 The paper presents an efficient and effective method for semantic classification of 3D point clouds that can handle unstructured and inhomogeneous data and is computationally efficient, making it possible to process millions of points in minutes. The key issue is careful handling of neighborhood relations, which leads to a fast and expressive feature set. The method outperforms the state of the art in terms of per point classification accuracy while being much faster to compute. This article was authored by T. Heckel, J. D. Wegner, and K. Schindler.