load_model() couldn't work on multipe same Custom layers

Like what is described in #6900, The code load_model() for a single custom layer works well: model = load_model('Network.hdf5', custom_objects= {'Myclass': Myclass})

but not works for the Model which contains multipe same Custom layers, like AlexNet, which contains 3 LRN, and LRN is my custom layer.

Since I will get something like that:

File "D:/AlexNet.py", line 114, in <module>
   model = load_model('AlexNet.h5', custom_objects= {'LRN': LRN})
 File "E:\Anaconda3\lib\site-packages\keras\engine\saving.py", line 260, in load_model
   model = model_from_config(model_config, custom_objects=custom_objects)
 File "E:\Anaconda3\lib\site-packages\keras\engine\saving.py", line 334, in model_from_config
   return deserialize(config, custom_objects=custom_objects)
 File "E:\Anaconda3\lib\site-packages\keras\layers\__init__.py", line 55, in deserialize
   printable_module_name='layer')
 File "E:\Anaconda3\lib\site-packages\keras\utils\generic_utils.py", line 145, in deserialize_keras_object
   list(custom_objects.items())))
 File "E:\Anaconda3\lib\site-packages\keras\engine\sequential.py", line 293, in from_config
   model.add(layer)
 File "E:\Anaconda3\lib\site-packages\keras\engine\sequential.py", line 193, in add
   self.build()
 File "E:\Anaconda3\lib\site-packages\keras\engine\sequential.py", line 233, in build
   name=self.name)
 File "E:\Anaconda3\lib\site-packages\keras\engine\network.py", line 237, in _init_graph_network
   self.inputs, self.outputs)
 File "E:\Anaconda3\lib\site-packages\keras\engine\network.py", line 1442, in _map_graph_network
   ' times in the model. 
ValueError: The name "LRN2D" is used 2 times in the model. All layer names should be unique.

Which is caused by multipe same Custom layers Since I am working on network prune experiment, it is a little complicated to use model.load_weights(), because the model architecture is always changing. Any idea about this? Appreciation in advance!

[farizrahman4u]

Note that the recommend way to write get_config() is by first calling the base class’s get_config() and the adding items to it:

def get_config(self):
   config = super(CustomLayer, self).get_config()
    config['alpha'] = self.alpha
    config['k'] = self.k
    return config

[farizrahman4u]

This is probably an issue with the Custom Layer. Make sure that the Custom Layer class is not explicitly setting the .name attribute.

For e.g, the following code works:

from keras.layers import *
from keras.models import *

class CustomLayer(Layer):
    def __init__(self, dim=10, **kwargs):
        self.dim = dim
        super(CustomLayer, self).__init__(**kwargs)

a = Input((2,))
b = CustomLayer(3)(a)
c = CustomLayer(4')(b)
model = Model(a, c)

model.save('test.h5')
model2 = load_model('test.h5')

Following will raise ValueError: The name "CustomLayer" is used 2 times in the model. All layer names should be unique.:

from keras.layers import *
from keras.models import *

class CustomLayer(Layer):
    def __init__(self, dim=10, **kwargs):
        self.dim = dim
        super(CustomLayer, self).__init__(**kwargs)
        self.name = 'CustomLayer'   ### DONT DO THIS!

a = Input((2,))
b = CustomLayer(3)(a)
c = CustomLayer(4')(b)
model = Model(a, c)

model.save('test.h5')
model2 = load_model('test.h5')