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

Bounding box not working when extracting a mesh with ns-export poisson #3579

Open
maurock opened this issue Jan 23, 2025 · 4 comments
Open

Comments

@maurock
Copy link

maurock commented Jan 23, 2025

Thank you very much for this awesome library! I'm encountering an issue when I extract a mesh from a Nerfacto representation using ns-export poisson. Instead of extracting the object, the resulting mesh represents a hemisphere surrounding the object. This suggests that the white background is being reconstructed, which could be part of the issue.

TO mitigate this problem, I used the --bounding-box parameters to restrict the volume:

ns-export poisson ... --bounding-box-max 0.1 0.1 0.1 --bounding-box-min -0.1 -0.1 -0.1 --use-bounding-box True

However, this did not change the output. I saw that older version of nerfstudio use --obb parameters rather than bounding boxes, so I am wondering whether there is a bug in the current configuration. Could you please advise on whether I am setting the boundaries incorrectly?

Thank you for your time!

@Charley-xiao
Copy link

Same issue. Did you resolve it? 😢

@maurock
Copy link
Author

maurock commented Jan 27, 2025

Hi @Charley-xiao, unfortunately I have not solved the issue yet. Looking at the code, I believe there is a bug in the --bounding-box parameters as they are not used. A suboptimal workaround I am using is to call one of the classes in nerfstudio/scripts/exporter.py. There, you can pass the obb parameters, and they work as intended. Although it is not the same as simply passing the boundaries of a bounding box.

I should mention that the quality of the extracted mesh is extremely poor. They're just floaters in my case, on a simple YCB object with known camera parameters. There might be something else I need to do to extract a proper mesh, and I'd be very grateful for any advice the authors might have. I have used 40 images in the training set and 10 in the test set.

@IsaacBerman
Copy link

IsaacBerman commented Jan 30, 2025

Yes, it seems to have broken in a newer version, causing really bad results on low iteration trainings (this didn't exist in old versions). Does anyone know if the authors have posted a solution anywhere?

@IsaacBerman
Copy link

This seems related/same issue: #3581

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