Testing a neural network after training it

hi i have a 2 class problem and im using vgg16 as a convnet to classify

im using deep network designer app and after training it i dont have an option to test it

i have 1000 pics for each class and about 230 as a test

i need to test with these pics and give the results

after i train it i have results and layer in my workshop which i save them

but i have no way to test it

also when i export the network after training and then run it, it just keeps retraining it

does it test it and i dont get it? or im doing something wrong

Answer 1

[Hari]

Hi Kamyar,

I understand that you have trained a VGG16 convolutional neural network using the Deep Network Designer app for a two-class classification problem, but you're unsure how to test it with your test dataset of 230 images.

I assume you have already saved the trained network and have access to the test dataset, and you're looking for a way to evaluate the network's performance on this dataset without retraining.

• Export the Trained Network: After training in the Deep Network Designer, ensure you export the trained network to the MATLAB workspace. This should provide you with a "trainedNet" variable or similar.
• Prepare the Test Dataset: Load and preprocess your test images to match the input size of VGG16 (224x224x3) and create an "augmentedImageDatastore".

testFolder = 'path_to_test_images'; % Path to your test images
testImages = imageDatastore(testFolder, 'IncludeSubfolders', true, 'LabelSource', 'foldernames');
inputSize = trainedNet.Layers(1).InputSize;
augTestImages = augmentedImageDatastore(inputSize(1:2), testImages);

• Classify Test Images: Use the "classify" function to predict the labels of your test images using the trained network.

predictedLabels = classify(trainedNet, augTestImages);
trueLabels = testImages.Labels;

• Evaluate the Model: Calculate the accuracy and other performance metrics using the predicted and true labels.

accuracy = sum(predictedLabels == trueLabels) / numel(trueLabels);
disp(['Test Accuracy: ', num2str(accuracy)]);

• Visualize Results: Optionally, display test images with predicted and true labels to visually assess the model's performance.

References:

• Refer to the documentation of "classify" for classifying images using a trained network: https://www.mathworks.com/help/deeplearning/ref/classify.html
• Refer to the documentation of "augmentedImageDatastore" for resizing and augmenting images: https://www.mathworks.com/help/deeplearning/ref/augmentedimagedatastore.html
• Refer to the documentation of "imageDatastore" for loading image data: https://in.mathworks.com/help/matlab/ref/matlab.io.datastore.imagedatastore.html

Hope this helps!