def __init__(self): super(discriminator_cls, self).__init__() self.image_size = 64 self.num_channels = 3 self.embed_dim = 1024 self.projected_embed_dim = 128 self.ndf = 64 self.B_dim = 128 self.C_dim = 16 self.netD_1 = nn.Sequential( # input is (nc) x 64 x 64 nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=False), nn.LeakyReLU(0.2, inplace=True), # state size. (ndf) x 32 x 32 nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=False), nn.BatchNorm2d(self.ndf * 2), nn.LeakyReLU(0.2, inplace=True), # state size. (ndf*2) x 16 x 16 nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=False), nn.BatchNorm2d(self.ndf * 4), nn.LeakyReLU(0.2, inplace=True), # state size. (ndf*4) x 8 x 8 nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=False), nn.BatchNorm2d(self.ndf * 8), nn.LeakyReLU(0.2, inplace=True), ) self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim) self.netD_2 = nn.Sequential( # state size. (ndf*8) x 4 x 4 nn.Conv2d(self.ndf * 8 + self.projected_embed_dim, 1, 4, 1, 0, bias=False), nn.Sigmoid() )
def __init__(self, dataset='youtubers'): super(discriminator, self).__init__() self.image_size = 64 self.num_channels = 3 self.embed_dim = 62 self.projected_embed_dim = 128 self.ndf = 64 self.B_dim = 128 self.C_dim = 16 self.dataset_name = dataset self.conv1 = SpectralNorm(nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=False)) self.conv2 = SpectralNorm(nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=False)) self.conv3 = SpectralNorm(nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=False)) self.conv4 = SpectralNorm(nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=False)) self.disc_linear = nn.Linear(self.ndf * 1, self.ndf) self.disc_linear2 = nn.Linear(31, 31) self.aux_linear = nn.Linear(4*4*512, self.embed_dim+1) self.softmax = nn.Softmax() self.sigmoid = nn.Sigmoid() self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim) self.netD_2 = nn.Sequential( # state size. (ndf*8) x 4 x 4 nn.Conv2d(self.ndf * 8, 1, 4, 1, 0, bias=False), #nn.Conv2d(self.ndf * 8 + self.projected_embed_dim, 1, 4, 1, 0, bias=False), #nn.Sigmoid() )
def __init__(self, improved=False, dataset='youtubers'):
super(discriminator, self).__init__()
self.image_size = 64
self.num_channels = 3
self.embed_dim = 1024
self.projected_embed_dim = 128
self.ndf = 64
self.dataset_name = dataset
if improved:
self.netD_1 = nn.Sequential(
# input is (nc) x 64 x 64
nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=False),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf) x 32 x 32
nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=False),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*2) x 16 x 16
nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=False),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*4) x 8 x 8
nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=False),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*8) x 4 x 4
)
else:
self.netD_1 = nn.Sequential(
# input is (nc) x 64 x 64
nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=False),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf) x 32 x 32
SpectralNorm(
nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=False)),
#nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*2) x 16 x 16
SpectralNorm(
nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1,
bias=False)),
#nn.BatchNorm2d(self.ndf * 4),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*4) x 8 x 8
SpectralNorm(
nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1,
bias=False)),
#nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*8) x 4 x 4
)
self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim)
#Uncomment first layer for concatenation and comment second. For projection do the opposit
#TODO: Handle this!!!
self.netD_2 = nn.Sequential(
nn.Conv2d(self.ndf * 8 + 64, 1, 4, 1, 0, bias=False)
#nn.Conv2d(self.ndf * 8 + self.projected_embed_dim, 1, 4, 1, 0, bias=False)
)
def __init__(self, dataset='youtubers'):
super(discriminator, self).__init__()
self.image_size = 64
self.num_channels = 3
self.embed_dim = 62
self.projected_embed_dim = 128
self.ndf = 64
self.B_dim = 128
self.C_dim = 16
self.dataset_name = dataset
self.conv1 = SpectralNorm(
nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=False))
self.conv2 = SpectralNorm(
nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=False))
self.conv3 = SpectralNorm(
nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=False))
self.conv4 = SpectralNorm(
nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=False))
self.netD_1 = nn.Sequential(
# input is (nc) x 64 x 64
nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=False),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf) x 32 x 32
nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=False),
nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*2) x 16 x 16
nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=False),
nn.BatchNorm2d(self.ndf * 4),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*4) x 8 x 8
nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=False),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True),
)
self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim)
self.netD_2 = nn.Sequential(
# state size. (ndf*8) x 4 x 4
#nn.Conv2d(self.ndf * 8 , 1, 4, 1, 0, bias=False),
#nn.Conv2d(self.ndf * 8 + self.projected_embed_dim, 1, 4, 1, 0, bias=False),
nn.Conv2d(self.ndf * 8 + self.projected_embed_dim,
1,
4,
1,
0,
bias=False),
#nn.Sigmoid()
)
def __init__(self):
super(discriminator, self).__init__()
self.image_size = 64
self.num_channels = 3
self.embed_dim = 2400 # compatible with skip thought (1024)
self.projected_embed_dim = 128
self.ndf = 64
self.B_dim = 128
self.C_dim = 16
self.netD_1 = nn.Sequential(
# input is (nc) x 64 x 64
nn.Conv2d(self.num_channels, self.ndf, 4, 2, 1, bias=True),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf) x 32 x 32
nn.Conv2d(self.ndf, self.ndf * 2, 4, 2, 1, bias=True),
nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*2) x 16 x 16
nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=True),
nn.BatchNorm2d(self.ndf * 4),
nn.LeakyReLU(0.2, inplace=True),
# state size. (ndf*4) x 8 x 8
nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=True),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.ndf * 8, self.ndf * 2, 1, 1, 0),
nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, True),
nn.Conv2d(self.ndf * 2, self.ndf * 2, 3, 1, 1),
nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, True),
nn.Conv2d(self.ndf * 2, self.ndf * 8, 3, 1, 1),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, True)
#output size (ndf*8) x 8 x 8
)
self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim)
self.netD_2 = nn.Sequential(
# state size. (ndf*8) x 4 x 4
nn.Conv2d(self.ndf * 8 + self.projected_embed_dim, self.ndf * 8,
1),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, True),
nn.Conv2d(self.ndf * 8, 1, 4, 1, 0, bias=True),
nn.Sigmoid())
def __init__(self):
super(discriminator2, self).__init__()
self.image_size = 64
self.num_channels = 3
self.embed_dim = 2400 # compatible with skip thought (1024)
self.projected_embed_dim = 128
self.ndf = 64
self.B_dim = 128
self.C_dim = 16
self.encode_img = nn.Sequential(
# state size = 3 x 128 x 128
nn.Conv2d(3, self.ndf, 4, 2, 1, bias=True), # 128 * 128 * ndf
nn.LeakyReLU(0.2, inplace=True),
# state size = ndf x 64 x 64
nn.Conv2d(self.ndf, self.ndf * 2, 3, 1, 1, bias=True),
nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.ndf * 2, self.ndf * 4, 4, 2, 1, bias=True),
nn.BatchNorm2d(self.ndf * 4),
nn.LeakyReLU(0.2, inplace=True), # 32 * 32 * ndf * 4
nn.Conv2d(self.ndf * 4, self.ndf * 8, 4, 2, 1, bias=True),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True), # 16 * 16 * ndf * 8
nn.Conv2d(self.ndf * 8, self.ndf * 8, 3, 1, 1),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True), # 8 * 8 * ndf * 8
nn.Conv2d(self.ndf * 8, self.ndf * 2, 4, 2, 1),
nn.BatchNorm2d(self.ndf * 2),
nn.LeakyReLU(0.2, inplace=True), # 4 * 4 * ndf * 2
nn.Conv2d(self.ndf * 2, self.ndf * 8, 3, 1, 1),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.ndf * 8, self.ndf * 8, 4, 2, 1),
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True)
# nn.Conv2d(self.ndf * 8, self.ndf * 16, 4, 2, 1, bias=True),
# nn.BatchNorm2d(self.ndf * 16),
# nn.LeakyReLU(0.2, inplace=True), # 8 * 8 * ndf * 16
# nn.Conv2d(self.ndf * 16, self.ndf * 32, 4, 2, 1, bias=True),
# nn.BatchNorm2d(self.ndf * 32),
# nn.LeakyReLU(0.2, inplace=True), # 4 * 4 * ndf * 32
# conv3x3(self.ndf * 32, self.ndf * 16),
# nn.BatchNorm2d(self.ndf * 16),
# nn.LeakyReLU(0.2, inplace=True), # 4 * 4 * ndf * 16
# nn.Conv2d(self.ndf * 16, self.ndf * 8, 3, 1, 1, bias=True),
# nn.BatchNorm2d(self.ndf * 8),
# nn.LeakyReLU(0.2, inplace=True) # 4 * 4 * ndf * 8
)
self.outlogitscond = nn.Sequential(
conv3x3(self.ndf * 8 + self.projected_embed_dim, self.ndf * 8),
nn.BatchNorm2d(self.ndf * 8), nn.LeakyReLU(0.2, inplace=True),
nn.Conv2d(self.ndf * 8, 1, kernel_size=4, stride=4), nn.Sigmoid())
# self.outlogits = nn.Sequential(
# nn.Conv2d(self.ndf * 8, 1, kernel_size=4, stride=4),
# nn.Sigmoid()
# )
self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim)
def __init__(self):
super(discriminator, self).__init__()
self.image_size = 64
self.num_channels = 3
self.embed_dim = 1024
self.projected_embed_dim = 128
self.ndf = 64
self.B_dim = 128
self.C_dim = 16
self.netD_1 = nn.Sequential(
# input is (nc) x 64 x 64
nn.Conv2d(self.num_channels,
self.ndf,
3,
stride=2,
padding=1,
bias=False),
nn.BatchNorm2d(self.ndf),
nn.ReLU(True),
# state size. (ndf) x 32 x 32
BasicBlock(self.ndf,
self.ndf * 2,
stride=2,
downsample=make_downsample(self.ndf, self.ndf * 2)),
BasicBlock(self.ndf * 2, self.ndf * 2),
# state size. (ndf*2) x 16 x 16
BasicBlock(self.ndf * 2,
self.ndf * 4,
stride=2,
downsample=make_downsample(self.ndf * 2, self.ndf * 4)),
BasicBlock(self.ndf * 4, self.ndf * 4),
# state size. (ndf*4) x 8 x 8
BasicBlock(self.ndf * 4,
self.ndf * 8,
stride=2,
downsample=make_downsample(self.ndf * 4, self.ndf * 8)),
BasicBlock(self.ndf * 8, self.ndf * 8),
# state size. (ndf*8) x 4 x 4
nn.BatchNorm2d(self.ndf * 8),
nn.LeakyReLU(0.2, inplace=True),
)
self.projector = Concat_embed(self.embed_dim, self.projected_embed_dim)
self.netD_2 = nn.Sequential(
# state size. (ndf*8) x 4 x 4
nn.Conv2d(self.ndf * 8 + self.projected_embed_dim,
1,
4,
1,
0,
bias=False),
nn.Sigmoid())
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)