CRYPTOMATTE

[Jolbertoquini]

Hi Guys,

I find this tools here really interesting for mattes do you think would be possible to make work with Octane I just wondering?
I know before I add chat about coverage pass with depth and motion blur maybe I need render two ID render and then create de difference but in this case doesn't work, so with OC3 still possible to add a coverage pass???

http://www.psyop.com/news/view/new-cryp ... yond-psyop

and the good news is free for now they try to add on render engines right now works with Arnold but I'm sure we could use specially Nuke users.Like me heehe

Download link :

https://github.com/Psyop/Cryptomatte

Very best,
JO

[calus]

Good stuff !!

Here is the direct link to the specification paper, to better understand what this is about :
https://github.com/Psyop/Cryptomatte/ra ... poster.pdf

and how this is used:
http://www.anderslanglands.com/alshader ... matte.html

"Cryptomatte, developed at Psyop, is a system that allows you to generate perfect mattes of an arbitrary number of named objects from a single OpenEXR AOV with just a few channels. The generated mattes exactly match the beauty image and can properly handle motion blur and depth of field."


So the Cryptomatte open source tool to extract the perfect matte is implemented in Nuke,
and the needed ID passes are implemented in the open-source Alshader rc 17 for Arnold.
(source here : https://bitbucket.org/anderslanglands/a ... c17.tar.gz )

Now all we need is to follow the same specification for the ID pass in Octane.
I guess maybe it's already possible to output the correct ID passes with scripted node texture and texture pass, or using OSL when it's there.
But obviously a native implementation for the ID passes would be the best...or maybe something like this is already at work for the Octane_Imager ?

Would like to ear what Octane team think about this !

[sohara20]

There is also this implementation from Guerilla Render https://www.youtube.com/watch?v=W_ltvSMrwnQ.

https://github.com/MercenariesEngineering/openexrid

and this
http://cgpress.org/archives/evotis-sub- ... ystem.html

[calus]

, it took only 4 hours for the Cryptomatte Matte ID pass system to be implemented in Blender Cycles...

https://developer.blender.org/D2106

[doca]

Cryptomatte is extremely useful!!! Octane is really missing some matte options, at least standalone that I use. I am trying fully to switch from Arnold to Octane, although Arnold is much more versatile in therms of small useful things around it in production, but such a thing like Cryptomatte is really making me think more. Please add Cryptomatte to Octane!!!

[abstrax]

I just read through the presentation and first I thought: Great finally a matte system that is useful and lets you separate various contributions to a pixel, but then I realized that this is not the case, i.e. they are still storing only one RGBA value per pixel. They have a whole bunch of information about the object/material IDs contributing to a pixel, but no way to figure out the RGBA value of that contribution. If you for example would like to change the colour of the pink flower to blue in the small cutout of figure 5 you still can't, because the compositing tool still only knows that say 30% of the pixel colour is contributed by the flour and 70% by the bunny in the background, but has no information about the colours of the flower and the bunny.

-> Did anyone try out the system? Does it actually help in post beside making the organization of matte IDs more convenient?

[Jolbertoquini]

I just read through the presentation and first I thought: Great finally a matte system that is useful and lets you separate various contributions to a pixel, but then I realized that this is not the case, i.e. they are still storing only one RGBA value per pixel. They have a whole bunch of information about the object/material IDs contributing to a pixel, but no way to figure out the RGBA value of that contribution. If you for example would like to change the colour of the pink flower to blue in the small cutout of figure 5 you still can't, because the compositing tool still only knows that say 30% of the pixel colour is contributed by the flour and 70% by the bunny in the background, but has no information about the colours of the flower and the bunny.

-> Did anyone try out the system? Does it actually help in post beside making the organization of matte IDs more convenient?

Hi Abstrax

I use similar system in our workflow. we use Mentalray Coverage pass and ID to achieve similar results we also have a tool for Nuke with similar options. We are not using ID from Octane because without "Coverage pass" it's impossible to get a nice alpha the alpha.

I understand your points but render but most of time it's for small tweaks and the many ID workflow is just hard and take a lot renders and space on the server.

it's much easier to provide two passes for 2D department.

I didn't try this system yet, because we are not Arnold User. We may have a licensing some where to try, but I toke the passes and tried with nuke and is looking totally fine and we get better results compare to our version with MR because this workflow, we can have a alpha from depth and motion blur this the reason we are quite excited to have something like this at Octane to only use Octane on hour Pipeline.

Please let us know if we could do something about it, I already create a topic before about this ID and alphas options.
But the answers was would be really hard to get Coverage pass from Octane. then one of developer propose me to render two ID pass one with less sampling and other with more and then calculate the difference and etc..

Doesn't work good as MR passes. or this new Arnold options.

Very best,
JO

[calus]

they are still storing only one RGBA value per pixel. They have a whole bunch of information about the object/material IDs contributing to a pixel, but no way to figure out the RGBA value of that contribution. If you for example would like to change the colour of the pink flower to blue in the small cutout of figure 5 you still can't, because the compositing tool still only knows that say 30% of the pixel colour is contributed by the flour and 70% by the bunny in the background, but has no information about the colours of the flower and the bunny.

Right,
there is no such thing as a perfect mask without subsample information.

This seems to be doing the right thing :
http://www.go-ghost.com/

[Jolbertoquini]

they are still storing only one RGBA value per pixel. They have a whole bunch of information about the object/material IDs contributing to a pixel, but no way to figure out the RGBA value of that contribution. If you for example would like to change the colour of the pink flower to blue in the small cutout of figure 5 you still can't, because the compositing tool still only knows that say 30% of the pixel colour is contributed by the flour and 70% by the bunny in the background, but has no information about the colours of the flower and the bunny.

Right,
there is no such thing as a perfect mask without subsample information.

This seems to be doing the right thing :
http://www.go-ghost.com/

Hi Calus,

Is great one but I'm not sure they support depth and motion blur as cryptomatte. this could be something interesting if they support that to.

best,
JO

[doca]

-> Did anyone try out the system? Does it actually help in post beside making the organization of matte IDs more convenient?

Yes, It really helps. I still have not used it in production, but tried demo. It can be seen immediately that compositing is much easier with it. I am not fun of material ID or object ID passes, actually, I think they are quiet useless in production, but Cryptomatte makes a difference. From the rendering point of view, as far as I know, there is no way to render all the mattes with beauty render in Octane as a pass or few passes, correct me if I'm wrong. I am talking about running render once, not as much times as I need mattes, especially because Octane is gpu based render and most of the users probably do work on one computer.