Code Generation for Networks

Hallo everyone,

when we want to generate a Code, we should choose a pretrained net like mobilenetv2() and have an entry-point function for this net, type("mobilenetv2 _predict.m") :

% Copyright 2017-2019 The MathWorks, Inc.

function out = mobilenetv2_predict(in)

%#codegen

persistent mynet;

if isempty(mynet)

mynet = coder.loadDeepLearningNetwork('mobilenetv2','mobilenetv2');

end

% pass in input

out = mynet.predict(in);

My question is: What if I train a standalone network for my project? How can I put it in this function to deploy it on Jetson nano?

Thank you very much

Answer 1

[Hariprasad Ravishankar]

Hello,

If you have a standlone network, you can save the network to a MAT file and specify the name of the MAT file as the first argument to coder.loadDeepLearningNetwork function as follows.

net = squeezenet; % net could be any custom SeriesNetwork, DAGNetwork or dlnetwork object

save mynet.mat net

function out = mpredict(in)

%#codegen

persistent net;

if isempty(net)

net = coder.loadDeepLearningNetwork('mynet.mat');

end

out = predict(net, in);

end

You can refer to the example link below to deploy your application to NVIDIA Jetson boards:

https://www.mathworks.com/help/supportpkg/nvidia/ug/sobel-edge-detection-using-webcam-on-nvidia-jetson.html

Here is an example video:

https://www.mathworks.com/support/search.html/videos/deploy-yolov2-to-an-nvidia-jetson-1578035533852.html?fq[]=asset_type_name:video&fq[]=category:deeplearning/index&page=1

Answer 2

[yazan doha]

Thank you for your Answer @Hariprasad Ravishankar

now when i use this function in GPU Coder as an Entry-Point Functions, so the next step is to provide a test file that calls the project entry-point functions.

should i convert my Matlab Code (train the Network) in a test file or??

Can you help me to convert my Matlab Code in a test file to define the Input??

[Hariprasad Ravishankar]

Hi Yazan,

You can write an entry point function that passes a single input or a batch of inputs to classfiy function. For example:

function out = mclassify(in)

%#codegen

persistent net;

if isempty(net)

net = coder.loadDeepLearningNetwork('mynet.mat');

end

out = classify(net, in);

You can then generate code and interface with it using MEX using GPU Coder as follows:

cfg = coder.gpuConfig('mex');

cfg.DeepLearningConfig = coder.DeepLearningConfig(TargetLibrary = 'cudnn');

codegen -config cfg -args {testInput} mclassify

This will generate a MEX file named mclassify_mex which you can invoke from your test file as follows:

idx2=randi([5,50]);

aug_idx2=augmentedImageDatastore([224 224], idx2);

o2= readimage(imds_test,idx2);

aug_o2=augmentedImageDatastore([224 224], o2);

result2=mclassify_mex(convnet,aug_o2);

Hari