Topic: Anti aliasing types: when to use what type.
posted by Vlado
archived on 22.6.2002
VRay already has built-in logic that controls simultaneous sampling of things like AA, DOF, GI, glossy reflections etc. If an effect (glossy reflection or area shadow, for example) is supersampled multiple times (because of DOF, for example), it will automatically reduce the quality of individual samples. This is a concept adopted everywhere in VRay for all fuzzy effects, not just DOF.
An important note to make - one that I have tried to mention whenever I can - is that if you have a lot of fuzzy effects (dof, glossy reflections/refractions, area shadows, direct GI, motion blur) do not use adaptive AA - use two-level or fixed rate aa. You can get substantial speed-ups from this simple rule.
Q: How could I have missed that???... I assumed the the opposite was true. I basically used Adaptive all the time. Can you explain a little more WHY this is true and HOW two level works differently? This may explain why one of my renders is taking 1 hour a frame. I have GI, area shadows blurry reflections.
A: Well, with fixed and two-level AA, VRay knows in advance how many image samples it is going to make when computing a pixel value. Fuzzy effects take this in account and reduce their sampling rate accordingly.
With Adaptive AA, it is not clear beforehand exactly how many samples will be taken within a pixel. In addition, Adaptive AA uses the already computed samples within a pixel to determine if more samples are needed, which means that the sample "quality" must be maintained - that is, the sub-pixel samples can't be of worse quality, because they will mislead the adaptive image sampler. To make the matters worse, fuzzy effects are usually noisy, which makes the adaptive sampler to take even more samples, thus increasing the rendering time further (although this can be avoided to some extent by increasing the aa threshold).
The two-level sampler on the other hand, uses the color of pixels as a whole to decide which pixels need to be supersampled. Therefore it doesn't matter if the subpixel samples are of worse quality - only the final average result will be taken into account.
This means that the adaptive sampler will, on the whole, produce smoother results; however the rendering times may get excessively high. When there are a lot of fuzzy effects, the fixed and adaptive samplers will perform much better, while maintaining a relatively constant image quality.
There are two images below with direct GI and motion blur (same settings for both images). One is rendered with fixed rate AA at 3 subdivs, and the other one - with adaptive AA at rate -1/2. Rendering times speak for themselves.
Q: Thank you Vlado... very helpful info. I gather from what you are saying, is that the real advantage to thr adaptive AA is when doing raytracing I notice that it si a lot faster than two level, especially with refraction.
A: Yes, adaptive AA is very good for smooth images where you don't have noise - if you use raytracing only or GI with irradiance map.
> print this entry