RPGAN: random paths as a latent space for GAN interpretability

Andrey Voynov, Artem Babenko

Keywords: generative models, GAN, interpretability

TL;DR: We introduce an alternative GAN design based on random routes in generator, which can serve as a tool for generative models interpretability.

Abstract: In this paper, we introduce Random Path Generative Adversarial Network (RPGAN) --- an alternative scheme of GANs that can serve as a tool for generative model analysis. While the latent space of a typical GAN consists of input vectors, randomly sampled from the standard Gaussian distribution, the latent space of RPGAN consists of random paths in a generator network. As we show, this design allows to associate different layers of the generator with different regions of the latent space, providing their natural interpretability. With experiments on standard benchmarks, we demonstrate that RPGAN reveals several interesting insights about roles that different layers play in the image generation process. Aside from interpretability, the RPGAN model also provides competitive generation quality and allows efficient incremental learning on new data.

Code: https://github.com/rpgan-ICLR2020/RPGAN