from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image, ImageFilter
import torch
import re

# Load TrOCR model
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")

def clean_ocr_text(text):
    print("[RAW OCR]", text)
    # Fix common misreads
    text = text.replace(",", ".").replace("s", "5").replace("o", "0").replace("O", "0")
    text = re.sub(r"[^\d\.kg]", "", text.lower())  # Keep digits, dot, kg
    print("[CLEANED OCR]", text)
    return text

def preprocess_image(image):
    # Enlarge + sharpen for better OCR
    image = image.resize((image.width * 2, image.height * 2), Image.BICUBIC)
    image = image.filter(ImageFilter.SHARPEN)
    return image

def extract_weight(image):
    try:
        image = preprocess_image(image)
        pixel_values = processor(images=image, return_tensors="pt").pixel_values
        generated_ids = model.generate(pixel_values)
        raw_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()

        cleaned = clean_ocr_text(raw_text)

        # Try matching decimal weight with unit
        match = re.search(r'(\d{1,3}\.\d{1,3})\s*(kg|g)', cleaned)
        if match:
            return f"{match.group(1)} {match.group(2)}"

        # Fallback: match only decimal number
        fallback = re.search(r'(\d{1,3}\.\d{1,3})', cleaned)
        if fallback:
            return f"{fallback.group(1)} g"  # Default to grams

        return f"No valid weight found | OCR: {cleaned}"
    except Exception as e:
        return f"Error: {str(e)}"