Vehicle Re-Identification: Pushing the limits of re-identification

Abner Ayala-Acevedo, Akash Devgun, Sadri Zahir, Sid Askary; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 291-296

In this paper, we present a series of techniques which help push the limits of vehicle re-identification. First, we establish a strong baseline by using one of the best person re-identification models and applying them to vehicle re-identification. Secondly, we show improvements in four key components of re-identification: 1) detection, 2) tracking, 3) model, 4) loss function. Finally, our improvements lead to the state-of-the-art in the vehicle re-identification dataset VeRi-776, with 85.20 mean Average Precision (mAP) and 96.60% Rank-1 accuracy. This represents a +17.65 mAP and +6.37 Rank-1 improvement over the literature.