Group Reidentification with Multigrained Matching and Integration

The task of re-identifying groups people underdifferent camera views is an important yet less-studied problem.Group re-identification (Re-ID) a very challenging sinceit not only adversely affected by common issues in traditionalsingle object Re-ID problems such as viewpoint and human posevariations, but it also suffers from changes group layout andgroup membership. In this paper, we propose novel conceptof granularity characterizing image multi-grained objects: individual persons sub-groups two andthree within group. To achieve robust Re-ID,we first introduce representations which can beextracted via the development separate schemes, i.e. onewith hand-crafted descriptors another with deep neuralnetworks. proposed representation seeks to characterize bothappearance spatial relations objects, isfurther equipped importance weights capture varia-tions intra-group dynamics. Optimal group-wise matching isfacilitated multi-order process turn,dynamically updates iterative fashion.We evaluated on three multi-camera datasets containingcomplex scenarios large dynamics, experimental resultsdemonstrating effectiveness our approach. published dataset be found \url{http://min.sjtu.edu.cn/lwydemo/GroupReID.html}