search

chungyiweng / humannerf

HumanNeRF turns a monocular video of moving people into a 360 free-viewpoint video.
MIT License
791 stars 87 forks source link

Transformation from posed space to canonical space #16

Closed caiyongqi closed 2 years ago

caiyongqi commented 2 years ago

It seems that remove the transformation due to the rest pose is required in the original SMPL model, such as: https://github.com/zju3dv/neuralbody/blob/333026fc12f33d5e732008e7094b442a0095a1e2/zju_smpl/smplmodel/lbs.py#L375 But this step doesn't seem to be present in your code, why is that? thanks. MotionBasisComputer in network_util.py: https://github.com/chungyiweng/humannerf/blob/7389c4049c0d5b14afdbac41a605ddad9d51dd4a/core/utils/network_util.py#L157

chungyiweng commented 2 years ago

Hello,

Looks like what they did is to compute the relative transformation of all joints w.r.t the root joint, while what we need is to get the transformation between the observation space and canonical space (i.e., equation 16).