Custom Weighted Classification Layer: Chnage in input value size

I am trying to change the input vlaue size from [1 1 x] to [50 50 x] (here x =1 or 2 or 3 or so on) in weightedClassificationLayer, but its giving error. I need to know where I can do changes so taht this function accept the different input values.

classdef weightedClassificationLayer < nnet.layer.ClassificationLayer

properties

% Row vector of weights corresponding to the classes in the

% training data.

ClassWeights

end

methods

function layer = weightedClassificationLayer(classWeights, name)

% layer = weightedClassificationLayer(classWeights) creates a

% weighted cross entropy loss layer. classWeights is a row

% vector of weights corresponding to the classes in the order

% that they appear in the training data.

% layer = weightedClassificationLayer(classWeights, name)

% additionally specifies the layer name

% Set class weights.

layer.ClassWeights = classWeights;

% Set layer name.

if nargin == 2

layer.Name = name;

end

% Set layer description

layer.Description = 'Weighted cross entropy';

end

function loss = forwardLoss(layer, Y, T)

% loss = forwardLoss(layer, Y, T) returns the weighted cross

% entropy loss between the predictions Y and the training

% targets T.

N = size(Y,4);

Y = squeeze(Y);

T = squeeze(T);

W = layer.ClassWeights;

loss = -sum(W*(T.*log(Y)))/N;

end

function dLdY = backwardLoss(layer, Y, T)

% dLdX = backwardLoss(layer, Y, T) returns the derivatives of

% the weighted cross entropy loss with respect to the

% predictions Y.

[H,Wi,K,N] = size(Y);

Y = squeeze(Y);

T = squeeze(T);

W = layer.ClassWeights;

dLdY = -(W'.*T./Y)/N;

dLdY = reshape(dLdY,[H Wi K N]);

end

end

end

to check the validity of this layer

classWeights = [0 1]

Size=size(classWeights)

layer = weightedClassificationLayer(classWeights);

numClasses = numel(classWeights)

validInputSize = [1 1 numClasses]

checkLayer(layer,validInputSize, 'ObservationDimension',4);

this works on Valid input size =[1 1 numClasses] but I am trying to change it to [x x numClasses] (x any number greater than 1)

Answer 1

[Divya Gaddipati]

As I understand, you want to change the validInputSize to “[x x numClasses]”, which implies a single prediction will be of size [x x 1]. Hence, you also need to reshape your classWeights (which is W in your code) to [x x numClasses].

Modified your code below:

classdef weightedClassificationLayer < nnet.layer.ClassificationLayer
    properties
        % Row vector of weights corresponding to the classes in the
        % training data.
        ClassWeights
    end
    methods
        function layer = weightedClassificationLayer(classWeights, name)
            % layer = weightedClassificationLayer(classWeights) creates a
            % weighted cross entropy loss layer. classWeights is a row
            % vector of weights corresponding to the classes in the order
            % that they appear in the training data.
            % layer = weightedClassificationLayer(classWeights, name)
            % additionally specifies the layer name
            % Set class weights.
            
            layer.ClassWeights = classWeights;
            % Set layer name.
            if nargin == 2
                layer.Name = name;
            end
            % Set layer description
            layer.Description = 'Weighted cross entropy';
        end
        
        function loss = forwardLoss(layer, Y, T)
            % loss = forwardLoss(layer, Y, T) returns the weighted cross
            % entropy loss between the predictions Y and the training
            % targets T.
            
            N = size(Y,4);
            Y = squeeze(Y);
            T = squeeze(T);
            W = layer.ClassWeights;
            
            %% Modified %%
            n = length(T(:))/2;
            W = repelem(W, 1, [n n]);
            W = reshape(W, size(T));
            
            prod = W.*(T.*log(Y));
            loss = -sum(prod(:))/N;
            %% Modified %%
        end
        function dLdY = backwardLoss(layer, Y, T)
            % dLdX = backwardLoss(layer, Y, T) returns the derivatives of
            % the weighted cross entropy loss with respect to the
            % predictions Y.
            [H,Wi,K,N] = size(Y);
            Y = squeeze(Y);
            T = squeeze(T);
            W = layer.ClassWeights;
            
            %% Modified %%
            n = length(T(:))/2;
            W = repelem(W, 1, [n n]);
            W = reshape(W, size(T));
            %% Modified %%
            
            dLdY = -(W.*T./Y)/N;
            dLdY = reshape(dLdY,[H Wi K N]);
           
        end
    end
end

[Raza Ali]

Thank you for the answer. I have tried with the above changes but now got another error..

Error using 'backwardLoss' in Layer weightedClassificationLayer. The function threw an error and could not be executed.

Caused by:

Error using repelem

In repelem(A, N1, N2, ..., Nn), Nj must be a scalar or a vector of length size(A, j).

[Divya Gaddipati]

Hi,

Can you send the validInputSize and classWeights you used to check the validity of the layer?

[Raza Ali]

My input is image; so input size will be 256 by 256 by 3 or 512 by 512 by 3 in size.

I have two outputs so; class weights will be any two numbers between 0 and 1.

[Divya Gaddipati]

Generally, such large sizes (like 256, 512) are not recommended to use for checkLayer. To speed up the tests, specify a smaller valid input size.

You can find more information here: https://www.mathworks.com/help/deeplearning/ref/checklayer.html#d117e116206

[Raza Ali]

For my case the data from the upper layer will be in this dimension (256x256x3 or 512x512x3) I cannot change the dimensions.

any other psossiblity to entertain the images of these size?