MotionMAE: Self-Supervised Video Representation Learning with Motion-Aware Masked Autoencoders (BMVC2024)