Dataset (Matrix) as an input in CNN

Hi hello, i am trying to create a CNN model but my inputs are not images, it's a matrix, i would like to know how is it possible to do that, i have searched about it but nothing, and when i am trying to train my network i amm getting this error

Error using trainNetwork

Number of observations in X and Y disagree.

Error in Analyse_Data__CNN (line 81)

net = trainNetwork(D_App,L_App,layers,options);

Thank you so much for helping me i will appreciate it a lot

[Dehia]

Salam Mehdi, peux tu s'il te plaît m'aider à faire ce que tu as fait. Je te remercie

Answer 1

[Image Analyst]

An image is a matrix so that should not be a problem. It looks like you don't have the same number of labeled ground truth matrices as you have training matrices.

[MEHDI HAMIDI]

can you tell me more sir how can i fix it

[Image Analyst]

How many matrices do you have for training? How many labeled ground truth items do you have? They should be the same number.

[MEHDI HAMIDI]

So basically i have 1 matrix (6100,1000) that im spliting into 2 one for train and one for test, i didn't understand what do you mean exactly by labeled ground truth items, but i do have another labeled matrix that contains (6100,2), thank you sir

Data = [XTrain;XTest]; %this one contains (6100,1000)
Label = [YTrain;YTest]; % this one contains (6100,2)
Ind = randperm(size(Data,1));
D_App = Data(Ind(1:4000),:); % this one for train
L_App = Label(Ind(1:4000),:);
D_Tst = Data(Ind(4001:end),:);% this one for test 
L_Tst = Label(Ind(4001:end),:);

[MEHDI HAMIDI]

I also would like to know, for my layers should i use

    imageInputLayer([4000 1000])

[Image Analyst]

I don't know what kind of deep learning training you're doing but for a standard type you need lots of images that are the same size. The size depends on the network architecture. If your image is a montage of lots of smaller images stitched together, then you get get them as separate images by cropping them out. You can't train a network on only one image!

[MEHDI HAMIDI]

you're right i totally forgot about this, it's just interpret it as a single image, and as you said you can't train a network on 1 image only, maybe if i make change my matrix to vectors it will make it

Answer 2

[manel marweni]

Dear Sir\madam,

Can you tell me please how can I see the output of each CNN layer with matlab, ( for example: I would like to extract the output of the convolution layer (to see the results) and make it as input to the next layer and so on).

With best regards.