---
license: other
tags:
  - visual-object-tracking
  - rgbt-tracking
  - pytorch
---

# ESMTrack

Model weights for **ESMTrack**.

- **Code**: https://github.com/LiShenglana/ESMTrack
- **Paper**:https://arxiv.org/abs/2609.37162

## Files

| File | Benchmark |
|:--|:--|
| `checkpoints/LasHeR_best_checkpoint.pth` | LasHeR |
| `checkpoints/VTUAV_best_checkpoint.pth` | VTUAV |
| `checkpoints/GTOT_best_checkpoint.pth` | GTOT |
| `checkpoints/RGBT210_best_checkpoint.pth` | RGBT210 |
| `checkpoints/RGBT234_best_checkpoint.pth` | RGBT234 |

## Usage

Download the checkpoints into the project root:

```bash
hf download ShenglanLiaaa/ESMTrack --include "checkpoints/*" --local-dir .
```

Then test with the checkpoint of the corresponding benchmark, e.g. LasHeR:

```bash
python tracking/test.py --tracker_name esmtrack --tracker_param dropmae_256_150ep --checkpoint checkpoints/LasHeR_best_checkpoint.pth --dataset_name lasher --threads 8 --num_gpus 2
```

Pretrained backbone weights (e.g. DropMAE `dropmae_k700_800E.pth`) are not redistributed here; please download them from the [DropMAE authors](https://github.com/jimmy-dq/DropMAE) and put them under `pretrained_networks/` (see the GitHub README).

If you find our work useful in your research, please consider citing:
```
@inproceedings{ESMTrack2026,
title={End-to-End Self-Supervised RGB-T Tracking without Modality Misleading},
author={Shenglan Li and Rui Yao and Kunyang Sun and Hong Jia and Yong Zhou and Javen Qinfeng Shi and Xinyu Zhang},
year={2026},
eprint={2609.37162},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2609.37162}
}
```