Ben Spencer and Mark W. Jones

Visual and Interactive Computing Group, Swansea University, UK

The caustics underneath the sphere and leaf are generated using an enhanced photon mapping algorithm described in Into the Blue: Better Caustics through Photon Relaxation, Ben Spencer and Mark W. Jones, Eurographics 2009. The advantage of the approach is that low-noise radiance estimates may be achieved using very low bandwidth kernels. The caustic photon map in this scene contains 120,000 photons and only 50 are used in each radiance estimate. We can achieve good-quality results with as few as 20 photons. This results in the reduction of proximity, topology and boundary bias and also reduces the time required to render caustic illumination. The scene was created using a scan of a real leaf, post-processed and overlaid onto a translucent scattering dielectric film perturbed using a fractal noise function. Clip, gloss and bump maps were then created and the whole scene rendered using our own global illumination rendering platform.

Arno Zinke

GfaR GmbH, University of Bonn, Germany

The image was rendered using an accurate and efficient path-tracing method for woven cloth based on light scattering from filaments. In contrast to traditional approaches all filaments are modeled individually using generalized cylinders and physically-based Bidirectional Curve Scattering Distribution Functions ("BRDF for fibers"). The highlights, the translucency and all other subtle scattering effects are captured accurately and are a direct consequence the BCSDF model and the weaving pattern. The cloth (base geometry) was simulated using a proprietary cloth solver.

Martin Bokeloh (1), Alexander Berner (1), Michael Wand (2), Hans-Peter Seidel (3) and Andreas Schilling (1)

(1) WSI/GRIS, Tübingen, Germany
(2) Saarland University, Germany
(3) Max-Planck Institute Informatik, Germany

We detect reoccuring parts such as doors and windows in point clouds. The dataset is a laser scan of the historical building "Zwinger" in Dresden, Germany (courtesy of Markus Wacker, HTW Dresden). Our algorithm reduces the model to a line-based representation, which is indicated on the left hand side of the image. We use this representation to detect symmetries automatically. All instances of the same type are colored in two variations of the same color to distinguish between neighboring instances. We also exploit the symmetries for reconstruction: For a given type of building block, we compute an average instance taking all found instancesinto account. This gives more details as one may notice in the gray parts of the building.

Eric Galin, Adrien Peytavie, Jerome Grosjean and Stephane Merillou

This scene shows an rocky scenery created with our Arches System. The corresponding paper Arches: a Framework for Modeling Complex Terrains will be presented at EG 2009.
The originality of our technique is that we can automatically generate terrains of arbitrary geometry and topology with different materials whereas previous methods are based on height field representations. Our method provides us with a coherent framework for creating arches, caves, cliffs or overhangs, and generating sand, rocks and different material layers.
Rocks and boulders have been generated automatically by instancing and tiling a very small set of rocks (there are only 70 different rocks in this image, about 6000 rocks instances). We need not compute collision detection and directly generate stable rock configuratons with a purely geometric approach.
We have developed a specific texturing method that avoids the computation of uv-coordinates and coded it as an HLSL shader so as to render our terrains of arbitrary geometry.
Our system has been linked with Maya and this scene rendered using Mental Ray. We achieve real time rendering with our HLSL shader (not with global illumination as in this scene though) on standard PCs.