Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
ViTMAE has a similar pretraining objective, except it must predict the pixels instead of visual tokens.