def __init__(self, in_ch, out_ch, downsampling='stride'):
super().__init__()
self.conv = ConvBlock(in_ch,
out_ch,
sampling='same',
normalization='instance')
self.down = ConvBlock(out_ch,
out_ch,
sampling=downsampling,
normalization='instance')
def __init__(self, input_dim, out_ch):
super().__init__()
self.input_dim = input_dim
self.conv1 = ConvBlock(input_dim,
out_ch,
sampling='same',
normalization='pixel')
self.conv2 = ConvBlock(out_ch,
out_ch,
sampling='same',
normalization='pixel')
def __init__(self):
super(SRCNN, self).__init__()
self.layers = nn.Sequential(
# ConvBlock(3, 64, kernel_size=9, padding=4),
nn.Conv2d(3, 64, kernel_size=(1, 1), stride=(1, 1), bias=False),
nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=4),
nn.Conv2d(3, 64, 256, kernel_size=(1, 1),stride=(1, 1), bias=False),
nn.Conv2d()
ConvBlock(64, 32, kernel_size=1, padding=0),
ConvBlock(32, 3, kernel_size=5, padding=2, activation=None))
def __init__(self, in_ch, out_ch):
super().__init__()
self.mb_stddev = MiniBatchStddev()
self.conv = ConvBlock(in_ch + 1,
out_ch,
sampling='same',
normalization='instance')
self.down = ConvBlock(out_ch,
out_ch,
kernel_size=4,
padding='valid',
normalization=None)
self.norm = torch.nn.LayerNorm(out_ch)
def __init__(self, in_ch, out_ch, upsampling='upsampling'):
super().__init__()
self.up = ConvBlock(in_ch,
in_ch,
sampling=upsampling,
normalization='pixel')
self.conv1 = ConvBlock(in_ch,
in_ch,
sampling='same',
normalization='pixel')
self.conv2 = ConvBlock(in_ch,
out_ch,
sampling='same',
normalization='pixel')
def discriminator(img_height, img_width):
model = keras.Sequential()
model.add(
keras.layers.experimental.preprocessing.Rescaling(
scale=1. / 127.5,
offset=-1,
input_shape=(img_height, img_width, 3)))
model.add(ConvBlock(128, 5))
model.add(ConvBlock(256, 5))
model.add(ConvBlock(512, 5))
model.add(ConvBlock(1024, 5))
model.add(layers.Flatten())
model.add(layers.Dense(1))
return model
def __init__(self, num_classes, attention=True, normalize_attn=True, init='default'):
super(AttnVGG_before, self).__init__()
self.attention = attention
# conv blocks
self.conv_block1 = ConvBlock(3, 64, 2)
self.conv_block2 = ConvBlock(64, 128, 2)
self.conv_block3 = ConvBlock(128, 256, 3)
self.conv_block4 = ConvBlock(256, 512, 3)
self.conv_block5 = ConvBlock(512, 512, 3)
# self.conv_block6 = ConvBlock(512, 512, 2, pool=True)
# self.dense = nn.Conv2d(in_channels=512, out_channels=512, kernel_size=int(im_size/32), padding=0, bias=True)
self.dense1 = nn.Conv2d(in_channels=512, out_channels=4096, kernel_size=7, padding=0, bias=True)
self.dense2 = nn.Conv2d(in_channels=4096, out_channels=4096, kernel_size=1, padding=0, bias=True)
self.dense3 = nn.Conv2d(in_channels=4096, out_channels=512, kernel_size=1, padding=0, bias=True)
# Projectors & Compatibility functions
if self.attention:
self.projector = ProjectorBlock(256, 512)
self.attn1 = LinearAttentionBlock(in_features=512, normalize_attn=normalize_attn)
self.attn2 = LinearAttentionBlock(in_features=512, normalize_attn=normalize_attn)
self.attn3 = LinearAttentionBlock(in_features=512, normalize_attn=normalize_attn)
# final classification layer
if self.attention:
self.classify = nn.Linear(in_features=512 * 3, out_features=num_classes, bias=True)
else:
self.classify = nn.Linear(in_features=512, out_features=num_classes, bias=True)
# initialize
if init == 'kaimingNormal':
weights_init_kaimingNormal(self)
elif init == 'kaimingUniform':
weights_init_kaimingUniform(self)
elif init == 'xavierNormal':
weights_init_xavierNormal(self)
elif init == 'xavierUniform':
weights_init_xavierUniform(self)
else:
print("Initializing Default weights")
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.normal_(m.weight, 0, 0.01)
nn.init.constant_(m.bias, 0)
def __init__(self, config):
super(Decoder, self).__init__()
channels = config.dec_channels
kernel_size = config.dec_kernel_size
stride = config.dec_stride
res_norm = 'none' # no adain in res
norm = 'none'
pad_type = 'reflect'
acti = 'lrelu'
layers = []
n_resblk = config.dec_resblks
n_conv = config.dec_up_n
bt_channel = config.dec_bt_channel # #channels at the bottleneck
layers += get_norm_layer('adain', channels[0]) # adain before everything
for i in range(n_resblk):
layers.append(BottleNeckResBlock(kernel_size, channels[0], bt_channel, channels[0],
pad_type=pad_type, norm=res_norm, acti=acti))
for i in range(n_conv):
layers.append(Upsample(scale_factor=2, mode='nearest'))
cur_acti = 'none' if i == n_conv - 1 else acti
cur_norm = 'none' if i == n_conv - 1 else norm
layers += ConvBlock(kernel_size, channels[i], channels[i + 1], stride=stride,
pad_type=pad_type, norm=cur_norm, acti=cur_acti)
self.model = nn.Sequential(*layers)
self.channels = channels
def __init__(self, in_ch):
super().__init__()
self.conv = ConvBlock(in_ch,
3,
sampling='same',
normalization=None,
activation=torch.nn.Tanh())
def __init__(self, *channels):
super().__init__()
self.convs = torch.nn.Sequential(*[
ConvBlock(channels[i], channels[i + 1], normalization='batch')
for i in range(len(channels) - 2)
])
self.up = UpModule(channels[-2], channels[-1])
def __init__(self, config):
super(PatchDis, self).__init__()
channels = config.disc_channels
down_n = config.disc_down_n
ks = config.disc_kernel_size
stride = config.disc_stride
pool_ks = config.disc_pool_size
pool_stride = config.disc_pool_stride
out_dim = config.num_classes
assert down_n + 1 == len(channels)
cnn_f = ConvLayers(kernel_size=ks, in_channels=channels[0], out_channels=channels[0])
for i in range(down_n):
cnn_f += [ActiFirstResBlock(kernel_size=ks, in_channels=channels[i], out_channels=channels[i], stride=stride, acti='lrelu', norm='none')]
cnn_f += [ActiFirstResBlock(kernel_size=ks, in_channels=channels[i], out_channels=channels[i + 1], stride=stride, acti='lrelu', norm='none')]
cnn_f += [get_conv_pad(pool_ks, pool_stride)]
cnn_f += [nn.AvgPool1d(kernel_size=pool_ks, stride=pool_stride)]
cnn_f += [ActiFirstResBlock(kernel_size=ks, in_channels=channels[-1], out_channels=channels[-1], stride=stride, acti='lrelu', norm='none')]
cnn_f += [ActiFirstResBlock(kernel_size=ks, in_channels=channels[-1], out_channels=channels[-1], stride=stride, acti='lrelu', norm='none')]
cnn_c = ConvBlock(kernel_size=ks, in_channels=channels[-1], out_channels = out_dim,
stride=1, norm='none', acti='lrelu', acti_first=True)
self.cnn_f = nn.Sequential(*cnn_f)
self.cnn_c = nn.Sequential(*cnn_c)
self.device = config.device
def __init__(self, in_ch=3):
super().__init__()
self.down1 = DownBlock(in_ch, 64, 64)
self.down2 = DownBlock(64, 128, 128)
self.down3 = DownBlock(128, 256, 256, 256)
self.down4 = DownBlock(256, 512, 512, 512)
self.down5 = DownBlock(512, 512, 512, 512)
self.up1 = UpBlock(512, 512, 512, 256)
self.up2 = UpBlock(512 + 256, 512, 256)
self.up3 = UpBlock(256 + 512, 512, 128)
self.up4 = UpBlock(128 + 256, 256, 64)
self.up5 = UpBlock(64 + 128, 128, 32)
self.last_conv = ConvBlock(32 + 64, in_ch, activation=torch.nn.Tanh())
def __init__(self,
num_classes,
normalize_attn=False,
init='kaimingNormal'):
super(AttnVGG_grid, self).__init__()
# conv blocks
self.conv_block1 = ConvBlock(3, 64, 2)
self.conv_block2 = ConvBlock(64, 128, 2)
self.conv_block3 = ConvBlock(128, 256, 3)
self.conv_block4 = ConvBlock(256, 512, 3)
self.conv_block5 = ConvBlock(512, 512, 3)
self.pool = nn.AvgPool2d(2, stride=1)
self.attn1 = GridAttentionBlock(256,
512,
256,
4,
normalize_attn=normalize_attn)
self.attn2 = GridAttentionBlock(512,
512,
256,
2,
normalize_attn=normalize_attn)
# final classification layer
self.classify = nn.Linear(in_features=512 + 512 + 256,
out_features=num_classes,
bias=True)
# initialize
if init == 'kaimingNormal':
weights_init_kaimingNormal(self)
elif init == 'kaimingUniform':
weights_init_kaimingUniform(self)
elif init == 'xavierNormal':
weights_init_xavierNormal(self)
elif init == 'xavierUniform':
weights_init_xavierUniform(self)
else:
raise NotImplementedError("Invalid type of initialization!")
def __init__(self,
im_size,
num_classes,
attention=True,
normalize_attn=True,
init='xavierUniform',
interpolate=False):
super(AttnVGG_after, self).__init__()
if interpolate:
self.interpolate = Interpolate(size=(im_size, im_size),
mode='nearest')
else:
self.interpolate = None
self.attention = attention
# conv blocks
self.conv_block1 = ConvBlock(3, 64, 2)
self.conv_block2 = ConvBlock(64, 128, 2)
self.conv_block3 = ConvBlock(128, 256, 3)
self.conv_block4 = ConvBlock(256, 512, 3)
self.conv_block5 = ConvBlock(512, 512, 3)
self.conv_block6 = ConvBlock(512, 512, 2, pool=True)
self.dense = nn.Conv2d(in_channels=512,
out_channels=512,
kernel_size=int(im_size / 32),
padding=0,
bias=True)
# Projectors & Compatibility functions
if self.attention:
self.projector = ProjectorBlock(256, 512)
self.attn1 = LinearAttentionBlock(in_features=512,
normalize_attn=normalize_attn)
self.attn2 = LinearAttentionBlock(in_features=512,
normalize_attn=normalize_attn)
self.attn3 = LinearAttentionBlock(in_features=512,
normalize_attn=normalize_attn)
# final classification layer
if self.attention:
self.classify = nn.Linear(in_features=512 * 3,
out_features=num_classes,
bias=True)
else:
self.classify = nn.Linear(in_features=512,
out_features=num_classes,
bias=True)
# initialize
if init == 'kaimingNormal':
weights_init_kaimingNormal(self)
elif init == 'kaimingUniform':
weights_init_kaimingUniform(self)
elif init == 'xavierNormal':
weights_init_xavierNormal(self)
elif init == 'xavierUniform':
weights_init_xavierUniform(self)
else:
raise NotImplementedError("Invalid type of initialization!")
def __init__(self, config, dim):
super(EncoderStyle, self).__init__()
channels = config.enc_cl_channels
channels[0] = config.style_channel_3d if dim == "3d" else config.style_channel_2d
kernel_size = config.enc_cl_kernel_size
stride = config.enc_cl_stride
self.global_pool = F.max_pool1d
layers = []
n_convs = config.enc_cl_down_n
for i in range(n_convs):
layers += ConvBlock(kernel_size, channels[i], channels[i + 1],
stride=stride, norm='none', acti='lrelu')
self.conv_model = nn.Sequential(*layers)
self.channels = channels
def __init__(self, config):
super(EncoderContent, self).__init__()
channels = config.enc_co_channels
kernel_size = config.enc_co_kernel_size
stride = config.enc_co_stride
layers = []
n_convs = config.enc_co_down_n
n_resblk = config.enc_co_resblks
acti = 'lrelu'
assert n_convs + 1 == len(channels)
for i in range(n_convs):
layers += ConvBlock(kernel_size, channels[i], channels[i + 1],
stride=stride, norm='in', acti=acti)
for i in range(n_resblk):
layers.append(ResBlock(kernel_size, channels[-1], stride=1,
pad_type='reflect', norm='in', acti=acti))
self.conv_model = nn.Sequential(*layers)
self.channels = channels
import tensorflow as tf
import os
import sys
import pathlib
current_dir = pathlib.Path(__file__).resolve().parent
sys.path.append(os.path.join(current_dir, '../'))
from blocks import ConvBlock
conv = ConvBlock(32,
kernel_size=(3, 3),
activation_='lrelu',
dilation_rate=(1, 1),
sampling='up',
normalization='batch',
spectral_norm=True)
x = tf.random.normal(shape=(10, 32, 32, 3))
y = conv(x, training=True)
y_ = conv(x, training=False)
# print(y)
def __init__(self,
im_size,
num_classes,
attention=True,
normalize_attn=True,
init='xavierUniform',
_base_features=64,
dropout=0.0):
super(AttnVGG_before, self).__init__()
#self.base_features = 64
self.base_features = _base_features
self.dropout = nn.Dropout(p=dropout)
self.attention = attention
# conv blocks
self.conv_block1 = ConvBlock(1, self.base_features, 2, dropout=dropout)
self.conv_block2 = ConvBlock(self.base_features,
self.base_features * 2,
2,
dropout=dropout)
self.conv_block3 = ConvBlock(self.base_features * 2,
self.base_features * 4,
3,
dropout=dropout)
self.conv_block4 = ConvBlock(self.base_features * 4,
self.base_features * 8,
3,
dropout=dropout)
self.conv_block5 = ConvBlock(self.base_features * 8,
self.base_features * 8,
3,
dropout=dropout)
self.conv_block6 = ConvBlock(self.base_features * 8,
self.base_features * 8,
2,
pool=True,
dropout=dropout)
self.dense = nn.Conv2d(in_channels=self.base_features * 8,
out_channels=self.base_features * 8,
kernel_size=int(im_size / 32),
padding=0,
bias=True)
# Projectors & Compatibility functions
if self.attention:
self.projector = ProjectorBlock(self.base_features * 4,
self.base_features * 8)
self.attn1 = LinearAttentionBlock(in_features=self.base_features *
8,
normalize_attn=normalize_attn)
self.attn2 = LinearAttentionBlock(in_features=self.base_features *
8,
normalize_attn=normalize_attn)
self.attn3 = LinearAttentionBlock(in_features=self.base_features *
8,
normalize_attn=normalize_attn)
# final classification layer
if self.attention:
self.classify = nn.Linear(in_features=(self.base_features * 8) * 3,
out_features=num_classes,
bias=True)
else:
self.classify = nn.Linear(in_features=self.base_features * 8,
out_features=num_classes,
bias=True)
# initialize
if init == 'kaimingNormal':
weights_init_kaimingNormal(self)
elif init == 'kaimingUniform':
weights_init_kaimingUniform(self)
elif init == 'xavierNormal':
weights_init_xavierNormal(self)
elif init == 'xavierUniform':
weights_init_xavierUniform(self)
else:
raise NotImplementedError("Invalid type of initialization!")
def __init__(self, in_ch):
super().__init__()
self.block1 = ConvBlock(in_ch,
96,
kernel_size=11,
sampling='stride',
normalization='batch',
stride=4,
padding='valid')
self.block2 = torch.nn.Sequential(
ConvBlock(96, 96, kernel_size=1, normalization='batch'),
ConvBlock(96, 96, kernel_size=1, normalization='batch'),
ConvBlock(96,
96,
kernel_size=3,
normalization='batch',
sampling='stride',
padding='valid'))
self.block3 = torch.nn.Sequential(
ConvBlock(96, 96, kernel_size=1, normalization='batch'),
ConvBlock(96, 96, kernel_size=1, normalization='batch'),
ConvBlock(96,
256,
kernel_size=5,
normalization='batch',
sampling='stride',
padding='valid'))
self.block4 = torch.nn.Sequential(
ConvBlock(256, 512, kernel_size=1, normalization='batch'),
ConvBlock(512, 1024, kernel_size=1, normalization='batch'),
ConvBlock(1024, 1024, kernel_size=1, normalization='batch'),
ConvBlock(1024, 1, kernel_size=1, activation=torch.nn.Sigmoid()))
def __init__(self, out_ch):
super().__init__()
self.conv = ConvBlock(3,
out_ch,
sampling='same',
normalization='instance')
def __init__(self, in_ch, out_ch):
super().__init__()
self.conv = ConvBlock(in_ch, out_ch)
def __init__(self, in_ch, out_ch):
super().__init__()
self.conv = ConvBlock(in_ch, out_ch, sampling='upsampling')
