July 28, 2005

Light Clouds, Camera Arrays
and Speedier Rendering at SIGGRAPH 2005

By Doug Ramsey

Computer-graphics researchers from UCSD’s Jacobs School of Engineering will be out in force at SIGGRAPH 2005. The leading annual conference and expo for computer graphics and interactive techniques takes place July 31 to August 4 at the L.A. Convention Center. This year’s event will attract a broad array of attendees, including faculty and student researchers from UCSD and other universities, as well as industry leaders and film-makers.

Only 98 research papers were selected from 461 submissions, and four of the accepted papers were authored or co-authored by researchers in professor Henrik Wann Jensen’s Computer Graphics Lab. Jensen co-authored three of the papers, all of which will be published in the Proceedings of ACM SIGGRAPH 2005.

Mimicking how light transits through a multi-layered leaf.

“It is great that these papers got accepted for SIGGRAPH, which is the premier conference for computer graphics research,” says Jensen. “The students have been working on these projects for a long time, and I am very happy that their hard work paid off.”

Building on their previous work to improve the lighting of translucent objects such as skin, Jensen and Ph.D. candidate Craig Donner are now tackling the diffusion of light in multi-layered translucent materials. “We are presenting a new and efficient technique to account for diffusion in thin slabs such as a leaf (pictured), where some of the light may exit the leaf rather than being completely absorbed or reflected,” says Donner. “We then extended this theory and applied it to multi-layered materials such as human skin, paint or paper.”

Professor Jensen – who shared an Academy Award last year for a light-diffusion technique first presented at SIGGRAPH 2001 – also co-authored two other new papers to be presented in Los Angeles.

Real-time rendering of complex model of the Sponza Atrium -- over 115,000 triangles at 30 to 50 frames per second. Watch video: 00:46

Jensen worked with research assistant Anders Wang Kristensen and Lund University collaborator Tomas Akenine-Möller on Precomputed Local Radiance Transfer for Real-Time Lighting Design. The paper introduces a new method for real-time relighting of scenes illuminated by local light sources. They do so by introducing the concept of unstructured light clouds – a compact representation of local lights in the model..

“Our results demonstrate real-time rendering of scenes with moving lights, dynamic cameras, glossy materials and global illumination,” explains Jensen. “The system handles indirect illumination efficiently for models with more than 100,000 triangles, so our algorithm offers a practical alternative to the traditional techniques used in real-time graphics.”

A third paper to be presented at SIGGRAPH 2005 is titled Wavelet Importance Sampling: Efficiently Evaluating Products of Complex Functions. Jensen’s co-authors are UCSD grad student Wojciech Jarosz, Lund’s Akenine-Möller and Petrik Clarberg (Lund and UCSD).

An image rendered using the new algorithm by sampling the product of the environment map and the BRDFs on the fly. Watch video: 04:15

‘Importance sampling’ is a mathematical technique to select a good distribution from which to simulate random variables, and is widely used in computer graphics for modeling lighting and other elements of a scene. For instance, it is now commonly used with bidirectional reflectance distribution function (BRDF), a measure of how much light is reflected when light makes contact with a certain material. But current techniques are limited because they do not efficiently permit modeling the lighting of scenes that may change over time.

The authors will present a new technique for importance sampling products of complex functions using wavelets. “We show how the product can be sampled on-the-fly without the need of evaluating the full product,” says Jarosz, who TA’d Jensen’s CSE 168 course, “Rendering Algorithms,” in the spring. “This makes it possible to sample products of high-dimensional functions even if the product of the two functions in itself is too memory consuming.”

One application of the new sampling technique is rendering of objects with measured BRDFs illuminated by complex distant lighting. “Our results demonstrate how the new sampling technique is more than ten times more efficient than the best previous techniques,” adds Jarosz. “In contrast to previous work, our method is capable of efficiently importance sampling the product of the lighting and the BRDF.”

Camera array configurations for high-resolution imaging (left) and high-speed video capture. [Mpeg-4 video only, downloadable here]

Another member of Jensen’s team, Ph.D. student and research assistant Neel Joshi, co-authored a paper on high-performance imaging using large camera arrays. His co-authors on that paper – featured in the SIGGRAPH session “Capturing Reality” – are from Stanford University, where Joshi received his M.S. in 2004. They explore a range of applications that become possible with the advent of inexpensive digital image sensors.

“This changes how we think about photography,” report Joshi and his co-authors. “Our goal was to explore the capabilities of an array of 100 custom video cameras that would be inexpensive to produce in the future.” The research also assumes cheap and easy processing of large numbers of images, and the ability to create photographs that combine information from a number of sensed images.

Media Contact: Doug Ramsey, (858) 822-5825