As pointed out in #3, although TreeSat is multi-label dataset, we were only taking the first label (in the filename) instead of all the labels in the labels json.
This PR updates the models to be multilabel, with the following results (paper to be updated):
As pointed out in #3, although TreeSat is multi-label dataset, we were only taking the first label (in the filename) instead of all the labels in the labels json.
This PR updates the models to be multilabel, with the following results (paper to be updated):