How to rotate and rescale a portion of a class images in an image datastore network ?

I have a network of 5 classes and looking to rotate and rescale a portion of images of specifc class , then train, test and validate the network

Answer 1

[Mohammad Sami]

You can use the augmentedImageDataStore. It will randomly apply the specified agumentation to the images.

https://www.mathworks.com/help/deeplearning/ref/augmentedimagedatastore.html

% From Example
imageAugmenter = imageDataAugmenter( ...
    'RandRotation',[-20,20], ...
    'RandXTranslation',[-3 3], ...
    'RandYTranslation',[-3 3])
augimds = augmentedImageDatastore(imageSize,XTrain,YTrain,'DataAugmentation',imageAugmenter);
% or 
augimds = augmentedImageDatastore(imageSize,imds,'DataAugmentation',imageAugmenter);

[Mohammad Sami]

You can combine multipe data stores using the combine function.

So create and augmenteddatastore with the images you want to rotate.

Create another datastore with rest of the images.

Then use the combine function to merge the two together.

Reference :

https://www.mathworks.com/help/matlab/ref/matlab.io.datastore.combine.html