Training a Convolutional Autoencoder

11 visualizzazioni (ultimi 30 giorni)
Aghata
Aghata il 23 Mag 2024
Risposto: ali kaffashbashi il 17 Ott 2024
I'm trying to train this simple convolutional autoencoder but I'm getting error on the training part. The error says the size of predictions and tragets are not the same. But When I check the network structure using the analyseNetwork function it seems that my input has the same size as my output. I can't find where is the error. Can someone help me?
Follows the code
datastore_MP = datastore("Tiles_MP1_100ov50\", "IncludeSubfolders",true, "LabelSource","foldernames");
images_MP = cell(numel(datastore_MP.Files), 1);
for i = 1:numel(datastore_MP.Files)
img_MP = readimage(datastore_MP, i);
[rows, cols] = size(img_MP);
images_MP{i} = img_MP;
end
encoderBlock = @(block) [
convolution2dLayer(3,2^(3+block), "Padding",'same')
reluLayer
maxPooling2dLayer(2,"Stride",2)
convolution2dLayer(3,2^(5+block), "Padding",'same')
reluLayer
maxPooling2dLayer(2,"Stride",2)];
net_E = blockedNetwork(encoderBlock,1,"NamePrefix","encoder_");
decoderBlock = @(block) [
transposedConv2dLayer(3,2^(5-block),"Stride",2)
reluLayer
transposedConv2dLayer(3,2^(1-block), "Stride",2)
reluLayer];
net_D = blockedNetwork(decoderBlock,1,"NamePrefix","decoder_");
inputSize = [100 100 1];
CAE = encoderDecoderNetwork(inputSize,net_E,net_D);
analyzeNetwork(CAE)
options = trainingOptions( "adam",...
"Plots","training-progress",...
"MaxEpochs", 100,...
"L2Regularization",0.001);
trainedCAE = trainnet(datastore_MP, CAE, "mse", options);

Risposte (2)

newhere
newhere il 23 Mag 2024
Hey, try changing 'trainnet' to 'trainNetwork'.
trainedCAE = trainNetwork(datastore_MP, CAE, "mse", options);
  1 Commento
Aghata
Aghata il 27 Mag 2024
Hello, Fatma.
I tried but this doesn't fix the problem. I've been trying to find a way to set the target size to be the same as the input or output, but without success.

Accedi per commentare.


ali kaffashbashi
ali kaffashbashi il 17 Ott 2024
I guess it tries to set your label sources (the folder names) as targets during the training. Hence, the input and output sizes become different. I reckon using the following code instead of your training line will solve your problem:
trainedCAE = trainnet(combine(datastore_MP,datastore_MP), CAE, "mse", options);

Categorie

Scopri di più su Simulink Functions in Help Center e File Exchange

Prodotti


Release

R2024a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by