Support for remove all normalization layers?

[lanqingxi]

Can you use model_surgery remove all the normalization layers ? or replace it with none object ?

[leondgarse]

Sorry, Haven't checked Github for a long time...
Yes, you could create a similar function like model_surgery.py convert_dense_to_conv that converting Norm layers to keras.layers.Activation('linear') layers. Or use fuse_layer_single_input function for removing them:

from kecam import model_surgery, models
mm = models.AotNet50()
# verbose=1 for checking which layers are removed
fuse_layer_condition=lambda xx: xx['class_name'] in ["BatchNormalization", "LayerNormalization"]
ee = model_surgery.model_surgery.fuse_layer_single_input(mm, fuse_layer_condition=fuse_layer_condition, verbose=1)

# Check result
ee.summary()