Joint Person Objectness and Repulsion for Person Search

Person search targets to the probe person from unconstrainted scene images, which can be treated as combination of detection and matching. However, existing methods based on Detection-Matching framework ignore objectness repulsion (OR) are both beneficial reduce effect distractor images. In this paper, we propose an OR similarity by jointly considering information. Besides traditional visual term, also contains term a term. The images that not contain boost performance improving ranking positive samples. Because has different ID with its neighbors, gallery having higher neighbors should have lower person. Based constraint, is proposed most similar Treating Faster R-CNN detector, evaluated PRW CUHK-SYSU datasets six description models. extensive experiments demonstrate effectively samples further search, e.g., improve mAP 92.32% 93.23% for CUHK-SYSY dataset, 50.91% 52.30% datasets.