How to solve this error?

Dear matlab team,

How to solve this error?

Invalid training data. For numeric array image input, predictors must be a 4-D array of images.

for this

trainedNet = trainNetwork(gaborResponses, imdsTrain.Labels, layers, options)

please guide us

[Arnav]

Dear sir,

I tried this code, but it resulted in error. As trial, I am using 3 class folders (1,2 and 3) with 10 gray images (total 30 images) of identical size 227x227x1 (of type uint8).

Also, We have created the network. The Gabor responses of size 227x227x4x30 have been computed on each 227x227 gray images (total 30 nos.) and on four orientations (gaborOrientation = [0 45 90 135];) and CNN layera number 25.

Prior to reshape statement, The size(gaborResponses) results in Gabor responses of size 227x227x4x30.

But when run

trainedNet = trainNetwork(gaborResponses, trainingImages.Labels, layers, options);

I still found this error

Invalid training data. For numeric array image input, predictors must be a 4-D array of images.

Pl. guide me to resolve the issue.

Answer 1

[Garmit Pant]

Hello Arnav

Given the function used and the error encountered, I gather that you are trying to train a neural network using the “trainNetwork” function and are encountering an error because of the format of the input training data.

The “trainNetwork” function expects either a datastore, a numeric array or a table as the input for the predictor. Given the error, the ‘gaborResponses’ parameter is a numeric array. For 2-D images, the function expects a ‘h-by-w-by-c-by-N’ numeric array, where h, w, and c are the height, width, and number of channels of the images, respectively, and N is the number of images.

Check the dimensionality of ‘gaborResponses’ using the following code snippet:

size(gaborResponses) 

If the input images are single channel images and ‘gaborResponses’ is 3-D numeric array, reshape it using the following code snippet:

gaborResponses = reshape(gaborResponses, height, width, 1, numberOfImages); 

This shall help you eliminate the error.

Additionally, the “trainNetwork” function is not recommended for use. Instead, the “trainnet” function is recommended. For further understanding, refer to the documentation of the “trainnet” function:

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

I hope you find the above explanation and suggestions useful!

[Arnav]

Dear sir,

Thanks for the prompt response. I tried your code, but it resulted in error. As trial, I am using 3 class folders (1,2 and 3) with 10 gray images (total 30 images) of identical size 227x227x1 (of type uint8).

We have chosen gaborSize = 15;

gaborOrientation = [0 45 90 135];

results in following error

Error using conv2

Invalid data type. First and second arguments must be numeric or logical.

Error in Gabor_Test1 (line 32)

gaborResponse = conv2(image, gaborFilters{j}, 'same');

gaborResponses(:, :, j, i) = gaborResponse;

Pl. guide me to resolve the issue.

[Walter Roberson]

The return type of the image function is a handle to matlab.graphics.primitive.Image

It is recommended to avoid naming a variable image

[Arnav]

Dear matlab team,

How to solve this error?

Number of observations in X and Y disagree. for this 
trainedNet = trainNetwork(gaborResponses, imdsTrain.Labels, layers, options)

Pl. guide me to resolve the issue.

[Walter Roberson]

@Arnav

The most common cause of that message, is if gaborResponses needs to be transposed to match the length of the labels.

[Arnav]

Dear sir,

As trial, I am using 3 class folders (1,2 and 3) with 10 gray images (total 30 images) of identical size 227x227x1 (of type uint8).

We have chosen gaborSize = 15;

gaborOrientation = [0 45 90 135];

calculating value of thisX = 227x227 uint8 and ThisFilter = 4x1 cell

1x1 gabor

1x1 gabor

1x1 gabor

1x1 gabor

For this

numImages = numel(imdsTrain.Files);

gaborResponses = zeros(gaborSize, gaborSize, length(gaborOrientation), numImages);

for i = 1:numImages

image = readimage(imdsTrain, i);

for j = 1:length(gaborOrientation)

gaborFilters{j} = gabor(gaborSize, gaborOrientation(j));

for k = 1 : size(image, 4)

ThisFilter = gaborFilters(:, :, :, k);

thisX = image(:, :, :, k);

gaborResponse = convn(thisX, ThisFilter, 'same');

end

gaborResponses = reshape(gaborResponse,227,227,[]);

gaborResponses(:, :, j, i) = gaborResponse;

end

end

results in following Error using double

Conversion to double from cell is not possible.

Error in convn (line 30)

B = double(B);

Pl. guide me to resolve the issue.