Resizing images to 224*224 for training VGG16 Model

Hi I'm training a VGG16 Model in the latest version of MATLAB R2023-a and I keep getting the same resizing error. I'm wondering if there is a new way to resize the images in a datastore in MATLAB or if the error is really on my improper execution of resizing in my code.

I have a folder of dataset with two sub folders which are my classes. This is the first part of my code with resizing:

% Load the dataset
imds = imageDatastore('DataSetFolder', 'IncludeSubfolders', true, 'LabelSource', 'foldernames');

% Define the desired input size
inputSize = [224 224];

% Resize images to 224x224
inputSize = [224 224];
augmentedIMDS = augmentedImageDatastore(inputSize, imds);

% Use the transform function to resize images
augmentedIMDS = transform(imds, @(x)imresize(x, inputSize));

% Split the dataset into training, validation, and test sets
[trainingData, validationData, testingData] = splitEachLabel(augmentedIMDS, 0.8, 01, 01);

% Load the pretrained VGG16 model
net = vgg16;

I keep getting error for this part of the code:

% Split the dataset into training, validation, and test sets
[trainingData, validationData, testingData] = splitEachLabel(augmentedIMDS, 0.8, 01, 01);

when i change the directory to imds the error is about resizing "Incorrect input size. the input size must have a size of [224 224 3]" and when i change it to augmentedIMDS the error is "Incorrect number or types of inputs or outputs for function 'splitEachLabel'."

Answer 1

[David Ho]

Hi Lalaine,

The "splitEachLabel" function only accepts ImageDatastores as an input. To make training, validation and test data of size [224 224], first split the imageDatastore, then transform it:

[imdsTrain, imdsValidation, imdsTest] = splitEachLabel(imds, 0.8, 0.1, 0.1);
inputSize = [224 224];
trainingData = augmentedImageDatastore(inputSize, imdsTrain);
validationData = augmentedImageDatastore(inputSize, imdsValidation);
testingData = augmentedImageDatastore(inputSize, imdsTest);