C++ language conversion of the captured neural network

After importing the trained model created by keras into MATLAB, I am trying to convert it to C++ and use it in Simulink.

I created the code to convert the keras model in .mat file into C++, but I get the following error.

Error: The library method'dltargets.mkldnn.mkldnnApi.updateBuildInfo' does not work. Cause: The number of outputs from the right-hand side of the equal sign is not enough to satisfy the substitution.

function out = myNet_predict(in) %#codegen
inputnet = 'netTransfer.mat';
persistent mynet;
if isempty(mynet)
    mynet = coder.loadDeepLearningNetwork(inputnet);
end
out = predict(mynet,in);
end

in = imread('two.png');
in = imresize(in,[1,784]);

cfg = coder.config('mex');
cfg.TargetLang = 'C++';
cfg.DeepLearningConfig = coder.DeepLearningConfig('mkldnn'); 
codegen -args {ones(1,784,1,'uint8')} -config cfg myNet_predict -report;

Can this be dealt with by rewriting the program? Or is it better to try another method?

[Hariprasad Ravishankar]

Hi Kubota,

Could you please share with us the version of MATLAB you are using?

Would it also be possible to share the network MAT file for us to investigate further?

Hari

[Kubota Junya]

I'm using R2020b version.

The network MAT file is "netTransfer.mat" attached. It was created by learning MNIST of keras.

[wang kai]

any updates. I had similar issues

Answer 1

[Sayan Saha]

This issue is likely because a supported C++ compiler is not set in MATLAB. You can invoke "mex -setup C++" to set the compiler from the list of supported ones as mentioned in https://www.mathworks.com/support/requirements/supported-compilers.html