Gdmn: group decision-making network for person re-identification

Person re-identification (re-ID) is a widely studied yet still challenging problem in computer vision. It aims to match images of the same pedestrian captured from different cameras. Recently, deep learning has been widely used for feature extraction and distance metric learning in re-ID. However, most of them only consider a certain aspect of the input data and thus will make certain mistakes during the testing process. In this paper, group decision-making (GDM) theory is introduced for comprehensive decision. Furthermore, a novel GDM network (GDMN) is proposed which consists of two sub-networks. First, proposal generation network can generate proposals based on baseline networks for the following decision-making process. Then, decision evaluation network evaluates all the proposals and makes the comprehensive decision. The proposed GDMN can analyze the merits and drawbacks of existing methods and make a better decision. The experimental results on public re-ID benchmarks show that our approach significantly improves the performance of the baseline methods and achieves competitive results compared with other state-of-the-art methods.