Subdivision + displacement test

[drBouvierLeduc]

edit : Damn, too slow at typing

Yeah, in the modifier you have to enable that option manually ("backface culling"), depending if you have glass materials or not.
So I guess in Indigo you would need to have some automatic checkup to see if the mat is transparent or not before applying bakface culling.
But I think we can live without backface culling !
Keep up the good work.

[BbB]

Mmmh, another completely non-educated comment from me. But as someone who needs to render at pretty high resolution, I'm a bit wary of solutions that mean more memory consumption and even slower render times.
The exhibitionist bathroom I posted earlier was rendered with a width of 1,600 pixel (as opposed to the 2,400 pixels I use normally). Even so, it crashed Indigo on my 32 bit systems and needed Indigo 64 bits and 4gig of RAM to render. And it was still far from clean after 20 hours.

[OnoSendai]

Yeah.. I'm hearing you BbB.
It's possible it may be 5 years too early for such a technique.
perhaps I need to wait until we all have > 32 cores and 16 GB RAM or whatever

[PureSpider]

perhaps I need to wait until we all have > 32 cores and 16 GB RAM or whatever

Don't... just don't

[OnoSendai]

After further convergence...

You can see a little bit of 'triangle facet aliasing' around the highlight on the sphere, but it's not too bad. And this was using ~300,000 tris, using something like 150 to 200 MB memory for the Indigo process.

Render speed is also tolerable. (> 200K samples/sec)

I think the worst case for a complete micro poly approach would be highly specular reflections from high curvature surfaces, like chromed taps or something like that. IN that case the triangles would have to be significantly smaller than 1 pixel, otherwise the mesh is going to look like a disco ball

EDIT: and unfortunately picking out the highlight for some kind of adaptive subdivision is impossible (in general) before the render takes place

[dougal2]

Looking pretty good! I can't wait to play with this!

[Kram1032]

great improvement so far

[alex22]

The problem you have looks somehow like the ones of realtime terrain renderers (like crysis too). There the triangle density depends on the distance to the camera and the jagginess of the terrain. Would be cool if you could build in such dependencies. Also it would be cool if the subdivision algo could cut irregular, too long triangles in two pieces to make the whole geometry more regular, instead of cutting every triangle the same.

[manitwo]

I think the worst case for a complete micro poly approach would be highly specular reflections from high curvature surfaces, like chromed taps or something like that.

just curious: what about subdivision -> displacement -> subdivision ?

[CTZn]

If can be adaptative (it is already, nice one) and dynamic (I mean wich can cope with variables amount of available memory (full REYES, to say)), yes*, or stick with shading normals. wehrk...

* One hundred times

[CTZn]

An example of dynamic data management, even thought I never heard about this one before. That's an dramatic case of management to me, single computer user.

This work effectively tackles issues related to load balancing and memory allocation by a master-slave paradigm

And if you where tempted to ask, no I haven't suscribed to their service

[OnoSendai]

This image uses Perlin noise of a single frequency as the displacement map:

[OnoSendai]

And this one uses Fractal Brownian Motion (FBM) noise.
Makes a pretty good asteroid generator, I think

[Kosmokrator]

niiiiiiiiiiiiice!!!!
keep it up man...
we waiting for some test version soon!

[drBouvierLeduc]

Yeah : test version !