It is an official repository of the paper entitled "Datasets and Benchmarks for Nanophotonic Structure and Parametric Design Simulations", which is presented at NeurIPS 2023 Datasets and Benchmarks track.
- nanophotonic_structures: a directory for nanophotonic structures and utils
- scripts: a directory for runnable scripts
- src: a directory for dataset generation, model training, model testing, and structure optimization
- generate_datset.sh: a script for dataset generation
- optimize_models_continuous.sh: a script for nanophotonic structure optimization over continuous spaces
- optimize_models_discrete.sh: a script for nanophotonic structure optimization over discrete spaces
- train_models.sh: a script for surrogate model training
- test_models.sh: a script for surrogate model testing
Before installing our package, you should install meep first. A guide to the installation of meep is provided in this link.
Our package can be installed from source or from source in an editable mode.
pip install .or
pip install -r requirements.txt
python setup.py developIf you want to create datasets by yourself, you can use scripts in the scripts directory.
Our datasets can be accessed via the Hugging Face repository.
@inproceedings{KimJ2023neuripsdb,
author={Kim, Jungtaek and Li, Mingxuan and Hinder, Oliver and Leu, Paul W.},
title={Datasets and Benchmarks for Nanophotonic Structure and Parametric Design Simulations},
booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
volume={36},
pages={4685--4715},
year={2023},
note={Datasets and Benchmarks Track}
}
We are open to any users who want to contribute to the project. You can refer to instructions to know how you can contribute to the project.
It is licensed under the MIT license.
We follow the code of conduct to create a diverse, inclusive, and postive community.