Wednesday, June 2

Thursday, June 3

Friday, June 4

Invited Talks

Leif Kobbelt: Point-Sampled Shape Representations

In point-based graphics, complex free form geometric objects are usually represented by an unstructured set of sample points that are scattered over the object's surface. However, in order to make such data sets useful for realtime graphics applications, a set of surface attributes is usually associated with each sample point, most importantly: a normal vector, color information, and a conservative estimate for the local sampling density. For
high quality rendering, these attributes are used to define small disks in object space (surface splats) such that a continuous surface can be displayed by accumulating or blending these splats in image space. From the geometric point of view it is interesting to investigate the approximation power and the shape flexibility
of these surface representations. In this context, point based shape representations can be considered as piecewise constant or piecewise linear surface approximations whose segments join in a C1 fashion. This is the important difference to triangle meshes which are piecewise linear C0 surfaces. This difference is the reason why some geometric computations are much easier and more efficient for point-based representations since no global
consistency requirements between the individual segments have to be taken into account.

In my talk I will discuss a few consequences of this observation and I will present algorithms that produce near optimal samplings of a given surface under different side conditions and different application specific requirements.

Turner Whitted: Are Points the Universal Modeling Primitive?

This talk is in two parts, one looking backwards at the reasoning leading up to the adoption of points as a display primitive (reconstructed primarily from Marc Levoy’s carefully preserved notes) and a second looking ahead at the general utility of points in graphics and imaging.

Historically, it is clear that the use of points as a universal representation was not a difficult concept to devise or justify given the trend towards richer graphics content with extreme geometric complexity. However, the implementation of an effective rendering processes was never simple. In an era of limited memory and computing power any simple, streamlined process was a compromise. On current hardware this is much less of a problem.

Going forward our emphasis should be focused as much on rendering quality as on content complexity. I claim that adaptively generated point samples are again ideal as a primitive element for high quality rendering, but at a cost in bandwidth and processing. Examples shown in the talk will illustrate the benefits as well as the costs.

Leonard McMillan: Architecture Futures for Point-Based Representations

New ideas can be either an impetus to revolution or a stimulus for evolution. In this talk I will speculate on how point-based modeling primitives might influence future directions in computer graphics architectures- both in terms of APIs and the underlying hardware. I will attempt to illuminate whether the incorporation of point-based modeling
primitives will require a fundamental rethinking of how we do the whole 3D computer graphics process including geometric transformations, shading calculations, sampling, and reconstruction, or, alternatively, whether
everything of interest will be accomplished in a fragment program on a next generation GPU. This task is particularly complicated by the fact that the trajectories of both graphics hardware and point-based modeling are moving targets.

I will also focus on identifying the unique opportunities afforded by new point-based representations. Point-based modeling might very well be the unifying concept that brings all other modeling approaches, polygon-based, higher-order surfaces, implicit surfaces and image-based models into one coherent framework. Moreover, point-based rendering might also be the lever that allows us to break the mold (some might say chains) of the traditional
graphics pipeline.