MUST

Paper

The paper " MUST: The First Dataset and Unified Framework for Multispectral UAV Single Object Tracking " has been accepted by CVPR2025.

Dataset

MUST, a large-scale and challenging dataset for Multispectral UAV Single Object Tracking, has been released.

Description:

MUST is the first large-scale and challenging dataset for Multispectral UAV Single Object Tracking, aimed at leveraging the advantages of spectral information to cope with restricted target spatial features. Addressing the data requirements of contemporary deep learning models, this dataset comprises 250 video sequences, totaling 43000 frames. Each frame contains 8 bands covering from visible light to near-infrared bands, with a spatial resolution of up to 1200×900 pixels. Furthermore, the dataset incorporates 12 key challenge attributes, making it highly representative of real world challenges encountered in practical applications.

Download:

Download link: https://pan.baidu.com/s/1TcuO5Xb0NGkgl4GCaJWiAA?pwd=must

Citation

If you use this benchmark in your research, please cite this project.

License

This repository contains two components with different licenses:

Our code is released under the .

The HRSSD dataset is licensed under . It is intended for academic research only. You must attribute the original source, and you are not allowed to modify or redistribute the dataset without permission.