def __init__(self, input_channel, n_blocks=9, dropout=False):
super(ResnetGenerator, self).__init__()
self.conv0 = ConvBN(input_channel, 32, 7, 1)
self.conv1 = ConvBN(32, 64, 3, 2, padding=1)
self.conv2 = ConvBN(64, 128, 3, 2, padding=1)
dim = 128
self.resnet_blocks = []
for i in range(n_blocks):
block = self.add_sublayer("generator_%d" % (i + 1),
ResnetBlock(dim, dropout))
self.resnet_blocks.append(block)
self.deconv0 = DeConvBN(dim,
32 * 2,
3,
2,
padding=[1, 1],
outpadding=[0, 1, 0, 1])
self.deconv1 = DeConvBN(32 * 2,
32,
3,
2,
padding=[1, 1],
outpadding=[0, 1, 0, 1])
self.conv3 = ConvBN(32,
input_channel,
7,
1,
norm=False,
act=False,
use_bias=True)
def __init__(self, input_channel, d_dims=64, d_nlayers=3):
super(NLayerDiscriminator, self).__init__()
self.conv0 = ConvBN(input_channel,
d_dims,
4,
2,
1,
norm=False,
use_bias=True,
relufactor=0.2)
nf_mult, nf_mult_prev = 1, 1
self.conv_layers = []
for n in range(1, d_nlayers):
nf_mult_prev = nf_mult
nf_mult = min(2**n, 8)
conv = self.add_sublayer(
'discriminator_%d' % (n),
ConvBN(d_dims * nf_mult_prev,
d_dims * nf_mult,
4,
2,
1,
relufactor=0.2))
self.conv_layers.append(conv)
nf_mult_prev = nf_mult
nf_mult = min(2**d_nlayers, 8)
self.conv4 = ConvBN(d_dims * nf_mult_prev,
d_dims * nf_mult,
4,
1,
1,
relufactor=0.2)
self.conv5 = ConvBN(d_dims * nf_mult,
1,
4,
1,
1,
norm=False,
act=None,
use_bias=True,
relufactor=0.2)
def __init__(self, dim, dropout=False):
super(ResnetBlock, self).__init__()
self.dropout = dropout
self.conv0 = ConvBN(dim, dim, 3, 1)
self.conv1 = ConvBN(dim, dim, 3, 1, act=None)