Skip to content

ocha221/PillDefect

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 

Repository files navigation

PillDefect

fold2 When tested on a hold-out set we see that the model has high accuracy (96%), excellent class separation and, importantly, avoids false-negatives. The model has good tolerance to different angles and lighting conditions, and correctly identifies the existence of at least one defective pill even when multiple passing ones are in the frame. This model acts as a classifier and does not segment images. No synthetic images were used during training/testing/validation, but some nano banana pro generations were used for the demo images below.

About

Classifiers for spotting defective pills using a ConvNeXt network and a custom dataset

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages