A Simple and Effective Deep Model for Person Re-identification

摘要

Person re-identification (re-ID), which aims to re-identify a person captured by one camera from another camera at any non-overlapping location, has attracted more and more attention in recent years. So far, it has been significantly improved by deep learning technology. A variety of deep models have been proposed in person re-ID community. In order to make the deep model simple and effective, we propose an identification model that combines the softmax loss with center loss. Moreover, various data augmentation methods and re-ranking strategy are used to improve the performance of the proposed model. Experiments on CUHK03 and Market-1501 datasets demonstrate that the proposed model is effective and has good results in most cases.