Black square artifacts from true refractions (Mantra PBR)

[OskarSwierad]

Hi,

What can cause such blocky screen-aligned artifacts?

The scene file: http://oskarswierad.com/public/houdini/HoudiniDispEarth02.hipnc.zip

First, I used just a Mantra Surface shader and got these glitches. So then I tried Ocean FX - the issue still remained, though.

It happens only with refractions and Physically Based Rendering. Raytrace engine doesn't have this problem.

edit: They seem to disappear above a certain environment brightness value?...

Oskar

Edited December 19, 2015 by OskarSwierad

[MENOZ]

if you turn off ray variance antialiasing it removes these artifacts. ray variance doesn't work very well in low light conditions as far as I know.

 

I don't know why they are blocky anyway. I remember it was a problem of sampling distribution, and was more uniform, usually you had jagged shapes, and was noisy, not so blocky. Maybe in H15 they changed some of the architecture behind and it changes the way it distributes / generates the samples.. I'm just making up things here.

Edited December 19, 2015 by MENOZ

[fathom]

i don't understand the ray variance anti-aliasing thing, but it sounds like it tries to pick where to send samples based on image content as it's rendering, rather than just a stochastic random sample across the entire bucket.  could be it's not very smart.

 

it'd be interesting to turn on some diagnostic image planes and see what the sampling looks like.

[tinyparticle]

Try increasing your min ray samples.

[dennis.albus]

Yes, min ray samples works wonders. I always have it at a minimum of 2 or 3. With just 1 minimum sample chances are the variance estimation is wrong due to too few stochastic samples.

As your reflection is very faint please also play with the Minimum Reflection Ratio "pbr_reflectratio". Upping the light samples might also work in your case.

 

Mantras variance estimation is indeed not very smart and produces artifacts quite frequently which can only be removed by upping the overall sampling quality. This seems to contradict the usefulness of an adaptive approach.

I tried to hack a better algorithm into the integrator, but it is not that easy to come up with a really good solution.

 

Cheers,

Dennis

[OskarSwierad]

Thanks for your replies.

 

Yes, min ray samples works wonders.

 

You're both right, it solves the problem.

So when the minimum samples amount is too small, it can lead to a completely dark block initially? that's why it discards the block later, not willing to invest any more samples?

Edited December 22, 2015 by OskarSwierad