LOD, how?

[gregvfx]

HI,

has someone figured how to do LOD with VRay?
I isolated a couple of rules that LOD objects should follow and I haven't tried yet but it seems difficult with the way MAX is handling traceSets...

LOD objects shouldn't see each other in refract and reflection
LOD objects shouldn't cast or receive shadows on each others but cast and receive the rest

Anyone?

That would be very useful for hair for example, you don't want to have the full set in one pixel...

[Art48]

Have you tried the Substitute modifier? It's not exactly LOD but might be okay, depending on your case.

What about the "Level of Detail" Utility? Never tried it, but remembered there was something like Level of Detail hidden there. As I see it from quickly skipping the parameters it is image-size-oriented, not distance-based but I might be wrong.

Further thoughts could include scripting something which makes 3 ProOptimizers for the particular object then enabling/disabling them based on a prerender script wired to camera distance or something similar.

[gregvfx]

switching objects at render time can be done with a callback.
That's not what I am looking for.
When I was working with guerrilla render, there is a special object type where you can input three different shapes and it is blending with stochastic transparency between them depending on the distance to the render view. It is also taking care of the rules I mentioned in my previous post so that you can use it in animation without any "Popping".
I will try and take a look at a "level of detail" but I doubt it is what I am looking for.

[Art48]

Yeah I think I know what you mean - I recently saw exactly that in the new Unreal Engine 4 update (4.11 maybe?), they had exactly that, it was amazing.

For a "non-popping" version I cannot imagine a viable solution at the moment with standard max/vray tools (Except maybe code your own plugin via VRay SDK )

[^Lele^]

Take a look at Pro Optimizer, if it's for geometric detail.
There isn't, currently, anything in V-Ray which will provide you with automated LoD generation or management, at least not for standard geometry (Fur for example has LoD management).
You ought to have, however, all the tools at your disposal, in Max, to be able to achieve what you want: trace sets can be handled with the VRay properties and the vrayOverrideMtl, and if you convert a geo to proxies you will be able to render it as a point cloud (generate it on export), while switching can be done with object properties and the visibility spinner (mayhap tied to whatever heuristic you prefer).
There is no light-linking in Max, everything is done in the light include/exclude lists (and when that fails, i'm afraid, with passes.).

As for hair, renders speed is generally down to screen coverage. If your hair system of choice can handle LoD, then V-Ray will have no problems rendering fewer hair as you instructed the hair system to generate.
Did you notice issues for distant, hairy characters?

Agreeably, it's convoluted, but hopefully things will be better soon.

[gregvfx]

We are using hair farm at the studio but I don't think there is a LOD option. Anyway the issue is also that using transparency isn't really a generic solution. When getting transparent you can see the backside and any overlapping self geometry. The way guerilla was handling it was by using a ratio of stochastic Rays seeing either objects and filtering the result into the pixel. Perfect. The concern with tracesets right now is that if two objects declare each other as non visible is creating a circular dependency, which I find unescesary and can seldom be useful. But still, would be convinient sometimes.
Is there any chance to implement a light linking system that could override max system?
I like its simplicity most of the time but when you want to have things done without multiplying the passes and tricks to get things done... I really don't like rendering the shadows separately and comp them can be a headache.

Like how is vray handling massive? There has to be a LOD system no?

[^Lele^]

I hear ya, on all counts.
As for massive, i have no idea, but it's not down to the renderer: provided the LoD system is in place from the host app, V-Ray'll rended what it's instructed to.
It may be that coming from Guerrilla has you used to the way they do things (trying to do it all themselves, including the host app, and all ancillary techs) while V-Ray's been historically more reliant on the underlying app feeding it the right data (I.e.: Katana, Maya, and so on).
It'll change, in places and for LoDs too, but probably not quickly enough for you to benefit right now.

As to writing a LL system in max, ouch.
Join the beta, and beg AD with us...

This said, are you facing some specific issue which is stopped by the lack of LoDs?

[gregvfx]

Actually I don't have an urgent need for it now but it could be something to add to the wish list of vray.
A vray object with inputs for three vray proxy that could blend the stochastic way depending on the size of the bounding box in pixels would be a must for vegetation. Maybe it is not really an issue to have millions of poly in a vray proxy but I guess it could still improve render time. I don't think we should wait for AD to do a system...

[gregvfx]

The thing is that guerilla is developed by friends. When we were having drinks after work we came with a lot of useful functionalities for immediate use. I was working in maya which I totally don't like but inside guerilla I was finding my wished functions. I find the community of VRay very close to that, minus the beers. The developers are really listening to the community.

Same goes for pheonixFD of course.

[^Lele^]

The devs (i'm surely NOT one of them! I lack the skills, bigtime.) do a lot more than listen, Greg: LoD research has been ongoing for a while.
The thing with it is that there are half a million approaches, and most of them are either problematic, or work only for very specific cases, so instead of doing the cheap thing (f.e. copying what others do), Chaos will come up with their own solution, however long it'll take to get it to a production-ready state.
It'll likely, whichever way they'll decide to go, raise a few eyebrows when it'll come out...
And hey, keep the questions and requests coming, for at the very least they are proof investing in some avenues of research wasn't entirely void of merit...

Come to Utrecht, at EUE, for beers. :P

[gregvfx]

Well for the beers it's a bit of a stretch! I am French based in Singapore!!!
But many thanks for the invitation!

[^Lele^]

LOL, ok, it may be too late, for this year.
You can still plan a holiday back home for june next year, though...