- Install and initialize Anaconda3
- Clone the
diffimpact
repo from this page. - 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
- From within the
diffimpact
directory, initialize and update the submodule for our ddsp fork with
git submodule init
git submodule update
- Go to our DDSP fork and install it in the pip environment with
cd ddsp
pip install -e .
cd ../
- Install the conda environment as kernel to Jupyter notebooks with
python -m ipykernel install --user --name=diffimpactenv
- Run
jupyter notebook
- Make sure each Jupyter notebook is set to the
diffimpactenv
kernel before running code - See
AnalysisBySynthesis.ipynb
andSourceSeparation.ipynb
for the analysis by synthesis experiment and the source separation experiment, respectively.
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}
}