How to define a custom loss function using SSIM

Hi All -- I'm relatively new to deep learning. I want to define and reconst a loss function.

I have study how to define the MAE loss function through define-custom-deep-learning-layers (https://www.mathworks.com/help/deeplearning/ug/define-custom-deep-learning-layers.html).

However, those methods are based on the computation of pixels to pixels.

I want to use the ssim as the loss function and write the code. But it does not work. Could you help me? Thank you!

function loss = forwardLoss(layer, Y, T)

% loss = forwardLoss(layer, Y, T) returns the SSIM loss between

% the predictions Y and the training targets T.

[m n R N] = size(Y);

all_loss = zeros(R,N);

for j = 1:N

for i = 1:R

temp1 = (Y(:,:,i,j));

temp2 = (T(:,:,i,j));

all_loss(i,j) = ssim(temp1,temp2); % The ssim() is to calculate the structure similarity of two gray images.

end

end

loss = 1 - mean(all_loss,'all');

end