Losses used

[SharhadBashar]

Hi
In your paper you mention you train for 20 epochs
the first 5 you use only classification loss
the next 15 self-supervised segmentation loss
Where are these losses defined?

[SharhadBashar]

Also in the paper you mention training for 20 epochs
but in the core.config.py file, __C.TRAIN.NUM_EPOCHS = 15
is that supposed to be 20?

[SharhadBashar]

Also in train.py you have these lines:
stat.add_val("loss") stat.add_val("loss_cls") stat.add_val("loss_fg") stat.add_val("loss_bce")
what are these losses?