AnimalTrack: A Benchmark for Multi-Animal Tracking in the Wild

Libo Zhang*        Junyuan Gao*        Zhen Xiao        Heng Fan
*Equal contribution        Corresponding author

Figure 1: Several sample videos in AnimalTrack (Note, the colors of boxes for each target are randomly generated).


  • 11/2022: All videos have been uploaded. Welcome to use AnimalTrack.
  • 11/2022: Our AnimalTrack (pdf) is accepted to International Journal of Computer Vision (IJCV).
  • 11/2022: The webpage of AnimalTrack has been launched.

About AnimalTrack

AnimalTrack aims to provide the first dedicated platform for studying the problem of multi-animal tracking (MAT) that is important for animal motion and behavior analysis. The proposed AnimalTrack has many features:
  • Dedicated MAT Platform: 58 sequences with around 25K frames provided in AnimalTrack.
  • Accurate Annotations: Each frame is manual annotated with careful inspection.
  • Dense Trajectories: The average number of tracks in AnimalTrack is more than 30.
  • Diverse Classes: We collect videos from ten animal categories in the wild.
Please kindly check out the benchmark details and download links at the Dataset Download page, and evaluation results at the Evaluation page.


If you use our AnimalTrack for research, please cite our work :)

AnimalTrack: A Benchmark for Multi-Animal Tracking in the Wild
Libo Zhang*, Junyuan Gao*, Zhen Xiao, and Heng Fan (*Equal contribution)
International Journal of Computer Vision (IJCV), 131: 496-513, 2023. (accepted)


The annotations of our AnimalTrack are licensed under a Creative Commons Attribution 4.0 License. We offer the AnimalTrack for non-commercial research purposes only. All data are obtained from Internet which are not the property of CAS or UNT. These organizations are not responsible for the content of these videos.


If you have any questions, please contact Junyuan Gao at