The Baseline Codes for the paper "The MI-Motion Dataset and Benchmark for 3D Multi-Person Motion Prediction"

Xiaogang Peng, Xiao Zhou, Yikai Luo, Hao Wen, Zizhao Wu*

Paper | Video | Data | Project Page

News

• [2022/06/01]: The codes of SocialTGCN and other baselines are released.

Pipeline of SocialTGCN

Getting Started

The MI-Motion dataset can be downloaded from and . You can also download the pretrained models of all the baselines in . More details could be found in the Project Page.

Prepare Dataset

After download the dataset, please prepare your data like this:

your_project_folder/
├── data/
│   ├── MI-Motion
│   │   ├── S0
│   │   ├── S1
│   │   ├── S2
│   │   ├── S3
│   │   ├── S4
│   ├── ├── ...
│   ├── preprocess_data.py
│   ├── ...
├── baselines/
│   ├── ...
├── util/
│   ├── ...

Preprocsess Data

cd data
python preprocess_data.py

Training

For any baseline method:

python baselines/train_{method}.py
python baselines/train_hri.py   #  example of training for HRI baseline

Evaluation

For any baseline method:

python baselines/eval_{method}.py

If you want evaluation for ultra-long-term prediction, use:

python baselines/eval_{method}.py --ultra-long 1

Visualization

python baselines/eval_{method}.py --vis 1  # for short-term and long term prediction
python baselines/eval_{method}.py --vis 1 --ultra-long 1 # for ultra-long-term prediction 

The rendered PNGs and GIFs are automatically saved in output folder of each baseline.

Acknowledgement

Many thanks to the previous works:

• HRI
• MRT
• TBIFormer

Bibtex

If this work is helpful for your research, please consider citing the following BibTeX entry.

@misc{xx,
      title={The MI-Motion Dataset and Benchmark for 3D Multi-Person Motion Prediction}, 
      author={Xiaogang Peng and Xiao Zhou and Yikai Luo and Hao Wen and Zizhao Wu},
      year={2023},
      eprint={xx.xx},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

License

MIT