Spectral Normalization for Generative Adversarial Networks

Takeru Miyato, Toshiki Kataoka, Masanori Koyama, Yuichi Yoshida
2/16/2018
cs.LGcs.CVstat.ML

Official code: pfnet-research/sngan_projection (1,102 stars, last commit 2019-11-12). 1 more implementation below.

Abstract

One of the challenges in the study of generative adversarial networks is the instability of its training. In this paper, we propose a novel weight normalization technique called spectral normalization to stabilize the training of the discriminator. Our new normalization technique is computationally light and easy to incorporate into existing implementations. We tested the efficacy of spectral normalization on CIFAR10, STL-10, and ILSVRC2012 dataset, and we 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.

View on arXivView PDF

Code Implementations(2)

pfnet-research/sngan_projectionOfficial100%

GANs with spectral normalization and projection discriminator

1,102201Shell, PythonFeb 2, 20186 years agoNOASSERTION
crcrpar/pytorch.sngan_projectionCommunity100%
MIT

An unofficial PyTorch implementation of SNGAN (ICLR 2018) and cGANs with projection discriminator (ICLR 2018).

15631Python, DockerfileSep 2, 20186 years agoMIT

Similar Papers

On the Effects of Batch and Weight Normalization in Generative Adversarial Networks

87% match1 repo222017

Penalty Gradient Normalization for Generative Adversarial Networks

87% match1 repo42023

Training Generative Adversarial Networks with Limited Data

87% match1 repo1.8k2020

Progressive Growing of GANs for Improved Quality, Stability, and Variation

86% match1 repo6.2k2017

Spectral Regularization for Combating Mode Collapse in GANs

84% match2 repos112019

CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization

83% match1 repo32024

Cite this paper

@article{miyato2018spectral,
  title  = {Spectral Normalization for Generative Adversarial Networks},
  author = {Takeru Miyato and Toshiki Kataoka and Masanori Koyama and Yuichi Yoshida},
  year   = {2018},
  eprint = {1802.05957},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1802.05957}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1802.05957.svg)](https://codeofpaper.com/paper/1802.05957)

Discussion