QMC gi and noise

[chinate]

Hi. This is my first msg on the forum. I've been reading it for a long time and i've learned a LOT from you guys. (Xcuse me for my english)

I've been usin im+lc on gi solution until some time a go cuz i didn't get de desired results on image, specially on image detail and some kind of low contrasted renders. Since i moved to QMc + GI dose problems were gone But i found out another problem even bigger: Noise. I've read QMC is a non-blurry solution or some like that. My experience is that it handles better with complex materials, all details are reveled sat and contrast gets better but its impossible to get that noise out. I can have QMC gi 100 subdivs qmc noise treshhold .0005 and 100 on mat and light subdivs that i get same noise than on standard values. 20 min vs 4h renders get the same result. What am i doing wrong?

The greatest difference i get is about image sampling Speacially on aa filer area is best for noise but worst for detail <> catmull-rom is greatest for detail but omg noise is terrible

[dalomar_cg]

Re: QMC gi and noise

Some things about DMC (or QMC, its the same thing)....

There's really very little reason to increase the DMC GI subdivisions beyond 8. It doesn't really do much to smooth the noise.

Sharpening filters will ALWAYS increase both noise and render times. I rarely use them and rarely recommend using them. They have their place, but its typically only for final renders. For my buck, the time is better spent adding a sharpening filter in photoshop.

The Noise threshold is a good way to decrease the noise of a DMC calculation. The lowest I tend to go is .001 lower than that and it gets out of control.

When working with DMC GI, I find the best image sampling method is Adaptive DMC. The two work very well together, and you can actually smooth out the noise of DMC gi with good AA settings within Adaptive DMC.

I know those weren't direct answers to your questions, but I think that will help you get rid of the noise. From your description, I think leaving the AA filter behind and switching over to Adaptive DMC should work for you.

[chinate]

Re: QMC gi and noise

Thnx for response. I've done some other tests. OK dalomar dmc gi subdivs is enough with default, thnx.
Other things; yes i use adaptetive dmc on image sampler. Im using a 5-20 rating for example, cuz i read a tuto where it said it was better 10-20 than 1-100 for getting good geometry sampling specially on far detailed geometry ex: architecture. Is that right?, is it affecting on noise issues? I'm telling that cos the main problem i get with noise is on corners ( less iluminated) with matte materials specially on withe walls. I found aout that adding a poor reflexion layer with .5 or so, on specularity reduces it. But most people dont do that, am i right? Should i add more poor lights than real ones to avoid my problems? do you do that?

I am an industrial designer (student) and i get perfect renders in product renders, which i control 'perfectly' in terms of how to iluminate 'small' objects, controlled scenes.

On the other hand i work for an interiorism company. and of course is there where i want the best results he ;D and of course is in those renders where i get trouble: Murphy

Ah, forgot to say (as my images dont appear), my problem is allways on the ceiling cuz light is there and rectangular light point to the floor.

Thnx in advance.

[chinate]

Re: QMC gi and noise

OMG. I've discovered what's up. Maybe you all did know but i didn't.
After lots of tests i came up with the solution: DMC sample noise treshold values does NOT affect anithing on render when using (at least) DMC + light cache and adaptative dmc image sampler.

So, the noise reductor is noise treshold on the image sampler (dmc)

another think i'll investigate is why the hell i get same render times with image sampling 5-20 or 5-120 ?¿

[dalomar_cg]

Re: QMC gi and noise

OMG. I've discovered what's up. Maybe you all did know but i didn't. After lots of tests i came up with the solution: DMC sample noise treshold values does NOT affect anithing on render when using (at least) DMC + light cache and adaptative dmc image sampler.

That is most certainly not the case. The Noise Threshold value in the DMC sampler references almost EVERY value that is calculated by V-Ray. It may be that there wasn't much response within your specific situation and the range you tested, but that setting is very important. This may wind up being more evident with things like glossy materials, DOF, complex lighting, or other illumination engines, but never the less.

So, the noise reductor is noise treshold on the image sampler (dmc)

When it comes to working with DMC and Adaptive DMC, the noise threshold within the image sampler does have a lot to do with how smooth things will get. I do find that adjusting the noise threshold can sometimes do more than adjust the max subdivisions within adaptive DMC.

another think i'll investigate is why the hell i get same render times with image sampling 5-20 or 5-120 ?¿

One of the things that makes V-Ray so fast is that it only takes the samples that it needs (the concept itself is called early termination). So, if V-Ray only needs 20 subdivisions to get an image that satisfies your given settings, then it will only take 20 subdivisions. This is where the noise threshold kicks in because that setting is what determines whether two values are considered similar enough, and therefore V-Ray can stop the calculation and move to the next one. Chances are with a higher noise threshold that jump from 20 to 120 didn't do much because the higher noise threshold was already reached within the 20 original subdivisions. With a lower noise threshold that increase may very well have a time increase associated with it.

Its also worth noting that going from 5-n instead of 1-n makes the sampling more reliable, so that might have help in the "level playing field" between the two settings.

[chinate]

Re: QMC gi and noise

ThnX! dalomar

Good explanation, now everything starts making sense for me!

[Micha_cg]

Re: QMC gi and noise

Damien, shouldn't be possible to set the subdivs of each material at 120 and the global noise threshold take what is needed? I remember me on test in the past that higher material caused a very long render time. Or was a failure in my test and the noisethreshold was to strong, so that all subdivs are used allways? Or was the test right and the reason for the higher render times was caused by adaptive amount 0.85? That could mean, that 15% are fixed sampling. So, it could be cool to get adaptive amount 1 for all DMC samplings except IM sampling, that should get adaptive amount 0.85 for a stable solutions. I remember me that this idea was discussed in the past, but we forget it because 0.85 was good working. Or is the 0.85 workflow the best, because the user can push the sampling where it is needed instead the global threshold? What do you think?

[dalomar_cg]

Re: QMC gi and noise

I would never recommend or set a material at 120 subdivisions. In my mind there isn't any conceivable reason for that high of a value. I'm also not sure why you wouldn't expect that to cause long render times.

Your getting Adaptive Amount and Noise Threshold mixed together, yet they control very different things. Noise Threshold controls the early termination of a given calculation and Adaptive Amount controls the Importance Sampling of a given calculation. Noise threshold will determine how similar two values are and Adaptive Amount will determine how much influence a given value has on continuing the calculation (based on its brightness). Ultimately, the only places where Adaptive Amount should really comes into play is the darker areas of the scene, and we know what comes of that. DMC is less suceptible to those artifacts than IR, so if you feel like pushing up the Adaptive Amount with DMC, then be my guest.

As to material subdivisions, this is one of the most likely values to be sampled to the number of subdivisions that is specified. The reason being is that there is typically more variation between each of the different samples that are taken. When combined together, they smooth themselves out, but when comparing them individually, they will almost always to noisy to pass the noise threshold.

[Micha_cg]

Re: QMC gi and noise

Since not only the mat subdivs control the noise of a material I thought (as I started with Vray), if the noise threshold determinate the noise intensity of material, than why not set a high mat subdiv and let the noise threshold decide, how much subdivs are needed. But during usage of Vray I found, that both, mat subdiv and noisethreshold, are needed for a clean blury reflection/refraction.

[dalomar_cg]

Re: QMC gi and noise

What's your definition of clean blurry reflections?

I currently follow the general workflow suggestions from Chris Nichols (Sway Studios, Gnomon Workshop DVDs). He suggests that most of the noise be resolved through the image sampler, and it makes a lot of sense. Everything has to pass through the image sampler...the only "loop holes" to that are IR and LC. So rather than high material subdivisions or DMC gi settings, if you have really tight antialiasing, then it will all end up smooth in the end.

So at the moment I actually lower my mat subds to 4 instead of 8 and do most of my "smoothing" through the max subds in my Adaptive DMC settings along with the Adaptive DMC noise threshold. This means that each sample that's taken for a given pixel will take less time, but that V-Ray will generally take more pixel samples because the material calculation is presumably less noise. When you switch the two (ie have a high mat subds and a lower AA) you will have to take a lot of material subds per pixel sample, and more times than not, you'll probably have to take a good percentage (if not all) of the pixel samples you've specified.

Typically I find that this works out a bit faster, although not 100% of the time. The more "noisy" the glossy calculations (either though surface variance or bump) the more likely this will be a bit slower. I however like how its much simpler to make a quick quality adjustment (I don't have to worry about individual mat subds), and I also like the look of Adaptive DMC with "noisy" calculations (DMC, glossies, DOF) as I think it gives it a little more sharpness and depth. Plus, a "noisy" calculation looks "better" with Adaptive DMC vs Adaptive Subdiv in my humble opinion.

[Micha_cg]

Re: QMC gi and noise

For me are is clean blury reflections a reflection without noise.
Is your prefered workflow is integrated in my starterkit 4? The adaptive DMC is set a at 1/32 and the noisethreshold allow get low noise. I found, that this setup avoid the material subdivs are full taken in acount. So, if a material is noisy, than I don't get only this easy controlled and I must use the global noise threshold, that affect all materials/effects in the scene. A 1/32 is quite easy to use and it helps best for sampling fine bump details.
If I remember me right the mat subdivs are not complete ignored and very high mat subdivs can be used to get a blury reflection clean.

Most I use a 1/4 or 1/6 setup with a medium noise threshold of 0.015. The renderings are quite fast and if I see a noise that I don't like, than I increase the mat subdivs of the noisy material only. I have the feeling, so I get the best speed.

So, it seems to be we can choose between two main worflow, one with more global control and one for more local control. And I have the impression a middle way could be a setup with 1/12(16) + 0.01 + matsubdivs where needed.

General is noise a problem for me, if render design products and I want to show the difference between a crisp plastic and a smooth satin plastic. A noisy satin plastic looks like a plastic with a crisp bump and that's the biggest problem, except the speed, that I see for unbiased render engines.

Attached some test images - interesting for me is the noise at the black parts.
left: 1-12 0.01 high mat subdivs
middle:1-16 0.01 high mat subdivs
right:1-64 0.007 low mat subdivs


At the 1-64 test I tried to get the fine noise of the 1-12, but without high mat subdivs. The global 1-64 setup works fine, but what is, if I a noise at the metalic carpaint is fine and the black plastic should be noisefree - than a setup with low image sampler values is faster and allow a direct control of single effects.
Both methods have advantages and disadvantages and I think it's good to know both.

My dream would be, to setup a high image sampler like 1/100 and don't lost the control of single DMC effects. But I found no way for it.

PS: I hope the differences at the example renderings are good enough visible.

[dalomar_cg]

Re: QMC gi and noise

The difference between the renderings is subtle, but I see it. I personally like the last one (1/64) as I think it captures some of the geometry of the tire a little better.

Is your prefered workflow is integrated in my starterkit 4?

I don't know what you mean by this question, but I myself haven't looked at the starter kit (I'm just a little past that stage ) If you're asking if you can include it, you can include any advice that I write on this forum or we exchange via email.

In those cases where I actually want a noisier result I'll go back into a "local adjustment" mode. I'll bring the material that I want the noise from down to 2 (never lower), and increase the other ones back to 8 or 12 if need be to get those a bit smoother. I also have to watch setting the AA to high for that.

[Micha_cg]

Re: QMC gi and noise

My question was, is the workflow you mean the workflow at starterkit 4? You don't write what image sampler subdivs you use. So I thought, it could be the same.

"I currently follow the general workflow suggestions from Chris Nichols (Sway Studios, Gnomon Workshop DVDs). He suggests that most of the noise be resolved through the image sampler, and it makes a lot of sense."

What do it mean for the settings? Values like 1/32 for the image sampler?

[ALTO]

Re: QMC gi and noise

Interesting thread, how about the Interpolation option for glossy refl/refraction, from what i know this is the way to solve the problem Micha mentioned, can we see that in next release?

[dalomar_cg]

Re: QMC gi and noise

From what I remember Chris doesn't suggest numbers...what he suggests is controlling the noise in the image through the image sampler as opposed to local settings. This relates to both material subdivisions, but also DOF, MoBlur, and (to some extent) lights.

As to numbers, its a combination of min/max subdivisions and the noise threshold. Generally I tend to lean towards the noise threshold as being the main determiner of the settings.