Atlas generative models and geodesic interpolation

作者：

Highlights：

• Interpreting hybrid discrete-continuous latent spaces of generative models as atlases

• This enables geodesic interpolation while linear interpolation is not well-defined

• Theoretically unifies models from popular paradigms as Atlas Generative Models

• Algorithm for graph based geodesic interpolation on any Atlas Generative Model

• Experimentally verifies concept of Atlas Wasserstein Autoencoder and graph geodesics

摘要

•Interpreting hybrid discrete-continuous latent spaces of generative models as atlases•This enables geodesic interpolation while linear interpolation is not well-defined•Theoretically unifies models from popular paradigms as Atlas Generative Models•Algorithm for graph based geodesic interpolation on any Atlas Generative Model•Experimentally verifies concept of Atlas Wasserstein Autoencoder and graph geodesics

论文关键词：Generative networks,Manifold learning,Manifold atlas,Geodesic interpolation

论文评审过程：Received 10 February 2021, Revised 29 January 2022, Accepted 11 March 2022, Available online 31 March 2022, Version of Record 25 April 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2022.104433