Instructions for running DiffImpact

1. Install and initialize Anaconda3
2. Clone the diffimpact repo from this page.
3. In the Anaconda base environment, change the working directory to this repo's folder, install the conda env, and activate it with

cd diffimpact
conda env create -f diffimpactenv.yml
conda activate diffimpactenv

4. From within the diffimpact directory, initialize and update the submodule for our ddsp fork with

git submodule init
git submodule update

5. Go to our DDSP fork and install it in the pip environment with

cd ddsp
pip install -e .
cd ../

6. Install the conda environment as kernel to Jupyter notebooks with

python -m ipykernel install --user --name=diffimpactenv

7. Run

jupyter notebook

8. Make sure each Jupyter notebook is set to the diffimpactenv kernel before running code
9. See AnalysisBySynthesis.ipynb and SourceSeparation.ipynb for the analysis by synthesis experiment and the source separation experiment, respectively.

Paper

Please check out our paper, "DiffImpact: Differentiable Rendering and Identification of Impact Sounds", published at the Conference on Robot Learning (CoRL), 2021.

If you use this code, please cite us:

  @inproceedings{clarke2022diffimpact,
	  title={Diffimpact: Differentiable rendering and identification of impact sounds},
	  author={Clarke, Samuel and Heravi, Negin and Rau, Mark and Gao, Ruohan and Wu, Jiajun and James, Doug and Bohg, Jeannette},
	  booktitle={Conference on Robot Learning},
	  pages={662--673},
	  year={2022},
	  organization={PMLR}
}