Learning an Effective Equivariant 3D Descriptor Without Supervision | IEEE Conference Publication | IEEE Xplore