Extracting features from one layer of dlnetwork model MATLAB 2023a

I have used the multi-input CNN network example on the following link :

https://se.mathworks.com/matlabcentral/fileexchange/74760-image-classification-using-cnn-with-multi-input-cnn?s_tid=srchtitle_multi%2520input%2520cnn_1

After the traing and getting the predction, I need to extract the features from one of the max pooling layers of the dlnet model.

Can you help by writing the code to do so?

I have tried to use activations function with the dlnet model but not working

https://se.mathworks.com/help/deeplearning/ref/seriesnetwork.activations.html

for example, using activation with SeriesNetwork or DAGNetwor , the code is

layer = "pool10";

featuresTrain = activations(net,augimdsTrain,layer,OutputAs="rows");

featuresTest = activations(net,augimdsTest,layer,OutputAs="rows");

What is the code to do the same if I have dlnetwork model?

Answer 1

[V Sairam Reddy]

Hi Mahmoud,

I understand that you want a way to visualise the activations in a dl network.

The 'activations' function in MATLAB is however only supported in SeriesNetwork or DAGNetwork object. Instead you can use the 'foward' function to get the activations in a dl network.

Please find the attached code below:

features = forward(dlNetwork, dlArray, 'Outputs', layerName);
% In your case:
features = forward(dlnet, dlarray(single(XUpper),'SSCB'), 'Outputs', 'pool10');

To know more about the forward function, please check the following link from the documentation:

https://www.mathworks.com/help/deeplearning/ref/dlnetwork.forward.html

[MAHMOUD EID]

Hi. The issue is how to pass the two inputs to the command predict. when I run the mentioned command , I have recived this error message @V Sairam Reddy

Error using deep.internal.network.dlnetwork/validateForwardInputs

Incorrect number of network inputs. Network has 2 inputs, but 1 inputs were passed.