mkl-dnn code generation failed with Index exceeding error

Hi, I'm trying to make mex function with my pretrained network.

I have followed this example: https://kr.mathworks.com/help/deeplearning/examples/deep-learning-prediction-with-intel-mkl-dnn.html

But I ran into this error msg:

??? Error generating code for network net_0. Index exceeds the number of array elements (0).

My SeriesNetwork object has layers as:

1 'input_6' Image Input 1x1x2 images

2 'conv2d_transpose_8' Transposed Convolution 32 4x4x2 transposed convolutions with stride [4 4] and cropping 'same'

3 'batch_normalization_20' Batch Normalization Batch normalization with 32 channels

4 're_lu_8' ReLU ReLU

5 'conv2d_transpose_9' Transposed Convolution 64 4x4x32 transposed convolutions with stride [2 2] and cropping 'same'

6 'batch_normalization_21' Batch Normalization Batch normalization with 64 channels

7 're_lu_9' ReLU ReLU

8 'conv2d_transpose_10' Transposed Convolution 128 4x4x64 transposed convolutions with stride [2 2] and cropping 'same'

9 'batch_normalization_22' Batch Normalization Batch normalization with 128 channels

10 're_lu_10' ReLU ReLU

11 'conv2d_transpose_11' Transposed Convolution 256 4x4x128 transposed convolutions with stride [2 2] and cropping 'same'

12 'batch_normalization_23' Batch Normalization Batch normalization with 256 channels

13 're_lu_11' ReLU ReLU

14 'conv2d_14' Convolution 1 4x4x256 convolutions with stride [1 1] and padding 'same'

15 'conv2d_14_tanh' Tanh Hyperbolic tangent

And I have a function:

function out = microstructure_generator(in)

persistent net

if isempty(net)

net = coder.loadDeepLearningNetwork('MSGen_z2_no_cond.mat', 'net');

end

out = net.predict(in);

Finally, I executed the following commands just like in the example above.

cfg = coder.config('mex');

cfg.TargetLang = 'C++';

cfg.DeepLearningConfig = coder.DeepLearningConfig('mkldnn');

codegen -config cfg microstructure_generator -args {ones(1,1,2,'single')} -report

I tried with adding an output layer or something, but nothing worked.

I think the error msg does not have enough information about the error. I don't get the reason for this. Please help me...

Answer 1

[Chaitanya Mallela]

Save the DAG network as .mat file

Try Codegen by modifying isempty function

function out = microstructure_generator(in)
persistent net
if isempty(net)
    net = coder.loadDeepLearningNetwork('MSGen_z2_no_cond.mat');
end
out = net.predict(in);

[sruthi gundeti]

I am facing the same issue , Did you get the solution ?

[Naga Sai Pavan Swaroop Ainapurapu]

same here. Did you find a solution?