distributed=args.distributed),
drop_last=True,
)
if get_rank() == 0 and wandb is not None and args.wandb:
wandb.init(project="dualstylegan")
ckpt = torch.load(args.encoder_path, map_location='cpu')
opts = ckpt['opts']
opts['checkpoint_path'] = args.encoder_path
if 'learn_in_w' not in opts:
opts['learn_in_w'] = True
if 'output_size' not in opts:
opts['output_size'] = 1024
opts = Namespace(**opts)
encoder = pSp(opts).to(device).eval()
encoder.latent_avg = encoder.latent_avg.to(device)
vggloss = VGG19().to(device).eval()
id_loss = id_loss.IDLoss(args.identity_path).to(device).eval()
print('Encoder model successfully loaded!')
instyle_dict = np.load(args.instyle_path, allow_pickle='TRUE').item()
exstyle_dict = np.load(args.exstyle_path, allow_pickle='TRUE').item()
path = args.image_path
instyles = []
exstyles = []
Simgs = []
for filename in instyle_dict.keys():
instyle = instyle_dict[filename]
exstyle = exstyle_dict[filename]
# load DualStyleGAN
generator = DualStyleGAN(1024, 512, 8, 2, res_index=6)
generator.eval()
ckpt = torch.load(os.path.join(MODEL_DIR, style_type, 'generator.pt'),
map_location=lambda storage, loc: storage)
generator.load_state_dict(ckpt["g_ema"])
generator = generator.to(device)
# load encoder
model_path = os.path.join(MODEL_DIR, 'encoder.pt')
ckpt = torch.load(model_path, map_location='cpu')
opts = ckpt['opts']
opts['checkpoint_path'] = model_path
opts = Namespace(**opts)
opts.device = device
encoder = pSp(opts)
encoder.eval()
encoder = encoder.to(device)
# load extrinsic style code
exstyles = np.load(os.path.join(MODEL_DIR, style_type,
MODEL_PATHS[style_type + '-S']["name"]),
allow_pickle='TRUE').item()
# load sampler network
icptc = ICPTrainer(np.empty([0, 512 * 11]), 128)
icpts = ICPTrainer(np.empty([0, 512 * 7]), 128)
ckpt = torch.load(os.path.join(MODEL_DIR, style_type, 'sampler.pt'),
map_location=lambda storage, loc: storage)
icptc.icp.netT.load_state_dict(ckpt['color'])
icpts.icp.netT.load_state_dict(ckpt['structure'])