app.py

import streamlit as st
from transformers import pipeline
from PIL import Image, ImageDraw
import io

# -------------------------
# Load models once (cached)
# -------------------------
@st.cache_resource
def load_pipelines():
    captioner = pipeline("image-to-text", model="nlpconnect/vit-gpt2-image-captioning")
    detector = pipeline("object-detection", model="facebook/detr-resnet-50")
    vqa = pipeline("visual-question-answering", model="Salesforce/blip-vqa-base")
    return captioner, detector, vqa

captioner, detector, vqa = load_pipelines()

# -------------------------
# Streamlit UI
# -------------------------
st.set_page_config(page_title="Vision Chatbot+", page_icon="🖼️")
st.title("🖼️ Vision Chatbot+")
st.write("Upload an image to get **captions, emojis, object detection (with boxes!), and Q&A**")

uploaded_file = st.file_uploader("Upload an image...", type=["jpg", "jpeg", "png"])

if uploaded_file:
    image = Image.open(uploaded_file).convert("RGB")
    st.image(image, caption="Uploaded Image", use_container_width=True)

    # ---- Captioning ----
    with st.spinner("Generating caption..."):
        caption = captioner(image)[0]["generated_text"]
    st.subheader("📝 Caption")
    st.success(caption)

    # ---- Emoji Mode ----
    emoji_map = {
        "dog": "🐶", "cat": "🐱", "ball": "⚽", "frisbee": "🥏",
        "man": "👨", "woman": "👩", "child": "🧒",
        "car": "🚗", "bicycle": "🚲", "horse": "🐎", "bird": "🐦",
        "food": "🍔", "drink": "🥤", "tree": "🌳"
    }
    emoji_caption = " ".join(
        emoji_map.get(word.lower(), word) for word in caption.split()
    )
    st.subheader("😎 Emoji Mode")
    st.info(emoji_caption)

    # ---- Object Detection ----
    with st.spinner("Detecting objects..."):
        detections = detector(image)

    st.subheader("🔍 Objects Detected")

    # Draw bounding boxes
    draw_img = image.copy()
    draw = ImageDraw.Draw(draw_img)

    for obj in detections:
        box = obj["box"]
        label = f"{obj['label']} ({obj['score']:.2f})"

        # Draw rectangle
        draw.rectangle(
            [(box["xmin"], box["ymin"]), (box["xmax"], box["ymax"])],
            outline="red", width=3
        )
        # Add label above box
        draw.text((box["xmin"], box["ymin"] - 10), label, fill="red")

        st.write(f"- {label}")

    st.image(draw_img, caption="Objects with bounding boxes", use_container_width=True)

    # ---- Download Button ----
    buf = io.BytesIO()
    draw_img.save(buf, format="PNG")
    byte_im = buf.getvalue()
    st.download_button(
        label="📥 Download Annotated Image",
        data=byte_im,
        file_name="annotated_image.png",
        mime="image/png"
    )

    # ---- Visual Question Answering ----
    st.subheader("❓ Ask a Question About the Image")
    user_q = st.text_input("Type your question (e.g., 'What is the dog doing?')")

    if user_q:
        with st.spinner("Thinking..."):
            answer = vqa({"question": user_q, "image": image})
        st.success(f"**Answer:** {answer[0]['answer']}")