I’m wondering if it is possible to use the cleaner to view the top losses so I can easily review those and delete them from the dataset if needed?
Basically use the output of interp.plot_top_losses
as the sort order for ImageClassifierCleaner
My thinking was if the model was incorrectly predicting specific images (birds) then perhaps removing those would help with accuracy?
I’m using the CUB_200_2011 dataset
I’m on chapter 3 so maybe something like this is covered later, but so far the course is extremely helpful!