def __init__(self,
input_channels=1,
verbose=False,
cshape=(128, 7, 7),
w_distance=True):
super(Discriminator, self).__init__()
self.channels = input_channels
self.cshape = cshape
self.iels = int(np.prod(self.cshape))
self.lshape = (self.iels, )
self.verbose = verbose
self.model = nn.Sequential(
nn.Conv2d(self.channels, 64, 4, stride=2, padding=1),
nn.LeakyReLU(0.02, inplace=True),
nn.Conv2d(64, 128, 4, stride=2, padding=1),
nn.LeakyReLU(0.02, inplace=True),
Reshape(self.lshape),
nn.Linear(self.iels, 1024),
nn.LeakyReLU(0.02, inplace=True),
nn.Linear(1024, 1),
)
if w_distance is False:
self.model = nn.Sequential(self.model, nn.Sigmoid())
init_weights(self)
if self.verbose:
print(self.model)
def __init__(self,
input_channels=1,
output_channels=64,
cshape=(128, 7, 7),
verbose=False):
super(Encoder, self).__init__()
self.cshape = cshape
self.iels = int(np.prod(self.cshape))
self.lshape = (self.iels, )
self.output_channels = output_channels
self.input_channels = input_channels
self.verbose = verbose
self.model = nn.Sequential(
nn.Conv2d(self.input_channels, 64, 4, stride=2, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(True),
nn.Conv2d(64, 128, 4, stride=2, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(True),
Reshape(self.lshape),
nn.Linear(self.iels, 1024),
nn.BatchNorm1d(1024),
nn.ReLU(True),
)
self.mu = nn.Linear(1024, self.output_channels)
self.con = nn.Linear(1024, self.output_channels)
init_weights(self)
if self.verbose:
print(self.model)
def __init__(self,
latent_dim=50,
x_shape=(1, 28, 28),
cshape=(128, 7, 7),
verbose=False):
super(Generator, self).__init__()
self.latent_dim = latent_dim
self.ishape = cshape
self.iels = int(np.prod(self.ishape))
self.x_shape = x_shape
self.output_channels = x_shape[0]
self.verbose = verbose
self.model = nn.Sequential(
nn.Linear(self.latent_dim, 1024),
nn.ReLU(True),
nn.Linear(1024, self.iels),
nn.BatchNorm1d(self.iels),
nn.ReLU(True),
#
Reshape(self.ishape),
)
# block_layers = []
# for i in range(6):
# block_layers += [ResNetBlock(128)]
# self.model = nn.Sequential(self.model, *block_layers)
self.model = nn.Sequential(
self.model,
nn.ConvTranspose2d(128, 64, 4, stride=2, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(True),
#
nn.ConvTranspose2d(64,
self.output_channels,
4,
stride=2,
padding=1),
nn.Sigmoid(),
)
init_weights(self)
if self.verbose:
print(self.model)
def __init__(self, latent_dim=50, input_dim=784, inter_dims=[500, 500, 2000], verbose=False):
super(Generator, self).__init__()
self.latent_dim = latent_dim
self.input_dim = input_dim
self.inter_dims = inter_dims
self.verbose = verbose
self.model = nn.Sequential(
*block(self.latent_dim, self.inter_dims[-1]),
*block(self.inter_dims[-1], self.inter_dims[-2]),
*block(self.inter_dims[-2], self.inter_dims[-3]),
nn.Linear(self.inter_dims[-3], self.input_dim),
nn.Sigmoid()
)
init_weights(self)
if self.verbose:
print(self.model)
def __init__(self, input_dim=784, inter_dims=[500, 500, 2000], latent_dim=10, verbose=False):
super(Encoder, self).__init__()
self.latent_dim = latent_dim
self.input_dim = input_dim
self.inter_dims = inter_dims
self.verbose = verbose
self.model = nn.Sequential(
*block(self.input_dim, self.inter_dims[0]),
*block(self.inter_dims[0], self.inter_dims[1]),
*block(self.inter_dims[1], self.inter_dims[2]),
)
self.mu = nn.Linear(self.inter_dims[-1], self.latent_dim)
self.con = nn.Linear(self.inter_dims[-1], self.latent_dim)
init_weights(self)
if self.verbose:
print(self.model)
def __init__(self, verbose=False, input_dim=784, inter_dims=[500, 500, 2000]):
super(Discriminator, self).__init__()
self.input_dim = input_dim
self.inter_dims = inter_dims
self.verbose = verbose
self.model = nn.Sequential(
nn.Linear(self.input_dim, self.inter_dims[0]),
nn.LeakyReLU(0.02, True),
nn.Linear(self.inter_dims[0], self.inter_dims[1]),
nn.LeakyReLU(0.02, True),
nn.Linear(self.inter_dims[1], self.inter_dims[2]),
nn.LeakyReLU(0.02, True),
nn.Linear(self.inter_dims[2], 1),
nn.Sigmoid()
)
init_weights(self)
if self.verbose:
print(self.model)
def __init__(self, input_dim=784, inter_dims=[500, 500, 2000], r=9, latent_dim=10, verbose=False):
super(Encoder_SMM, self).__init__()
self.latent_dim = latent_dim
self.input_dim = input_dim
self.inter_dims = inter_dims
self.r = r
self.verbose = verbose
self.model = nn.Sequential(
*block(self.input_dim, self.inter_dims[0]),
*block(self.inter_dims[0], self.inter_dims[1]),
*block(self.inter_dims[1], self.inter_dims[2]),
)
self.mu = nn.Linear(self.inter_dims[-1], self.latent_dim)
self.con = nn.Linear(self.inter_dims[-1], self.latent_dim)
self.v = nn.Linear(self.inter_dims[-1], 1)
init_weights(self)
self.v.weight.data.uniform_(0.01, 0.03)
if self.verbose:
print(self.model)
