Mattias Heinrich and Max Blendowski
please also see the accomponying MICCAI paper: "Closing the Gap between Deep and Conventional Image Registration using Probabilistic Dense Displacement Networks" (https://arxiv.org/pdf/1907.10931) 3D code is available at https://github.com/multimodallearning/pdd_net
There is also a Kaggle kernel available with the provided 2D TCIA multilabel dataset at: www.kaggle.com/mattiaspaul/learn2reg-tutorial