Page 2 of 2

Re: it was better before

Posted: Sat Nov 10, 2018 1:39 pm
by sarmath
lacilaci wrote: Sat Nov 10, 2018 12:50 pm On every one good spectral render you can find 10 just as good or better from rgb renderer https://corona-renderer.com/gallery

there is no secret ingredient. It's colors and tonemapping, and luxcore could definitely do a bit better here too.

But hey, there will be a spectral build of cycles so maybe we'll see some nice comparisons there. If there will be also some materials and shaders that can take the advantage. But other than some specific effects like dispersion I don't think it's worth it.
I dont want to start here spectral vs rgb rendering war :lol: Just think that spectral renderers output is somehow more natural and let me qoute our master Dade from this forum http://ompf2.com/viewtopic.php?t=1981:

"In my opinion, aside from considerations on the rendering quality where spectral rendering is likely to be always better, you have to factor also the impact on rendering times: not because it is computational more expansive than RGB/XYZ rendering but because it generates more noise. After all, you have yet another dimension to sample (i.e. wavelengths)."

So yep this days no one wants to wait 20h for the final render. But if someone make Luxcorerender spectral one would be nice becouse we dont have opensource option now :lol: Dade?

Re: it was better before

Posted: Sat Nov 10, 2018 2:16 pm
by Mango3
I'm speaking in a relative way as you can see i mention Arion render / Maxwell / Indigo and theses one seem to behave in this kind of super accuracy light transport and shading. you can really see they are different in that level but they are not Renderer from the past.

In the contrary some other renderer seem to rely to much on cheating for everything....
You talk about a subjective feeling...difficult to pin down..
Let me say something objective about the mentioned render engines and spectral vs RGB.
All render engines use shortcuts or biases to calculate the light distribution (ray clamping, bounce limits, approximations in the calculation of reflected contributions (Schlick's approximation) etc. Some to more some to less extent. None of these render engines accounts for instance for polarization, which you would get from an intensely lit sky or from reflected sunlight on a water surface and none of them would be able to render bi-spectral effects like a fluorescenct paint.

Spectral rendering is difficult, not only the render engine core needs to support it also the surface and volume shader, light sources and detectors (camera response) would need to be able to handle spectral data. Almost all the images you posted above use a HDRI for image based lighting. Problem is, there is no data format for a spectral HDRI. Using RGB components for lighting with a spectral renderer won't give you much advantage to a non-spectral one. This means the images above won't tell you much about the difference to RGB render engines.
Another point, many gallery images are probably post-processed, which can change an image appearance markedly, making it hard to judge the effects that you look after.

Re: it was better before

Posted: Sat Nov 10, 2018 2:41 pm
by lacilaci
Mango3 wrote: Sat Nov 10, 2018 2:16 pm
I'm speaking in a relative way as you can see i mention Arion render / Maxwell / Indigo and theses one seem to behave in this kind of super accuracy light transport and shading. you can really see they are different in that level but they are not Renderer from the past.

In the contrary some other renderer seem to rely to much on cheating for everything....
You talk about a subjective feeling...difficult to pin down..
Let me say something objective about the mentioned render engines and spectral vs RGB.
All render engines use shortcuts or biases to calculate the light distribution (ray clamping, bounce limits, approximations in the calculation of reflected contributions (Schlick's approximation) etc. Some to more some to less extent. None of these render engines accounts for instance for polarization, which you would get from an intensely lit sky or from reflected sunlight on a water surface and none of them would be able to render bi-spectral effects like a fluorescenct paint.

Spectral rendering is difficult, not only the render engine core needs to support it also the surface and volume shader, light sources and detectors (camera response) would need to be able to handle spectral data. Almost all the images you posted above use a HDRI for image based lighting. Problem is, there is no data format for a spectral HDRI. Using RGB components for lighting with a spectral renderer won't give you much advantage to a non-spectral one. This means the images above won't tell you much about the difference to RGB render engines.
Another point, many gallery images are probably post-processed, which can change an image appearance markedly, making it hard to judge the effects that you look after.
Regarding post processing. Most modern engines have tools that allow for good post processing and in-built camera responses and custom luts, white balance control etc. For many purposes this might be all you need and ideally you would be able to avoid further tweaking.
I personally didn't use heavy post in years. 95% is done with rendering. That includes corona, cycles and even latest luxcore I just output 16bit tiffs and usually small contrast at best(well for luxcore there is this extra step of denoising with nvidia denoiser)

Re: it was better before

Posted: Sat Nov 10, 2018 9:20 pm
by Sharlybg
There are many things that can give an extra natural look to an image. But to my point of view and what i use to see in my daily works/test/practice/research is Light bounce quality. And this is more obvious when it come to interior rendering. Speaking of that particular case it is obvoius to said that if we consider X complex interior project done by a very talented professional artist and add some limitation about light bounce and engine algorithm settings you will end up with this :

From more natural look to more CGI look : Max bounces 64 Min bounce 2


Bidirectional guided path engine : 32 + 32

simple Bidirectional engine : 32 + 32

Guided path engine + (IDLCS) : 32

Simple path engine : 8

Simple path engine : 4

Simple path engine : 2


For example heavy optimisation for speed matter like low clamping / low bouncing ray tend most of the time to cut a lot in term of hyper realism (not talking about pretty or plausible renders). I believe that sometime smart algorithm and high bouncing settings can bring an extra obvious boost to final raw output that can't be seen otherway.

There is a correlation between corona default ray deph and image quality i thing : (25 by default )
Default deph ray corona.jpg
There is also a correlation between brighter indigo output and light bounce. just saw it in the past with personal test ( Walls was so bright even at low light intensity :roll: ) : This is a quote from indigo features description page : https://www.indigorenderer.com/unbiased-rendering
Optimised for maximum quality
Indigo's CPU engine runs with a ray depth of many thousands by default.

While other, biased renderers commonly use a default of around 4-8 bounces and would struggle with a setting that high, Indigo is purpose-built for it and delivers great performance.
There is also another things i think could impact final output independly of artsits level. YES "energy conservation". here again it is about light bouncing quality/intensity . With something like this feature don't tell me that an interior scen will look the same even if you turn ON/OFF energy conservation. if light become darker and darker or less dark as it bounce arround object sure the output will be impacted. Add to that it mostly affect Diffuse or matt surfaces and ....

http://www.cgchannel.com/2018/09/isotro ... e-ifx-4-0/
Multiple scattering improves rendering of matte surfaces
New shading features in Clarisse iFX 4.0 include multiple scattering for reflected and transmitted light.

The software’s previous single scattering model minimised computation times, but caused surfaces to darken unrealistically as their roughness values increased.

The new implementation (shown above) approximates true multiple scattering – “while [it isn’t] strictly Lambertian diffuse reflection, it’s very close” – and is fully energy-conserving.

According to Isotropix, it has no impact on render time, and will be enabled by default
180906_ClarisseiFX4_MultipleScattering.jpg
look at the final ball this is not tone mapping issue.

Again on Energy conservation and albedo strengh i always look at theses images with huge " :?: " in my head. Please look how bright is object even with diffuse white set to only 80%. we are not at 100% and it look like supernova ("yes i exagerated to make you see what i mean). from here :

https://coronarenderer.freshdesk.com/su ... is-albedo-

corona albedo.jpg

Some renderer are very senssible to setting like light bounce setting / albedo value. BTW i will retry to perform some test about that light behaviour question with today renderers state but i'm pretty sure of the results.