Learning Subject-Discriminative Features

Abstract

In this section, we introduce algorithms for online descriptor extraction and adaptive candidate matching. The online algorithm leverages the pose prior $\mathbf {P}(\theta )$ and classifier $\mathbf {W}$ learned offline using training data in the previous section. In previous chapters, we discuss offline-learned distance metric $\mathbf {W}$, signatures and $\mathbf {P}(\theta )$. In this chapter, we are going to dig into boosting human re-identification performance online.