J_regressor_extra vs J_regressor_h36m

JiangWenPL / multiperson

Code repository for the paper: "Coherent Reconstruction of Multiple Humans from a Single Image" in CVPR'20

https://jiangwenpl.github.io/multiperson/

376 stars 50 forks source link

J_regressor_extra vs J_regressor_h36m #35

Closed kidkych closed 2 years ago

kidkych commented 2 years ago

What is the difference between the two joint regressors? From what I can tell J_regressor_extra is used to extract smpl paramters during smplifyx fitting operation for dataset generation, while J_regressor_h36m is used solely in loss calculation.

Since J_regressor_extra is used to fit SMPL keypoints to gt keypoints would it not also work for loss calculation?

JiangWenPL commented 2 years ago

I guess that was just related to the legacy code from HMR and SPIN. There might be slight difference on the definition of joints across different dataset, even though they have the same name.