In Defense of the Classification Loss for Person Re-Identification

The recent research for person re-identification has been focused on two trends. One is learning the part-based local features to form more informative feature descriptors. The other one is designing effective metric learning loss functions such as the Triplet loss family. We argue that learning global features with classification loss could achieve the same goal, even with a simple and cost-effective architecture design. We propose a person re-id framework featured by channel grouping and multi-branch strategy, which divides the global feature into multiple channel groups and learns the discriminative channel group features by multi-branch classification layers. In extensive experiments, our network outperforms state-of-the-art person re-id frameworks in terms of both accuracy and inference cost.