http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
사람-의상 군집화 알고리즘 기반 의상 변경 상황에서의 비지도 도메인 적응형 사람 재식별 기술
김윤한(Yoon-Han Kim),오민영(Minyoung Oh),홍승빈(Seung-Bin Hong),심재영(Jae-Young Sim) 대한전자공학회 2024 대한전자공학회 학술대회 Vol.2024 No.6
Most of existing person re-identification (ReID) methods only consider short-term scenarios such that all the instances with the same person identity wear consistent clothes in dataset. Recently, there has been a lot of researches in the field of clothes-changing ReID (CC-ReID) to solve a more realistic problem, assuming that the persons change their clothes over time. However, it is much harder to collect labeled datasets for CC-ReID. To deal with this problem, we propose a novel unsupervised domain adaptive setting for CC-ReID, which is based on the clustering methods. Meanwhile, existing clustering algorithms depend on the person identity as well as the characteristics of data such as the clothes, light conditions, camera settings, etc. These may cause poor clustering performance in CC-ReID. Therefore, we propose a Cross Clustering Module (CCM) that performs the pseudo labeling by considering not only the person features but also the clothing features. Experimental results verify the proposed method improves the performance of CC-ReID by a large margin on LTCC dataset.