ocr_cpu.py

import os
from transformers import AutoModel, AutoTokenizer
import torch
import re

# Load model and tokenizer
model_name = "srimanth-d/GOT_CPU"  # Using GOT model on CPU
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, return_tensors='pt')

# Load the model
model = AutoModel.from_pretrained(
    model_name,
    trust_remote_code=True,
    low_cpu_mem_usage=True,
    use_safetensors=True,
    pad_token_id=tokenizer.eos_token_id,
)

# Ensure the model is in evaluation mode and loaded on CPU
device = torch.device("cpu")
model = model.eval()

# OCR function to extract text
def extract_text_got(uploaded_file):
    """Use GOT-OCR2.0 model to extract text from the uploaded image."""
    temp_file_path = 'temp_image.jpg'

    try:
        # Save the uploaded file temporarily
        with open(temp_file_path, 'wb') as temp_file:
            temp_file.write(uploaded_file.read())

        print(f"Processing image from path: {temp_file_path}")

        ocr_types = ['ocr', 'format']
        results = []

        # Run OCR on the image
        for ocr_type in ocr_types:
            with torch.no_grad():
                print(f"Running OCR with type: {ocr_type}")
                outputs = model.chat(tokenizer, temp_file_path, ocr_type=ocr_type)

                if isinstance(outputs, list) and outputs[0].strip():
                    return outputs[0].strip()  # Return the result if successful
                results.append(outputs[0].strip() if outputs else "No result")

        # Combine results or return no text found message
        return results[0] if results else "No text extracted."

    except Exception as e:
        return f"Error during text extraction: {str(e)}"

    finally:
        # Clean up temporary file
        if os.path.exists(temp_file_path):
            os.remove(temp_file_path)
            print(f"Temporary file {temp_file_path} removed.")

# Function to clean extracted text (removes extra spaces and handles special cases for Hindi and English)
def clean_text(extracted_text):
    """
    Cleans extracted text by removing extra spaces and handling language-specific issues (Hindi, English, Hinglish).
    """
    # Normalize spaces (remove multiple spaces)
    text = re.sub(r'\s+', ' ', extracted_text)

    # Handle special cases based on Hindi, English, and Hinglish patterns
    text = re.sub(r'([a-zA-Z]+)\s+([a-zA-Z]+)', r'\1 \2', text)  # For English
    text = re.sub(r'([ा-ह]+)\s+([ा-ह]+)', r'\1\2', text)  # For Hindi (conjoining Devanagari characters)
    
    # Remove trailing and leading spaces
    return text.strip()