Cleanup noise in Elements: Reflection and Shadow - Also LOOONG render times

[zeronove]

so then how do you suggest I fix it?

what's the value of the clr threshold? ...maybe it's too high now; I'll try to drop down its value

[Morne]

it's on 0.005 (also tried 0.003)
making that figure bigger introduces MORE noise
havent tried 0.001 yet, I seem to recall somewhere some1 mentioned that its pointless to go below 0.003

I'm reading that dmc tut link you posted, maybe can pickup something from there

[Morne]

this just doesnt make sense to me. In that tutorial it says:
"But, since the nature of DMC algorithm, by raising Max DMC setting really high, let's say 50, adaptiveness will not play any role after some point and you can just forget about it."
but then 3 sentances on it says the opposite:
"a lot of glossy reflections etc... Vray will try to cast minimum number of samples needed to get a clean result. In these kinds of situations it's best to keep DMC values really high"

so as far as I understand so far, your subdivs in material gets divided by max in dmc image sampler. So if I have 512 in my meterial and max of 512 in my dmc image sampler, it means it will only get 1 subdivs, but if I make max dmc only 2, then it will get 256 samples. So what do I want? 1 sample or 256? now I'm totally confused

...and reading some more now

[Morne]

ok done reading that, seems dmc image sampler is better of with something like 1,4 or maybe 2,6 depending. Then raise material subdivs untill clean.
good. Now where does BF subdivs fit into all that? Do I also want HIGH BF settings, or lower like with DMC? I'm going to attempt doing it backwards. Switch off all reflection, 1st get clean GI, then switch on reflection and try and get clean reflections

seems this thread is turning into me ranting on and "writing stuff that comes to mind into my diary", rather than a forum discussion
chaosgroup, where are you guys? slaving away getting vray ready for 2013 I geuss ?

[zeronove]

I agree that it's confusing; I'm fighting with this kind of adp. DMC issues starting from the first 2.0 release and I still don't have a safe path to follow similar to what I've with the 1.5.
Based on my test what is written there about mat subds gets divided by max in dmc image sampler seems to be verified

[Morbid Angel]

hey here is my result. Imo its pretty clean, with under 2 hours for a render using pure brute force/lc. As I said before, with brute force you will not get lightning fast render time. But to me this is pretty acceptable, apply those settings send to the farm and done.

Btw i had to modify your geometry, the tunnel walls needed to be shelled, otherwise lightcache was bleeding in from the top.

Also if render time is such an issue for you, I would bake the lighting of the main interior objects (walls and so on) to a texture and use that instead of computing true gi from everything. That will indeed be rendering really fast and look great. Though you cannot modify the lighting or position of objects that much after.

[super gnu]

that does look -extremely- clean.. id be interested to know how much switching to imap/lc could bring that time down whilst keeping those lovely glossies you have... currently juggling too many jobs to test now but if i get a minute.

[Morbid Angel]

great thing about this setup is, the noise threshold value is really sensitive. You measure the noise difference of nearest pixels, and lower the noise threshold until the difference is only few numbers apart, at that point you are guaranteed a clean image.

Of course using irmap will give you much faster render time, not as crisp though in terms of shadows (nothing new to irmap).

[zeronove]

anyone so kind to repost a 2011 file?

thanks

[dlparisi]

You measure the noise difference of nearest pixels.

Would you mind elaborating a little more on this? How do you measure the noise difference?

[Morbid Angel]

using vray frame buffer just rightclick/hold over the flat area where you see noise in your image and while holding your rmb slide the mouse side to side (or whatever dir you want) and see the float values change. The higher the float difference between each pixel, numerically shows the noise intensity i.e. strong noise, strong value difference, no noise, the difference in pixels should be almost the same.
Based on this observation you can then determine weather or not to lower the noise threshold, I typically render with .85/0.01, as I think those default values are really great and not to be reckoned with. After giving dmc sample enough sample to work with, you can the lower both adaptive amount and adaptive threshold, in this particular render its .7/0.007

[super gnu]

why do you lower the adaptive amount? doesnt that reduce vray's rendering efficiency?

[dlparisi]

Thanks Dmitry. I thought you were using something to directly measure the noise value against the noise threshold. Still useful though - never thought to actually sample the pixels to determine the noise value rather than just subjectively looking at it.

[Morbid Angel]

why do you lower the adaptive amount? doesnt that reduce vray's rendering efficiency?

quite the opposite. I find that it takes longer for dmc sampler to adaptively figure out how many rays/sampler particular object needs to trace during render time, as oppose to giving it more samples upfront, and reducing adaptivity.

[Morbid Angel]

Here is another example. This one is at full hd 1920. I also lowered the noise threshold even more and got an even cleaner image. I think if you have a small farm such render with DR would be quite fast and even at 4k you can manage a decent render.