human_fitting

Fitting SMPL-X to RenderPeople mesh model.

Example

[mesh_data]

• rp_aaron_posed_014_OBJ
  • rp_aaron_posed_014_100k.mtl
  • rp_aaron_posed_014_100k_textured.obj
  • rp_aaron_posed_014_dif.jpg

[data_root]

• rp_{subj_name} (subject_dir)
  • depth
  • images
  • keypts
  • skin
  • pcd.ply
  • pcd_body.ply
  • pcd_normal.npy
  • pose3d.npz
  • skin_label.npy

0. Setting

$ docker pull jbjeong/research:human_fitting
$ conda activate py37

1. Mesh rendering

docker: jbjeong/research:human_fitting

$ cd {home}/human_fitting
$ python mesh_rendering.py --mesh_dir mesh_data --out_dir data_root

Output -> {depth, images, pcd.ply, pcd_normal.npy}

2. Skin detection

[code from: https://github.com/Jeanvit/PySkinDetection] docker: jbjeong/research:human_fitting

$ cd PySkinDetection/src
$ python main.py --dataset data_root

Output -> {skin}

$ cd {home}/human_fitting
$ python labeling_skin_point.py --dataset data_root

Output -> {pcd_body.ply, skin_label.npy}

3. Pose estimation

docker: cwaffles/openpose-python

$ docker pull cwaffles/openpose-python
$ cd /openpose
$ cp {home}/human_fitting/openpose_run.py . 
$ python openpose_run.py --root {home}/human_fitting/data_root

Output -> {keypts}

4. Pose triangulation

docker: jbjeong/research:human_fitting

$ cd {home}/human_fitting
$ python triangulation_mp.py --dataset data_root

Output -> {pose3d.npz}

5. SMPL-X fitting

Download SMPL-X model. (https://github.com/vchoutas/smplx#downloading-the-model).
Put the SMPL-X model named 'models' in smpl-fit folder.
-human_fitting
  -smplx-fit
    -models
      -smplx
        -SMPLX_MALE.npz
        -...
docker: jbjeong/research:human_fitting

$ cd {home}/human_fitting/smplx-fit
$ python fit_smplx_torch.py --root /root/code/human_fitting/data_root

Output -> {

• result_step0.obj
• result_step1.obj
• result_step2.obj
• result_step3.obj
• result_step4.obj
• result_step5.obj
• result_step6.obj
• result_step7.obj
• result_step_face.obj
• result_step_fingers.obj
• result_step_wrist.obj
• output.obj
• output_aux.npz }