Skip to content

Latest commit

 

History

History
 
 

bytetrack

README.md

ByteTrack: Multi-Object Tracking by Associating Every Detection Box

Abstract

Multi-object tracking (MOT) aims at estimating bounding boxes and identities of objects in videos. Most methods obtain identities by associating detection boxes whose scores are higher than a threshold. The objects with low detection scores, e.g. occluded objects, are simply thrown away, which brings non-negligible true object missing and fragmented trajectories. To solve this problem, we present a simple, effective and generic association method, tracking by associating every detection box instead of only the high score ones. For the low score detection boxes, we utilize their similarities with tracklets to recover true objects and filter out the background detections. When applied to 9 different state-of-the-art trackers, our method achieves consistent improvement on IDF1 score ranging from 1 to 10 points. To put forwards the state-of-the-art performance of MOT, we design a simple and strong tracker, named ByteTrack. For the first time, we achieve 80.3 MOTA, 77.3 IDF1 and 63.1 HOTA on the test set of MOT17 with 30 FPS running speed on a single V100 GPU.

Citation

@inproceedings{zhang2021bytetrack,
  title={ByteTrack: Multi-Object Tracking by Associating Every Detection Box},
  author={Zhang, Yifu and Sun, Peize and Jiang, Yi and Yu, Dongdong and Yuan, Zehuan and Luo, Ping and Liu, Wenyu and Wang, Xinggang},
  journal={arXiv preprint arXiv:2110.06864},
  year={2021}
}

Results and models on MOT17

Please note that the performance on MOT17-half-val is comparable with the performance reported in the manuscript, while the performance on MOT17-test is lower than the performance reported in the manuscript.

The reason is that ByteTrack tunes customized hyper-parameters (e.g., image resolution and the high threshold of detection score) for each video in MOT17-test set, while we use unified parameters.

Method Detector Train Set Test Set Public Inf time (fps) HOTA MOTA IDF1 FP FN IDSw. Config Download
ByteTrack YOLOX-X CrowdHuman + half-train half-val N - - 78.6 79.2 12909 21024 666 config model | log
ByteTrack YOLOX-X CrowdHuman + half-train test N - 61.7 78.1 74.8 36705 85032 2049 config model | log