adaptive subdivision AA, higher setting=more noise

[Gijs]

The following occurs in RC2. QMC settings default (noise tresh. 0.01)
domelight with 20 subdivisions and material with 20 subdivisions.

Look how shadow noise increases with each step. the noise in the material decreases though.

[vlado]

Are you sure this is with adaptive subdivision and not adaptive qmc?

Best regards,
Vlado

[Gijs]

sorry...it's adaptive qmc.

[vlado]

In this case, please see my reply in this thread:

http://www.chaosgroup.com/forum/phpB...ic.php?t=17652

Best regards,
Vlado

[Gijs]

I have been reading that thread, but it still confuses me. If you say that more of the sampling is handled by the image sampler, why then does the material noise decrease, but shadow noise increase when the max samples are increased?

Also how would you explain that the noise increases going from 16 to 32 subdivisions, while the noise treshold remains the same?

It makes me wonder, is the adaptive QMC designed to work with 1/4 and should we just stick to that, and solve noise issues only by changing the noise treshold, material settings and/or global subdivisions?

[vlado]

Also how would you explain that the noise increases going from 16 to 32 subdivisions, while the noise treshold remains the same?

Because the Noise threshold is too high, which causes less rays to be traced.

It makes me wonder, is the adaptive QMC designed to work with 1/4 and should we just stick to that, and solve noise issues only by changing the noise treshold, material settings and/or global subdivisions?

You can use whatever you want for the adaptive QMC samples, but you may have to lower the Noise threshold.

Best regards,
Vlado

[studioDIM]

Also how would you explain that the noise increases going from 16 to 32 subdivisions, while the noise treshold remains the same?

It's very dependant from MATERIAL settings as well.
Especially when you have glossies, if the adaptive amount is small, with 1-4 the material sampling (in my scene was at 32 for the glossy refs) would take priority, and the AA would do the "easy" bits on top of the good job the material did.
Raising AA max samples biases the sampler towards the AAser, skipping the material sampling more, and producing -with the other settings left at the same values- a noisier result (16 samples max , versus the 32 in the material).

I *think* lol

Lele

[Gijs]

Thanks for the explanation.
At first it seems a bit counter intuitive that, when the noise treshold is the limiting factor, that the shadow on a simple diffuse material is silky smooth with 1/4 and extremely noisy with 1/32.
It looks to me that high adaptive QMC doesn't work well with blurry shadows, because in the example it is clear that increasing the QMC AA settings improves the blurry reflections, but shadow quality decreases.

I think the final conclusion is that the optimization of the Vray light shadows cannot be matched by the qmc AA image sampler when it takes sampling over from the vray light, whereas for sampling materials the image sampler does a better job taking over from the material.

[vlado]

I think the final conclusion is that the optimization of the Vray light shadows cannot be matched by the qmc AA image sampler when it takes sampling over from the vray light, whereas for sampling materials the image sampler does a better job taking over from the material.

It can be matched, of course, but it would need a lower noise threshold. Let me see if I can come up with some images...

Best regards,
Vlado