Training question

Dear Organizer

In the official mention is the final four outputs. But in the data of train_label.csv only NRG and RG are provided, I would like to ask how to handle the rest of the prediction problem. I did not understand the meaning of the official article.

Dear participant,

The only labels provided are indeed 'referable glaucoma' and 'no referable glaucoma'. The test set, however, also contains ungradable cases next to these classes. It is one of the challenge tasks to come up with a solution for detecting these ungradable cases without using any of this ungradable data from training.

Best regards, Coen

Apology, I did not get thing at all. Does it mean our training data has two classes and our test has three class? So the output layer of our model has three classes? Can you explain it

Indeed, the training data has two classes and the test set has three. Your model does not necessarily need to have three output classes, that is up to the participants. They could also use a different way to compute the ungradability outputs, as is often the case with out-of-distribution detection methods, for example.