The discriminator network contains five volumetric convolutional layers with the following configuration:
- 3D convolutional layers: 5
- Channels: 64, 128, 256, 512, 1
- Kernel sizes: 4, 4, 4, 4, 4
- Strides: 2, 2, 2, 2, 1
- Activations: Leaky ReLU, Leaky ReLU, Leaky ReLU, Leaky ReLU, Sigmoid
- Batch normalization: Yes, Yes, Yes, Yes, None
- Pooling layers: No, No, No, No, No
- Linear layers: No, No, No, No, No
The input and output of the network are as follows:
- Input: A 3D image with shape (64, 64, 64)
- Output: The probability of the input data belonging to either the real or the fake class
The flow of the tensors and the input and output shapes of the tensors for each layer in the discriminator network are shown in the following diagram. This will provide you with a better understanding of the discriminator network:
The discriminator network mostly mirrors the generator network. An important difference is that it uses LeakyReLU instead of ReLU as the activation function. Also, the sigmoid layer at the end of the network is for binary classification and predicts whether the provided image is real or fake. The last layer has no normalization layer, but the other layers use batch normalization to regularize the input.