shadow and spec flicker problems

[sharktacos]

Questions about flicker:

#1 Hair: We are getting flicking on the spec pass for our hair using the VrayMtlHair. This material has no samples like the vrayMaterial does. So what is the best way to deal with this? The GI pass is flicker-free.

Here's our globals settings:
min: 4
max: 10
thres: .005

We can of course up these, but I'm thinking that since the flicker is local to the fur/hair that it would better to tweak that directly. Just can't see where...

thanks!

[joconnell]

If you've got no spec / ref samples in the vray hair mat and you've got shadow flicker then your best bet is to up your light samples. For highlights, when you have the option in your material, you can either up your reflection samples so it'll clean up, or up your light samples so the material is reflecting a cleaner object / light source.

[sharktacos]

If you've got no spec / ref samples in the vray hair mat and you've got shadow flicker then your best bet is to up your light samples. For highlights, when you have the option in your material, you can either up your reflection samples so it'll clean up, or up your light samples so the material is reflecting a cleaner object / light source.

Thanks, we can do this with the shadow quality of the lights.

For the spec we can't increase the sample quality of the light because it's a directional light which does not have any samples (unlike all the vray lights). So we still have the issue with the spec.

[joconnell]

Ah - might be worth switching to a vray rectangular light with a heavy directional value? As usual, you'll get the nicest results sticking with purely vray bits.

[sharktacos]

Ah - might be worth switching to a vray rectangular light with a heavy directional value? As usual, you'll get the nicest results sticking with purely vray bits.

The directional light is for sun light. I'd think a rectangular light would not work for that as it would only effect a certain area and have very diffused shadows if it was scaled larger.
I considered using a vray sun, but don't like the way it colors the light with ozone and such. I'd like to have the light color be pure white.

[sharktacos]

Actually, now that I look at it, the vray sun light does not have any sampling either (just shadow samples, which the directional light also has). So I'm assuming that the nature of these lights, which have no decay, means they don't need samples. So I'm back to the hair material and the spec flicker.

[Morbid Angel]

I don't think it would matter if you raise the samples on the hair specular (if it had any), since the hair is so thin, intersecting a really small object (light) like a sun which generally is small, would be quite difficult. I've done a project not long ago with fur, where I had similar issue, no matter what I did, I had flickering in the hair. It came down to the only way to solve that particular issue was to use universal settings method. Which would be for me 3/100 at 1/0.008 *(depending on your initial render times of course) you could lower that value even more. This all though too a while to render a frame, produced a stable result.

[sharktacos]

When you say "3/100 at 1/0.008" I assume this means:

min subdivs: 3
max subdivs: 100
threshhold .008

Is that right? I'm not quite sure what the 1/0.008 part means...

[Morbid Angel]

Yes, dmc min 3, max 100, adaptive threshold 1 (full adaptivity) and noise threshold of 0.008 or whatever gets a good result. By being full adaptive, it may take much longer in certain areas of the image to achieve necessary sampling, but in my experience it's what needs to happen to get a clean result in such cases. Either this, or raising the max samples to about 3/24 and then dropping the threshold to a really low value like 0.002 or 0.001, which pretty much will be the same, but perhaps will produce cleaner image in faster time.

[sharktacos]

Yes, dmc min 3, max 100, adaptive threshold 1 (full adaptivity) and noise threshold of 0.008 or whatever gets a good result. By being full adaptive, it may take much longer in certain areas of the image to achieve necessary sampling, but in my experience it's what needs to happen to get a clean result in such cases. Either this, or raising the max samples to about 3/24 and then dropping the threshold to a really low value like 0.002 or 0.001, which pretty much will be the same, but perhaps will produce cleaner image in faster time.

Understood, thanks. We currently have it set to min 3, max 10, thresh .005, adaptivity 0.85. Will try adjusting that accordingly.

I have to say that this all makes me wish that there were sample settings on the the hairMtl so that we could instead have the adaptive amount at say 0.8 and then bump up the samples on the hair material really high so it gets lots of samples. This is how I would approach flickering on everything else, rather than going for the "Nederhorst" method of 100% adaptivity which is a lot less efficient than a targeted approach. I believe this has become the prefered approach for most folks in fact. Rather than rendering with full adapativity, you instead raise the samples on the material or light where it is needed. It's just that here you can't.

So I guess I'm asking for a feature request. I wish the hairMtl had sampling. I'd be curious to hear from Vlado if there is a reason why this would not be a good thing.

P.S. I also wish the hair material had material ID attributes.

P.P.S Oh, and I'd like a pony too!

[sharktacos]

Yes, dmc min 3, max 100, adaptive threshold 1 (full adaptivity) and noise threshold of 0.008 or whatever gets a good result. By being full adaptive, it may take much longer in certain areas of the image to achieve necessary sampling, but in my experience it's what needs to happen to get a clean result in such cases. Either this, or raising the max samples to about 3/24 and then dropping the threshold to a really low value like 0.002 or 0.001, which pretty much will be the same, but perhaps will produce cleaner image in faster time.

Oh, and Dmitry, I wanted to ask a question about your statement that you can lower the max samples with a lower threshhold. My understanding is that the threshhold looks at the differences between neighboring pixels and determines if they differ enough (based on the threshhold value). Once this has been determined Vray then applied the max samples to it. If there are 24 available it could get that many, and if it's set to 100 it could get that many. So lowering the threshold to .001 (thus making it more sensitive) would raise the likelihood of a pixel being flagged as needing samples, but the lower max sample value of 24 would mean that there was less available to give it. Having the threshhold set higher (to .008 )would mean less pixels would be flagged as needing samples, but there would be more samples to help with (max 100). So less need with more available (max 100, thresh .008 ), or more need with less available (max 24 thresh .001).

In short, I'm not following why lowing the max samples and dropping the threshhold would help. I suspect that you know a lot more about this than I do, and that I'm just not getting something here, so I'd appreciate it if you could help me to understand this better and so see what it is I'm not getting. Thanks!

P.S. writing the number 8 with an end parentheses apparently automatically makes a smiley face so initially my post read: (threshold .00

[Morbid Angel]

I like to think of it like this: high sensitivity does not need lot of samples, lower sensitivity may need more samples. It really is a case dependent, for example for a longest time I've used a method much like what you described, raising shader/light subdivs rather then dmc samples, but then I came across a specific situation, when this exact method would result in a huge render time, in my case very complex scene which required cg dof. So I've started using the universal method and that worked well, but still created long times. I then spent some time trying to figure out if I could improve it, and came up with this solution. My renders still take about 4-6 hours per frame, but its much better from projected 20 or so before