WaveDiff

A differentiable data-driven wavefront-based PSF modelling framework.

This repository includes:

A differentiable PSF model entirely built in Tensorflow.
A numpy-based PSF simulator here.
All the scripts, jobs and notebooks required to reproduce the results in arXiv:2203.04908 and arXiv:2111.12541.

For more information on how to use the WaveDiff model through configurable scripts see the long-runs directory's README.

Proposed framework

A schematic of the proposed framework can be seen below. The PSF model is estimated (trained) using star observations in the field-of-view.

Install

wf-psf is pure python and can be easily installed with pip. After cloning the repository, run the following commands:

$ cd wf-psf
$ pip install .

The package can then be imported in Python as import wf_psf as wf. We recommend using the release 1.2.0 for stability as the current main branch is under development.

Requirements

numpy [>=1.19.2]
scipy [>=1.5.2]
TensorFlow [==2.4.1]
TensorFlow Addons [==0.12.1]
Astropy [==4.2]
zernike [==0.0.31]
opencv-python [>=4.5.1.48]
pillow [>=8.1.0]
galsim [>=2.3.1]

Optional packages:

matplotlib [=3.3.2]
seaborn [>=0.11]

Reproducible research

arXiv:2203.04908 Rethinking data-driven point spread function modeling with a differentiable optical model (2022)

Submitted.

Use the release 1.2.0.
All the scripts, jobs and notebooks to reproduce the figures from the article can be found here.
The trained PSF models are found here.
The input PSF field can be found here.
The script used to generate the input PSF field is this one.
The code required to run the comparison against pixel-based PSF models is in this directory.
The training of the models was done using this script. In order to match the script's option for the different models with the article you should follow:
- poly->WaveDiff-original
- graph->WaveDiff-graph
- mccd->WaveDiff-Polygraph

Note: To run the comparison to other PSF models you need to install them first. See RCA, PSFEx and MCCD.

arXiv:2111.12541 Rethinking the modeling of the instrumental response of telescopes with a differentiable optical model (2021)

NeurIPS 2021 Workshop on Machine Learning and the Physical Sciences.

Use the release 1.2.0.
All the scripts, jobs and notebooks to reproduce the figures from the article can be found here.

Citation

If you use wf-psf in a scientific publication, we would appreciate citations to the following paper:

Rethinking data-driven point spread function modeling with a differentiable optical model, T. Liaudat, J.-L. Starck, M. Kilbinger, P.-A. Frugier, arXiv:2203.04908, 2022.

The BibTeX citation is the following:

@misc{https://doi.org/10.48550/arxiv.2203.04908,
  doi = {10.48550/ARXIV.2203.04908},
  
  url = {https://arxiv.org/abs/2203.04908},
  
  author = {Liaudat, Tobias and Starck, Jean-Luc and Kilbinger, Martin and Frugier, Pierre-Antoine},
  
  keywords = {Instrumentation and Methods for Astrophysics (astro-ph.IM), Computer Vision and Pattern Recognition (cs.CV), FOS: Physical sciences, FOS: Physical sciences, FOS: Computer and information sciences, FOS: Computer and information sciences},
  
  title = {Rethinking data-driven point spread function modeling with a differentiable optical model},
  
  publisher = {arXiv},
  
  year = {2022},
  
  copyright = {arXiv.org perpetual, non-exclusive license}
}

martinkilbinger/wf-psf