Error with layer when trying to train my own CNN

1 次查看（过去 30 天）

显示更早的评论

Alizée Perrin 2017-12-6

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/371482-error-with-layer-when-trying-to-train-my-own-cnn

评论： sasmita mahakud 2019-6-28

在 MATLAB Online 中打开

Hello,

I am training my own neural network (like in this example) with my own dataset, to classify images.

Problem is, I have this error message when I try to start the training :

Error using trainNetwork (line 140)
Invalid training data. The output size (250000) of the last layer doesn't match the number of classes (1).
Caused by:
    Error using nnet.internal.cnn.util.TrainNetworkDataValidator/assertCorrectResponseSizeForOutputLayer (line
    217)
    Invalid training data. The output size (250000) of the last layer doesn't match the number of classes (1).

Here is my code :

digitDatasetPath = 'ChinaSet_AllFiles/CXR_png';
digitData = imageDatastore(digitDatasetPath,...
           'IncludeSubfolders',true,'LabelSource','foldernames');
% creating datastore with files from my folder  
trainingNumFiles = 500;
rng(1) % For reproducibility
[trainDigitData,testDigitData] = splitEachLabel(digitData,...
        trainingNumFiles,'randomize');
% splits the image files in digitData into two new datastores, trainDigitData and testDigitData
layers = [imageInputLayer([600 600 1]);
convolution2dLayer(500,50);
reluLayer();
maxPooling2dLayer(2,'Stride',2);
fullyConnectedLayer(500*500);
softmaxLayer();
classificationLayer()];
options = trainingOptions('sgdm','MaxEpochs',50,...
  'InitialLearnRate',0.0001);
convnet = trainNetwork(trainDigitData,layers,options);  % starts the training

Does anyone knows where it comes from ?

Thanks in advance!

2 个评论
显示无隐藏无

Maria Duarte Rosa 2017-12-15

在 MATLAB Online 中打开

Hi Alizee,

The outputSize of the fullyConnectedLayer should be equal to the number of classes in your dataset. Currently you have outputSize = 500*500. If your digit dataset contains 10 classes then you should use fullyConnectedLayer(10). You can also count the number of classes from the data:

numClasses = numel(categories(trainDigitData.Labels));

Then use this variable in the fully connected layer:

fullyConnectedLayer(numClasses).

Thanks