BLISS is a Bayesian method for deblending and cataloging light sources. BLISS provides
- Accurate estimation of parameters in blended field.
- Calibrated uncertainties through fitting an approximate Bayesian posterior.
- Scalability of Bayesian inference to entire astronomical surveys.
BLISS uses state-of-the-art variational inference techniques including
- Amortized inference, in which a neural network maps telescope images to an approximate Bayesian posterior on parameters of interest.
- Variational auto-encoders (VAEs) to fit a flexible model for galaxy morphology and deblend galaxies.
- Wake-sleep algorithm to jointly fit the approximate posterior and model parameters such as the PSF and the galaxy VAE.
-
To use and install
bliss
you first need to install poetry. -
Then, install the fftw library (which is used by
galsim
). With Ubuntu you can install it by running
sudo apt-get install libfftw3-dev
- Install git-lfs if you haven't already installed it for another project:
git-lfs install
- Now download the bliss repo and fetch some pre-trained models and test data from git-lfs:
git clone https://github.com/prob-ml/bliss.git
- To create a poetry environment with the
bliss
dependencies satisified, run
cd bliss
poetry install
poetry shell
- Verify that bliss is installed correctly by running the tests both on your CPU (default) and on your GPU:
pytest
pytest --gpu
- Finally, if you are planning to contribute code to this repository, consider installing our pre-commit hooks so that your code commits will be checked locally for compliance with our coding conventions:
pre-commit --install
- BLISS now includes a galaxy model based on a VAE that was trained on Galsim galaxies.
- BLISS now includes an algorithm for detecting, measuring, and deblending galaxies.
- BLISS already includes the StarNet functionality from its predecessor repo: DeblendingStarFields.
Runjing Liu, Jon D. McAuliffe, Jeffrey Regier, and The LSST Dark Energy Science Collaboration. Variational Inference for Deblending Crowded Starfields. arXiv:2102.02409, 2021.