-1

In keras.oi under BatchNormalization, Arguments: axis: Integer, the axis that should be normalized (typically the features axis). For instance, after a Conv2D layer with data_format="channels_first", set axis=1 in BatchNormalization.

Does it mean that it only can be used after a Conv2D layer? If my model has Conv3D layers, may I still use BatchNormalization(axis=1)?

Xu Zhang
  • 17
  • 2

1 Answers1

0

If you have a conv3D, then you have have to be careful to chose the right axis for normalization. As it is mentioned in the documentation, if you choose "channels first" for your conv3d, then you can do BatchNormalization with axis=1 on a 3D vector (since the position of the channels will be the same as for the conv2D)

Nakor
  • 1,484
  • 2
  • 13
  • 23