Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"
Apache License 2.0
1.02k stars 123 forks source link

How to get J_regressor_h36m_correct.npy #80

Closed buptxyb666 closed 11 months ago

buptxyb666 commented 1 year ago

I have some problems when processing AMASS data, the J_regressor_h36m_correct.npy obtained from https://github.com/hongsukchoi/Pose2Mesh_RELEASE/tree/master/data/Human36M cannot be read by np.load(), can you provide a available J_regressor_h36m_correct.npy for me? Furthermore, i found there is no 2D data used in pre-train stage, which is inconsistent with the paper?

ShirleyMaxx commented 1 year ago

Hi, for the first question, I've just downloaded the J_regressor_h36m_correct.npy file from https://github.com/hongsukchoi/Pose2Mesh_RELEASE/tree/master/data/Human36M and it can be correctly loaded by np.load(), the numpy version I used is 1.21.5. Maybe you could have another try.

Walter0807 commented 1 year ago

Furthermore, i found there is no 2D data used in pre-train stage, which is inconsistent with the paper?

We use 2D dataset InstaVariety.

buptxyb666 commented 1 year ago

Hi, for the first question, I've just downloaded the J_regressor_h36m_correct.npy file from https://github.com/hongsukchoi/Pose2Mesh_RELEASE/tree/master/data/Human36M and it can be correctly loaded by np.load(), the numpy version I used is 1.21.5. Maybe you could have another try.

It still doesn't work with numpy 1.21.5.

ShirleyMaxx commented 1 year ago

Hi, for the first question, I've just downloaded the J_regressor_h36m_correct.npy file from https://github.com/hongsukchoi/Pose2Mesh_RELEASE/tree/master/data/Human36M and it can be correctly loaded by np.load(), the numpy version I used is 1.21.5. Maybe you could have another try.

It still doesn't work with numpy 1.21.5.

It's weird😕 Could you please paste the error information here?

dajiaohuang commented 4 months ago

install human_motion_prior from github instead of pypi