Awesome
MeshLifter
MeshLifter: Weakly supervised approach for 3D human mesh reconstruction from a single 2D pose based on loop structure
To download data files like annotation for Human3.6M https://drive.google.com/drive/folders/1aWzMfDqTOAVnSxYB2Gc_XoyfgW6m653v?usp=sharing
And Pretrained Model https://drive.google.com/file/d/1ZZZJ55vJ8fdRmgzepnHHZYAm_Xm7eRj3/view?usp=sharing
Human3.6M dataset and Parsed file https://drive.google.com/drive/folders/1kgVH-GugrLoc9XyvP6nRoaFpw3TmM5xK
demo.py : To visualize 2D pose(input), reconstructed 3D Mesh and corresponding 2D images(Not input)
test.py : To quantitatively evaluated this project
And you have to change the path of annoatation file, pretrained model and smpl files..
<img src="https://user-images.githubusercontent.com/54057384/88774980-13d66d80-d1bf-11ea-88dc-3df984c2f174.png" width="512">