def __init__(self, k1=2, c1=40, k2=2, c2=100, d1=96, d2=10):
super(Generator, self).__init__()
self.convt1 = ComplexConvTranspose2d(c1, 1, k1, 1, padding=0)
self.bn = ComplexBatchNorm2d(c1)
self.convt2 = ComplexConvTranspose2d(c2, c1, k2, 1,
padding=0) # k = 2,p' = k - 1
self.c2 = c2
self.fc1 = ComplexLinear(d1, 2 * 2 * c2)
self.fc2 = ComplexLinear(10, d1)
def __init__(self, k1=2, c1=40, k2=2, c2=100, k3=3, c3=3, d1=96, d2=10):
super(Encoder_f, self).__init__()
self.conv1 = ComplexConv2d(1, c1, k1, 1, padding=0)
self.bn1 = ComplexBatchNorm2d(c1)
self.conv2 = ComplexConv2d(c1, c2, k2, 1, padding=0)
self.bn2 = ComplexBatchNorm2d(c2)
self.conv3 = ComplexConv2d(c2, c3, k3, 1, padding=0)
self.fc1 = ComplexLinear(2 * 2 * c3, d1)
self.fc2 = ComplexLinear(d1, d2)
self.c3 = c3
def __init__(self):
super(ComplexGenerator, self).__init__()
# torch.nn.ConvTranspose2d(in_channels, out_channels, kernel_size, stride=1,
# padding=0, output_padding=0, groups=1, bias=True, dilation=1)
# input(N, C_in, H_in, W_in), output(N, C_out, H_out, W_out)
# H_out=(H_in−1)×stride[0]−2×padding[0] + kernel_size[0] + output_padding[0]
self.size = 16384
self.size1 = int(self.size * 0.125)
self.size2 = int(self.size * 0.25)
self.size3 = int(self.size * 0.5)
self.dense1 = ComplexLinear(100, self.size1)
self.batchnorm1 = ComplexBatchNorm1d(self.size1)
self.dense2 = ComplexLinear(self.size1, self.size2)
self.batchnorm2 = ComplexBatchNorm1d(self.size2)
self.dense3 = ComplexLinear(self.size2, self.size3)
self.batchnorm3 = ComplexBatchNorm1d(self.size3)
self.dense4 = ComplexLinear(self.size3, self.size)
def __init__(self):
super(ComplexDiscriminator, self).__init__()
# torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1,
# padding=0, dilation=1, groups=1, bias=True)
# input(N, C_in, H_in, W_in), output(N, C_out, H_out, W_out)
# H_out=[H_in + 2×padding[0] - dilation[0]×(kernel_size[0]−1) − 1]/stride[0] + 1
self.conv1 = ComplexConv2d(1, 64, 4, 2, 1)
self.conv2 = ComplexConv2d(64, 128, 4, 2, 1)
self.conv3 = ComplexConv2d(128, 256, 4, 2, 1)
self.conv4 = ComplexConv2d(256, 512, 4, 2, 1)
self.conv5 = ComplexConv2d(512, 1024, 4, 2, 1)
self.dense = ComplexLinear(1024 * 4 * 4, 1)
def __init__(self):
super(ComplexGenerator, self).__init__()
# torch.nn.ConvTranspose2d(in_channels, out_channels, kernel_size, stride=1,
# padding=0, output_padding=0, groups=1, bias=True, dilation=1)
# input(N, C_in, H_in, W_in), output(N, C_out, H_out, W_out)
# H_out=(H_in−1)×stride[0]−2×padding[0] + kernel_size[0] + output_padding[0]
self.dense = ComplexLinear(100, 1024 * 4 * 4)
self.deconv1 = ComplexConvTranspose2d(
1024, 512, 4, 2, 1) # (1024, 4, 4) --> (512, 8, 8)
self.deconv2 = ComplexConvTranspose2d(
512, 256, 4, 2, 1) # (512, 8, 8) --> (256, 16, 16)
self.deconv3 = ComplexConvTranspose2d(
256, 128, 4, 2, 1) # (256, 16, 16) --> (128, 32, 32)
self.deconv4 = ComplexConvTranspose2d(
128, 64, 4, 2, 1) # (128, 32, 32) --> (64, 64, 64)
self.deconv5 = ComplexConvTranspose2d(
64, 1, 4, 2, 1) # (64, 64, 64) --> (1, 128, 128)
def __init__(self, encoder, decoder):
super(VAE, self).__init__()
self.encoder = encoder
self.decoder = decoder
self._enc_mu = ComplexLinear(200, latent_dim)
self._enc_log_sigma = ComplexLinear(200, latent_dim)