transform = transforms.Compose([
transforms.Resize((opt.input_size, 2 * opt.input_size)),
transforms.ToTensor(),
transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))
])
train_dataset = datasets.ImageFolder(opt.train_dir, transform=transform)
train_loader = DataLoader(train_dataset,
batch_size=opt.batch_size,
shuffle=True)
# network
G = Generator(opt.ngf)
D = Discriminator(opt.ndf)
G.init_weight(mean=0.0, std=0.02)
D.init_weight(mean=0.0, std=0.02)
if opt.use_gpu:
G.cuda()
D.cuda()
# loss
bce_loss = nn.BCELoss()
l1_loss = nn.L1Loss()
# optimizer
G_optimizer = optim.Adam(G.parameters(),
lr=opt.lrG,
betas=(opt.beta1, opt.beta2))
D_optimizer = optim.Adam(D.parameters(),
lr=opt.lrD,
