app.py

import gradio as gr
from paddleocr import PaddleOCR
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from PIL import Image
import os

# --- مدل‌ها ---
try:
    model_name = "m3hrdadfi/mt5-small-finetuned-grammar-synthesis"
    
    ocr_model = PaddleOCR(lang='fa', use_textline_orientation=True)
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    nlp_model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    
except Exception as e:
    raise gr.Error(f"خطا در بارگذاری مدل‌ها: لطفاً این خطا را به توسعه دهنده گزارش دهید. خطا: {str(e)}")

# --- توابع پردازش ---
def run_ocr(image):
    image_path = image.name
    result = ocr_model.ocr(image_path, cls=True)
    texts = [line[1][0] for line in result[0]] if result else []
    os.remove(image_path)
    return " ".join(texts)

def postprocess_text(text):
    inputs = tokenizer(text, return_tensors="pt", max_length=512, truncation=True)
    outputs = nlp_model.generate(**inputs)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# --- رابط کاربری ---
with gr.Blocks() as app:
    gr.Markdown("## سیستم OCR فارسی با پردازش پیشرفته متن")
    
    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="filepath", label="تصویر ورودی")
            process_btn = gr.Button("پردازش تصویر")
        
        with gr.Column():
            raw_output = gr.Textbox(label="متن استخراج شده")
            processed_output = gr.Textbox(label="متن پردازش شده")
    
    def process_image(img):
        # Avoid running OCR twice
        raw = run_ocr(img)
        processed = postprocess_text(raw)
        return raw, processed

    process_btn.click(
        fn=process_image,
        inputs=image_input,
        outputs=[raw_output, processed_output]
    )

if __name__ == "__main__":
    app.launch()