Mehdi S. M. Sajjadi, Giambattista Parascandolo, Arash Mehrjou, Bernhard Schölkopf
TL;DR通过引入一种简单的方法,使真实数据分布经过一个‘透镜’传达给辨别器,让生成器逐步揭示出更多细节特征,改善了 GAN 训练的质量、稳定性和收敛速度,对各种 GAN 架构如 DCGAN、LSGAN、WGAN-GP 都可行。
Abstract
generative adversarial networks (GANs) have been shown to produce realistic
samples from high-dimensional distributions, but training them is considered
hard. A possible explanation for training instabilities is