(1) The code is tested on: python 3.6
pytorch: conda install pytorch torchvision cudatoolkit=8.0 -c pytorch
(2) For training the model: python train.py
(3) C.D. (Before and After) is saved in Vis_xxx and a sample of qualitative testing registration results is saved in Vis_xxx.mat.
(4) For comparison with CPD, we use the package: https://github.com/siavashk/pycpd
(5) geotnf is modified based on the code from paper: Convolutional neural network architecture for geometric matching,Ignacio Rocco, Relja Arandjelović, Josef Sivic.