search

jiahaoLjh / HumanDepth

Code for "HDNet: Human Depth Estimation for Multi-Person Camera-Space Localization"
MIT License
28 stars 3 forks source link

How to obtain the 3D pose? #1

Closed zhangyahu1 closed 4 years ago

zhangyahu1 commented 4 years ago

Thanks for sharing your nice work!

From your paper, you obtain the 3D pose in the camera coordinate according to the 2D pose in the pixel coordiante and the depth. Could you please give me some details about how to obtain 3D joint locations?

jiahaoLjh commented 4 years ago

Hi @zhangyahu1

Our framework only estimates the 3D joint location for the root joint. To obtain the full 3D pose, we adopt the same 3D pose estimator as in the ICCV 2019 work by Moon et al. You may refer to the code repo here: https://github.com/mks0601/3DMPPE_POSENET_RELEASE

zhangyahu1 commented 4 years ago

Thanks!

zhangyahu1 commented 4 years ago

Hi @jiahaoLjh,

How do you set self.bin_start and self.bin_end for calculating the depth? Dose they depend on the specific dataset?

jiahaoLjh commented 4 years ago

Hi @zhangyahu1,

We set bin_start and bin_end based on the statistics of the dataset to guarantee that the bins range is large enough to cover the depths of all persons in the dataset.

zhangyahu1 commented 4 years ago

Thanks!