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

How to perform low resolution segmentation #111

Closed
519830100 opened this issue Mar 24, 2020 · 3 comments
Closed

How to perform low resolution segmentation #111

519830100 opened this issue Mar 24, 2020 · 3 comments

Comments

@519830100
Copy link

Thanks for sharing the code. I want to ask how to get the ground truth label if the output is 4s feature map, (ie. how to get the 4s ground truth)

@chrischoy
Copy link
Contributor

It selects a label randomly within the voxel. Other ways to select label would be selecting argmax, or using the label distribution itself. However, currently these alternatives are not implemented yet.

@519830100
Copy link
Author

I don't know if I put it clear. I want to build an FPN architecture, and I want to know the instance or semantic label for every output tower. Is there any fast implementation that I can refer to?

@chrischoy
Copy link
Contributor

I see. It is a bit tricky to implement such models, but it is possible with get_kernel_map with kernel size 1 and sparse_quantize.

First, you use the sparse quantize with quantization 4*s to create labels and associate coordiates. At the last layer that you compute loss, you reorder the coordinates with get_kernel_map.

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