Ideas on making V-Ray's sampling UI more intuitive.

[joconnell]

It still mystifies me why people want to make their lives so complicated... they don't care about how many samples any other render engine makes, it beats me why it is such big deal here.

It's always the nagging though that things could be made faster and faster given a set of magic numbers I think, like if we happened upon a perfect situation that we'd be able to get full 3d motion blur and dof, per pixel gi and lots of glossy everything at 5 minutes a frame for a hd interior

With the other engines you just resign yourself to the fact that it'll be slow and move on with your life. With vray it's very possible to get very quick results with a lot of expensive rendering effects, it just takes a bit of effort to get it!

[vlado]

That being said, i do not ever disable shading subdivs division.

Hehe, and to think that I added that option for you specifically Would make me think twice before trying something like that again.

But non the less, one really needs to have some sort of feel for the sampling balance, and some skill of prediction to be successful with Vray. At least if optimal rendertimes and quality are desired.

That's true; but as you pointed out, getting the hang of it is not that hard and comes with a bit of practice. At least you have the option to reduce render times when you need to.

Best regards,
Vlado

[LudvikKoutny]

To express what i mean even better, here's one more post.

The first problem is that if right now, you were to remove sampling settings from Vray, there would be many scenarios, many scenes, that would simply render a lot longer than they really need. So removal of those settings is beneficial only when there's some internal mechanism, which can decide right sampling importance itself.

Second problem is that as long as there those settings are, people will always remain under the impression that they need to touch them in order to have optimal speed/quality ratio. Which is actually true, due to the problem above, but there aren't that many people who actually know how to do it right, because it's really not so simple, and there are no rules of thumb to apply as it really varies between the different scenes a lot.

[LudvikKoutny]

Hehe, and to think that I added that option for you specifically Would make me think twice before trying something like that again.

Yes, you did. And i not only wrote quite a few posts, but also posted quite a few examples that it does not work, or at very least does not do what i expected it would do. It was in this thread -> http://forums.chaosgroup.com/showthr...-correct/page2

Even when i disable it, it still does not behave the way it should. There are still cases where increasing some sampling value improves quality with same, or even better rendertime, and vice-versa. What i originally wanted, before you implemented the checkbox, was a good old Mental Ray style sampling, where increasing some sampling value always means longer rendertime, and better quality. Whatever disable shading subdivs checkbox does, it's definitely not this

[vlado]

Whatever disable shading subdivs checkbox does, it's definitely not this

No, it's not. There is still some adaptivity based on the intensity of materials and lights that remains.

Best regards,
Vlado

[jstrob]

Even when i disable it, it still does not behave the way it should. There are still cases where increasing some sampling value improves quality with same, or even better rendertime, and vice-versa. What i originally wanted, before you implemented the checkbox, was a good old Mental Ray style sampling, where increasing some sampling value always means longer rendertime, and better quality. Whatever disable shading subdivs checkbox does, it's definitely not this

It can't be like that because even if you uncheck the checkbox, there are still 2 samplers and they still have to divide the job between each other. The shading subdiv division only rebalance the secondsary rays automatically when you increase the primary rays subdivs, if you stop rebalancing automatically you will have to do it manually most of the times but that doesn't mean the render time will increase when you increase the subdiv. Cause the render time can go lower if you use the right sampler to do the right job. So if you tell the sec rays sampler to work more by letting him send more samples it can unload the primary ray sampler and decrease render time.

[LudvikKoutny]

Yup, then that's probably what creates the possibility of having either noisier image than it needs to be in given time, or having slower rendertime than it needs to be to achieve given noise level

Anyways, i did not intend to hijack this thread, so to post something more constructive, here is my (naive) idea on how to make Vray sampling more intuitive:

1, Remove all subdiv values from materials, maps, lights, and brute force GI.

2, Remove Global DMC rollout and Image sampler type sensitive rollout (Adaptive image sampler, Adaptive subdivision or Progressive specific rollouts, depending on which option is selected in image sampler type)

3, Replace this with a single rollout called Image sampler, that would have following settings

- Dropdown to choose between progressive and bucket render mode

- A single value box to set Max. amount of AA samples

- A single value box to set Max. amount of secondary (branching) samples

- Noise (cutoff) threshold value

- Time limit value (would be frozen when bucket mode selected)



That's it. All it would really take. 5 knobs.

So in practice, you would have some reasonable defaults, and then you would render an image...

See noise on edges, fine geometry, motion blur or Dof? Simply increase Max AA samples.

See noise in reflections, refractions, SSS, direct or indirect light? Simply increase Max. secondary samples value.

Don't need areas of your image to be super clean? Increase noise threshold?


Of course this is only going to work with some reliable under the hood adaptivity mechanism, which can prevent wasting of samples. But i think current mechanism Vray has would suffice for that, if tweaked a bit

[jstrob]

No, it's not. There is still some adaptivity based on the intensity of materials and lights that remains.

Best regards,
Vlado

And yes Vlado, that one remains a bit mysterious for me. That's what is called importance sampling. But is that importance sampling completely automatic or is there any input value that can have a direct influence on it?

[LudvikKoutny]

It can't be like that because even if you uncheck the checkbox, there are still 2 samplers and they still have to divide the job between each other. The shading subdiv division only rebalance the secondsary rays automatically when you increase the primary rays subdivs, if you stop rebalancing automatically you will have to do it manually most of the times but that doesn't mean the render time will increase when you increase the subdiv. Cause the render time can go lower if you use the right sampler to do the right job. So if you tell the sec rays sampler to work more by letting him send more samples it can unload the primary ray sampler and decrease render time.

See? This is exactly the kind of wacky science I, as a user, do not want to cope with.

Don't get me wrong. Vray is by far a lot better renderer than Mental Ray, but with Mental Ray, i am always sure i am getting most of what it offers, and I am the ultimate master of the decision of where the samples should go. And with something like Corona for example, i am always sure i am getting most of what it offers, without need to twist around any knobs at all

[vlado]

And yes Vlado, that one remains a bit mysterious for me. That's what is called importance sampling. But is that importance sampling completely automatic or is there any input value that can have a direct influence on it?

It is controlled by the "adaptive amount" parameter. If you set it to 0.0, then V-Ray will not do any adaptivity at all for secondary rays.

Best regards,
Vlado

[jstrob]

It is controlled by the "adaptive amount" parameter. If you set it to 0.0, then V-Ray will not do any adaptivity at all for secondary rays.

Best regards,
Vlado

LOL! That was exactly what I was suspecting! So now I understand another big point! Thanks!

Can you tell us what is the exact math the adaptative amount is doing to the importance sampling?

And do the primary rays also do importance sampling? (I could test this with some DOF)

[jstrob]

It is controlled by the "adaptive amount" parameter. If you set it to 0.0, then V-Ray will not do any adaptivity at all for secondary rays.

Best regards,
Vlado

And I just reread the doc: "Amount - controls the extent to which the number of samples depends on the importance of a blurry value. It also controls the minimum number of samples that will be taken. A value of 1.0 means full adaptation; a value of 0.0 means no adaptation. "

Now I understand the meaning of "importance of the blurry value". You were talking about "importance sampling". I should reread the doc over and and over again! i will not read them the same way now that I know everything explained by Akin and cie. I really forgot about that one. And importance is not only brightness but also distance etc.

[devin]

I'm glad this is being discussed, I've been using Vray for about 7 years and although I love it I get tired of all the constant changing and balancing of settings that goes on. I've always wondered why these things aren't automated and why there setup as they are, how changing one setting can have such a dramatic impact on render times while others seem to do nothing. It seem like the engine is overly complicated on purpose and people take a lot of pride in the fact that they've mastered all of it's quirks. I know a lot of people would like to simplify things, one of reasons Corona's getting so much attention right now is because it's pretty easy to operate and I really want to see Vray embrace this. Just my 2 cents.

[vlado]

I've always wondered why these things aren't automated

Because then people complain that they don't have control... besides, there is no easy way to automate that. Adaptive sampling generally means that the render engine tries to guess how to get to the result faster, so by its very definition it involves guesswork to some extent.

and why there setup as they are, how changing one setting can have such a dramatic impact on render times while others seem to do nothing. It seem like the engine is overly complicated on purpose and people take a lot of pride in the fact that they've mastered all of it's quirks. I know a lot of people would like to simplify things, one of reasons Corona's getting so much attention right now is because it's pretty easy to operate and I really want to see Vray embrace this. Just my 2 cents.

You CAN run V-Ray in this mode - use progressive render mode and you'll be fine. Like I've said many times before, IT IS NOT NEEDED TO PLAY WITH THE SETTINGS to such extreme confusion. Every time I've gotten a slow to render scene, going back to the default settings made things a lot faster and cleaner.

But, if people prefer to spend their time tinkering with settings, what am I supposed to do? Well, the obvious answer is, remove the settings V-Ray will still work just fine and don't think that I haven't considered it seriously... The only thing that is stopping me is that sometimes every minute you shave off of a render is really important and it is possible in that case, to find settings that render the scene faster than the default. But, just because V-Ray gives you options to get faster renders, doesn't mean it makes sense to use them always.

Best regards,
Vlado

[devin]

Vlado I understand it's got to be difficult to keep everyone happy. It's too bad there's not an option that would let you toggle between basic settings and advanced settings so the user could choose what level they wanted to work with.