Skip to content

Draft: Fix #25 BCE loss not working

nicolas david requested to merge NDavid/odeon-landcover:debug_binary_seg into master

This solution introduced two changes :

  • use odeon BCELoss instead of native pytorch BCELoss
  • pass target tensor conversion to long or float from nn/training_engine.py to each loss in nn/losses.py

another solution could be to only make a check in nn/training_engine.py to check the class/instance of the loss used.

Merge request reports