This is the code for the paper
Wei Mao, Miaomiao Liu, Richard Hartley, Mathieu Salzmann. Contact-aware Human Motion Forecasting. In NeurIPS 22.
tested on pytorch == 1.8.1
For the original dataset please contact the authors of Long-term Human Motion Prediction with Scene Context.
python process_gta_dataset.py
The file structure of data folder should look like below.
data
├── GTA-IM-Dataset
│ ├── 2020-06-03-13-31-46
│ │ ├── 0000.jpg
│ │ ├── 0000.png
│ │ ├── 0000_id.png
│ │ ......
│ ├── 2020-06-04-23-14-02
│ ......
│
├── data_v2_downsample0.02
│ ├── 2020-06-04-23-14-02_r013_sf0.npz
│ ├── 2020-06-04-23-08-16_r013_sf0.npz
│ └── 2020-06-04-23-08-16_r013_sf1.npz
└──
cfg/
. These configs correspond to pretrained models in results
.run.sh
file.results
folder.Run the following code to visualize the results.
python eval_gta_vis.py
The interface should look like this
If you use our code, please cite our work
@inproceedings{mao2022contact,
title={Contact-aware Human Motion Forecasting},
author={Mao, Wei and Liu, Miaomiao and Hartley, Richard and Salzmann, Mathieu},
journal={NeurIPS},
year={2022}
}
The overall code framework (dataloading, training, testing etc.) is adapted from DLow.
The 3D scene encoding framework (code in ./pcvnn/
) is adapted from PVCNN paper
Part of the visualization code and the data processing code is from GTA-IM dataset.
MIT