This repo is a collection of the implementations of many GANs. In order to make the codes easy to read and follow, I minimize the code and run on the same MNIST dataset.

What does the MNIST data look like?

Toy implementations are organized as following:

1. Base Method

• Deep Convolutional GAN (DCGAN)

2. Loss or Structure Modifications

• Least Squares GAN (LSGAN)
• Wasserstein GAN (WGAN)
  • Gradient Penalty (WGAN gp)
  • Wasserstein Divergence (WGAN_div)
• Self-Attention GAN (SAGAN)
• Progressive-Growing GAN (PGGAN)

3. Can be Conditional

• Conditional GAN (CGAN)
• Auxiliary Classifier GAN (ACGAN)
• InfoGAN

4. Image to Image Transformation

• Context-Conditional GAN (CCGAN)
• CycleGAN
• Pix2Pix
• Super-Resolution GAN (SRGAN)
• StyleGAN
• WIP

$ git clone https://github.com/MorvanZhou/mnistGANs
$ cd mnistGANs/
$ pip3 install -r requirements.txt

Generative Adversarial Nets

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

code - gif result

Least Squares Generative Adversarial Networks

code - gif result

Wasserstein GAN

code - gif result

Improved Training of Wasserstein GANs

code - gif result

Wasserstein Divergence for GANs

code - gif result

Self-Attention Generative Adversarial Networks

code - gif result

PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION

code - gif result

Conditional Generative Adversarial Nets

code - gif result

Conditional Image Synthesis with Auxiliary Classifier GANs

code - gif result

InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets

code - gif result

Semi-Supervised Learning with Context-Conditional Generative Adversarial Networks

code - gif result

Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks

code - gif result

Image-to-Image Translation with Conditional Adversarial Networks

code - gif result

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

code - gif result

A Style-Based Generator Architecture for Generative Adversarial Networks

code - gif result