Deep feature embedding learning for person re-identification based on lifted structured loss

Zhangping He, Cheolkon Jung, Qingtao Fu, Zhendong Zhang

Abstract: Person re-identification (re-id) aims at matching the same individual in videos captured by multiple cameras, and much progress has been made in recent years due to large scale pedestrian data sets and deep learning-based techniques. In this paper, we propose deep feature embedding learning for person re-id based on lifted structured loss. Triplet loss is commonly used in deep neural networks for person re-id. However, the triplet loss-based framework is not able to make full use of the batch information, and thus needs to choose hard negative samples manually that is time-consuming. To address this problem, we adopt lifted structured loss for deep neural networks that makes the network learn better feature embedding by minimizing intra-class variation and maximizing inter-class variation. Extensive experiments on Market-1501, CUHK03, CUHK01 and VIPeR data sets demonstrate the superior performance of the proposed method over state-of-the-arts in terms of the cumulative match curve (CMC) metric.