Publications

Modeling and animation of geometry lie at the very heart of many computer graphics applications.  As such, the field of geometric modeling has been the subject of many research articles in recent years.  Efficient, high-quality representation of deforming surfaces, however, remains a major challenge.  In this article, we offer a solution to this challenge by introducing a novel data structure and algorithms that allow for implicit but compact representations of deforming geometry at very high resolutions.  We demonstrate the usefulness of this data structure with a series of important graphics applications.

In this sketch, we introduce the Compact RLE Level Set. This representation is new structure that combines the benefits of the two previously presented sparse regular level set representations: the versatile RLE Sparse Level set of [Houston et al. 2004] and the near-optimally efficient DT-Grid of [Nielsen and Museth 2005]. We then present many important applications for computer graphics.

Several scenes in the film Cursed called for wispy smoke to interact
closely with characters. Simple particle systems failed to
capture the characteristic motion of wispy smoke, while existing
smoke simulators generally produced smoke of a diffuse nature,
more appropriate for explosions or large flames than cigarette or
incense smoke. This sketch describes our implementation of a flexible,
artist-friendly smoke simulator capable of producing realistic
wispy smoke for a production environment.

Creating the Tar Monster in Scooby Doo 2 presented a unique
challenge, because the desired effect of a continually flowing textured
character with expressive features had never been done before.
Starting from a fluid simulator as in [Enright et al. 2002], we
developed the liquid skin technique which applies a fluid layer over
an animated character. In addition, the facial animation was preserved
by using localized morphing, whereby a specified portion of
the simulation is made to match the Tar Monster geometry. The
result is a character from whom fluid constantly emanates, with
texture sliding down its body and fluid splashing during vigorous
arm gestures. Similar previous work includes [Sumner et al. 2003],
where the “TX” character is gradually liquefied. Our method, while
producing a comparable result in that texture is applied to fluid flow
on a character, uses quantities defined over the volume rather than
particles for control and texture coordinates.

This paper introduces a novel scalable level set representation, the RLE
(run-length encoded) sparse level set, and discusses the application of
representing implicitly a character animated via traditional means.

A few recent papers, [Enright 2002; Foster 2001; Foster 1996], have
described the development of a fluid simulation method useful for
computer graphics. We have implemented a system based on these
papers, and propose an extension that augments and simplifies the
handling of occlusions (impermeable solid objects). In the referenced
papers, occlusions are treated differently based on whether or not they are
moving, and discussion of how to deal with complex occlusions efficiently
is minimal. The technique we have developed unifies the treatment of
static and dynamic objects, and is able to better represent fluid-occlusion
interactions at the low simulation resolutions used.