errors in using transform for imagedatastore in deep learning

A part of source code:

dsnew_trn = transform(ds_trn, @(x) GFR2filter_crop(x, kerfcn, s1, s2, inputSize));

dsnew_tst = transform(ds_tst, @(x) GFR2filter_crop(x, kerfcn, s1, s2, inputSize));

trainedNet = trainNetwork(dsnew_trn,lgraph,opts);

ds_trm: original imagedatastore for training; read images by using a custom function, output: double type

ds_tst: original imagedatasotre for testing; the same to the above.

batch size: 64

function dataout = GFR2filter_crop(dataIn, kerfcn, s1, s2, inputSize)

NoImg = size(dataIn);

dataout = cell(NoImg,1);

for xxx=1:NoImg

Img = dataIn{xxx};

%}

I_spatial = Img(:,:,1);

R1 = imfilter(I_spatial, kerfcn);

trasImg(:,:,1)=R1(s1:s1+inputSize(1)-1,s2:s2+inputSize(2)-1);

I_spatial = Img(:,:,2);

R2 = imfilter(I_spatial, kerfcn);

trasImg(:,:,2)=R2(s1:s1+inputSize(1)-1,s2:s2+inputSize(2)-1);

I_spatial = Img(:,:,3);

R3 = imfilter(I_spatial, kerfcn);

trasImg(:,:,3)=R3(s1:s1+inputSize(1)-1,s2:s2+inputSize(2)-1);

dataout{xxx}=trasImg;

end

-----------------------------------

Error message:

Error using trainNetwork (line 170)

Invalid transform function defined on datastore.

Error in alaska2_deeplearningxx (line 74)

trainedNet = trainNetwork(dsnew_trn,lgraph,opts);

Caused by:

Error using nnet.internal.cnn.util.NetworkDataValidator/assertDatastoreHasResponses (line 140)

Invalid transform function defined on datastore.

Error using cell

Size inputs must be scalar.

[Frank Liu]

Who can help to solve the problem? What is the data type of the "dataIn" in the function dataout = GFR2filter_crop(dataIn, kerfcn, s1, s2, inputSize) since it inputs the batch size of images for training?

[Bhargavi Maganuru]

Makesure the input to the transform is a datastore . Could you mention the version of MATLAB you are using?

[Frank Liu]

Hi, I already solve the proveblem. Regarding the transform function in matlab, it has the fixed input signature as the follows,

dsnew = transform(ds,@fcn)
dsnew = transform(ds,@fcn,'IncludeInfo',IncludeInfo)

therefore, the original functions

dsnew_trn = transform(ds_trn, @(x) GFR2filter_crop(x, kerfcn, s1, s2, inputSize));
dsnew_tst = transform(ds_tst, @(x) GFR2filter_crop(x, kerfcn, s1, s2, inputSize));

has been modified into the followings. Meanwhile, kerfcn, s1, s2 and inputSize are set as global variables.

dsnew_trn = transform(ds_trn, @GFR2filter_crop,'IncludeInfo',IncludeInfo);
dsnew_tst = transform(ds_tst, @GFR2filter_crop,'IncludeInfo',IncludeInfo);