class Ladder(torch.nn.Module):
def __init__(self, encoder_sizes, decoder_sizes, encoder_activations,
encoder_train_bn_scaling, noise_std, use_cuda):
super(Ladder, self).__init__()
self.use_cuda = use_cuda
decoder_in = encoder_sizes[-1]
encoder_in = decoder_sizes[-1]
self.se = StackedEncoders(encoder_in, encoder_sizes, encoder_activations,
encoder_train_bn_scaling, noise_std, use_cuda)
self.de = StackedDecoders(decoder_in, decoder_sizes, encoder_in, use_cuda)
self.bn_image = torch.nn.BatchNorm1d(encoder_in, affine=False)
def forward_encoders_clean(self, data):
return self.se.forward_clean(data)
def forward_encoders_noise(self, data):
return self.se.forward_noise(data)
def forward_decoders(self, tilde_z_layers, encoder_output, tilde_z_bottom):
return self.de.forward(tilde_z_layers, encoder_output, tilde_z_bottom)
def get_encoders_tilde_z(self, reverse=True):
return self.se.get_encoders_tilde_z(reverse)
def get_encoders_z_pre(self, reverse=True):
return self.se.get_encoders_z_pre(reverse)
def get_encoder_tilde_z_bottom(self):
return self.se.buffer_tilde_z_bottom.clone()
def get_encoders_z(self, reverse=True):
return self.se.get_encoders_z(reverse)
def decoder_bn_hat_z_layers(self, hat_z_layers, z_pre_layers):
return self.de.bn_hat_z_layers(hat_z_layers, z_pre_layers)
def __init__(self, encoder_sizes, decoder_sizes, encoder_activations,
encoder_train_bn_scaling, noise_std, use_cuda):
super(Ladder, self).__init__()
self.use_cuda = use_cuda
decoder_in = encoder_sizes[-1]
encoder_in = decoder_sizes[-1]
self.se = StackedEncoders(encoder_in, encoder_sizes, encoder_activations,
encoder_train_bn_scaling, noise_std, use_cuda)
self.de = StackedDecoders(decoder_in, decoder_sizes, encoder_in, use_cuda)
self.bn_image = torch.nn.BatchNorm1d(encoder_in, affine=False)
def __init__(self, encoder_in, encoder_sizes, decoder_in, decoder_sizes,
image_size, encoder_activations, encoder_train_bn_scaling,
encoder_bias, noise_std):
super(Ladder, self).__init__()
self.se = StackedEncoders(encoder_in, encoder_sizes,
encoder_activations,
encoder_train_bn_scaling, encoder_bias,
noise_std)
self.de = StackedDecoders(decoder_in, decoder_sizes, image_size)
self.bn_image = torch.nn.BatchNorm1d(image_size, affine=False)