22109095 / SimOWT

This repository is an official implementation of the paper A Simple Baseline for Open-World Tracking via Self-training.

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

1. Installation: Please refer to install.md for more details.
2. Data preparation: Please refer to data.md for more details.
3. 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.