def get_model(model_name, z_dim, configs=None):
if model_name == 'dc':
D = GoodDiscriminator()
G = GoodGenerator()
elif model_name == 'dcBN':
D = GoodDiscriminatorbn()
G = GoodGenerator()
elif model_name == 'dcD':
D = GoodDiscriminatord()
G = GoodGenerator()
elif model_name == 'DCGAN':
D = DC_discriminator()
G = DC_generator(z_dim=z_dim)
elif model_name == 'Resnet32':
D = ResNet32Discriminator(n_in=3, num_filters=128, batchnorm=True)
G = ResNet32Generator(z_dim=z_dim, num_filters=128, batchnorm=True)
elif model_name == 'Resnet':
D = ResNetDiscriminator(in_channel=configs['image_channel'],
insize=configs['image_size'],
num_filters=configs['feature_num'],
batchnorm=configs['batchnorm_d'])
G = ResNetGenerator(z_dim=z_dim,
outsize=configs['image_size'],
num_filters=configs['feature_num'],
batchnorm=configs['batchnorm_g'])
elif model_name == 'ResnetWBN':
D = ResNet32Discriminator(n_in=3, num_filters=128, batchnorm=False)
G = ResNet32Generator(z_dim=z_dim, num_filters=128, batchnorm=True)
elif model_name == 'DCGAN-WBN':
D = DC_discriminatorW()
G = DC_generator(z_dim=z_dim)
elif model_name == 'dcSN':
D = GoodSNDiscriminator()
G = GoodGenerator()
elif model_name == 'mnist':
D = dc_D()
G = dc_G(z_dim=z_dim)
elif model_name == 'dc32':
D = dcD32()
G = dcG32(z_dim=z_dim)
elif model_name == 'DCGANs':
D = DCGAN_D(insize=configs['image_size'],
channel_num=configs['image_channel'],
feature_num=configs['feature_num'],
n_extra_layers=configs['n_extra_layers'])
G = DCGAN_G(outsize=configs['image_size'],
z_dim=z_dim,
nc=configs['image_channel'],
feature_num=configs['feature_num'],
n_extra_layers=configs['n_extra_layers'])
else:
print('No matching result of :')
print(model_name)
return D, G
config = vars(parser.parse_args())
print(config)
print('numpy calculation')
device = torch.device('cuda:0')
model = config['model']
z_dim = config['z_dim']
if model == 'dc':
G = GoodGenerator()
elif model == 'ResGAN':
G = ResNet32Generator(z_dim=z_dim, num_filters=128, batchnorm=True)
elif model == 'DCGAN':
G = DC_generator(z_dim=z_dim)
elif model == 'mnist':
G = dc_G(z_dim=z_dim)
elif model == 'dc32':
G = dcG32(z_dim=z_dim)
else:
raise ValueError('No matching generator for %s' % model)
G.to(device)
G.eval()
evalor = evalor(G=G,
z_dim=z_dim,
model_dir=config['model_dir'],
device=device,
log_path=config['logdir'],
dim=config['dim'])
evalor.eval_metrics(begin=config['begin'],
end=config['end'],
step=config['step'],
is_flag=config['eval_is'],
fid_flag=config['eval_fid'],