I'm doing a transfer learning, and notice that there're no difference between:
letterModel = Model(inputs=trained.get_layer('input').output, outputs = output)
and
letterModel = Model(inputs=trained.input, outputs=output)
Is it true? I can't understand how the output of Input layer can be used as input of new model.
Your assumption seems to be correct. trained.get_layer('input').output and trained.input are equal Tensors. To verify, I loaded a Keras Model object:
>>> import tensorflow as tf
>>> model = tf.keras.models.load_model('model.h5')
1) Getting the first Input Layer from the model and printing its output tensor.
Note: You used
model.get_layer( 'input' )to get the input layer for the model. I have usedmodel.layers[0]for the same purpose.
>>> model.layers[0].output
<tf.Tensor 'input_1:0' shape=(?, 49152) dtype=float32>
2) Likewise, printing the model input tensor.
>>> model.input
[<tf.Tensor 'input_1:0' shape=(?, 49152) dtype=float32>, <tf.Tensor 'input_2:0' shape=(?, 49152) dtype=float32>]
The model printed a list of two Tensors as the model has two
Inputlayers. As you can see the 1st Tensor is equal to the Tensor printed in ( 1 ).
3) We can check this via a Boolean expression.
>>> model.input[0] == model.layers[0].output
True
