복장이 변화된 사람 재식별을 위한 컬러 레이블 기반 분기 구조 모델 = Color Labels-based Branching Structure Model for Cloth-changing Person Re-identification

다국어 초록 (Multilingual Abstract)

Person re-identification involves the task of identifying a specific person among multiple images obtained from different locations.
Re-identifying individuals with complex clothing changes poses a greater challenge as clothing information, a key feature in conventional person re-identification, is altered. Attempts have been made to address this issue by either adding non-clothing features or excluding features related to clothing from training. However, such approaches have limitations in performance improvement as they exclude valid features of clothing. In contrast, this paper proposes a technique that utilizes valid features of clothing information through a color label-based branching structure model for re-identification. The proposed method is compared to the performance of state-of-the-art methods using the LTCC dataset, which consists of clothing change data. The results show superior performance based on Rank-1 and mAP metrics.