/WGAN-GP

Primary LanguagePython

WGAN-GP

Wasserstein Generative Adversial Network Gradient Penalty

Collaborator

Prayushi Mathur, Vraj Patel

This repository contains the files related to the latest generative network, WGAN-GP. The data used was self-generated.

Introduction

WGAN-GP is the latest generative adversial network which uses gradient penalty instead of the weight clipping to enforce the Lipschitz constraint. The data used was self-generated containing 8x8 noisy patches. This model learnt the noise distribution over the image and generated the similar results.

Application

It can generate noise to make a dataset of noisy images for the further better training of denoising autoencoders and denoising CNN.