def __init__(self,
num_features=32,
dim_z=128,
bottom_width=7,
activation=F.relu,
num_classes=0,
distribution='normal'):
super(Omniglot_Generator, self).__init__()
self.num_features = num_features
self.dim_z = dim_z
self.bottom_width = bottom_width
self.activation = activation
self.num_classes = num_classes
self.distribution = distribution
self.l1 = nn.Linear(dim_z, 4 * num_features * bottom_width**2)
self.block2 = Block(num_features * 4,
num_features * 2,
activation=activation,
upsample=True,
num_classes=num_classes)
self.block3 = Block(num_features * 2,
num_features,
activation=activation,
upsample=True,
num_classes=num_classes)
self.b4 = nn.BatchNorm2d(num_features)
self.conv4 = nn.Conv2d(num_features, 1, 1, 1)
def __init__(self, num_features=64, dim_z=128, bottom_width=4,
activation=F.relu, num_classes=0, distribution='normal'):
super(VGG_Generator, self).__init__()
self.num_features = num_features
self.dim_z = dim_z
self.bottom_width = bottom_width
self.activation = activation
self.num_classes = num_classes
self.distribution = distribution
self.l1 = nn.Linear(dim_z, 4 * num_features * bottom_width ** 2)
self.block2 = Block(num_features * 4, num_features * 2,
activation=activation, upsample=True,
num_classes=num_classes)
self.block3 = Block(num_features * 2, num_features,
activation=activation, upsample=True,
num_classes=num_classes)
self.block4 = Block(num_features, num_features // 2,
activation=activation, upsample=True,
num_classes=num_classes)
self.block5 = Self_Attn(num_features // 2)
self.block6 = Block(num_features // 2, num_features // 4,
activation=activation, upsample=True,
num_classes=num_classes)
self.block7 = Self_Attn(num_features // 4)
self.b4 = nn.BatchNorm2d(num_features // 4)
self.conv4 = nn.Conv2d(num_features // 4, 3, 1, 1)
def __init__(self,
num_features=64,
dim_z=128,
bottom_width=4,
activation=F.relu,
num_classes=0,
distribution='normal'):
super(ResNetGenerator, self).__init__()
self.num_features = num_features
self.dim_z = dim_z
self.bottom_width = bottom_width
self.activation = activation
self.num_classes = num_classes
self.distribution = distribution
self.l1 = nn.Linear(dim_z, 16 * num_features * bottom_width**2)
self.block2 = Block(num_features * 16,
num_features * 16,
activation=activation,
upsample=True,
num_classes=num_classes)
self.block3 = Block(num_features * 16,
num_features * 8,
activation=activation,
upsample=True,
num_classes=num_classes)
self.block4 = Block(num_features * 8,
num_features * 4,
activation=activation,
upsample=True,
num_classes=num_classes)
self.block5 = Block(num_features * 4,
num_features * 2,
activation=activation,
upsample=True,
num_classes=num_classes)
self.block6 = Block(num_features * 2,
num_features,
activation=activation,
upsample=True,
num_classes=num_classes)
self.b7 = nn.BatchNorm2d(num_features)
self.conv7 = nn.Conv2d(num_features, 3, 1, 1)