def forward(self, input, label):
network = torch.cat([input, label], dim=1)
network = self.gc_full(network)
network = ReLU(inplace=True)(network)
network = network.view(-1, 512, 1, 1)
network = ReLU()(self.bn2(self.dconv1(network)))
network = ReLU()(self.bn3(self.dconv2(network)))
network = ReLU()(self.bn4(self.dconv3(network)))
network = Sigmoid()(self.dconv4(network))
return network
def forward(self, inputs):
"""Return the output of a forward pass of the ToyNetwork
:param inputs: batch of input images, of shape
(batch_size, n_channels, height, width)
:type inputs: torch.Tensor
:return: outputs of a ToyNetwork model, of shape
(batch_size, n_classes)
:rtype: torch.Tensor
"""
layer = self.conv(inputs)
layer = ReLU()(layer)
layer = layer.view(layer.size(0), -1)
outputs = self.linear(layer)
return outputs
def forward(self, inputs):
"""Return the output of a forward pass of AlexNet
:param inputs: batch of input images, of shape
(batch_size, height, width, n_channels)
:type inputs: torch.Tensor
:return: outputs of an AlexNet model, of shape
(batch_size, n_classes)
:rtype: torch.Tensor
"""
# === convolutional block 1 === #
layer = self.conv1(inputs)
layer = ReLU()(layer)
layer = MaxPool2d(kernel_size=(3, 3), stride=(2, 2))(layer)
# === convolutional block 2 === #
layer = self.conv2(layer)
layer = ReLU()(layer)
layer = MaxPool2d(kernel_size=(3, 3), stride=(2, 2))(layer)
# === convolutional blocks 3, 4, 5 === #
layer = self.conv3(layer)
layer = ReLU()(layer)
layer = self.conv4(layer)
layer = ReLU()(layer)
layer = self.conv5(layer)
layer = ReLU()(layer)
layer = MaxPool2d(kernel_size=(3, 3), stride=(2, 2))(layer)
# === dense layers (2) === #
layer = layer.view(layer.size(0), -1)
layer = self.linear1(layer)
layer = ReLU()(layer)
layer = Dropout(0.5)(layer)
layer = self.linear2(layer)
layer = ReLU()(layer)
layer = Dropout(0.5)(layer)
# === output layer === #
outputs = self.linear3(layer)
return outputs
def forward(self, inputs):
"""Return the outputs from a forward pass of the network
:param inputs: batch of input images, of shape
(BATCH_SIZE, n_channels, height, width)
:type inputs: torch.Tensor
:return: outputs of SimpleCNN, of shape (BATCH_SIZE, NUM_CLASSES)
:rtype: torch.Tensor
"""
layer = self.conv1(inputs)
layer = ReLU()(layer)
layer = self.conv2(layer)
layer = ReLU()(layer)
layer = MaxPool2d(kernel_size=(2, 2))(layer)
layer = Dropout(0.25)(layer)
layer = layer.view(layer.size(0), -1)
layer = self.linear1(layer)
layer = Dropout(0.5)(layer)
outputs = self.linear2(layer)
return outputs
