act65 / mri-reconstruction

Compressed sensing and machine learning

mri-reconstruction

This repo contains my notes and code while exploring how to use learned priors for compressed sensing. The work was done in collaboration with Paul Teal (while I was working as a part-time research assistant).

Docs

I have written a couple of documents detailing my work and thoughts on;

• learned priors for comressed sensing
• density estimation
• reconstruction with guarantees

Code

To install you will need to run

git clone https://github.com/act65/mri-reconstruction.git
cd mri-reconstruction
python setup.py install

You should then be able to run the scripts (to train new models). For example;

python scripts/train_infovae.py --logdir={path} \\
--n_hidden=4 --width=64 --epochs=100 --beta=1.0 \\
--learning_rate=0.0001 --batch_size=8

And to run the notebooks;

jupyter lab

and then navigate to http://localhost:8888/lab.