Rotation equivariant vector field networks  (to appear in ICCV 2017)
Using MatConvNet 1.0-beta25 the intallation can be done by running
vl_contrib('RotEqNet'). For older versions it can also be unziped manually.
Run the setup script
setup_RotEqNet.m from Matlab and then
main.m to download the MNIST-rot dataset, train and test a rotation invariant classifier.
Anders Waldeland, from the Norwegian Computing Center, has been working on a Pytorch remake of RotEqNet! The code is available here.