render sampling quality

[sharktacos]

As you can see in the above image, I am getting some flickering (little white specs at the top). This is caused by a Maya spot light. The sampling is: adaptive DMC with min: 1 max:4 When I raise the sampling to min: 1 max:25 then the flickering goes away, but the render time shoots up too.

My question is whether there are better approaches to dealing with this? Besides these spec dots, the rest of the render looks fine. It seems like vray is calculating these little dots, and then smoothing them out with the high sampling rate. Would it be possible for example to tell vray to just not even calculate such minute detail in the first place in this particular case where I don't really want it? (That is, I want a blurry reflections, but quicker).

Thanks for the insights!

[vlado]

Have you tried turning on the "sub-pixel mapping" option in the Color Mapping settings?

Best regards,
Vlado

[joconnell]

Try leaving your aa low, but putting up the glossy reflection subdivs on that material. The AA in vray isn't always the most efficient way to clean up a scene, sometimes a few more material samples will do it. Don't be worried about using high numbers in your material subdivs either, try 16 / 24 / 32 / 48 or whatever works, vray might never actually get up to using those higher numbers. Here's a good article on some of the dmc stuff - http://interstation3d.com/tutorials/...yfing_dmc.html

[MoonDoggie]

I've read that article now twice. I'm still figuring parts of it out. Really great read.

[sharktacos]

Have you tried turning on the "sub-pixel mapping" option in the Color Mapping settings?

Best regards,
Vlado

sub-pixel mapping does not have any visible effect in the render for that area of this scene.

[sharktacos]

Try leaving your aa low, but putting up the glossy reflection subdivs on that material. The AA in vray isn't always the most efficient way to clean up a scene, sometimes a few more material samples will do it. Don't be worried about using high numbers in your material subdivs either, try 16 / 24 / 32 / 48 or whatever works, vray might never actually get up to using those higher numbers. Here's a good article on some of the dmc stuff - http://interstation3d.com/tutorials/...yfing_dmc.html

Thanks for the article link, I look forward to reading it.
I tried keeping the globals low (min:1 max:4) and cranking the material's reflection subDivs way up (from 8 to 128 ) but did not see any difference in the render. Am I missing something?

[MoonDoggie]

Keeping your max at 4 won't allow the renderer to use all of the subdivisions set for the materials, I believe. It is also tied to your adaptive amount and noise threshold as well. As the article states (paraphrase) if you have your adaptive amount at 1.0 (fully adaptive) this can lead to strangeness because it is dropping samples before it has a good idea whether to keep sampling.

I almost always use: 2, 6-8. Then I control the noise with the clr threshold on the dmc sampler. ~0.003 will be very smooth and max out the max subdivs (often). If noise still isn't gone, the max subdivs need to go up or material subdivs do.

[sharktacos]

Keeping your max at 4 won't allow the renderer to use all of the subdivisions set for the materials, I believe. It is also tied to your adaptive amount and noise threshold as well. As the article states (paraphrase) if you have your adaptive amount at 1.0 (fully adaptive) this can lead to strangeness because it is dropping samples before it has a good idea whether to keep sampling.

I almost always use: 2, 6-8. Then I control the noise with the clr threshold on the dmc sampler. ~0.003 will be very smooth and max out the max subdivs (often). If noise still isn't gone, the max subdivs need to go up or material subdivs do.

Thanks, those are good insights.

I figured out what the problem was on my render: the geometry did not have enough resolution (smoothing) in it for the floor planks. I swapped it out with a simple poly plane and used a bump map for the planks and it renders fine with the low settings.

[sharktacos]

As the article states (paraphrase) if you have your adaptive amount at 1.0 (fully adaptive) this can lead to strangeness because it is dropping samples before it has a good idea whether to keep sampling.

So I have a question about this: I have the adaptive amount set to 1.0 following the Nederhorst settings (same advice is also given in the Universal Settings). If I change them to 0.85 (which I think is the default) then the rendering time goes up dramatically (from 40 seconds to 4 minutes per frame) without improving the image quality. As I understood the article (and it is entirely possible that I am missing something here!) there is a fail-safe with the adaptive min samples that solves any "strangeness" that can result:

"If adaptiveness is set to 1 (fully adaptive), even if you have 100 reflection samples available to sample glossy reflection, Vray will cast only 1-2 reflection samples and it won't be able to make a smart decision if it needs to take more samples. To prevent this, Min Samples is the bottom limit of number of samples that needs to be taken before adaptiveness kicks in. Vray will always take those 8 samples for every glossy value in the scene before it starts to be adaptive."

So since that fail-safe is there, I would think keeping the adaptive amount at 1.0 (fully adaptive) is a good idea since it gives clean results with a much faster speed.

[joconnell]

It's a slightly awkward one and I think there's a few other bits of info we could get from vlado or from vray render statistics - for example it's clear about how aa and material samples work together with your max aa and material subdivisions firing out rays along the normal of your polygons but I'd be wondering how things like the area light samples come into play. Is it a case where the camera rays hit a polygon, then that polygon sees which light is hitting it and then uses the amount of subdivs set in the light subdivs for that pixel?

Shark in your case, the adaptive amount isn't the last word on quality, it's a combination of that and the noise threshold. The adaptive amount will change between brute force higher samples (0.0) and totally adaptive (1.0) which will control how much of the render is based on adaptive sampling (you're totally correct, the more adaptive the better and faster) but it's the noise threshold that controls what quality is good enough. What you might be finding with the quality / speed change when you drop your adaptive amount is that vray is starting to use more aa samples as you drop from 1.0, and if the noise in the render is being caused by material sampling then aa isn't the most efficient way to clean it up - hence your render gets worse and slower.

It kind of seems to me at the minute that a lower adaptive amount will push your aa samples up towards the max number and when that happens vray will start to lower your material subdivs to try and speed things up. The main problem is that sometimes you'll get better results by having higher material subdivs so you have to trick vray into this by turning down your aa!