def __init__(self, downs, n_res, input_dim, dim, norm, activ, pad_type):
super(ContentEncoder, self).__init__()
self.model = []
self.model += [
Conv2dBlock(input_dim,
dim,
7,
1,
3,
norm=norm,
activation=activ,
pad_type=pad_type)
]
for i in range(downs):
self.model += [
Conv2dBlock(dim,
2 * dim,
4,
2,
1,
norm=norm,
activation=activ,
pad_type=pad_type)
]
dim *= 2
self.model += [
ResBlocks(n_res,
dim,
norm=norm,
activation=activ,
pad_type=pad_type)
]
self.model = nn.Sequential(*self.model)
self.output_dim = dim
def __init__(self, ups, n_res, dim, out_dim, res_norm, activ, pad_type):
super(Decoder, self).__init__()
self.model = []
self.model += [
ResBlocks(n_res, dim, res_norm, activ, pad_type=pad_type)
]
for i in range(ups):
self.model += [
nn.Upsample(scale_factor=2),
Conv2dBlock(dim,
dim // 2,
5,
1,
2,
norm='in',
activation=activ,
pad_type=pad_type)
]
dim //= 2
self.model += [
Conv2dBlock(dim,
out_dim,
7,
1,
3,
norm='none',
activation='tanh',
pad_type=pad_type)
]
self.model = nn.Sequential(*self.model)
def __init__(self, hp):
super(GPPatchMcResDis, self).__init__()
assert hp['n_res_blks'] % 2 == 0, 'n_res_blk must be multiples of 2'
self.n_layers = hp['n_res_blks'] // 2
nf = hp['nf']
cnn_f = [Conv2dBlock(3, nf, 7, 1, 3,
pad_type='reflect',
norm='none',
activation='none')]
for i in range(self.n_layers - 1):
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_f += [nn.ReflectionPad2d(1)]
cnn_f += [nn.AvgPool2d(kernel_size=3, stride=2)]
nf = np.min([nf * 2, 1024])
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_c = [Conv2dBlock(nf_out, hp['num_classes'], 1, 1,
norm='none',
activation='lrelu',
activation_first=True)]
self.cnn_f = nn.Sequential(*cnn_f)
self.cnn_c = nn.Sequential(*cnn_c)
def __init__(self):
super(DisModel, self).__init__()
self.n_layers = 6
self.final_size = 1024
nf = 16
cnn_f = [Conv2dBlock(1, nf, 7, 1, 3,
pad_type='reflect',
norm='none',
activation='none')]
for i in range(self.n_layers - 1):
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_f += [nn.ReflectionPad2d(1)]
cnn_f += [nn.AvgPool2d(kernel_size=3, stride=2)]
nf = np.min([nf * 2, 1024])
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_c = [Conv2dBlock(nf_out, self.final_size, IMG_HEIGHT//(2**(self.n_layers-1)), IMG_WIDTH//(2**(self.n_layers-1))+1,
norm='none',
activation='lrelu',
activation_first=True)]
self.cnn_f = nn.Sequential(*cnn_f)
self.cnn_c = nn.Sequential(*cnn_c)
self.bce = nn.BCEWithLogitsLoss()
def __init__(self, num_writers):
super(WriterClaModel, self).__init__()
self.n_layers = 6
nf = 16
cnn_f = [Conv2dBlock(1, nf, 7, 1, 3,
pad_type='reflect',
norm='none',
activation='none')]
for i in range(self.n_layers - 1):
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_f += [nn.ReflectionPad2d(1)]
cnn_f += [nn.AvgPool2d(kernel_size=3, stride=2)]
nf = np.min([nf * 2, 1024])
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_c = [Conv2dBlock(nf_out, num_writers, IMG_HEIGHT//(2**(self.n_layers-1)), IMG_WIDTH//(2**(self.n_layers-1))+1,
norm='none',
activation='lrelu',
activation_first=True)]
self.cnn_f = nn.Sequential(*cnn_f)
self.cnn_c = nn.Sequential(*cnn_c)
self.cross_entropy = nn.CrossEntropyLoss()
def __init__(self, batch_size, downs, ind_im, dim, latent_dim, norm, activ, pad_type):
super(ClassModelEncoder, self).__init__()
s_s_layers = []
dim_size = dim
for i in range(downs[0]):
if i == 0:
s_s_layers.append(Conv2dBlock(ind_im, dim, 4, 2, 1, norm=norm, activation=activ, pad_type=pad_type))
else:
dim = dim*2
s_s_layers.append(Conv2dBlock(dim // 2, dim, 4, 2, 1, norm=norm, activation=activ, pad_type=pad_type))
self.enc_s_s = nn.Sequential(*s_s_layers)
dim = dim_size
s_c_layers = []
for i in range(downs[1]):
if i == 0:
s_c_layers.append(Conv2dBlock(ind_im, dim, 4, 2, 1, norm=norm, activation=activ, pad_type=pad_type))
else:
dim = dim*2
s_c_layers.append(Conv2dBlock(dim // 2, dim, 4, 2, 1, norm=norm, activation=activ, pad_type=pad_type))
dim = dim * 2
s_c_layers.append(Conv2dBlock(dim // 2, dim, 3, 1, 1, norm=norm, activation=activ, pad_type=pad_type))
s_c_layers.append(ResBlocks(2, dim, norm=norm, activation=activ, pad_type=pad_type))
self.enc_s_c = nn.Sequential(*s_c_layers)
self.linear_s = nn.Linear(dim*2, latent_dim)
self.linear_c = nn.Linear(dim, latent_dim)
self.csb = torch.randn(batch_size, dim).cuda()
def __init__(self, downs, n_res, input_dim, dim, norm, activ, pad_type):
super(ContentEncoder, self).__init__()
s_c_layers = []
for i in range(downs):
if i == 0:
s_c_layers.append(Conv2dBlock(input_dim, dim, 4, 2, 1, norm=norm, activation=activ, pad_type=pad_type))
else:
dim = dim * 2
s_c_layers.append(Conv2dBlock(dim // 2, dim, 4, 2, 1, norm=norm, activation=activ, pad_type=pad_type))
dim = dim * 2
s_c_layers.append(Conv2dBlock(dim // 2, dim, 3, 1, 1, norm=norm, activation=activ, pad_type=pad_type))
s_c_layers.append(ResBlocks(n_res, dim, norm=norm, activation=activ, pad_type=pad_type))
self.model = nn.Sequential(*s_c_layers)
self.output_dim = dim
def __init__(self, hp):
super(Generator, self).__init__()
self.mlp = nn.Sequential(
Conv2dBlock(in_dim=hp['in_dim_mlp'],
out_dim=1024,
ks=1,
st=1,
padding=0,
norm='none',
activation='lrelu',
dropout=0.1),
Conv2dBlock(in_dim=1024,
out_dim=960,
ks=1,
st=1,
padding=0,
norm='none',
activation='lrelu',
dropout=0.1),
Conv2dBlock(in_dim=960,
out_dim=864,
ks=1,
st=1,
padding=0,
norm='none',
activation='lrelu',
dropout=0.3),
Conv2dBlock(in_dim=864,
out_dim=784,
ks=1,
st=1,
padding=0,
norm='none',
activation='lrelu',
dropout=0.5),
Conv2dBlock(in_dim=784,
out_dim=720,
ks=1,
st=1,
padding=0,
norm='none',
activation='lrelu',
dropout=0.5),
nn.Conv2d(720, hp['out_dim_mlp'], 1, stride=1, padding=0),
)
self.sigmoid = nn.Sigmoid()
def __init__(self, ups, n_res, dim, out_dim, res_norm, activ, pad_type):
super(Decoder, self).__init__()
self.model = []
self.model += [ResBlocks(n_res, dim, res_norm, activ, pad_type=pad_type)]
for i in range(ups):
self.model.append(AdainUpResBlock(dim, activation=activ, pad_type=pad_type))
dim = dim // 2
self.model.append(Conv2dBlock(dim, out_dim, 3, 1, 1, norm='none', activation='tanh', pad_type=pad_type))
self.model = nn.Sequential(*self.model)
def __init__(self, hp):
super(GPPatchMcResDis, self).__init__()
#InceptionBlock = Conv2dBlock
assert hp['n_res_blks'] % 2 == 0, 'n_res_blk must be multiples of 2'
self.n_layers = hp['n_res_blks'] // 2
nf = hp['nf']
input_channels = hp['input_nc']
cnn_f = [
Conv2dBlock(input_channels,
nf,
KERNEL_SIZE_7,
1,
3,
pad_type='reflect',
norm='none',
activation='none')
]
for i in range(self.n_layers - 1):
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_f += [nn.ReflectionPad2d(1)]
cnn_f += [nn.AvgPool2d(kernel_size=3, stride=2)]
nf = np.min([nf * 2, 1024])
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_c = [
Conv2dBlock(nf_out,
hp['num_classes'],
1,
1,
norm='none',
activation='lrelu',
activation_first=True)
]
self.cnn_f = nn.Sequential(*cnn_f)
self.cnn_c = nn.Sequential(*cnn_c)
#self.register_backward_hook(debug.printgradnorm)
self.debug = Debugger()
def __init__(self, args):
super(Refiner, self).__init__()
c_up = args.c_up // 2 # 32
down = args.down # 2
self.model = [
Conv2dBlock(6, c_up, 7, 1, 3, norm='in',
pad_type='reflect'), # RGB + target
]
for i in range(down):
self.model.append(
Conv2dBlock(c_up,
2 * c_up,
4,
2,
1,
norm='in',
pad_type='reflect'))
c_up *= 2
self.model.append(
ResBlocks(5,
c_up,
norm='in',
activation='relu',
pad_type='reflect'))
for i in range(down):
self.model.append(
UpConv2dBlock(c_up,
norm='in',
activation='relu',
pad_type='reflect'))
c_up //= 2
self.model.append(
Conv2dBlock(c_up,
3,
7,
1,
padding=3,
norm='none',
activation='none',
pad_type='reflect'))
self.model = nn.Sequential(*self.model)
def __init__(self, downs, n_res, input_dim, dim, norm, activ, pad_type):
super(ContentEncoder, self).__init__()
#InceptionBlock = Conv2dBlock
self.model = []
self.model += [
InceptionBlock(input_dim,
dim,
KERNEL_SIZE_7,
1,
3,
norm=norm,
activation=activ,
pad_type=pad_type)
]
"""
for i in range(downs):
self.model += [InceptionBlock(dim, 2 * dim, KERNEL_SIZE_4, 1,#2,
1,
norm=norm,
activation=activ,
pad_type=pad_type)]
if (i == downs-1):
self.model += [
nn.MaxPool2d(KERNEL_SIZE_4, 2, padding=1)
]
else:
self.model += [
nn.MaxPool2d(KERNEL_SIZE_4, 2, padding=0)
]
dim *= 2
"""
for i in range(downs):
self.model += [
Conv2dBlock(dim,
2 * dim,
KERNEL_SIZE_4,
2,
1,
norm=norm,
activation=activ,
pad_type=pad_type)
]
dim *= 2
self.model += [
ResBlocks(n_res,
dim,
norm=norm,
activation=activ,
pad_type=pad_type,
inception=True)
]
self.model = nn.Sequential(*self.model)
self.output_dim = dim
def __init__(self, downs, ind_im, dim, latent_dim, norm, activ, pad_type):
super(ClassModelEncoder, self).__init__()
self.model = []
self.model += [
Conv2dBlock(ind_im,
dim,
7,
1,
3,
norm=norm,
activation=activ,
pad_type=pad_type)
]
for i in range(2):
self.model += [
Conv2dBlock(dim,
2 * dim,
4,
2,
1,
norm=norm,
activation=activ,
pad_type=pad_type)
]
dim *= 2
for i in range(downs - 2):
self.model += [
Conv2dBlock(dim,
dim,
4,
2,
1,
norm=norm,
activation=activ,
pad_type=pad_type)
]
self.model += [nn.AdaptiveAvgPool2d(1)]
self.model += [nn.Conv2d(dim, latent_dim, 1, 1, 0)]
self.model = nn.Sequential(*self.model)
self.output_dim = dim
def __init__(self, hp):
super(GPPatchMcResDis, self).__init__()
assert hp['n_res_blks'] % 2 == 0, 'n_res_blk must be multiples of 2'
self.n_layers = hp['n_res_blks'] // 2 # 5
nf = hp['nf'] # 64
cnn_f = [
Conv2dBlock(3,
nf,
7,
1,
3,
pad_type='reflect',
norm='none',
activation='none')
] # 第一个卷积层提取特征，输出batchsize*128*128*64维度向量
for i in range(
self.n_layers -
1): # 128->256->512->1024,输出batchsize*8*8*1024维度向量，在原论文条件下
nf_out = np.min([nf * 2, 1024])
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu', 'none')]
cnn_f += [nn.ReflectionPad2d(1)]
cnn_f += [nn.AvgPool2d(kernel_size=3, stride=2)]
nf = np.min([nf * 2, 1024])
nf_out = np.min([nf * 2, 1024]) # 1024
cnn_f += [ActFirstResBlock(nf, nf, None, 'lrelu', 'none')]
cnn_f += [ActFirstResBlock(nf, nf_out, None, 'lrelu',
'none')] # 再经过两个ResBlock,得到1024深度
cnn_c = [
Conv2dBlock(nf_out,
hp['num_classes'],
1,
1,
norm='none',
activation='lrelu',
activation_first=True)
] # 最后再经过一个卷积层得到各个种类的概率：输出batchsize*patch*patch*num_class的四维向量
self.cnn_f = nn.Sequential(*cnn_f)
self.cnn_c = nn.Sequential(*cnn_c)
def __init__(self, hp):
super(Discriminator, self).__init__()
self.fc = Conv2dBlock(in_dim=hp['in_dim_fc'],
out_dim=hp['out_dim_fc'],
ks=1,
st=1,
padding=0,
norm=hp['norm_fc'],
activation=hp['activ_fc'],
dropout=hp['drop_fc'])
self.pred = nn.Conv2d(hp['out_dim_fc'], 1, 1, stride=1, padding=0)
self.sigmoid = nn.Sigmoid()
self.cls = nn.Conv2d(hp['out_dim_fc'],
hp['out_dim_cls'],
1,
stride=1,
padding=0)
