How to train features that have been extracted by using GoogleNet?

Hello,

I extracted features by using GoogleNet, but I do not know how to train it to create classifier?

Thanks in advance

[rcjr15]

How did you extracted the features using Googlenet. Kindly help.

[sahar alhaddad]

I use this way to extract features:

net = googlenet()

%Extract features

featureLayer = 'pool5-drop_7x7_s1';

trainingFeatures = activations(net, trainingSet, featureLayer);

[rcjr15]

I am getting error while using the above command. Can someone help me with this? @Jatin Waghela @Mathworks

[sahar alhaddad]

Have you define the "trainingSet"?

[rcjr15]

Yes. I have defined trainingSet.

[sahar alhaddad]

what is the error that you have got?

Answer 1

[Jatin Waghela]

If I understood you correctly, you would like to transfer learning to retrain GoogLeNet to create a classifier.

Please refer to the below documentation link which gives more information on Pretrained GoogLeNet convolutional neural network:

https://www.mathworks.com/help/nnet/ref/googlenet.html

[sahar alhaddad]

Actually, I extract features by using activation, now I do not know how to train it to get a classifier.

fitecec and other methods do not work. I think I need to understand the output of activation.

[keke zhang]

Hi,I want to know how to extract features from GoogLeNet, the function 'activations' are not work, the object function of DAGNetwork only has three functions, and there not exist the function 'activations'.https://cn.mathworks.com/help/nnet/ref/dagnetwork.html?s_tid=srchtitle

[rcjr15]

Yes, the activations function is not working and how to extract the features learned by the googlenet ? In Alexnet we could use activation (Alexnet,trainimages,'fc7') in order to extract the features. Any similar function for googlenet?

[sahar alhaddad]

I use this way to extract features:

net = googlenet()

%Extract features

featureLayer = 'pool5-drop_7x7_s1';

trainingFeatures = activations(net, trainingSet, featureLayer);

[rcjr15]

I am getting error while using the above command. Can someone help me with this? @Jatin Waghela

[Sivaramakrishnan Rajaraman]

Has anyone tried to extract features from layers other than pool5-drop_7*7_s1? If so what is the syntax?

[umit kacar]

https://www.mathworks.com/matlabcentral/answers/379635-error-using-activations-with-googlenet-in-r2017b#answer_302390

Answer 2

[michael scheinfeild]

use some layer as the feature . then you can use it for other classifier

    net = googlenet;
    inputSize = net.Layers(1).InputSize;
    imds = imageDatastore(dbpathSave)
    augimdsTrain = augmentedImageDatastore(inputSize(1:2),imds );
    %%Extract Image Features
    layer = 'loss3-classifier';%1000
    featuresTrain = activations(net,augimdsTrain,layer,'OutputAs','rows');