search

zhan-xu / RigNet

Code for SIGGRAPH 2020 paper "RigNet: Neural Rigging for Articulated Characters"
GNU General Public License v3.0
1.36k stars 189 forks source link

Is it possible to process A-pose mesh #47

Closed CrossLee1 closed 3 years ago

CrossLee1 commented 3 years ago

Thanks for your provided code. I got some environment problem and havenot successfully run it to get the result. I see your input mesh are all in T-pose, is it possible to process other posed mesh, like A-pose？Whats the result.

zhan-xu commented 3 years ago

Yes. Models in neutral pose should be fine. The only constraint here is the symmetry constraint. Only if the model is symmetric and normalized, it should work.

CrossLee1 commented 3 years ago

Thanks for your relpy. Two more questions: 1/ I see the predicted joints usually have duplicate ones, such as root, root_dup_0, root_dup_1, why? 2/ can we use the SMPL skeleton structure as input, and use the network to predict the joints location and skinning weights.

zhan-xu commented 3 years ago

  1. This is for the skinning weights prediction when a joint has multiple children joints. Mesh surface is bind to "bones" instead of "joints". If one joint has three children, these three bones can control up to three parts of the mesh. By duplicate joints, I split joints with common parent to different bones. Each duplicated parent joint represents a bone.

  2. This is not supported right now. We are aware that people are interested in fixed skeleton and will work on this later.

tshrjn commented 1 year ago

Is there Now a way to use RigNet with SMPL?

zhan-xu commented 1 year ago

Hi @tshrjn, how do you want to use RigNet with SMPL? SMPL has its own joint regressor and skinning. They also have SMPLX if you want more details.

tshrjn commented 1 year ago

Hi @zhan-xu I wanted to use RigNet to rig a 3D humanoid like asset in a SMPL skeleton and skinning weights?