Spectral bounding: Strictly satisfying the 1-Lipschitz property for generative adversarial networks
作者：
Highlights：
• The paper starts by highlighting the need for spectral bounding of weights in the discriminator for GAN training.
• The paper proposes to perform quick spectral bounding by using the 1 norm and infinity norms of the weight matrices to normalize the weights of the models.
• Extensive experimental results on CIFAR-10 and ImageNet dataset demonstrate that our approach can maintain more successfully the balance between generators and discriminators encountered prior to a Nash equilibrium having been reached. In so doing we can obtain a robust GAN model which accurately captures features of the statistical distribution for data samples used in training.
摘要
•The paper starts by highlighting the need for spectral bounding of weights in the discriminator for GAN training.•The paper proposes to perform quick spectral bounding by using the 1 norm and infinity norms of the weight matrices to normalize the weights of the models.•Extensive experimental results on CIFAR-10 and ImageNet dataset demonstrate that our approach can maintain more successfully the balance between generators and discriminators encountered prior to a Nash equilibrium having been reached. In so doing we can obtain a robust GAN model which accurately captures features of the statistical distribution for data samples used in training.
论文关键词：Generative adversarial networks,1-Lipschitz constraint,Spectral bounding,Image generation
论文评审过程：Received 11 June 2019, Revised 7 November 2019, Accepted 16 December 2019, Available online 24 December 2019, Version of Record 5 June 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.107179
