Learning Multi-Attention Context Graph for Group-Based Re-Identification.

Learning to re-identify a group of people across camera systems has important applications in video surveillance. However, most existing methods focus on person re-identification (re-id), ignoring the fact that people often walk in groups. In this work, we consider employing context information for group re-id. On the one hand, group re-id is more challenging than single person re-id, since it requires both a robust modeling of individual person and full awareness of global group structures. On the other hand, person re-id can be greatly enhanced by incorporating visual context, a task which we formulate as group-aware person re-id. In this paper, we propose a novel unified framework to simultaneously address the above tasks, i.e., group re-id and group-aware person re-id. Specifically, we construct a context graph to exploit dependencies among different people. A multi-level attention mechanism is developed to formulate both intra- and inter-group context, with an additional self-attention module for robust graph-level representations. Meanwhile, to facilitate the deployment of deep learning models on these tasks, we build a new group re-id dataset containing 3.8K images with 1.5K annotated groups. Extensive experiments on the novel dataset as well as three existing datasets clearly demonstrate the effectiveness of the proposed framework.