Smooth, exact rotational symmetrization for deep learning on point clouds

We propose a general symmetrization method that adds rotational equivariance to any given model while preserving all the other requirements.