Settings to Reproduce Reported AUROC

[ChewZY]

Hi @basveeling , we are trying to reproduce the results of your GDensenet (using your gcnn repo), which achieved the AUROC of 96.3 from your paper.

May I know what are exact settings of your GDensenet? From your paper, we only knows about the batch size (64), optimizer (Adam with initial learning rate of 1e-3), learning rate reduction scheme (reduce by 50% after 20 epochs of no improvements on validation loss).

In addition, from what I read, the training was only done using 312*64 = 19,968 train images and 40,000 validation images. May I know how you sampled it from your provided train and validation set, which consist of 262,144 and 32,768 images respectively?

Appreciate if you could assist us in reproducing the results.

Thanks!