Automatic "psyop" ID mattes

[monsteradurm]

@monsteradmurm:
You do not render and MMREs anymore. Only MtlID or ObjectID.

I understand that this is instead of rendering multi-matte render elements. I said multimatteID RE pass. Which if i use the expressions you described ( which are identical to the first one i had already written ) produces the exact same result - a matte without aliasing. Which is why I thought I needed a coverage pass.

If i render a materialId pass, with deepexr export for imageformat - i get a blank image.

[instinct]

I understand that this is instead of rendering multi-matte render elements. I said multimatteID RE pass. Which if i use the expressions you described ( which are identical to the first one i had already written ) produces the exact same result - a matte without aliasing. Which is why I thought I needed a coverage pass.

If i render a materialId pass, with deepexr export for imageformat - i get a blank image.

Tough to troubleshoot without any information. Basically you need to have a scene with materials that have materialIDs other than 0 and render to exr using the vray raw file output and turn on deep in the file dialog. That's it basically. I can try and provide a minimal sample. Don't know when i'll find the time though.

Cheers,
Thorsten

[monsteradurm]

VRay raw output

that is probably the issue then as ive never heard or seen anything pertaining to a vray raw output option in maya.

[instinct]

that is probably the issue then as ive never heard or seen anything pertaining to a vray raw output option in maya.

I have no idea how/if Maya supports that kind of deep output as we're not using Maya here :/

[monsteradurm]

I have no idea how/if Maya supports that kind of deep output as we're not using Maya here :/

Well this topic was originally posted in the vray for maya forum...

Anyway, as it turns out the coverage pass is what was screwing up my deep expression. Im not sure what setting was adjusted, but it had additional lines intersecting accross the z axis. Reverting to an older save output it correctly, and i could use it to multiply against the multimatteID pass to get sampled mattes.

With that in mind i tend to agree with Vlado. The cryptomatte plugin looks to be largely irrelevant. It is easy enough to spew out a quick text file with some json/ascii data pertaining to id names (such as shader/object names) that correspond to
the Id information. From then on its just a quick nuke script to create the appropriate layers in nuke and point them to the ids with a deep expression.

I know vray has an extra attributes section with the image format so perhaps the data could even be stored in there?


**Edit: Switching the attribute 'Deep merge node' to 'By z-depth' is what caused the coverage map to become unusable in this case.

[instinct]

**Edit: Switching the attribute 'Deep merge node' to 'By z-depth' is what caused the coverage map to become unusable in this case.

Ah, that makes sense. That will lead to samples for seperate objects to be merged if they are close enough. Whereas "by render id" makes for perfect object separation.

Cheers,
Thorsten

[jasonhuang1115]

Hey Vlado, is the Cryptomatte implementation actually on the to do list for potential nightly build sometime soon or?

[vlado]

Yes, yes, yes; if one more person asks me, I'll put it in giant letters on the front page.

Best regards,
Vlado

[jasonhuang1115]

Yes, yes, yes; if one more person asks me, I'll put it in giant letters on the front page.

Best regards,
Vlado

Thanks for the confirmation, vlado. Sorry if I sound pushy. Would be totally fine with me if you decided not to implement after considerations. Just want to know what to expect as I am investigating an alternative solution for our current multimatte based comp. Evotis is on my list too.

I read prman is looking at openexrid instead considering the mattes for hair, grass, and volumetrics. Maybe they have some similar concerns on cryptomatte too?

Thanks.

[Neilg]

There seem to be some misconceptions in regards to deep images and how VRay can store information in deep.

Do you know if fusion can work with deep images the same way, and do you have any resources that go over this?
Sounds like something I would like to have, but we dont use nuke...

[instinct]

I haven't used Fusion since DF 2.5 or so, so i would not really know, but i don't think so

[sdelalla]

In regards to fusion, there is a lot of info about the cryptomatte implementation and deep over in this thread:

http://www.steakunderwater.com/wesuc...0ef9512b34de58

[RichardBlank]

In a couple of projects we have been "forced" to use deep images. Our comp artists does not like them as much as I do. We haven't really found a satisfying way to multipass comp deep images. Is it possible to multipass comp deep images in nuke without going from DeepToImage-DeepFromImages witch makes the deep mattes look bad?

[vlado]

Why would it make the deep mattes look bad? Can you post a screenshot of your Nuke network?

Best regards,
Vlado

[instinct]

Well to be honest it's pretty pointless. While DeepToImage and DeepFromImage are valid in quite some cases they should not be your main means of comp. There is quite a lack of nodes that are needed to do proper deep comp in nuke. We have created most that we need inhouse. We'd welcome a broader initiative of other people using at and vendors supporting it. VRay support for deep is actually pretty stellar in my eyes. Nuke's not so much. And everyone else hardly existent. It's a chicken egg problem. The Foundry say they won't waste resources as long as only few use it, and few will be using it because the stock toolset is lacking.

I am not sure what you mean by "looking bad" and i'd be interested in seeing screenshots or an example too. But by going flat you are flatting all your deepsamples, and while recolor applies your changes back to the samples you will not get the separation of samples as you had before. You will get the samples, but the split RGB values will be eradicated i guess.

Cheers,
Thorsten