Many multi-object tracking (MOT) methods follow the framework of "tracking by detection", which associates the target objects-of-interest based on the detection results. However, due to the separate models for detection and association, the tracking results are not optimal.Moreover, the speed is limited by some cumbersome association methods to achieve high tracking performance. In this work, we propose an end-to-end MOT method, with a Gaussian filter-inspired dynamic search region refinement module to dynamically filter and refine the search region by considering both the template information from the past frames and the detection results from the current frame with little computational burden, and a lightweight attention-based tracking head to achieve the effective fine-grained instance association. Extensive experiments and ablation study on MOT17 and MOT20 datasets demonstrate that our method can achieve the state-of-the-art performance with reasonable speed.
@article{arxiv.2203.10729,
title = {DSRRTracker: Dynamic Search Region Refinement for Attention-based Siamese Multi-Object Tracking},
author = {JiaXu Wan and Hong Zhang and Jin Zhang and Yuan Ding and Yifan Yang and Yan Li and Xuliang Li},
journal= {arXiv preprint arXiv:2203.10729},
year = {2023}
}
Comments
The paper contained some errors in the legends and visualisations, such as incorrectly using the visualisations of the next generation model we studied. We have rewritten our paper on its next-generation model based on that paper. Since we do not want readers to misunderstand the next-generation paper due to the errors in this preprint paper, we have decided to withdraw this preprint paper