import inside worker to avoid pickling issues
     from training.data.autolabel_report import AutoLabelReport
     from shared.pdf import PDFDocument
-    from training.yolo.annotation_generator import FIELD_CLASSES
+    from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
     from training.processing.document_processor import process_page, record_unmatched_fields
 
     start_time = time.time()
@@ -342,7 +342,8 @@ def main():
     from shared.ocr import OCREngine
     from shared.matcher import FieldMatcher
     from shared.normalize import normalize_field
-    from training.yolo.annotation_generator import AnnotationGenerator, FIELD_CLASSES
+    from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
+    from training.yolo.annotation_generator import AnnotationGenerator
 
     # Handle comma-separated CSV paths
     csv_input = args.csv

packages/training/training/processing/autolabel_tasks.py

@@ -90,7 +90,7 @@ def process_text_pdf(task_data: Dict[str, Any]) -> Dict[str, Any]:
     import shutil
     from training.data.autolabel_report import AutoLabelReport
     from shared.pdf import PDFDocument
-    from training.yolo.annotation_generator import FIELD_CLASSES
+    from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
     from training.processing.document_processor import process_page, record_unmatched_fields
 
     row_dict = task_data["row_dict"]
@@ -208,7 +208,7 @@ def process_scanned_pdf(task_data: Dict[str, Any]) -> Dict[str, Any]:
     import shutil
     from training.data.autolabel_report import AutoLabelReport
     from shared.pdf import PDFDocument
-    from training.yolo.annotation_generator import FIELD_CLASSES
+    from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
     from training.processing.document_processor import process_page, record_unmatched_fields
 
     row_dict = task_data["row_dict"]

packages/training/training/processing/document_processor.py

@@ -15,7 +15,8 @@ from training.data.autolabel_report import FieldMatchResult
 from shared.matcher import FieldMatcher
 from shared.normalize import normalize_field
 from shared.ocr.machine_code_parser import MachineCodeParser
-from training.yolo.annotation_generator import AnnotationGenerator, FIELD_CLASSES
+from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
+from training.yolo.annotation_generator import AnnotationGenerator
 
 
 def match_supplier_accounts(

packages/training/training/yolo/annotation_generator.py

@@ -9,43 +9,12 @@ from pathlib import Path
 from typing import Any
 import csv
 
-
-# Field class mapping for YOLO
-# Note: supplier_accounts is not a separate class - its matches are mapped to Bankgiro/Plusgiro
-FIELD_CLASSES = {
-    'InvoiceNumber': 0,
-    'InvoiceDate': 1,
-    'InvoiceDueDate': 2,
-    'OCR': 3,
-    'Bankgiro': 4,
-    'Plusgiro': 5,
-    'Amount': 6,
-    'supplier_organisation_number': 7,
-    'customer_number': 8,
-    'payment_line': 9,  # Machine code payment line at bottom of invoice
-}
-
-# Fields that need matching but map to other YOLO classes
-# supplier_accounts matches are classified as Bankgiro or Plusgiro based on account type
-ACCOUNT_FIELD_MAPPING = {
-    'supplier_accounts': {
-        'BG': 'Bankgiro',  # BG:xxx -> Bankgiro class
-        'PG': 'Plusgiro',  # PG:xxx -> Plusgiro class
-    }
-}
-
-CLASS_NAMES = [
-    'invoice_number',
-    'invoice_date',
-    'invoice_due_date',
-    'ocr_number',
-    'bankgiro',
-    'plusgiro',
-    'amount',
-    'supplier_org_number',
-    'customer_number',
-    'payment_line',  # Machine code payment line at bottom of invoice
-]
+# Import field mappings from single source of truth
+from shared.fields import (
+    TRAINING_FIELD_CLASSES as FIELD_CLASSES,
+    CLASS_NAMES,
+    ACCOUNT_FIELD_MAPPING,
+)
 
 
 @dataclass

packages/training/training/yolo/dataset_builder.py

@@ -101,7 +101,8 @@ class DatasetBuilder:
         Returns:
             DatasetStats with build results
         """
-        from .annotation_generator import AnnotationGenerator, CLASS_NAMES
+        from shared.fields import CLASS_NAMES
+        from .annotation_generator import AnnotationGenerator
 
         random.seed(seed)
 

packages/training/training/yolo/db_dataset.py

@@ -18,7 +18,8 @@ import numpy as np
 from PIL import Image
 
 from shared.config import DEFAULT_DPI
-from .annotation_generator import FIELD_CLASSES, YOLOAnnotation
+from shared.fields import TRAINING_FIELD_CLASSES as FIELD_CLASSES
+from .annotation_generator import YOLOAnnotation
 
 logger = logging.getLogger(__name__)