BriefGPT.xyz
Mar, 2021
通过对比鉴别器强化增强训练GANs
Training GANs with Stronger Augmentations via Contrastive Discriminator
HTML
PDF
Jongheon Jeong, Jinwoo Shin
TL;DR
本研究提出了ContraD方法,将对比性表示学习方案融入生成对抗网络鉴别器中,使得生成器以更强的数据增强方式工作而不会增加训练不稳定性,并且在对比学习中也能受益。实验结果表明,GANs with ContraD在FID和IS方面表现更好,还能通过简单的潜在采样诱导许多条件生成模型。
Abstract
Recent works in
generative adversarial networks
(GANs) are actively revisiting various
data augmentation
techniques as an effective way to prevent
→