Attention Deep Model With Multi-Scale Deep Supervision for Person Re-Identification

As an important part of intelligent surveillance systems, person re-identification (PReID) has drawn wide attention the public in recent years. Many deep learning-based PReID methods have used or multi-scale feature learning modules to enhance discrimination learned features. However, mechanisms may lose some information. Moreover, models usually embed module into backbone network, which increases complexity testing network. To address two issues, we propose a supervision with model for PReID. Specifically, introduce reverse remedy information losing issue caused by module, and layer train The proposed are only at training phase discarded during test phase. Experiments on Market-1501, DukeMTMC-reID, CUHK03 MSMT17 datasets. demonstrate that our notably beats other competitive state-of-the-art models.