Masked Autoencoders Are Scalable Vision Learners

Abstract

This paper introduces and evaluates the idea described in “Masked Autoencoders Are Scalable Vision Learners”, and reports empirical results that helped shape subsequent work in self-supervised, vision.