- why this is multi-class problem when there is 1 label per image?
- why do we have UNK when there are no images corresponding to this class?
To 1. The label is one-hot. 0 0 1 0… means it corresponds to the third category.
To 2. The test dataset (with no public ground truth) contains images that are not in one of the eight categories of the training dataset.
1 Like