app.py

import gradio as gr
import torch
import numpy as np
import cv2
from PIL import Image
from transformers import BlipProcessor, BlipForConditionalGeneration, SamModel, SamProcessor
import time

# Set device to GPU if available
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load Florence BLIP model (Public Model - No Authentication Required)
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base").to(device)

# Load SAM model (Public Model - No Authentication Required)
sam_processor = SamProcessor.from_pretrained("facebook/sam-vit-base")
sam_model = SamModel.from_pretrained("facebook/sam-vit-base").to(device)

def process_image(image):
    start_time = time.time()
    
    # Convert and resize image
    pil_image = Image.fromarray(image).resize((512, 512))  # Resize to optimize processing
    print("✅ Image loaded and resized.")

    # Generate caption using Florence
    try:
        inputs = processor(pil_image, return_tensors="pt").to(device)
        with torch.no_grad():
            out = model.generate(**inputs)
        description = processor.decode(out[0], skip_special_tokens=True)
        print(f"📝 Florence Captioning done in {time.time() - start_time:.2f} sec")
    except Exception as e:
        print(f"❌ Error in Florence: {e}")
        return "Failed to generate description.", image
    
    # Process Image for SAM
    try:
        encoding = sam_processor(images=pil_image, return_tensors="pt").to(device)
        with torch.no_grad():
            outputs = sam_model(**encoding)
        
        # Extract segmentation mask
        mask = outputs.pred_masks[0, 0].cpu().numpy()
        mask_overlay = image.copy()
        mask_overlay[mask > 0.5] = [0, 255, 0]  # Green overlay for segmentation
        print(f"🎨 SAM Segmentation done in {time.time() - start_time:.2f} sec")
    except Exception as e:
        print(f"❌ Error in SAM: {e}")
        return description, image

    return description, mask_overlay

# Gradio Interface
demo = gr.Interface(
    fn=process_image,
    inputs=gr.Image(type="numpy"),
    outputs=[gr.Textbox(label="Image Description"), gr.Image(label="Segmented Image")],
    title="Florence + SAM Image Processing",
    description="Upload an image to get its description using Florence and segmentation using SAM (loaded from Hugging Face)."
)

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)