A multi-scale network with multi-view correlation for vehicle re-identification

A multi-scale network with multi-view correlation for vehicle re-identification

연구 분야: Software Development

학회: Multimedia Systems

Vehicle Re-identification (Re-ID) aims to identify the same vehicle across different cameras, which plays an important role in modern traffic management systems. Vehicle Re-ID faces two main challenges: (1) Intra-instance discrepancy. The appearance of the same vehicle can vary greatly with different viewpoints. (2) Inter-instance similarity. Different vehicles of the same model often have similar appearances. To address these two challenges, we propose a Multi-Scale Network with Multi-View Correlation (MSMV-Net). Firstly, the network is divided into four branches to extract different and effective features from global, local, and viewpoint perspectives. Secondly, a multi-scale feature extraction module is designed based on the characteristics of the vehicle Re-ID task. Different scales of local features are extracted using a local partition strategy, and the scale size can be adjusted across different local branches to obtain features at multiple granularities. Finally, a multi-view correlation module is proposed to associate the features from different viewpoints using the perspective information extracted by the network. Through metric learning with vector space constraints, the network is guided to learn the shared features of vehicles from different viewpoints. Extensive experiments on three large-scale datasets demonstrate that our method is superior to state-of-the-art approaches. The code is publicly available at https://github.com/zgywsh/MSMV-Net.git.