LuxRender makes use of new realistic sky-dome model
Developers of LuxRender have just released an update that supports Blender 2.64 and introduces a new method of realistic sky-dome rendering.
The new model uses materials of a SIGGRAPH 2012 paper “An Analytic Model for Full Spectral Sky-Dome Radiance” by Lukáš Hošek and Alexander Wilkie.
An excerpt from the SIGGRAPH paper
That looks pretty realistic, doesn't it?
The work actually started in 2006 when Georg Zotti, Alexander Wilkie and Werner Purgathofer evaluated various approaches to physical modeling of sky-dome (mostly Preetham and Perez models from 1990s) and made some measurements of their own.
As the result of the research, in 2007 they posted “A Critical Review of the Preetham Skylight Model”. The review compared the Preetham model, the ISO/CIE 2003 Standard General Sky luminance distributions and a few measurements taken by the researchers using a Minolta LS-110 Luminance Meter on a tripod with coordinate indications.
Here's some boring scientific plotting you've been dying to see. Red area is where Preetham model suddenly issues negative zenith luminance values. Oopsie-daisies.
The review, essentially, stated that the Preetham model was prone to the following errors:
- A very clear atmosphere only has a moderate brightening along the horizon, and not the wide, extremely bright zone, leading to a totally unrealistic “horizon glow”.
- The Preetham model does not properly reproduce the noticeable darkening of the sky in the antisolar hemisphere, when the sun is low.
- The brightness peak towards the sun is not as steep as it can be measured or is modeled by the CIE Clear-Sky models.
The review ended with a statement that an analytic skylight model was required for outdoor scenes, when the skylight should be used as light source, and the sky as visible background.
After that most of the time Alexander was involved with quite different researches, including a very interesting recent one on interactive cloud rendering using temporally-coherent photon mapping. But he finally returned to this work on sky-dome model with Lukáš Hošek.
What's up with that new sky-dome model?
For their own model Lukáš and Alexander used the same general approach as the Perez and Preetham models, but made several modifications that amount to significantly more realistic rendition of sunsets and high atmospheric turbidity setups. The new model also accounts for ground albedo, and handles each spectral component independently.
An excerpt from the paper that demonstrates ground albedo taken into consideration
The code written by Lukáš Hošek was ported to LuxRender by Jean-Philippe Grimaldi. Here is a quick synthetic test that Jason Clarke posted in LuxRender's forum few weeks ago:
One of examples by J the Ninja
Apparently the new model also effectively ditches a blueish cast from sky-dome that was a problem for the Preetham-based sky1 model. I guess we'll see many more sky2-based render pretty soon.
Is that all? Well, the researchers stated that there is room for further improvements. For instance, future work will include an investigation of after-sunset conditions and polarisation patterns found on the sky-dome for conditions of low turbidity. They also plan to investigate the effect that non-monotonic aerosol distributions in the atmosphere have on the luminance patterns seen on the sky-dome.
Other changes in LuxRender 1.1
The new version makes use of a few major improvements in Blender 2.64, such as the new tile rendering and the new color management features. Of course, the LuxBlend25 extension also makes the new sky2 model available to Blender users.
There's also a considerable speed-up for CPU rendering, from 10 to 40%, mostly noticeable when using instances.