observationDim should be either 2, 3, 4 or 5

Hi, sorry about noob question.

I am trying to adapt the TrainNetworkOnImageAndFeatureDataExample.mlx to my use case. I my use case I only have the training image data and the categories, the "X2Train" feature input is absent. Thus my layers now look as follows:

[h,w,numChannels,numObservations] = size(trainData);

numFeatures = 1;

numClasses = numel(categories(trainLabels));

imageInputSize = [h w numChannels];

filterSize = 5;

numFilters = 16;

layers = [

imageInputLayer(imageInputSize,Normalization="none")

convolution2dLayer(filterSize,numFilters)

batchNormalizationLayer

reluLayer

fullyConnectedLayer(50)

flattenLayer

fullyConnectedLayer(numClasses)

softmaxLayer

classificationLayer

];

%concatenationLayer(1,2,Name="cat")

lgraph = layerGraph(layers);

When running the trainNetwork I get an error:

observationDim should be either 2, 3, 4 or 5

But I have no clue whatsoever what I need to change here.

Answer 1

[Sai Kiran]

Hi,

The error is due to mismatch in the dimensions of two matrices while multiplying them.

The layers doesn't have any errors when we pass another dataset.

Please check the dimensions of 'trainData' and 'trainLabels' and the command window , it shows where the error has occured at a specific line in the code.

I hope it helps!