WASB: Widely Applicable Strong Baseline for Sports Ball Detection and Tracking

Code & dataset repository for the paper: Widely Applicable Strong Baseline for Sports Ball Detection and Tracking

Shuhei Tarashima, Muhammad Abdul Haq, Yushan Wang, Norio Tagawa

We present Widely Applicable Strong Baseline (WASB), a Sports Ball Detection and Tracking (SBDT) baseline that can be applied to wide range of sports categories ⚽ 🎾 🏸 🏐 🏀 .

teaser.mp4

News

• [11/23/2023] Our BMVC2023 proceeding is available! Thank you, BMVC2023 organizers!
• [11/23/2023] Evaluation codes of DeepBall, DeepBall-Large and BallSeg are added!
• [11/21/2023] Evaluation codes of TrackNetV2, ResTrackNetV2 and MonoTrack are added!
• [11/17/2023] Repository is released. Now it contains evaluation codes of pretrained WASB models only. Other models will be coming soon!
• [11/09/2023] Our arXiv preprint is released.

Installation and Setup

Tested with Python3.8, CUDA11.3 on Ubuntu 18.04 (4 V100 GPUs inside). We recommend to use the Dockerfile provided in this repo (with -it option when running the container).

• See GET_STARTED.md for how to get started with SBDT models.
• See MODEL_ZOO.md for available model weights.

Citation

If you find this work useful, please consider to cite our paper:

@inproceedings{tarashima2023wasb,
	title={Widely Applicable Strong Baseline for Sports Ball Detection and Tracking},
	author={Tarashima, Shuhei and Haq, Muhammad Abdul and Wang, Yushan and Tagawa, Norio},
	booktitle={BMVC},
	year={2023}
}