J_regressor_h36m_smpl.npy

mks0601 / Hand4Whole_RELEASE

Official PyTorch implementation of "Accurate 3D Hand Pose Estimation for Whole-Body 3D Human Mesh Estimation", CVPRW 2022 (Oral.)

MIT License

314 stars 31 forks source link

Open neuripsasd opened 2 years ago

neuripsasd commented 2 years ago

Hi thank you for the nice work!

Can you provide J_regressor_h36m_smpl.npy used in PW3D #26?

mks0601 commented 2 years ago

Hi, actually, use should use Pose2Pose branch (https://github.com/mks0601/Hand4Whole_RELEASE/tree/Pose2Pose/data) for the 3DPW evaluation.

neuripsasd commented 2 years ago

Thank you for the answer.

I've used the released model (snapshot_6.pth.tar) to test on 3DPW dataset, and the results are quite far from the accuracy reported in the paper.

These are the results I got: MPJPE: 526.83 mm PA MPJPE: 404.11 mm MPVPE: 544.21 mm PA MPVPE: 402.07 mm

Do you have any idea?

mks0601 commented 2 years ago

Pose2Pose branch? main branch?

neuripsasd commented 2 years ago

pose2pose branch

mks0601 commented 2 years ago

I guess you used snapshot_6.pth.tar from main branch, not Pose2Pose branch.