Hi, great work!

I noticed that the pos_weight in binary_classifier_region_abnormal.py is set to 6.0 because normal regions are six times more prevalent than abnormal regions. My question is whether it would be more reasonable to set pos_weight as a vector, as the normal/abnormal ratio for the lungs is likely different from that of the diaphragm.

Additionally, is there any specific rationale behind choosing -1 for logits > -1 ?

Thanks!