An Empirical Study on Disentanglement of Negative-free Contrastive Learning