Need help for extracting CNN features by mdCNN toolbox

[Sadani Adem]

Hello Hagay garty,

Thanks a lot for sharing your mdCNN toolbox with us

I'm an university student in the first year of master ( IA field). i have a projet to classify Minist data by using CNN as a features extraction tool then, all the extracted feature will be fed into a multi-class support vector machine. 

My question: I run your CNN model, the accuracy was great. 

If i want to use CNN+SVM, where can I find the extracted features after running the CNN program? 

in the Net struct , I found as an output for the last convolution layer, three saved variables (index, z , activation); is the output "activation" describes the extracted features? and how can i found labels that correspond to CNN features?

thanks a lot for helping me

i'm waiting for your reply

[Hagay Garty]

Hi, Sadani,

Please refer to the doc file under "classify a single sample" there is a sample code on how to extract the predicted class.

Basically you are right, the activation of the last layer is the extracted features. The predicted label is the index of the largest element in the activation vector of the last layer. (when using classification output layer)

Hagay