app.py

import gradio as gr
import torch
import cv2
import numpy as np
from PIL import Image
import os
# Import ACGPN-specific modules (adjust based on actual repository structure)
# Note: You may need to copy relevant ACGPN code into the Space or modify imports
from models.acgpn import ACGPN  # Hypothetical import; replace with actual model class
from utils.preprocessing import preprocess_image, parse_human  # Hypothetical preprocessing utilities

# Set device to CPU
device = torch.device("cpu")

# Load pre-trained ACGPN model
def load_model():
    model = ACGPN()  # Initialize model (adjust parameters as per ACGPN docs)
    checkpoint_path = "checkpoints/acgpn_checkpoint.pth"  # Path to pre-trained weights
    model.load_state_dict(torch.load(checkpoint_path, map_location=device))
    model.to(device)
    model.eval()
    return model

model = load_model()

# Function to process images and generate try-on
def virtual_try_on(person_image, cloth_image):
    try:
        # Convert Gradio inputs (PIL Images) to numpy arrays
        person_img = np.array(person_image)
        cloth_img = np.array(cloth_image)

        # Preprocess images (resize, normalize, etc.)
        person_processed, person_mask = preprocess_image(person_img, is_person=True)
        cloth_processed = preprocess_image(cloth_img, is_person=False)

        # Parse human pose and segmentation (using ACGPN utilities)
        pose_map, parse_map = parse_human(person_processed)

        # Convert to tensors
        person_tensor = torch.from_numpy(person_processed).float().to(device)
        cloth_tensor = torch.from_numpy(cloth_processed).float().to(device)
        pose_tensor = torch.from_numpy(pose_map).float().to(device)
        parse_tensor = torch.from_numpy(parse_map).float().to(device)

        # Run inference
        with torch.no_grad():
            output = model(person_tensor, cloth_tensor, pose_tensor, parse_tensor)
            output = output.cpu().numpy()

        # Post-process output
        output_img = (output * 255).astype(np.uint8)
        output_img = Image.fromarray(output_img)

        return output_img
    except Exception as e:
        return f"Error: {str(e)}"

# Gradio interface
iface = gr.Interface(
    fn=virtual_try_on,
    inputs=[
        gr.Image(type="pil", label="Upload Person Image"),
        gr.Image(type="pil", label="Upload Clothing Image"),
    ],
    outputs=gr.Image(type="pil", label="Try-On Result"),
    title="ACGPN Virtual Try-On",
    description="Upload a person image and a clothing image to see the virtual try-on result.",
)

# Launch the app
if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)