Better triplet loss variants for retrieval

Hi,

I am wondering that is there any metric learning method for retrieval tasks work better than triplet loss and contrastive loss?

Thanks a lot