How to add a trained deep neural network into a MATLAB function for C++ code generation using MATLAB Coder?

Question

Philip Kanjirathingal Kurian 2019-8-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/476097-how-to-add-a-trained-deep-neural-network-into-a-matlab-function-for-c-code-generation-using-matlab

评论： Philip Kanjirathingal Kurian 2019-8-15

As part of my project, I have to include a trained deep neural network into a MATLAB function in order to classify different types of arrows using a webcam. The code is given below. And when i try to convert the MATLAB code using MATLAB Coder it shows an error of Maximum recursion limit of 500 reached. What is the reason this error occuring and how can i avoid this error.

function [Vx,Vy,Omega] = Landmarkerkennung(Image)
net = coder.loadDeepLearningNetwork('convnetwork1'); %Is this how to include a trained network?
%convnetwork1 is the name of my trained Network which is saved in MATLAB
% Vx, Vy,Omega are the linear and angular velocities which has to be fed to the robot. And input is the image from the camera of a robot.
%Workspace.
%UNTITLED Summary of this function goes here
%   Detailed explanation goes here
% Continous loop for live video from thr Robotino Camera    
    while true
            img = Image;                                % Capturing new Image in 'img'
            picture = imresize(img,[227,227]);          % Resizing the picture (227 227). Pixel required for Alexnet Algorithm
            label = classify(net, picture);             % Classify the picture. Labels of the pictures are stored in 'label'
            
% Calling the respective function according to the label that appears on
% the camera frame
            if label == 'straight'
                    Omega = 0;
                    Vx = 75;
                    Vy = 0;
%                 [Vx,Vy,Omega] = gostraight();
                elseif label == 'left'
                        disp('Left');
                        Omega = 50;
                        Vx = 0;
                        Vy = 0;
                        pause(2);                    % Pause the execution of the Programm for 2 second
%                     [Vx,Vy,Omega] = gosharpleft();
                elseif label == 'right'
                        disp('Right');
                        Omega = -50;
                        Vx = 0;
                        Vy = 0;
                        pause(2);                    % Pause the execution of the Programm for 2 second
%                     [Vx,Vy,Omega] = gosharpright();
                elseif label == 'right'
                        disp('Right');
                        Omega = -50;
                        Vx = 0;
                        Vy = 0;
                        pause(2);                    % Pause the execution of the Programm for 2 second
            end
    end
end

When i convert the MATLAB function i get this error.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Mukund Sankaran 2019-8-14

2
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/476097-how-to-add-a-trained-deep-neural-network-into-a-matlab-function-for-c-code-generation-using-matlab#answer_387583

在 MATLAB Online 中打开

You are seeing that error message because you've named your function 'convnetwork1' and in line #5 of that function, you are calling coder.loadDeepLearningNetwork('convnetwork1' ).

The call to coder.loadDeepLearningNetwork can take either the name of a MAT-file or the name of a function existing in the MATLAB path as the first argument.

When codegen runs, it hits line #5, which invokes the function 'convnetwork1' again, and this continues in a recursive loop until the call stack overflows, which is why you see the error 'Maximum recursion limit of 500 reached'

You have a couple of options to load your trained network as I see it. Hope this helps!

1. Call coder.loadDeepLearningNetwork with the name of a MAT-file that contains your network. Here's an example

% foo.m
function out = foo()
    out = coder.loadDeepLearningNetwork('convnetwork.mat') % Contains your trained network
end
% Now you can codegen foo.m

2. Write a function that returns your trained network and provide the name of that function while calling coder.loadDeepLearningNetwork. Here's an example

% foo.m
function out = foo() 
    out = coder.loadDeepLearningNetwork('convnetwork');
end
% convnetwork.m
function out = convnetwork()
    out = alexnet; % Return your network
end
% Now you can codegen foo.m 

3 个评论
显示 1更早的评论隐藏 1更早的评论

Philip Kanjirathingal Kurian 2019-8-14

在 MATLAB Online 中打开

Thanks for the fast reply. I tried your solution.

function [Vx,Vy,Omega] = Landmarkerkennung(Image)
    function out = foo()%#codegen
    out = coder.loadDeepLearningNetwork('looprun1.mat');
    end
%UNTITLED Summary of this function goes here
%   Detailed explanation goes here
% Continous loop for live video from thr Robotino Camera    
    while true
            img = Image;                                % Capturing new Image in 'img'
            picture = imresize(img,[227,227]);          % Resizing the picture (227 227). Pixel required for Alexnet Algorithm
            label=predict(foo,picture);
%             label = classify(net, picture);             % Classify the picture. Labels of the pictures are stored in 'label'
            
% Calling the respective function according to the label that appears on
% the camera frame
            if label == 'straight'
                    Omega = 0;
                    Vx = 75;
                    Vy = 0;
                elseif label == 'left'
                        disp('Left');
                        Omega = 50;
                        Vx = 0;
                        Vy = 0;
                        pause(2);                    % Pause the execution of the Programm for 2 second
                elseif label == 'right'
                        disp('Right');
                        Omega = -50;
                        Vx = 0;
                        Vy = 0;
                        pause(2);                    % Pause the execution of the Programm for 2 second
            end
    end
end

But there are some other errors poping up when i generate the code.

When i checked, to classify the input image using the trained network, there are two options.

Using the function classify() and
Using the function predict()

I tried the classify() function first, but the function was not recognised in MATLAB Coder. So now i used the predict() function in order to classify the input image. But then i get the first error in the image. What should i change in my code in order to avoid the error?

Mukund Sankaran 2019-8-14

The error says that the input passed to the "predict" function must be fixed size for codegen to work.

In your example, you pass in a variable that contains the image to the Landmarkerkennung function, which is later resized and passed to the predict function.

In the "Define Input Types" step, you must make sure that the input named "Image" is fixed size i.e. no dimension can be variable size ('up to' or 'Inf'). It must be fixed size: eg: uint8(227 x 227 x 3)

Philip Kanjirathingal Kurian 2019-8-15

Thanks alot Mukund Sankaran for the help.

My code generation did work.

请先登录，再进行评论。

How to add a trained deep neural network into a MATLAB function for C++ code generation using MATLAB Coder?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

采纳的回答

3 个评论
显示 1更早的评论隐藏 1更早的评论

更多回答（0 个）

另请参阅

类别

标签

产品

版本

Community Treasure Hunt

How to add a trained deep neural network into a MATLAB function for C++ code generation using MATLAB Coder?

0 个评论 显示 -2更早的评论隐藏 -2更早的评论

采纳的回答

3 个评论 显示 1更早的评论隐藏 1更早的评论

更多回答（0 个）

另请参阅

类别

标签

产品

版本

Community Treasure Hunt

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

3 个评论
显示 1更早的评论隐藏 1更早的评论