How can I concatenate an additional input data in Alexnet with the output of the last dropout layer using Pytorch implementation?

Question

Here is the implementation architecture

class AlexNet(nn.Module):
def __init__(self, num_classes=10):
    super(AlexNet, self).__init__()
    #1
    self.features= nn.Sequential(
    nn.Conv2d(3, 96, kernel_size=11, stride=4, padding=0),
    nn.ReLU(inplace=True),
    nn.MaxPool2d(kernel_size=3, stride=2),
    #2
    nn.Conv2d(96, 256, kernel_size=5, stride=1, padding=2),
    nn.ReLU(inplace=True),
    nn.MaxPool2d(kernel_size=3, stride=2),
    #3
    nn.Conv2d(256, 384, kernel_size=3, stride=1, padding=1),
    nn.ReLU(inplace=True),
    #4
    nn.Conv2d(384, 384, kernel_size=3, stride=1, padding=1),
    nn.ReLU(inplace=True),
    #5
    nn.Conv2d(384, 256, kernel_size=5, stride=1, padding=2),
    nn.ReLU(inplace=True),
    nn.MaxPool2d(kernel_size=3, stride=2),
    )
    self.avgpool= nn.AvgPool2d(6)
    self.classifier= nn.Sequential(
        nn.Dropout(), nn.Linear(256*6*6, 4096), #128*2*2, 1024
    nn.ReLU(inplace=True), nn.Dropout(),torch.cat((nn.Dropout(),PIs_features)),
    nn.Linear(4096, num_classes))
    
def forward(self, x):
    x= self.features(x)
    x=x.view(x.size(0), 256*6*6)
    x= self.classifier(x)
    return x

So I wanna implement say 'y' input data with the output of the last dropout layer 'nn.dropout()' in the self.classifier.

Thanks in advance.

Answer 1

You can do so in the forward definition by simply calling torch.cat((x, y), 1) to concatenate the two feature vectors together.

class AlexNet(nn.Module):
  def __init__(self, num_classes=10):
      super().__init__()
        #1
      self.features= nn.Sequential(
        nn.Conv2d(3, 96, kernel_size=11, stride=4, padding=0),
        nn.ReLU(inplace=True),
        nn.MaxPool2d(kernel_size=3, stride=2),
        #2
        nn.Conv2d(96, 256, kernel_size=5, stride=1, padding=2),
        nn.ReLU(inplace=True),
        nn.MaxPool2d(kernel_size=3, stride=2),
        #3
        nn.Conv2d(256, 384, kernel_size=3, stride=1, padding=1),
        nn.ReLU(inplace=True),
        #4
        nn.Conv2d(384, 384, kernel_size=3, stride=1, padding=1),
        nn.ReLU(inplace=True),
        #5
        nn.Conv2d(384, 256, kernel_size=5, stride=1, padding=2),
        nn.ReLU(inplace=True),
        nn.MaxPool2d(kernel_size=3, stride=2))
      
      self.avgpool= nn.AvgPool2d(6)
      self.classifier= nn.Sequential(
          nn.Dropout(), 
          nn.LazyLinear(4096),
          nn.ReLU(inplace=True), 
          nn.Dropout())
      
      self.fc = nn.LazyLinear(num_classes)
      
  def forward(self, x, y):
      x = self.features(x)
      x = self.avgpool(x)
      x = x.flatten(1)
      x = torch.cat((x, y), 1)
      x = self.classifier(x)
      return x

Additionally, I have replaced the fully connected nn.Linear layers with LazyLayer. But you can replace them with fixed neurons if you prefer.