importNetworkFromKeras

Import Keras 3 network as MATLAB network

Since R2026a

collapse all in page

    Syntax

    net = importNetworkFromKeras(modelFolder)

    Description

    Add-On Required: This feature requires the Deep Learning Toolbox Converter for TensorFlow Models add-on.

    net = importNetworkFromKeras(modelFolder) imports a Keras 3 model from the folder modelFolder. The function returns the network net as a dlnetwork object.

    Note

    The importNetworkFromKeras function is recommended for importing models created using the Keras 3 API and saved using the matlabsaver.py utility function. See Save Keras 3 models using matlabsaver.py for more.

    example

    Examples

    collapse all

    This example uses:

    Open Live Script

    This example provides a digitsKeras3Model.zip file containing a trained Keras 3 model saved using the matlabsaver.py utility function. Copy the matlabsaver file to the directory where the Keras model is defined.

    matlabsaverPath = which("matlabsaver.py");
copyfile(matlabsaverPath,<Keras model directory>);

    Save the trained Keras 3 model in Python.

    import matlabsaver
...
matlabsaver.save_for_matlab(trainedModel, "digitsKeras3Model")

    Unzip digitsKeras3Model.zip containing the pretrained Keras 3 network.

    if ~exist('digitsKeras3Model.zip','dir')
    unzip('digitsKeras3Model.zip')
end
modelFolder = './digitsModel';

    Specify the class names.

    classNames = {'0','1','2','3','4','5','6','7','8','9'};

    Use the imported network to predict class labels.

    net = importNetworkFromKeras(modelFolder)
    net = 
  dlnetwork with properties:

         Layers: [10×1 nnet.cnn.layer.Layer]
    Connections: [9×2 table]
     Learnables: [6×3 table]
          State: [0×3 table]
     InputNames: {'input_layer'}
    OutputNames: {'dense'}
    Initialized: 1

  View summary with summary.


    info = analyzeNetwork(net)
    info = 
  NetworkAnalysis with properties:

    TotalLearnables: 34826
          LayerInfo: [10×7 table]
             Issues: [0×3 table]
          NumErrors: 0
        NumWarnings: 0
       AnalysisDate: 24-Jan-2026 18:04:50

    Read the image you want to classify and display the size of the image. The image is a grayscale (one-channel) image with a size of 28-by-28 pixels.

    digitDatasetPath = fullfile(toolboxdir('nnet'),'nndemos','nndatasets','DigitDataset');
I = imread(fullfile(digitDatasetPath,'9','image8020.png'));
size(I)
    ans = 1×2

    28    28

    Display the input size of the network. In this case, the image size matches the network input size. If the sizes do not match, you must resize the image using this command: imresize(I,netInputSize(1:2)).

    netInputSize = net.Layers(1).InputSize
    netInputSize = 1×3

    28    28     1

    Convert the image to a formatted, single-precision dlarray object. Format the images with the dimensions 'SSCB' (spatial, spatial, channel, batch). In this case, the batch size is 1 and you can omit it ('SSC').

    I_dlarray = dlarray(single(I),'SSC');

    Classify the sample image and find the predicted label.

    prob = predict(net,I_dlarray);
[~,label] = max(prob);

    Display the image and the classification result.

    imshow(I)
title(['Classification result ' classNames{label}])

    Figure contains an axes object. The hidden axes object with title Classification result 9 contains an object of type image.

    Input Arguments

    collapse all

    Name of the Keras model folder, specified as a character vector or string scalar. modelFolder must be in the current folder, or you must include a full or relative path to the folder. modelFolder must be created using the matlabsaver.py utility function. See Save Keras 3 models using matlabsaver.py for more.

    Example: "MobileNet"

    Example: "./MobileNet"

    Output Arguments

    collapse all

    Keras 3 network, returned as a dlnetwork object. In some cases, the software indicates that you need to initialize the imported network.

    More About

    collapse all

    Alternative Functionality

    Use importNetworkFromKeras to import a network created using the Keras 3 API or TensorFlow versions 2.16 and later. This is the preferred way of importing networks with Keras 3 specific features such as Keras Ops. If you are using TensorFlow versions 2.15 and earlier or can downgrade to Keras 2, use the importNetworkFromTensorFlow function. This will result in more layers imported as MATLAB layers capable of code generation.

    References

    [1] Keras: The Python Deep Learning library https://keras.io.

    Version History

    Introduced in R2026a

    See Also

    | |

    Topics