Mining Better Samples for Contrastive Learning of Temporal Correspondence | IEEE Conference Publication | IEEE Xplore