def encoder(self, x):
h1= F.elu(self.e1(x))
h3= F.elu( self.e3(self.e2(h1)) )
h5= F.elu(self.e5(self.e4(h3)))
h6=h5.view(-1, 256)
h61= self.e61(h6)
h62= self.e62(h6)
return h61, h62
def encode(self, x):
x = F.elu(self.conv1(x))
x = F.avg_pool2d(x, kernel_size=2, stride=2)
x = F.elu(self.conv2(x))
x = F.avg_pool2d(x, kernel_size=2, stride=2)
x = F.elu(self.conv3(x))
x = x.reshape(-1, 256)
x = self.fc_log_sigma(x)
return x[:, :latent_dimension], x[:, latent_dimension:]
def forward(self, x):
out = self.conv1(x)
out = F.elu(self.conv1_bn(out))
out = self.conv2(out)
out = F.elu(self.conv2_bn(out))
out = self.conv3(out)
out = F.elu(self.conv3_bn(out))
out = out.view(out.size()[0], -1)
out = F.elu(self.fc1(out))
out = F.softmax(self.fc2(out), dim=-1)
return out
def forward(self, x):
out = self.conv1(x)
conv1 = F.elu(self.conv1_bn(out))
out = self.conv2(conv1)
conv2 = F.elu(self.conv2_bn(out))
out = self.conv3(conv2)
conv3 = F.elu(self.conv3_bn(out))
out = conv3.view(conv3.size()[0], -1)
out = F.elu(self.fc1(out))
out = F.softmax(self.fc2(out), dim=-1)
return out, conv1.detach(), conv2.detach(), conv3.detach()
def decode(self, z):
x = F.elu(self.fc_decoder(z))
x = x.reshape(z.shape[0], 256, 1, 1)
x = F.elu(self.conv4(x))
x = self.upsampling(x)
x = F.elu(self.conv5(x))
x = self.upsampling(x)
x = F.elu(self.conv6(x))
x = self.conv7(x)
return torch.sigmoid(x)
def decoder(self, z):
h1= F.elu(self.d1(z))
h1= h1.view(h1.size(0), 256, 1, 1)
h2= F.elu(self.conv1(h1))
h3=self.upsample(h2)
h4= F.elu(self.conv2(h3))
h5= self.upsample(h4)
h6= F.elu(self.conv3(h5))
h7= self.conv4(h6)
h10= F.sigmoid(h7)
return h10
def lovasz_loss_flat(logits, labels, error_func):
"""
Binary Lovasz hinge loss
logits: [P] Variable, logits at each prediction (between -\infty and +\infty)
labels: [P] Tensor, binary ground truth labels (0 or 1)
ignore: label to ignore
"""
if len(labels) == 0:
# only void pixels, the gradients should be 0
return logits.sum() * 0.
errors = error_func(logits, labels)
errors_sorted, perm = torch.sort(errors, dim=0, descending=True)
perm = perm.data
gt_sorted = labels[perm]
grad = lovasz_grad(gt_sorted)
# loss = torch.dot(F.relu(errors_sorted), Variable(grad))
loss = torch.dot(F.elu(errors_sorted) + 1, Variable(grad))
return loss
