def __init__(self):
super(Autoencoder, self).__init__()
self.encoder = nn.Sequential(
_ConvLayer(3, 128),
_ConvLayer(128, 256),
_ConvLayer(256, 512),
_ConvLayer(512, 1024),
Flatten(),
nn.Linear(1024 * 4 * 4,1024),
nn.Linear(1024,1024 * 4 * 4),
Reshape(),
_UpScale(1024, 512),
)
self.decoder_A = nn.Sequential(
_UpScale(512, 256),
_UpScale(256, 128),
_UpScale(128, 64),
Conv2d(64, 3, kernel_size=5, padding=1),
nn.Sigmoid(),
)
self.decoder_B = nn.Sequential(
_UpScale(512, 256),
_UpScale(256, 128),
_UpScale(128, 64),
Conv2d(64, 3, kernel_size=5, padding=1),
nn.Sigmoid(),
)
def __init__(self, ndf=128):
super(Discriminator, self).__init__()
self.ndf = ndf
self.conv1 = Conv2d(3, ndf, 5, stride=2)
self.conv2 = nn.utils.spectral_norm(Conv2d(ndf, ndf*2, 5, stride=2))
self.conv3 = nn.utils.spectral_norm(Conv2d(ndf*2, ndf*4, 5, stride=2))
self.conv4 = nn.utils.spectral_norm(Conv2d(ndf*4, ndf*8, 3, stride=2))
self.fc5 = nn.Linear(4*4*ndf*8, 1)
def __init__(self):
super(Discriminator, self).__init__()
self.conv1 = Conv2d(3, 128, 5, stride=2)
self.conv2 = Conv2d(128, 256, 5, stride=2)
self.bn2 = nn.BatchNorm2d(256)
self.conv3 = Conv2d(256, 512, 5, stride=2)
self.bn3 = nn.BatchNorm2d(512)
self.conv4 = Conv2d(512, 1024, 3, stride=2)
self.bn4 = nn.BatchNorm2d(1024)
self.fc5 = nn.Linear(4 * 4 * 1024, 1)
def __init__(self, z_dim):
super(Generator, self).__init__()
self.fc1 = nn.Linear(z_dim, 8 * 8 * 1024)
self.up2 = nn.Upsample(scale_factor=2, mode='nearest')
self.conv2 = Conv2d(1024, 512, 3)
self.bn2 = nn.BatchNorm2d(512)
self.up3 = nn.Upsample(scale_factor=2, mode='nearest')
self.conv3 = Conv2d(512, 256, 5)
self.bn3 = nn.BatchNorm2d(256)
self.up4 = nn.Upsample(scale_factor=2, mode='nearest')
self.conv4 = Conv2d(256, 3, 5)
def __init__(self, z_dim, ndf=128):
super(Encoder, self).__init__()
self.ndf = ndf
self.conv1 = Conv2d(3, ndf, 5, stride=2)
self.conv2 = Conv2d(ndf, ndf * 2, 5, stride=2)
self.bn2 = nn.BatchNorm2d(ndf * 2)
self.conv3 = Conv2d(ndf * 2, ndf * 4, 5, stride=2)
self.bn3 = nn.BatchNorm2d(ndf * 4)
self.conv4 = Conv2d(ndf * 4, ndf * 8, 3, stride=2)
self.fc5_mu = nn.Linear(4 * 4 * ndf * 8, z_dim)
self.fc5_logvar = nn.Linear(4 * 4 * ndf * 8, z_dim)
def __init__(self, z_dim):
super(GeneratorZ, self).__init__()
self.conv1 = Conv2d(3, 128, 5, stride=2)
self.bn1 = nn.BatchNorm2d(128)
self.conv2 = Conv2d(128, 256, 5, stride=2)
self.bn2 = nn.BatchNorm2d(256)
self.conv3 = Conv2d(256, 512, 5, stride=2)
self.bn3 = nn.BatchNorm2d(512)
self.conv4 = Conv2d(512, 1024, 3, stride=2)
self.bn4 = nn.BatchNorm2d(1024)
self.fc5 = nn.Linear(4 * 4 * 1024, z_dim)
def __init__(self, z_dim):
super(Encoder, self).__init__()
self.conv1 = Conv2d(3, 128, 5, stride=2)
self.conv2 = Conv2d(128, 256, 5, stride=2)
self.bn2 = nn.BatchNorm2d(256)
self.conv3 = Conv2d(256, 512, 5, stride=2)
self.bn3 = nn.BatchNorm2d(512)
self.conv4 = Conv2d(512, 1024, 3, stride=2)
self.bn4 = nn.BatchNorm2d(1024)
self.fc5_mu = nn.Linear(4*4*1024, z_dim)
self.fc5_logvar = nn.Linear(4*4*1024, z_dim)
def __init__(self, z_dim, ndf=128):
super(Generator, self).__init__()
self.ndf = ndf
self.fc1 = nn.Linear(z_dim, 8 * 8 * ndf * 8)
self.conv1 = Conv2d(ndf * 8, ndf * 4, 3)
# (512, 8, 8)
self.up2 = nn.Upsample(scale_factor=2, mode='nearest')
self.conv2 = Conv2d(ndf * 4, ndf * 2, 3)
self.bn2 = nn.BatchNorm2d(ndf * 2)
# (256, 16, 16)
self.up3 = nn.Upsample(scale_factor=2, mode='nearest')
self.conv3 = Conv2d(ndf * 2, ndf, 5)
self.bn3 = nn.BatchNorm2d(ndf)
# (128, 32, 32)
self.up4 = nn.Upsample(scale_factor=2, mode='nearest')
self.conv4 = Conv2d(ndf, 3, 5)
def __init__(self, input_features, output_features):
super(_UpScale, self).__init__()
self.add_module(
"conv2_", Conv2d(input_features,
output_features * 4,
kernel_size=3))
self.add_module("leakyrelu", nn.LeakyReLU(0.1, inplace=True))
self.add_module("pixelshuffler", _PixelShuffler())
def __init__(self, input_features, output_features):
super(_ConvLayer, self).__init__()
self.add_module('conv2', Conv2d(input_features, output_features,
kernel_size=5, stride=2))
self.add_module('leakyrelu', nn.LeakyReLU(0.1, inplace=True))
def __init__(self, in_channels, out_channels):
super(UpScale, self).__init__()
self.add_module('conv2d_',
Conv2d(in_channels, out_channels * 4, kernel_size=3))
self.add_module('leakyrelu', nn.LeakyReLU(0.1, inplace=True))
self.add_module('pixelshuffle', PixelShuffle())
def __init__(self, in_channels, out_channels):
super(ConvLayer, self).__init__()
self.add_module(
'conv2d_',
Conv2d(in_channels, out_channels, kernel_size=5, stride=2))
self.add_module('leakyrelu', nn.LeakyReLU(0.1, inplace=True))