def __init__(self, opt):
super(BaseEncoder, self).__init__()
self.enc_channels = opt.enc_channels
self.num_downsamples = opt.num_downsamples
self.padding_layer = get_padding(opt.enc_padding)
self.activ_layer = get_activ(opt.enc_activ)
self.norm_layer = get_norm(opt.enc_norm)
self.p_dropout = opt.enc_dropout
self._build_layers(opt)
def __init__(self,
in_channels,
out_channels,
activ='gelu',
norm='instance'):
super().__init__()
self.in_channels = in_channels
self._out_channels = out_channels
self.activ_layer = get_activ(activ)
self.norm_layer = get_norm(type)
self._build_layers()
def __init__(self, opt):
super(BasePatchDiscriminator, self).__init__()
self.dsc_channels = opt.dsc_channels
self.dsc_layers = opt.dsc_layers
self.dsc_scales = opt.dsc_scales
self.norm_layer = get_norm(opt.dsc_norm)
self.activ_layer = get_activ(opt.dsc_activ)
self.padding_layer = get_padding(opt.dsc_padding)
self.p_dropout = opt.dsc_dropout
self.num_scales = len(self.dsc_scales)
self._build_layers(opt)
def __init__(self, in_channels, out_channels, num_blocks, num_layers, opt):
super(ScalingResidualBlock, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.num_blocks = num_blocks
self.num_layers = num_layers
self.num_channels = max(self.in_channels, self.out_channels)
self.padding_layer = get_padding(opt.latent_padding)
self.activ_layer = get_activ(opt.latent_activ)
self.norm_layer = get_norm(opt.latent_norm)
self.p_dropout = opt.latent_dropout
self._build_layers(opt)
def __init__(self,
in_channels,
out_channels,
dropout=0.,
activ='gelu',
norm='instance',
padding='reflection'):
super().__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.norm_layer = get_norm(norm)
self.activ_layer = get_activ(activ)
self.padding_layer = get_padding(padding)
self.p_dropout = dropout
self._build_layers()
def __init__(self, opt, out_channels, type='continous'):
super(BaseDecoder, self).__init__()
self.out_channels = out_channels
self.enc_channels = opt.enc_channels
self.dec_channels = opt.dec_channels
self.num_upsamples = opt.num_downsamples
# these are layer constructors, not implicit layers
self.norm_layer = get_norm(opt.dec_norm)
self.activ_layer = get_activ(opt.dec_activ)
self.padding_layer = get_padding(opt.dec_padding)
self.p_dropout = opt.dec_dropout
if type == 'continous':
self.activ_final = nn.Tanh()
if type == 'probabilities':
self.activ_final = nn.Sigmoid()
self._build_layers(opt)
def __init__(self, opt):
super(AggregatedLargeDilationEncoder, self).__init__()
self.enc_channels = opt.enc_channels
self.num_downsamples = opt.num_downsamples
self.dilations = opt.enc_dilations
final_dilation = self.dilations[-1]
for i in range(1, self.num_downsamples + 1):
# for each downsample add an increasing dilation
# why? no reason
self.dilations = self.dilations + [final_dilation + i * 2]
self.dil_channels = opt.dil_channels
self.padding_layer = get_padding(opt.enc_padding)
self.activ_layer = get_activ(opt.enc_activ)
self.norm_layer = get_norm(opt.enc_norm)
self.p_dropout = opt.enc_dropout
self._build_layers(opt)
def __init__(self, opt, out_channels, type='continous'):
super(AggregatedLargeDilationDecoder, self).__init__()
self.out_channels = out_channels
self.enc_channels = opt.enc_channels
self.dec_channels = opt.dec_channels
self.num_upsamples = opt.num_downsamples
self.dil_channels = opt.dil_channels
self.dilations = opt.dec_dilations
final_dilation = self.dilations[-1]
for i in range(1, self.num_upsamples + 1):
self.dilations = self.dilations + [final_dilation + i * 2]
# these are layer constructors, not implicit layers
self.norm_layer = get_norm(opt.dec_norm)
self.activ_layer = get_activ(opt.dec_activ)
if type == 'continous':
self.activ_final = nn.Tanh()
if type == 'probabilities':
self.activ_final = nn.Sigmoid()
self.padding_layer = get_padding(opt.dec_padding)
self.p_dropout = opt.dec_dropout
self._build_layers(opt)
def __init__(self,
in_channels,
out_channels,
dil_channels,
dilations,
dropout=0.,
activ='gelu',
norm='instance',
padding='reflection',
residual=False):
super().__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.dil_channels = dil_channels
self.dilations = dilations
self.num_dilations = len(dilations)
self.residual = residual
self.norm_layer = get_norm(norm)
self.activ_layer = get_activ(activ)
self.padding_layer = get_padding(padding)
self.p_dropout = dropout
self._build_layers()