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

The class 0 of semantic labels #19

Open
sysu19351158 opened this issue Feb 21, 2024 · 2 comments
Open

The class 0 of semantic labels #19

sysu19351158 opened this issue Feb 21, 2024 · 2 comments

Comments

@sysu19351158
Copy link

Thanks for your excellent work. I read the code carefully. And I notice that when calculating the CE loss and the miou, the class 0 seems to be ignored, why?

@danielS91
Copy link
Member

danielS91 commented Feb 22, 2024

Class 0 is the void class that indicates an invalid pixel or a missing annotation (for the datasets we consider). Thus, pixels of the void class should be ignored during loss and metric calculation.
Note that our semantic head also does not predict this class, it can only occur after panoptic merging in predictions (if --instance-offset-distance-threshold is used).

@sysu19351158
Copy link
Author

Thanks a lot! I understand now.

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