Robust joint learning network: improved deep representation learning for person re-identification

Existing person re-identification methods, which based on deep representation learning, mostly only focus on either global feature or local feature. This obviously ignores the joint advantages and the correlation between global and local features. In this paper, we test and verify the benefits of jointly learning local and global features in a network based on the Convolutional Neural Network (CNN). Specifically, we give distinct weights to global loss and local loss when considering their different influence on our research, then we innovatively combine two losses into one loss. Besides, we propose a novel and strong network to learn part-level features with unified partition. Experimental results on three person ReID data sets, show that our method outperforms existing deep learning methods.