error using importONNXNetwork, importONNXNetwork cannot be called from compiled applications.

I used importONNXNetwork in MATLAB R2023b to load a .onnx file, then converted it into a DLL using MATLAB Compiler, and called it under Win64 through Visual Studio 2019. However, when running, I encountered an error: "Error using importONNXNetwork. importONNXNetwork cannot be called from compiled applications." What could be the issue?

Answer 1

[Walter Roberson]

The issue is that you are not permitted to call importONNXNetwork in a compiled application.

You have to break the code into two pieces. The first piece is not to be compiled; it loads the required network, trains it, and saves the trained network into a .mat file. The second piece is to be compiled; it loads the trained network from the .mat file and uses it to predict() or classify()

In compiled applications, you are limited to loading a trained network and using the trained network. Compiled applications cannot import networks or manipulate the layers or train networks.

[Walter Roberson]

You need to load the onnxnet in code that is not compiled, and save it as a .mat file, and load the .mat file in the code that is compiled.

[Alexander Mclean]

Thanks Bro.

I did saved the .onnx file as a .mat file and successfully loaded it with load().net to obtain a DAGNetwork in the MATLAB code and it ran correctly. Then After compiling the whole function(include net = load('xxx.mat').net ) into a DLL and calling it from a C++ console program, the following warning and error were displayed:

'Warning: The variable 'net' originally saved as a DAGNetwork could not be instantiated as an object and will be read in as uint32.

>Location: FunctionName (line 20)

Warning: Data sample time is assumed to be 1 second. To specify a different value for the data sample time, consider providing data using a timetable or an iddata object.

>Location: idpack.iodata.extractRefDataFromAnalysisInputList>localDeduceSampling (line 412)

>Location: idpack.iodata.extractRefDataFromAnalysisInputList (line 108) Location: idmodel.parseCompareResidPredictInputs (line 66)

>Location: predict (line 83)

> Location: FunctionName (line 43)

Error using predict The number of inputs and outputs of the model must match that of the data.'

What's going on here? Is there a problem with a detail?"

[Walter Roberson]

You need to add

%#function DAGNetwork

as a comment in the code to be deployed.

[Alexander Mclean]

Thank you, brother, you are a genius.The problem is solved！ ^_^