is Supersampling Phyical Correct ??

[Sebastian]

I read these Articels here and ask myself if this is a Physical correct Method or if it's some kind of biased Methode, does anyone know ???

German Article

English Articel

[Kram1032]

It simply makes Antialiasing sharper but less artefacty.

It's as if you'd render on higher res and make it smaller, afterwards. - No bias there

The german Wiki also says, that it's a Monte Carlo thingy, and Monte Carlo == unbiased

[Knaxknarke]

It's always better to take more samples, so you get less variance. One problem could be if you have some "smart" strategy about the sample points. I'm not sure, but smart adaptive sampling and interpolating could introduce bias.

Any experts reading? Is QMC "biased" ? There are no real random number series with a deterministic computer, so what is the exact definition of biased rendering?

[Knaxknarke]

BTW: supersampling the eye ray to do antialiasing should be no problem, but selection sampling directions for further bounces along the light path could be a problem. Having some rule when to "cheat" is like using a loaded dice. It changes your pdf, so the experiment is distorted.

[Anthony]

I thought Indigo gets naturally anti-aliased anyway because of how its made because its all sampley and stuff.

[Kram1032]

the only biased thing in indigo is, as far as I know, the num consec rejections - the lower they are, the more biased, indigo gets

If you can get your hands on a laser, a polariser, a PBS (Polarizing Beam Splitter) and two light detectors, you could make an absolutely random process

even with different randomness for 0 and 1, depending on how you set the polariser^^ - if you set it to 45°, you get 50:50 0 and 1

- actually, a thing, that you could build yourself with some materials... there was something on wiki, I think, to make your own laser (not sure, if there also are pule lasers, as needed, for this, though) the light detectors would be "simple" solar cells... the polariser shouldn't be too hard to get. No idea about the PBS, though... something like this:


then, you only need to find a way, to use that information for indigo^^

[oodmb]

QMC may not actualy be completely random, but for statistical purposes its random (which is why scenes eventualy converge). I use the jitter algorithm.

[OnoSendai]

the only biased thing in indigo is, as far as I know, the num consec rejections - the lower they are, the more biased, indigo gets

If you can get your hands on a laser, a polariser, a PBS (Polarizing Beam Splitter) and two light detectors, you could make an absolutely random process

even with different randomness for 0 and 1, depending on how you set the polariser^^ - if you set it to 45°, you get 50:50 0 and 1

- actually, a thing, that you could build yourself with some materials... there was something on wiki, I think, to make your own laser (not sure, if there also are pule lasers, as needed, for this, though) the light detectors would be "simple" solar cells... the polariser shouldn't be too hard to get. No idea about the PBS, though... something like this:


then, you only need to find a way, to use that information for indigo^^

RQMC is actually better than samples picked uniformly, over, say, the unit interval, because the actualy distribution of drawn samples will be flatter.
So real random numbers = lose.