generative-pnp-allerton

Code to implement methods and experiments for Generative Plug-and-Play: Posterior Sampling for Inverse Problems, available at https://arxiv.org/abs/2306.07233.

If you use this code in your work, please cite

C.A. Bouman and G.T. Buzzard, "Generative Plug-and-Play: Posterior Sampling for Inverse Problems," arxiv preprint:2306.07233, 2023.

Installation on MacOS and Linux:

Assuming you have python and conda installed, then from a terminal open in the generative-pnp-allerton directory, enter

cd dev_scripts
yes | source clean_install_all.sh

This will install and activate the conda enviroment gpnp.

If you prefer, you can create a pip virtual environment. From a terminal, change directory to the generative-pnp-allerton directory, then enter

yes | python -m venv gpnp-venv
source gpnp-venv/bin/activate
pip install -r requirements.txt
source gpnp-venv/bin/activate

This will create and activate the pip virtual environment gpnp-venv.

Experiments:

After creating a virtual environment as above, the figures from the paper can be reproduced with the following procedure.

From a terminal, change to the experiments directory and activate the virtual environment using either conda activate gpnp or source gpnp-venv/bin/activate.

1. Butterfly example: python gpnp_generation.py
2. CT example: python gpnp_generation_svmbir.py

When finished, you can use conda deactivate (or deactivate for a pip virtual environment).