Generator

A GAN generator takes a random noise vector as input and produces a generated image. To make its architecture more reusable, you will pass both input and output shapes as parameters to the model. This way, you can use the same model with different sizes of input noise and images of varying shapes.

You will find torch.nn imported already imported for you as nn. You can also access a custom gen_block() function which returns a block of: linear layer, batch norm, and ReLU activation. You will use it as a building block for the generator.

def gen_block(in_dim, out_dim):
    return nn.Sequential(
        nn.Linear(in_dim, out_dim),
        nn.BatchNorm1d(out_dim),
        nn.ReLU(inplace=True)
    )

This exercise is part of the course

Deep Learning for Images with PyTorch

View Course

Exercise instructions

Define self.generator as a sequential model.
After the last gen_block, add a linear layer with the appropriate input size and the output size of out_dim.
Add a sigmoid activation after the linear layer.
In the forward() method, pass the model's input through self.generator.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

class Generator(nn.Module):
    def __init__(self, in_dim, out_dim):
        super(Generator, self).__init__()
        # Define generator block
        self.generator = ____(
            gen_block(in_dim, 256),
            gen_block(256, 512),
            gen_block(512, 1024),
          	# Add linear layer
            ____
            # Add activation
            ____
        )

    def forward(self, x):
      	# Pass input through generator
        return ____

Edit and Run Code