HTR/hcr.py

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 utils import notification_queue, log_print
from all_models import models

def text(image_cv):
    try:
        # Get model instance from singleton
        model, processor = models.get_trocr_model()
            
        if not isinstance(image_cv, list):
            image_cv = [image_cv]
            
        t = ""
        total_images = len(image_cv)
        log_print(f"Processing {total_images} image(s) for text extraction")
        
        for i, img in enumerate(image_cv):
            try:
                log_print(f"Processing image {i+1}/{total_images}")
                
                # Validate image
                if img is None:
                    log_print(f"Skipping image {i+1} - Image is None", "WARNING")
                    continue
                    
                # Convert to RGB
                img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
                image = Image.fromarray(img_rgb)
                
                # Get pixel values
                pixel_values = processor(image, return_tensors="pt").pixel_values
                if torch.cuda.is_available():
                    pixel_values = pixel_values.to(models.device)
                    
                # Generate text
                generated_ids = model.generate(pixel_values)
                generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
                
                # Clean up the text
                cleaned_text = generated_text.replace(" ", "")
                t = t + cleaned_text + " "
                
                log_print(f"Successfully extracted text from image {i+1}: {cleaned_text}")
                
                # Clean up CUDA memory
                if torch.cuda.is_available():
                    del pixel_values
                    del generated_ids
                    torch.cuda.empty_cache()
                    
            except Exception as e:
                log_print(f"Error processing image {i+1}: {str(e)}", "ERROR")
                continue
                
        return t.strip()
        
    except Exception as e:
        error_msg = f"Error in text function: {str(e)}"
        log_print(error_msg, "ERROR")
        notification_queue.put({
            "type": "error",
            "message": error_msg
        })
        return ""
    finally:
        # Release model reference
        models.release_trocr_model()