Padding (p=3) does not increase de Nc. Thus the answer could be 67x67x32 or 64x64x32. Since the question does not show the filter (f), I tried to find it. But it does not match with the answers shown.
Can you explain to me the correct answer?
They are only talking about padding, not filtering. If you have 61 x 61 for height and width and use pad = 3, the padding happens on both sides, right? So that adds 6 and gives you 67 x 67 x 32, since padding does not modify the “channel” dimension, only the H and W dimensions.
Note that at least the last part of the explanation I just gave was also included in the error response that they gave you shown above.
So, in this case padding is just a count on how many lines (Lines of Pixels so to speak) we add on either side of the image, 3 horizontally and 3 vertically? This idea is never discussed in the lectures, and I also can’t seem to find it anywhere online
Yes, you have the definition right for padding. But this is discussed in the lectures in full detail. I suggest you watch them again if you missed all that on the first time through.
Alright, I will surely do! Thank you for your prompt replies!
You can use the “interactive transcript” feature to find a particular point in the lectures. Just click on the transcript and then use the browser text search (^F) to find “padding”.
reminds of HTML padding of images, where the padding in pixels is applied to all sides if you specify just one number for padding
I don’t know what you mean by “padding in relation to a filter”. Please give us a reference to which lecture and the time offset you are referring to.
The discussion of padding of input images or inputs at any layer is discussed in this lecture in Week 1. The title of the lecture is … wait for it … “Padding”. I didn’t actually need the interactive transcript feature in order to find it.
I just had, what the germans call, a “Denkfehler”. Everything is clear now. thank you!