Person Re-Identification Based On Deep Learning

Person re-identification(ReID)aims to retrieve particular persons across different cameras.Although great progress has been made with the success of deep learning in recent years,it still suffers from the challenges brought by the change of pedestrian post variations,different resolutions and similar appearances.So it is necessary to locate the discriminative features of pedestrians in the process of identification.Considering the fact that it is not always easy to find clues that distinguish different pedestrians,this paper proposes a self-supervision mechanism.That is,the model locates the informative regions firstly,and then let the network judge whether the informative regions is reasonable,so as to promote the network to extract better features.A large number of experiments show that the method proposed in this paper help the model find discriminative clues,and the strategy makes significant improvement over the baselines on two commonly used datasets(Market1501 and DukeMTMC-reID).