How to change extra-data files

[FinMan28]

Hello, can you please tell me where the files body_info_f.npz and body_info_npz in extra-data come from? How are they produced and how to change their content? I would like to change pose and betas of the model.

[liren2515]

If you mean changing the pose and betas of the body, you can get them from any SMPL related body mesh dataset, such as 3DPW. Here is a survey for it https://github.com/tinatiansjz/hmr-survey?tab=readme-ov-file#datasets. You don't need the generate the data as the one in body_info_f.npz.

Since the tfs_c_inv we used here is to align the rest pose of CLOTH3D to the rest pose of SMPL.

data_body = np.load(os.path.join(extra_dir, 'body_info_f.npz'))
tfs_c_inv = torch.FloatTensor(data_body['tfs_c_inv']).to(device)

The rest pose of CLOTH3D is defined as below instead of a zero vector as that defined in SMPL.

pose = pose.reshape(1, 24, 3)
pose[0, 1, 2] = 0.15
pose[0, 2, 2] = -0.15
pose = pose.reshape(1, 72)

[Yashaswini2039]

If you mean changing the pose and betas of the body, you can get them from any SMPL related body mesh dataset, such as 3DPW. Here is a survey for it https://github.com/tinatiansjz/hmr-survey?tab=readme-ov-file#datasets. You don't need the generate the data as the one in body_info_f.npz.

Since the tfs_c_inv we used here is to align the rest pose of CLOTH3D to the rest pose of SMPL.

data_body = np.load(os.path.join(extra_dir, 'body_info_f.npz'))
tfs_c_inv = torch.FloatTensor(data_body['tfs_c_inv']).to(device)

The rest pose of CLOTH3D is defined as below instead of a zero vector as that defined in SMPL.

pose = pose.reshape(1, 24, 3)
pose[0, 1, 2] = 0.15
pose[0, 2, 2] = -0.15
pose = pose.reshape(1, 72)

I have an .obj file containing posed SMPL, how can i use that for draping.
As i checked it out u have taken the pose component from sample-pose.pt file and i wanned to know can i extract that pose parameters from my posed_smpl.obj file

[liren2515]

If you mean changing the pose and betas of the body, you can get them from any SMPL related body mesh dataset, such as 3DPW. Here is a survey for it https://github.com/tinatiansjz/hmr-survey?tab=readme-ov-file#datasets. You don't need the generate the data as the one in body_info_f.npz.
Since the tfs_c_inv we used here is to align the rest pose of CLOTH3D to the rest pose of SMPL.

data_body = np.load(os.path.join(extra_dir, 'body_info_f.npz'))
tfs_c_inv = torch.FloatTensor(data_body['tfs_c_inv']).to(device)

The rest pose of CLOTH3D is defined as below instead of a zero vector as that defined in SMPL.

pose = pose.reshape(1, 24, 3)
pose[0, 1, 2] = 0.15
pose[0, 2, 2] = -0.15
pose = pose.reshape(1, 72)

I have an .obj file containing posed SMPL, how can i use that for draping. As i checked it out u have taken the pose component from sample-pose.pt file and i wanned to know can i extract that pose parameters from my posed_smpl.obj file

Yes, you can. You need to fit smpl model to your obj mesh to get the corresponding \theta and \beta.

[Yashaswini2039]

If you mean changing the pose and betas of the body, you can get them from any SMPL related body mesh dataset, such as 3DPW. Here is a survey for it https://github.com/tinatiansjz/hmr-survey?tab=readme-ov-file#datasets. You don't need the generate the data as the one in body_info_f.npz.
Since the tfs_c_inv we used here is to align the rest pose of CLOTH3D to the rest pose of SMPL.

data_body = np.load(os.path.join(extra_dir, 'body_info_f.npz'))
tfs_c_inv = torch.FloatTensor(data_body['tfs_c_inv']).to(device)

The rest pose of CLOTH3D is defined as below instead of a zero vector as that defined in SMPL.

pose = pose.reshape(1, 24, 3)
pose[0, 1, 2] = 0.15
pose[0, 2, 2] = -0.15
pose = pose.reshape(1, 72)

I have an .obj file containing posed SMPL, how can i use that for draping. As i checked it out u have taken the pose component from sample-pose.pt file and i wanned to know can i extract that pose parameters from my posed_smpl.obj file

Yes, you can. You need to fit smpl model to your obj mesh to get the corresponding \theta and \beta.

Can you please provide mode resources on how to do it