Key Components of Deep Metric Learning

Abstract

The emergence of early metric learning algorithms improved the distance-based classifier, the distance-based clustering and the performance of feature dimensionality reduction. Compared with classic metric learning, deep metric learning can effectively use massive training data, can perform non-linear mapping of input features, and can effectively use deep learning models and technologies that have developed rapidly in recent years. In this survey we review recent deep metric learning. We review commonly used loss functions, and point key components of deep metric learning.