Error saying Layer 'conv2d_top': Invalid input data. Complex numbers are not supported".

Hemanth

2023 1 5

0 个回答

10 次查看（30 天）

显示更早的评论

0 个投票

Iam exploring with "Unsupervised day to dusk image translation using Unit", code is provided by matlab in the below link https://in.mathworks.com/help/images/unsupervised-day-to-dusk-image-translation-using-unit.html. I worked with my own dataset and in training divided the single image into six blocks , each blocks are individually trained. While training 2nd block of 1 st image,I got the error like.Please help me in resolving this.

function [genGrad,genLoss,images] = modelGradientGen(gen,discA,discB,ImageA,ImageB,lossWeights)
    
    [ImageAA,ImageBA,ImageAB,ImageBB] = forward(gen,ImageA,ImageB);
    hidden = forward(gen,ImageA,ImageB,Outputs="encoderSharedBlock");
    OutputSize= [150  100] ;
    ImageAA = dlresize(ImageAA,'OutputSize', OutputSize);
    ImageAB = dlresize(ImageAB,'OutputSize', OutputSize);
    ImageBA = dlresize(ImageBA,'OutputSize', OutputSize);
    ImageBB = dlresize(ImageBB,'OutputSize', OutputSize);
    
 [~,ImageABA,ImageBAB,~] = forward(gen,ImageBA,ImageAB);
    cycle_hidden = forward(gen,ImageBA,ImageAB,Outputs="encoderSharedBlock");
    
    ImageABA = dlresize(ImageABA,'OutputSize',OutputSize);
    ImageBAB = dlresize(ImageBAB,'OutputSize',OutputSize);
    
    % Calculate different losses
    selfReconLoss = computeReconLoss(ImageA,ImageAA) + computeReconLoss(ImageB,ImageBB);
    hiddenKLLoss = computeKLLoss(hidden);
    cycleReconLoss = computeReconLoss(ImageA,ImageABA) + computeReconLoss(ImageB,ImageBAB);
    cycleHiddenKLLoss = computeKLLoss(cycle_hidden);
   
    outB = forward(discB,ImageAB);
    outA = forward(discA,ImageBA); 
    
      advLoss = computeAdvLoss(outB)+computeAdvLoss(outA); 
    % Calculate the total loss of generator 
    genTotalLoss = ...
        selfReconLoss*lossWeights.selfReconLossWeight + ...
        hiddenKLLoss*lossWeights.hiddenKLLossWeight + ...
        cycleReconLoss*lossWeights.cycleConsisLossWeight + ...
        cycleHiddenKLLoss*lossWeights.cycleHiddenKLLossWeight+ ...
        advLoss*lossWeights.advLossWeight; 
    
    % Update the parameters of generator
    genGrad = dlgradient(genTotalLoss,gen.Learnables); 
    
    % Convert the data type from dlarray to single
    genLoss = extractdata(genTotalLoss);
    images = {ImageAA,ImageAB,ImageB,ImageBA};
end