How can I export Roboflow annotation to work in Matlab

Hello! I'm using annotations to create bounding boxes on my images to train a model. To export the dataset created on Roboflow, we can export in different ways, like COCO segmentation Json files, or TXT YOLO oriented bounding boxes, or CSV tensorflow/ CSV keras. The downloaded files comes with the images and labels created in train, test and validation. So I have the images, the labels with coordinates from each image that I'm creating the Annotation. But I don't know how am I work with those files on Matlab. Anyone can help me with this problem?

[Animesh]

Hey @Luiz Augusto Meleiro

The following MATLAB answer might help:

https://www.mathworks.com/matlabcentral/answers/1833448

[Luiz Augusto Meleiro]

Thank for your attention Rahul!

Answer 1

[Rahul]

Hi @Luiz Augusto Meleiro,

I understand that you are trying to use the annotated data obtained from "Roboflow" in MATLAB.

One of the approaches you can utilize for your case is that you export the data in XML or JSON or CSV formats which can be converted easily to be used in MATLAB using functions like "xml2struct", "jsondecode", "csvread" as mentioned in this answer: https://www.mathworks.com/matlabcentral/answers/1833448-how-to-export-roboflow-annotation-to-mat-file

Another method that you can use is to use "insertObjectAnnotation" function where you can pass the images with their corresponding labels and annotation coordinates to obtain annotated images in MATLAB.

I = imread('test1.png');
position = [23,56,134,88];
RGB = insertObjectAnnotation(I,"rectangle",position,'l1',TextBoxOpacity=0.9,FontSize=18);
imshow(RGB)

% This is just an example 

You can refer to the following documentations for the detailed instructions on how to use these functions:

"xml2struct": https://www.mathworks.com/matlabcentral/fileexchange/28518-xml2struct

"xmlread": https://www.mathworks.com/help/matlab/ref/xmlread.html#mw_9a3a5305-86fd-41b1-a9b0-8113f0653321

"jsondecode": https://www.mathworks.com/help/releases/R2024a/matlab/ref/jsondecode.html?searchHighlight=jsondecode&s_tid=doc_srchtitle

"csvread": https://www.mathworks.com/help/matlab/ref/csvread.html

"insertObjectAnnotation": https://www.mathworks.com/help/vision/ref/insertobjectannotation.html#d126e276766

Note: You can utilize the MATLAB Image Labeller app if required to annotate image in MATLAB

https://www.mathworks.com/help/vision/ug/get-started-with-the-image-labeler.html

Hope this helps! Thanks.

[Luiz Augusto Meleiro]

Hello, thanks for the information, but the annotations that we are creating is not rectangle, is more complicated shape. I attached the photos and the coordinates that are created is in the another attached, so I'm facing this challenge, how to export that data to matlab and create the same shape there. Hope you can help me again, thank you!

Answer 2

[Kenneth]

For those who want rectangular ROIs in MATLAB, you can convert it from YOLO using the following code:

function gTruth = annotateFromYOLO(fileName, imgName)
    % fileName - filename of annotation file
    % imgName - name of the image reference

    % Create the data source from the image name
    dataSource = groundTruthDataSource(imgName);

    % Define the labels
    ldc = labelDefinitionCreator();
    addLabel(ldc, 'GradeJK', 'Rectangle', 'Group', 'AbacaFibers', 'Description', 'Grade JK abaca fibers');
    addLabel(ldc, 'GradeSI', 'Rectangle', 'Group', 'AbacaFibers', 'Description', 'Grade S-I abaca fibers');
    addLabel(ldc, 'GradeSS2', 'Rectangle', 'Group', 'AbacaFibers', 'Description', 'Grade S-S2 abaca fibers');
    labelDefs = create(ldc);

    % Read YOLO data from the file
    fileID = fopen(fileName, 'r');
    formatSpec = '%f';
    yoloData = fscanf(fileID, formatSpec);
    fclose(fileID); % Close the file after reading

    % Extract bounding box data
    center_x = yoloData(2);
    center_y = yoloData(3);
    width = yoloData(4);
    height = yoloData(5);

    % Convert to [x, y, w, h] format
    x = center_x - width / 2;
    y = center_y - height / 2;
    w = width;
    h = height;

    % Create the transposed bounding box data
    transposedBBoxData = [x, y, w, h];

    % Read the image to get dimensions
    img = imread(imgName);
    [imgHeight, imgWidth, ~] = size(img);

    % Scale bounding box data based on image dimensions
    % Assuming YOLO data is normalized between 0 and 1
    % Convert normalized coordinates to pixel coordinates
    bBoxDataScaled = transposedBBoxData .* [imgWidth, imgHeight, imgWidth, imgHeight];

    % Create table with scaled bounding box data
    labelNames = {labelDefs.Name{yoloData(1) + 1}};
    labelData = table(bBoxDataScaled, 'VariableNames', labelNames);

    % Create the ground truth object
    gTruth = groundTruth(dataSource, labelDefs, labelData);

    % Construct the .mat filename
    [~, baseFileName, ~] = fileparts(fileName);
    matFileName = strcat(baseFileName, '.mat');

    % Save the ground truth object to a .mat file
    save(matFileName, 'gTruth');

    % Display the gTruth object to verify it was created successfully
    disp(gTruth);
end

Hope this helps!