Harmonization Moyer et al.

This repo is based on code by Moyer et al. The code was modified to support the TractoInferno dataset. Other changes were also needed, as decribed in this document.

Setup

Create a virtualenv using requirements.txt, and use a virtualenv when running the scripts.

requirement_versions.txt lists versions of packages that work as of this writing.

How to launch training

Here is how to launch training with default hyperparameters:

export PYTHONPATH=$(pwd)  # As required by original Moyer implementation
python src/runner.py <TRACTOINFERNO_DATASET_PATH>

Note that the first time this script is run, the dataset is preprocessed to extract the voxel neighborhoods.

To see what hyperparameters/arguments are available, run python src/runner.py -h.

How to evaluate

python src/posthocadv.py <ZS_FILE>

where ZS_FILE is a file created at the end of training, with filename as such: trainval_zs__{comet_experiment.id}__{epoch}.pth.