def __init__(self,
logger,
use_cpu=False,
num_gpu=1,
gan_type='WGAN_GP',
gan_k=1,
lr_dis=1e-4,
train_crop_size=40):
super(AdversarialLoss, self).__init__()
self.logger = logger
self.gan_type = gan_type
self.gan_k = gan_k
self.device = torch.device('cpu' if use_cpu else 'cuda')
self.discriminator = discriminator.Discriminator(train_crop_size *
4).to(self.device)
if (num_gpu > 1):
self.discriminator = nn.DataParallel(self.discriminator,
list(range(num_gpu)))
if (gan_type in ['WGAN_GP', 'GAN']):
self.optimizer = optim.Adam(self.discriminator.parameters(),
betas=(0, 0.9),
eps=1e-8,
lr=lr_dis)
else:
raise SystemExit('Error: no such type of GAN!')
self.bce_loss = torch.nn.BCELoss().to(self.device)
def __init__(self, args, gan_type):
super(Adversarial, self).__init__()
self.gan_type = gan_type
self.gan_k = args.gan_k
self.discriminator = discriminator.Discriminator(args, gan_type)
if gan_type != 'WGAN_GP':
self.optimizer = utility.make_optimizer(args, self.discriminator)
else:
self.optimizer = optim.Adam(
self.discriminator.parameters(),
betas=(0, 0.9), eps=1e-8, lr=1e-5
)
self.scheduler = utility.make_scheduler(args, self.optimizer)
def __init__(self, args, gan_type):
super(Adversarial, self).__init__()
self.gan_type = gan_type
self.gan_k = args.gan_k
self.aprx_epochs = args.aprx_epochs
self.aprx_training_dir = args.aprx_training_dir
self.aprx_training_dir_HR = args.aprx_training_dir_HR
self.batch_size = args.batch_size
self.patch_size = args.patch_size
self.discriminator = discriminator.Discriminator(args, gan_type)
self.optimizer = utility.make_optimizer(args, self.discriminator)
self.scheduler = utility.make_scheduler(args, self.optimizer)
self.a_counter = 0
def __init__(self, args, gan_type):
super(Adversarial, self).__init__()
self.gan_type = gan_type
self.gan_k = args.gan_k
self.dis = discriminator.Discriminator(args)
if gan_type == 'WGAN_GP':
# see https://arxiv.org/pdf/1704.00028.pdf pp.4
optim_dict = {
'optimizer': 'ADAM',
'betas': (0, 0.9),
'epsilon': 1e-8,
'lr': 1e-5,
'weight_decay': args.weight_decay,
'decay': args.decay,
'gamma': args.gamma
}
optim_args = SimpleNamespace(**optim_dict)
else:
optim_args = args
self.optimizer = utility.make_optimizer(optim_args, self.dis)