The Wusi Basketball Training Dataset is proposed in Social Motion Prediction with Cognitive Hierarchies (NeurIPS 2023). It is a multi-person 3D motion dataset with a special focus on strategic interactions. This new dataset contains 3D skeleton representation of 5 people in highly intense basketball drills recorded at 25 FPS.
Please check our paper and project page for more details.
Requirements:
- Numpy
Please read carefully the license agreement before you download and/or use the Wusi dataset. By downloading and/or using the dataset, you acknowledge that you have read these terms and conditions, understand them, and agree to be bound by them.
-
We provide the processed training and testing split in numpy (sequence, people, frame, keypoints).
training.npy: (7125, 5, 50, 45)testing.npy: (1782, 5, 50, 45)
-
If you wish to handle the full motion sequences and process them on your own, please download the
Wusifolder todata/and check the processing guidance below. -
If you wish to fit the 3D poses to SMPL for better visual effect, please refer to this repo.
The script (process_wusi.py) reads the undivided data and cut them up into sequences, you can specify the sequence length by argument sequence_len and stride by the argument stride. After processing the undivided data, it shall tell you the sequence number and sequence length. Then, processed data will be divided into training and testing set. You could specify the ratio of training dataset by argument ratio, the remaining should automatically become test set.
cd data
python precess_wusi.py --stride=[your stride] --sequence_len=[your sequence len] --ratio=[your ratio]If you use this dataset, please cite the corresponding NeurIPS 2023 paper:
@inproceedings{zhu2023social,
title={Social Motion Prediction with Cognitive Hierarchies},
author={Zhu, Wentao and Qin, Jason and Lou, Yuke and Ye, Hang and Ma, Xiaoxuan and Ci, Hai and Wang, Yizhou},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023}
}