This repository contains the code of Stable Deep MRI Reconstruction using Generative Priors.
The framework assumes the environment variables EXPERIMENTS_ROOT (needed for training) and DATASETS_ROOT (needed for training and evaluation) to be set.
EXPERIMENTS_ROOT is the output base-directory for training can be any directory on the machine.
DATASETS_ROOT should contain the fastmri dataset, i.e. the directory $DATASETS_ROOT/fastmri/multicoil_train (e.g.) should exist.
To train the model, run python train.py output_dir.
The first argument is the experiment output directory, i.e. checkpoints and losses etc. will be saved to $EXPERIMENTS_ROOT/output_dir/.
All evaluation code is found in evaluate.py.
The if __name__ == '__main__': block lists evaluation functions along with annotations indicating the corresponding table or figure in the paper.
Pretrained models and other data needed for evaluation can be found here.