About train.py

[yearing1017]

firstly,thanks your code. I have some questions as follow:

1. in danet.py, the output includes three feature maps, then in train.py, the code calculate the loss value separately, and add the three loss value together, but as followed code, as every epoch_loss is the sum of three loss, so shouldn't return epoch_loss / len(train_loader)/3?

 epoch_loss += loss.item()
 return epoch_loss / len(train_loader)

2. in train.py, why the code calculate the loss value separately? is it for the self-learning weight gamma in attention.py ?

3. if just calculate the loss between feats_sum = pam_out + cam_out and label, is ok?

[yearing1017]

@yiskw713