Sampling tutorial for V-Ray 3.20

[SR_JK_2010]

Maybe im missing something (sorry) but in this case (MSR takes the extra samples needed overriding the UI) then why is there an MSR UI input? Its not needed?

[super gnu]

For the next release of V-Ray I'm experimenting with a prepass to figure out a good MSR value per pixel, depending on the geometric detail in it. Hopefully it could remove the need to tinker with these setting at all.

Best regards,
Vlado

that sounds like a great idea. i wonder if such a prepass could be used to control other aspects of the render.. automatic exposure for example.

[vlado]

that sounds like a great idea. i wonder if such a prepass could be used to control other aspects of the render.. automatic exposure for example.

Probably, so long as it is still fast enough.

Best regards,
Vlado

[^Lele^]

I have absolutely NO idea if it still works, but you could try and hack-load my exposimeter script from the stone age.
The issue with auto-exposure is simple, and yet potentially deal-breaking: what exactly is the "correct" exposure?
Further: would you then auto-expose per frame (as the prepass works now), leading to incurable flickering (much like in stop-motion. unavoidable in the digital realm. post is needed to balance it.), or would you have a parameter to specify exposure should change every NTh frame?
Exposing every so many frames would cure the flickering, as exposure levels interpolate, but it would be wholly oblivious as to what happened in the shot.
f.e.: you walk into a building from the outside, and the 6th frame after entering gets re-exposed, not the first.

Further still: how should the exposimeter know that bright source dead in the center of your image is a light, and not your subject, or a light AND your subject, and choose to consider or reject it for the exposure calculations?

dSLR cameras have the benefit of CCD ranges being limited, we don't.

I think, personally, and after a few years both coding, and using, the exposimeter, and catering for users' needs with it, that it ought to be left out of the prepass (should that, ever, see the light of day.).

[joconnell]

Regardless you'd almost need to do a full render, get exposure values and then do something like a sample over the range to get an animated curve for your exposure - then how would you animate it? I'd say the most sensible is go with iso so the look of each frame won't change drastically - the machines aren't smart enough yet

[super gnu]

to be honest, im not sure id ever use auto-exposure, seems better to do in post. i was just trying to think of some other potential uses for a prepass like the one being discussed.

if you were to do it in-render, youd need to smooth out any changes ( interpolation over frames) with an adjustable offset, so when, for example, going from outside to inside, it starts to brighten -after- you get inside, and not before. definitely sounds like a post job.

[^Lele^]

indeed, the amount of data to gather across animations may be prohibitive, and there's the slight issue of having then the variable exposure baked in the images (great, if it's the right one, a PITA if it ain't, for whatever reason.).

[joconnell]

And even if you do it as a post thing, you're exposing up or down to levels that vray hasn't been told to sample for - Auto things are dangerous

[^Lele^]

Auto things are dangerous

I wholeheartedly agree.
Not to mention Truck-things, Plane-things, and ship-things.
Try get hit by one of those, for danger size.

[joconnell]

If I have to get hit by one of those things, hopefully it's a really big one that takes me out in one go as opposed to horrible injuries

(on that note, happy friday )

[SR_JK_2010]

simple scene render times MSR v AA

Hi Everyone - I built a very simple scene to test some settings out. It is a box with an opening in the roof (vray white material) and vray plane floor with checker pattern (to test AA, receive subdivs for this plane set to 0) and chrome spheres within to test reflections. It is lit by an exterior dome light so the interior is not well lit and i did some test renders to see the affect of MSR v AA in this scene. The default setting renders at 1m 54 sec and I used the sample rate pass to try and get a clean (low AA) result first upping MSR. MSR of 100 gave a good result at 3m 57s, i then tired MSR 30 with wall subdiv multiplier of 4 but that rendering was 60 seconds longer so looks like more MSR rays is faster than upping the multiplier. Next i went back to MSR 6 and upped AA to 115 to get a sample pass to match the image quality of the MSR version. The render time was almost exactly the same at 3m 52 sec. Im guessing this scene is almost 50 / 50 areas that need anti-aliasing (checker) v MSR but thought you'd be interested - i did this very quickly so i didnt get too much into noise balancing optimizing, was just looking at MSR v AA speed, needs more testing but thought you would be interested -

[vlado]

The next service pack will feature an improved adaptive image sampling method that makes fiddling with subdivs and MSR even less necessary.

Best regards,
Vlado

[^Lele^]

Fair Warning: this is written in Vlado-speak.
Expect surprises... ^^

[WilleViz]

The next service pack will feature an improved adaptive image sampling method that makes fiddling with subdivs and MSR even less necessary.

Best regards,
Vlado

Sweet music to my ears as a new Vray customer!

[SR_JK_2010]

Great news Vlado - the system is already much simpler than previously ...