I am new in deep learning and I try to understand some details about CNN. Could you explain to me what is the effect of the filter size e.g 5*5 vs 7*7 on features extraction and locality structure ?


If you work in natural language processing, you can view the filter size in a CNN as the n-gram size.

If you work in computer vision, you can view the filter size in a CNN as the receptive field.

See How to work multiple filter region sizes: 2, 3 and 4 in CNN? to get a better intuition.

how to fix the size of the filter.

The choice of the size of the filter depends on the dataset. I am not aware of any rule of thumb, typically one would start with a size of the filter that has been used in similar works. Some ideas: What are the state-of-the-art methods to determine parameters in CNN, NN, RNN, or any deep learning models . E.g. in {1} we used Gaussian Processes to automatically choose the filter height.


  • $\begingroup$ I mean how to fix the size of the filter. What is the effect on extracting features when we use a small filter size vs big filter size ? $\endgroup$ – BetterEnglish Nov 14 '16 at 23:09
  • $\begingroup$ @BetterEnglish added in answer. $\endgroup$ – Franck Dernoncourt Nov 15 '16 at 1:03

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.