I believe the results are for augmenting training data, not for humans to look at.