Evaluating semantic segmentation results error

I used the documentation to evaluate my results but I get this error, can someone tell me what's wrong please , my input is a .mat file

Evaluating semantic segmentation results

----------------------------------------

* Selected metrics: global accuracy, class accuracy, IoU, weighted IoU, BF score.

* Processed 0 images.Error using semanticSegmentationMetrics>iAssertCategoricalsHaveSameCategories

The categorical data returned by dsResults and dsTruth must have the same

categories.

my code is below

dataSetDir = fullfile('LungTS','preprocessedDataset');

testImagesDir = fullfile(dataSetDir,'imagesTest');

imdReader = @(x) matRead(x);

imds = imageDatastore(testImagesDir, ...

'FileExtensions','.mat','ReadFcn',imdReader);

%imds = imageDatastore(testImagesDir);

classNames = ["nodule" "background"];

labelIDs = [255 0];

labelReader = @(x) matRead(x);

testLabelsDir = fullfile(dataSetDir,'labelsTest');

pxdsTruth = pixelLabelDatastore(testLabelsDir,classNames,labelIDs, ...

'FileExtensions','.mat','ReadFcn',labelReader);

net = load('trained3DUNet-16-Sep-2022-14-34-13-Epoch-250.mat');

net = net.net;

pxdsResults = semanticseg(imds,net,"WriteLocation",tempdir);

metrics = evaluateSemanticSegmentation(pxdsResults,pxdsTruth);

metrics.ClassMetrics

metrics.ConfusionMatrix

cm = confusionchart(metrics.ConfusionMatrix.Variables, ...

classNames, Normalization ='row-normalized');

cm.Title = 'Normalized Confusion Matrix (%)';

imageIoU = metrics.ImageMetrics.MeanIoU;

figure

histogram(imageIoU)

title('Image Mean IoU')

[minIoU, worstImageIndex] = min(imageIoU);

minIoU = minIoU(1);

worstImageIndex = worstImageIndex(1);

worstTestImage = readimage(imds,worstImageIndex);

worstTrueLabels = readimage(pxdsTruth,worstImageIndex);

worstPredictedLabels = readimage(pxdsResults,worstImageIndex);

worstTrueLabelImage = im2uint8(worstTrueLabels == classNames(1));

worstPredictedLabelImage = im2uint8(worstPredictedLabels == classNames(1));

worstMontage = cat(4,worstTestImage,worstTrueLabelImage,worstPredictedLabelImage);

worstMontage = imresize(worstMontage,4,"nearest");

figure

montage(worstMontage,'Size',[1 3])

title(['Test Image vs. Truth vs. Prediction. IoU = ' num2str(minIoU)])

[minIoU, worstImageIndex] = min(imageIoU);

minIoU = minIoU(1);

worstImageIndex = worstImageIndex(1);

worstTestImage = readimage(imds,worstImageIndex);

worstTrueLabels = readimage(pxdsTruth,worstImageIndex);

worstPredictedLabels = readimage(pxdsResults,worstImageIndex);

worstTrueLabelImage = im2uint8(worstTrueLabels == classNames(1));

worstPredictedLabelImage = im2uint8(worstPredictedLabels == classNames(1));

worstMontage = cat(4,worstTestImage,worstTrueLabelImage,worstPredictedLabelImage);

worstMontage = imresize(worstMontage,4,"nearest");

figure

montage(worstMontage,'Size',[1 3])

title(['Test Image vs. Truth vs. Prediction. IoU = ' num2str(minIoU)])

[maxIoU, bestImageIndex] = max(imageIoU);

maxIoU = maxIoU(1);

bestImageIndex = bestImageIndex(1);

bestTestImage = readimage(imds,bestImageIndex);

bestTrueLabels = readimage(pxdsTruth,bestImageIndex);

bestPredictedLabels = readimage(pxdsResults,bestImageIndex);

bestTrueLabelImage = im2uint8(bestTrueLabels == classNames(1));

bestPredictedLabelImage = im2uint8(bestPredictedLabels == classNames(1));

bestMontage = cat(4,bestTestImage,bestTrueLabelImage,bestPredictedLabelImage);

bestMontage = imresize(bestMontage,4,"nearest");

figure

montage(bestMontage,'Size',[1 3])

title(['Test Image vs. Truth vs. Prediction. IoU = ' num2str(maxIoU)])

evaluationMetrics = ["accuracy" "iou"];

metrics = evaluateSemanticSegmentation(pxdsResults,pxdsTruth,"Metrics",evaluationMetrics);

metrics.ClassMetrics

Answer 1

[Shantanu Dixit]

Hi Amira,

The error message you encountered suggests that the ground truth and predicted labels in the "pixelLabelDatastore" ('pxdsTruth' and 'pxdsResults') do not have matching categories. Specifically, the categories of the predicted data ('pxdsResults') and the true labels ('pxdsTruth') must be identical for the "evaluateSemanticSegmentation" function to work correctly.

You can verify the categories of the ground truth ('pxdsTruth') and predicted labels ('pxdsResults') through

I = read(pxdsTruth);
R = read(pxdsResults);
CR = categories(I{1});
cI = categories(C{1});

Additionally check that '.mat' files are being correctly read. You can verify that the custom 'matRead' function correctly reads the data and converts it into categorical labels with the expected classes.

function data = matRead(filePath)
    matData = load(filePath); 
    fieldName = fieldnames(matData); 
    data = matData.(fieldName{1});
    % Ensure the data is categorical and matches the class names
    data = categorical(data, categoryValues, categoryNames); %% based on specific values and category names
end

You can also refer to the following MathWorks documentation which may be useful:

'pixelLabelDatastore': https://www.mathworks.com/help/vision/ref/pixellabeldatastore.html

'categorical': https://www.mathworks.com/help/matlab/ref/categorical.html

'evaluateSemanticSegmentation': https://www.mathworks.com/help/vision/ref/evaluatesemanticsegmentation.html#d126e288402

Hope this helps!