We annotate both static and dynamic 3D scene elements with rough bounding primitives and transfer this information into the image domain, resulting in dense semantic & instance annotations for both 3D point clouds and 2D images.