Understanding Dimensional Collapse in Contrastive Self-supervised Learning