Predicting Long-Term Skeletal Motions by a Spatio-Temporal Hierarchical Recurrent Network

This is the pytorch version, python code of our paper Predicting Long-Term Skeletal Motions by a Spatio-Temporal Hierarchical Recurrent Network.
Please follow the introduction below to reproduce our results .

Publication

Our paper is accepted by the 24th European Conference on Artificial Intelligence (ECAI2020) and the submitted version is available at arXiv. We will release the final version later.

See the accepted papers here.

See the published version here

Required packages

• Anaconda is highly recommend

• Pytorch >= 1.2

• Matplotlib = 3.0.1

• FFMPEG

  Note that any Matplotlib version > 3.0.1 cannot guarantee the correct operation of the program due to some compatibility issues.

Dataset download

• H3.6m dataset

  cd src
  sh ./data/h3.6m/download_h3.6m.sh

• Mouse dataset

  cd src
  sh ./data/Mouse/download_mouse.sh

Reproduce our results

We save our model in the checkpoint folder. Our code will search a checkpoint automatically according to your settings.

cd src
python train.py --dataset Human --training False --visualize True

Train our network

The main file can be found in train.py.

cd src
python train.py

This command will train our network with default settings, i.e. Human dataset and all actions.

All settings are listed below:

For more detail configurations, you could refer config.py

Acknowledge

Some code is adopted or modified from BII-wushuang, una-dinosauria, and asheshjain399.

Thanks for their great works! :)

Citation

If you find this useful, please cite our work as follows:

Not available.

Questions

If you have any question or bug in my code, you can contact me with issues or email: hjf@cqu.edu.cn