1
$\begingroup$

I know that the first layer uses a low-level filter to see the edge information. As the layer gets deeper, it will represent high-level (abstract) information. Is it because the combinations of filters used in the previous layer are used as filters in the next layer? ("Does the combination of the previous layer's filters make the next layer's filters?) If so, are the combinations determined in advance?

$\endgroup$
1
$\begingroup$

I guess you are making mistake about the filters. After applying filters to the input of each layer, the output will be used as the input of the next layer. The first layer's filters try to find the edges in the image and their output show whether those edges exist in a specified position or not. Next layer filters try to find patterns in the outputs of the previous layer which shows the existence of edges. Due to the point that each filter is a window and specifies a receptive field on the input, it finds patterns in the input which are more abstract and more complicated than the previous layers' activations.

| improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Thank you for your answer. I understood! Because of this site, I confused the concept of the entire CNN process. Thanks again! $\endgroup$ – hyunsu jeong Jul 19 '18 at 4:56

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.