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

Setup.py #7

Open
medif opened this issue Oct 18, 2020 · 4 comments
Open

Setup.py #7

medif opened this issue Oct 18, 2020 · 4 comments

Comments

@medif
Copy link

medif commented Oct 18, 2020

hi, is the file src is named knn in the code because i had the error
import lightconvpoint.knn as nearest_neighbors
ModuleNotFoundError: No module named 'lightconvpoint.knn'
?

@medif
Copy link
Author

medif commented Oct 18, 2020

what is the structure of the folder to segment NPM3D? most of the module in lightconvpoint are not found
is there any folder named knn in lightconvpoint

@medif medif closed this as completed Oct 20, 2020
@aboulch
Copy link
Collaborator

aboulch commented Nov 24, 2020

Hello,
Sorry for this very late response.
Yes, in order to import lightconvpoint.knn, you need to import torch first.
This is something that will be changed.

@medif medif reopened this Nov 24, 2020
@medif
Copy link
Author

medif commented Nov 24, 2020

Hi,
I have question about the way you select the radius can you please explain why you choose:

  • the EMA method
  • 8192 as the number of the support point for the larges scales datasets (as i understand 8192 support points are selected randomly after the selection the voxel size like in the formula cited in page 7)
  • the way you select the support point , as i understand it do not take into account the density of points
    can i have further details please?

@aboulch
Copy link
Collaborator

aboulch commented Jan 28, 2021

Hello,
Sorry for very late response.

  • What is EMA ?
  • We select the points randomly in each voxel, the whole selection process for quantized sampling is not purely random.
  • In quantized sampling, the density of the point is taken into account implicitly: as we select in the voxels, we favors low density locations. However if the density variations are very high and the number of support is high, you still might end up with density problems.

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

2 participants