fastai_class_label_cleaning.md

How do I start the fastai class label cleaner?

You can start it by executing

import ipywidgets
from fastai.vision.widgets import ImageClassifierCleaner
cleaner = ImageClassifierCleaner(learn)
cleaner

How do I save the results of the cleaning process?

Assuming your dataset is structured with samples in folders with the corrosponding class labels you can directly update your dataset by running

for idx in cleaner.delete():
    cleaner.fns[idx].unlink()

to delete the items marked for deletion and running

for idx, cat in cleaner.change():
    print(cat)
    print(cleaner.fns[idx])
    shutil.move(str(cleaner.fns[idx]), 'data/train/' + cat)

to move the files into the directory which matches their new label. You will likely need to adjust the destination path in the shutil.move command.