lawrencekiba/Keras-GAN

Keras implementations of Generative Adversarial Networks.

Keras-GAN

About

Keras implementations of Generative Adversarial Networks (GANs) suggested in research papers. If dense layers gets the job done for a model I will often prefer them over conv layers. The reason for this is that I would like to enable people without GPUs to test these implementations out. These models are in some cases simplified versions of the ones ultimately described in the papers, but I have chosen to focus on getting the core ideas covered instead of getting every layer configuration right. However, because of this the results will not always be as nice as in the papers.

Table of Contents

• Keras-GAN
  • About
  • Table of Contents
  • Implementations
    • Auxiliary Classifier GAN
    • Adversarial Autoencoder
    • Bidirectional GAN
    • Boundary-Seeking GAN
    • Context-Conditional GAN
    • Context Encoder
    • Coupled GANs
    • Deep Convolutional GAN
    • DualGAN
    • Generative Adversarial Network
    • InfoGAN
    • LSGAN
    • Semi-Supervised GAN
    • Wasserstein GAN

Implementations

AC-GAN

Implementation of Auxiliary Classifier Generative Adversarial Network.

Code

Paper: https://arxiv.org/abs/1610.09585

Adversarial Autoencoder

Implementation of Adversarial Autoencoder.

Code

Paper: https://arxiv.org/abs/1511.05644

BiGAN

Implementation of Bidirectional Generative Adversarial Network.

Code

Paper: https://arxiv.org/abs/1605.09782

BGAN

Implementation of Boundary-Seeking Generative Adversarial Networks.

Code

Paper: https://arxiv.org/abs/1702.08431

CC-GAN

Implementation of Semi-Supervised Learning with Context-Conditional Generative Adversarial Networks.

Inpainting using a GAN where the generator is conditioned on a randomly masked image. In this implementation images of dogs and cats taken from the Cifar-10 dataset are used. These images are of very low resolution and the results are therefore not as nice as in the implementation described in the paper. In this implementation I have also decided to combine the adversarial loss with an l2 loss which measures the generated image's similarity to the original images. These losses are weighted similar to the approach described by Pathak et al. (2016)
in their paper Context Encoders: Feature Learning by Inpainting.

Code

Paper: https://arxiv.org/abs/1611.06430

Context Encoder

Implementation of Context Encoders: Feature Learning by Inpainting.

Code

Paper: https://arxiv.org/abs/1604.07379

CoGAN

Implementation of Coupled generative adversarial networks.

Code

Paper: https://arxiv.org/abs/1606.07536

DCGAN

Implementation of Deep Convolutional Generative Adversarial Network.

Code

Paper: https://arxiv.org/abs/1511.06434

DualGAN

Implementation of DualGAN: Unsupervised Dual Learning for Image-to-Image Translation.

Code

Paper: https://arxiv.org/abs/1704.02510

GAN

Implementation of Generative Adversarial Network with a MLP generator and discriminator.

Code

Paper: https://arxiv.org/abs/1406.2661

InfoGAN

Implementation of InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets.

Code

Paper: https://arxiv.org/abs/1606.03657

LSGAN

Implementation of Least Squares Generative Adversarial Networks.

Code

Paper: https://arxiv.org/abs/1611.04076

SGAN

Implementation of Semi-Supervised Generative Adversarial Network.

Code

Paper: https://arxiv.org/abs/1606.01583

WGAN

Implementation of Wasserstein GAN (with DCGAN generator and discriminator).

Code

Paper: https://arxiv.org/abs/1701.07875