problems of using deploy fuction to create yolov2 c++ code for raspberry pi

hi,

here i use the matlab code deploy function to create the yolov2 c++ code in raspberry pi, but some problems occured after the code compliation.

when i use the armcompute version v19.02 with debug=1 to complie and build the code, there is an error when i deploy the code in matlab:

'terminate called after throwing an instance of 'std::runtime_error'terminate called after throwing an instance of 'std::runtime_error'
what(): in validate_arguments src/core/NEON/kernels/NEDepthwiseWeightsReshapeKernel.cpp:92: biases->dimension(0) != input->dimension(2)

if i set the debug=0, this erro will disappear and the code can be successfully generated. but when i use the generated code to run on the raspberry pi, the elf progamme can be started but there is nothing response from the device. i also try deploy sample in this link in my environment, it can successfully work.

i setup my environment follow the process below:

1. install arm-software computelibrary
   git config --global http.sslverify false
   git clone https://github.com/Arm-software/ComputeLibrary.git
   cd ComputeLibrary
   git tag -l
   git checkout v19.02
   
  -(to build the arm_compute library and make the libarm_compute.so, libarm_compute_core.so in 'lib folder')
   sudo apt-get install scons
   scons debug=0 neon=1 opencl=0 arch=armv7a
   mkdir lib
   mv /build/*.so /lib/

2.install userland
  download userland on 'https://github.com/raspberrypi/userland' and unzip it to opt folder
  (or git clone https://github.com/raspberrypi/userland)
  cd /opt/userland;
  sudo ./buildme
  sudo ldconfig  
  
3.install nanomsg-1.0.0
   wget https://github.com/nanomsg/nanomsg/archive/1.0.0.zip
   sudo unzip 1.0.0.zip -d /tmp
   cd /tmp/nanomsg-1.0.0 ; 
   mkdir build ;
   cd build
   sudo cmake ..  ;
   sudo cmake --build .
   sudo cmake --build . --target install
   sudo ldconfig

4. sudo apt-get install libasound2-dev
   sudo apt-get install libsdl1.2-dev

5. edit arm computelibrary in /.bashrc
   # ~/.bashrc: executed by bash(1) for non-login shells.
   # see /usr/share/doc/bash/examples/startup-files (in the package bash-doc)
   # for examples
   # If not running interactively
   case $- in
      *i*) ;;
        *) export ARM_COMPUTELIB=~/ComputeLibrary
           export PATH=${PATH}:~/ComputeLibrary
           return;;
   esac

the fuction code and deploy.m i use is below here:

   function  yolov2_detect_arm_vid()

persistent yolov2obj;

%for raspberry pi opencv library build
opencv_linkflags = '`pkg-config --cflags --libs opencv`';
coder.updateBuildInfo('addLinkFlags',opencv_linkflags);

if isempty(yolov2obj)
    yolov2obj = coder.loadDeepLearningNetwork('yolov2.mat','yolov2');
end

host = raspi();
cam = webcam(host,1);
imgsize = [416,416,3];
fps = 0;

for k = 1:1800
    
    img = snapshot(cam);
    img = imresize(img,imgsize(1:2));
    tic;
    [bboxes,~,labels] = detect(yolov2obj,img);
    elapsedtime = toc;
    fps = 0.9*fps + 0.1*(1/elapsedtime);
    fprintf('fps: %f\n',fps);
    
    if isempty(bboxes) || isempty(labels)
        img = insertText(img, [5,5],fps);
        %imshow(img);
        displayImage(host,img);
    else
        img = insertObjectAnnotation(img,'rectangle',bboxes,labels);
        img = insertText(img,[5,5],fps);
        %imshow(img);
        displayImage(host,img);
    end
end 

   end

deploy.m

board = targetHardware('Raspberry Pi');
board.DeviceAddress = 'xxx';
board.Username = 'xx';
board.Password = 'xx';
board.BuildDir = '/home/pi/matlab_arm_codegen/';

board.CoderConfig.TargetLang = 'C++';

dlcfg = coder.DeepLearningConfig('arm-compute');
dlcfg.ArmArchitecture = 'armv7';
dlcfg.ArmComputeVersion = '19.02'; 
board.CoderConfig.DeepLearningConfig = dlcfg; 

deploy(board,'yolov2_detect_arm_vid');

is anyone can help with this problem?

best regards

[Reeno Joseph]

Hi liheng mo,

Can you please clarify the below items.

1. OpenCV based APIs are not used in “yolov2_detect_arm_vid”. Do you really need those openCV linkage flags?
2. There is an example that does classification with resnet-50. (Link). Is it working fine?
3. Raspberry Pi support package comes with a customized Raspbian image (ARM compute 19.05 included). You can try that if you don’t have any version-specific dependency. Just follow the hardware setup screens.

I’ll check the issue related to debug flag and will get back to you.

[Meruzhan Hovhannisyan]

I also tried this technique, nothing good came out. When compiling, it returns the error "Calling the code generation implementation function 'code.internal.const' from the user function 'insertObjectAnnotation/validateAndParseOptInputs_cg' is not supported. Please contact technical support for help."