[Feature Request] Skinning with Dual Quaternions

[Maniac Choche]

I've already created a Jira about it. According to the papers the implementation is very simple and only on a shader level. Most of the leading 3D packages already have plugins to support it.

https://jira.secondlife.com/browse/CTS-724

Share your thoughts :)

[Lilith Heart]

Skinning with Dual Quaternions ....... yummy ... want it ......  but want lower PE for large objects first.

Mesh is only HALF ALIVE with the super mega high PEs even for ultra optimised large mesh objects

[Prep Linden]

---

Maniac Choche wrote:

I've already created a Jira about it. According to the papers the implementation is very simple and only on a shader level. Most of the leading 3D packages already have plugins to support it.

https://jira.secondlife.com/browse/CTS-724



Share your thoughts

---

Now that's pretty darn efficient - something like this would be a nice replacement for our linear skinning approach and would also lay the foundation for some nice improvements like muscle bulges and  generally better visual skinning effects.

Definitely noted and I'll be reading it thoroughly tonight - thanks!

 

-prep

[guY Ralior]

Way above my head but the one thing I noticed was that the plugin for Maya only goes to version 8.5 and the implementation in Blender link was down but it looks like it comes out with a much better result on the weight map.

[Gaia Clary]

For your info: This has been implemented in Blender since release 2.46

    http://www.blender.org/development/release-logs/blender-246/skinning/

It can be enabled in the armature modifier by checking the "Preserve Volume" checkmark:

[edit] This method must also be supported by the 3D-engine (the SL-viewer) to take effect. So having it implemented in Blender has no benefit on viewing the rigged meshes in SL.

[Drongle McMahon]

Does it need support in Collada? Or is it just the same data treated differently?

[Maniac Choche]

Actually there is no need for support inside your 3d software of choice. It's cool to have one to be able to see the result in real-time, but the method only interprets the data that is already inside the COLLADA in a different way ( current is linear ). That why the implementation should be easy ( *guessing here* ), because the changes will be only inside the SL Viewer's vertex shaders. Because of the more complex calculations this can slow down the rendering a bit, but in comparison to the other available methods ( log-matrix and spherical blending ) this one is really fast. The current practical fix for "candy wrapping" problem is to add extra bones ( twist bones ) to the skeleton. This can't be achieved in SL at this point. So dual quaternions can be a cheap way to avoid those problems and it can be available only to users on Ultra graphic settings, for example. 

[Gaia Clary]

From what i know about it, it uses the exact same weight data but treets the mesh bending differently. So it sounds like its just a matter of the implementation, hence something to put into the SL-viewer i guess... [edit] as maniac says ...

[guY Ralior]

Wow way sweet, then if you are able to get a not perfect but exceptable weight map now then with that on the viewer side you would have an excellent result, I am so glad they are looking into it that would be a nice feature.