ocr_engine.py

from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image

# Load model + processor once
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)

    # Decode output
    text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

    # Keep only numbers + decimal
    weight = ''.join(filter(lambda x: x in '0123456789.', text))

    return weight.strip() if weight else "No valid weight detected"