【论文简介】Circle Loss: A Unified Perspective of Pair Similarity Optimization
CircleLoss:AUnifiedPerspectiveofPairSimilarityOptimization旷世cvpr2020的一篇文章,站在更高的视角,统一了deepfeaturelearning的两大基础loss:基于class-levellabel的loss(如softmax+crossentropy)和基于pair-wiselabel的loss(如tripletloss),指出了