The training images are of size 224×224×3 but the input layer expects images of size 224×224×1.

I want to train a D.L network (googlenet) with images with "224×224×1" of size (they are panchromatic). I follow the steps described in the "mathworks" as described bellow but I get this error: " The training images are of size 224×224×3 but the input layer expects images of size 224×224×1." So what I should do in order to train my network ? Many thanks.

Below is the code.

clear all
close all
%faceDatasetPath = fullfile('c:','FaceDataset');
imds = imageDatastore("D:\test", ...
    'IncludeSubfolders',true,'LabelSource','foldernames');
%display some samples from  the dataset
figure;
perm = randperm(213,20);
for i = 1:20
    subplot(4,5,i);
    imshow(imds.Files{perm(i)});
    title(imds.Labels(perm(i)));
end

numTrainFiles = 6;
[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomize');
[no_of_TrainImages ~]=size(imdsTrain.Files);
[no_of_TestImages ~]=size(imdsValidation.Files);
layers = [
    
imageInputLayer([224 224 1])
    
    convolution2dLayer(3,8,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,16,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,64,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    fullyConnectedLayer(7)
    softmaxLayer
    classificationLayer];

options = trainingOptions('sgdm', ...
    'InitialLearnRate',0.01, ...
    'MaxEpochs',10, ...
    'Shuffle','every-epoch', ...
    'ValidationData',imdsValidation, ...
    'ValidationFrequency',10, ...
    'Verbose',false, ...
    'Plots','training-progress');
net = trainNetwork(imdsTrain,layers,options);

YPred = classify(net,imdsValidation);
YValidation = imdsValidation.Labels;

accuracy = sum(YPred == YValidation)/numel(YValidation)

Answer 1

[Matt J]

imageInputLayer([224 224 3])

[Muhammad]

Its not working. I have single class input only which I am using. I am not giving multiclass input.

[Matt J]

Your datastore appears to contain 3-channel images, not 1-channel as you intend. The problem is in the input data format that you are presenting to the network, not the code.

[Muhammad]

Can you tell me how can I change the datastore foe 1-channel images.

[Matt J]

The first question is whether you actually have panchromatic image files in the folders under "D:\test". You think you do, but the Matlab code disagrees. What have you done to verify this?

[Muhammad]

Yes, I have panchromatic image in the folder. I want to predicts the gestational age.

[Matt J]

I suggest you attach one of the images from the folder so that we can study how it reads in.

[Muhammad]

I attached the file. See the attached file.

[Matt J]

As you can see, 3-channels:

A=imread('IM-0061-0010.jpg');

whos A

  Name        Size                Bytes  Class    Attributes

  A         224x224x3            150528  uint8              

[Matt J]

Since all 3 channels of your images are the same, one thing you can do is modify the ReadFcn property of your datastore to read only one of the channels:

imds = imageDatastore("D:\test", ...
    'IncludeSubfolders',true,'LabelSource','foldernames','ReadFcn',@customreader);


function img=customreader(fname)

 img=imread(filename);
 img=img(:,:,1);
 
end

[Muhammad]

Can you please modify my code? I spent lots of time but couldn't do that. Please modify the below code.

clear all
close all
%faceDatasetPath = fullfile('c:','FaceDataset');
imds = imageDatastore("D:\test", ...
    'IncludeSubfolders',true,'LabelSource','foldernames');
%display some samples from  the dataset
figure;
perm = randperm(213,20);
for i = 1:20
    subplot(4,5,i);
    imshow(imds.Files{perm(i)});
    title(imds.Labels(perm(i)));
end
numTrainFiles = 6;
[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomize');
[no_of_TrainImages ~]=size(imdsTrain.Files);
[no_of_TestImages ~]=size(imdsValidation.Files);
layers = [
    
imageInputLayer([224 224 1])

convolution2dLayer(3,8,'Padding','same')
batchNormalizationLayer
reluLayer
maxPooling2dLayer(2,'Stride',2)

convolution2dLayer(3,16,'Padding','same')
batchNormalizationLayer
reluLayer
maxPooling2dLayer(2,'Stride',2)

convolution2dLayer(3,32,'Padding','same')
batchNormalizationLayer
reluLayer
maxPooling2dLayer(2,'Stride',2)

convolution2dLayer(3,64,'Padding','same')
batchNormalizationLayer
reluLayer
maxPooling2dLayer(2,'Stride',2)

fullyConnectedLayer(7)
softmaxLayer
classificationLayer];
options = trainingOptions('sgdm', ...
    'InitialLearnRate',0.01, ...
    'MaxEpochs',10, ...
    'Shuffle','every-epoch', ...
    'ValidationData',imdsValidation, ...
    'ValidationFrequency',10, ...
    'Verbose',false, ...
    'Plots','training-progress');
net = trainNetwork(imdsTrain,layers,options);
YPred = classify(net,imdsValidation);
YValidation = imdsValidation.Labels;
accuracy = sum(YPred == YValidation)/numel(YValidation)

[Matt J]

I have told you what to change in the code. It's the one line that creates the imds. There's nothing else I would change.

[Muhammad]

I am just confused what should I write in the below code

function img=customreader(fname) // fname what should I write fname

img=imread(filename); // filename ??

img=img(:,:,1); // what I write here

[Matt J]

fname should have been filename.

function img=customreader(filename)  
 img=imread(filename);            
 img=img(:,:,1);  
end

[Muhammad]

Can you please tell me what will be the input into the function filename?

function img=customreader(filename) // (filename) what will be the input here

function img=customreader(filename)  
img=imread(filename);  
img=imread(filename); 
 img=img(:,:,1);  
end

[Image Analyst]

The imds object will handle that. You do not need to explicitly call customreader() yourself, so don't worry about it. imds will know what to do.

[Muhammad]

I run this code now but getting this error

Error: File: testfetalMri.m Line: 16 Column: 1

Function definitions in a script must appear at the end of the file.

Move all statements after the "customreader" function definition to before the first local function definition.

clear all
close all
%faceDatasetPath = fullfile('c:','FaceDataset');
%imds = imageDatastore("D:\test", ...
   % 'IncludeSubfolders',true,'LabelSource','foldernames');

imds = imageDatastore("D:\test", ...
    'IncludeSubfolders',true,'LabelSource','foldernames','ReadFcn',@customreader);
function img=customreader(filename)
 img=imread(filename);
 img=img(:,:,1);
 
end

%display some samples from  the dataset
figure;
perm = randperm(213,20);
for i = 1:20
    subplot(4,5,i);
    imshow(imds.Files{perm(i)});
    title(imds.Labels(perm(i)));
end
numTrainFiles = 6;
[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomize');
[no_of_TrainImages ~]=size(imdsTrain.Files);
[no_of_TestImages ~]=size(imdsValidation.Files);
layers = [
    
imageInputLayer([224 224 1])
    
    convolution2dLayer(3,8,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,16,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,64,'Padding','same')
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,'Stride',2)
    
    fullyConnectedLayer(7)
    softmaxLayer
    classificationLayer];
options = trainingOptions('sgdm', ...
    'InitialLearnRate',0.01, ...
    'MaxEpochs',10, ...
    'Shuffle','every-epoch', ...
    'ValidationData',imdsValidation, ...
    'ValidationFrequency',10, ...
    'Verbose',false, ...
    'Plots','training-progress');
net = trainNetwork(imdsTrain,layers,options);
YPred = classify(net,imdsValidation);
YValidation = imdsValidation.Labels;
accuracy = sum(YPred == YValidation)/numel(YValidation)

[Matt J]

Move customreader to the end of the file as the error message says.

Answer 2

[Image Analyst]

You should write a little script to convert all of your images to gray scale and save them in a different folder so you don't overwrite your originals. Or else change your input layer to accept color images, as Matt showed you.

Answer 3

[Walter Roberson]

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

Use an augmentedImageDataStore with size 224 224 1 and 'ColorPreprocessing', 'rgb2gray' . This will automatically resize any image to the right size and will convert to grayscale if needed.