C4W1L04 Padding

[@kvishnudev]

Andew is one of the most knowledgeable person on machine learning out there. His explanation is much based on theory. Thank you very much sharing the valuable info in youtube

[@seanmenzies1986]

Such great explanations to what's going on under the hood. Knowing is one thing, teaching is an entirely different animal. Kudos to the teacher for having mastered both!

[@shahriarrahman8425]

Was struggling with some conceptions and this lecture provided insights I needed. Thank you so much!

[@vipinmakde2426]

When the stride is greater than 1 and we are using SAME padding concept, thus the size of the input image is retained. For example : input = 16, filter = 5, stride =5. Is value of P =(5-1)/2=2 ? Thus makeing the output of size [(n+2p-f)/stride]+1= 4 It can be seen clearly that output size is not retained.

[@farmanshah1647]

The formula for calculating p in case of stride > 1 should be: Derivation: (n + 2p - f)/s + 1 = n (n + 2p - f + s) / s = n n + 2p -f + s = ns Formula: p = [n(s-1) + f - s] / 2 In your case where: input = 16, filter = 5, stride =5. p = [16(5-1) + 5 - 5] / 2 p = 64 / 2 = 32 Recheck: Output = (n + 2p - f)/s + 1 = (16 + 64 - 5) / 5 + 1 = 16

[@marcbroadus]

Sir Andrew, you're the boss!

[@aramroshani6197]

Thanks for your great videos, but please can you explain which parmeter(s) determine the value for each member of the filters, I mean why do you use (1,1,1 / 0,0,0 / -1,-1,-1) but not other numbers. Thanks advance.

[@hammad365]

He explained that in the previous video, and explained other methods that use other numbers, or parameters that could be learnt too

[@rachadlakis1]

Thank you.

[@anjanakesavan5615]

Thank you so much for the explanation on zero padding. Why do we use 0 for padding and not 1 ?

[@Payah-sy8qw]

because we want the pixel values ​​we want to filter using the kernel, still get the actual pixel values ​​when convoluted as they are not affected by the numbers in the padding

[@rahuldey6369]

An extremely insightful lecture as always have been. But when to use Valid padding and when to use Same it wasn't clear, I mean in which scenarios we'll consider them and why?

[@Sayied-s7d]

odd numbers are good

[@Shewanee25]

What is the difference between reflective padding and zero padding ?

[@rahulpalli36]

in reflective padding , the edge pixels are added onto the outside copying the pixels from the edge of the image.in zero padding just add the zero pixels

[@rohanshetty1016]

***Note: 5:58 - Valid Convolution actually means that the input image size is 'valid' for this operation and there's no need for padding...

[@gorgolyt]

No, that's not what it means. Your comment really makes no sense in context, a convolution can be applied to any image of any size, there is no concept of a "valid image size for the operation". The video is correct.

[@danieltheshark]

thank you so much... Just Thank you !