Abstract

In contrast to category-level or cluster-level classififiers, exemplar SVM [17] is successfully applied to classifying (or detecting) a target object as well as transferring instancelevel annotations. The method, however, is formulated in a highly biased classifification problem where only one positive sample is contrasted with a substantial number of negative samples, which makes it diffificult to properly determine the regularization parameters balancing two types of costs derived from positive and negative samples. In this paper, we present two novel viewpoints toward exemplar SVM in addition to the original defifinition. From these proposed viewpoints, we can give light on an intrinsic structure of exemplar SVM, reducing two parameters into only one as well as providing clear intuition on the parameter, in order to free us from exhaustive parameter tuning. We can also clarify how the classififier geometrically works so as to produce homogeneous classifification scores of multiple exemplar SVMs which are comparable to each other without calibration. In addition, we propose a novel feature transformation method based on those viewpoints which contributes to general classifification tasks. In the experiments on object detection and image classifification, the proposed methods regarding exemplar SVM exhibit favorable performance.