How do I control the output class of combine() with multiple image and trasnformedimage datastores?

I am trying to train an image to image regression network. I have a series of input and output images in different folders. I create seperate imagedatastore for each folder than use combine() to create the input for the network. When I use combine on the two datastores as long as one of them is a pure imageDataStore the output is a cell array as trainnetwork() expects, but if both imagedatastores are also transformed than the output becomes an array concating the two images.

Minimal reproducible code:

dataDir = fullfile(toolboxdir('vision'),'visiondata','triangleImages');
imageDir = fullfile(dataDir,'trainingImages');
imds = imageDatastore(imageDir);
targetSize = [224,224];
imdsTransformed = transform(imds,@(x)  imresize(x,targetSize));
combinedDS1 = combine(imdsTransformed, imds);
combinedDS2 = combine(imdsTransformed, imdsTransformed);

response1 = preview(combinedDS1);
response2 = preview(combinedDS2);
whos response1

  Name           Size            Bytes  Class    Attributes

  response1      1x2             51408  cell               

whos response2

  Name           Size             Bytes  Class    Attributes

  response2      8x448             3584  uint8              

Given that the folders are dozens of gigabytes and I only want to transform the output image to do some proof of concept and parameter tuning exiperements, I don't want to save whole new batch of images to disk. Any way I can get combine() to concate two transformedDataStore and output a cell array?

Answer 1

[Matt J]

imdsTransformed = transform(imds,@(x)  {imresize(x,targetSize)}  );

[Christopher Greulich]

Thanks! I can't believe I didn't try this. I tried, before submitting the question, using "cell()" instead of "{}" and was getting "Invalid Transform function defined on datastore.".

Glad it was this simple.