Update app.py

app.py

@@ -1,19 +1,32 @@
 import fitz  # PyMuPDF for PDF processing
 from PIL import Image
-from transformers import pipeline
+from transformers import pipeline, AutoProcessor, AutoModelForImageTextToText
 import streamlit as st
 import os
 import re
 from docx import Document
 from langdetect import detect
 
-trocr_pipeline = pipeline("image-to-text", model="microsoft/trocr-base-printed")
+# Load Qwen model for image-to-text
+processor = AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct")
+model = AutoModelForImageTextToText.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct")
+
+# Load translation model
 translator = pipeline("translation", model="Helsinki-NLP/opus-mt-mul-en")
 
 
 def extract_text_from_image(image):
-    result = trocr_pipeline(image)
-    return result[0]['generated_text'] if result else ""
+    """Extract text from image using Qwen and TrOCR fallback."""
+    # Convert image to RGB format for processing
+    image = image.convert("RGB")
+
+    # Use Qwen processor and model
+    pixel_values = processor(image, return_tensors="pt").pixel_values
+    result = model.generate(pixel_values)
+    decoded_text = processor.batch_decode(result, skip_special_tokens=True)[0]
+    
+    # Ensure extracted text is clean
+    return decoded_text.strip()
 
 
 def extract_from_pdf(pdf_path):
@@ -47,7 +60,7 @@ def translate_text(text):
     if detected_language == "en":
         return "The text is already in English."
 
-    chunks = [text[i:i + 15000] for i in range(0, len(text), 15000)]
+    chunks = [text[i:i + 500] for i in range(0, len(text), 500)]
     translated_text = ""
     for chunk in chunks:
         translated_chunk = translator(chunk, max_length=400)