NeurIPS 2020

GANSpace: Discovering Interpretable GAN Controls

Meta Review

This paper studies interpretable controls for generative models using PCA directions in latent space. The proposed technique is natural and applied to Stylegan and BigGAN showing very good empirical results. Despite concerns by some reviewers that PCA directions are not necessarily the right thing to do, they are reasonable and work in practice, so we overall think this paper is a suitable contribution for NeurIPS. Please include the mentioned related works in the final manuscript.