Why is rendering of simple BSDF material slow and inaccurate?


#1

Hi everyone.
I have the project to analyse the influence of using different types of glass on the daylight distribution in a greenhouse.
In this context I thought it would make sense, i.e. give maximum flexibility, to model the glass as a BSDF material (e.g. as can be exported from LBNL Window). I’ve set up a simple test scene: A “Gendaylit” sky, a horizontal glass pane, and somewhat underneath the glass pane an additional horizontal plane on which I compute the light distribution. (Btw I’m using the Blender vi-suite add-on for creating the scene and running the radiance simulation). For test reasons, I first simply took a BSDF for clear glass, i.e. only specular, no diffuse component, i.e. the matrix is diagonal. Now, even with this most trivial BSDF I don’t get a homogeneous distribution of light on my test plane, even when increasing the radiance parameters to seemingly extremely high values.
Why wouldn’t this work? Maybe I lack understanding on how BSDF materials are computed in radiance. I’d be grateful for any advise on this question.

(Ultimately I want to model frosted glass. Would somebody have an advise on how to best model frosted glass? I need physically accurate results. That requires for instance that the diffuse lobe is around the spectral direction and changes with incidence angle.)


#2

Hi Daniel,

the data-driven BSDF model is ment to replicat irregular scattering,
e.g. arbitrary BSDF. Therefore, Radiance can make little to no
assumptions when rendering it, and basically has to send out as many
arbitrary rays as you need to sample the BSDF at a resolution fine
enough to reach the light source. This means that you are either slow or
inaccurate - but at least not both, as the subject line suggests :wink:

The BSDF that you get from Window has a rather low resolution, so it may
be not sufficient to distinguish the effect of different kinds of
frosted glass. If that is what you want, you’d need high resolution BSDF
models and give Radiance some time to render it (unless you are ready to
try the Photon Map, which may be more efficient).

On the other hand, if you are looking at illuminance, and given that
your “aperture” is practically infinite in a glass house, you would see
little to no difference when the resolution changes (other than if you
are interested in shadow patterns by frames or supporting structure). In
fact, you may require nothing but than direct-hemispherical transmission
for several incident directions to fit a trans model. That would also be
much faster in rendering. There are some publications that you may find
interesting (there is more, just a quick look-up):

  • Christoph F. Reinhart, Marilyne Andersen. Development and validation
    of a Radiance model for a translucent panel. Energy and Buildings
    38, pp. 890-904 (2006). DOI: 10.1016/j.enbuild.2006.03.006
  • Lars O. Grobe, David Geisler-Moroder, Stephen Wittkopf. Transfer of
    measured transmission distribution data into Radiance. 9th Radiance
    Workshop, Freiburg, DE (2010).
  • Peter Apian-Bennewitz. Radiance plastic, metal, trans parameters
    based on material data. 10th Radiance Workshop, Berkeley, US (2011).
  • Andreas Noback, Lars O. Grobe, Franziska Lang. Modeling the effects
    of daylight scattering by window glass: The case of 6th century
    Hagia Sophia in Istanbul. 21st International Congress of the
    Association Internationale pour I’Histoire du Verre, Istanbul, TR
    (2018). DOI: 10.5281/zenodo.1414007

Cheers, Lars.