The best Anti-Aliasing for v-ray renderings

[Morbid Angel]

usually 1920x1080p or 2048x1556 2k. And some times 4k.

[R_Cyph]

usually 1920x1080p or 2048x1556 2k. And some times 4k.

So if you were working at a lower resolution, would you need higher settings?


anychance of you uploading those AA comparisons ?

[Morbid Angel]

well if you work with lower resolutions, it can be positive and negative at the same time.
Positive, because some of the finer details will be overlooked, and or smoothed out by the filtering.
Negative, because if you have some bright highlites on detailed surfaces it will need more samples to anti-aliase to produce steady frame-to-frame transitions, whereas on larger res frames such things are resolved better. It used to be in old mental ray, version 2, where people rendered x2 or x4 the original size, because it could not some times solve detail. Still cant actually in some cases. MR is limited to adaptive subdivision which often can suffer.

I will post the samples in a bit...

[Morbid Angel]

Ok, here are the tests.
Here is one of the worst possible situations for any render. A multi depth low frequency glossy reflection.
In this case, single area light, full reflective shader with glossy set to 0.5. A light is reflected in teapots, as well as the wall, and the teapots are reflected back into the wall and vise-versa. The frames are gamma corrected with 0.4545. I have rendered the scene in each case with the following settings:
DMC 1/4, 0.01, DMC 1/32 0.01, DMC 1/32 0.001, DMC 3/6 0.01, adaptive Subdiv 0/3, clr thresh 0.05. None of the settings have reached a flicker free solution. However the render times climb to 20-30 min per frame.
DMC at 1/32 0.001 has reached a better quality sampling at longer times then adaptive subdivision, however there was some precision loss, notice the black buckets where vray determined that there were no bright pixels there, even though the noise thresh was set to 0.001 (the lowest possible).



DMC 1/4
DMC 1/32
DMC 1/32 0.001
DMC 3/6
Adaptive Subdivision 0/3 0.05

[insleight]

On my screen, I notice a slight difference between 1/32 .001 and 0/3 .05 in terms of image quality. Though, that might be the compression. The biggest difference however, I see, is in render time. A nearly 50% increase in render time going from adaptive sub to dmc. Can the case be made for high end animations including moblur/dof for using adaptive sub? Particularly if you increase your adaptive sub values to higher than tested. I imagine at least a 20% increase in time going from adaptive sub to dmc to acheive the same image quality that you are looking for?

[insleight]

Or because of HD resolutions, being smaller than most stills, still need that subpixel calculation to get accurate/smooth results? Morbid, you mentioned the difference between high/hd resolutions in terms of quality with your extensive testing. Do the test renders you provided change any of that?

[Morbid Angel]

Im not sure I understand the question?
My test is simply showing a complex condition, and one of the worst possible anti-aliasing situations. And that even the highest aa settings will not work.

[insleight]

Ok, heh. Ill try to be clear and to the point. For average conditions, not the most complex: For a render with high adaptive dmc to achieve a good result OR adaptive subdivision to reach an equally good result, which one will render faster if the quality of the render is 1:1 between the methods?

[R_Cyph]

Ok, here are the tests.
Here is one of the worst possible situations for any render. A multi depth low frequency glossy reflection.
In this case, single area light, full reflective shader with glossy set to 0.5. A light is reflected in teapots, as well as the wall, and the teapots are reflected back into the wall and vise-versa. The frames are gamma corrected with 0.4545. I have rendered the scene in each case with the following settings:
DMC 1/4, 0.01, DMC 1/32 0.01, DMC 1/32 0.001, DMC 3/6 0.01, adaptive Subdiv 0/3, clr thresh 0.05. None of the settings have reached a flicker free solution. However the render times climb to 20-30 min per frame.
DMC at 1/32 0.001 has reached a better quality sampling at longer times then adaptive subdivision, however there was some precision loss, notice the black buckets where vray determined that there were no bright pixels there, even though the noise thresh was set to 0.001 (the lowest possible).



DMC 1/4
DMC 1/32
DMC 1/32 0.001
DMC 3/6
Adaptive Subdivision 0/3 0.05

Thanks for the post, very enlightening.

I have a few questions.

1) If this was rendered at HD or 4K res would 1/32 @ .001 be enough for a flicker free solution?

2) would raising your global subdivision multiplier or raising the mat subdiv fix the flickering even at 1/32?

3) when you run into situations like this. Do you just use increase ur AA to 1/64 or something? or just deal with the not so perfect results?

[Morbid Angel]

Ok, heh. Ill try to be clear and to the point. For average conditions, not the most complex: For a render with high adaptive dmc to achieve a good result OR adaptive subdivision to reach an equally good result, which one will render faster if the quality of the render is 1:1 between the methods?

Vlado may be able to answer this one more correctly. However I think adaptive DMC will be better, because it can deal with certain aspects faster/better then adaptive subdivision.
Plus adaptive subdivision is limited to a number of samples, 0/2 (1/16), 1/3 (4/64).

Thanks for the post, very enlightening.

I have a few questions.

1) If this was rendered at HD or 4K res would 1/32 @ .001 be enough for a flicker free solution?

2) would raising your global subdivision multiplier or raising the mat subdiv fix the flickering even at 1/32?

3) when you run into situations like this. Do you just use increase ur AA to 1/64 or something? or just deal with the not so perfect results?

1 - I need to test this...

2 - possibly, however the rendertime will skyrocket.

3 - It depends. In this case, the best thing to do is to nutralize the multibounce reflection. That is, terminate the glossy at a first or second ray depth. With much less rays to trace between each other, you will get a much cleaner image with similar look, not the same, but similar.

[Morbid Angel]

Thanks for the post, very enlightening.

I have a few questions.

1) If this was rendered at HD or 4K res would 1/32 @ .001 be enough for a flicker free solution?

So I ran a quick test and indeed it seems to produce a waaay better quality image at 4k with 1/32 at @0.001
However the render time was projected to be 32 hours per frame.

[victor.nsy]

Ok, here are the tests.
Here is one of the worst possible situations for any render. A multi depth low frequency glossy reflection.
In this case, single area light, full reflective shader with glossy set to 0.5. A light is reflected in teapots, as well as the wall, and the teapots are reflected back into the wall and vise-versa. The frames are gamma corrected with 0.4545. I have rendered the scene in each case with the following settings:
DMC 1/4, 0.01, DMC 1/32 0.01, DMC 1/32 0.001, DMC 3/6 0.01, adaptive Subdiv 0/3, clr thresh 0.05. None of the settings have reached a flicker free solution. However the render times climb to 20-30 min per frame.
DMC at 1/32 0.001 has reached a better quality sampling at longer times then adaptive subdivision, however there was some precision loss, notice the black buckets where vray determined that there were no bright pixels there, even though the noise thresh was set to 0.001 (the lowest possible).

Yo Mobid,

Could you post the scene which you tested? Max8.0? I couldn't reproduce the
noise... Thanks in advance!

[Morbid Angel]

I only have it in max 2009

http://www.godray.org/newdirectory/aa_tests.rar

[R_Cyph]

well thanks for the schooling

btw just looked at your demo reel. Nice stuff there.

if you don't mind me asking where do you goto school? or how did u train to be such a good technical director

[Morbid Angel]

I actually went to VFS (Vancouver Film School) but I dont want to credit them for anything. Mostly just years of experience and desire to learn.