app.py

import torch
from transformers import Blip2Processor, Blip2ForConditionalGeneration
from PIL import Image
import gradio as gr
from io import BytesIO
import requests

# -----------------------------
# Device and Model Setup
# -----------------------------
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

model_name = "Salesforce/blip2-flan-t5-xl-coco"  # public model
processor = Blip2Processor.from_pretrained(model_name)
model = Blip2ForConditionalGeneration.from_pretrained(model_name)
model.to(device)

# -----------------------------
# Inference Function
# -----------------------------
def analyze_eo_image(image, question):
    if image is None or question.strip() == "":
        return "Please upload an EO image and ask a question."

    if image.mode != "RGB":
        image = image.convert("RGB")

    inputs = processor(image, question, return_tensors="pt").to(
        device, torch.float16 if device=="cuda" else torch.float32
    )

    out = model.generate(
        **inputs,
        max_new_tokens=80,
        do_sample=True,
        temperature=0.7
    )

    return processor.decode(out[0], skip_special_tokens=True)

# -----------------------------
# Optional: URL input
# -----------------------------
def analyze_eo_url(url, question):
    try:
        response = requests.get(url)
        image = Image.open(BytesIO(response.content)).convert("RGB")
        return analyze_eo_image(image, question)
    except Exception as e:
        return f"Error loading image: {e}"

# -----------------------------
# Beautiful Gradio Layout
# -----------------------------
with gr.Blocks(title="🌍 EO Image Analysis") as demo:

    gr.Markdown(
        """
        # 🌍 Earth Observation Image Analysis
        Ask questions about EO images using a multimodal AI model.
        Powered by BLIP-2 + FLAN-T5.
        **Examples:** "Identify land cover types", "Where is the river?", "Has urban area expanded?"
        """
    )

    with gr.Tabs():
        with gr.Tab("Upload Image"):
            with gr.Row():
                with gr.Column(scale=1):
                    img_input = gr.Image(type="pil", label="Upload EO Image")
                    question_input = gr.Textbox(label="Ask a question about the image", placeholder="E.g. Where is the river?")
                    submit_btn = gr.Button("Analyze 🌟")
                with gr.Column(scale=1):
                    output_text = gr.Textbox(label="AI Answer", interactive=False)

            submit_btn.click(analyze_eo_image, inputs=[img_input, question_input], outputs=output_text)

        with gr.Tab("Use Image URL"):
            with gr.Row():
                with gr.Column(scale=1):
                    url_input = gr.Textbox(label="Enter Image URL")
                    url_question = gr.Textbox(label="Ask a question about the image")
                    url_btn = gr.Button("Analyze 🌟")
                with gr.Column(scale=1):
                    url_output = gr.Textbox(label="AI Answer", interactive=False)

            url_btn.click(analyze_eo_url, inputs=[url_input, url_question], outputs=url_output)

    gr.Markdown(
        "💡 Tip: Use clear, simple questions for best results. Supports natural language queries about EO images."
    )

demo.launch(share=True)