My network is returning both 100 percent training and validation. What seems to be the issue?

Question

Ramon Miguel Legaspi 2021-11-29

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1598019-my-network-is-returning-both-100-percent-training-and-validation-what-seems-to-be-the-issue

回答： Prateek Rai 2022-3-2

myfolder = 'C:\Users\MigueLegaspi\Downloads\fall dataset\rgb';
dataDir = fullfile(myfolder);
imdir = fullfile(dataDir);
myfolder2 = 'C:\Users\MigueLegaspi\Downloads\Validation\rgb';
dataDir2 = fullfile(myfolder2);
imdir2 = fullfile(dataDir2);
imds = imageDatastore(imdir, "IncludeSubfolders",true ,"LabelSource","foldernames");
imds2 = imageDatastore(imdir2,"IncludeSubfolders",true,"LabelSource","foldernames");
numTrainfiles =6963;
numValidfiles = 6598;
[imdsTrain] = splitEachLabel(imds,numTrainfiles,'randomized');
[imdsValidation] = splitEachLabel(imds2,numValidfiles,'randomized');
%definingarchitecture
numClasses = numel(categories(imdsTrain.Labels));
numClasses2 = numel(categories(imdsValidation.Labels));
layers = [
    imageInputLayer([240 320 3])
    convolution2dLayer(3,16,'Padding',1)
    batchNormalizationLayer
    reluLayer
    maxPooling2dLayer(2,"Stride",2)
    convolution2dLayer(3,32,'Padding',1)
    batchNormalizationLayer
    reluLayer
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer];
%trainetwork
options = trainingOptions('sgdm', ...
    'MaxEpochs',4, ...
    'MiniBatchSize',64,...
    'ValidationData',imdsValidation, ...
    'ValidationFrequency',30, ...
    'Verbose',false, ...
    'Plots','training-progress');
net = trainNetwork(imdsTrain,layers,options);
YPred = classify(net,imdsValidation);
yvalidation = imdsValidation.Labels;
accuracy = mean(Ypred == yvalidation);

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Prateek Rai 2022-3-2

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1598019-my-network-is-returning-both-100-percent-training-and-validation-what-seems-to-be-the-issue#answer_907640

To my understanding, you are trying to train the network and the network is returning both 100 percent training and validation.

One possible reason for this could be that your datastore is not created properly. Your folder might contain only one subfolder and hence creating only one class in imds which might be leading to 100 percent training and validation accuracy.

Try checking the dataset once. You can refer to the imageDatastore MathWorks Documentation page to learn more about creating a datastore for image data.