11
$\begingroup$

In the paper Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network by Christian Ledig et al., the distance between images (used in the loss function) is calculated from feature maps extracted from the VGG19 network. The two used in the article are (a bit confusingly) called VGG22 and VGG54.

What are these feature maps?

What do the designations "22" and "54" mean?

Improve this question
$\endgroup$
4
  • $\begingroup$ 19 is the number of layers. Probably the rest mean the same? $\endgroup$
    – Alex
    Commented Dec 31, 2017 at 17:21
  • $\begingroup$ If it was only that simple... ;-) These are designations of mapping from the VGG19, not networks in their own right. $\endgroup$
    – Lafayette
    Commented Jan 1, 2018 at 9:46
  • $\begingroup$ I never read the paper. This is the first thing that comes to mind when i see the acronym. $\endgroup$
    – Alex
    Commented Jan 1, 2018 at 15:05
  • $\begingroup$ Your assumption is indeed reasonable, but they do say that it is not the case, only the VGG19 network is used. $\endgroup$
    – Lafayette
    Commented Jan 1, 2018 at 15:46

1 Answer 1

Reset to default
4
$\begingroup$

Reading the article, it seems like they define VGG54 as the loss calculated from the euclidean distance between the $\phi_{5,4}$ feature maps derived from both the high and low resolution images using the VGG19 network. Where $\phi_{i,j}$ is defined as "the feature map obtained by the j-th convolution (after activation) and before the i-th max-pooling layer within the VGG19 network".

Improve this answer
$\endgroup$
3
  • $\begingroup$ I assume the same is true for VGG22 - that is, it being the loss calculated from ϕ2,2. Is that right? $\endgroup$
    – Lafayette
    Commented Feb 22, 2018 at 9:45
  • 1
    $\begingroup$ That's correct :) $\endgroup$
    – Carlos S. Na
    Commented Feb 22, 2018 at 17:12
  • 1
    $\begingroup$ Can you please elaborate on "the feature map obtained by the j-th convolution (after activation) and before the i-th max-pooling layer within the VGG19 network"? $\phi_{5,4}$ means $4^{th}$ layer before $5^{th}$ max-pooling layer right? But $4^{th}$ layer has so many filters (I think 512). So we would have 512 feature spaces. Which one to choose from this? Also what does "after activation" mean? $\endgroup$
    – Nagabhushan S N
    Commented Nov 20, 2018 at 17:31

Your Answer

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

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