Artifact-Resilient Real-Time Holography

Project Page | Paper

Victor Chu, Oscar Pueyo-Ciutad, Ethan Tseng, Florian Schiffers, Grace Kuo, Nathan Matsuda, Albert Redo-Sanchez, Douglas Lanman, Oliver Cossairt, Felix Heide

We introduces a novel method for generating artifact-resilient phase holograms in real-time. This code implements the following:

• A model for simulating pre and post-pupil obstructions (a common source of holographic artifacts).
• A differentiable metric to quantify holographic artifact resilience (see rayleigh_distance_loss in holo_utils.py).
• Training and inference of a real-time neural network to create pseudo-random phase holograms that are inherently artifact-resilient.

This code builds on Neural-Holography and Pado repositories.

Installation

1. Create and Activate a Conda Environment

conda create -n arh python=3.10 -y
conda activate arh

2. Install Required Packages With pip

pip install uv

uv pip install -r requirements.txt

Training RealTime ARH

Place your training data in the data directory. Then run the following commands:

python train_ARH.py --channel=0 --run_id=experiment_red
python train_ARH.py --channel=1 --run_id=experiment_green
python train_ARH.py --channel=2 --run_id=experiment_blue

Evaluating Pretrained Models

For evaluation, we provide four pretrained models for each color channel (R, G, and B). You can run inference using the provided scripts:

./ensemble_inference_2d.sh
./ensemble_inference_3d.sh

Our RGBD evaluation data comes from Split Lohmann Multifocal Displays.

Citation

@article{Chu2025RealTime,
author = {Chu, Victor and Pueyo-Ciutad, Oscar and Tseng, Ethan and Schiffers, Florian and Kuo, Grace and Matsuda, Nathan and Redo-Sanchez, Albert and Lanman Douglas and Cossairt, Oliver and Heide, Felix},
title = {Artifact-Resilient Real-Time Holography},
year = {2025},
issue_date = {December 2025},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {44},
number = {6},
issn = {0730-0301},
url = {https://doi.org/10.1145/3763361},
doi = {10.1145/3763361},
journal = {ACM Trans. Graph.},
month = dec,
articleno = {219},
numpages = {13}
}

License

Our code is licensed under BSL-1. By downloading the software, you agree to the terms of this License.