About to export .ply file for manul alignmnt

[SHEN YIBO]

For scene, i guess that it should use "data_folder/sparse/points3D.txt or points3D.bin".
but for human pose model, i have no idea about which file should i use to export as .ply.

[Wei Jiang]

We export the vertices of the posed mesh as ply file, so it's basically a point cloud of human meshes.
You can also refer to #42

[SHEN YIBO]

We export the vertices of the posed mesh as ply file, so it's basically a point cloud of human meshes. You can also refer to #42

So, i should generate posed mesh by SFU pose data first?

[Wei Jiang]

Yes, one trick you can do is add a dummy else branch here with 0 translation, 0 rotation and 1 scaling,

ml-neuman/render_reposing.py

Lines 50 to 59 in 0149d25

 def get_manual_alignment(opt): 
     if os.path.basename(opt.scene_dir) == 'bike' and opt.motion_name == 'jumpandroll': 
         manual_trans = np.array([0.08, 0.12, 0.4]) 
         manual_rot = np.array([95.8, 10.4, 1.8]) / 180 * np.pi 
         manual_scale = 0.14 
     else: 
         manual_trans = np.array([0, 0, 0]) 
         manual_rot = np.array([0, 0, 0]) / 180 * np.pi 
         manual_scale = 1 
     return manual_trans, manual_rot, manual_scale 

It will give you the posed mesh from a mocap sequence.
The meshes can be obtained from this line:

ml-neuman/render_reposing.py

Line 144 in 0149d25

 raw_verts, Ts = read_human_poses(opt, scene) 

[SHEN YIBO]

Yes, one trick you can do is add a dummy else branch here with 0 translation, 0 rotation and 1 scaling,

ml-neuman/render_reposing.py

Lines 50 to 59 in 0149d25

 def get_manual_alignment(opt): 
     if os.path.basename(opt.scene_dir) == 'bike' and opt.motion_name == 'jumpandroll': 
         manual_trans = np.array([0.08, 0.12, 0.4]) 
         manual_rot = np.array([95.8, 10.4, 1.8]) / 180 * np.pi 
         manual_scale = 0.14 
     else: 
         manual_trans = np.array([0, 0, 0]) 
         manual_rot = np.array([0, 0, 0]) / 180 * np.pi 
         manual_scale = 1 
     return manual_trans, manual_rot, manual_scale 

It will give you the posed mesh from a mocap sequence.
The meshes can be obtained from this line:

ml-neuman/render_reposing.py

Line 144 in 0149d25

 raw_verts, Ts = read_human_poses(opt, scene) 

Thanks very much!!