Import Keras network From Python to Matlab

Hi,

I want to import a trained TensorFlow Keras model from Python to Matlab. I already have a '.h5' file with this model, but I keep getting the error 'Keras network with multiple inputs or multiple outputs is not supported when one of the network inputs is sequential' when I use the function ImportKerasNetwork or ImportKerasLayers.

Is there any way in solving this? Or is there any other way to import this model into Matlab?

Answer 1

[Sivylla Paraskevopoulou]

You can try to import your TensorFlow model by using the importTensorFlowNetwork function, which is a newer function than importKerasNetwork and supports more options.

You have to save your .h5 model in the SavedModel format to import it by using the importTensorFlowNetwork function.

[Tristan Milleville]

Hi, I have tried this but this gives me the same error again. I think i have to convert the model to a Functional model to make it work

[Sivylla Paraskevopoulou]

The importTensorFlowNetwork function can import TensorFlow networks created with the sequential or functional API, but not models with subclassing.

[Tristan Milleville]

I am using a sequential LSTM network so this shoud not be a problem i guess?

[Sivylla Paraskevopoulou]

The function supports importing LSTM networks. This does not necessarily mean that it supports all the layers in your TensorFlow model for conversion to MATLAB layers. To learn more, read these sections on Autogenerated Custom Layers and Placeholder Layers.

[Tristan Milleville]

It seems that my network is bit too complicated to insert into Matlab. I will connect Matlab with Python and call the Python script from Matlab to run this model. This will be the easiest solution. Thank you for the support!

[Sivylla Paraskevopoulou]

You are welcome! For this workflow, you might find this GitHub example helpful: Image Classification in MATLAB Using TensorFlow.