vedaldi
Some questions in CNN output formula
Hi! Very glad to be the first person starting an issue!
I am a new learner to CNNs and recently I found some problem when reading the MatConvNet documentation or VGG Convolutional Neural Networks Practical.
In Part 1.1 of the practical there is a question "Note that x is indexed by i+i′ and j+j′, but that there is no plus sign between k and k′. Why?" This is also my question.
(1) In that formula, why does x has four dimensions but f has only three dimensions? When defining x,y and f we found clearly that x and y are 3-dimensional, f is 4-dimensional. Is there any typo in this formula? (2) Why do the first two dimensions of x are i+i′ and j+j′? Does that mean that (row-1, col-1) of x is not involved in the convolution process? If so, I don't think it is right.
Hi, that’s because the convolution applied spatially, but not done across feature channels. This is in principle possible, but it would amount to account for “channel” as a third coordinate analogous to space, which would be odd. For example, there is usually no preferential order in features (although this rule is not always true: there is in local cross-channel normalisation).
On 17 Mar 2015, at 05:11, Zhiyang Wang [email protected] wrote:
Hi! Very glad to be the first person starting an issue!
I am a new learner to CNNs and recently I found some problem when reading the MatConvNet documentation or VGG Convolutional Neural Networks Practical.
In Part 1.1 of the practical there is a question "Note that x is indexed by i+i′ and j+j′, but that there is no plus sign between k and k′. Why?" This is also my question.
(1) In that formula, why does $x$ has four dimensions but $f$ has only three dimensions? When defining x,y and f we found clearly that x and y are 3-dimensional, f is 4-dimensional. (2) Why do the first two dimensions of $x$ are i+i′ and j+j′? Does that mean that (row-1, col-1) of x is not involved in the convolution process? If so, I don't think it is right.
— Reply to this email directly or view it on GitHub https://github.com/vedaldi/practical-cnn/issues/1.
Thanks for answering, Andrea. I have verified my understandings using my own code and it matches the result in MatConvNet, which means my understandings are right!
What I am arguing is that I think the formula is wrong. It should be like
in the VGG practical or
in the manual of MatConvNet.
Thanks again for helping!
Zhiyang
Hi, except for the bias term b_k and the fact that the indexes have been renamed, is there any difference between the two formulas?
On 19 Mar 2015, at 02:17, Zhiyang Wang [email protected] wrote:
Thanks for answering, Andrea. I have verified my understandings using my own code and it matches the result in MatConvNet, which means my understandings are right!
What I am arguing is that I think the formula is wrong. It should be like https://cloud.githubusercontent.com/assets/7233963/6723299/e277b444-ce20-11e4-8ad1-e8161f19da32.jpg in the VGG practical or https://cloud.githubusercontent.com/assets/7233963/6723304/f531aff4-ce20-11e4-8f58-f4b5c5a34219.jpg in the manual of MatConvNet.
Thanks again for helping!
Zhiyang
— Reply to this email directly or view it on GitHub https://github.com/vedaldi/practical-cnn/issues/1#issuecomment-83267848.
Hi @vedaldi you didn't understand what I mean!
The above two formulas are surely right and they make no difference. But in the VGG practical and the MatConvNet manual, the two formulas are wrong.
Specifically, in the practical, your formula is
However it should be
Similarly, in the manual, the formula is
The right form should be
Hope you can find the differences and understand what I mean.
i have a question regarding to this expression that what does i' and j' signifies
