Nobuyuki Umetani, Autodesk Research
Exploring Generative 3D Shapes Using Autoencoder Networks
Abstract: We propose a new algorithm for converting unstructured triangle meshes into ones with a consistent topology for machine learning applications. We combine the orthogonal depth map computation and the shrink wrapping approach to efficiently and robustly parameterize the triangle geometry regardless of some imperfections such as inverted faces, holes, and self-intersections. The converted mesh is consistently and compactly parameterized and thus is suitable for machine learning. We use an autoencoder network to extract the manifold of shapes in the same category to explore and synthesize a variety of shapes. Furthermore, we introduce a direct manipulation interface to directly navigate the synthesis. We demonstrate our approach with over one thousand car shapes represented in unstructured triangle meshes.
Biography: Nobuyuki Umetani is a research scientist at Autodesk Research. Previously, he was a postdoctoral researcher in Autodesk Research and Disney Research Zurich. He received his Ph.D. degree in 2012 from The University of Tokyo under the supervision of Takeo Igarashi. He also spent one year at Columbia University and in TU Delft, and spent three months in Microsoft Research Asia and at UCL. He won the Microsoft Research Asia fellowship in 2011. The principal research question he addresses through his studies is: how to integrate real-time physical simulation into interactive geometric modeling procedure to facilitate creativity. He is broadly interested in physics simulation, especially the finite element method, applied for computer animation, biomechanics, and mechanical engineering.