Efficient Methods for Parameterizing and Rendering Large Models

Speaker:	Dr. Pedro Sander
		Application Research Group
		ATI Research

Title:		"Efficient Methods for Parameterizing and Rendering
		 Large Models"

Date:		Monday, 6 March 2006

Time:		4:00pm - 5:00pm

Venue:		Lecture Theatre F
		(Leung Yat Sing Lecture Theatre, ear lift nos. 25/26)
		HKUST


ABSTRACT:

Texture-mapping is a traditional technique that has several applications
in computer graphics. It uses surface signals to achieve a variety of
rendering effects, including color mapping, bump mapping (where surface
normals are the signal), displacement mapping (geometry), and
self-shadowing. To allow texture-mapping, a surface must first be
parameterized onto the texture domain by assigning texture coordinates to
its vertices. In the first part of this talk, I will present new metrics
and algorithms to parameterize a 3D surface onto the 2D domain for the
purpose of texture-mapping. As opposed to previous methods, our new
metrics directly minimize geometry and signal undersampling in the texture
domain. I will also present a novel parameterization algorithm that is
inspired on multigrid methods and uses hierarchical techniques to
accelerate the parameterization process and improve results. These methods
allow for more efficient use of the texture-map, yielding significantly
better results than previous approaches for storing signals over surfaces.
Since these methods were published, several new parameterization
algorithms based on this work have been introduced.

The second part of this talk focuses on how to use this parameterization
framework to render texture-mapped level-of-detail meshes. First, I will
describe a method to parameterize a progressive mesh (PM) so that the same
texture-map can be used for all levels of detail. Then, I will present a
new view-dependent level of detail rendering system designed with modern
GPU architectures in mind. Our method is the first out-of-core system to
support texture mapping, including a mechanism for texture level of
detail. Our method is suitable for all classes of GPUs that provide basic
vertex shader programmability, and is applicable for both out-of-core and
instanced geometry. I will conclude showing a demo of the Parthenon, which
uses all of the above mentioned techniques to efficiently render a
multi-million polygon texture-mapped model in real-time.


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

Dr. Pedro V. Sander is a member of the Application Research Group of ATI
Research, where he researches new rendering algorithms for upcoming latest
generation graphics hardware. Prior to joining ATI, Dr. Sander received a
Bachelor of Science degree in Computer Science from Stony Brook University
in 1998, and a Master of Science and Ph.D. in Computer Science from
Harvard University in 1999 and 2003, respectively. While pursuing his
Ph.D. at Harvard, Dr. Sander was a Microsoft Research Fellow from 2000 to
2002. Within computer graphics, his main areas of interest are real-time
rendering and geometry processing.