MixMAE: Mixed and Masked Autoencoder for Efficient Pretraining of Hierarchical Vision Transformers | IEEE Conference Publication | IEEE Xplore