Re-identifying Pedestrians via Part Based Method

Person re-identification (Re-id) has received more and more attention due to its wide applications and various approaches were proposed to solve the challenging task. In this paper, we adopt part-based model as the main method to perform person Re-id on the self-made dataset. In order to improve accuracy, transfer learning is applied in both a direct and indirect way by leveraging original public dataset Market-1501 and its transferred counterpart respectively. The transferred dataset is obtained by making the pedestrians in the original public dataset show similar styles to those of the self-made dataset while keeping their identities unchanged. Experimental results demonstrate the effectiveness of transfer learning and the superiority of applying transfer learning indirectly against applying transfer learning directly.