IMPRINT

Code for IMPRINT: Interactional Dynamics-aware Multi-agent Motion Prediction using Multimodal Context.

Instruction

Our implementation requires download skeleton and rgb data for here. In addition, we suggest extracting rgb features separately for efficient training. We have provided a feature_extract.py script. Replace the path to video file and the path to store extracted features.

Training IMPRINT

./run_ntu.sh

If you find this code useful, and want to cite our paper, please use the following bibtex:

@article{yasar2023imprint,
  title={IMPRINT: Interactional Dynamics-aware Motion Prediction in Teams using Multimodal Context},
  author={Yasar, Mohammad Samin and Islam, Md Mofijul and Iqbal, Tariq},
  journal={ACM Transactions on Human-Robot Interaction},
  year={2023}
}