type=str,
help='pretrained model checkpoint')
parser.add_argument('--message',
default='message',
type=str,
help='pretrained model checkpoint')
parser.add_argument('--epochs', default=101, type=int, help='train epochs')
parser.add_argument('--train', default=True, type=bool, help='train')
args = parser.parse_args()
save_path = args.save_path + f'{args.message}_{time_str}'
if not os.path.exists(save_path):
os.mkdir(save_path)
logger = Logger(f'{save_path}/log.log')
logger.Print(args)
train_data, val_data, test_data = load_cisia_surf(train_size=args.batch_size,
test_size=args.test_size)
model = Model(pretrained=False, num_classes=2)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(),
lr=0.01,
momentum=0.9,
weight_decay=5e-4)
scheduler = lr_scheduler.ExponentialLR(optimizer, gamma=0.95)
if use_cuda:
model = model.cuda()
criterion = criterion.cuda()
eval_history = []
parser = argparse.ArgumentParser(description='face anto-spoofing')
parser.add_argument('--batch-size', default='128', type=int, help='train batch size')
parser.add_argument('--test-size', default='64', type=int, help='test batch size')
parser.add_argument('--save-path', default='./logs/', type=str, help='log save path')
parser.add_argument('--checkpoint', default='model.pth', type=str, help='pretrained model checkpoint')
parser.add_argument('--message', default='message', type=str, help='pretrained model checkpoint')
parser.add_argument('--epochs', default=101, type=int, help='train epochs')
parser.add_argument('--train', default=True, type=bool, help='train')
args = parser.parse_args()
save_path = args.save_path + f'{args.message}_{time_str}'
if not os.path.exists(save_path):
os.mkdir(save_path)
logger = Logger(f'{save_path}/log.log')
logger.Print(args.message)
train_data, val_data, test_data= load_cisia_surf(train_size=args.batch_size,test_size=args.test_size)
model = Model(pretrained=False,num_classes=2)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9,weight_decay=5e-4)
scheduler = lr_scheduler.ExponentialLR(optimizer, gamma=0.95)
ct_loss = CenterLoss(num_classes=2, feat_dim=512)
optimzer4ct = optim.SGD(ct_loss.parameters(), lr =0.01, momentum=0.9,weight_decay=5e-4)
scheduler4ct = lr_scheduler.ExponentialLR(optimzer4ct, gamma=0.95)
if use_cuda:
model = model.cuda()
criterion = criterion.cuda()
