This paper enhances contrastive learning by introducing the variable similarity of augmented views.