search

hongsukchoi / TCMR_RELEASE

Official Pytorch implementation of "Beyond Static Features for Temporally Consistent 3D Human Pose and Shape from a Video", CVPR 2021
MIT License
274 stars 39 forks source link

smpl 매개변수 정보 #11

Open liuhaorandezhanghao opened 2 years ago

liuhaorandezhanghao commented 2 years ago

smpl의 포즈 매개변수는 프레임당 75로 23*4+3입니다. 이 매개변수는 어떻게 계산됩니까? 나는 모든 3이 관절의 회전 벡터이고, 벡터의 모듈러스 길이가 회전 각도(라디안)이고, 벡터의 방향이 회전 방향이라는 통일성을 복원하고 싶습니다. 비슷한 작업을 해보셨나요? 이런 식으로 이론상 잘못된 것은 없지만 실제 결과는 옳지 않습니다.

hongsukchoi commented 2 years ago

Hi @liuhaorandezhanghao,

I think translation in English would be easier for me to understand. Anyway, in this repo, smpl pose parameters have 72 dimension, where 23*3 is the axis-angle rotation of smpl joints, and the remaining 3 is the axis-angle global rotation.

I am not sure what you want to do, but maybe you can checkout the codes here for transforming the rotation representation.

liuhaorandezhanghao commented 2 years ago

Hi @liuhaorandezhanghao,

I think translation in English would be easier for me to understand. Anyway, in this repo, smpl pose parameters have 72 dimension, where 23*3 is the axis-angle rotation of smpl joints, and the remaining 3 is the axis-angle global rotation.

I am not sure what you want to do, but maybe you can checkout the codes here for transforming the rotation representation.

What I mean is that this parameter is expressed in axial angle, and unity happens to be a function of axial angle. I input parameters to restore a specific posture without the desired output posture.