NEUTRON: Neural Particle Swarm Optimization for Material-Aware Inverse Design of Structural Color

This is the official repository accompanying the paper NEUTRON: Neural Particle Swarm Optimization for Material-Aware Inverse Design of Structural Color. link

Installation

1. clone the repository to your local machine: git clone https://github.com/hammer-wang/NEUTRON.git
2. create a conda envionrment using the provided env configuration file: conda env create -f environment.yml

Download the dataset

Download the train/val/test splits from link. Store the files to ./simulation/multilayer_data/sRGB_400K/.

Training model

1. activate the conda environment: conda activate meta-learning
2. run the bash script bash ./exp_script/run_best.sh
   The model checkpoint will be automatically save to the folder ./log/ for downstream evaluations.

You can also download a trained model direclty from Google Drive

Cr design experiment

Please refer to the provided paper_figures.ipynb notebook.

Photo reconstruction

Please refer to the bash script ./exp_scripts/reconstruct_imgs.sh.

---

If you find this repository useful for your research, please consider citing us as:

@article{wang2022neutron,
  title={NEUTRON: Neural particle swarm optimization for material-aware inverse design of structural color},
  author={Wang, Haozhu and Guo, L Jay},
  journal={iScience},
  volume={25},
  number={5},
  pages={104339},
  year={2022},
  publisher={Elsevier}
}