Awesome

ml-optics-WACV2023

This repository contains code and some associated data from our WACV 2023 paper titled "Learning incoherent light emission steering from metasurfaces using generative models". A link to the paper can be found here: <url>

The repository contains the following files

1. vae.py -- Python code used to train the 1D VAE used to generate pump patterns
2. gen_images.py and gen_VAE_database.py -- Python codes used to generate a training set of 1D profiles for the VAE. These codes generate 1D Bezier curves of 3840 pixels
3. ax_surrogate_model.py -- Python code used to perform Bayesian optimization on pump patterns, optimizing for directivity
4. nn_surrogate_for_expt.py -- Python code that uses some experimental data to create a surrogate model that the Bayesian optimization uses
5. nn_-3.0.pth -- Example neural network surrogate model generated from the training code
6. -3.0mm.p -- Raw experimental data, generated by sampling the VAE trained using VAE.py for points in the latent dimension, using the VAE decoder to generate 1D pump patterns, and passing them to the experimental setup for evaluation

Note: Coupling to experimental setup API not included in this version of the code

Contact saadesa@sandia.gov for extra instructions or further clarifications on this repository