Currently, I am reading Rethinking Model Scaling for Convolutional Neural Networks. The authors are talking about a different way of scaling convolutional neural networks by scaling all dimensions simultaneously and relative to each dimension. I understand the scaling methods regarding the depth of a network (# layers) and the resolution (size of the input image).

What I was stumbling is the concept of the network's width (# channels). What is meant by the width or the number of channels of a network? I don't think it is the number of color channels, or is this the case? The number of color channels was the only link I found regarding the terms "ConvNets" and "number of channels".

  • $\begingroup$ It's not clear to me the relationship between "network's width" and "number of channels". Do the authors use the term "width" to refer to the "number of channels" in the paper? That would be very strange. $\endgroup$ – nbro Jun 23 at 16:59

It is exactly that - the number of color channels or any other analogue to color that you use.

| improve this answer | |
  • $\begingroup$ Ok, I did not know this.. I thought, the max color space of an image with space = 3 would be 3. But instead, I could also synthetically increase the color space to 32 or some other number, correct? $\endgroup$ – Tobitor Jun 23 at 13:34
  • 1
    $\begingroup$ Yes, that's right. And you also can use CNNs not only for regular images, you can analyze any data that is intrinsically 2-dimentional. $\endgroup$ – oleg.mosalov Jun 23 at 13:42
  • $\begingroup$ Ok, perfect! Thank you very much :) $\endgroup$ – Tobitor Jun 23 at 13:56

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.