If I build a DataBlock and my get_y
function cannot find a label for an item is there a way to remove it from the pipeline (like label it unlabeled) and then remove it. My intuition is that this is against the concept of the data pipeline and I should make sure that I only get items for which I have labels, is that correct?
Look at the multicat chapter in the fastbook repo. You’ll need to use BCEWithLogitLoss instead of CrossEntropyLoss. You’d want your model to predict an ‘unknown’ classes for objects in the image whose prediction logits are below a certain threshold in your metric