GAN Model Collection

Overview

GAN (generative adversial networks) and some variants implementation.

List of GAN and its variants:

• InfoGan: infomative GAN model.
• AeGan: auto-encoder GAN model.
• WGan-GP: Wasserstein GAN model with gradient peanlty.
• LsGan: least-square GAN model.
• Gan-cls: LSGAN model with features and labels together in input of discriminator.

Requirements

• Python 3.5+
• TensorFlow 1.3.0+

Usages

To train/infer/generate GAN models, you may need to run codes by following the steps in each model folder:

the below steps are based on InfoGAN model, others have similar steps,

Train GAN models

> python Train_Mnist.py 

Infer (classify) a new sample in GAN models

> python Inference_Mnist.py

Generate samples

> python Generate_Mnist.py

Plot images

> python Plot.py