1
$\begingroup$

I have used a pre-trained CNN to extract features from training and test images sets. The same CNN was used for all images. The CNN includes normalization layers.

Before training a classifier (SVM etc) on these feature vectors, should they be normalized? I.e:

scaler3 = preprocessing.StandardScaler()
X = scaler3.fit_transform(X_train)
Xt = scaler3.transform(X_test)

Normally, I would, but I am unsure in this case as they derive from the same CNN. Thanks.

$\endgroup$
1
$\begingroup$

I would go with No for the following reason. Normally you use CNN with the last layer being softmax. And clearly you don't do anything to the output of CNN before feeding it to Softmax. In your case, you change Softmax to SVM so I think you should treat it the same. And what you mentioned reminds me of batch normalization. You can read about batch normalization here http://arxiv.org/abs/1502.03167

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.