SimOWT: A Simple Baseline for Open-World Tracking via Self-training
This repository is the project page for the paper A Simple Baseline for Open-World Tracking via Self-training.
Highlight
- SimOWT is accepted to ACMMM 2023.
Overview
We propose SimOWT, a simple baseline for Open-World Tracking(OWT). Our method demonstrates state-of-the-art result on the TAO-OW benchmark.
Demo
simowt.mp4
Results
Getting started
- Installation: Please refer to install.md for more details.
- Data preparation: Please refer to data.md for more details.
- Training, testing and Model zoo: Please refer to train&test&model_zoo.md for more details.
Citing SimOWT
If you find SimOWT useful in your research, please consider citing:
@inproceedings{10.1145/3581783.3611695,
author = {Wang, Bingyang and Li, Tanlin and Wu, Jiannan and Jiang, Yi and Lu, Huchuan and He, You},
title = {A Simple Baseline for Open-World Tracking via Self-Training},
booktitle = {Proceedings of the 31st ACM International Conference on Multimedia},
year={2023}
}
Acknowledgments
- Thanks IDOL for providing the strong baseline for Multi-Object Tracking.