Hello,thank u for sharing your code,which is helpful and inspring.But I notice that in MaskModule,you use dice_coeff as mask_loss, is this mask_loss used in the phase of training teacher model?Can you explain how the mask_loss work in the phase?Looking forward your reply.
|
mask_loss = dice_coeff(mask_cA.flatten(2)) + dice_coeff(mask_cD.flatten(2)) \ |
Hello,thank u for sharing your code,which is helpful and inspring.But I notice that in MaskModule,you use dice_coeff as mask_loss, is this mask_loss used in the phase of training teacher model?Can you explain how the mask_loss work in the phase?Looking forward your reply.
FreeKD/mmrazor/models/losses/freekd.py
Line 94 in e0874f5