V-Ray Scene Optimisation Script

[CCX]

Yes I second that.. Great Script indeed.
the whole thread has cleared a lot of confusion i had so far.

Thanks a ton Lele.

[joconnell]

My point wasn't really based on the assumption that the MSR divides or reduces sampling in lights and materials it was based on the assumption that a higher MSR results in lower AA subdivisions (in a situation where you have adaptive subdiv AA).
Therefore if you don't fix your AA (as Lele has suggested in the previous post) then higher MSR = less camera rays = potential to compromise geometric detail/outlines/DOF/textures (or anything that you need camera rays for)

Ah okay I get you now.

So with the adaptive DMC anti aliaser, the only thing that it can use to judge whether to go up to the next level of quality is the contrast between a pixel and the pixel beside it. If there's a big difference between these two pixels, you get more AA being used. In the case of geometry, the main way you'll need more AA is if one object is against a brighter or darker background. Say for example you have a black object against a white background, then you'll have a really high level of contrast and you'll get loads more AA until vray is satisfied that it's met the quality you've asked for. Same thing with noise / shading issues. You can see in Lele's much lower quality render tests (the little 100 x 100 pixels ones) that it almost looks like a frame of black and white noise. The adaptive dmc sampler looks at the contrast in these pixels the exact same way - it doesn't matter if it's seeing the difference between colour values in pixels that are on the flat faces of an object (what we'd consider "shading") or on the edges / silhouette of the object (what we'd consider "geometry"). All adaptive dmc sees is noisy pixels and goes to solve them.

Just because you're using really high shading rates and thus giving the dmc sampler really clean results to work with, they won't have any effect on the edge / silhouette pixels where your object is on top or behind another object. This is a bit of an all over the place explanation, I'll try and come back with something a bit better!

[^Lele^]

Yes I second that.. Great Script indeed.
the whole thread has cleared a lot of confusion i had so far.

Thanks a ton Lele.

Thank YOU.
I'm glad it helped understanding that labour of love that V-Ray is a bit better.
For me, trying to get into the inner workings of the engine is one of the main reasons to work in the industry i do, and it's never ending a job.
Particularly as often enough Vlado and co. come up with some novel approach, and turn what iI held for granted head over heels.
Wich only add to the delight, eheh ^^

[^Lele^]

Just because you're using really high shading rates and thus giving the dmc sampler really clean results to work with, they won't have any effect on the edge / silhouette pixels where your object is on top or behind another object.

Actually, this is EXACTLY what is going to happen using MSR and a low min/high max rate.
MSR WILL shift the sampling towards less camera rays, and so, with high enough MSR, you WILL get worse edges (unless, of course, one sets very low thresholds globally).
Here's two pictures of an alpha edge, see for yourself.
All to default, no image filtering.
AA 1-100, MSR of 1


Same, but MSR of 1024

You can tell there are indeed differences, as less camera rays are cast there.
Which is another reason for having a fixed AA method with a min rate which isn't of 1, regardless of the scene contents, and high enough to have the alpha edges clean.
The sample rate element for that image also shows the same behavior: higher MSR, lower camera rays (and ofc quicker renders.).

[Paul Oblomov]

Meanwhile, I think a found a bug. Single threaded progressive render... Region reder. Fixed with lowering ray bundle size to 32. Also - when fast scrolling between render elements looks like their menus slowly falloff

[^Lele^]

"do you have a noise threshold for it, other than 0.0?" (Cit.)

[Paul Oblomov]

Lowered bundle size and it all works nicely It was region render...

[CCX]

One more thing Lele,
I use vray in maya
I downloaded your max script and run it 3dsmax to
see how the settings are related based on the presets.
I was just curious how the "resulting Noise threshold" is calculated ?
which is really fine considering the clr threshold being constant at 0.01
or is it that i might have missed one of your posts in the thread.

[whtnoise]

Thanks for this Lele!

- Jesse

[joconnell]

Actually, this is EXACTLY what is going to happen using MSR and a low min/high max rate.
MSR WILL shift the sampling towards less camera rays, and so, with high enough MSR, you WILL get worse edges (unless, of course, one sets very low thresholds globally).
Here's two pictures of an alpha edge, see for yourself.
All to default, no image filtering.
AA 1-100, MSR of 1

You can tell there are indeed differences, as less camera rays are cast there.
Which is another reason for having a fixed AA method with a min rate which isn't of 1, regardless of the scene contents, and high enough to have the alpha edges clean.
The sample rate element for that image also shows the same behavior: higher MSR, lower camera rays (and ofc quicker renders.).

Eeeeenteresting - must have a bit of a fiddle with it. Got a few bits with fine detail which should provide a test.

[^Lele^]

One more thing Lele,
I use vray in maya
I downloaded your max script and run it 3dsmax to
see how the settings are related based on the presets.
I was just curious how the "resulting Noise threshold" is calculated ?
which is really fine considering the clr threshold being constant at 0.01
or is it that i might have missed one of your posts in the thread.

Eheh, it's in the tooltip, and in the thread, but man the thread has turned into a mess for my own faults (and i wrote the formula WRONG, but just in the tooltip AND the info panel. sorry, corrected it and re-uploading the script, which you'll find in the first post of the thread.), so i agree it may need further explanation.

Basically, 1 bit can express one more stop of brightness.
For example, from 8 to 9 bits, you can express from 256 to 512 levels of brightness (just as in DSLR raw formats, right? a 12bit raw goes to 4096, as for the three quarter CCD cameras. the full frame ones allow RAW files at 14 bits, which is 16384 levels of brightness).
Given that oftentimes my post production friends like to fiddle with renders as they see fit (read: "this is too dark. let me brighten it a bit. Ah! look at this render, it's SO noisy!"), i often have to cater for that type of accident before it happens, as it's quite impractical, not to mention expensive, re-rendering a 500 frames sequence.
This is why i express the "target" noise level as bits.
The maths to turn that into a float value for the (DMC ONLY!) noise threshold is simple enough: 1.0/(2^noiseBits)
Which, to all intents and purposes, and with enough sampling allowed, WILL make your renders largely noise-proof for the targetet stop level.
As you can see from the 4 times brighter render i posted above, which rendered with a 10 bit noise threshold, the concept is holding quite well: even two stops up, there is no visible noise.

Given that the vast part of the speed gains is achieved by NOT triggering camera rays, that is the reason why i do not use the same threshold for the AA sampler.
The actual 0.01 value there is the default, but i should be changing that (although it is irrelevant when min/max are identical, of course) to 1/256 (or 0.00390625), as most of the times the alpha channel doesn't need anything more than 8bit of precision (although it gets promoted to 16 bit when exporting EXRs).

Version 1.48 of the script will do both things.
grab it in the first post of the thread.

And of course, let me know if i managed to confuse you more. ^^

[kosso_olli]

To be honest, I have no idea what this is all about. None of my scenes are rendering any faster, if they render at all. Still having the problem of stuck buckets. Some of them will move eventually, and the result is tad bit cleaner I have to admit. I don't even need renders that clean, so I tried lowering the settings. But I can't get any faster speed out of this... Still, the only preset that is rendering at acceptable speed is the rough one, but it is way to noisy.

[joconnell]

It's always horses for courses though. Lele's standards are incredibly high on renders as his compositors may fuck around with the look a huge amount, so he wants to give them the latitude to do that and never have his render come back. You've got a pretty refined process for your work so you probably get your renders 90% correct from the off so they won't be colour corrected as much and thus won't need the level of refinement that something like Lele's VFX preset goes to. Likewise if we look at Grant Warwick's optimisation tutorial that's totally fine for his work which is single image print frames, but the level of noise and AA would cause issues for me as a vfx operator dealing with fizzing noise from frame to frame.

Your mileage may vary

[rajdarge]

@lele script still turns on "getsetsfromax" and sets vrayVFB to default res at 640x480, though the calculations of the lc seems to reflect the values prior to running the script.

[^Lele^]

Oh?
Then i just introduced a bug, eheh, will check the code later.
Sorry for the delay replying, i had a slight issue driving back home from a nearby town.
At around 130km/h the engine decided it didn't want one of the four pistons, so it dropped it from the bottom of the engine.
I thank the hundreds of hours poured into driving simulators and karts for real, as without them, i'd likely would never have been able to fix that silly bug.
Give me a night's rest, as i feel i sorely need it, to make sure i won't botch the script, rather than fixing it.