from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
import cv2
import os
import torch
import sys
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from main import notification_queue

# Initialize model and processor at module load time
MODEL_NAME = "microsoft/trocr-large-handwritten"
try:
    notification_queue.put({
        "type": "info",
        "message": "Initializing TrOCR model..."
    })
    processor = TrOCRProcessor.from_pretrained(MODEL_NAME)
    model = VisionEncoderDecoderModel.from_pretrained(MODEL_NAME)
    if torch.cuda.is_available():
        model = model.to('cuda')
    notification_queue.put({
        "type": "success",
        "message": "TrOCR model initialized successfully"
    })
except Exception as e:
    error_msg = str(e)
    notification_queue.put({
        "type": "error",
        "message": f"Error initializing TrOCR model: {error_msg}"
    })
    processor = None
    model = None

def text(image_cv):
    try:
        # Check if model is initialized
        if processor is None or model is None:
            notification_queue.put({
                "type": "error",
                "message": "TrOCR model not initialized properly"
            })
            return ""
            
        if not isinstance(image_cv, list):
            image_cv = [image_cv]
            
        t = ""
        total_images = len(image_cv)
        notification_queue.put({
            "type": "info",
            "message": f"Processing {total_images} image(s)..."
        })
        for i in image_cv:
            img_rgb = cv2.cvtColor(i, cv2.COLOR_BGR2RGB)
            image = Image.fromarray(img_rgb)

            pixel_values = processor(image, return_tensors="pt").pixel_values
            generated_ids = model.generate(pixel_values)

            generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
            t = t + generated_text.replace(" ", "") + " "
        return t
        
    except Exception as e:
        error_msg = str(e)
        notification_queue.put({
            "type": "error",
            "message": f"Error in text function: {error_msg}"
        })
        return ""