Cryptomatte edge issue?

1,429 views
Skip to first unread message

T MUSH

unread,
Jul 25, 2016, 12:21:11 PM7/25/16
to alshaders
Hey guys, trying out cryptomatte and gotta say its great so far but I'm getting an edge outline when using the mask from the cryptomatte node, even when using the example files? Am I just being a bit dull and thats how its meant to be or is there a step I'm missing out? 

thanks
TM
cwm

Anders Langlands

unread,
Jul 25, 2016, 5:00:14 PM7/25/16
to T MUSH, alshaders
Do you have an example nuke script using the example files I could take a look st?
--
You received this message because you are subscribed to the Google Groups "alshaders" group.
To unsubscribe from this group and stop receiving emails from it, send an email to alshaders+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Tafadzwa Mushawasha

unread,
Jul 26, 2016, 7:27:32 AM7/26/16
to Anders Langlands, alshaders
​I used the example Buddha scene as a test, hit render, took the passes into nuke and dropped a crypt node on it but Im getting an edge issue, what am I doing wrong?
many thanks

s.des...@gmail.com

unread,
Sep 26, 2016, 11:57:33 AM9/26/16
to alshaders, andersl...@gmail.com
Hello, I've just tested cryptoMatte on a simple scene with some cubes and I also have a problem with the outline edges in nuke.
Is there a special workflow in nuke to use the crytptomatte ?

At the moment, I connect a grade to my beauty pass. 
I plug the alpha generated with crypotmatte into the grade mask slot.
If I increase multiply or gain, I see an outline on my cube.

Thx,

Seb

Fabrice Altman

unread,
Jan 16, 2017, 8:42:57 AM1/16/17
to alshaders
Hi,

we have exactly the same problem with edges. Maya, AL, Nuke all up to date. Any progress made on this?

Cheers,
F.

Jonah Friedman

unread,
Jan 16, 2017, 11:47:56 AM1/16/17
to Fabrice Altman, alshaders
Hi Fabrice,

You may have some kind of real issue, or you may just be running into the limits of trying to modify an image with a matte. Cryptomattes are accurate, as accurate as any other kind of matte, but they aren't magic- they can only produce a single alpha channel which indicates how much of the pixel is occupied by the object in question. 

If you want me to look at something, please do the following test: Render a regular RGB matte for whatever it is, and compare the results to the cryptomatte. If things are the same, you don't have an issue but are just running into the limits of mattes. If the Cryptomatte is different, then please do send me a reproduction. 

Copy and pasted from another thread about the general limits of any kind of matte (including deep ones): 

Your edge pixels contain both your grid, and your yellow cube. Let's say, a pixel is divided up 50-50. Your matte tells you that pixel is 50% cube. Great, so if you darken your yellow cube, you darken that pixel 50% as much as the pixels in the center. However, what you've really done to that pixel is darkened both the grid and the yellow cube half as much as the other yellow cube pixels, creating a fringe. A good compositor can also sometimes work around these edge issues but it is case by case. However, these mattes are exactly as accurate as RGB mattes. 

Sample compositing (read: deep images) can potentially solve this, however using deep mattes used to modify shallow images does not. You would need deep beauty images as well as deep mattes to separate the samples properly. 

 

--
You received this message because you are subscribed to the Google Groups "alshaders" group.
To unsubscribe from this group and stop receiving emails from it, send an email to alshaders+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.



--

JONAH FRIEDMAN


psyop_logo_signature.png


PSYOP

NYC | LA

OFFICE: +1 (212) 533-9055

MOBILE: +1 (347) 873-3992

45 HOWARD ST, 5TH FLOOR, NEW YORK NY 10013

b-facebook-128.png   b-twitter-128.png   b-instagram-128.png

Reply all
Reply to author
Forward
0 new messages