Uncertainty Modeling with Second-Order Transformer for Group Re-identification

Group re-identification (G-ReID) focuses on associating the group images containing same persons under different cameras. The key challenge of G-ReID is that all cases intra-group member and layout variations are hard to exhaust. To this end, we propose a novel uncertainty modeling, which treats each image as distribution depending current layout, then digs out potential features by random samplings. Based original features, modeling can learn better decision boundaries, implemented two modules, variation module (MVM) (LVM). Furthermore, second-order transformer framework (SOT), inspired fact position in coped with task. SOT composed intra-member inter-member module. Specifically, extracts first-order token for member, learns feature above tokens, be regarded tokens. A large number experiments have been conducted three available datasets, including CSG, DukeGroup RoadGroup. experimental results show proposed outperforms previous state-of-the-art methods.