Skip to content

Latest commit

 

History

History
50 lines (35 loc) · 1.84 KB

README.md

File metadata and controls

50 lines (35 loc) · 1.84 KB

SportsHHI

This is the official repository of the CVPR 2024 paper SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos.

Download

SportsHHI Dataset

Baseline Method & Evaluation

Environment setup

conda create --name sports_hhi python=3.8 -y
conda activate sports_hhi
conda install pytorch==1.12.1 torchvision==0.13.1 cudatoolkit=10.2 -c pytorch  # **This** command will automatically install the latest version PyTorch and cudatoolkit, please check whether they match your environment.
pip install -U openmim
mim install mmengine
mim install mmcv==2.0.0
mim install mmdet
pip install einops
pip install numpy==1.23.5

Training & Testing

We provide the training and testing code for our baseline method. Please first specify data_dir and work_dir in configuration files in configs folder.

For training, run bash train.sh $CONFIG $GPU_NUM. $CONFIG should be a configuration file in configs folder. $GPU_NUM should be the number of gpus for training.

For testing, run bash test.sh $CONFIG $CHECKPOINT $GPU_NUM. $CONFIG should be a configuration file in configs folder. $CHECKPOINT is the checkpoint to test. $GPU_NUM should be the number of gpus for training.

Acknowledgments

Our implementation of baseline method is developed based on the mmaction2 repository.

Citation

If you find our code or paper useful, please cite as

@misc{wu2024sportshhi,
      title={SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos}, 
      author={Tao Wu and Runyu He and Gangshan Wu and Limin Wang},
      year={2024},
      eprint={2404.04565},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}