HongwenZhang / PyMAF-X

[TPAMI 2023] PyMAF-X: Towards Well-aligned Full-body Model Regression from Monocular Images
https://www.liuyebin.com/pymaf-x
Other
224 stars 28 forks source link

Why is pkl pose output of shape (1, 72) when normally shape of pose is (1,63) what is the difference? #5

Open AIMads opened 1 year ago

AIMads commented 1 year ago

Hey great work, i try load pose from this method into the blender smplx addon, but it requires the pose being of shape (1,63), but output from this method is either pose with shape (1,72) or body_pose in smplx_paramters of shape (1,216) what are these extra values for?

HongwenZhang commented 1 year ago

Hi, the vector with the shape 63 denotes the pose parameters of 21 body joints (the first 22 joints except the root joint). The vector with the shape 72 denotes the pose parameters of 24 joints (all SMPL joints). The joint names of SMPL and SMPLX can be found at: https://meshcapade.wiki/SMPL#skeleton-layout

liangxuy commented 1 year ago

Thank you for your great work of MyMAF-X. Since the body pose of SMPL_X is not compatible with SMPL, so that it may be not suitable to obtain the body pose of SMPL-X by selecting 21 body poses from the 24 body poses of SMPL. So that, I wonder how to obtain the exact body pose of the SMPL-X model, and also the global orientation and the root translation parameters?