R-MAE: Regions Meet Masked Autoencoders