Skip to content

BCENa loss is modification of BCE loss to handle unknown categories

Notifications You must be signed in to change notification settings

KornelDylski/bcena

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 

Repository files navigation

BCENa loss function

Notebook contains experiments for BCENa loss, which is tweaked BCE loss that takes into account unknown and unseen before categories. Checks whether changing the target vector of an unknown category to an empty vector would be more effective (or less) than treating it the same as the others.

Code is written in pytorch with fastai2
Dataset used for experiments is Imagenett

labs/bcena-experiments.ipynb - notebook containing experiment

bcena-loss - python module with BCENa loss and customized metrics

About

BCENa loss is modification of BCE loss to handle unknown categories

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published