Cryptomatte reflections/refractions sampling

[Codi]

Hello,

we are testing Cryptomatte and have found that the masks on reflections and refractions are very noisy. If we try to generate the same masks with Multimatte render element they look much better. Is it a known limitation of Cryptomatte? I have tried raising min/max subdivs and lowering the noise threshold but the results seem to be the same. Is there a setting to avoid this problem?





We are testing it in VRay version 3.60.05 in Max 2018.

[fixeighted]

This is not supported by Cryptomatte afaik
I didn't even realise that multimatte produced the result you've shown...although it's not really a matte I guess.
I found this post in another forum with some info from a dev re. this. https://forum.isotropix.com/viewtopic.php?f=5&t=5019
Athough it occurs to me that if cryptomatte could 'perceive' anything seen through refraction to be opaque, then it could work, as it does have the capacity to matte blurred objects already. But then maybe I'm not thinking straight

[Codi]

Thanks for the pointer fixeighted. Now I realize it's stated on the docs:

Compared to the Multimatte Render Element, Cryptomatte offers the following:

• Does not require a setup with object IDs, etc.
• Only requires a fixed number of additional render elements, typically four.
• Cannot be propagated through reflections or refractions.

How do comp people handle those cases, combining Cryptomatte and Multimatte? with wirecolor?
Any chance to have support for reflections/refractions in the future?

[fixeighted]

Well in that post I linked it suggests..... 'You will have to do split between front and backside of the glass and then apply a refraction matte for the distortion.' which I guess would work just fine...just not as exact or usefully in one file. But hey!...crypto is still the most awesome thing regardless

As for if it'll be supported in the future, I have no idea but hope so. That same post comment by one of the devs seems to suggest it may be possible, though that post was from 2014 so
if it's not implemented even now then maybe there is something critical that prevents it....

[fixeighted]

..hmm...oddly I just found this comment by someone calling themselves "Vlado'....no idea who that may be

"We've done some work on that for the next version of V-Ray, at least as far as refraction is concerned (objects visible behind glass etc)."

This was from the Github post in Dec 2017.

A post just above that one was from one crypto dev saying it is in fact possible but handled largely by the renderer.

Is this still being worked on Vlado?

https://github.com/Psyop/Cryptomatte/issues/48

[Codi]

Well in that post I linked it suggests..... 'You will have to do split between front and backside of the glass and then apply a refraction matte for the distortion.' which I guess would work just fine...just not as exact or usefully in one file. But hey!...crypto is still the most awesome thing regardless

Yeah, I meant how to achieve it in a single render. Agree that, despite this problem, it is pretty useful

..hmm...oddly I just found this comment by someone calling themselves "Vlado'....no idea who that may be
"We've done some work on that for the next version of V-Ray, at least as far as refraction is concerned (objects visible behind glass etc)."

I couldn't find any mention to it on release notes, but I speculate the comment refers to some change made in VRay between versions 3.60.03 and 3.60.05. The former wasn't able to render mattes on reflections/refractions, in 3.60.05 you can do it using "all channels" material properties, only that they seem to lack samples.

Might be wrong tough, hope there is some way to achieve this.

[Codi]

Any word from the devs regarding this behaviour? Are there plans to improve support for reflections/refractions in the future?