About noise in general

[joconnell]

You're quite right in what you're thinking but perhaps by setting the subdivs so high, will it mean that your min aa pass end up using too many rays when you're rendering at aa 1 and taking max material samples? That's one of the reasons that I'd like to know how vray loops between the dmc sampler and the aa sampler up through the levels of quality!

[cecofuli]

Jérôme,
it's possible divide the subdivs by Refraction/reflection channel?
Now we change all Subdivs
Thanks!

[DanielBrew]

I'm having a request for guys wich are at ease with maths... :

I'm trying to simplify this tool, allowing the user to enter only 2 values : subdivs for gloss 0.0 and subdivs for gloss 1.0.
All the other values will be computed by the script, in different manners.

for example, the user entered 8 subdivs for gloss=1.0 and 64 subdivs for gloss = 0.0
In linear method, that's quite easy : ((max - min)+1) * glossyvalue or something like that.
in this case, glossy with 0.5 value will have ((64 - 8 ) + 1) * 0.5 = 28 subdivs.

but it would be nice to have also other methods, like log, expo, power, etc.. to compute the intermediate values...
Someone have an interesting formula here ?

What about using a curveControl to interpolate between values? It might not be quite as accurate but visually it's easier to understand.



Here's the code for anyone who wants to try it:

try(destroydialog vrSubs) catch()

rollout vrSubs "Set VRay Material Subdivs"
(
	on vrSubs open do
	(
		vrSubs.theCurve = vrSubs.msa.curves[1]
		vrSubs.theCurve.color = red
		vrSubs.subdMax.value = 64
		vrSubs.subdMin.value = 8
		vrSubs.setCCVal 1 vrSubs.subdMax.value
		vrSubs.setCCVal 2 vrSubs.subdMin.value
	)

	local theCurve
	
	fn getCcPoint curve ptIndex = curve.points[ptIndex] --because getPoint is broken in Max2012
	fn setCcPoint curve ptIndex newPoint = (curve.points[ptIndex] = newPoint) --because setPoint is broken in Max2012
	
	fn setCCVal i yVal =
	(
		local pt1 = vrSubs.getCcPoint vrSubs.theCurve i
		pt1.value = [0,yVal]
		vrSubs.setCcPoint vrSubs.theCurve i pt1
		zoom vrSubs.msa #all
	)
	
	fn setSpins ccIndex ptIndex =
	(
		local selCurve = vrSubs.msa.curves[ccIndex]
		local newPt = vrSubs.getCcPoint selCurve ptIndex
		
		if ptIndex == 1 do vrSubs.subdMax.value = newPt.value.y
		if ptIndex == selCurve.numPoints do vrSubs.subdMin.value = newPt.value.y
	)
	
	fn setVrSubs =
	(
		for vrm in getClassInstances vrayMtl do vrm.reflection_subdivs = getValue vrSubs.theCurve 0 vrm.reflection_glossiness lookup:true
		for vrm in getClassInstances vrayCarpaintMtl do vrm.subdivs = getValue vrSubs.theCurve 0 vrm.base_glossiness lookup:true
		for vrm in getClassInstances vrayFlakesMtl do vrm.subdivs = getValue vrSubs.theCurve 0 vrm.flake_glossiness lookup:true
		for vrm in getClassInstances vrayFastSSS2 do vrm.specular_subdivs = getValue vrSubs.theCurve 0 vrm.specular_glossiness lookup:true
	)
	
	spinner subdMax "Subdivs @ Glossiness = 0.0:" pos:[10,410] range:[0,128,64] type:#integer fieldwidth:30 --controller:(Bezier_Float())
	spinner subdMin "Subdivs @ Glossiness = 1.0:" pos:[210,410] range:[0,128,8] type:#integer fieldwidth:30 --controller:(Bezier_Float())
		
	CurveControl msa pos:[5,5] width:390 height:390 x_range:[0,1] y_range:[0,128] numCurves:1
	uiFlags:#(#drawBG, #drawgrid, #ruler, #upperToolbar, #showReset, #constrainY, #xvalue) --#lowerToolbar,
	commandMode:#move_xy
	
	button setVrSubsBut "Set VRayMlt Reflection Subdivs" pos:[5,435] width:390
	
	on msa ptChanged ccI ptI do vrSubs.setSpins ccI ptI
	on subdMax entered do vrSubs.setCCVal 1 vrSubs.subdMax.value
	on subdMin entered do vrSubs.setCCVal vrSubs.theCurve.numPoints vrSubs.subdMin.value
	on setVrSubsBut pressed do setVrSubs()
)
CreateDialog vrSubs width: 400 style:#(#style_titlebar, #style_minimizebox, #style_sysmenu)

[Subburb]

wow ! nice approach !

[DanielBrew]

wow ! nice approach !

Thanks
If you want to make it more accurate you could make the dialog bigger. Also if you want to position the points exactly there is an optional lower toolbar you can add by adding #lowerToolbar to the array of uiFlags.

[cecofuli]

here a small test i made a test with Bertrands Chicago scene,
19min with bruteforce-LC using my method at 1600px, 1h 38h at 5000px wide.
of course it is not fully textured and has no furniture elements,
for the test i put a light gray glossy reflecting material (ior 1.6, intensity 0.5) on all objects, and hdri outside.

brute force, glossy reflection and one hdri dome light, all a subdiv of 256,
AA 1x100x 0.004, DMC all default(threshold 0.01)
colormapping: subpixelmapping on (on makes it dramatically faster for me even though not 100% correct in light then), reinhard 1/0.25/2.2

http://images-stefan.s3.amazonaws.co...BRLC_19min.jpg

i am not sure this says some for fully textured and crowded scenes, but it seems quite ok from noise to me,
AA thrh. at 0.003 would make it more clean of course, render slightly longer

Stefan

Sorry if I if I raise this thread. But, as many users, we fight every days with the noise problem.
I read all what we can read about DMC, noise. etc... The "Stefan method" is good, but if you read well, he wrote: colormapping: subpixelmapping on
with subpixel mapping off, I think with VRay 2.0, with BF+LC (old school or US) it's impossible have good rendering time and no noise (or a few noise). It's impossible (or you have +10 good rendernodes for a single images)
But it's crazy to use 10 rendernode, in the 2012, for a single "normal" image
Again Stefan said: "not 100% correct in light" I reply "yes" only if you don't have glossy, two side material. Otherwise, the differences are very notable (highlight more realistic)
I hope in some "fix" and "development" in the new VRay
We, VRay Users, wrote million thread here and, at the end, no a real solution with sub-pixel mapping ON. Or:

• we have huuuuuuuge rendering time (+50 hours)

• or a lot of noise.

• or, with sub-pixel mapping on, we lose nice highlight

Also, retrace threshold kills the rendering time. In my test x2 time, if you want that this option is slightly visible.

This is my conclusion, after days of test.

New opinions are VERY WELCOME!

[vlado]

Unfortunately I don't have an answer for you. We are researching various options for more efficient sampling, but none has lead to a practical solution so far.

Best regards,
Vlado

[cecofuli]

Hi Vlado, the good news is that you are researching various options. This means that there is an obvious problem.
And Chaosgroup is working on it. We are waiting for a new sampling Good luck!

[super gnu]

i think faster sampling would always be near the top of the to-do list at chaos.. even if everyone was happy with it. i understand people are saying it was faster in earlier builds but ive not noticed any particular slowdown in the later versions myself. having said that i rarely use brute force. we have the lovely imap. with the right settings it can still be pin sharp and faster too. imho, if youre gonna use brute force may as well use maxwell (or rt of course...)

[vlado]

This means that there is an obvious problem.

Nope, no problems as far as I can tell - things are perfectly correct. But it does not mean that things can't be improved

Best regards,
Vlado

[vlado]

i understand people are saying it was faster in earlier builds but ive not noticed any particular slowdown in the later versions myself.

It may only have been faster because it was buggy before... After the fix, for most of our tests, there wasn't a great difference in render times, and many were exactly the same.

Best regards,
Vlado

[vlado]

Also, keep in mind that subpixel mapping ON will *always* be wrong. Our research is directed at less noise with that option disabled.

Best regards,
Vlado

[cecofuli]

Ehehe.. sorry... I wanted say: this means that there is a space for improve the sampling engine

super gnu: VRay is so famous for the flexibility:

I want speed, but a lot of test rendering, splotches (sometime) and human time (change settings etc): IM+LC.
I want quality and "one click" solution with high rendering time: BF+LC

But, as we can see, many people, like me, after seeing the rendering time so high and noisily, we started to think: why not Maxwell or Octane?
This is not the road. I want to use VRay for both solutions

[Morbid Angel]

I have a solution for you: time travel to the future

[cecofuli]

Also, keep in mind that subpixel mapping ON will *always* be wrong. Our research is directed at less noise with that option disabled.

Best regards,
Vlado

Yes Vlado, I know. And sometimes is not possible to have a correct render with subpixel mapping ON (diamond shader with aberration for example).
Also, in the old VRay, if I remember correctly, subpixel mapping was ON by default and unexposed in the UI.
But in my last project (big interior, with a lot of little windows and glossy wood everywhere) with BF+LC With (no universal setting) I had 39 hours at 2500px, with LC retrace on.
With "only" subpixel mapping ON and retrace OFF, now I've 11 hours with I7 920 (and less noise).
Ok, the rendering is less "realistic", a little bit darker, and the highlight less beauty. But the client doesn't see the difference.
But, inside me, I understand that this is no "correct" and I could make better rendering. And I don't like think this.