dltk.models.gan package

dltk.models.gan.dcgan module

class dltk.models.gan.dcgan.DCGAN(discriminator_filters=(64, 128, 256, 512), generator_filters=(512, 256, 128, 64, 1), discriminator_strides=((1, 1, 1), (2, 2, 2), (1, 1, 1), (2, 2, 2)), generator_strides=((7, 7, 7), (2, 2, 2), (2, 2, 2)), relu_leakiness=0.01, generator_activation=<function identity>, name='dcgan')

Bases: dltk.core.modules.base.AbstractModule

Convolutional Autoencoder

This module builds a convolutional autoencoder with varying number of layers and hidden units.

class Discriminator(filters, strides, relu_leakiness, name)

Bases: dltk.core.modules.base.AbstractModule

class Generator(filters, strides, output_activation, name)

Bases: dltk.core.modules.base.AbstractModule

dltk.models.gan.wdcgan module

class dltk.models.gan.wdcgan.WDCGAN(discriminator_filters=(64, 128, 256, 512), generator_filters=(512, 256, 128, 64, 1), discriminator_strides=((1, 1, 1), (2, 2, 2), (1, 1, 1), (2, 2, 2)), generator_strides=((7, 7, 7), (2, 2, 2), (2, 2, 2)), relu_leakiness=0.01, generator_activation=<function identity>, clip_val=0.01, improved=True, name='wdcgan')

Bases: dltk.models.gan.dcgan.DCGAN

Deep convolution generative network trained with a wasserstein loss based on https://github.com/igul222/improved_wgan_training

Module contents