README.md

---
license: apache-2.0
tags:
  - object-detection
  - multi-object-tracking
  - yolox
  - sports-tracking
  - pedestrian-tracking
datasets:
  - MOT17
  - SportsMOT
  - DanceTrack
  - VisDrone
pipeline_tag: object-detection
---

# YOLOX Detection Models for Multi-Object Tracking

A collection of fine-tuned **YOLOX** detector weights used as the detection backbone in [BoxMOT](https://github.com/mikel-brostrom/boxmot) tracking pipelines.

## Available Models

| File | Fine-tuned On | Evaluated On | MOTA | IDF1 | HOTA | IDs | FPS | Experiment |
|------|--------------|--------------|------|------|------|-----|-----|------------|
| `yolox_x_MOT17_ablation.pt` | CrowdHuman + MOT17 train first half | MOT17 half val | 76.6 | 79.3 | — | 159 | 29.6 | [`yolox_x_ablation.py`](https://github.com/ifzhang/ByteTrack/blob/main/exps/example/mot/yolox_x_ablation.py) |
| `yolox_x_MOT17_test.pt` | CrowdHuman + MOT17 full train + Cityperson + ETHZ | MOT17 train | 90.0 | 83.3 | — | 422 | 29.6 | [`yolox_x_mix_det.py`](https://github.com/ifzhang/ByteTrack/blob/main/exps/example/mot/yolox_x_mix_det.py) |
| `yolox_x_MOT20_ablation.pt` | CrowdHuman + MOT20 train first half + Widerperson | — | — | — | — | — | — | [`yolox_x_dance_val.py`](https://github.com/KyujinShim/TrackTrack/blob/main/YOLOX/exps/yolox_x_dance_val.py) |
| `yolox_x_MOT20_test.pt` | CrowdHuman + MOT20 full train | MOT20 train | 93.4 | 89.3 | — | 1057 | 17.5 | [`yolox_x_mix_det.py`](https://github.com/ifzhang/ByteTrack/blob/main/exps/example/mot/yolox_x_mix_det.py) |
| `yolox_x_sportsmot.pt` | SportsMOT train + val | SportsMOT test | 96.3 | 79.8 | 77.2 | — | — | [`yolox_x_ch_sportsmot.py`](https://github.com/hsiangwei0903/Deep-EIoU/blob/main/yolox/yolox_x_ch_sportsmot.py) |
| `yolox_x_dancetrack.pt` | DanceTrack full train + CrowdHuman + Widerperson | DanceTrack test | 93.6 | 67.8 | 66.5 | — | — | [`yolox_x_dance_test.py`](https://github.com/KyujinShim/TrackTrack/blob/main/YOLOX/exps/yolox_x_dance_test.py) |
| `yolox_x_visdrone.pt` | VisDrone-MOT trainval | VisDrone-MOT test-dev | 52.3 | 69.0 | — | 1052 | 19.4 | [`yolox_x_u2mot_visdrone.py`](https://github.com/alibaba/u2mot/blob/main/exps/example/u2mot/yolox_x_u2mot_visdrone.py) |

All models are YOLOX-X (`depth=1.33`, `width=1.25`), input size 800×1440 (except VisDrone: 896×1600), single pedestrian class (except VisDrone: 10 classes). Size ~756 MB (VisDrone: 806 MB).

## Origin & Citation

The DanceTrack and MOT20 YOLOX-X weights were trained following the procedure described in:

> **Focusing on Tracks for Online Multi-Object Tracking**
> Kyujin Shim, Kangwook Ko, Yujin Yang, Changick Kim
> *Korea Advanced Institute of Science and Technology (KAIST)*
> CVPR 2025

The SportsMOT YOLOX-X weights were trained following the procedure described in:

> **Iterative Scale-Up ExpansionIoU and Deep Features Association for Multi-Object Tracking in Sports**
> Hsiang-Wei Huang, Cheng-Yen Yang, Jiacheng Sun, Pyong-Kun Kim, Kwang-Ju Kim, Kyoungoh Lee, Chung-I Huang, Jenq-Neng Hwang
> *Information Processing Lab, University of Washington; ETRI; National Center for High-Performance Computing*
> [arXiv:2306.13074](https://arxiv.org/abs/2306.13074) (2023)

The VisDrone YOLOX-X weights were trained following the procedure described in:

> **Uncertainty-aware Unsupervised Multi-Object Tracking**
> Kai Liu, Sheng Jin, Zhihang Fu, Ze Chen, Rongxin Jiang, Jieping Ye
> *Zhejiang University; Alibaba DAMO Academy*
> [arXiv:2307.15409](https://arxiv.org/abs/2307.15409) (2023)

The MOT17 YOLOX-X weights follow the ByteTrack-style training recipe originally proposed in:

> **ByteTrack: Multi-Object Tracking by Associating Every Detection Box**
> Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Zehuan Yuan, Ping Luo, Wenyu Liu, Xinggang Wang
> [arXiv:2110.06864](https://arxiv.org/abs/2110.06864) (2021)

## Usage with BoxMOT

These models are automatically downloaded when running BoxMOT benchmarks:

```bash
# MOT17 evaluation (auto-downloads yolox_x_MOT17_ablation.pt)
boxmot eval --benchmark mot17 --split ablation --tracker boosttrack

# SportsMOT evaluation (auto-downloads yolox_x_sportsmot.pt)
boxmot eval --benchmark sportsmot --split ablation --tracker boosttrack
```

Or use directly in Python:

```python
from boxmot import Boxmot

# SportsMOT with YOLOX-X detector
tracker = Boxmot(detector="yolox_x_sportsmot", tracker="boosttrack")
results = tracker.track(source="path/to/video.mp4")
```

## License

The YOLOX architecture is released under the [Apache 2.0 License](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/LICENSE).
Weights are provided for research purposes. Please cite the original papers when using these models.

```bibtex
@inproceedings{shim2025tracktrack,
  title={Focusing on Tracks for Online Multi-Object Tracking},
  author={Shim, Kyujin and Ko, Kangwook and Yang, Yujin and Kim, Changick},
  booktitle={CVPR},
  year={2025}
}

@article{huang2023deep,
  title={Iterative Scale-Up ExpansionIoU and Deep Features Association for Multi-Object Tracking in Sports},
  author={Huang, Hsiang-Wei and Yang, Cheng-Yen and Sun, Jiacheng and Kim, Pyong-Kun and Kim, Kwang-Ju and Lee, Kyoungoh and Huang, Chung-I and Hwang, Jenq-Neng},
  journal={arXiv preprint arXiv:2306.13074},
  year={2023}
}

@article{liu2023u2mot,
  title={Uncertainty-aware Unsupervised Multi-Object Tracking},
  author={Liu, Kai and Jin, Sheng and Fu, Zhihang and Chen, Ze and Jiang, Rongxin and Ye, Jieping},
  journal={arXiv preprint arXiv:2307.15409},
  year={2023}
}

@article{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}
}

@article{ge2021yolox,
  title={YOLOX: Exceeding YOLO Series in 2021},
  author={Ge, Zheng and Liu, Songtao and Wang, Feng and Li, Zeming and Sun, Jian},
  journal={arXiv preprint arXiv:2107.08430},
  year={2021}
}
```