sub pixel mapping

[super gnu]

so.. why does it make such a difference to images and rendertime? i understand the concept of applying colour mapping to individual samples then averaging them to get final pixel colour, rather than, with it off, averaging the samples -then- colour mapping... but..but..

i cant see how this should make any difference to the final colour of the pixel, or the rendertime....?

[joconnell]

Okay say you've got black being 0 and white being 1 in terms of colour values. Anti aliasing works by smoothing along the edges where colours of different values meet so with a simple anti aliaser if a black edge meets a white edge you'll probably get pixels with a value of 0.5 to make a soft transition between the two. With floating point colour, hdri maps lighting the scene and vray area lights using multipliers much higher than 1.0, you're inevitably going to get some pixels that have a really high brightness value, often way higher than our white value of 1.0. The problem in terms of render time is that vray sees an edge where some black pixels with a value of 0 meet some really bright pixels with a value of 10, lets say either a light or a reflection of a bright light source. The anti aliaser starts working and picks a colour value in the middle of these two and it ends up with 5.0 as the middle value to smooth things out. This value then goes back to the vray sampler, vray checks it against it's colour or dmc sampler threshold and sees if its below the threshold you've set for noise. It's not so vray goes to the next level of sampling, does a comparison between 0 and our new 5.0 pixel and again comes back with 2.5. The same threshold process occurs again, 2.5 isn't under our noise threshold so vray keeps sampling until it either gets the value below the noise or you run out of samples and it doesn't look good enough.

If you turn on sub pixel mapping, what vray will do is tone map the colour samples first so that our pixels with a value of 10 might get brought down way closer to 1.0 early on, and it means that it takes far less steps for vray to get below it's noise threshold. You'll end up with less harsh highlights in your scene so it's far easier to sample and it'll be quicker, but it's losing some of the range of brightness you might need for your final image to be realistic.

[super gnu]

nicely explained

[joconnell]

Hate when I do that