Cryptomatte and DR

[Billabong]

We have been having this issue with rendering Cryptomattes using Deadline's Distributed rendering. All of our other passes render out fine. While the render is going the crytomatte looks the way it should, but as soon as its done and its going through the denoise phase, the CM pass looks like a bunch of dropped buckets. We are trying to determine is this is a deadline or Corona issue. I feel like if it was Deadline, then all the passes would look like the attached image.

 Has anyone else had any sort of issues like this?

 Thank you for any help on this.

-B

[TomG]

You can always ignore how the Cryptomatte looks in the VFB etc. The pass itself is not "Cryptomatte" just a quick visual representation. So the question is, are you saving to CXR? If not then you won't get Cryptomatte data. If you are saving to CXR, when you rename it to EXR and load it into something that can recognize Cryptomatte data, is it working?

[Billabong]

This is what it looks like after saving a .cxr, renaming the file to be an exr and opening that up in PS

[TomG]

If you watched the video, you can see that opening it in raw Photoshop is not going to show the Cryptomatte data, as PS doesn't understand it (it does terribly with EXRs). Did you have EXR-IO installed, and use the settings recommended in the video? (because once loaded, Cryptomatte data is not one layer any more, but one layer per masked item).

[Billabong]

Here are some screengrabs of everything i did. Also notice the error I get about the amount layers the CM data wants to include. Plus when opening this exr. it is giving me a ton of layers I did not add or for that matter ever seen before.

[TomG]

That ton of layers - that's the cryptomatte :) There is one layer for each "thing" that is in cryptomatte (what the "thing" is depends on whether that's by object, etc.) Each of those layers you have never seen before is a mask for a particular part of the image. All this should match what is in the video, btw.

In terms of that number of layers, I'd return to Corona and change what you are dividing the Cryptomatte by. That is way too many layers to work with.

[TomG]

At a guess, it was using an ID of Name and there are lots of objects with unique names in there (something where every leaf has a unique name, or there are 100,000 trees each with a unique name). Best idea is setting up Layers and using the Layer option, though you may find Hierarchy works well enough to reduce the number of cryptomatte layers.
https://support.chaos.com/hc/en-us/articles/4954305024529-How-to-use-the-Corona-Cryptomatte-3ds-Max-

[Billabong]

Hey Tom,

 So it does seem like this is a Deadline DR issue. I sent the same render to the farm and you can see in the attachment, I was given what I needed. My deepest apologies for your troubles.

[TomG]

No troubles at all! Always happy to help, and Cryptomatte is very different from the other masks so it often raises questions. Interesting that running it through Deadline gives an unusual or even "wrong" result with so many layers. It might be something we can look into with the Deadline team (may help to raise a ticket with us at https://support.chaos.com/hc/en-us/requests/new now you have narrowed down the cause - provide a link to this thread in the ticket, if you would). Sorry for your own troubles in having to dig so deep to find out exactly what was going on!

[Billabong]

That would be fantastic if you guys could do that! I'll make the ticket now.

[maru]

Sorry for the delay! I took us a while, but we have managed to reproduce the problem. It is now logged for our dev team to investigate with a sample scene and other necessary data. Thank you for reporting this.
It seems that unfortunately Cryptomatte cannot be currently used with Corona DR (with Deadline or without it). The only workarounds I can recommend right now are:
- Rendering Cryptomattes without Corona DR (locally only)
or
- Using Corona DR, but without Cryptomattes (using other masking render elements)
I will be sure to update this thread once we have another solution/workaround available for this and when it is fixed.

(Internal ID=1183721762)