Generative Models for Complex Shapes

--------------------------------------------------------------------
               ***Joint Seminar***
--------------------------------------------------------------------
The Hong Kong University of Science & Technology

Department of Computer Science and Engineering
Center of Visual Computing and Image Science
--------------------------------------------------------------------
Speaker:        Dr. Vladlen Koltun
                Department of Computer Science
                Stanford University

Title:          "Generative Models for Complex Shapes"

Date:           Monday, 12 November 2012

Time:           4:00pm - 5:00pm

Venue:          Lecture Theatre F (near lifts 25/26), HKUST

Abstract:

I will discuss computational modeling of complex object categories. Can we
characterize the space of all chairs, or airplanes, or single-family
homes? Can such a characterization be generative, allowing the synthesis
of valid new chairs, or airplanes, or single-family homes? I will present
recent work that answers these questions in the affirmative.

Our key idea is to treat shapes from complex domains as assemblies of
components, and to learn relationships between components that
characterize plausible shapes from particular domains. For example, an
airplane may be composed from a fuselage, wings, stabilizers, engines, and
other components. The presence and shape of all these parts are
interrelated by complex probabilistic dependencies: the presence of a jet
engine implies a decreased likelihood of propellers, and well as a
decreased likelihood of the plane being a biplane. It also increases the
likelihood of a smooth and aerodynamic fuselage shape. I will present
representations that model these networks of relationships, and provide a
general approach to generative modeling of the structure
and semantics of three-dimensional shapes from complex domains. These
representations lead to new kinds of easy-to-use 3D modeling tools.


********************
Biography:

Vladlen Koltun is a faculty member in the Computer Science Department at
Stanford University, working in computer graphics, computer vision, and
machine learning. His prior work in theoretical computer science was
recognized with the NSF CAREER Award, the Alfred P. Sloan Fellowship, and
the Machtey Award.