Axis-angle coordinate system direction

[YuyanHuang]

docs/joint_order.md gives the axis-angle coordinate direction of pred_hand_pose as: But I found through smplx that its T-pose is: But if it is in the direction of the above coordinate system, then the human body will be upside down?

[penincillin]

@YuyanHuang Interesting, when I set zero-poses, the resulting human poses are truly upside down. I guess the discrepancy is caused by the different coordinate system cross different renderer.

[YuyanHuang]

@penincillin So the real coordinate direction of the “wrist axis-angle” of pred_hand_pose is indeed:

In addition, I calculated the direction vector of the hand and palm based on some key points of pred_joints_smpl, and then calculated the axis-angle of the wrist based on these two vectors, and found that it is very different from the wrist axis-angle of pred_hand_pose. The included angle is greater than 100 degrees.

pred_joints_smpl is calculated based on pred_hand_pose, why is there such a big difference between the two wrist axis-angles？Thank you!

[penincillin]

@YuyanHuang

• The axis-angle values pred_hand_pose is the relative rotation. For wrist angle, it is the rotation relative to the parent of wrist in the kinematics tree which is the elbow. You might need to double check whether the axis-angle you calculated from pred_joints_smpl is the global rotation are relative rotation.
• The axis-angle has ambiguity, two vectors with different values can represent the same rotation. You might also need to check this.