Authors: David R. Wessels*, David M. Knigge*, Riccardo Valperga, Samuele Papa, Sharvaree Vadgama, Efstratios Gavves, Erik J. Bekkers
**equal contribution *
This is the reproducibility repo for the paper "Grounding Continuous Representations in Geometry: Equivariant Neural Fields". All experiments in the paper should be reproducible using the code in this repository. Data for the experiments is mostly downloadable via torch-dataloaders. However, the celeb-A dataset should be downloaded via the following https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html. Shapenet data is downloaded via the https://www.shapenet.org/ website.
To install the requirements, we use conda. We recommend creating a new environment for the project.
conda create -n enf-jax python=3.11
conda activate enf-jax
Install the relevant dependencies.
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
pip install --upgrade "jax[cuda12]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
pip install optax flax orbax wandb matplotlib trimesh hydra-core tqdm netCDF4 py-pde
conda install -c conda-forge dedalus
With pyenv, you can install python 3.11 as follows:
pyenv install 3.11.1
pyenv virtualenv 3.11.1 enf-jax
pyenv activate enf-jax
pip install -r requirements.txt
pip install -e .
We list relevant components of the repository here:
enf/
contains the code for the Equivariant Neural Field.enf/steerable_attention/invariant/
contains the code for the bi-invariants used in the experiments.experiments/
contains the code for all experiments in the paper.
We specify commands per experiment in the experiments readme.
If you find this code useful, please consider citing our paper:
@article{wessels2024grounding,
title={Grounding Continuous Representations in Geometry: Equivariant Neural Fields},
author={Wessels, David R and Knigge, David M and Papa, Samuele and Valperga, Riccardo and Vadgama, Sharvaree and Gavves, Efstratios and Bekkers, Erik J},
journal={arXiv preprint arXiv:2406.05753},
year={2024}
}