search

open-mmlab / mmhuman3d

OpenMMLab 3D Human Parametric Model Toolbox and Benchmark
https://mmhuman3d.readthedocs.io/
Apache License 2.0
1.2k stars 129 forks source link

Question about hybrik code #207

Open dldaisy opened 2 years ago

dldaisy commented 2 years ago

Dear authors, Thanks for the great work. I have a question about the below code. It seems that you have adapted the original code to implement naive hybrik. You change the final_pose_skeleton to relative postition in if train branch as indicated below: https://github.com/open-mmlab/mmhuman3d/blob/2ccdd2484353a77efe27508d207f6ae5349506ef/mmhuman3d/models/utils/inverse_kinematics.py#L71 However, in the code snippet below, you seemed subtract the rotated parent position from the relative final skeleton(instead of absolute position) to get a relative joint position, which doesn't make sense to me: https://github.com/open-mmlab/mmhuman3d/blob/2ccdd2484353a77efe27508d207f6ae5349506ef/mmhuman3d/models/utils/inverse_kinematics.py#L135 Can you explain why you can use relative final_pose_skeleton here? Thanks a lot!

ttxskk commented 2 years ago

Hi @dldaisy, thanks for you point out this. This line looks like a redundant code because the element of the children map will not be -3. I have noticed that the official codes have made some changes, and I will check it. For naive HybrIK, the authors have released it in this issue.

dldaisy commented 2 years ago

That's True. Thanks for your kind reply!