Additional training data update

We have just made a change in the rules of the challenge regarding the use of additional training data. We first stated that the use of additional training data is not allowed. We have now decided that only the use of additional fundus image data (including weights pretrained on fundus image data) is prohibited. Hence, pretraining on images such as those from ImageNet is allowed.

As we received some questions about this matter, we would like to make the rule above more explicit:

The use of additional fundus image data (including weights pretrained on fundus image data and in a pre-processing step, such as optic disk segmentation) is prohibited.