Open cbyzju opened 1 year ago
https://github.com/Jeff-sjtu/HybrIK/blob/main/hybrik/utils/presets/simple_transform_3d_smpl_cam.py#L475
neural network benefit from the relatively continous space of quat during train stage, I wonder why you change the optimize target from quat to 3x3 rotmats.
https://github.com/Jeff-sjtu/HybrIK/blob/main/hybrik/utils/presets/simple_transform_3d_smpl_cam.py#L475
neural network benefit from the relatively continous space of quat during train stage, I wonder why you change the optimize target from quat to 3x3 rotmats.