0
$\begingroup$

I'm doing some experimentation with a CNN and have 2 conv layers with 32 and 64 filters respectively. I started out with 3x3 kernel sizes and noticed that when I increased it to 5x5, 7x7 etc the time per epoch shot up! Why is this? Is this simply due to the fact that more element wise calculations are necessary per pixel in the image?

Improve this question
$\endgroup$
1
  • $\begingroup$ Have you tried plotting your time per epoch vs kernel size (9, 25, 49)? If you see something close to a straight line then the answer to your question is "yes" $\endgroup$
    – Valentas
    Jun 23, 2022 at 7:52

1 Answer 1

Reset to default
0
$\begingroup$

It is expected to train a model slower as parameters increase. Here you add more computational cost by increasing kernel size.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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