BME BSDF data, conclusions

conclusions, F.A.Q. and discussion

Have other BSDF models been fitted ?

Yes, other function types, (e.g. Cauchy and Laplace functions) have been tested and used as BSDF models. Other BSDF models could be documented here, depending on available time and interest of clients.

Why aren't there more complex materials shown, like acrylic twin-wall-sheets, honeycombs or micro-structured surfaces ?

Yes, measurements of these are not a problem and have been done here. However, the first reason of this website was to illustrate frequently used material models . Complex examples may follow.

Plots have a linear scale and the BSDF data seems to have low dynamic range.

The pgII gonio-photometer has > 1:10⁷ dynamic range, as shown in the raw data and the logarithmic plots for the metal model (e.g. anofol606). Most plots have a linear scale, since a) it is generally faster to understand at first sight and b) deviations between standard models and data are clearly visible even on a linear scale. For more detailed examinations, log scales are used for 2D and 3D plots. See literature reference on the instrument web site.

How about BSDF data in the infrared ?

Measurements with the pgII cover the wavelength range up to 2.5µm, using different sensors. Please contact us if this data is of interest to you.

Your description of material is grey, samples are coloured.

Please see introduction. Primary emphasis of these pages lies on the shape of the BSDF.

Images rendered with these parameters don't have details on their surface

For texturing, bump- or image mapping, and why they aren't used here, please see introduction.

Is more demo data available ?

Older data has been available on our sister site www.pab-opto.de: LED candle-power-distributions, transmission data and reflection data. Please note that those specific data-sets had been generated using first-generation pgII hard- and software, they are considered historic and may serve as illustration.

discussion specific to Radiance rendering program

For some materials, parameters for plastic are outside their reasonable range (e.g. RGB > 1 ), this can't be right.

During fitting for plastic, parameters are not bound. If the fitting algorithm finds that the error χ is lower with an RGB parameter >1 , then this has to be regarded as a fact. What happens is a tradeoff between matching the peak and matching the ''flat'' area. The mathematical conclusion is that the model itself is not describing the data in an optimal way, otherwise its parameters would remain inside their intended ranges.
Here's an empirical and practical solution: The metal_ps models have their first RGB parameter a1 kept constant during fitting, so the direct-hemispherical reflection value ρ is kept fixed.

The given parameters for plastic and metal vary with incident direction, why ?

A mathematical model should have few and constant parameters, depending on the material characteristics only. The dependency of the plastic model on the incident angle theta_in for most materials is not ideal.

Radiance plasfunc allows user supplied functions for specularity. Why not use these ?

Correct, plasfunc, metfunc and transfunc allow the specular part of the BSDF to be defined using a fairly powerful functional language in user supplied files. However, no specular rays are spawn when a ray hits these materials. This would need an mathematical inversion step of arbitrary user-supplied function, which is numerically difficult and mathematically-explicitly impossible. So, in practice, these surface do not generate a blurred mirror image of their surroundings. They only work with light materials, and require setting Radiance parameters for source subdivision and jitter, e.g. -ds 0.01 -dj 0.9 .

our suggestions to Radiance plastic,metal Gaussian Ward models

• Deviation between data and standard models can be fairly large, which leads to corresponding errors in simulation. A factor of ten between a model with optimised parameters and data of non-metal surfaces was not uncommon. Sometimes this is close to the peak, an angular area which might matter in glare analysis. This should be subject to research on the software side how to improve the models.
• Models of metal surfaces exhibit large (≈ factor 100) deviations in the back-scatter direction. This introduces fairly large errors when models are used un-guided and un-checked in simulations.
• The majority of materials tested showed a dependency of specularity and roughness with incident angle. This should be cared for with an revised model.
• The vast majority of materials tested showed a fall-off in the BSDF with large outgoing angles (≈ >75°). Hence the assumption of a truly constant part of the BSDF is not supported by this data.
• For shiny materials, the shape of peaks at their bottom, is not well reproduced by the Gaussian model.

Contact author per email, see pgII gonio-photometer that was used for collecting BSDF data.

Links: technical details. model details, sample gallery, BME start page