Multi-View Fusion based Attribute Network for Multi-Query Vehicle Re-identification
Weijun Zhang, Renjian Li, Xiaoyi Zhou, Hanqin Shi · 2023
In recent years, most re-identification methods use only a single query image to match the target vehicle. However, the limitation of single query image information hinders the performance of vehicle re-identification. In this case, we propose a Multi-View Fusion based Attribute Network (MVANet) for the Multi-Query Vehicle re-identification task. On the one hand, we fuse the information of multiple vehicle views by the Viewpoint Fusion Module to improve the information representation, thus solving the challenge of large intra-class variability caused by viewpoint differences. On the other hand, we use the Attribute Embedding Module to embed attribute information into visual features to address the challenge of small intra-class variability caused by similarity in appearance. In addition, we construct a new dataset HFVeRi for the Multi-Query Vehicle re-identification task.