Usage of subdivs (directional light)

[Micha_cg]

Hallo,

last I had an exterior scene and anything was good, except for the noisy shadows of the sun shadows (dir light). I render my scene in full adaptive mode (Vlados universal method) at noise level 0.02. In this case I don't liked to use a stronger noise level, because I need more subdivs for the sun shadow only, the other parts of the scene are looking good. I tested to set the sun subdivs at a high count, but it dosn't help. What can I do in this case?

Today I start a test for this scene type. My problem, if I use the Adaptive QMC image sampler, than the subdivs settings of the directional light show no effect. Is it inpossible to set a min subdiv level at the light options?

-Micha

[Micha_cg]

Usage of subdivs (directional light)

I will try to explain my problem with a simple example: a ground plane, a cuboid and a directional lightsource with a radius.



The first image show a smooth shadow. The Vray engine seems to ignore the adaptive subdivs and seems to use the shadow subdivs of the light.
Now we make a little step to the universal method, we set adaptive QMC max subdivs at 44 - the shadow looks very noisy - my problem of the exterior scene.



I can set the adaptive amount at 0.1 ... 1, nothing change. Also min samples 99 dosn't help. The only option that help, is the noise threshold. A noisethreshold 0.002 bring back the smooth shadow, but the rendertimes increase at 56s - the whole image is very high sampled. My goal is to use a noise threshold 0.01 for the whole scene and only very critical parts like the shadow should get extra subdivs per subdiv option. What can be done?

[flipside]

Usage of subdivs (directional light)

I have no idea why that happens, but why do you use the 1/44 method? Simply leave it at 1/4 then

You have much more control if you set all subdivs locally instead of using that universal method. That method is not aimed at speed, so if you want that, don't use it.

I always use 1/4, and qmc treshold 0.02 with other qmc sampler options at default.

Then for glossy reflective materials, use subdivs of 16 or higher (20,24 or smth). For light subdivs, also use 16 or higher (depending on the size of the light, sometimes I use 30).

If you use QMC GI in first bounce, you will need to increase that subdivs too.

[Gijs]

Usage of subdivs (directional light)

I remember Vlado mentioned something about that when the adaptive max setting is higher than material or light subdivisions, it will take over the sampling and then all noise is controlled through the qmc sampler.

[Micha_cg]

Usage of subdivs (directional light)

Thanks for the hints Wouter and Gijs.

@Wouter: I try to find a way to use the universal method (good for beginners too), but run in the problem, that my sun shadow was very noisy. Now, I wonder me, that I can not get both together. I have the feeling, something dosn't work like it should work. But I happy to read your hints for the subdiv values. I think, it's time to leave the universal method.

@Gijs: Oh, this is bad. So, the universal method is very limited. No chance to get more local samples at this way.

I replied in the thread. One other thing, I see that you have the min samples in your qmc sampler at 4. Try this at 8 or 16 and see if your noise problem goes away. In the manual there is something about this that says that when the min samples is higher, the early termination process will be more reliable.

The min samples dosn't show any effect. Interesting for me, the min samples show no effect in any case. A bug?

[flipside]

Usage of subdivs (directional light)

I also never noticed the effect of min samples, also in max... Maybe I didn't try extremes.

[Gijs]

Usage of subdivs (directional light)

I must say I have done a few experiments yesterday, and it seems it is very difficlut to get noisefree shadows without exessive rendertimes when the max value of AA subdivision is high. I also noticed no differences (in this simple case) with the min samples. There could be something wrong here...

[joe_b]

Usage of subdivs (directional light)

Micha - you only experianced problems with the directional light's shadows? I know that inherently the soft shadows are calculated differently for direct lights. I believe the radius when used for direct lights is used as some kind of angle comparison - whereas I think the other lights use the radius as a scene-unit distance. I don't know if this difference in calculations could account for this odd sampling behavior? I guess if you don't experiance issues with soft shadows for any other light source then I should ask Vlado to look into it.

[Micha_cg]

Usage of subdivs (directional light)

I'm not sure I understand you right Joe, but I tested a spot light too now - same effect. If I set adaptive QMC 1/100 (noise 0.01), than I find no way to get a smooth shadow except for I use a stronger noise level.

[joe_b]

Usage of subdivs (directional light)

Oh ok - I thought you said it was a problem only with the direct light. nm