Dual Attentive Features for Person Re-identification

Person re-identification (re-ID) plays a vital role in intelligent video surveillance and is the task of retrieving particular persons across different cameras. Effective convolutional features is a key component in person re-ID but how to learn powerful features is still a challenging task due to pose variation, occlusion, and similar appearance among different persons. Considering this fact, we propose a novel dual attention module network which consists of spatial attention and channel-wise attention. Dual attention module add weights on image features in spatial and channel. Attentive features generated by the proposed network are able to better enhance the foreground objects while eliminate background clutter. Intensive experimental results have been prove that the proposed dual attention module network outperforms the state-of-the-art on Market-1501 and DukeMTMC-reID datasets.