Resize images in ImageDataStore on GPU

I have trained a neural net (based on AlexNet) and want to classify new images by means of this neural net. First I read the images into an ImageDataStore but for AlexNet the dimension of the images have to be 227 x 227. Can I resize all images in the ImageDataStore on GPU? I currently use below code for the resizing but am wondering if this is the "fastest" way.

newSample.ReadFcn = @(newSample)imresize(imread(newSample),[227 227])

Thanks for your help,

Raf

Answer 1

[Joss Knight]

A ReadFcn is fine for classifying one image at a time. For classifying large batches of images it will slow things down considerably. You should resize all your data in batches before calling classify:

batchData = imresize(imds.read(), [227 227]);

If you have a large dataset to classify, it may be best simply to resize all the images and save them to new files.

For training, in R2017b you can use the augmentedImageSource.

[RP]

Thanks for your suggestion Joss. I'm fairly new to Matlab and noticed that only my first image ends up in batchData. Do I have to create a for loop around the batchData resizing?

[Joss Knight]

Sorry. Set the ReadSize to the size of the batch and concatenate:

imds.ReadSize = 128;
batch = imds.read();
batch = cat(4, batch{:});

Answer 2

[ENVER AKBACAK]

Hi,

Set the number of images to be read in imds by executing:

imds.ReadSize = numpartitions(imds)

Then resize all images by:

imds.ReadFcn = @(loc)imresize(imread(loc),[227,227]);

thats it, all your images in imds is now [227,227]

check some of them randomly:

img = readimage(imds,5); imshow(img); size(img)

img = readimage(imds,222); imshow(img); size(img)

[Anup Gurung]

Hi I am trying to do the same thing, i want to reduce the size of the images i have in my folders. But its not working for me. Can you help me?

imds = imageDatastore('Shapes','IncludeSubfolders',true,'Labelsource','foldernames');

% % Count the label and how many images are there in each of the label % images

labelCount = countEachLabel(imds); % % Resize images in an image datastore

imds.ReadSize = (imds)25; imds.ReadFcn = @(loc)imresize(imread(loc),[32,32,1]);

Answer 3

[ENVER AKBACAK]

Try this

imds = imageDatastore('Shapes','IncludeSubfolders',true,'Labelsource','foldernames');

imds.ReadSize = numpartitions(imds)

imds.ReadFcn = @(loc)imresize(imread(loc),[32,32]);

[Rituraj Kushwaha]

what is loc?

[ENVER AKBACAK]

It is the parameter (input) of anonymous function. You can use any proper name instead. Each images in imds passed the function over this parameter. For example sqr= @y y.^2;

c=sqr(5);

c=25

here, 5 passed the function through y, like loc.

Answer 4

[benhadid ahmed]

try this you make a function to read a picture and resize it like this function img = CrerDataStore( file ) % Import image img = imread(file); % resize image img = imresize(img,[227 227]); % Convert grayscale to color (RGB) if you need img = repmat(img,[1 1 3]); end

after this you make your call function in your program like this imds= imageDatastore(pathToImages,'ReadFcn',@CrerDataStore); img=readimage(imds,1); size(img)