Hi!
I wanted to ask you about specific "eval trick" that you put in here:
|
# trick: eval have effect on BatchNorm only |
It seems to put all modules in "self.cb_linears" into eval model, which generally sounds like a good idea, however I don't see any such batch-norm modules inside cb_linears.
Am I correct that this effectively does nothing? Or am I missing something?
Thank you!
Hi!
I wanted to ask you about specific "eval trick" that you put in here:
CBNetV2/mmdet/models/backbones/cbnet.py
Line 380 in 7233410
It seems to put all modules in "self.cb_linears" into eval model, which generally sounds like a good idea, however I don't see any such batch-norm modules inside cb_linears.
Am I correct that this effectively does nothing? Or am I missing something?
Thank you!