How to add Distance transformation Map in loss function at classification layer.

Hi Everyone, I am trying to insert distance map information in loss fucntion. i am doing this in classification layer of CNN.

but when I calculate the disance map using "bwdist(Y)" commnad, during training process the MATLAB produce error

"Error using 'forwardLoss' in Layer ClassificationLayer. The function threw an error and could not be executed".

"Expected input image to be a 2-D real-valued, non-sparse gpuArray with underlying class uint8, uint16, uint32, int8, int16, int32, logical, single or double".

How can I add Distance transformation Map in loss fucntion. or how to resolve this issue?

[Madhav Thakker]

Can you share your code?

[Raza Ali]

The classification layer code:

%%%%%%%%%%%%%%%%%%

classdef CEDLossLayer < nnet.layer.ClassificationLayer

properties

% Row vector of weights corresponding to the classes in the

% training data.

Beta=0.7;

end

methods

function layer = CEDLossLayer(name)

% layer = CEDLossLayer(name) creates a

% Set layer name.

if nargin == 2

layer.Name = name;

end

% Set layer description

layer.Description = 'cross entropy';

end

function loss = forwardLoss(layer, Y, T)

% loss = forwardLoss(layer, Y, T) returns the cross entropy loss between the predictions Y and the training

% targets T.

N = size(Y, 4) * size(Y, 1) * size(Y, 2);

Y = squeeze(Y);

T = squeeze(T);

addpath ('D:\Ditsnave Transform')

W=WeightMap(T(:,:,1));

W = gather(W);

loss_i = ((layer.Beta).*W.*T .* log(nnet.internal.cnn.util.boundAwayFromZero(Y)))+((1-layer.Beta).*(1-W).*(1-T) .* log(1-(nnet.internal.cnn.util.boundAwayFromZero(Y))));

loss = -sum( sum( sum( sum(loss_i, 3).*(1./N), 1), 2));

end

function dLdY = backwardLoss(layer, Y, T)

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

% cross entropy loss with respect to the

% predictions Y.

N = size(Y, 4) * size(Y, 1) * size(Y, 2);

Y = squeeze(Y);

T = squeeze(T);

addpath ('D:\Ditsnave Transform')

W=WeightMap(T(:,:,1));

W = gather(W);

dLdY= (-(W.*T./nnet.internal.cnn.util.boundAwayFromZero(Y))).*(1./N);

% dLdY= -(1./N).*((((layer.Beta).*T)./nnet.internal.cnn.util.boundAwayFromZero(Y))-((1-layer.Beta).*(1-T))./(1-nnet.internal.cnn.util.boundAwayFromZero(Y)));

end

end

end

[Raza Ali]

%% Weight Map Function

function [weight]=WeightMap(gt);

% class balance weights w_c(x)

uvals=unique(gt);

wmp=zeros(1,length(uvals));

for uv=1:length(uvals)

wmp(uv)=1/sum(gt(:)==uvals(uv));

end

% this normalization is important!

%background pixels must have weight 1

wmp=wmp/max(wmp);

% wc=double(gt);

% wc=uint8(gt);

wc=zeros(size(gt));

for uv=1:length(uvals)

wc(gt==uvals(uv))=wmp(uv);

end

% cells instances for distance computation

cells=bwlabel(gt==1, 4);

% cells distance map

bwgt=zeros(size(gt));

maps=zeros(size(gt,1),size(gt,2),max(cells(:)));

if max(cells(:))>=2

for ci=1:max(cells(:))

maps(:,:,ci)=bwdist(cells==ci);

end

maps=sort(maps,3);

d1=maps(:,:,1);

d2=maps(:,:,2);

bwgt=10*exp(-((d1+d2).^2)./(2*25) ).*(cells==0)

end

% unet weights

weight=wc + bwgt;

end