Clarify Instructions Regarding Training Data

Hello,

Is using a pretrained network (trained on fundus images but not labelled glaucoma images) as part of the preprocessing allowed? For example to detect the optic disk.

Thank you

Dear participant,

Thank you for your question.

Please see the new announcement here: https://grand-challenge.org/forums/forum/airogs-609/topic/additional-training-data-update-200/.

For fairness, we do not allow the use of additional fundus image data in your solution, so unfortunately a network pretrained on unlabeled fundus images for preprocessing is also not allowed.

Best regards, Coen de Vente

Are we allowed to play with provided training set - ie, do some manual mask labeling for segmentation?

I am also interested in the response to this question.

Hi @temirgali,

Yes, it is allowed to make and use your own manual annotations on the train set.

Best regards, Coen de Vente

I want to clarify it again. Our manual annotations on the provided training dataset are allowed, such as manual segmentation of disk. Is that true?

Dear austin,

Yes, it is allowed to make and use your own manual annotations on the train set, such as manual segmentations of the optic disk.

Best regards, Coen de Vente