Harmonious Multi-branch Network for Person Re-identification with Harder Triplet Loss

Recently, advances in person re-identification (Re-ID) has benefitted from use of the popular multi-branch network. However, performing feature learning a single branch with uniform partitioning is likely to separate meaningful local regions, and correlation among different branches not well established. In this article, we propose novel harmonious network (HMBN) relieve these intra-branch inter-branch problems harmoniously. HMBN various stripes on learn coarse-to-fine pedestrian information. We first replace partition horizontal overlapped cover regions between adjacent branch. then incorporate attention module make all interact by modeling spatial contextual dependencies across branches. Finally, order train more effectively, harder triplet loss introduced optimize triplets manner. Extensive experiments are conducted three benchmark datasets — DukeMTMC-reID, CUHK03, Market-1501 demonstrating superiority our proposed over state-of-the-art methods.