0

I trained a FCNN model using tensorflow with keras backend. The model consists of two separate networks (trained in same graph).

Two mid-layers of this model has same shape ((?,512, 24, 16), 'channel_first'). I would like to extract these two layers (based on 2nd response Keras, How to get the output of each layer?) and map them to each other using a dense layer:

intermediate_layer_net = Model(inputs=model.input,
                             outputs=model.get_layer(name='net1_bn7').output)
intermediate_layer_net2 = Model(inputs=model.input,
                             outputs=model.get_layer(name='net2_bn5').output)

mid_model_in = Input(shape=(512, 24, 16))
mid_model_dense = Dense(512,activation=None, use_bias=True, kernel_initializer="glorot_uniform", bias_initializer="zeros", kernel_regularizer=None, bias_regularizer=None, activity_regularizer=None, name='mid_dense')(mid_model_in)

mid_model = Model(inputs=[mid_model_in], outputs=[mid_model_dense])

mid_model.compile(loss=['mean_squared_error'], optimizer='Adam', metrics=['mae'])

history_mid= mid_model.fit(intermediate_layer_net, intermediate_layer_net2, batch_size=10, epochs=50, validation_split=0.2, shuffle=True)

When I run the code above I get this error: AttributeError: 'Model' object has no attribute 'shape'

how do I pass the shape? shouldn't the shape be included from the first model from which I extract the output layer?

Farnaz
  • 454
  • 6
  • 23
  • 1
    You need to pass NumPy arrays, `tf.data.Dataset` ( other types described [here](https://www.tensorflow.org/api_docs/python/tf/keras/Model#fit) ) to the `mid_model.fit()` method for arguments `x` and `y`. You are passing a instance of `Model` in `mid_model.fit()` – Shubham Panchal Jul 07 '21 at 01:38

0 Answers0