loss value is very huge

[myhussien]

Hi,

Thank you for sharing this code. I was debugging the code to get an idea of how loss is computed, and I understand the BCE part. However, the loss printed is around ~650 using the BaselinePretrained .pth.tar. Even thought eh AUC is 95% and CLE is 11 pixels, but the loss printed is too high.

What does this value represent? is there a normalization/scaling needed to get it in the 0.09 range expected?

Thanks,