PDA

View Full Version : Has Anyone Seen This– Adaptive Pixel Subdivision



gerry_g
08-06-2005, 05:59 PM
http://www.spinquad.com/forums/showthread.php?t=7607

otacon
08-06-2005, 06:07 PM
Looks cool.

Kuzey
08-06-2005, 08:25 PM
Interesting...I wonder if you can use that in Modeler. Say, add detail to only parts of the model you want while keeping the rest simple?

Can't wait to see more on 9 :D

Kuzey

UnCommonGrafx
08-06-2005, 08:57 PM
Open the mind cuz thinking out of the box is going to be even more needed.

Think about it: since modelling tasks are now shared betwixt the two, indeed, this is exactly what you should be able to do.

Man, lots more to learn...

Kuzey
08-06-2005, 09:15 PM
My mind is so open things are flying out and other things are falling in :help: :help:

Sometimes "should" does not mean it will happen but that was the old school, the new Newtek is tearing down the walls.

Kuzey

BloodQuest
08-06-2005, 11:45 PM
I wonder if you can use that in Modeler. Say, add detail to only parts of the model you want while keeping the rest simple?
Kuzey

There's an option to save out the subdivided mesh as an object or object sequence.

duke
08-07-2005, 05:53 AM
Wonder how it handles animation in regards to "poly popping", seeing as the subdivision is dependent on the camera.

theo
08-07-2005, 08:03 AM
There's an option to save out the subdivided mesh as an object or object sequence.

Lordy- an object sequence saved from Layout with the amount of polys you are talking about Bloody would make my CPUs form their own trade union and banish me from taking any part in the creative process whatsoever.

Frankly I cannot imagine what you could possible do with an object sequence saved out with LOD info part of the mesh.

I am also think along the lines of Kuzey- But since the detail appears to be dependent on the camera I am not sure where a benefit could be derived bringing the object back into Modeller. Unless I have missed an important detail.

Anyways the whole thing is beautiful so keep up the phenom work Blood.

Lightwolf
08-07-2005, 08:20 AM
Frankly I cannot imagine what you could possible do with an object sequence saved out with LOD info part of the mesh.
Actually, looking at the current SDK, this I how I assume they coded their plugin. It basically does an object replacement (as far as LW is concerned from disk) with an optimized mesh. However, the object doesn't actually reside on disk but gets meshed and created by a special loader plugin (Yeah, I thought about a similar problem some time ago).
Saving the meshes would have to advantage of saving a bit of processing time in return for more time spent on object i/o when rendering.
I don't know how it handles sub-frames though, especially concerning motion blur of changing meshes and popping. It would basically have to re-create the adaptive mesh for every motion blur sampling pass, which means that you'd also end up with more than one object per frame on disk...

Cheers,
Mike

Karmacop
08-07-2005, 08:28 AM
Theo, you could also save a mesh to disk to export to another program that doesn't have this feature.

Lightwolf, they probably only change on integer frames, we all know that's the easy path, and it shouldn't cause too many visual inconsistencies if moving at a speed that'd make the polys pop too much ;)

Lightwolf
08-07-2005, 08:49 AM
Lightwolf, they probably only change on integer frames, we all know that's the easy path, and it shouldn't cause too many visual inconsistencies if moving at a speed that'd make the polys pop too much ;)
I hope they don't in the final imlementation... Popping won't be noticeable at high speeds, but if you intend to do a gracefull swoop over some terrain... pop,pop,pop ;)

Cheers,
Mike

Karmacop
08-07-2005, 09:54 AM
Then you just subdivide more ;)

Well hopefully they'll test this before they release it and implement subframes if needed, but here's my reasoning:

There's always going to be some popping, and the only way to fix it is to subdivide more. Then, if you're moving fast enough to get very noticeable popping you'll have motion blur on, which will hide (hopefully) the popping.

Of course the best way to make sure is to check practicle examples :)

Lightwolf
08-07-2005, 10:24 AM
There's always going to be some popping, and the only way to fix it is to subdivide more.
Not quite, you can use a technique called vertex morphing that completely prevents popping.

Basically, you know how many subdivisions you need for any polygon and you know the rules for when to use how many.
What you do is you tesselate to one subdivision level higher than the current one, but don't move the new vertices to their displaced position, only gradually once you approach the higher amount of subdivisons. So new polygons don't pop into existance, but gradually "move" in to their displaced position. Most game engines use a similar technique for terrain (Actually, so does google maps), trilinear mip-mapping follows the same principle, only concerning textures.

Cheers,
Mike

theo
08-07-2005, 03:59 PM
Theo, you could also save a mesh to disk to export to another program that doesn't have this feature.

Make sense- although camera angles would have to match fairly closely or what would be the purpose?

Captain Obvious
08-07-2005, 04:37 PM
That looks downright awesome.

Karmacop
08-07-2005, 04:54 PM
Make sense- although camera angles would have to match fairly closely or what would be the purpose?

Well you can use plugins to get camera motion/data in and out of maya as far as I know, so you could create optimised meshes in lightwave and export the object sequencealong with the camera data. Other than that yes, there's not much point :)

Thanks for the info about morphing Lightwolf, I didn't know about that :)

Kuzey
08-07-2005, 05:17 PM
There's an option to save out the subdivided mesh as an object or object sequence.

I presume the object is frozen when it's saved out. I think it would be more useful if it kept it's subd state, so you can tab on and off when you load it back into Modeler..keeping the object as simple as possible. But if you could do that kind of thing within Modeler then that would be the icing on the cake.

:D :D

Kuzey

theo
08-07-2005, 05:40 PM
I presume the object is frozen when it's saved out. I think it would be more useful if it kept it's subd state, so you can tab on and off when you load it back into Modeler.

I guess my point about APS as I have interpreted it to mean from Blood's comments is if APS is more-or-less an advancement in LOD'ing then there really is no true benefit for this type of APS for use in Modeller. In most cases this advanced method of mapping pixel subdivision to a camera angle is not something that can be utilized within Modeller, I would suppose.

My interpretation of BloodQuest's comments could be completely wrong and if so I would appreciate a clarification BloodQuest.

My guess, Kuzey, is that you have interpreted APS on this thread to mean a ZBrush type of displacement WITHOUT the LOD factor. And if you are correct then this certainly is more than a beautiful development- though I am not sure there is such a state as "more than beautiful".

My- this is fascinating though.

Kuzey
08-07-2005, 06:02 PM
My guess, Kuzey, is that you have interpreted APS on this thread to mean a ZBrush type of displacement WITHOUT the LOD factor. And if you are correct then this certainly is more than a beautiful development- though I am not sure there is such a state as "more than beautiful".

My- this is fascinating though.

Yes it is fascinating and it would be interesting to see how it develops.

I haven't used ZBrush so I wouldn't know what it does ;)

Kuzey

Karmacop
08-07-2005, 07:40 PM
Yeah, I think the division is based on the camera distance and angle, not on how much the texture needs to subdivide the object.

And part of the subdivision is that it gets converted to triangles when changing resolution, so I think it's keeping it as subd as possible .. though I guess it could not triangulate and keep it as a 5 point poly ...

Captain Obvious
08-07-2005, 07:57 PM
Yeah, I think the division is based on the camera distance and angle, not on how much the texture needs to subdivide the object.
Yes, that was how I figured it as well. Perhaps the "shape" of the displacement texture effects the subdivision as well, but the key feature is that it's based on the distance. This isn't really a new concept, though. I've seen a couple of renderers doing this before. It's really useful, to say the least. The more you zoom in, the greater the mesh detail. Heck, you don't even need bump maps with this stuff. ;)

BloodQuest
08-07-2005, 08:01 PM
Take a look at the thread on spinquad, which includes a list of how many polys are generated at various settings.

I can think of times it might be useful to save it out at a low setting for further tweaking in modeler.

Simon Coombs

PS Must reiterate that I'm an end-user, not developer, so can take no credit for this stuff - just trying to show what it can do.

duke
08-07-2005, 09:49 PM
Take a look at the thread on spinquad, which includes a list of how many polys are generated at various settings.

I can think of times it might be useful to save it out at a low setting for further tweaking in modeler.

Simon Coombs

PS Must reiterate that I'm an end-user, not developer, so can take no credit for this stuff - just trying to show what it can do.

Isn't there some kind of NDA stopping you from yappin' about LW beta's? :\

Karmacop
08-07-2005, 10:31 PM
He was at sig and got to play with LW9 then, I don't think they made him sign an nda just for that :)

BloodQuest
08-08-2005, 02:48 AM
Isn't there some kind of NDA stopping you from yappin' about LW beta's? :\

I've not discussed anything here we didn't cover in the siggraph demos.