Multimatte interesecting objects issue

[thanulee]

Hi, i ve noticed sometimes that multimatte element has aliasing issues. Seems when objects intersect with different object id?
Any thoughts?
see image below.

thanks!

[joconnell]

Your aa seems really low on this? Are you doing a material override or trying to make everything black in your scene for utility renders? What happens if you raise min aa to 2?

[thanulee]

My aa is fine. Ofc if i go unnecessary high it might fix. If i had a better quality image, u d notice that aa is perfect except where 2 objects intersect.
I use multimatte element with object ids 1 + 2 for R and G.

There must be a solution for this. I dont wanna create new scene/overrides/extra setups, just to have proper mattes. Cause ofc if i use a pitch black scene and use vraylight 255 red and green, everything is fine, if i use scanline is fine etc etc. There are infinite ways to do this, but i want the efficient one, cause i have changes in my scene all the time and that means that my mattes should render perfect along with the rest of my passes so get instant replaced in my composition.

Any other thoughts on this? Cause it seems like a bug.

thanks a lot

[thanulee]

Here is one more screenshot to see it better

[joconnell]

What does your beauty pass look like?

[thanulee]

Dont have it here, its perfectly fine, i wouldnt post otherwise. This is a problem i ve bumped into many times and i thought its time to ask

[joconnell]

The sampling of a multimatte is directly driven by the level of anti aliasing of the beauty pass so if you've got lower sampling along that joining edge in your beauty, it'll mean lower sampling of the multimatte too. I wanted to see the beauty just to check what issue is there that's making vray not bother with more aa along that edge. Are the two objects the same colour so that in the beauty they blend into each other?

[thanulee]

I ll post it 2morrow its perfectly fine though, more aa would be excessive for this image. There is something else going on i suspect not related to AA.

Cause if i just use vray light pure red, vray light pure green and rest pitch black, there will be 0 aa issue in the image i suspect. I ll confirm 2morrow.

I had the exact same issue in another file, and when i just did this vray light method, then all my mattes were perfect. But as said, this is completely unwanted. When working alone is fine, but when u have to handle ur files to others to continue, they do not have to fiddle with 1 scene for beauty and 1 scene for mattes. Or they cannot go everytime and after beauty renders, assign different materials to different objects all the time, mistakes are made like this, project folders get unessesary loaded with extra files and eventually after 20 client changes, ur just screwed

[joconnell]

Yep. If you assign the pure green and pure red what'll happen is when vray renders the beauty, there'll be such contrast along the edge where the two objects meet, it'll automatically use more than the standard 1 aa in those areas and you'll get a clean edge. Typically vray uses contrast in the scene as a way to decide if something gets low or high aa, if you've got something really low contrast and flat, vray won't do much but if there's loads of areas where different values meet then it'll start to raise up the aa. Again the multimatte will use whatever amount of sampling was used for the beauty hence me thinking that the two objects are very similar in colour and kind of camouflaging the edge where they meet and thus vray isn't bothered making more effort along that edge. Where the objects lie against the empty background though there's a tonne of contrast and thus it'll do way more sampling on those pixels.

There was talk of having an option in the multimatte element so that it had an influence on the beauty aa sampling (similar to the consider for anti aliasing tick box in the extratex element) but there's been nothing done about this yet.

[thanulee]

Hmm yes makes sense what ur saying here, now that i remember the most recent cases i had this issue. Dont know if thats actually the reason, but yea, seems legit explanation.
I need some dev word on this though, how they gonna tackle an issue like this. Cause even raising the AA if there is not enough contrast from the beauty, wont raise the samples anyways, unless u do fixed sampling or boost the minimum samples (and that is out of the question in 90% of the scenes :P ) ty for the help btw

[joconnell]

Yep you're totally right on the contrast issue. In that regard the adaptive subdivision sampler was great as rather than just using the pixel colours to decided if you got more aa or not, it'd also use object edges and object normals as extra input to decide on whether sampling was increased or not.

One sneaky trick that might work for you though - the extratex map has a feature called "consider for antialiasing". This used to cause a lot of slow down issues for me as I'd use it for the vray dirt map to render an AO passes for the compositors and what was happening was the edge contrast of the dirt map was also being used to drive the anti aliasing levels of the beauty (hence "consider for anti aliasing") and making my beauty renders slower. You might be able to use this to your advantage though. If you out in an extratex map, put a vray mutlisubtex map into the map slot and make sure the map is set to "random by node". If you leave the "consider for anti aliasing" turned on in the extratex map, it'll assign a random grey value to each object in your scene and might give you more contrast along those meeting edges which will force the aa sampler to work harder.

[thanulee]

Thank John! i ll try that see what happens. I was wondering since u mentioned it, if its possible to override specific objects with vray light material in extra texture. That will guarantee me perfect selections.