Need some help. Many thanks.

• The training set showed an imbalance in both NRG/RG labels and ten glaucoma feature labels, did the hidden test set do the same?

• Can you provide baseline results for the multilabel classification of the ten glaucoma characteristics? My baseline results for multi-label classification are very poor and I'm not sure what the problem is, for example inaccurate labels, is this possible?

• It is necessary to provide more medical knowledge about the ten glaucoma features (e.g., how these features relate to fundus images), I think we should focus more on using computer method to solve medical problems...

Hello,

These are your responses, respectively:

1) Yes.

2) No.

3) For more information about the dataset please see this paper.

Thanks!