New adaptive method and optimisation prepass early results

[Neilg]

This looks interesting!

It is in the nightlies already, along with the new Hosek sky model.

Side note, but does this simulate dusk/accurately represent the sun below the horizon?

[^Lele^]

Side note, but does this simulate dusk/accurately represent the sun below the horizon?

Not below the horizon, as far as i remember from a very early implementation.
But don't quote me on it!

[jacksc02]

ooooooo Lele you tease!
Looking forward to trying this out.
The Hosek sky model looks really interesting with my early tests.........

[Dave_Wortley]

Give me sky models with clouds please.

[Morne]

You see now, you do other things for a week and come back to the forum and all this good news is floating around. I'm happy to sign up for testing!

[SR_JK_2010]

This sounds a fantastic way forward, cant wait!

[3LP]

Oh yeah....

Stan

[WilleViz]

Wow, this sounds really awesome!

Is this being optimized for mainly for Irradiance Map/Light Cache GI or Brute Force/Light Cache GI? Until now I usually goes with IM/LC to keep the noise down on Arch Viz interiors, but if this will help combat the Brute Force noise on big plain areas, like white walls, color me beyond impressed! Well, I'm impressed nonetheless already!

Oh, one small question: If the pre-pass takes some time to calculate, would it make sense to save that calculation to disc and reuse it, if you're doing a lot of test renders? It should be re-calculated for final render of cause, so best to keep that option under Expert. But I don't know whether that makes any sense.

[peterguthrie]

sounds amazing, happy to test it out!

[Moriah]

Guys this is sick! Can't wait to test it out!

[vlado]

What I want you to remember out of all this, is that the new adaptive method is much better at achieving consistent noise levels across the whole image and in different sampling situations, so there is a lot less need to tweak the MSR or any subdivs. In fact, the new build by default ignores any individual subdivs except for irradiance map and some special textures like VRayDirt. (This can be changed of course, if you want to fine-tune everything).

The MSR optimization itself may or may not be enabled by default, I haven't decided yet. It can shave off a bit of render time, but I think the approach needs more work.

Best regards,
Vlado

[Mokiki]

So by default material and light subdivs will be ignored and practially Vray will be in Auto-Pilot controlled by the new sampling/prepass method?

[^Lele^]

That is correct.
In that local subdivisions are ignored by default, and sampling is governed by MSR/AA.
This has nothing to do with the new sampler, though, or the prepass.

Just to be perfectly clear:
The new sampler is (much!) better at distributing where rays will be cast, between dark and bright areas, and across varying lighting situations.
The prepass is responsible for the balancing of AA rays versus the MSR ones.

What the prepass does can be already achieved by hand (say, you know fur will need aa rays, and you reduce its subdivs multiplier to 0 in the vray properties).
That's why it may not appear in the next SP, while Vlado works at making it more robust across a variety of situations (there is no magic, you see. :P).
The same(ish) result can already be obtained by a human brain with comparatively little effort.

[Badbullet]

I'd also like to know if the pre-passes could be saved. And would they be an image format of some sort, or a special format like light cache and irradiance currently are.

And if they are saved as images, would I be able to modify those frames? Say, multiply a depth map onto it in AE, and then use that as the pre-pass map? One of my requests was to be able to use a map to control sampling, using a depth map to give less samples to what is off in the distance as it would be blurred out in post using Frischluft's Lenscare.

http://forums.chaosgroup.com/showthr...ld-map-control

[vlado]

And if they are saved as images, would I be able to modify those frames? Say, multiply a depth map onto it in AE, and then use that as the pre-pass map? One of my requests was to be able to use a map to control sampling, using a depth map to give less samples to what is off in the distance as it would be blurred out in post using Frischluft's Lenscare.

This particular prepass will not work for this purpose; the prepass controls the rays for shading effects, however without modifying the noise threshold of the image sampler, you will still get the same result - the image sampler will just put more camera rays to compensate for the lower shading rate. So giving you access to that particular prepass is somewhat pointless.

We could add a map for the noise threshold, which probably would be more useful for you.

Best regards,
Vlado