PDA

View Full Version : Pixar's OpenSubdiv Presentation



erikals
12-13-2012, 10:51 PM
Fast...!!

https://vimeo.com/55032699

Jarno
12-13-2012, 11:20 PM
Not shown, the pre-processing time needed after any change to the topology of the mesh, which is substantial for any complex mesh.
When I last looked at it a few months ago, subdividing a 32*32 regular grid to level 6 (about 4 million quads) tool 6 seconds on my machine. And that would have to be done every time you add or remove a vertex, edge, or polygon from the base mesh.
And it has no support for texture maps except PTex style maps.

---JvdL---

Lewis
12-14-2012, 03:25 AM
I see him changing mesh LIVE, no wait ? He is adding bevels/smooth shifts. adding edges.. all looks interactive to me in this video?

erikals
12-14-2012, 07:20 AM
myeah, but only to 20 polygons or so...
i've asked at the vimeo page, let's see...

Skonk
12-14-2012, 07:52 AM
He talks about the issue in the video; sounds like he's saying that they have pre calculated a bunch of stuff for the demos but it shows the "promise" of what could be done with it.

erikals
12-14-2012, 05:58 PM
ok, got a reply, should be faster now...


OpenSubdiv 6 hours ago

Erikals- the preprocessing time in the beta released over the summer is proportional to the total number of final points drawn. In the release candidate made available this week with adaptive tessellation the preprocessing time is much less because we generate cubic patches and let the hardware tessellate those on the GPU. In regular regions (all quads with vertices having 4 incoming edges) it generates one patch per face during preprocessing and doesn't subdivide at all, and around extraordinary vertices or creases it subdivides a bit.

Snosrap
12-14-2012, 09:56 PM
Not shown, the pre-processing time needed after any change to the topology of the mesh, which is substantial for any complex mesh.
When I last looked at it a few months ago, subdividing a 32*32 regular grid to level 6 (about 4 million quads) tool 6 seconds on my machine. And that would have to be done every time you add or remove a vertex, edge, or polygon from the base mesh.
And it has no support for texture maps except PTex style maps.

---JvdL--- So are you saying that the LW group is not interested and is not actively involved with this tech?

probiner
12-14-2012, 10:09 PM
Glad to see this (http://forums.newtek.com/showthread.php?117211-Catmull-Clark-Edge-Weight-Freeze-Optimization) alive :) Now... I want to play with it :D
So many room for improvement for subds, glad to see this is somewhat moving on and they are going in the way to deal with subDs without as much approximation. Looking forward to see that apllied to retopo.

Cheers

Jarno
12-14-2012, 10:49 PM
I me personally am saying that I do not believe that as it stands currently that OpenSubdiv is a viable replacement for the subdivision system in LW.

People see these sorts of technologies and immediately start jumping up and down that it has to be added to LW because it would make things so fast. In reality this can't often be done because, for example (talking generally here, not specifically to OpenSubdiv):

It requires a workflow entirely different from that currently used in LW.
It requires an expensive pre-processing step.
It does not support some important aspects (e.g. UV texture maps with discontinuous entries).
It does not support the sort of meshes produced by LW (e.g. N-gons or polygons with holes in them).
It requires a mesh datastructure entirely different from that used by LW.
It does not solve the actual speed bottleneck in LW (e.g. the time taken to do subdivision in LW is actually dominated by general mesh handling issues, not the subd code itself).


There are however some speed improvements for subdivision coming up in 11.5.

---JvdL---

Tobian
12-15-2012, 07:53 AM
Well that's a shame, that you aren't considering it though perhaps you can take some ideas away from the technology?

I am not sure IF LW should support meshes with holes in them. If it causes the overall performance of CC subD's to slow down they should be dropped. It wouldn't even occur to me to use them, except say by accident, when using a knife operation, but then I'd fix that.

CC's don't seem to support UV's properly either, so yeah... and if mesh handling issues are the major bottleneck, then I guess work on those :) Ok pointing out the obvious I know, but it does seem a shame to have to reinvent the wheel. The Pixar subdivision forms are much more pleasant, and have more complex sharpness and edges etc. Don't get me started on edge sharpness, it's a deal-breaker for using CC's (along with UV issues) for anything other than initial messing about, so it is about more than just the accelerated nature of the OpenSubDiv technology.

I am REALLY looking forward to the 'speed improvements' in 11.5 though , all that said! I hope they have managed to quash some of the long standing buts with LW modeler, and using SubD's and sped it up a little too!

jwiede
12-15-2012, 10:57 AM
Well then what about the other Pixar Subdivision Surface technology (the same one Luxology and Autodesk licensed)? The for-cost-licensed one works quite well for those products, and being able to interchange subd meshes between apps (with assurance that shape will remain constant) is a big plus. That tech also seems quite fast, doesn't seem to suffer from the same set of limitations as OpenSubdiv. It certainly doesn't hurt that having been repeatedly integrated into other 3D pkgs has yielded a rather solid, mature implementation, nor that the "pSub" tech is more or less a defacto "standard" in the 3D market at this point.

JCG
12-15-2012, 07:41 PM
That tech also seems quite fast, doesn't seem to suffer from the same set of limitations as OpenSubdiv.

I may have misunderstood them, but in the video they mentioned that there was no difference at all between their internal implementation an the free one, and that they had not left any of their subdiv patents out of it.

Snosrap
12-15-2012, 09:17 PM
I me personally am saying that I do not believe that as it stands currently that OpenSubdiv is a viable replacement for the subdivision system in LW.

People see these sorts of technologies and immediately start jumping up and down that it has to be added to LW because it would make things so fast. In reality this can't often be done because, for example (talking generally here, not specifically to OpenSubdiv):

It requires a workflow entirely different from that currently used in LW.
It requires an expensive pre-processing step.
It does not support some important aspects (e.g. UV texture maps with discontinuous entries).
It does not support the sort of meshes produced by LW (e.g. N-gons or polygons with holes in them).
It requires a mesh datastructure entirely different from that used by LW.
It does not solve the actual speed bottleneck in LW (e.g. the time taken to do subdivision in LW is actually dominated by general mesh handling issues, not the subd code itself).


There are however some speed improvements for subdivision coming up in 11.5.

---JvdL--- Very interesting - thanks for being candid Jarno. Personally I don't care what kind of SubD technology you guys use in LW as long as it becomes way better than it is. :)

fahr
12-15-2012, 11:56 PM
The important thing for me is to see the authentic pixar sds algorithms implemented. So we get nondistorted uvs, edge weighting, etc. So even if hardware acceleration is too difficult to implement, we could still be able to make and work with prman sds models that are compatible with every other app that implements it. Gpu acceleration is more of a "nice to have", imho.

Chris S. (Fez)
12-16-2012, 02:44 AM
Very interesting - thanks for being candid Jarno. Personally I don't care what kind of SubD technology you guys use in LW as long as it becomes way better than it is. :)

Agreed 100%. Just relieved to see Jarno/Newtek actively researching/developing subds.

Surrealist.
12-16-2012, 03:10 AM
I think the world of 3D is moving ahead at a rapid pace. And it is a challenge to keep up with it. I'd just like to see NT change whatever they have to at the basic level so that new technologies and current workflows can be embraced. I think really that is what most of us want to see. We are seeing that in many ways with NewTek embracing GOZ and so on. Pizar Subd is one of the big ones that we'd like to see happen I think. And that is where I believe we are coming from as a community. Maybe I am wrong.

Or let me put it this way.

If NT was to announce that by version 12 the plan is to embrace some current technology such as Ptex and Pixar SubD, it would take a lot of notice. A lot. And if it actually happened - if the workflow enhancements were made- then you'd see a new found raise in the interest in rendering in LightWave. And that has been the bread and butter it seems with LightWave as of late.

Snosrap
12-16-2012, 09:40 AM
then you'd see a new found raise in the interest in rendering in LightWave. And that has been the bread and butter it seems with LightWave as of late. Yeah - you bring up a good point. Whatever what might make bringing content over from other apps into LW sure would be a plus.

jwiede
12-16-2012, 02:16 PM
I may have misunderstood them, but in the video they mentioned that there was no difference at all between their internal implementation an the free one, and that they had not left any of their subdiv patents out of it.
Ah, well if that's the case, that's even more reason to embrace OpenSubDiv, and the de-facto industry standard it embodies. That just makes the decision not to pursue PSubs even more disappointing, IMO.

hazmat777
07-15-2013, 02:03 PM
Not long to wait to see the updates! Can't wait!

http://graphics.pixar.com/siggraph2013.html#opensubdiv

erikals
07-15-2013, 02:31 PM
older video >
http://youtu.be/Y-3L9BOTEtw#t=7m07s


newer video >
http://youtu.be/xFZazwvYc5o#t=5m34s

OnlineRender
07-15-2013, 02:39 PM
blenders getting it .........can we have it please DAD please

hrgiger
07-15-2013, 03:12 PM
blenders getting it .........can we have it please DAD please

Its been hinted at. Which means I wouldn't be surprised if it showed up in LW12.

erikals
07-15-2013, 05:19 PM
might speed up LightWave...

hrgiger
07-15-2013, 05:31 PM
Well I think it would be a wise move considering we are long overdue for a fix for our own CC subdivision surface implementation. CORE had subdivision surfaces that were powered by the GPU. Well, I don't think we're going to find a better subdivision surface then Pixar's OpenSubdiv.

DogBoy
07-16-2013, 01:33 AM
Its been hinted at. Which means I wouldn't be surprised if it showed up in LW12.

When was that?

jwiede
07-16-2013, 04:19 AM
Well, I don't think we're going to find a better subdivision surface then Pixar's OpenSubdiv.

They're certainly not likely to find one as mature in implementation, with as many features, and a complete verification suite ready to go as part of the deal. That last part is immensely important, BTW, as it basically ensures quality level and reliable interchange across implementations.

- - - Updated - - -


When was that?

Yep, I'd also like to know where they mentioned that?

Surrealist.
07-16-2013, 04:40 AM
I do think it is something that is coming. It is a large project in the works it seems. It will be nice when they do actually get it finalized and implemented across multiple platforms. It is a great tech.

hrgiger
07-16-2013, 08:09 AM
When was that?

im away from my computer right now and i hate navigating on my mobile so i dont have a link right now... but basically there was another thread on open subdiv a while ago. rob commented on it and said they were looking at it. it reminded me very much of the time he commented on GoZ before we knew that was coming to LightWave. i would say at the very least they see the appeal of it for LightWave.

jwiede
07-16-2013, 11:26 AM
im away from my computer right now and i hate navigating on my mobile so i dont have a link right now... but basically there was another thread on open subdiv a while ago. rob commented on it and said they were looking at it. it reminded me very much of the time he commented on GoZ before we knew that was coming to LightWave. i would say at the very least they see the appeal of it for LightWave.

Ah, okay, "looking at it" isn't quite the same as "working on it" in my book. Still, I do agree their looking into it is a good first step.

hrgiger
07-16-2013, 12:52 PM
Ah, okay, "looking at it" isn't quite the same as "working on it" in my book. Still, I do agree their looking into it is a good first step.

they usually dont comment on something unless theres something to it. at least thats been my impression.

art
07-16-2013, 01:11 PM
This is the thread, I believe. I remember it too.
http://forums.newtek.com/archive/index.php/t-129923.html

hrgiger
07-16-2013, 01:19 PM
This is the thread, I believe. I remember it too.
http://forums.newtek.com/archive/index.php/t-129923.html

yes thank you art. rob quoted someone who said email this to the lightwave developers quick! And then said weve been looking at this for quite a while.

that sounds to me more the just passive interest.

Marcia
07-18-2013, 11:35 AM
Read in 3dWorld this morning that Poser has implemented Pixar SubD in their new release. I don't use Poser products (not even sure who owns them this year), and this may only be news to me, but thought I'd pass it along anyway.