Light Cache in interior animations

[edgar_cg]

Re: Light Cache in interior animations

Thanks, but for some reason, it does not recognize the file:

"Unrecognized irradiance map format"

Any idea?

Thanks again

[dalomar_cg]

Re: Light Cache in interior animations

Yea, I think I ran into this a little while ago...I'll ask Joe if we can get an updated version of the IR viewer for download as I'm sure its just an issue with the version we have for download not supporting the newer maps.

[Micha_cg]

Re: Light Cache in interior animations

Experiment: caravan interior 360° 1000x500 LC sample size 0.02

Tested different LC subdivs and render time:

LC 1000 - 2:38 (my standard value)
LC 200 - 1:56 (light loss of shadow details)
LC 50 - 1:57 (many artefacts at details and reflections - LC pass need ~1s only)
LC 2000 - 4:27 (a very little bit better details than LC 1000)

The subdiv count dosn't show so much effect at the render quality. The LC quality could be important for the quality of the reflections, if "use for glossy rays" is enabled. The test scene based on many blury reflection, but no clear reflection.

LC 1000 + size 0.005 2:47 -> approx. the same quality like LC 2000

I forget to enable 4 thread for the LC pass. If I would use all cores of the quadcore, than the LC pass would be faster. The biggest part of time is needed for IM and final pass.

I think, LC+IM is the better combination than QMC+IM. For an animation I would test the main settings to find the lowest speed for good quality.
I try to get the lowest rendertime for my testscene with good quality:
LC - 10s
IM - 55s
final - 55s

We see, IM and final pass are the critical parts and LC at medium settings dosn't matter. For animations, the IM can be cached and is no render time problem anymore. And for the final pass, the LC "use for glossy rays" is a help too.

It's interesting, could it be, that we could stick at LC for secondary engine allways? Damien, any tests from your side? I'm curious to know, for what sec engine QMC could be used. I thought too, that QMC could help to bring down the render times, but my tests show me that the LC is quite good. OK, I use one scene only, but's a scene without any direct light source, only a GI environment (a sky color gradient image without sun). Could be nice, if we could say: use LC allways, it's the best.
So, we need to choose the primary engine only and that is easy to decide - IM for speed and low noise, QMC for high details and light noise.

Edgar, let us know, what you get with your scene.

PS: without reflections my scene need 1:19 only, maybe you can render your scene with diffuse materials only.

[dalomar_cg]

Re: Light Cache in interior animations

The subdiv count dosn't show so much effect at the render quality. The LC quality could be important for the quality of the reflections, if "use for glossy rays" is enabled. The test scene based on many blury reflection, but no clear reflection.

I've found about the same thing, that the subdivisions value doesn't effect a whole lot once you get to a certain quality level. Personally I've found that I can get good base results at about 4-500 and if need be for finals I'll go up to 1000 or 1500 (if I've got allot of glossies)

I forget to enable 4 thread for the LC pass. If I would use all cores of the quadcore, than the LC pass would be faster. The biggest part of time is needed for IM and final pass.

Well depending on the amount of time it actually takes to calculate the samples, setting the phases to 4 may not speed things up. This is because merging each one of those passes is a single threaded operation, thus leaving the other cores idle. The time needed to merge those phases is almost always constant (from my tests at least), so the quicker LC can actually be calculated, the less of a need for another phase. For interiors and lots of glossy/bounced light this isn't the case, so match the phases to the number of cores. Also more phase will introduce slightly more noise.

It's interesting, could it be, that we could stick at LC for secondary engine allways? Damien, any tests from your side? I'm curious to know, for what sec engine QMC could be used.

The last time I made the conscious decision to use QMC over LC was doing some animations last december...and the only reason for that was because I was pressed for time and didn't want to worry about putting together the precalcs. I really haven't found any reason not to use LC for secondaries, so although I won't advocate making it a fixed thing, it doesn't change to much in my setup. As I've said before, sample size is the only thing that I've really tweeked in my LC setup, and I keep that at either 1% or .5% (scale always as screen).

On another note, since the sr I've made allot of use of Detail Enhancement. So very rarely do a make the complete switch to QMC. Thats really my best/fastest setup IM(typically at lower settings, but with)+DE and LC for secondary bounces. Also I will typically stick with Adaptive Subdivision over AQMC...In my mind it take a bit to many samples for aQMC to smooth out some of its noise.

[Micha_cg]

Re: Light Cache in interior animations

AQMC vs. ASubdiv:
In the past my tests with ASubdivs makes me not happy, textures lost details and are looking blury. I tested it at my current caravan test scene with cloth and wood textures.

final pass only
AQMC 54s
ASubdivs 57s (textures are looking bad)

I used the default Asubdiv setting. Do I something wrong, is it possible to get more quality and less render time with Asubdiv? Or is it good for scenes without textures only?

An DE problem was, that if my scene used blury reflections, than at places there QMC is used, the reflections are looking noisy sometimes. Good, I could use more DE subdivs, but this increase the render time. I ask me, could it be possible, that DE is used for lighting only and not for reflections? Make it sense? I like the detailed shadows of DE, but not that it touch reflection sampling. Attached a test from an old project that I found on my harddisc. But ...

... I tested DE at my current scene and the final pass jump from 54s to 91s, this isn't so much, because the IM pass could be speed up by lower settings. The images looks good, no noise problem. Could it be, that something was wrong with my lavatory scene? Hmm, I tested the lavatory scene today again - and it works perfect, no additional noise at DE QMC areas.

The DE QMC needs very low subdivs - IM subdivs 80 and DE subdivs multiplier 0.05 (caravan scene) - the image looks very good. Whats your experience, how much subdivs need the DE QMC? (And why is it dependent from the IM?)

[Micha_cg]

Re: Light Cache in interior animations

... I found the reason for that the scene show no noise today - the DE radius is smaller now and DE isn't active at the critical part. A bigger radius and the noise is back. So, the problem stay: DE influence reflection sampling. Damien, do you think, DE could be used for GI lighting calculation only and not for reflection sampling? I would post a wish for Vlado, if it makes sense.

[dalomar_cg]

Re: Light Cache in interior animations

The default settings for ASubD are not really the best. Anytime you have a negative min rate, your going to be undersampling things, so thats where the issue comes in. Although I leave it at -1 for tests, my finals always have a min rate thats either 0 or 1 (unless its at a high resolution <2000)

When you set up DE you've got to realize that it changes how the IM gets calculated as well, so you actually don't need as high IM settings. This goes for both min/max rate and the number of h.subdiv. The IM calc is only being used for general overall lighting information, so that can be done relatively quickly. Higher max rates aren't need because the DE (qmc) is used for the details, so a more refined IM isn't needed.

That being said, the settings for DE are going to be based off of your IM settings (as you probably already know). I try to set the multiplier to something that is going to end up having the DE subdivisions right around 8. Since I typically use 40 for h.SubD my DE multiplier is about .2 . After that you can just use the radius value to control how much QMC is used. The larger the radius the more of the QMC calculations. In my opinion 60 is quite big, so I'll lower that to either 30 or 40.

So here's how my IM+DE settings typically look
Min/Max: -5 or -4/-2
H.SubD: 40
DE: Enabled Scale: Screen
Radius: 40
Multiplier: .2

do you think, DE could be used for GI lighting calculation only and not for reflection sampling?

I highly doubt it. Once the lighting calculation is handed off to QMC, it will need to sample almost everything to get the right result. So asking to exclude it from sampling reflections might actually lead to incorrect lighting results.

[Micha_cg]

Re: Light Cache in interior animations

Thanks Damien. I rendered my scene with ASubDiv 0 too, but it's slower than AQMC. Good to know, that the DE multiplier can be so low.

I found an interesting effect: it's not the reflection sampling, that looks noisy, it's the sampling of the emitter material under the lavatory. For a test I global disabled reflections and render DE on/off - the noise is visible at the DE rendering.
Other settings, a bug or is some Vray code refinement needed?

[dalomar_cg]

Re: Light Cache in interior animations

I found an interesting effect: it's not the reflection sampling, that looks noisy, it's the sampling of the emitter material under the lavatory.

That wouldn't surprise me...its tricky to get the right sampling for Emitters sometimes. The sampling for the emitters themselves are actually based on the mesh, therefore objects with finer meshes will have better sampling then course meshes. Below is a quick example. The emitter on the left is one face and the one on the right is 20x20 (so 400 I guess) faces. Both have the same material and there is no other light in the scene. The differences are subtle, but they are there. This is also a fairly straight forward example as well.

[Micha_cg]

Re: Light Cache in interior animations

Very subtle - I filter the image with the highpass photoshop filter and correct the levels, but I can't see the difference. Maybe it's better visible in other scenes. I will keep it in my mind.

I create a simple example scene for the emitter DE noise problem. Without DE the scene need 10s, with noisy DE 16s and with fine DE 22s. It's a pity, that the emitter need a high DE subdivs multiplier for emitters only. Maybe Vlado could help and improve the Vray code, so we get low noise and high speed.

If I replace the emitter with a rect light, than the DE renderings is noisefree at low settings and ready after 13s. Perfect, but often, if I set emitters, than I can not replace the emitters.

www.simulacrum.de/download/DE_noise.rar

[dalomar_cg]

Re: Light Cache in interior animations

Comparing DE with regular IM numbers is like comparing apples to oranges. In other words, IM is always going to be smoother and faster because thats just how IM works. Comparing DE to QMC is more of a fair comparison because thats really what your trying to get around, complete use of the qmc algorithm. So as with QMC it can be pretty noisy in certain situations. Best smooth it out as you would QMC with either more samples (higher subD multiplier) or better AA settings.

Also, Micha, its not as simple as asking Vlado to just optimize the code, its much more complex then that. And there aren't too many places where he can just change it an make it better (its already so damn good)...sorry, it just doesn't work like that.

[Micha_cg]

Re: Light Cache in interior animations

OK, thanks for your effort, Damien.

[dalomar_cg]

Re: Light Cache in interior animations

NP...its always good to bring more info on vray to the masses...its quite dense, and I've been fortunate enough to finally begin understanding how the whole thing really works...now all I need to do is learn C++ to help Joe out (but that isn't happening for another 8-10 months) ;D

[Micha_cg]

Re: Light Cache in interior animations

Sounds great, good luck.

[Micha_cg]

Re: Light Cache in interior animations

My last observation: HDRIs and Physical sky create DE noise too, if not high DE settings are used. So, I think, the general rule is: non raytraced lightsources create more DE noise or need stronger DE settings.
So, some times it's better to use high quality IM settings and not DE.