Matrix dimension conversion for image input layer

Hi

I have been working on a machine learning task where I have stuck at a point where I need to manage size of matrices for next computations. I am quite unsure how to do it or is it possible to do it..

Currently, I have a matrix, A of size 1000 x 768. I am going to fed this data as an input to the image input layer to train a network. I found that image input layert accepts input in the dimension of [x y z].

I would like to convert the data of A to [1 21 768] size. Is it possible to do such reshaping?

I have tried one of the way like this:

[a,b]=size(A)
newA=reshape(A.', 1,1,a,b)

But it turns out that the data was converted to 1 x 1 x 768 format which I am not looking for.

What should be the ideal solution of this?

thanks,

Answer 1

[KALYAN ACHARJYA]

Sir, you can convert the size of the image from 2D to 3D using the following way (The image isualization is not affected)

ori_image=imread('file_name'); %Original Gray Image 
image_modied=cat(3,ori_image,ori_image,ori_image); % Image with rxcxslices

"Currently, I have a matrix, A of size 1000 x 768. I am going to fed this data as an input to the image input layer to train a network. I found that image input layert accepts input in the dimension of [x y z]."

%A in [x,y] format
modified_data=cat(3,A,A,A); % [x,y,z] format

[Basu Sau]

Hi Kalyan

Thanks for your help. I have tried your suggestion and it reutrns 1000 x 768 x 3 size of matrix. But I have tried to manage the matrix A to 1 x 21 x 768 size. Is it possible or my idea needs to be reshaped?

[KALYAN ACHARJYA]

If you want to change 1000 x 768 to 1 x 21 x 768 using reshape, here number of the elements must be same? Are the input features remain invariaent after the operation?

Have you checked on pen and paper?