Page 3 of 6

Re: spectral rendering

Posted: Sat Oct 17, 2020 5:13 pm
by Theo_Gottwald
My hint for the developers:
For Spectral rendering there will be the need to rework all materials for the frequency dependent attributes.
If you take the constants from the POV-Include files, that i have linked above, you can reach a very natural
spectral result. Even to the level that several ISO-Light configurations will break in the same way as they do in "real life"
in a light prism.
The values have been carefully developed that time from those people.
So maybe no need to re-develope the wheel just take these values.
Doing so introducing a basic spectral render capability for Luxray will just be a bit of this and that (and rendering stuff internally several times with the different material attributes for each light frequency).

Re: spectral rendering

Posted: Sat Oct 17, 2020 5:33 pm
by MetinSeven
Sharlybg wrote:
Sat Oct 17, 2020 4:43 pm
So the best thing will be Spectral rendering on demand and accelerated By GPU.Maybe we can open a bounty for that one :idea:
Sounds like a good plan.

Re: spectral rendering

Posted: Sun Oct 18, 2020 7:22 am
by Theo_Gottwald
Having said that, "Spectral Render" may also be implemented in several ways with more or less Quality and speed.
Because you can render just 3 Frequencies, or you could render 20. (Nature renders unlimited many). :-)

Re: spectral rendering

Posted: Sun Oct 18, 2020 8:08 am
by Sharlybg
Theo_Gottwald wrote:
Sun Oct 18, 2020 7:22 am
Having said that, "Spectral Render" may also be implemented in several ways with more or less Quality and speed.
Because you can render just 3 Frequencies, or you could render 20. (Nature renders unlimited many). :-)
That is an interesting insight. I think such parameter should investigated tested and hard coded to a lumit where it doesn't provide any additional quality.
Maybe this is the reason why indigo look like RGB nowdays.
Is there a way to know Maxwell frequencies limit ?

Re: spectral rendering

Posted: Sun Oct 18, 2020 12:46 pm
by Sharlybg
This statement from light tracer render engine can explain why more often people fail to show the advantage of spectral Vs RGB renderer.
At the same time, spectral rendering is most apparent when used with measured spectral data for both light sources and materials. But, in an artist-driven scene, the light source spectra are more usually reconstructed from RGB sliders in some way, leading to very smooth spectra, where it is very difficult to see actual visual differences between RGB and spectral rendering. Moreover, spectral rendering is computationally more expensive and generates more noise. For these reasons, Light Tracer is built on the RGB ray tracing engine.
BTW this paper seem to find a less consuming but yet close approach to spectral quality without the penalty : http://www.anyhere.com/gward/papers/PicturePerfect.pdf

Besides the open question of how many spectral samples
to use, there are other practical barriers to applying full spectral rendering in commercial software.
First, there is the general dearth of spectral reflectance data on which to base a
spectral simulation. This is consistent with the lack of any
kind of reflectance data for rendering. We are grateful to
the researchers who are hard at work making spectral data
available3 19
, but the ultimate solution may be to put the necessary measurement tools in the hands of people who care
about accurate color rendering. Hand-held spectrophotometers exist and may be purchased for the cost of a good laser
printer, but few people apply them in a rendering context,
and to our knowledge, no commercial rendering application
takes spectrophotometer data as input

The second practical barrier to spectral rendering is white
balance. This is actually a minor issue once you know how
to address it, but the first time you render with the correct
source and reflectance spectra, you are likely to be disappointed by the strong color cast in your output. This is due
to the change in illuminant from the simulated scene to the
viewing condition, and there is a well-known method to correct for this, which we will cover in Section 2.
The third practical barrier to the widespread acceptance of
spectral rendering is what we call the “data mixing problem.”
What if the user goes to the trouble of acquiring spectral reflectances for a set of surfaces, but they also want to include
materials that are characterized in terms of RGB color, or
light sources that are specified to a different spectral resolution? One may interpolate and extrapolate to some extent,
but in the end, it may be necessary to either synthesize a
spectrum from RGB triples a la Smits’ method14
, or reduce
all the spectral data to RGB values and fall back on three
component rendering again.
The fourth practical barrier to full spectral rendering is
cost. In many renderings, shading calculations dominate the
computation, even in RGB. If all of these calculations must
be carried out at the maximum spectral resolution of the input, the added cost may not be worth the added benefit.
Nice quote here from first mentionned paper (Picture Perfect RGB Rendering Using Spectral Prefiltering and Sharp Color Primaries) : As we are mainly interested by the look of spectral vs accuracy and considering the size of our team ressources.
In this paper, we present a method that has the same overall accuracy as Peercy’s technique, but without the computational overhead. In fact, no modification at all is required
to a conventional RGB rendering engine, which multiplies
and sums its three color components separately throughout
the calculation. Our method is not subject to point sampling
problems in spiked source or absorption spectra, and the use
of an RGB rendering space all but eliminates the data mixing problem mentioned earlie

Re: spectral rendering

Posted: Sun Oct 18, 2020 12:59 pm
by Sharlybg
This thing deserve a try look at the error level compared to full spectral and naive RGB
Picture Perfect RGB Rendering Using Spectral Prefiltering.jpg

Re: spectral rendering

Posted: Sun Oct 18, 2020 2:00 pm
by MetinSeven
Interesting!

Re: spectral rendering

Posted: Sun Oct 18, 2020 8:05 pm
by Theo_Gottwald
, or reduce
all the spectral data to RGB values and fall back on three
component rendering again.
They suggest exactly what i said above. They want render a R,G and B Sample and each with the frequency dependent Material values.
Then they sum it together. I think that is also what they did with POV-Sample above.

Re: spectral rendering

Posted: Mon Oct 19, 2020 5:27 am
by lighting_freak
Hello all,

I totally appreciate the discussion about spectral renderings. I agree to all of your pros and cons... my latest thread could point to one more situation that seems to be limited to current RGB solution.

https://forums.luxcorerender.org/viewt ... =4&t=2678

I would like to see a "scientific mode" that allows spectral visualisation. For start the definitions could be done by look up tables that specify volumes, surfaces, light sources etc.

In future nice and user friendly editors may steal the fear of spectral definitions for "artists".

By the way we have lots of spectral information and also some measurement devices to grab the necessary input data. I would be glad to support the community in that way.

BR

Re: spectral rendering

Posted: Tue Oct 20, 2020 7:17 pm
by FarbigeWelt
Three Sided Prism
Three Sided Prism
:ugeek: