Skip to content

Latest commit

 

History

History
24 lines (21 loc) · 1.35 KB

MAKE_DATA.md

File metadata and controls

24 lines (21 loc) · 1.35 KB

Data Documentation

We provide a scripts for rendering training data from human scans, many thanks Ruizhi Shao for sharing this code. Take THuman2.0 as an example.

  • Download THuman2.0 scan data from This Link and the SMPL-X fitting parameters from This Link. Then spilt the THuman2.0 scans into train set and validation set. We use theSMPL-X parameters to normalization the orientation of human and it is not essential. Comment L133-L140 in render_data.py if you do not need to normalization the orientation of human scans in THuman2.0.
./Thuman2.0
├── THuman2.0_Smpl_X_Paras/
├── train/
│   ├── 0004/
│   │   ├── 0004.obj
│   │   ├── material0.jpeg
│   │   └── material0.mtl
│   ├── 0005
│   ├── 0007
│   └── ...
└──val
    ├── 0000
    ├── 0001
    ├── 0002
    └── ...
  • Set the correct thuman_root and save_root in render_data.py. Reset the cam_nums , scene_radius and camera parameters to make the training data similar to your targeting real-world scenario.