Hi everyone, the V-SENSE [v-sense.scss.tcd.ie] lab of Trinity College Dublin has developed auto-skeleton tools [github.com] that can be used to quickly create skeletal structures from human character meshes. This work was part of the EU SAUCE Project [www.sauceproject.eu] and in collaboration with the DNEG crowd team.

• The tool can extract a skeleton from a human character mesh in any pose if the rest pose (does not support t-pose yet) of the same character is provided.
  
• The generated skeleton works seamlessly with KineFX workflow inside Houdini.
  
• The skeletal structure is compatible with the CMU Graphics Lab Motion Capture Database [mocap.cs.cmu.edu] for any re-targeting.
  

Any feedback is welcome for future improvements. Thank you.

–Pisut

Hi again everyone,

We need your help with the evaluation for the current and future development. If you have a minute to spare, could you answer this simple 4-item questionnaire [forms.gle], please? Thank you.

here is the video demonstrating a use case of our skeleton tools.


OTLs can be downloaded from Orbolts so they can be used with full-commercial licenses:
https://www.orbolt.com/asset/SAUCE::sauce_skeleton_from_rest
https://www.orbolt.com/asset/SAUCE::sauce_skeleton_from_pose

Your feedback is greatly appreciated.

Did you try the Erik character? https://www.sidefx.com/contentlibrary/erik-rig/ [www.sidefx.com]

At first I thought it was just because it wasn't a watertight mesh. But it still didn't work after I fixed this. I've attached my attempt at it.

edward
Did you try the Erik character? https://www.sidefx.com/contentlibrary/erik-rig/ [www.sidefx.com]

At first I thought it was just because it wasn't a watertight mesh. But it still didn't work after I fixed this. I've attached my attempt at it.

Thanks for checking it out. The current version doesn't work with the t-pose or the character that have their arms raised high like Erik. There are also these two papers from this year SIGGRAPH that are very interesting.

RigNet: Neural Rigging for Articulated Characters
https://zhan-xu.github.io/rig-net/ [zhan-xu.github.io]

Skeleton-Aware Networks for Deep Motion Retargeting
https://deepmotionediting.github.io/retargeting [deepmotionediting.github.io]

Ah, I missed that part in your description. However, T-poses are pretty common though. Is it theoretically possible to add support for these? For example, if it was specified as a parameter by the user on the node.

edward
Ah, I missed that part in your description. However, T-poses are pretty common though. Is it theoretically possible to add support for these? For example, if it was specified as a parameter by the user on the node.

We definitely can add that and have a switch in the next iteration.

Indeed these are interesting... it would be amazing to have those under Houdini's roof

VSENSE_2020

edward
Did you try the Erik character? https://www.sidefx.com/contentlibrary/erik-rig/ [www.sidefx.com]

At first I thought it was just because it wasn't a watertight mesh. But it still didn't work after I fixed this. I've attached my attempt at it.

Thanks for checking it out. The current version doesn't work with the t-pose or the character that have their arms raised high like Erik. There are also these two papers from this year SIGGRAPH that are very interesting.

RigNet: Neural Rigging for Articulated Characters
https://zhan-xu.github.io/rig-net/ [zhan-xu.github.io]

Skeleton-Aware Networks for Deep Motion Retargeting
https://deepmotionediting.github.io/retargeting [deepmotionediting.github.io]