How to solve this error: "Error using DAGNetwork/activations (line 245) Incorrectly defined MiniBatchable Datastore. Error in read method of C:\Program Files\MATL​AB\R2020b\​toolbox\ma​tlab\datas​toreio\+ma​tlab\+io\+​datastore\​@ImageData​store\read​.m"

Hi,

I have the following code to extract the features from certain layer of ResNet101 deep learning model. However, after training the network, I am unable to extract the features from the layer specified below.

imds=imageDatastore('C:\Users\Manisha\Test', 'IncludeSubfolders', true, 'LabelSource','foldernames'); % There are two subfolders

tbl = countEachLabel(imds);

minSetCount = min(tbl{:,2});

imds = splitEachLabel(imds, minSetCount, 'randomize');

tbl = countEachLabel(imds)

[imdsTrain, imdsTest] = splitEachLabel(imds, 0.75, 'randomize');

net = resnet101;

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

lgraph = layerGraph(net);

newFCLayer = fullyConnectedLayer(numClasses,'Name','new_fc','WeightLearnRateFactor',15,'BiasLearnRateFactor',15);

lgraph = replaceLayer(lgraph,'fc1000',newFCLayer);

newClassLayer = classificationLayer('Name','new_classoutput');

lgraph = replaceLayer(lgraph,'ClassificationLayer_predictions',newClassLayer);

lgraph = replaceLayer(lgraph,'ClassificationLayer_fc1000',newClassLayer);

tbl1 = countEachLabel(imdsTrain)

tbl2 = countEachLabel(imdsTest)

inputSize = net.Layers(1).InputSize;

augimdsTrain = augmentedImageDatastore(inputSize(1:2),imdsTrain);%'DataAugmentation',imageAugmenter);

imageAugmenter = imageDataAugmenter('RandRotation',[-90,90])

augimdsTest = augmentedImageDatastore(inputSize(1:2),imdsTest, 'DataAugmentation',imageAugmenter);

options = trainingOptions('adam', ...

'ExecutionEnvironment','gpu',...

'MiniBatchSize',12, ...

'MaxEpochs',20, ...

'InitialLearnRate',1e-4, ...

'Shuffle','every-epoch', ...

'ValidationFrequency',10, ...

'Verbose',true, ...

'Plots','training-progress');

trainedNet = trainNetwork(augimdsTrain,lgraph,options);

featureLayer = 'pool5'

trainingFeatures = activations(trainedNet, augimdsTrain, featureLayer, ...

'MiniBatchSize', 12, 'OutputAs', 'rows'); % error in this line

label_train = [zeros(tbl1.Count(1),1); ones(tbl1.Count(1),1)];

testFeatures = activations(trainedNet, augimdsTest, featureLayer, ...

'MiniBatchSize', 12, 'OutputAs', 'rows');

label_test = [zeros(tbl2.Count(1),1); ones(tbl2.Count(2),1)];

Answer 1

[Madhav Thakker]

Hi Manisha,

If you want your custom datastore to be MiniBatchable, the read function MUST output a 2 column table, as noted in this documentation link. https://in.mathworks.com/help/deeplearning/ug/develop-custom-mini-batch-datastore.html#mw_1fbbfc62-d6e2-4e7c-843f-67b467135050

Hope this helps.