/WGAN-LP-tensorflow

Reproduction code for WGAN-LP

Primary LanguagePythonMIT LicenseMIT

WGAN-LP-tensorflow

Report on arXiv

Reproduction code for the following paper:

Title:	
On the regularization of Wasserstein GANs
Authors:	
Petzka, Henning; Fischer, Asja; Lukovnicov, Denis
Publication:	
eprint arXiv:1709.08894
Publication Date:	
09/2017
Origin:	
ARXIV
Keywords:	
Statistics - Machine Learning, Computer Science - Learning
2017arXiv170908894P

Original Paper on arXiv

Repository structure

data_generator.py

  • provides a class that generates the sample data needed for learning.

reg_losses.py

  • defines the sampling method and loss term for regularization.

model.py

  • implements 3-layer neural networks for a generator and a critic.

trainer.py

  • a pipeline for model learning and visualization.