About Training

[KevinKai-0717]

I would like to ask, during the network training process, is it necessary to freeze ResNet50 to keep the parameters of the feature extraction network unchanged, and only train the parameters of the prediction network?

[pjl1995]

It is not necessary.

[LittleQQ]

i can train model with multi-gpu ,while evaling appear error : TypeError: forward() missing 1 required positional argument: ‘inputs‘,how can solve it?thanks very much