Unified Sampling simplified

[bart]

I'd like to start an artist thread on using unified sampling in a more brute force way to simplify scene set up and control.

In its simplest form, brute force means turning down all local sampling numbers. We could think about general areas of local sampling control.

Lighting - area light samples
Lens - depth of field samples
Materials - glossy samples, diffuse samples, FG samples
Motion blur - time samples (automatically set to 1 when using unified sampling)

Pure brute force would mean 1 sample, 1 ray path, but to take advantage of potential tradeoffs inbetween, we could scale them to some other amounts.

Use these settings to start:
"samples min" 1
"samples max" 300
"samples quality" 0.2
"samples error cutoff" 0
"unified sampling" on

Reduce your local sampling and start turning up samples quality as needed.

There's a lot of material on unified sampling on this blog:
http://elementalray.wordpress.com/20...or-the-artist/

But if you'd like to step along here in this thread in small steps, we are emphasizing reduced local sampling to begin with.

[thorsten hartmann]

Hi bart,

i understand your workflow for the first ray, but what is with second ray.

If i reduce the Glossy Samples to 1 Sample, work the first ray ok, but not in a other Refraktion or Reflection ray.

What can i do.

cheers
hot chip

[bart]

1 ray for all bounces. With absorption in materials, not all ray paths will necessarily bounce light.

If you reduce glossy samples to 1 ray, it will remain 1 ray for all bounces. So either the trace depth would limit when it stops, or a different light path, or absorption. When using this approach, generally you can turn up trace depth higher.

BTW, what materials are you using?

[thorsten hartmann]

Hi Bart,

i use A&D Material. I will later upload a screen shot, that better explain my Problem.

cheers
hot chip

[bart]

So glossy samples can be set to 1 or 2 and there is a local trace depth setting on that as well which one should ensure to be set high enough for your tastes. I'd set it and the global setting up above 4 in using this approach. What is the glossiness value? Looking forward to seeing the screen shot.

[rlevene]

are you suggesting we use Brute Force Final Gather mode with this then? And set final gather rays to 1? density and interpolation will be ignored then correct? or is desity still necessary?

cheers,

Richard

[Remydrh]

For now, no. Final Gather/IP, etc are still expensive as brute force. (IP tends to be less expensive than FG due to importance sampling) However, using brute force FG/IP etc should be less expensive than it has been in the past.

Using a few other tools like the fgshooter, etc you should still be able to get away with fast interpolated FG for indirect illumination. This goes back to combining techniques between brute force and interpolated schemes.

One of the reasons we pointed to brute force for shaders is that we could achieve higher quality with better speeds than older methods combined with Unified Sampling. If you try brute force FG for example, your render times will indeed become very expensive again (but still better than regular AA)

So to try and maintain the best of both worlds, direct lighting and shaders benefit most from brute force settings. Area lights may still need some more samples, as well as something like SSS samples internally for the shader. But otherwise, complex scenes perform really well like this.

I think as things evolve, you'll be able to control much more just using Unified Quality. But for now this is a bit of a transition period where you can experiment and combine methods in preparation for things like BSDF shaders and EDF lighting. You'll have the flexibility of mental ray with the simplicity of something like iray.

[Kzin]

i use brute force fg at the moment and i think it works not that bad.
the problem with low samples is that the representation of the lighting is really coarse with low fg samples.
and you cant replace fg samples with aa samples, its not possible. you can save fg samples with unified, thats right, but its not a solution ala arnold there all those sampling things are really unified.

BUT, bart, if you and the other guys at ARC would implement importance sampling for brute force fg i think this would speed up the rendering alot and would give us better lighting with low samples. any chance for this?

[thorsten hartmann]

Hi Bart,

ok ihave test it, and this new values give me a good result. But it need a long time to render. I render most stils without real motionblur or real z-depth. At this time i have the best result with photon+fg, adaptive AA and my Sample Curve for all Shaders in the Scene. I think i must more leran about unified sampling.

cheers
thorsten

[bart]

kzin, I'm not sure exactly what you are saying about fg samples replaced with aa samples. And the sentence after it doesn't quite make sense to me either, with what I know. Could very well be that we need to provide more information about usage, which is what I'm trying to do with this thread.

Regarding importance sampled FG, which you could actually use with FG+importons, we are working on many related things which will take more and more advantage of unified's flexibility. And the issue you point up seems a bit more like importance sampling the light if using an ibl based system, which turns an environment into a light.

That is what I want to stress, the flexibility. It should be able to do any type of algorithm, spanning multiple rendering products out there, allowing the user to trade off ease-of-use with fine tuning for repeated speed optimizations. So I'm not sure how correct your statements are re: mental ray limitations with unified sampling.

It is up to us to help you, so I'm not letting myself off the hook. But there's lots of stuff that can be done now, that isn't quite communicated effectively and there's lots of stuff coming to even make it better.