I have an image dataset where the folder structure is as follows- there are 900 folders (all of which will be classes) and in each folder, we have a varying number of images(each image has been resized to the same pixel intensity).

Now when we construct the 4d matrix to be fed to the deep learning model it should be of the dimension-


900 because that's the total number of classes
105 x 105 because I have resized each image to that 
now A should be the number of images in each folder.

In my case the number of images in each folder is varying so my question is-

If I keep the original varying number of images in each folder that means in my 4d array A would be the dimension of the largest folder size which means for folders of lesser number of images it will have a zero 2d matrix. So is that technically right or should I make all the folder sizes same?

  • $\begingroup$ Correct me if I'm wrong, but in a siamese network you will feed "pairs" of images to the network, right? So you need to decide on the number of pairs generated from each class. Without seeing the actual network it is hard to answer your question. $\endgroup$ – Peter May 16 at 11:26

Your Answer

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

Browse other questions tagged or ask your own question.