Error forming mini-batch for network input

i want to train a cnn with a folder of images with the size [h w c], the imageInputLayerargument is the same size of the images (h w c), but when training the network matlab says:

Error using trainnet (line 46)

Error forming mini-batch for network input "imageinput". Data interpreted with format "SSCB". To specify a

different format, use the InputDataFormats option.

Caused by:

Dimensions of arrays being concatenated are not consistent.

[Matt J]

We c have no way of knowing what you did. Please attach materials needed to reproduce it.

[Walid]

Thanks for the replay, this is my code for multiclass (3 classes) classification using CNN based on image files stored in three subfolders in the main folder: data.:

clear all; close all; clc
% Define the path to your main data folder
dataFolder = 'C:\data';
% Create an image datastore
imds = imageDatastore(dataFolder, ...
    'IncludeSubfolders', true, ...
    'LabelSource', 'foldernames');
% View the class names and the number of images per class
%imds=aug_imds;
labelCount = countEachLabel(imds);
disp(labelCount);
% Split the datastore into training and validation sets (e.g., 70% for training, 30% for validation)
[imdsTrain, imdsValidation] = splitEachLabel(imds, 0.7, 'randomized');
inputSize = [570 714 3];
augimdsT = augmentedImageDatastore(inputSize,imdsTrain,'ColorPreprocessing','rgb2gray');
augimdsV = augmentedImageDatastore(inputSize,imdsValidation,'ColorPreprocessing','rgb2gray');
numClasses = 3;
layers = [
    imageInputLayer(inputSize)
    convolution2dLayer(5,5)
    batchNormalizationLayer
    reluLayer
    fullyConnectedLayer(numClasses)
    softmaxLayer];
%
options = trainingOptions("sgdm", ...
    MaxEpochs=4, ...
    ValidationData=imdsValidation, ...
    ValidationFrequency=30, ...
    Plots="training-progress", ...
    Metrics="accuracy", ...
    Verbose=false);
% train
net = trainnet(augimdsT,layers,"crossentropy",options);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

I receive the following message:

Error forming validation data mini-batch.

Caused by:

Error forming mini-batch for network input "imageinput". Data interpreted with format "SSCB". To specify

a different format, use the InputDataFormats option.

Dimensions of arrays being concatenated are not consistent.

Answer 1

[Matt J]

This is not enough to reproduce the error. You haven't provided input images. My guess, however, is that there are some files in C:\data that are not 570x714x3.

[Walid]

Hi, the is the download link of the data folder: data

[Walid]

there is some image with different size, so i used augmentedImageDatastore function to resize them to a unique size !

[Walid]

i uploaded the files in matlab drive: data in matlab drive

[Matt J]

You need,

inputSize = [570 714 1];

and

ValidationData=augimdsV,...

[Walid]

Thank you for the response,

performing the code, Matlab return this error:

Error using trainnet (line 46)

Unable to apply function specified by 'MiniBatchFcn' value.

Error in march (line 47)

net = trainnet(imdsTrain,layers,"crossentropy",options);

Caused by:

Error using deep.internal.train.createMiniBatch (line 22)

Error forming mini-batch for network input "imageinput". Data interpreted with format "SSCB". To specify

a different format, use the InputDataFormats option.

Dimensions of arrays being concatenated are not consistent.

my code is attached. in this Link

[Matt J]

your link goes to new code. However, your original code, as presented in your question, should be working now with the changes I propose above. If the code is working for you as well, please Accept-click the answer to indicate that the question is resolved.

As for the new code, it is pretty clear why it doesn't work. You removed the resizing done by the augmentedDatastores so, per our discussion above, it is not going to be possible to concatenate them.

[Walid]

the original code does'nt work ! the same error !

[Matt J]

It works for me. Did you incorporate my fixes?