I am getting an error rate of 0.037 yet when I run DatasetFormatter().from_toplosses(learn_cln) I get indexes for all the images(train + validation). This looks highly unusual as in the next step I have to manually verify their cleanliness. Am I doing something wrong?
I am getting the same issue. from_toplosses is returning the whole ImageList. Has anyone else encountered and solved this problem?
Watching the Lesson 2 video Jeremey mentions that both from_toplosses and from_similars does return ALL indices in order from the the most significant to least for each function. So when calling from_toplosses the first images shown by the image cleaner are the the ones ranked for most loss. Not the most intuitive and probably helpful if that was in the docs.