Introduction

This project uses GAN to generate desired curve, which implemented by keras. In [1][2], there are implementations by tensorflow and pytorch respectively; this project uses keras. There is one thing i want to mention: in this project, i use different loss function than original paper; see loss function and reasons in [4].

Run code

python GAN.py

Recent update

Added Wasserstein GAN. Even the reasons why WGAN performs better than GAN are complicated, the differences of code by using keras are minor. In fact, only few lines need to be changed in this example.

Run code

python WGAN.py

Future work

Implement different type of GAN.

Reference

[1]Tensorflow-Tutorial [2]PyTorch-Tutorial [3]Keras-GAN [4]ganhacks