We have seen in previous lectures how to convolve with 3x3 kernel but here we have 28x28 (which are actually 6 kernels of 5x5) kernel, what are the values of 28x28 kernel and how does convolution works with this size of kernel?
Please see this
I still have confusion!
Q1- 3x3 kernel looks like:
-1, 0, 1
-1, 0, 1
-1, 0, 1
What does that 28x28 look like?
Q2- A 3x3 kernel we place it at top left of image and move it one step to right till end to convolve. Do we do same with 28x28 kernel to convolve?
In lenet architecture, 28x28 refers to the output of convolving 32x32 input with six 5x5 kernels with stride = 1. Please look at the output shapes in the model summary.
2 Likes