Update ocr_engine.py

ocr_engine.py

@@ -1,22 +1,27 @@
 from transformers import TrOCRProcessor, VisionEncoderDecoderModel
 from PIL import Image
 
-# Load model + processor once
+# Load model + processor
 processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-stage1")
 model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-stage1")
 
 def extract_weight(image: Image.Image) -> str:
-    # Convert image to RGB just in case
     image = image.convert("RGB")
-
-    # Preprocess + generate prediction
     pixel_values = processor(images=image, return_tensors="pt").pixel_values
     generated_ids = model.generate(pixel_values)
+    full_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
 
-    # Decode output
-    text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+    # Extract digits
+    weight = ''.join(filter(lambda x: x in '0123456789.', full_text))
 
-    # Keep only numbers + decimal
-    weight = ''.join(filter(lambda x: x in '0123456789.', text))
+    # Check for unit in original OCR text
+    unit = "grams"  # default
+    if "kg" in full_text.lower():
+        unit = "kg"
+    elif "g" in full_text.lower():
+        unit = "grams"
 
-    return weight.strip() if weight else "No valid weight detected"
+    if weight:
+        return f"{weight} {unit}"
+    else:
+        return "No valid weight detected"