Tryed Replace a TensorFlow-Keras Layer in a pretrained Network (Error netUpdated = initialize(net))

I try to import a pretrained Network from the Savedmodel format. When I use the initialize Function I get the following Error:

Example inputs: Incorrect number of example network inputs. 0 example network inputs provided but network has 1 inputs including 1

unconnected layer inputs.

Layer 'input_1': Unconnected input. Each input must be connected to input data or to the output of another layer.

So I tryed to replace the TensorFlow-Keras Layers (Placeholders) to get a fully connection.

1 'input_1' Input Verification This layer verifies the input to 'input_1' has size [1 875 1] and data format 'UUU'.

2 'tf.__operators__.getitem' PLACEHOLDER LAYER Placeholder for 'SlicingOpLambda' Keras layer

3 'tf.__operators__.getitem_1' PLACEHOLDER LAYER Placeholder for 'SlicingOpLambda' Keras layer

4 'tf.math.subtract' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

5 'tf.math.multiply' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

6 'tf.__operators__.getitem_2' PLACEHOLDER LAYER Placeholder for 'SlicingOpLambda' Keras layer

7 'tf.__operators__.getitem_3' PLACEHOLDER LAYER Placeholder for 'SlicingOpLambda' Keras layer

8 'tf.math.subtract_1' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

9 'tf.math.multiply_1' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

10 'tf.compat.v1.pad' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

11 'tf.compat.v1.pad_1' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

12 'tf.concat' PLACEHOLDER LAYER Placeholder for 'TFOpLambda' Keras layer

13 'conv1' 1-D Convolution 96 7×3 convolutions with stride 3 and padding 'same'

But the Problem is that there are no Corresponding Deep Learning Toolbox Layers. Are there more possibilities to convert TensorFlow-Keras Layers or to replace them? I tryed already to import the model as ONNX and Keras Format.

Answer 1

[Sourabh]

Greetings Mathias,

The list of supported ‘Keras’ layers that the Deep Learning Toolbox Converter supports can be found in the ‘Supported Keras Layers’ section of this documentation page:

• https://www.mathworks.com/help/deeplearning/ref/importkerasnetwork.html#mw_c791d156-e8c2-46d1-bc40-8261cbefb024

If the network contains a layer that isare not supported, then ‘importKerasNetwork’ returns an error message. In this case, you can still use ‘importKerasLayers’ to import the network architecture and weights.

If you import a custom ‘TensorFlow-Keras' layer or if the software cannot convert a ‘TensorFlow-Keras' layer into an equivalent built-in MATLAB layer, you can use ‘importTensorFlowNetwork’ or ‘importTensorFlowLayers’, which try to generate a custom layer. For example, ‘importTensorFlowNetwork’ and ‘importTensorFlowLayers’ generate a custom layer when you import a ‘TensorFlow-Keras' Lambda layer.

The following documentation pages might help you further:

• https://www.mathworks.com/help/deeplearning/ref/importkerasnetwork.html
• https://www.mathworks.com/help/deeplearning/ref/importtensorflownetwork.html

You can try and use the ‘replaceLayer’ function to replace placeholder layers in your imported model:

• https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layergraph.replacelayer.html

Hope this helps.