Ching-Hang Chen
email: [email protected]
This matlab code demo the 3D human pose estimation from single RGB image with given 2D pose landmarks. The 2D pose can be derived by state-of-the-art CNN such as CPM (https://github.com/shihenw/convolutional-pose-machines-release) or Hourglass
Before start, check out the demo videos!
https://www.youtube.com/watch?v=T_gEx7eQqEg
https://www.youtube.com/watch?v=uGBKZ7QQlDI
Download the 3D pose library from the following link: https://drive.google.com/file/d/0BxZCS0CAHYZpRjA2S1FsWDNhcEU/view?usp=sharing&resourcekey=0-sQrEU466ZG3Ty-ulY-cpaQ
Usage:
- Open Matlab, set and include current directory to "Release", put "3D_library.mat" in "Release" folder
- Run demo.m (few examples are provided with associated 2D pose by CPM)
- Implement 2D pose estimation (ex. CPM or Hourglass) to run your own examples
- Change the input image in demo.m (line 26) and load your own 2D pose estimation (line 32)
3D Human Pose Estimation = 2D Pose Estimation + Matching http://arxiv.org/abs/1612.06524