Noise threshold per object or material?

[cecofuli]

Yes Peter, I see this. But for the interior the rendering time will be x3 time with or without sub pixel mapping ON/OFF... so, 5 hours vs 15 hours... With IM+LC or the old BF+LC, we don't have this huge difference in rendering time and looks good (in terms of brightness)

[Morne]

With IM+LC or the old BF+LC, we don't have this huge difference in rendering time and looks good (in terms of brightness)

what do you mean "the old" BF+LC?

[cecofuli]

Not the "new" US ( AA 1 100, Adaptive amount 1.00 noise 0.01 etc... )
But: AA 1 4, Adaptive amount 0,85 noise 0.005 etc. and play with all subdivs setting for shader, lights, DOF etc...
I know, it's more easy the US, you must change only a few parameter, but you have less control.
For example, in my test, with the "old method" I can have low rendering time with the same noise. For example:

BF + LC:
BF: 32 subdivs
LC (as usual)
AA 1 4
All subdivs shader 32 ( I change with script)
All light subdivs 32 ( I change with script)
0.85 and 0.01
In static rendering I prefer turn off all the filter and use more subdivs.
In this case, I can use sub pixel mapping OFF without too much slowdown in rendering time.

[zeronove]

ciao Francesco,

do you have a test scene to illustrate this?
I'm asking because nowdays for me it's very hard to use so low AA values
I usually work with IM/LC, could be BF as primary the reason of my AA value going to the roof?

[cecofuli]

Ok zeronove, lancio dei test abbastanza estremi a poi li posto qui.
Ok zeronove, I do some extreme test and I will post here.

[cecofuli]

Ok, this is a very simple scene. I want to focus the attention on the ceiling, the right wall and the curtain.





As you can you see, we can to compare (for quality vs rendering time) the:

(1) OLD BF+LC at 64 subdivs vs US with 0.005 (b) and (f)
(2) OLD BF+LC at 32 subdivs vs US with 0.01 (c) and (e)

The rendering time is near the same
In the old method we have less noise in the ceiling, but it's not possible to remove the noise in the curtain (2SideMat), also with 64 subdivs. I don't know why... (EDIT: maybe the bmp filtering I forgot to change from "None" in "pyramidal"!)

At the end, I think I was wrong and yes, if you want a "similar" unbiased method, the US is the best.
About the jagged edges (see the arrow in the rendering), unfortunately, with US we cannot use Sub-pixel mapping and Clamp ON, or we will have a dark rendering as a "g" Rendering.
So, we must use a AA filter with +20% rendering time. And sometimes this cannot solve the problem...

PS: The beige wall and ceiling have a fine bump map. Also, the dark big lamp in the center has a fine textures for the diffuse and bump. So, some "noise" comes from here.
I think with a neutral white shader I could have more clean rendering. Just to know, this is the "drawing" from the project manager =)

[Morbid Angel]

just curious the fine bump on the ceiling, is it filtered? if its not will create more noise, also...to me if you cant see the bump, why have it there? it does not make the image better, but create more grain.

[cecofuli]

Hi Morbid, yes is filtered, as usual with all default bmp (Pyramidal). As you know, we don't use the filtering only in the opacity maps. (opps... in the white,diffuse, translucent curtain I have the filtering OFF due to copy/paste from opacity slot!!!)
I use the fine bump because at 4000x3000 you could see in the near walls. In this scene, the pillar in the center, for example. This small reception is only a little part of a big interior project. Just for test, now, the same image at 2500x1500 px with x3 I7 esacore 3.8 Ghz with the US and 0.005 give me 6 hours =)

[Morbid Angel]

Hi Morbid, yes is filtered, as usual with all default bmp (Pyramidal). As you know, we don't use the filtering only in the opacity maps. (opps... in the white,diffuse, translucent curtain I have the filtering OFF due to copy/paste from opacity slot!!!)
I use the fine bump because at 4000x3000 you could see in the near walls. In this scene, the pillar in the center, for example. This small reception is only a little part of a big interior project. Just for test, now, the same image at 2500x1500 px with x3 I7 esacore 3.8 Ghz with the US and 0.005 give me 6 hours =)

Couple of things I can suggest. Try without the bump completely and see if noise is gone. If it is, then try convert your bump to tiled exr, and use vray hdr loader, may help and also exr stores much more accurate data in terms of color, as oppose to srgb/8 bit.

[super gnu]

may i also suggest putting the bump in a fallof map set to distance blend? that way you can have it visible in the foreground where you need it, but fading to nothing as you move away from camera so it doesnt give you grain.

[cecofuli]

Thanks for the suggestion. I will try in my next project

[zeronove]

ciao Francesco,

thanks a lot for this ultra detailed test; anyway I'm very surprised about your results using a not so high mtl subds and quite ordinary AA settings; they look not so different from settings in use pre 2.0. Actually I always tend to boost AA a lot more than 1/4 to solve antialiasing and, as consequence, I need to raise mtl/lights subds as well; I'm still confused about 2.0, maybe I need to go back to test it deeper.

I've just forgot to ask it, do u have a samplerate element already saved for your test renders, just curious to see how it looks, thanks in advance

[cecofuli]

Here the render pass