app.py

import streamlit as st
from transformers import PaliGemmaProcessor, PaliGemmaForConditionalGeneration
from PIL import Image
import torch
import os

def load_model():
    """Load PaliGemma2 model and processor."""
    processor = PaliGemmaProcessor.from_pretrained("google/paligemma2")
    model = PaliGemmaForConditionalGeneration.from_pretrained("google/paligemma2")
    return processor, model

def process_image(image, processor, model):
    """Extract text from image using PaliGemma2."""
    # Preprocess the image
    inputs = processor(images=image, return_tensors="pt")
    
    # Generate predictions
    with torch.no_grad():
        generated_ids = model.generate(**inputs)
        text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    
    return text

def main():
    # Set page configuration
    st.set_page_config(page_title="Text Reading with PaliGemma2", layout="centered")
    st.title("Text Reading from Images using PaliGemma2")

    # Load model and processor
    with st.spinner("Loading PaliGemma2 model... This may take a few moments."):
        processor, model = load_model()
    st.success("Model loaded successfully!")

    # User input: upload image
    uploaded_image = st.file_uploader("Upload an image containing text", type=["png", "jpg", "jpeg"])

    if uploaded_image is not None:
        # Display uploaded image
        image = Image.open(uploaded_image)
        st.image(image, caption="Uploaded Image", use_column_width=True)

        # Extract text button
        if st.button("Extract Text"):
            with st.spinner("Processing image..."):
                extracted_text = process_image(image, processor, model)
                st.success("Text extraction complete!")
                st.subheader("Extracted Text")
                st.write(extracted_text)

    # Footer
    st.markdown("---")
    st.markdown("**Built with [PaliGemma2](https://huggingface.co/google/paligemma2) and Streamlit**")

if __name__ == "__main__":
    main()