Committee-Voting mechanism based on Graph for Semisupervised Person Re-identification

Abstract

Recently, person re-identification technology has made great progress, mainly thanks to a large amount of annotated data. However, the current million-dollar annotations are becoming increasingly difficult to scale. Therefore, we work on the problem of semi-supervised person re-identification using only a small amount of annotated data. Inspired by QBC, we propose the committee-voting mechanism based on graph (CVG), which consists of two modules, the committee, and the chairman. The committee proposes different opinions and constructs multiple sets of information pairs for unlabelled data, and the chairman is responsible for aggregating the opinions of all committee members and making the final decision. As a sound selection mechanism, which effectively improves the accuracy of semisupervised person re-identification. Extensive experiments on three large-scale datasets demonstrate the effectiveness of the proposed method.