search

Jeff-sjtu / HybrIK

Official code of "HybrIK: A Hybrid Analytical-Neural Inverse Kinematics Solution for 3D Human Pose and Shape Estimation", CVPR 2021
MIT License
1.21k stars 147 forks source link

question about Human3.6M dataset. #56

Closed Frank-Dz closed 1 year ago

Frank-Dz commented 2 years ago

As mentioned in the paper, you are using posenet's h36m dataset. And the training process adopts supervision on joint rotation. Are you using the rotation annotated by posenet?

Are you using the fitted_3d_pose: 17x2 dimensional joint coordinates (in world coordinate system) from the fitted mesh provided by posenet for evaluation?

Thank you very much!

YHDang commented 2 years ago

As mentioned in the paper, you are using posenet's h36m dataset. And the training process adopts supervision on joint rotation. Are you using the rotation annotated by posenet?

Are you using the fitted_3d_pose: 17x2 dimensional joint coordinates (in world coordinate system) from the fitted mesh provided by posenet for evaluation?

Thank you very much!

Hello. Did you download the posenet's h36m dataset successfully?

Jeff-sjtu commented 2 years ago

As mentioned in the paper, you are using posenet's h36m dataset. And the training process adopts supervision on joint rotation. Are you using the rotation annotated by posenet?

Are you using the fitted_3d_pose: 17x2 dimensional joint coordinates (in world coordinate system) from the fitted mesh provided by posenet for evaluation?

Thank you very much!

No, we use the moshed data for both training and evaluation.