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

[DPS]

Sounds good to me.

[jdrouse]

Arnold is commercial now. Arnold's sampling and noise reduction workflow couldn't be easier and people are going to love it. Side by side Vray's current front end needs clear, unified values. This is it.

I think this would be a very welcome change to old users who are not yet mastering v-ray's finer, deeper qualities. New users would benefit greatly. I really believe some of these interface improvements will simplify a lot of misconceptions.

I've used Arnold. I love Vray. The heat is on for this.

[vlado]

Arnold is commercial now. Arnold's sampling and noise reduction workflow couldn't be easier and people are going to love it. Side by side Vray's current front end needs clear, unified values. This is it.

I think this would be a very welcome change to old users who are not yet mastering v-ray's finer, deeper qualities. New users would benefit greatly. I really believe some of these interface improvements will simplify a lot of misconceptions.

I've used Arnold. I love Vray. The heat is on for this.

I don't quite a agree here (and by that I mean that setting up an optimized V-Ray render is not any more slow or complicated than setting up an optimized Arnold render), but I've no desire to go back to that right now. We will probably revisit this again at one point, but for the moment things will stay as they are as there are some more urgent things that need attention.

Best regards,
Vlado

[LudvikKoutny]

I have to say that both 2.4 and 3.0 version do a great job at being as counter-intuitive as possible. What puzzles me even more is that every time i change sampling to be either more eye-ray oriented, or more secondary-ray oriented, it always finds some way to compensate for it and sample additional stuff i do not need/want. Any time i try to optimize anything, it every single time turns against me, then i end up using some sort of brute force settings and suck up the excessive rendertimes.

That was not the case in Mental Ray... even with unified sampling... when i lowered some samples somewhere, it was always simply faster, but noisier... if i raised some samples, it was slower but cleaner. So i had a control over what i do. With Vray, there is so many paradoxes i already lost a track. I often get in ridiculous scenarios whare raising Max AA subdivs makes image more noisy, or where raising light or glossy samples makes image cleaner + render faster. I understand that there is some hidden mechanism to balance sampling... which benefits users who do not know what they do, but that thing is a real obstacle if you are trying to actually understand what you are doing and take advantage of it.

So in the end, Vray has even MORE sampling controls than other renderers, and yet it offers LESS control and freedom when setting the sampling up.

[vlado]

The question is, how to improve this, without making a bigger mess of everything... I guess we could add a hidden option that turns off the sample balancing. You can test it if works ok for you and if so, we can put it in the UI. If you don't have access to the 3.0 nightly builds, email us to vray@chaosgroup.com - the option will be there for the next nightly and is called "dmc_balance_subdivs", it is on by default but you can turn it off from MaxScript.

Best regards,
Vlado

[vlado]

Like this... I hope you will find it useful...



Best regards,
Vlado

[LudvikKoutny]

Our studio already has access to dailies AFAIK, so i will tell person in charge of it to get the last daily

The proposal sounds fair, but only as a proof of concept. Even if sampling works without some black magic going on under the hood, it will just add another button to the button hell that is already there. I believe everyone wanted their knob so they got it... and that is why Vray ended up the way it is today (there is even a checkbox with label "do not use" on it )

Here is my proposal of close-to-ideal scenario.

Vray has quite good adaptive sampling... a good way to detect where samples need to go. If i enable use DMC sampler threshold, then all i have to care about is noise threshold in DMC sampler. Lower values mean slower but cleaner, larger values mean faster but noisier.

Now there are image sampler settings. Min subdivs and max subdivs. Universal settings approach is a proof of concept that these would not even need to be there. Of course, in many scenes, you do not need so many max subdivs. Vray is capable of separating different render elements for very little overhead. So maybe a good combination of a few elements could be used to compute adaptivity from, and then Min/Max subdiv values could be derived from those data without need of user input. That would remove need for Image sampler settings and it's threshold would be derived from DMC settings. It does not make that much of a sense to have that separated, especially in case amount of eye rays would be decided automatically.

Next are local sampling settings... light, glossy, and brute force samples... I think if they all were set equally, removed from UI, and adaptivity of DMC sampler got pushed one step further, it could do the job without need for users to balance samples on every single light, material or GI solution. I know it may sound crazy, but i do not think it is impossible to automatically and adaptively decide how much branching is needed. Especially with as smart adaptivity decision method as Vray has. That would remove need for Global subdivs multiplier.

Adaptivity value could stay. It would drive how much of ray branching decision is based on adaptivity vs how many rays are being shot constantly, regardless of adaptivity

Not sure if min. samples should stay. It seems quite rigid way of error prevention. I think it could be replaced with something internal, that would be a bit more flexible... but nothing as radical as sample balancing that is currently present.

After this, Vray would basically have just one knob... noise threshold.... that would define ratio between speed and quality, and other one, for advanced users, that would define ration between adaptive and more brute force approach.

Of course, this is just ideal dreaming, and has quite far away from practice, when it comes to Vray.


Oh, and last one... Light cache subdivs... I know i was told that i should not use width of output resolution as a driving factor for Light Cache accuracy. But it just does not make any sense. There is no relation between Light Cache and resolution. I mean, if i set LC subdivs to 500, and then render 640*480 image and 4961*3508 image, LC calculation takes roughly the same. But i just can not believe that LC accuracy sufficient for 640*480 res would suffice for 5k res as well. So i still need to touch Light Cache every single time i change resolution. That is a bit of pain. In other renderers, you usually just define basic values like density and accuracy of samples, but at least for density, it is always somehow tied to the render resolution, so with increased resolution, you compute irradiance map adequate for that resolution.

[LudvikKoutny]

And thanks for the checkbox, i will give it a try as soon as i am able to

[vlado]

Our studio already has access to dailies AFAIK, so i will tell person in charge of it to get the last daily

That would be in tomorrows' builds though.

After this, Vray would basically have just one knob... noise threshold.... that would define ratio between speed and quality, and other one, for advanced users, that would define ration between adaptive and more brute force approach. Of course, this is just ideal dreaming, and has quite far away from practice, when it comes to Vray.

It is an admirable goal, and to be honest I have thought about some kind of automatic deduction of sampling values, but this requires research and time, which we don't have for the 3.0 release. But there are always service packs...

For the moment, this is the best that I can do for those crying rivers for Arnold- or mental ray-style sampling.

Oh, and last one... Light cache subdivs... I know i was told that i should not use width of output resolution as a driving factor for Light Cache accuracy. But it just does not make any sense.

You have looked at the example images here: http://docs.chaosgroup.com/display/V...ScaleParameter - do they help to clarify this?

Best regards,
Vlado

[LudvikKoutny]

Regarding LC... Yes, i took a look at examples... and i see it clearly... yet still do not understand it. Basically what i see is that when i have small resolution, a light cache sample may be as small as a few pixels, and when i increase resolution, light cache sample may grow as large as dozens of pixels. That means with increasing resolution, i am losing detail, especially on geometrically dense places. So i guess i should avoid touching subdivs and touch sample size instead. But then again:

1, sample size is not tied to changing resolution in any way, so that has to be done manually.

2, making sample size smaller will make smaller, but also less accurate samples, as it compensates increase in sample density by decreasing their accuracy (From practical observation, lowering sample size does not affect LC calc time either). That is not really helpful either. That just means i should touch two knobs when changing resolution, instead of one.

Or i just got even more confused, than i was.

And as for the research... i am not a CG programmer, so i can not really give any specific suggestion. But for example data for calculating adaptivity for casting eye rays could be derived from certain render elements such as normal pass, diffuse color pass (to antialias textures properly) etc... and passes that are unaffected by noise from secondary rays. And adaptivity for casting secondary eye rays could be computed from those passes, which are riddled with secondary rays noise, such as refraction, reflection, global illumination, and so on. Could be taken a step further and direct lighting sampling adaptivity could be derived from lighting pass... It would require some progressive rendering though, to get that initial pass. But that's in 3.0, so it could work. Then again... i might be completely wrong, so if you can correct me, then i would be only happy to learn new facts

[vlado]

Regarding LC... Yes, i took a look at examples... and i see it clearly... yet still do not understand it. Basically what i see is that when i have small resolution, a light cache sample may be as small as a few pixels, and when i increase resolution, light cache sample may grow as large as dozens of pixels. That means with increasing resolution, i am losing detail, especially on geometrically dense places.

You keep the relative detail though. If you render a large-res print, it is very unlikely that you will still want each light cache sample to still be a few pixels in size.

In general, this can be implemented in two ways: the light cache can be relative to the resolution (as it is now), or it can be in absolute pixels (like the irradiance map is now). Both ways are still valid, and you will always have to adjust one of them depending on what you want to achieve. So if we fix your specific case, we will introduce extra work for other people (those that now have to reduce the irradiance map min/max rate for large-res images).

Whichever way we chose, some users will have more work to do.

Best regards,
Vlado

[RockinAkin]

Like this... I hope you will find it useful...

[ATTACH=CONFIG]18128[/ATTACH]

Best regards,
Vlado

Being able to toggle the division of shading subdivs is definitely a step in the right direction. Thanks Vlado!
(I still think 'min shading rate' belongs down in the Global DMC section... )

[vlado]

Any time i try to optimize anything, it every single time turns against me, then in end up using some sort of brute force settings and suck up the excessive rendertimes.

Btw, I've always wanted to say that... but what happened to the "artists' time is cheaper than machine time" talk? That's exactly what the universal set up is for...

Best regards,
Vlado

[vlado]

Being able to toggle the division of shading subdivs is definitely a step in the right direction.

You have access to the nightly builds, play with it and see how it goes. I still don't think this approach makes things any simpler or easier, but I'd love to be proven wrong

Best regards,
Vlado

[grantwarwick]

Recon442,

I can send to you a link for my mastering vray video that covers the fundamentals of optimising Vray (I even explain why Vray is trying to optimise things for us)
At the end of the day, VRay definitely needed that option to disable the link between Max aa and Global subdivs so thats exciting but really, Vray isn't that hard and you are being a little melodramatic about how difficult it is. (I used to be the exact same)
I'd say it's more a lack of detailed documentation requiring a lot of trial and error so I do feel your pain.