def __init__(self,
num_features,
num_classes,
activation=F.relu,
dim_emb=128):
super(SNResNetConcatDiscriminator, self).__init__()
self.num_features = num_features
self.num_classes = num_classes
self.dim_emb = dim_emb
self.activation = activation
self.block1 = OptimizedBlock(3, num_features)
self.block2 = Block(num_features,
num_features * 2,
activation=activation,
downsample=True)
self.block3 = Block(num_features * 2,
num_features * 4,
activation=activation,
downsample=True)
if num_classes > 0:
self.l_y = utils.spectral_norm(nn.Embedding(num_classes, dim_emb))
self.block4 = Block(num_features * 4 + dim_emb,
num_features * 8,
activation=activation,
downsample=True)
self.block5 = Block(num_features * 8,
num_features * 16,
activation=activation,
downsample=True)
self.l6 = utils.spectral_norm(nn.Linear(num_features * 16, 1))
self._initialize()
def __init__(self, num_features=64, num_classes=0, activation=F.relu):
super(SNResNetProjectionDiscriminator, self).__init__()
self.num_features = num_features
self.num_classes = num_classes
self.activation = activation
self.block1 = OptimizedBlock(1, num_features)
self.block2 = Block(num_features,
num_features * 2,
activation=activation,
downsample=True)
self.block3 = Block(num_features * 2,
num_features * 4,
activation=activation,
downsample=True)
self.block4 = Block(num_features * 4,
num_features * 8,
activation=activation,
downsample=True)
self.block5 = Block(num_features * 8,
num_features * 16,
activation=activation,
downsample=True)
self.l6 = utils.spectral_norm(nn.Linear(num_features * 16, 1))
if num_classes > 0:
self.l_y = utils.spectral_norm(
nn.Embedding(num_classes, num_features * 16))
self._initialize()
def __init__(self, num_features=64, num_classes=0, activation=F.relu):
super(VGG_Discriminator, self).__init__()
self.num_features = num_features
self.num_classes = num_classes
self.activation = activation
self.block1 = OptimizedBlock(3, num_features // 4)
self.block2 = Block(num_features // 4,
num_features // 2,
activation=activation,
downsample=True)
self.block3 = Block(num_features // 2,
num_features,
activation=activation,
downsample=True)
self.block4 = Block(num_features,
num_features * 2,
activation=activation,
downsample=True)
self.block5 = Block(num_features * 2,
num_features * 4,
activation=activation,
downsample=True)
self.l4 = utils.spectral_norm(nn.Linear(num_features * 4, 1))
self.linear_mi = nn.Linear(num_features * 4, num_classes)
self.linear_c = nn.Linear(num_features * 4, num_classes)
if num_classes > 0:
self.l_y = utils.spectral_norm(
nn.Embedding(num_classes, num_features * 4))
self._initialize()
def __init__(self,
num_features=64,
num_classes=0,
activation=F.relu,
transform_space=None):
super(SNResNetProjectionDiscriminator, self).__init__()
self.num_features = num_features
self.num_classes = num_classes
self.activation = activation
self.block1 = OptimizedBlock(3, num_features)
self.block2 = Block(num_features,
num_features * 2,
activation=activation,
downsample=True)
self.block3 = Block(num_features * 2,
num_features * 4,
activation=activation,
downsample=True)
self.block4 = Block(num_features * 4,
num_features * 8,
activation=activation,
downsample=True)
self.block5 = Block(num_features * 8,
num_features * 16,
activation=activation,
downsample=True)
self.l6 = utils.spectral_norm(nn.Linear(num_features * 16, 1))
if transform_space is not None:
# self.transformer = nn.Transformer(d_model=1024,nhead=4,num_encoder_layers=4).encoder
encoder_layer = nn.TransformerEncoderLayer(d_model=1024, nhead=8)
transformer_encoder = nn.TransformerEncoder(encoder_layer,
num_layers=4)
self.transformer = transformer_encoder
if transform_space == 'batch':
self.transform_space = 'batch'
elif transform_space == 'embeddings':
self.transform_space = 'embeddings'
else:
print(
"You specified a transform space but did not provide an appropriate one. Exiting"
)
exit()
else:
#transform_space is None
self.transform_space = None
if num_classes > 0:
self.l_y = utils.spectral_norm(
nn.Embedding(num_classes, num_features * 16))
self._initialize()
def __init__(self, num_features=64, num_classes=0, activation=F.relu):
super(Animal_Discriminator, self).__init__()
self.num_features = num_features
self.num_classes = num_classes
self.activation = activation
self.block1 = OptimizedBlock(3, num_features // 8) # 128
self.block2 = Block(num_features // 8,
num_features // 4,
activation=activation,
downsample=True) # 64
self.block3 = Block(num_features // 4,
num_features // 2,
activation=activation,
downsample=True) # 32
self.block4 = Block(num_features // 2,
num_features,
activation=activation,
downsample=True) # 16
self.block5 = Self_Attn(num_features) # 8
self.block6 = Block(
num_features,
num_features * 2, # 8
activation=activation,
downsample=True)
self.block7 = Block(
num_features * 2,
num_features * 4, # 4
activation=activation,
downsample=True)
self.l4 = utils.spectral_norm(nn.Linear(num_features * 4, 1))
if num_classes > 0:
self.l_y = utils.spectral_norm(
nn.Embedding(num_classes, num_features * 4))
self._initialize()
def __init__(self, num_features=32, num_classes=0, activation=F.relu):
super(Omniglot_Discriminator, self).__init__()
self.num_features = num_features
self.num_classes = num_classes
self.activation = activation
self.block1 = OptimizedBlock(1, num_features)
self.block2 = Block(num_features,
num_features * 2,
activation=activation,
downsample=True)
self.block3 = Self_Attn(num_features * 2)
self.block4 = Block(num_features * 2,
num_features * 4,
activation=activation,
downsample=True)
self.l4 = utils.spectral_norm(nn.Linear(num_features * 4, 1))
if num_classes > 0:
self.l_y = utils.spectral_norm(
nn.Embedding(num_classes, num_features * 4))
self._initialize()