import os
import torch
import gradio as gr
from PIL import Image
from process import load_seg_model, get_palette, generate_mask

# Device selection
device = 'cuda' if torch.cuda.is_available() else 'cpu'

def load_model():
    """Load model with Hugging Face Spaces compatible paths"""
    model_dir = 'model'
    checkpoint_path = os.path.join(model_dir, 'cloth_segm.pth')
    
    # Verify model exists (must be pre-uploaded to HF Spaces)
    if not os.path.exists(checkpoint_path):
        raise FileNotFoundError(
            f"Model not found at {checkpoint_path}. "
            "Please upload the model file to your Space's repository."
        )
    
    try:
        net = load_seg_model(checkpoint_path, device=device)
        palette = get_palette(4)
        return net, palette
    except Exception as e:
        raise RuntimeError(f"Model loading failed: {str(e)}")

# Initialize model (will fail fast if there's an issue)
net, palette = load_model()

def process_image(img: Image.Image) -> Image.Image:
    """Process input image and return segmentation mask"""
    if img is None:
        raise gr.Error("Please upload an image first")
    
    try:
        return generate_mask(img, net=net, palette=palette, device=device)
    except Exception as e:
        raise gr.Error(f"Processing failed: {str(e)}")

# Gradio interface
title = "Cloth Segmentation Demo"
description = """
Upload an image to get cloth segmentation using U2NET.
"""

with gr.Blocks() as demo:
    gr.Markdown(f"## {title}")
    gr.Markdown(description)
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(type="pil", label="Input Image")
            submit_btn = gr.Button("Process", variant="primary")
        with gr.Column():
            output_image = gr.Image(type="pil", label="Segmentation Result")
    
    submit_btn.click(
        fn=process_image,
        inputs=input_image,
        outputs=output_image
    )

if __name__ == "__main__":
    demo.launch()