Progressive Data Mining and Adaptive Weighted Multi-Model Ensemble for Vehicle Re-Identification

Yongli Sun, Wenpeng Li, Hua Wei, Longtao Zhang, Jiahao Tian, Guangze Sun, Gang Wang, Junliang Cao, Zhifeng Zhao, Junfeng Ding; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 4201-4206

Abstract

In this paper, we introduce our solution to the vehicle re-identification (vehicle ReID) track2 in AI City Challenge 2021. As the key point of intelligent Traffic System, vehicle ReID has been a challenging task due to the higher intra-class and inter-class errors which are owing to variable vehicle orientation, camera and lighting. To reduce this error, at first, we innovatively propose a progressive data mining method to obtain more valid data from testing set. Then, we use the image to the mean of each tracklet method in the matching stage which can ensure the precision of image matching by reducing the error with the information of tracklets. Besides, we propose an adaptive weighted ensemble method which effectively improve the model capability. Finally, our method achieves 0.6533 in the mAP score which yields 4th place in the competition.

Related Material

[pdf]
[bibtex]
@InProceedings{Sun_2021_CVPR, author = {Sun, Yongli and Li, Wenpeng and Wei, Hua and Zhang, Longtao and Tian, Jiahao and Sun, Guangze and Wang, Gang and Cao, Junliang and Zhao, Zhifeng and Ding, Junfeng}, title = {Progressive Data Mining and Adaptive Weighted Multi-Model Ensemble for Vehicle Re-Identification}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2021}, pages = {4201-4206} }