Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Something strange in the expoting obj file #5

Closed
Yuanhong-Zheng opened this issue Dec 23, 2022 · 6 comments
Closed

Something strange in the expoting obj file #5

Yuanhong-Zheng opened this issue Dec 23, 2022 · 6 comments

Comments

@Yuanhong-Zheng
Copy link

Thank you so much for your awesome work,sir!
But something strange happen in the exporting obj file like this:
img

The input is a standard T-pose photo,but the output is skew(including the arms and legs).How could it happen?Maybe my picture is not so suitable for the model to predict. So are there certain requirements or something to pay attention to for the input images?
Thank you again!

@Marianoeli
Copy link

I also encountered this problem, is there anything need to pay attention to?

@YuliangXiu
Copy link
Owner

YuliangXiu commented Dec 23, 2022

I have discussed this failure case in ECON's paper. This issue mainly comes from off-the-shelf HPS (Human Pose & Shape Estimation), which were trained on pseudo 3D ground truth, have bias towards mean pose, or to say, the bent legs / arms.

Apart from replacing the HPS approach with one trained on synthetic data with REAL grond-truth data, another workaround is to re-posed the reconstruction with pre-defined SMPL-X T-Pose parameters as follows:

  1. change the avatarizer.py#L168

from

new_pose = smpl_out_lst[2].full_pose

to

new_pose = smpl_out_lst[0].full_pose
  1. repose the ECON's reconstruction
python -m apps.avatarizer -n {filename}

@Yuanhong-Zheng
Copy link
Author

Thank you so much,I will try it!

@Yuanhong-Zheng
Copy link
Author

Yuanhong-Zheng commented Jan 5, 2023

@YuliangXiu For some reason,I don't want to use the visible face from SMPL-X.But when I set ["hand"] or [] in config/econ.yaml and ran python -m apps.avatarizer -n {filename} (I had changed smpl_out_lst[2] to smpl_out_lst[0]),the output file {filename}_econ_pose.obj was wrong.It was only about 400KB and I coudn't open it.But the {filename}_econ_da.obj was normal.
Is this a normal phenomenon?How can I solve the problem.Is there any way that after running avatarizer.py,my hands are from SMPL-X and my face is actually mine.Thanks!

@YuliangXiu
Copy link
Owner

@YuliangXiu For some reason,I don't want to use the visible face from SMPL-X.But when I set ["hand"] or [] in config/econ.yaml and ran python -m apps.avatarizer -n {filename} (I had changed smpl_out_lst[2] to smpl_out_lst[0]),the output file {filename}_econ_pose.obj was wrong.It was only about 400KB and I coudn't open it.But the {filename}_econ_da.obj was normal. Is this a normal phenomenon?How can I solve the problem.Is there any way that after running avatarizer.py,my hands are from SMPL-X and my face is actually mine.Thanks!

I just updated the code to fix this bug.

@YuliangXiu
Copy link
Owner

@Marianoeli @Yuanhong-Zheng

The issue of bending legs get resolved finally, please check out tricks.md/bending-legs.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants