![](https://img-blog.csdnimg.cn/20201014180756757.png?x-oss-process=image/resize,m_fixed,h_64,w_64)
度量学习
文章平均质量分 89
大坡山小霸王
这个作者很懒,什么都没留下…
展开
-
SoftTriple Loss
目录19-ICCV-SoftTriple Loss:Deep Metric Learning Without Triplet SamplingSoftTriple LossMultiple CentersAdaptive Number of Centers1)SoftMax loss is equivalent to a smoothed triplet loss where each class has a single center.现实中一个类不只有一个中心,例如鸟有很多姿势(从细粒度角度解释)。扩展原创 2022-06-29 14:15:53 · 1066 阅读 · 1 评论 -
Margin Based Loss
目录17-ICCV-Sampling Matters in Deep Embedding LearningPreliminariescontrastive losstriplet losshard negative miningsemi-hard negative miningDistance weighted samplingMargin based lossRelationship to isotonic regression正样本尽可能近,负样本被固定距离α隔开visually diverse cla原创 2022-06-28 23:02:29 · 1213 阅读 · 0 评论 -
N-pair Loss
目录16-NIPS-improved deep metric learning with multi-class n-pair loss objectiveMultiple Negative Examples(N+1)-tuplet lossmulti-class N-pair loss (N-pair-mc)one-vs-one N-pair loss (N-pair-ovo)Hard negative class miningL2 norm regularization of embedding vec原创 2022-06-28 15:45:38 · 2647 阅读 · 0 评论 -
Proxy-NCA Loss、Proxy Anchor Loss
目录17-ICCV-No Fuss Distance Metric Learning using ProxiesProxy Ranking Loss20-CVPR-Proxy Anchor Loss for Deep Metric Learning Proxy Anchor Loss|P|原创 2022-06-13 22:07:05 · 2227 阅读 · 0 评论 -
Normalized Softmax Loss
目录19-BMVC-Classification is a Strong Baseline for Deep Metric LearningSoftmax LossLayer Normalization类别平衡采样binary embeddings19-ICCV-SoftTriple Loss:Deep Metric Learning Without Triplet SamplingSoftTriple Loss1) we establish that classification is a strong原创 2022-06-12 22:14:58 · 864 阅读 · 0 评论 -
对比损失(Contrastive Loss)、三元损失(Triplet Loss)
Triplet Loss Triplet Selection离线:每n步使用最近的网络,子集在线:mini-batch所有A-P pairs和困难N-P pairs避免f(x)=0 semi-hard原创 2022-06-01 16:30:53 · 2483 阅读 · 0 评论