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

Question about Refined T-Pose #25

Closed Xiaoming-Zhao closed 2 years ago

Xiaoming-Zhao commented 2 years ago

Hi @chungyiweng , congrats on this great work and thanks a lot for open-sourcing the code. I have a question about the data preprocessing for data in the wild and hope you can give me a hand. Specifically, there is a line to offset tpose_joints with pelvis_pos: https://github.com/chungyiweng/humannerf/blob/7389c4049c0d5b14afdbac41a605ddad9d51dd4a/tools/prepare_wild/prepare_dataset.py#L72-L73

After this step, the tpose_joints for mesh_infos of each frame will differ from canonical_joints. However, I do not find some code snippets to compensate such difference later. Probably I miss something. May I know whether you mind helping me understand why this is needed? Thanks a lot in advance.

chungyiweng commented 2 years ago

Hello Xiaoming,

I did this just for ease of implementation -- you can refer to the issue to get more details.

The offset is compensated by storing pelvis_pos as a global translation (i.e., Th) that will be applied after global rotation (i.e., Rh).