How can I test the models?

Hello guys. I'm new to MATLAB. I have performed fine-tuned of AlexNet and GoogleNet pre-trained networks to fit with my dataset.

How can i split the whole data into training, validation, and testing phase?

How can I test both models with same images for accuracy comparison?

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

%This is the code for AlexNet%

OutputFolder = fullfile('Dataset');

imds = imageDatastore('Dataset','IncludeSubfolders',true,'LabelSource','foldernames');

[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomize');

net = alexnet
analyzeNetwork(net)
inputSize = net.Layers(1).InputSize
layersTransfer = net.Layers(1:end-3);
numClasses = numel(categories(imdsTrain.Labels))
layers = [
    layersTransfer
    fullyConnectedLayer(numClasses,'WeightLearnRateFactor',10,'BiasLearnRateFactor',10)
    softmaxLayer
    classificationLayer];

pixelRange = [-30 30];
imageAugmenter = imageDataAugmenter( ...
    'RandXReflection',true, ...
    'RandXTranslation',pixelRange, ...
    'RandYTranslation',pixelRange);
augimdsTrain = augmentedImageDatastore(inputSize(1:2),imdsTrain, ...
    'DataAugmentation',imageAugmenter);

augimdsValidation = augmentedImageDatastore(inputSize(1:2),imdsValidation);

options = trainingOptions('sgdm', ...
    'MiniBatchSize',16, ...
    'MaxEpochs',5, ...
    'InitialLearnRate',1e-4, ...
    'Shuffle','every-epoch', ...
    'ValidationData',augimdsValidation, ...
    'ValidationFrequency',3, ...
    'Verbose',false, ...
    'Plots','training-progress');

netTransfer = trainNetwork(augimdsTrain,layers,options);
[YPred,scores] = classify(netTransfer,augimdsValidation);
YValidation = imdsValidation.Labels;
accuracy = mean(YPred == YValidation)

figure
for i = 1:4
    subplot(2,2,i)
    I = readimage(imdsValidation,idx(i));
    imshow(I)
    label = YPred(idx(i));
    title(string(label) + ", " + num2str(100*max(probs(idx(i),:)),3) + "%");
end

Thank you.

Answer 1

[Vinayak]

Hi Olalekan

Fine tuning a pre-trained model can improve the model performance, and your code achieves the requested problem effectively.

You have correctly split dataset using `splitEachLabel`.

I would suggest you to load both the models AlexNet and GoogleNet within the same script, or use a .mat file to use the same dataset across both files.

Train the fine tuned models using the same train dataset. And as you have the test set, you can use it for testing both models.

Sample code:

imds = imageDatastore(datasetFolder, 'IncludeSubfolders', true, 'LabelSource', 'foldernames');
trainRatio = 0.7;
[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomize');
alexNet = alexnet;
googleNet = googlenet;
% Training Code similar to your current code.
% Validate both trained models with imdsValidation
[YPredAlexNet, scoresAlexNet] = classify(alexNet, imdsValidation);
accuracyAlexNet = mean(YPredAlexNet == imdsValidation.Labels);
disp(['AlexNet Validation Accuracy: ' num2str(accuracyAlexNet * 100) '%']);
 
[YPredGoogleNet, scoresGoogleNet] = classify(googleNet, imdsValidation);
accuracyGoogleNet = mean(YPredGoogleNet == imdsValidation.Labels);
disp(['GoogleNet Validation Accuracy: ' num2str(accuracyGoogleNet * 100) '%']);

You may refer to the following docs on `spliteachlabel`

• https://www.mathworks.com/help/matlab/ref/matlab.io.datastore.imagedatastore.spliteachlabel.html

Also refer to this article to discover more on Transfer Learning with MATLAB:

• https://www.mathworks.com/discovery/transfer-learning.html?s_tid=srchtitle_support_results_1_transfer%2520learning