Comparative Study of Various Losses for Vehicle Re-identification

In this paper, we tackle the problem of vehicle reidentification, which has extensive applications in traffic analysis such as anomaly detection, congestion pricing and tolling. While previous methods extract visual features from the images and then use spatio-temporal regularization to further refine the results, our method focuses on extracting purely visual features from vehicle images and then further employs a re-ranking technique to improve results. We evaluate the proposed pipeline on the VeRi and CityFlow (NVIDIA AI City Challenge 2019) datasets. Experiments show that our pipeline achieves state of the art performance on the VeRi dataset. We also perform extensive analysis on each step of the pipeline and demonstrate how they increase overall performance.