mirror of
https://github.com/iperov/DeepFaceLab.git
synced 2025-08-14 10:46:59 -07:00
SAEHD:
Maximum resolution is increased to 640. ‘hd’ archi is removed. ‘hd’ was experimental archi created to remove subpixel shake, but ‘lr_dropout’ and ‘disable random warping’ do that better. ‘uhd’ is renamed to ‘-u’ dfuhd and liaeuhd will be automatically renamed to df-u and liae-u in existing models. Added new experimental archi (key -d) which doubles the resolution using the same computation cost. It is mean same configs will be x2 faster, or for example you can set 448 resolution and it will train as 224. Strongly recommended not to train from scratch and use pretrained models. New archi naming: 'df' keeps more identity-preserved face. 'liae' can fix overly different face shapes. '-u' increased likeness of the face. '-d' (experimental) doubling the resolution using the same computation cost Examples: df, liae, df-d, df-ud, liae-ud, ... Improved GAN training (GAN_power option). It was used for dst model, but actually we don’t need it for dst. Instead, a second src GAN model with x2 smaller patch size was added, so the overall quality for hi-res models should be higher. Added option ‘Uniform yaw distribution of samples (y/n)’: Helps to fix blurry side faces due to small amount of them in the faceset. Quick96: Now based on df-ud archi and 20% faster. XSeg trainer: Improved sample generator. Now it randomly adds the background from other samples. Result is reduced chance of random mask noise on the area outside the face. Now you can specify ‘batch_size’ in range 2-16. Reduced size of samples with applied XSeg mask. Thus size of packed samples with applied xseg mask is also reduced.
This commit is contained in:
Colombo
parent
9fd3a9ff8d
commit
0c2e1c3944
14 changed files with 513 additions and 572 deletions
|
|
@ -1,54 +1,46 @@
|
|||
from core.leras import nn
|
||||
tf = nn.tf
|
||||
|
||||
class DeepFakeArchi(nn.ArchiBase):
|
||||
class DeepFakeArchi(nn.ArchiBase):
|
||||
"""
|
||||
resolution
|
||||
|
||||
|
||||
mod None - default
|
||||
'uhd'
|
||||
'quick'
|
||||
"""
|
||||
def __init__(self, resolution, mod=None):
|
||||
def __init__(self, resolution, mod=None, opts=None):
|
||||
super().__init__()
|
||||
|
||||
|
||||
if opts is None:
|
||||
opts = ''
|
||||
|
||||
if mod is None:
|
||||
class Downscale(nn.ModelBase):
|
||||
def __init__(self, in_ch, out_ch, kernel_size=5, dilations=1, subpixel=True, use_activator=True, *kwargs ):
|
||||
def __init__(self, in_ch, out_ch, kernel_size=5, *kwargs ):
|
||||
self.in_ch = in_ch
|
||||
self.out_ch = out_ch
|
||||
self.kernel_size = kernel_size
|
||||
self.dilations = dilations
|
||||
self.subpixel = subpixel
|
||||
self.use_activator = use_activator
|
||||
super().__init__(*kwargs)
|
||||
|
||||
def on_build(self, *args, **kwargs ):
|
||||
self.conv1 = nn.Conv2D( self.in_ch,
|
||||
self.out_ch // (4 if self.subpixel else 1),
|
||||
kernel_size=self.kernel_size,
|
||||
strides=1 if self.subpixel else 2,
|
||||
padding='SAME', dilations=self.dilations)
|
||||
self.conv1 = nn.Conv2D( self.in_ch, self.out_ch, kernel_size=self.kernel_size, strides=2, padding='SAME')
|
||||
|
||||
def forward(self, x):
|
||||
x = self.conv1(x)
|
||||
if self.subpixel:
|
||||
x = nn.space_to_depth(x, 2)
|
||||
if self.use_activator:
|
||||
x = tf.nn.leaky_relu(x, 0.1)
|
||||
x = tf.nn.leaky_relu(x, 0.1)
|
||||
return x
|
||||
|
||||
def get_out_ch(self):
|
||||
return (self.out_ch // 4) * 4 if self.subpixel else self.out_ch
|
||||
return self.out_ch
|
||||
|
||||
class DownscaleBlock(nn.ModelBase):
|
||||
def on_build(self, in_ch, ch, n_downscales, kernel_size, dilations=1, subpixel=True):
|
||||
def on_build(self, in_ch, ch, n_downscales, kernel_size):
|
||||
self.downs = []
|
||||
|
||||
last_ch = in_ch
|
||||
for i in range(n_downscales):
|
||||
cur_ch = ch*( min(2**i, 8) )
|
||||
self.downs.append ( Downscale(last_ch, cur_ch, kernel_size=kernel_size, dilations=dilations, subpixel=subpixel) )
|
||||
self.downs.append ( Downscale(last_ch, cur_ch, kernel_size=kernel_size) )
|
||||
last_ch = self.downs[-1].get_out_ch()
|
||||
|
||||
def forward(self, inp):
|
||||
|
|
@ -79,346 +71,33 @@ class DeepFakeArchi(nn.ArchiBase):
|
|||
x = tf.nn.leaky_relu(inp + x, 0.2)
|
||||
return x
|
||||
|
||||
class UpdownResidualBlock(nn.ModelBase):
|
||||
def on_build(self, ch, inner_ch, kernel_size=3 ):
|
||||
self.up = Upscale (ch, inner_ch, kernel_size=kernel_size)
|
||||
self.res = ResidualBlock (inner_ch, kernel_size=kernel_size)
|
||||
self.down = Downscale (inner_ch, ch, kernel_size=kernel_size, use_activator=False)
|
||||
|
||||
def forward(self, inp):
|
||||
x = self.up(inp)
|
||||
x = upx = self.res(x)
|
||||
x = self.down(x)
|
||||
x = x + inp
|
||||
x = tf.nn.leaky_relu(x, 0.2)
|
||||
return x, upx
|
||||
|
||||
class Encoder(nn.ModelBase):
|
||||
def on_build(self, in_ch, e_ch, is_hd):
|
||||
self.is_hd=is_hd
|
||||
if self.is_hd:
|
||||
self.down1 = DownscaleBlock(in_ch, e_ch*2, n_downscales=4, kernel_size=3, dilations=1)
|
||||
self.down2 = DownscaleBlock(in_ch, e_ch*2, n_downscales=4, kernel_size=5, dilations=1)
|
||||
self.down3 = DownscaleBlock(in_ch, e_ch//2, n_downscales=4, kernel_size=5, dilations=2)
|
||||
self.down4 = DownscaleBlock(in_ch, e_ch//2, n_downscales=4, kernel_size=7, dilations=2)
|
||||
else:
|
||||
self.down1 = DownscaleBlock(in_ch, e_ch, n_downscales=4, kernel_size=5, dilations=1, subpixel=False)
|
||||
def on_build(self, in_ch, e_ch):
|
||||
self.down1 = DownscaleBlock(in_ch, e_ch, n_downscales=4, kernel_size=5)
|
||||
|
||||
def forward(self, inp):
|
||||
if self.is_hd:
|
||||
x = tf.concat([ nn.flatten(self.down1(inp)),
|
||||
nn.flatten(self.down2(inp)),
|
||||
nn.flatten(self.down3(inp)),
|
||||
nn.flatten(self.down4(inp)) ], -1 )
|
||||
else:
|
||||
x = nn.flatten(self.down1(inp))
|
||||
return x
|
||||
|
||||
lowest_dense_res = resolution // 16
|
||||
return nn.flatten(self.down1(inp))
|
||||
|
||||
lowest_dense_res = resolution // (32 if 'd' in opts else 16)
|
||||
|
||||
class Inter(nn.ModelBase):
|
||||
def __init__(self, in_ch, ae_ch, ae_out_ch, is_hd=False, **kwargs):
|
||||
def __init__(self, in_ch, ae_ch, ae_out_ch, **kwargs):
|
||||
self.in_ch, self.ae_ch, self.ae_out_ch = in_ch, ae_ch, ae_out_ch
|
||||
super().__init__(**kwargs)
|
||||
|
||||
def on_build(self):
|
||||
in_ch, ae_ch, ae_out_ch = self.in_ch, self.ae_ch, self.ae_out_ch
|
||||
if 'u' in opts:
|
||||
self.dense_norm = nn.DenseNorm()
|
||||
|
||||
self.dense1 = nn.Dense( in_ch, ae_ch )
|
||||
self.dense2 = nn.Dense( ae_ch, lowest_dense_res * lowest_dense_res * ae_out_ch )
|
||||
self.upscale1 = Upscale(ae_out_ch, ae_out_ch)
|
||||
|
||||
def forward(self, inp):
|
||||
x = self.dense1(inp)
|
||||
x = self.dense2(x)
|
||||
x = nn.reshape_4D (x, lowest_dense_res, lowest_dense_res, self.ae_out_ch)
|
||||
x = self.upscale1(x)
|
||||
return x
|
||||
|
||||
@staticmethod
|
||||
def get_code_res():
|
||||
return lowest_dense_res
|
||||
|
||||
def get_out_ch(self):
|
||||
return self.ae_out_ch
|
||||
|
||||
class Decoder(nn.ModelBase):
|
||||
def on_build(self, in_ch, d_ch, d_mask_ch, is_hd ):
|
||||
self.is_hd = is_hd
|
||||
|
||||
self.upscale0 = Upscale(in_ch, d_ch*8, kernel_size=3)
|
||||
self.upscale1 = Upscale(d_ch*8, d_ch*4, kernel_size=3)
|
||||
self.upscale2 = Upscale(d_ch*4, d_ch*2, kernel_size=3)
|
||||
|
||||
if is_hd:
|
||||
self.res0 = UpdownResidualBlock(in_ch, d_ch*8, kernel_size=3)
|
||||
self.res1 = UpdownResidualBlock(d_ch*8, d_ch*4, kernel_size=3)
|
||||
self.res2 = UpdownResidualBlock(d_ch*4, d_ch*2, kernel_size=3)
|
||||
self.res3 = UpdownResidualBlock(d_ch*2, d_ch, kernel_size=3)
|
||||
else:
|
||||
self.res0 = ResidualBlock(d_ch*8, kernel_size=3)
|
||||
self.res1 = ResidualBlock(d_ch*4, kernel_size=3)
|
||||
self.res2 = ResidualBlock(d_ch*2, kernel_size=3)
|
||||
|
||||
self.out_conv = nn.Conv2D( d_ch*2, 3, kernel_size=1, padding='SAME')
|
||||
|
||||
self.upscalem0 = Upscale(in_ch, d_mask_ch*8, kernel_size=3)
|
||||
self.upscalem1 = Upscale(d_mask_ch*8, d_mask_ch*4, kernel_size=3)
|
||||
self.upscalem2 = Upscale(d_mask_ch*4, d_mask_ch*2, kernel_size=3)
|
||||
self.out_convm = nn.Conv2D( d_mask_ch*2, 1, kernel_size=1, padding='SAME')
|
||||
|
||||
def forward(self, inp):
|
||||
z = inp
|
||||
|
||||
if self.is_hd:
|
||||
x, upx = self.res0(z)
|
||||
x = self.upscale0(x)
|
||||
x = tf.nn.leaky_relu(x + upx, 0.2)
|
||||
x, upx = self.res1(x)
|
||||
|
||||
x = self.upscale1(x)
|
||||
x = tf.nn.leaky_relu(x + upx, 0.2)
|
||||
x, upx = self.res2(x)
|
||||
|
||||
x = self.upscale2(x)
|
||||
x = tf.nn.leaky_relu(x + upx, 0.2)
|
||||
x, upx = self.res3(x)
|
||||
else:
|
||||
x = self.upscale0(z)
|
||||
x = self.res0(x)
|
||||
x = self.upscale1(x)
|
||||
x = self.res1(x)
|
||||
x = self.upscale2(x)
|
||||
x = self.res2(x)
|
||||
|
||||
m = self.upscalem0(z)
|
||||
m = self.upscalem1(m)
|
||||
m = self.upscalem2(m)
|
||||
|
||||
return tf.nn.sigmoid(self.out_conv(x)), \
|
||||
tf.nn.sigmoid(self.out_convm(m))
|
||||
|
||||
elif mod == 'quick':
|
||||
class Downscale(nn.ModelBase):
|
||||
def __init__(self, in_ch, out_ch, kernel_size=5, dilations=1, subpixel=True, use_activator=True, *kwargs ):
|
||||
self.in_ch = in_ch
|
||||
self.out_ch = out_ch
|
||||
self.kernel_size = kernel_size
|
||||
self.dilations = dilations
|
||||
self.subpixel = subpixel
|
||||
self.use_activator = use_activator
|
||||
super().__init__(*kwargs)
|
||||
|
||||
def on_build(self, *args, **kwargs ):
|
||||
self.conv1 = nn.Conv2D( self.in_ch,
|
||||
self.out_ch // (4 if self.subpixel else 1),
|
||||
kernel_size=self.kernel_size,
|
||||
strides=1 if self.subpixel else 2,
|
||||
padding='SAME', dilations=self.dilations )
|
||||
|
||||
def forward(self, x):
|
||||
x = self.conv1(x)
|
||||
|
||||
if self.subpixel:
|
||||
x = nn.space_to_depth(x, 2)
|
||||
|
||||
if self.use_activator:
|
||||
x = nn.gelu(x)
|
||||
return x
|
||||
|
||||
def get_out_ch(self):
|
||||
return (self.out_ch // 4) * 4 if self.subpixel else self.out_ch
|
||||
|
||||
class DownscaleBlock(nn.ModelBase):
|
||||
def on_build(self, in_ch, ch, n_downscales, kernel_size, dilations=1, subpixel=True):
|
||||
self.downs = []
|
||||
|
||||
last_ch = in_ch
|
||||
for i in range(n_downscales):
|
||||
cur_ch = ch*( min(2**i, 8) )
|
||||
self.downs.append ( Downscale(last_ch, cur_ch, kernel_size=kernel_size, dilations=dilations, subpixel=subpixel) )
|
||||
last_ch = self.downs[-1].get_out_ch()
|
||||
|
||||
def forward(self, inp):
|
||||
x = inp
|
||||
for down in self.downs:
|
||||
x = down(x)
|
||||
return x
|
||||
|
||||
class Upscale(nn.ModelBase):
|
||||
def on_build(self, in_ch, out_ch, kernel_size=3 ):
|
||||
self.conv1 = nn.Conv2D( in_ch, out_ch*4, kernel_size=kernel_size, padding='SAME')
|
||||
|
||||
def forward(self, x):
|
||||
x = self.conv1(x)
|
||||
x = nn.gelu(x)
|
||||
x = nn.depth_to_space(x, 2)
|
||||
return x
|
||||
|
||||
class ResidualBlock(nn.ModelBase):
|
||||
def on_build(self, ch, kernel_size=3 ):
|
||||
self.conv1 = nn.Conv2D( ch, ch, kernel_size=kernel_size, padding='SAME')
|
||||
self.conv2 = nn.Conv2D( ch, ch, kernel_size=kernel_size, padding='SAME')
|
||||
|
||||
def forward(self, inp):
|
||||
x = self.conv1(inp)
|
||||
x = nn.gelu(x)
|
||||
x = self.conv2(x)
|
||||
x = inp + x
|
||||
x = nn.gelu(x)
|
||||
return x
|
||||
|
||||
class Encoder(nn.ModelBase):
|
||||
def on_build(self, in_ch, e_ch):
|
||||
self.down1 = DownscaleBlock(in_ch, e_ch, n_downscales=4, kernel_size=5)
|
||||
def forward(self, inp):
|
||||
return nn.flatten(self.down1(inp))
|
||||
|
||||
lowest_dense_res = resolution // 16
|
||||
|
||||
class Inter(nn.ModelBase):
|
||||
def __init__(self, in_ch, ae_ch, ae_out_ch, d_ch, **kwargs):
|
||||
self.in_ch, self.ae_ch, self.ae_out_ch, self.d_ch = in_ch, ae_ch, ae_out_ch, d_ch
|
||||
super().__init__(**kwargs)
|
||||
|
||||
def on_build(self):
|
||||
in_ch, ae_ch, ae_out_ch, d_ch = self.in_ch, self.ae_ch, self.ae_out_ch, self.d_ch
|
||||
|
||||
self.dense1 = nn.Dense( in_ch, ae_ch, kernel_initializer=tf.initializers.orthogonal )
|
||||
self.dense2 = nn.Dense( ae_ch, lowest_dense_res * lowest_dense_res * ae_out_ch, kernel_initializer=tf.initializers.orthogonal )
|
||||
self.upscale1 = Upscale(ae_out_ch, d_ch*8)
|
||||
self.res1 = ResidualBlock(d_ch*8)
|
||||
|
||||
def forward(self, inp):
|
||||
x = self.dense1(inp)
|
||||
x = self.dense2(x)
|
||||
x = nn.reshape_4D (x, lowest_dense_res, lowest_dense_res, self.ae_out_ch)
|
||||
x = self.upscale1(x)
|
||||
x = self.res1(x)
|
||||
return x
|
||||
|
||||
def get_out_ch(self):
|
||||
return self.ae_out_ch
|
||||
|
||||
class Decoder(nn.ModelBase):
|
||||
def on_build(self, in_ch, d_ch):
|
||||
self.upscale1 = Upscale(in_ch, d_ch*4)
|
||||
self.res1 = ResidualBlock(d_ch*4)
|
||||
self.upscale2 = Upscale(d_ch*4, d_ch*2)
|
||||
self.res2 = ResidualBlock(d_ch*2)
|
||||
self.upscale3 = Upscale(d_ch*2, d_ch*1)
|
||||
self.res3 = ResidualBlock(d_ch*1)
|
||||
|
||||
self.upscalem1 = Upscale(in_ch, d_ch)
|
||||
self.upscalem2 = Upscale(d_ch, d_ch//2)
|
||||
self.upscalem3 = Upscale(d_ch//2, d_ch//2)
|
||||
|
||||
self.out_conv = nn.Conv2D( d_ch*1, 3, kernel_size=1, padding='SAME')
|
||||
self.out_convm = nn.Conv2D( d_ch//2, 1, kernel_size=1, padding='SAME')
|
||||
|
||||
def forward(self, inp):
|
||||
z = inp
|
||||
x = self.upscale1 (z)
|
||||
x = self.res1 (x)
|
||||
x = self.upscale2 (x)
|
||||
x = self.res2 (x)
|
||||
x = self.upscale3 (x)
|
||||
x = self.res3 (x)
|
||||
|
||||
y = self.upscalem1 (z)
|
||||
y = self.upscalem2 (y)
|
||||
y = self.upscalem3 (y)
|
||||
|
||||
return tf.nn.sigmoid(self.out_conv(x)), \
|
||||
tf.nn.sigmoid(self.out_convm(y))
|
||||
elif mod == 'uhd':
|
||||
|
||||
class Downscale(nn.ModelBase):
|
||||
def __init__(self, in_ch, out_ch, kernel_size=5, dilations=1, subpixel=True, use_activator=True, *kwargs ):
|
||||
self.in_ch = in_ch
|
||||
self.out_ch = out_ch
|
||||
self.kernel_size = kernel_size
|
||||
self.dilations = dilations
|
||||
self.subpixel = subpixel
|
||||
self.use_activator = use_activator
|
||||
super().__init__(*kwargs)
|
||||
|
||||
def on_build(self, *args, **kwargs ):
|
||||
self.conv1 = nn.Conv2D( self.in_ch,
|
||||
self.out_ch // (4 if self.subpixel else 1),
|
||||
kernel_size=self.kernel_size,
|
||||
strides=1 if self.subpixel else 2,
|
||||
padding='SAME', dilations=self.dilations)
|
||||
|
||||
def forward(self, x):
|
||||
x = self.conv1(x)
|
||||
if self.subpixel:
|
||||
x = nn.space_to_depth(x, 2)
|
||||
if self.use_activator:
|
||||
x = tf.nn.leaky_relu(x, 0.1)
|
||||
return x
|
||||
|
||||
def get_out_ch(self):
|
||||
return (self.out_ch // 4) * 4 if self.subpixel else self.out_ch
|
||||
|
||||
class DownscaleBlock(nn.ModelBase):
|
||||
def on_build(self, in_ch, ch, n_downscales, kernel_size, dilations=1, subpixel=True):
|
||||
self.downs = []
|
||||
|
||||
last_ch = in_ch
|
||||
for i in range(n_downscales):
|
||||
cur_ch = ch*( min(2**i, 8) )
|
||||
self.downs.append ( Downscale(last_ch, cur_ch, kernel_size=kernel_size, dilations=dilations, subpixel=subpixel) )
|
||||
last_ch = self.downs[-1].get_out_ch()
|
||||
|
||||
def forward(self, inp):
|
||||
x = inp
|
||||
for down in self.downs:
|
||||
x = down(x)
|
||||
return x
|
||||
|
||||
class Upscale(nn.ModelBase):
|
||||
def on_build(self, in_ch, out_ch, kernel_size=3 ):
|
||||
self.conv1 = nn.Conv2D( in_ch, out_ch*4, kernel_size=kernel_size, padding='SAME')
|
||||
|
||||
def forward(self, x):
|
||||
x = self.conv1(x)
|
||||
x = tf.nn.leaky_relu(x, 0.1)
|
||||
x = nn.depth_to_space(x, 2)
|
||||
return x
|
||||
|
||||
class ResidualBlock(nn.ModelBase):
|
||||
def on_build(self, ch, kernel_size=3 ):
|
||||
self.conv1 = nn.Conv2D( ch, ch, kernel_size=kernel_size, padding='SAME')
|
||||
self.conv2 = nn.Conv2D( ch, ch, kernel_size=kernel_size, padding='SAME')
|
||||
|
||||
def forward(self, inp):
|
||||
x = self.conv1(inp)
|
||||
x = tf.nn.leaky_relu(x, 0.2)
|
||||
x = self.conv2(x)
|
||||
x = tf.nn.leaky_relu(inp + x, 0.2)
|
||||
return x
|
||||
|
||||
class Encoder(nn.ModelBase):
|
||||
def on_build(self, in_ch, e_ch, **kwargs):
|
||||
self.down1 = DownscaleBlock(in_ch, e_ch, n_downscales=4, kernel_size=5, dilations=1, subpixel=False)
|
||||
|
||||
def forward(self, inp):
|
||||
x = nn.flatten(self.down1(inp))
|
||||
return x
|
||||
|
||||
lowest_dense_res = resolution // 16
|
||||
|
||||
class Inter(nn.ModelBase):
|
||||
def on_build(self, in_ch, ae_ch, ae_out_ch, **kwargs):
|
||||
self.ae_out_ch = ae_out_ch
|
||||
self.dense_norm = nn.DenseNorm()
|
||||
self.dense1 = nn.Dense( in_ch, ae_ch )
|
||||
self.dense2 = nn.Dense( ae_ch, lowest_dense_res * lowest_dense_res * ae_out_ch )
|
||||
self.upscale1 = Upscale(ae_out_ch, ae_out_ch)
|
||||
|
||||
def forward(self, inp):
|
||||
x = self.dense_norm(inp)
|
||||
if 'u' in opts:
|
||||
x = self.dense_norm(x)
|
||||
x = self.dense1(x)
|
||||
x = self.dense2(x)
|
||||
x = nn.reshape_4D (x, lowest_dense_res, lowest_dense_res, self.ae_out_ch)
|
||||
|
|
@ -433,8 +112,7 @@ class DeepFakeArchi(nn.ArchiBase):
|
|||
return self.ae_out_ch
|
||||
|
||||
class Decoder(nn.ModelBase):
|
||||
def on_build(self, in_ch, d_ch, d_mask_ch, **kwargs ):
|
||||
|
||||
def on_build(self, in_ch, d_ch, d_mask_ch ):
|
||||
self.upscale0 = Upscale(in_ch, d_ch*8, kernel_size=3)
|
||||
self.upscale1 = Upscale(d_ch*8, d_ch*4, kernel_size=3)
|
||||
self.upscale2 = Upscale(d_ch*4, d_ch*2, kernel_size=3)
|
||||
|
|
@ -450,6 +128,15 @@ class DeepFakeArchi(nn.ArchiBase):
|
|||
self.upscalem2 = Upscale(d_mask_ch*4, d_mask_ch*2, kernel_size=3)
|
||||
self.out_convm = nn.Conv2D( d_mask_ch*2, 1, kernel_size=1, padding='SAME')
|
||||
|
||||
if 'd' in opts:
|
||||
self.out_conv1 = nn.Conv2D( d_ch*2, 3, kernel_size=3, padding='SAME')
|
||||
self.out_conv2 = nn.Conv2D( d_ch*2, 3, kernel_size=3, padding='SAME')
|
||||
self.out_conv3 = nn.Conv2D( d_ch*2, 3, kernel_size=3, padding='SAME')
|
||||
self.upscalem3 = Upscale(d_mask_ch*2, d_mask_ch*1, kernel_size=3)
|
||||
self.out_convm = nn.Conv2D( d_mask_ch*1, 1, kernel_size=1, padding='SAME')
|
||||
else:
|
||||
self.out_convm = nn.Conv2D( d_mask_ch*2, 1, kernel_size=1, padding='SAME')
|
||||
|
||||
def forward(self, inp):
|
||||
z = inp
|
||||
|
||||
|
|
@ -460,13 +147,53 @@ class DeepFakeArchi(nn.ArchiBase):
|
|||
x = self.upscale2(x)
|
||||
x = self.res2(x)
|
||||
|
||||
|
||||
if 'd' in opts:
|
||||
x0 = tf.nn.sigmoid(self.out_conv(x))
|
||||
x0 = nn.upsample2d(x0)
|
||||
x1 = tf.nn.sigmoid(self.out_conv1(x))
|
||||
x1 = nn.upsample2d(x1)
|
||||
x2 = tf.nn.sigmoid(self.out_conv2(x))
|
||||
x2 = nn.upsample2d(x2)
|
||||
x3 = tf.nn.sigmoid(self.out_conv3(x))
|
||||
x3 = nn.upsample2d(x3)
|
||||
|
||||
if nn.data_format == "NHWC":
|
||||
tile_cfg = ( 1, resolution // 2, resolution //2, 1)
|
||||
else:
|
||||
tile_cfg = ( 1, 1, resolution // 2, resolution //2 )
|
||||
|
||||
z0 = tf.concat ( ( tf.concat ( ( tf.ones ( (1,1,1,1) ), tf.zeros ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ),
|
||||
tf.concat ( ( tf.zeros ( (1,1,1,1) ), tf.zeros ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ) ), axis=nn.conv2d_spatial_axes[0] )
|
||||
|
||||
z0 = tf.tile ( z0, tile_cfg )
|
||||
|
||||
z1 = tf.concat ( ( tf.concat ( ( tf.zeros ( (1,1,1,1) ), tf.ones ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ),
|
||||
tf.concat ( ( tf.zeros ( (1,1,1,1) ), tf.zeros ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ) ), axis=nn.conv2d_spatial_axes[0] )
|
||||
z1 = tf.tile ( z1, tile_cfg )
|
||||
|
||||
z2 = tf.concat ( ( tf.concat ( ( tf.zeros ( (1,1,1,1) ), tf.zeros ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ),
|
||||
tf.concat ( ( tf.ones ( (1,1,1,1) ), tf.zeros ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ) ), axis=nn.conv2d_spatial_axes[0] )
|
||||
z2 = tf.tile ( z2, tile_cfg )
|
||||
|
||||
z3 = tf.concat ( ( tf.concat ( ( tf.zeros ( (1,1,1,1) ), tf.zeros ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ),
|
||||
tf.concat ( ( tf.zeros ( (1,1,1,1) ), tf.ones ( (1,1,1,1) ) ), axis=nn.conv2d_spatial_axes[1] ) ), axis=nn.conv2d_spatial_axes[0] )
|
||||
z3 = tf.tile ( z3, tile_cfg )
|
||||
|
||||
x = x0*z0 + x1*z1 + x2*z2 + x3*z3
|
||||
else:
|
||||
x = tf.nn.sigmoid(self.out_conv(x))
|
||||
|
||||
|
||||
m = self.upscalem0(z)
|
||||
m = self.upscalem1(m)
|
||||
m = self.upscalem2(m)
|
||||
if 'd' in opts:
|
||||
m = self.upscalem3(m)
|
||||
m = tf.nn.sigmoid(self.out_convm(m))
|
||||
|
||||
return tf.nn.sigmoid(self.out_conv(x)), \
|
||||
tf.nn.sigmoid(self.out_convm(m))
|
||||
|
||||
return x, m
|
||||
|
||||
self.Encoder = Encoder
|
||||
self.Inter = Inter
|
||||
self.Decoder = Decoder
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue