Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners