This codebase accompanies the paper "Dynamic Neural Relational Inference" from CVPR 2020.

This code was written using the following packages:

• PyTorch 1.2.0
• numpy 1.16.4
• transforms3d 0.3.1 (For Motion Capture data processing)
• pandas (for InD data processing)

To run this code, you should pip install it in editable mode. This can be done using the following command:

pip install -e ./

Scripts train models can be found in the run_scripts directory.

Datasets:

• Motion Capture: the datasets can be downloaded from http://mocap.cs.cmu.edu/search.php?subjectnumber=118 and http://mocap.cs.cmu.edu/search.php?subjectnumber=35. For subject 35, you need trials 1-16 and 28-34. For subject 118, you need trials 1-30.
• Basketball: The original data can be accessed here: https://github.com/ezhan94/multiagent-programmatic-supervision. Some preprocessing has been done to get it used into the form used by this code; for convenience, these files can be found here.
• InD: Data must be requested from here: https://www.ind-dataset.com/
• Synth: this code includes the synth data, as well as code used to generate it.

Attribution: Some portions of this code are based on the code for the paper "Neural Relational Inference for Interacting Systems." This code can be found at https://github.com/ethanfetaya/NRI

If you use this code or this model in your work, please cite us:

@inproceedings{dNRI,
  title={Dynamic Neural Relational Inference},
  author={Graber, Colin and Schwing, Alexander},
  booktitle={The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2020},
}