warp_design_on_dress.py

import os
import torch
import torch.nn.functional as F
from torchvision import transforms
from PIL import Image
import numpy as np
from networks import GMM, TOM, load_checkpoint, Options
import torchvision.transforms.functional as TF

def prepare_inputs(dress_path, design_path, height=256, width=192):
    """Prepare and normalize input images"""
    transform = transforms.Compose([
        transforms.Resize((height, width)),
        transforms.ToTensor(),
        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
    ])
    
    dress_img = Image.open(dress_path).convert('RGB')
    design_img = Image.open(design_path).convert('RGB')
    
    dress_tensor = transform(dress_img).unsqueeze(0)
    design_tensor = transform(design_img).unsqueeze(0)
    
    # Create mask (assume design has transparent background)
    design_arr = np.array(design_img)
    if design_arr.shape[2] == 4:  # Has alpha channel
        mask = (design_arr[:, :, 3] > 0).astype(np.float32)
    else:
        mask = np.ones((design_arr.shape[0], design_arr.shape[1]), dtype=np.float32)
    
    mask_img = Image.fromarray((mask * 255).astype(np.uint8))
    mask_tensor = TF.to_tensor(TF.resize(mask_img, (height, width))).unsqueeze(0)
    
    return dress_tensor, design_tensor, mask_tensor

def create_agnostic(dress_tensor):
    """Create agnostic representation of dress"""
    return dress_tensor.clone()

def run_design_warp_on_dress(dress_path, design_path, gmm_ckpt, tom_ckpt, output_dir):
    os.makedirs(output_dir, exist_ok=True)
    
    # Prepare inputs
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    dress_tensor, design_tensor, design_mask = prepare_inputs(dress_path, design_path)
    agnostic = create_agnostic(dress_tensor)
    
    # Initialize models
    opt = Options()
    gmm = GMM(opt).to(device)
    tom = TOM(opt).to(device)
    
    # Load checkpoints
    load_checkpoint(gmm, gmm_ckpt)
    load_checkpoint(tom, tom_ckpt)
    
    # Move tensors to device
    agnostic = agnostic.to(device)
    design_tensor = design_tensor.to(device)
    design_mask = design_mask.to(device)
    
    # GMM Processing
    with torch.no_grad():
        gmm.eval()
        grid, _ = gmm(agnostic, design_mask)
        warped_design = F.grid_sample(design_tensor, grid, padding_mode='border', align_corners=True)
        warped_mask = F.grid_sample(design_mask, grid, padding_mode='zeros', align_corners=True)
    
    # TOM Processing
    with torch.no_grad():
        tom.eval()
        # Prepare TOM input: [agnostic, warped_design, warped_mask]
        tom_input = torch.cat([agnostic, warped_design, warped_mask], dim=1)
        p_rendered, m_composite = tom(tom_input)
        
        # Final composition
        tryon = warped_design * m_composite + p_rendered * (1 - m_composite)
    
    # Save output
    tryon = tryon.squeeze().permute(1, 2, 0).cpu().numpy()
    tryon = (tryon * 0.5 + 0.5) * 255
    tryon = tryon.clip(0, 255).astype(np.uint8)
    
    output_path = os.path.join(output_dir, "warped_design.jpg")
    Image.fromarray(tryon).save(output_path)
    
    return output_path