def __init__(self, config: HiDDenConfiguration):
super(Encoder, self).__init__()
self.H = config.H
self.W = config.W
self.conv_channels = config.encoder_channels
self.num_blocks = config.encoder_blocks
self.cover_dependent = config.cover_dependent
# self.residual = config.residual
if self.cover_dependent == 1:
layers = [ConvBNRelu(3, self.conv_channels)]
else:
layers = [ConvBNRelu(config.message_length, self.conv_channels)]
for _ in range(config.encoder_blocks - 1):
layer = ConvBNRelu(self.conv_channels, self.conv_channels)
layers.append(layer)
self.conv_layers = nn.Sequential(*layers)
if self.cover_dependent == 1:
self.after_concat_layer = ConvBNRelu(
self.conv_channels + 3 + config.message_length,
self.conv_channels)
else:
self.after_concat_layer = ConvBNRelu(config.message_length,
self.conv_channels)
self.final_layer = nn.Conv2d(self.conv_channels, 3, kernel_size=1)
self.final_tanh = nn.Tanh()
self.factor = 10 / 255
def __init__(self, config: HiDDenConfiguration):
super(Discriminator, self).__init__()
layers = [ConvBNRelu(3, config.discriminator_channels)]
for _ in range(config.discriminator_blocks-1):
layers.append(ConvBNRelu(config.discriminator_channels, config.discriminator_channels))
layers.append(nn.AdaptiveAvgPool2d(output_size=(1, 1)))
self.before_linear = nn.Sequential(*layers)
self.linear = nn.Linear(config.discriminator_channels, 1)
def __init__(self, config: HiDDenConfiguration):
super(Encoder, self).__init__()
self.conv_channels = config.encoder_channels
self.num_blocks = config.encoder_blocks
layers = [ConvBNRelu(config.input_channels, self.conv_channels)]
for _ in range(config.encoder_blocks-1):
layer = ConvBNRelu(self.conv_channels, self.conv_channels)
layers.append(layer)
self.conv_layers = nn.Sequential(*layers)
self.after_concat_layer = ConvBNRelu(self.conv_channels + config.input_channels + config.message_length, self.conv_channels)
self.final_layer = nn.Conv2d(self.conv_channels, config.input_channels, kernel_size=1)
def __init__(self, config: HiDDenConfiguration):
super().__init__()
self.channels = config.decoder_channels
self.num_blocks = config.decoder_blocks
layers = [ConvBNRelu(3, self.channels)]
for _ in range(self.num_blocks - 1):
layer = ConvBNRelu(self.channels, self.channels)
layers.append(layer)
layers.append(ConvBNRelu(self.channels, config.message_length))
layers.append(nn.AdaptiveAvgPool2d(output_size=(1, 1)))
self.layers = nn.Sequential(*layers)
self.linear = nn.Linear(config.message_length, config.message_length)