Re-ranking Person Re-identification with Adaptive Hard Sample Mining
Person re-identification(re-ID)aims at searching a specific person among non-overlapping cameras,which can be considered as a retrieval process,and the result is presented as a ranking list.There always exists the phenomenon that true matches are not the first rank,mainly owing to that they are more similar to other persons.In this paper,we use an adaptive hard sample mining method to re-train the selected samples in order to distinguish similar persons,which is applied for re-ranking the re-ID results.Specifically,in re-training stage,we divide the negative samples into three levels according to their ranking information.Meanwhile,we propose a coarse-fine tuning mechanism,which adaptively inflicts different punishment on the negative samples with the ranking information.Therefore,we can obtain a more valid metric,which is suitable for the re-ranking task to distinguish the easily-confused samples.Experimental results on VIPeR,PRID450S and CUHK03 datasets demonstrate the effectiveness of our proposed algorithm.
Re-ranking Hard sample mining Metric learning Adaptive
Chuchu Han Kezhou Chen Jin Wang Changxin Gao Nong Sang
Key Laboratory of Ministry of Education for Image Processing and Intelligent Control,School of Automation,Huazhong University of Science and Technology,Wuhan 430074,China
国际会议
广州
英文
3-14
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）