Towards Shape Understanding through Non-Parametric Shape Warping

Abstract

This paper pursues the idea of understanding shapes of unknown objects through establishing correspondence with points from the surface of known objects. A lot of geometryrelated knowledge, such as functionally correct grasps or object constellations, could thus be transferred from known shapes to novel shapes of the same or a similar category. As one critical module in such a system, this paper considers warping of surfaces in 3D across significant and possibly non-smooth shape variations. Results are shown for some objects from the Princeton Shape Benchmark and a range scan.