/DISA-universal-multimodal-registration

Training code for the MICCAI 2023 paper "DISA: DIfferentiable Similarity Approximation for Universal Multimodal Registration"

Primary LanguagePythonMIT LicenseMIT

DISA: DIfferentiable Similarity Approximation for Universal Multimodal Registration

Paper -- ArXiv

This repository contains the training code for our MICCAI 2023 paper DISA: DIfferentiable Similarity Approximation for Universal Multimodal Registration

Data

Preprocessed training data extracted from the "Gold Atlas - Male Pelvis - Gentle Radiotherapy" (Nyholm et al. 2017) dataset can be downloaded from Google Drive.

The zip archive should be extracted in a folder called Data so that the npz files have a path Data/*.npz

Training

docker build . -t disa
docker run -it --rm --gpus all --volume "$(pwd)/Data":/data --volume "$(pwd)/Output":/output --shm-size=32gb disa

After each epoch model weights are saved in the Output folder