person re-identification has achieved great progress with deep convolutional
neural networks. However, most previous methods focus on learning individual
appearance feature embedding, and it is hard for the models to handle difficult
situations with different illumination, large pose v