问题起源于实验室是同学问:triplet loss比contrastive loss的改进是什么,但尴尬的是我不知道这两个是什么…于是想写写见到的loss function
contrastive loss
L = 1 2 N ∑ n = 1 N y d 2 + ( 1 − y ) m a x ( m a r g i n − d , 0 ) 2 L=\frac{1}{2N} \sum_{n=1}^N yd^2+(1−y)max(margin−d,0)^2 L=2N1∑n=1Nyd2+(1−y)max(margin−d,0)2
提出于:Dimensionality Reduction by Learning an Invariant Mapping,luan sheng
triple loss
L = m a x ( d ( a , p ) − d ( a , n ) + m a r g i n , 0 ) L=max(d(a,p)−d(a,n)+margin,0) L=max(d(a,p)−d(a,n)+margin,0)
提出于:FaceNet: A Unified Embedding for Face Recognition and Clustering,目的是学习较好的脸的embedding