WebUAV-3M: A Benchmark for Unveiling the Power of Million-Scale Deep UAV Tracking [ArXiv][IEEE Xplore]
Unmanned aerial vehicle (UAV) tracking is of great significance for a wide range of applications, such as delivery and agriculture. Previous benchmarks in this area mainly focused on small-scale tracking problems while ignoring the amounts of data, types of data modalities, diversities of target categories and scenarios, and evaluation protocols involved, greatly hiding the massive power of deep UAV tracking. In this work, we propose WebUAV-3M, the largest public UAV tracking benchmark to date, to facilitate both the development and evaluation of deep UAV trackers. WebUAV-3M contains over 3.3 million frames across 4,500 videos and offers 223 highly diverse target categories. Each video is densely annotated with bounding boxes by an efficient and scalable semi-automatic target annotation (SATA) pipeline. Importantly, to take advantage of the complementary superiority of language and audio, we enrich WebUAV-3M by innovatively providing both natural language specifications and audio descriptions. We believe that such additions will greatly boost future research in terms of exploring language features and audio cues for multi-modal UAV tracking. In addition, a fine-grained UAV tracking-under-scenario constraint (UTUSC) evaluation protocol and seven challenging scenario subtest sets are constructed to enable the community to develop, adapt and evaluate various types of advanced trackers. We provide extensive evaluations and detailed analyses of 43 representative trackers and envision future research directions in the field of deep UAV tracking and beyond. The dataset, toolkits and baseline results are available at this page.
- Omni-Modality: visual bounding box, language, and audio annotations
- Large-Scale: 4500 videos, over 3 million densely annotated frames
- Diverse Categories: 12 superclasses, more than 200 target classes, over 60 motion classes
- Efficient Labeling: semi-automatic target annotation pipeline
- Rigorous Evaluation: UTUSC protocol
- Multiple-Baseline: 43 representative trackers
- Unified Dataset: train, validation, and test sets
- Multi-Task Covered: nighttime tracking, adversarial examples, multimodal tracking, data imbalances
- Mar. 25, 2024: The audio annotations V1.0, superclass, object classes and motion classes released.
- Dec. 31, 2022: The journal paper of WebUAV-3M accepted by IEEE TPAMI was available.
- Dec. 31, 2022: The Language Annotations V1.0 released.
- Aug. 9, 2022: The latest version of the paper published.
- Aug. 1, 2022: WebUAV-3M Evaluation Toolkits V1.0 released.
- Aug. 1, 2022: Baseline Results released.
- Aug. 1, 2022: The WebUAV-3M dataset V1.0 released.
- Evaluation Toolkits
- Video Sequences of WebUAV-3M Dataset
- Baseline Results
- Language Annotations
- Audio Annotations
- Superclass, Object Classes and Motion Classes
The WebUAV-3M dataset contains 4500 videos, divided into three sets (Train/Val/Test)
The dataset download and file organization process are as follows:
-
To download the dataset via Baidu Pan or Google Drive, please complete a Google Form first (require a VPN in Chinese Mainland), then a download link will be automatically sent to your email.
-
Download the whole dataset through Baidu Pan, the extraction code is UAV3.
-
Download the whole dataset through Google Drive.
-
Check the number of videos in each set.
-
The Train set should include 3520 videos (621G)
-
The Val set should include 200 videos (28G)
-
The Test set should include 780 videos (170G)
-
-
We also provide the adversarial examples sub-set (WebUAV-3M-AE) to evaluate the robustness of trackers (optional).
- The WebUAV-3M-AE should include 100 (clean) + 500 (with adversarial examples) videos (186G)
-
We plan to release the audio annotations soon. Stay tuned.
-
Based on users’ feedback and needs, we plan to develop and update this dataset gradually.
-
Run the unzipping script, and delete the script after decompression.
bash UnzipWebUAV3M-Train.sh
bash UnzipWebUAV3M-Val.sh
bash UnzipWebUAV3M-Test.sh
bash UnzipWebUAV3M-AE.sh (optional)
For Overall, Attribute, Accuracy and UTUSC Protocol evaluations in OPE using Pre, nPre, AUC, cAUC and mAcc metrics:
# Step1. Run experiments on dataset
# Step2. Put the results in WebUAV-3M_Evaluation_Toolkit/results/Baseline_Results
# Step3. Report tracking performance
python WebUAV-3M_Overall_Evaluation.py
python WebUAV-3M_Attribute_Evaluation.py
python WebUAV-3M_Accuracy_Evaluation.py
python WebUAV-3M_UTUSC_Protocol.py
Precision plot | Normalized precision plot |
---|---|
![]() |
![]() |
Success plot | Complete success plot |
![]() |
![]() |
The experiments are implemented using PyTorch or MATLAB with an Intel (R) Xeon (R) Gold 6230R CPU @ 2.10GHz and three NVIDIA RTX A5000 GPUs on an Ubuntu 18.04 server.
If you find the dataset and toolkits useful in your research, please consider citing:
@ARTICLE{10004511,
author={Zhang, Chunhui and Huang, Guanjie and Liu, Li and Huang, Shan and Yang, Yinan and Wan, Xiang and Ge, Shiming and Tao, Dacheng},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={WebUAV-3M: A Benchmark for Unveiling the Power of Million-Scale Deep UAV Tracking},
year={2023},
volume={45},
number={7},
pages={9186-9205},
doi={10.1109/TPAMI.2022.3232854}
}
Thanks for the great [GOT-10k toolkit]
Feedbacks and comments are welcome! Feel free to contact us via [email protected] or [email protected] or [email protected].