Penalty Gradient Normalization for Generative Adversarial Networks

Tian Xia
6/23/2023
cs.CVcs.LG

Official code: sumorday/EDGAN-PyTorch (4 stars, last commit 2025-01-22).

Abstract

In this paper, we propose a novel normalization method called penalty gradient normalization (PGN) to tackle the training instability of Generative Adversarial Networks (GANs) caused by the sharp gradient space. Unlike existing work such as gradient penalty and spectral normalization, the proposed PGN only imposes a penalty gradient norm constraint on the discriminator function, which increases the capacity of the discriminator. Moreover, the proposed penalty gradient normalization can be applied to different GAN architectures with little modification. Extensive experiments on three datasets show that GANs trained with penalty gradient normalization outperform existing methods in terms of both Frechet Inception and Distance and Inception Score.

View on arXivView PDF

Code Implementations(1)

sumorday/EDGAN-PyTorchOfficial100%

Official implementation for Penalty Gradient Normalization for Generative Adversarial Networks

41PythonFeb 17, 20231 years ago

Similar Papers

A New Formulation of Lipschitz Constrained With Functional Gradient Learning for GANs

85% match1 repo6612025

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

85% match1 repo32024

Local Convergence of Gradient Descent-Ascent for Training Generative Adversarial Networks

85% match2023

Parallelly Tempered Generative Adversarial Nets: Toward Stabilized Gradients

84% match2024

Nested Annealed Training Scheme for Generative Adversarial Networks

83% match2025

Error analysis of generative adversarial network

83% match2023

Cite this paper

@article{xia2023penalty,
  title  = {Penalty Gradient Normalization for Generative Adversarial Networks},
  author = {Tian Xia},
  year   = {2023},
  eprint = {2306.13576},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2306.13576v1}
}

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/2306.13576.svg)](https://codeofpaper.com/paper/2306.13576)

Discussion