Grey Image Colorization using deep learning

Hello, I'm trying to colorize an grey image using Matlab. there is multiple example in pyhton etc. I can't find any example in mathlab. I'm sharing the codes I got until today.

When training is over I got a little colored images but mostly yelow. And traning progress down below as an image. Can you tell me how can I read this graph? How can I understant it's fail or still learning ?

What should I do ? Some examples using Lab colorstyle instead of RGB. How can I train using Lab?

And I want to learn this codes meanings :

miniBatchSize = 16;
patchSize = [250 250];
patchds = randomPatchExtractionDatastore(greyimds,Trainimds,patchSize, ....
'PatchesPerImage',64);
patchds.MiniBatchSize = miniBatchSize;

When I scale up miniBatchSize my training fail because of GPU. Why?

When I scale up encoderDepth, traning can't work because it's ask for 2*Image height/width or something like that? How can I calculate it ? What it's means?

lgraph = unetLayers([250 250 3] , 3,'encoderDepth',1);

And last question: How can I get score of tranings success rate? I try some codes but they don't let me because it's not a classification problem.

Thanks for everything.

imagesDir = '.';
greyImagesDir = fullfile(imagesDir,'greyImages');
trainImagesDir = fullfile(imagesDir,'trainImages');
exts = {'.jpg','.bmp','.png'};


greyimds = imageDatastore(blurredImagesDir)
Trainimds = imageDatastore(trainImagesDir)



miniBatchSize = 16;
patchSize = [250 250];
patchds = randomPatchExtractionDatastore(greyimds,Trainimds,patchSize, ....
'PatchesPerImage',64);
patchds.MiniBatchSize = miniBatchSize;

lgraph = unetLayers([250 250 3] , 3,'encoderDepth',1);
lgraph = lgraph.removeLayers('Softmax-Layer');
lgraph = lgraph.removeLayers('Segmentation-Layer');
lgraph = lgraph.addLayers(regressionLayer('name','regressionLayer'));
lgraph = lgraph.connectLayers('Final-ConvolutionLayer','regressionLayer');

maxEpochs = 30;
epochIntervals = 1;
initLearningRate = 0.1;
learningRateFactor = 0.1;
l2reg = 0.0001;
options = trainingOptions('sgdm', ...
    'Momentum',0.9, ...
    'InitialLearnRate',initLearningRate, ...
    'LearnRateSchedule','piecewise', ...
    'LearnRateDropPeriod',10, ...
    'LearnRateDropFactor',learningRateFactor, ...
    'L2Regularization',l2reg, ...
    'MaxEpochs',maxEpochs ,...
    'MiniBatchSize',miniBatchSize, ...
    'GradientThresholdMethod','l2norm', ...
    'Plots','training-progress', ...
    'GradientThreshold',0.01);
    
modelDateTime = datestr(now,'dd-mmm-yyyy-HH-MM-SS');
    net = trainNetwork(patchds,lgraph,options);
    save(['trainedNet-' modelDateTime '-Epoch-' num2str(maxEpochs*epochIntervals) ...
            'ScaleFactors-' num2str(234) '.mat'],'net','options');
        
            
          
------------------Colorize image using code below---
        
Idegray = activations(net,testimage,'regressionLayer');
figure; imshow(Idegrey)
Iapprox = rescale(Idegrey);
Iapprox = im2uint8(Iapprox);
imshow(Iapprox)
title('Colored Image')

       
         

Answer 1

[Tolga Arslan]

I found answer of EncoderDepth problem :

Network input image size, specified as a:

2-element vector in the format [height, width].

3-element vector in the format [height, width, depth]. depth is the number of image channels. Set depth to 3 for RGB images, 1 for grayscale images, or to the number of channels for multispectral and hyperspectral images.

Note
Each encoder section has a 2x2 maxPooling2dLayer that halves the image size. The height and width of the input image must be a multiple of 2D, where D is the value of EncoderDepth.

If I understant true : I'm working with 250x250 image size. I have to use 3 for EncoderDepth, My input can gray scale but output is RGB. If I make it 3, 2³ =8 250/8 : 31,25. It's mean fails. I have to setup by image sizes 256x256.

Please comment if I understant correctly this situation. I will try and let you know asap.