DeepID2:Deep Learning Face Representation by Joint Identification-Verification

人脸识别最具挑战性的地方在于减少类内差异同时增大类间差异
The key challenge of face recognition is to develop effective feature repre-sentations for reducing intra-personal variations while enlarging inter-personaldifferences.

DeepID2:Deep Learning Face Representation by Joint Identification-Verification

这篇论文提出了两种信号(实际是loss函数)identification signalverification signal来达到上述目的。

identification signal

用于增大类间差异,实际上就是通常的交叉熵损失,用与在训练当中分出不同类别。
DeepID2:Deep Learning Face Representation by Joint Identification-Verification

verification signal

用于减少类内差异,训练时给出同一个人的两张图片,目的是使这两张图片得到的特征向量的距离尽可能小。
由于衡量两个向量的距离有多种方式,因此verification signal可以有多种不同的形式。基于L1距离L2距离或者余弦相似度

DeepID2:Deep Learning Face Representation by Joint Identification-Verification
DeepID2:Deep Learning Face Representation by Joint Identification-Verification