SN-GAN

A new weight normalization technique that generally improves GAN training

Released in: Spectral Normalization for Generative Adversarial Networks

Contributor:

by

Homepage

Summary

One of the challenges in the study of generative adversarial networks is the instability of its training. This paper proposes a novel weight normalization technique called spectral normalization to stabilize the training of the discriminator. The new normalization technique is computationally light and easy to incorporate into existing implementations. The authors tested the efficacy of spectral normalization on CIFAR10, STL-10, and ILSVRC2012 dataset, and experimentally confirmed that spectrally normalized GANs (SN-GANs) is capable of generating images of better or equal quality relative to the previous training stabilization techniques.

2018

Year Released

Key Links & Stats

SN-GAN (spectral normalization GAN) in PyTorch

Spectrally Normalised GAN

Spectral Normalization for Generative Adversarial Networks

@inproceedings{DBLP:conf/iclr/MiyatoKKY18, author = {Takeru Miyato and Toshiki Kataoka and Masanori Koyama and Yuichi Yoshida}, title = {Spectral Normalization for Generative Adversarial Networks}, booktitle = {6th International Conference on Learning Representations, {ICLR} 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings}, publisher = {OpenReview.net}, year = {2018}, url = {https://openreview.net/forum?id=B1QRgziT-}, timestamp = {Thu, 04 Apr 2019 13:20:10 +0200}, biburl = {https://dblp.org/rec/conf/iclr/MiyatoKKY18.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }

ML Tasks

  1. General
  2. Image Generation

ML Platform

  1. Pytorch

Modalities

  1. General
  2. Still Image

Verticals

  1. General

CG Platform

  1. Not Applicable

Related organizations

Preferred Networks, Inc.

Ritsumeikan University

National Institute of Informatics, Japan