We release a large-scale benchmark for Visble-thermal UAV Tracking.
- The dataset is available on VTUAV.
- Our paper is accepted by CVPR2022!!
- Three versions (full dataset, RGB split and mask split) are available. Please refer to our project page.
- Large-scale: We collected nearly 1.7 million well-aligned RGB-T image pairs with 500 sequences for unveiling the power of RGB-T tracking(the largest RGB-T tracking benchmark so far).
- High-diversity:13 sub-classes and 15 scenes cross 2 cities.
- Multi-task evaluation: Our benchmark is designed for evaluating both short-term tracking, long-term tracking and tracking with segmentation.
- Hierarchical attribute annotation: Sequence-level attribute annotation for 13 typical challenges. Additionally, we provide frame-level attribute for training challenge-aware trackers.
import gdown
url = "https://drive.google.com/drive/folders/1GwYNPcrkUM-gVDAObxNqERi_2Db7okjP?usp=sharing"
gdown.download_folder(url, quiet=False, use_cookies=False)
MSR | MPR | |
---|---|---|
DAFNet | 45.8 | 62.0 |
ADRNet | 46.6 | 62.2 |
FSRPN | 54.4 | 65.3 |
mfDiMP | 55.4 | 67.3 |
HMFT | 62.7 | 75.8 |
MSR | MPR | |
---|---|---|
ADRNet | 17.5 | 23.5 |
DAFNet | 18.8 | 25.3 |
FSRPN | 27.2 | 31.5 |
mfDiMP | 31.4 | 36.6 |
HMFT | 35.5 | 41.4 |
HMFT_LT | 46.1 | 53.6 |
The sequence-level attribute annotation can be found in BaiduDisk(code:h24u) and GoogleDrive.
The evaluation toolkit can be found in BaiduDisk(code:99j9) and GoogleDrive.
Note: The dataset should be extracted into the same folder.
- Modify the variable "basePath" in GenerateMat_ST.m and GenerateMat_LT.m and move your results into "BB_results" folder
- run GenerateMat_ST.m and GenerateMat_LT.m to generate the report files for short-term and long-term tracking
- If only overall performance is needed, directly run plot_ST.m and plot_LT.m to generate the MSR and MPR curves.
- If both overall and attribute-based performance are needed, change the "attrDisplays" and run plot_ST.m and plot_LT.m to generate the MSR and MPR curves.
- Modify the variable "basePath" in GenerateMat_ST_RGB_only.m and GenerateMat_LT_RGB_only.m and move your results into "BB_results_RGB" folder
- run GenerateMat_ST_RGB_only.m and GenerateMat_LT_RGB_only.m to generate the report files for short-term and long-term tracking
- If only overall performance is needed, directly run plot_ST_RGB_only.m and plot_LT_RGB_only.m to generate the MSR and MPR curves.
- If both overall and attribute-based performance are needed, change the "attrDisplays" and run plot_ST_RGB_only.m and plot_LT_RGB_only.m to generate the MSR and MPR curves.
If you find this benchmark useful, please cite
@InProceedings{Zhang_CVPR22_VTUAV,
author = {Zhang Pengyu and Jie Zhao and Dong Wang and Huchuan Lu and Xiang Ruan},
title = {Visible-Thermal UAV Tracking: A Large-Scale Benchmark and New Baseline},
booktitle = {Proceedings of the IEEE conference on computer vision and pattern recognition},
year = {2022}
}
If you have any questions, feel free to contact me.