Documentation about skeleton and joint angles

[matteorr]

If I understand correctly, in prepare_data_short_term you load the joint angles for each joint of the human skeleton defined in dataset_h36m.py.

Could you please clarify the following:

• Are the offsets obtained by averaging the poses all subjects (training, test or both?)? If not, how are they computed?
• How did you compute the inverse kinematics to get the joint angles from the individual 3D poses?

Thanks a lot!

[dariopavllo]

Hi,

• If I recall correctly, the offsets are those of Subject 1. They can be computed from the joint positions in the original Human3.6m dataset (not the one supplied with this repo, which contains only rotations).
• The joint angles are provided by the Human3.6m dataset and have most likely been computed using their mocap system. I think more information can be found in their paper. We did not perform IK on our own.