def forward(self, x):
x = self.layer00(x)
x = x[:, 64:70] # Extract at hardcoded z_gaze indices
x = F.selu_(x)
x = self.layer01(x)
x = F.selu_(x)
x = self.layer02(x)
x = F.normalize(x, dim=-1) # Normalize
return x
def forward(self, x):
x = F.selu_(self.bn1(self.Conv_1(x)))
x = F.selu_(self.bn2(self.Conv_2(x)))
x = F.selu_(self.bn3(self.Conv_3(x)))
x = x.view((x.shape[0], -1))
x = F.selu_(self.bn4(self.l4(x)))
x = F.selu_(self.bn5(self.l5(x)))
x = self.l6(x)
if (not self.wass):
x = torch.sigmoid(x)
return x
def forward(self, x):
x = F.selu_(self.bn1(self.Conv_1(x)))
x = F.selu_(self.bn2(self.Conv_2(x)))
x = F.selu_(self.bn3(self.Conv_3(x)))
x = x.view((x.shape[0], -1))
x = F.selu_(self.bn4(self.l4(x)))
z_img = self.l5(x)
z = z_img.view(z_img.shape[0], -1)
zn = z[:, 0:self.latent_dim]
zc_logits = z[:, self.latent_dim:]
zc = F.softmax(zc_logits, dim=1)
return zn, zc, zc_logits
def forward(self, x):
x = self.rnb_1(x)
x = self.rnb_2(x)
x = self.rnb_3(x)
x = self.rnb_4(x)
x = self.rnb_5(x)
x = self.rnb_6(x)
x = self.rnb_7(x)
x = self.rnb_8(x)
x = self.rnb_9(x)
x = self.rnb_10(x)
x = self.rnb_11(x)
x = self.rnb_12(x)
x = self.rnb_13(x)
x = self.rnb_14(x)
x = self.rnb_15(x)
x = self.rnb_16(x)
x = self.rnb_17(x)
x = self.rnb_18(x)
x = x.view((x.shape[0], -1))
x = F.selu_(self.bn_19(self.linear_19(x)))
z_img = self.linear_20(x)
z = z_img.view(z_img.shape[0], -1)
zn = z[:, 0:self.latent_dim]
zc_logits = z[:, self.latent_dim:]
zc = F.softmax(zc_logits, dim=1)
return zn, zc, zc_logits
def forward(self, zn, zc):
z = torch.cat((zn, zc), 1)
x = F.selu_(self.bn1(self.l1(z)))
x = F.selu_(self.bn2(self.l2(x)))
x = x.view([x.shape[0]] + self.ishape)
x = self.rnb_1(x)
x = self.rnb_2(x)
x = self.rnb_3(x) # 64x4x4
x = F.interpolate(x, scale_factor=2, mode='bicubic') # 64x8x8
x = self.rnb_4(x)
x = self.rnb_5(x)
x = self.rnb_6(x) # 32x8x8
x = F.interpolate(x, scale_factor=2, mode='bicubic') # 32x16x16
x = self.rnb_7(x)
x = self.rnb_8(x)
x = self.rnb_9(x) # 16x16x16
x = F.interpolate(x, scale_factor=2, mode='bicubic') # 16x32x32
x = self.rnb_10(x)
x = self.rnb_11(x)
x = self.rnb_12(x) # 8x32x32
x = F.interpolate(x, scale_factor=2, mode='bicubic') # 8x64x64
x = self.rnb_13(x)
x = self.rnb_14(x)
x = self.rnb_15(x) # 4x64x64
x = F.interpolate(x, scale_factor=2, mode='bicubic') # 4x128x128
x = self.rnb_16(x)
x = self.rnb_17(x)
x = self.rnb_18(x) # 2x128x128
x = F.interpolate(x, scale_factor=2, mode='bicubic') # 2x256x256
x = self.rnb_19(x)
x = self.rnb_20(x)
x = self.rnb_21(x) # 2x256x256
x_gen = torch.sigmoid(self.Conv_gen_final(x)) # 1x256x256
return x_gen
def forward(self, x):
x = self.rnb_1(x)
x = self.rnb_3(x)
x = self.rnb_4(x)
x = self.rnb_6(x)
x = self.rnb_7(x)
x = self.rnb_9(x)
x = self.rnb_10(x)
x = self.rnb_12(x)
x = self.rnb_13(x)
x = self.rnb_15(x)
x = self.rnb_16(x)
x = self.rnb_18(x)
x = x.view((x.shape[0], -1))
x = F.selu_(self.bn_19(self.linear_19(x)))
x = F.selu_(self.bn_20(self.linear_20(x)))
x = self.linear_21(x)
if (not self.wass):
x = torch.sigmoid(x)
return x
def forward(self, zn, zc):
z = torch.cat((zn, zc), 1)
x = F.selu_(self.bn1(self.l1(z)))
x = F.selu_(self.bn2(self.l2(x)))
x = x.view([x.shape[0]] + self.ishape)
x = F.selu_(self.bn3(self.Conv_1(x)))
x = F.interpolate(x, scale_factor=2, mode='bicubic')
x = F.selu_(self.bn4(self.Conv_2(x)))
x = F.interpolate(x, scale_factor=2, mode='bicubic')
x = F.selu_(self.bn5(self.Conv_3(x)))
x = F.interpolate(x, scale_factor=2, mode='bicubic')
x = F.selu_(self.bn6(self.Conv_4(x)))
x = F.interpolate(x, size=64, mode='bicubic')
x = F.selu_(self.bn7(self.Conv_5(x)))
x_gen = torch.sigmoid(self.Conv_6(x))
return x_gen
def forward(self, x):
for name, layer in self.layers[:-1]:
x = layer(x)
if self.activation_type == 'relu':
x = F.relu_(x)
elif self.activation_type == 'leaky_relu':
x = F.leaky_relu_(x)
elif self.activation_type == 'elu':
x = F.elu_(x)
elif self.activation_type == 'selu':
x = F.selu_(x)
elif self.activation_type == 'tanh':
x = torch.tanh_(x)
elif self.activation_type == 'sigmoid':
x = torch.sigmoid_(x)
elif self.activation_type == 'none':
pass
else:
raise ValueError('Unknown activation function "%s"' % self.activation_type)
x = self.layers[-1][1](x) # No activation on output of last layer
x = F.normalize(x, dim=-1) # Normalize
return x