niftynet.network.simple_gan module

class GenericGAN(generator, discriminator, name='generic_GAN')

Bases: niftynet.layer.base_layer.TrainableLayer

### Description Generic Generative Adversarial Network

### Diagram

RANDOM NOISE –> [GENERATOR] –> [DISCRIMINATOR] –> fake logits TRAINING SET ——————> [DISCRIMINATOR] –> real logits

### Constraints

layer_op(random_source, population, is_training)

class SimpleGAN(name='simple_GAN')

Bases: niftynet.network.simple_gan.GenericGAN

### Description Specification of generator and discriminator for generic gan

### Building blocks [GENERATOR] - See ImageGenerator class below [DISCRIMINATOR] - See ImageDiscriminator class below

### Diagram

RANDOM NOISE –> [GENERATOR] –> [DISCRIMINATOR] –> fake logits TRAINING SET ——————> [DISCRIMINATOR] –> real logits

### Constraints

class ImageGenerator(hidden_layer_channels, name)

Bases: niftynet.layer.base_layer.TrainableLayer

### Description

### Diagram

### Constraints

__init__(hidden_layer_channels, name)

Parameters:

• hidden_layer_channels –
• name – layer name

layer_op(random_source, image_size, is_training)

Parameters:

• random_source – tensor, random noise to start generation
• image_size – output image size
• is_training – boolean, True if network is in training mode

Returns:

tensor, generated image

class ImageDiscriminator(hidden_layer_channels, name)

Bases: niftynet.layer.base_layer.TrainableLayer

### Description

### Diagram

### Constraints

__init__(hidden_layer_channels, name)

Parameters:

• hidden_layer_channels – array, number of output channels for each layer
• name – layer name

layer_op(image, is_training)

Parameters:

• image – tensor, input image to distriminator
• is_training – boolean, True if network is in training mode

Returns:

tensor, classification logits