Attention driven person re-identification

Highlights：

• It designs a novel multi-branch network architecture that learns precise and discriminative person ReID features under the guidance of intra-attention and inter-attention.

• It designs a novel intra-attention network that learns discriminative features from precisely aligned global whole-body images and body-part images concurrently and independently.

• It designs a novel inter-attention module that fuses discriminative features of the global whole-body images and local body-part images adaptively for optimal person ReID.

• It develops an end-to-end trainable deep network system that achieves superior person ReID performance across a number of widely used benchmarking datasets.