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 .
Paper and revisions
Our paper is available at arXiv
Based on the version V1 at arXiv, we have several revisions.
Redrew Figure 4:
Modified the caption of Figure 4:
exchange the locations of the word 'orange' and 'blue'.
Anaconda is highly recommend
Pytorch >= 1.2
Matplotlib = 3.0.1
Note that any Matplotlib version > 3.0.1 cannot guarantee the correct operation of the program due to some compatibility issues.
cd src sh ./data/h3.6m/download_h3.6m.sh
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:
|--gpu||||[.., .., ..]||GPU device ids, list|
|--training||True||True, False||train or test|
|--action||all||all, walking, ....||see more in the code|
|--dataset||Human||Human, Mouse||choose dataset|
|--visualize||False||True, False||visualize predictions or not, only usable for testing|
For more detail configurations, you could refer config.py
Thanks for their great workings! :)
If you find this useful, please cite our work as follows:
If you have any question or bug in my code, you can contact me with issues or email: email@example.com