sample_data/untitled4.py

# -*- coding: utf-8 -*-
"""Untitled4.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1LWcAEqZ865KCYhK-crQ3RKcFJEcH4xhD
"""

from huggingface_hub import login

login(token="hf_xxxx")

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

# -----------------------------
# 1️⃣ Device and model setup
# -----------------------------
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

# Use a smaller model for memory efficiency
model_name = "Salesforce/blip2-flan-t5-xl-coco"  # Updated model with preprocessor_config.json
processor = Blip2Processor.from_pretrained(model_name)
model = Blip2ForConditionalGeneration.from_pretrained(model_name)
model.to(device)

# -----------------------------
# 2️⃣ Inference function
# -----------------------------
def analyze_eo_image(image, question):
    """
    Analyze Earth Observation (EO) image with a natural language question.
    """
    try:
        if image is None or question.strip() == "":
            return "Please upload an EO image and ask a question."

        # Convert image to RGB if needed
        if image.mode != "RGB":
            image = image.convert("RGB")

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

        # Generate answer
        out = model.generate(
            **inputs,
            max_new_tokens=80,  # allow slightly longer explanations
            do_sample=True,
            temperature=0.7
        )

        # Decode and return
        answer = processor.decode(out[0], skip_special_tokens=True)
        return answer

    except Exception as e:
        return f"Error: {e}"

# -----------------------------
# 3️⃣ Optional: Analyze image from URL
# -----------------------------
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}"

# -----------------------------
# 4️⃣ Gradio Interface
# -----------------------------
interface = gr.Interface(
    fn=analyze_eo_image,
    inputs=[
        gr.Image(type="pil", label="Upload EO Image"),
        gr.Textbox(label="Ask a Question about the EO Image")
    ],
    outputs="text",
    title="🌍 EO Image Analysis with Multimodal GPT-OSS",
    description="""
    Ask questions about Earth Observation (EO) images.
    Powered by BLIP-2 + FLAN-T5 (small, memory-efficient).
    Examples: "Identify land cover types", "Where is the river?",
    "Has urban area expanded?".
    """,
    allow_flagging="never"
)

# Launch the interface (shareable link)
interface.launch(share=True)

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)

from huggingface_hub import login

login(token="hf_xxxx")