Deploying YoloV3 to Jetson Nano

function outImg = YoloOnJetson(cameraName,resolution)

%   Copyright 2021-2022 The MathWorks, Inc.
hwobj = jetson;
camObj = camera(hwobj,cameraName,resolution);
dispObj = imageDisplay(hwobj);
configFilePath = "E:\JetsonWork\George  Camer Object Detection\ssd_mobilenet_v2_coco.config";

persistent yolov3Obj;

if isempty(yolov3Obj)
    yolov3Obj = coder.loadDeepLearningNetwork('darknet53-coco.mat');
end
% Read the contents of the configuration file as text
fid = fopen(configFilePath, 'r');
configText = fscanf(fid, '%c', inf);
fclose(fid);

% Load the pre-trained YOLOv3 object detector
detector = yolov3ObjectDetector('darknet53-coco');

while ishandle(1)
    
% Capture the image from the camera on hardware.
    img = snapshot(camObj);    
    % Detect objects in the frame
    [bboxes, scores, labels] = yolov3Obj.detect(detector, img);
    
    % Annotate the image with bounding boxes and labels
    img = insertObjectAnnotation(img, 'rectangle', bboxes, cellstr(labels));
    
    % Display the annotated image
    % Display the annotated image
    image(dispObj,img);

end

I am trying to run this code in order to generate an .exe of this code to my Jetson Nano Developer Kit

>> inputArgs = {coder.Constant(camName),coder.Constant(camResolution)};
codegen('-config ',cfg,'-args',inputArgs,'YoloOnJetson','-report');
The 'yolov3ObjectDetector' class does not support code generation.

Error in ==> YoloOnJetson Line: 20 Column: 12
Code generation failed: View Error Report

Error using codegen

I usually get this error, I do have the darknet53-coco.mat file in my file directory, I did generate it somehow playing around.

It's werid because yolov3ObjectDetector shall be compatible with code generation, what can I do?

Answer 1

[Ahmed Tamer]

Fixed, thanks.

Answer 2

[Walter Roberson]

if isempty(yolov3Obj)
    yolov3Obj = coder.loadDeepLearningNetwork('darknet53-coco.mat');
end

That step looks good: you are loading a pre-trained detector that should be of the right class.

detector = yolov3ObjectDetector('darknet53-coco');

But there you are trying to create a new object of that class. Creating a new object of the class is not supported in deployed code: all you can do in deployed code is load an already-trained detector and use the detect() method of that pre-trained detector.

[Ahmed Tamer]

I do believe that

detector = yolov3ObjectDetector('darknet53-coco');

Is a part to identify the detector variable in my code, because if you go down

% Detect objects in the frame
    [bboxes, scores, labels] = yolov3Obj.detect(detector, img);
    

You can see that It is used here, so once I comment the command that you highlighted, I do get this error

codegen('-config ',cfg,'-args',inputArgs,'YoloOnJetson','-report');
Undefined function or variable 'detector'.

Error in ==> YoloOnJetson Line: 27 Column: 49
Code generation failed: View Error Report