Group affinity guided deep hypergraph model for person re‐identification

Person re‐identification aims at searching in a large gallery image database for images of the same identity as probe image, which can also be treated as a retrieval task in which the pairwise affinity is often used to rank the retrieved images. However, most existing methods of person re‐identification only consider pairwise affinity but ignore the group affinity information. Some frameworks incorporate group affinity into the testing phase, which is not end‐to‐end trainable for deep neural networks. In this Letter, a powerful deep learning based framework for person re‐identification is presented, which aims at fully utilising group affinity information for more discriminative features. Specifically, hypergraph model for group information mining is novelly proposed in the training phase, which significantly improves the performance of features acquired by the baseline networks. Extensive experiments on two popular benchmarks demonstrate the effectiveness of the authors' model and applicability on different baseline networks.