importONNXLayers, importONNXNetwork, and importONNXFunction can't import my onnx model?

As of March 3, 2021, importONNXLayers, importONNXNetwork, and importONNXFunction in the latest official version of onnx importer/exporter cannot import my onnx model?

My model link: R3D_50.onnx

Extraction code: kuni

net = importONNXLayers('R3D_50.onnx',...
    'OutputLayerType','classification') 
analyzeNetwork(net)

importONNXFunction ：

params = importONNXFunction('R3D_50.onnx',...
    'resnet3D_50');

In both cases, the following internal errors occurred:

Error using nnet.internal.cnn.onnx.onnxmex

Invalid MEX-file

'C:\ProgramData\MATLAB\SupportPackages\R2020b\toolbox\nnet\supportpackages\onnx\+nnet\+internal\+cnn\+onnx\onnxmex.mexw64':

动态链接库(DLL)初始化例程失败。

Error in nnet.internal.cnn.onnx.ModelProto (line 31)

ModelPtr = onnxmex(int32(FuncName.EdeserializeFromFile), filename);

Error in nnet.internal.cnn.onnx.importONNXLayers (line 7)

modelProto =

nnet.internal.cnn.onnx.ModelProto(Filename);

Error in importONNXLayers (line 48)

Layers = nnet.internal.cnn.onnx.importONNXLayers(modelfile, varargin{:});

Error in testONNX2 (line 1)

net = importONNXLayers('D:\python_work\video_classification_pytorch\model\R3D_50.onnx',...

Error using nnet.internal.cnn.onnx.onnxmex

netron visual onnx model:

Answer 1

[cui,xingxing]

As of March 15, 2021, using the latest version of MATLAB 2021a+Deep Learning Toolbox Converter for ONNX Model Format (version21.1.0) can import R3D_50.onnx, but the prerequisite is:

mv C:\ProgramData\MATLAB\SupportPackages\ R2021a\bin\win64\onnxpb.dll C:\ProgramData\MATLAB\SupportPackages\R2021a\toolbox\nnet\supportpackages\onnx\+nnet\+internal\+cnn\+onnx\

[frederic tchuisseu]

how to do this on macos please

[cui,xingxing]

Functions affected by this bug are listed below.

1. importONNXNetwork
2. importONNXLayers
3. importONNXFunction
4. exportONNXNetwork
5. importCaffeNetwork
6. importCaffeLayers

• Email
• Follow
• Provide feedback

Workaround

As a workaround, move the DLL to the directory where the MEX-file is located.

If you use Deep Learning Toolbox Converter for ONNX Model Format, type the following at the MATLAB command line.

>> source = fullfile( matlabshared.supportpkg.getSupportPackageRoot, "bin", "win64", "onnxpb.dll" );

>> destination = fullfile( matlabshared.supportpkg.getSupportPackageRoot, "toolbox", "nnet", "supportpackages", "onnx", "+nnet", "+internal", "+cnn", "+onnx", "onnxpb.dll" );

>> movefile( source, destination );

If you use Deep Learning Toolbox Importer for Caffe Models, type the following at the MATLAB command line.

>> source = fullfile( matlabshared.supportpkg.getSupportPackageRoot, "bin", "win64", "caffelib.dll" );

>> destination = fullfile( matlabshared.supportpkg.getSupportPackageRoot, "toolbox", "nnet", "supportpackages", "caffe_importer", "+nnet", "+internal", "+cnn", "+caffe", "caffelib.dll" );

>> movefile( source, destination );

Reference here:

https://www.mathworks.com/support/bugreports/2465087

Answer 2

[Ting Su]

Hi

Can you tell us which release of MATLAB you are using?

Ting