streamlit_app.py

import streamlit as st
from transformers import AutoProcessor, AutoModelForVision2Seq
from PIL import Image
import torch
import io

# Load model and processor once
@st.cache_resource
def load_model():
    model_id = "HuggingFaceTB/SmolVLM2-2.2B-Instruct"
    processor = AutoProcessor.from_pretrained(model_id)
    model = AutoModelForVision2Seq.from_pretrained(model_id).to("cuda" if torch.cuda.is_available() else "cpu")
    return processor, model

processor, model = load_model()

# Streamlit UI
st.title("Aadhaar Card Information Extractor")
uploaded_file = st.file_uploader("Upload Aadhaar card image", type=["jpg", "png", "jpeg"])

if uploaded_file is not None:
    image = Image.open(uploaded_file).convert("RGB")
    st.image(image, caption="Uploaded Aadhaar Card", use_column_width=True)

    if st.button("Extract Info"):
        with st.spinner("Extracting..."):
            prompt = (
                "You are an AI system for extracting information from Indian Aadhaar cards. "
                "From the image, extract and return a structured JSON with:\n"
                "- Name\n"
                "- Father's Name\n"
                "- Date of Birth\n"
                "- Gender\n"
                "- Aadhaar Number\n"
                "- Address (Street, Locality, District, State, PIN)\n"
                "- QR code data (if visible)\n"
                "- Bounding box of photograph as [x1, y1, x2, y2]\n"
                "Respond only with JSON."
            )

            inputs = processor(images=image, text=prompt, return_tensors="pt").to(model.device)
            outputs = model.generate(**inputs, max_new_tokens=512)
            result = processor.batch_decode(outputs, skip_special_tokens=True)[0]
            st.code(result, language="json")