Fig 4.

Generator and discriminator networks: This figure illustrates the generator network (top row) and the discriminator network (bottom row). The generator consists of three convolutional layers (conv) with a rectified linear activation function (ReLU), followed by nine residual blocks, two transpose convolutional layers, and a final convolutional layer with a tanh activation function. The discriminator consists of five convolutional layers and classifies images into two categories: real or fake. The black numbers on top of the layers represent the number of feature channels. Below each array, the colored numbers denote the convolutional kernel size (#x#), the size of the stride s and the size of potential zero-padding zp.