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.

Sign up for GitHub

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

Jump to bottom

Avoid NaNs in RetinaNet training by adding an epsilon to normalizer #1683

Closed
ianstenbit opened this issue Apr 6, 2023 · 1 comment · Fixed by #1684
Closed

Avoid NaNs in RetinaNet training by adding an epsilon to normalizer #1683

ianstenbit opened this issue Apr 6, 2023 · 1 comment · Fixed by #1684
Assignees
@LukeWood

Comments

@ianstenbit
Copy link
Contributor

If a training row has no boxes, the normalizer is 0, causing division by 0 and thus NaNs.

For YOLOv8 I'm just using an epsilon factor for now.

@LukeWood
@LukeWood
Copy link
Contributor

LukeWood commented Apr 7, 2023

Surprised this wasn't problematic in the RetinaNet, thanks for the fix. I will add it to RetinaNet too

@LukeWood LukeWood mentioned this issue Apr 7, 2023
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@LukeWood LukeWood

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Add epsilon for NaN prevention + test case to RetinaNet LukeWood/keras-cv
2 participants
@ianstenbit @LukeWood