Building a U-Net: forward method
With the encoder and decoder layers defied, you can now implement the forward() method of the U-net. The inputs have already been passed through the encoder for you. However, you need to define the last decoder block.
The goal of the decoder is to upsample the feature maps so that its output is of the same height and width as the U-Net's input image. This will allow you to obtain pixel-level semantic masks.
This exercise is part of the course
Deep Learning for Images with PyTorch
Exercise instructions
- Define the last decoder block, using
torch.cat()to form the skip connection.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
def forward(self, x):
x1 = self.enc1(x)
x2 = self.enc2(self.pool(x1))
x3 = self.enc3(self.pool(x2))
x4 = self.enc4(self.pool(x3))
x = self.upconv3(x4)
x = torch.cat([x, x3], dim=1)
x = self.dec1(x)
x = self.upconv2(x)
x = torch.cat([x, x2], dim=1)
x = self.dec2(x)
# Define the last decoder block with skip connections
x = ____
x = ____
x = ____
return self.out(x)