It is a known fact that preprocessing images using CV techniques will improve CNN performance (see this answer).

But what happens when you feed in the entire image and the filtered image randomly to the network? Would the Neural Network learn to focus on the relevant aspects of an unfiltered image?

If yes, please explain how randomly processed images improve the CNN's performance/sample efficiency.

  • $\begingroup$ The question in the title seems to be different and more general than the question in the body of your post. Could you please clarify what your main question is here and how the two questions are related? $\endgroup$
    – nbro
    Jul 14 at 19:16
  • $\begingroup$ @nbro Thank you. Made the corrections. $\endgroup$ Jul 14 at 20:04
  • $\begingroup$ Ok, it's still not clear to me whether you're talking about "data augmentation", i.e. where you artificially generate data by applying some transformations to your original data so that to increase the size of your dataset. Then you would feed either the original images or some transformed image. So, would your question be: "Why is data augmentation useful?". If yes, please, edit your post to reformulate the question by mentioning "data augmentation" specifically. $\endgroup$
    – nbro
    Jul 16 at 12:32
  • $\begingroup$ To me, what's still confusing in your post is this sentence: "when you feed in the entire image and the filtered image randomly to the network". It's not clear whether you're considering the case where you feed 2 images to the neural network each time: one is the original one and the other is the transformed one (chosen randomly??). $\endgroup$
    – nbro
    Jul 16 at 12:34

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.