Global Illumination Test Scenes (V 0.9)

Overview (Abstract)

The global illumination community has discussed having a database of scenes that could be used to compare and validate different global illumination algorithms. We present a set of test scenes for global illumination algorithms and propose that they be the beginning of such a database. These scenes are designed to be as simple as possible and still test a particular aspect of energy transport. The scenes are available on a web site in a variety of formats, along with images and pixel radiance data. We feel that the simplicity and availability of the models will make it easier for the community to use them, and that the field will benefit from a standard set of scenes. Additionally, we discuss classes of models with analytic solutions.

The paper giving more details of this work can be found here.

Description of Data

Geometry

All geometry was modeled in Maya. Solids were used for all object, however single rectangles were used for lights. The geometry is primarily composed of scaled, translated, and rotated cubes. One object in the geometry scene is a thin wedge. All models were created using a scale of meters. Maya represents everything internally in centimeters, and unfortunately uses centimeters as the unit when converting to .obj and VRML2. The name of each scene below is linked to a figure describing it.

Materials

There are only three different materials used in these models. The luminaire material specifies that the emitting polygon has an emitted radiance of 1. Total emissive power on the luminaire is PI * Area.

diffuse50

An ideal diffuse (Lambertian) material that has a reflectance of 50%.

diffuseLuminaire1

An ideal diffuse (Lambertian) emitter, with a radiance of 1. Not all lights are the same size, so total power of the luminaires varies.

mirror

An ideal specular mirror. No energy is absorbed. Only used in the caustic model.

Reference Data Format

The goal here was simple and easy to parse. Also, it should be able to handle floating point (or higher) precision for exactly one channel, and be easy to use for doing numeric comparisons between different runs. The solution was a file with lines of "#" delimited comments, hopefully with some information about how the file was generated. Next comes a line with the width and height as whitespace separated integers. Last comes width * height lines of single float/double per line (in ascii) radiance values. A bit large, but easily compressible.

Submitting Data

Send me a message with links pointing to the reference data/images you have for the scenes you have tested. A link to a description of the renderer would be useful as well.

Scenes

Reference data will be available soon, now that the reference data format has been finalized.

Need to package everything up into .tar files.

Need to convert to MGF.

Renderers

DALI

Photon map and Monte Carlo Renderer written by Henrik Wann Jensen

PRIS

Monte Carlo Renderer, an overview is available.