Equation 3.2 we said in an off-hand comment that the size of the convolution filter had no impact on the number of applications when using the Same padding. Explain how this can be.

The decision on where to stop is called padding because when going horizontally with Same padding, by the time we stop we have to be using "imaginary" pixels. The left-hand side of the filter is within the image boundary, but the right-hand side is not. In TF the imaginary pixels have value zero. So with Same padding we need to pad the boundary of the image with imaginary pixels. With Valid padding almost never do we need actual padding since we stop convolving before any part of the filter moves beyond the image edge. When padding is required (with Same padding), the padding is applied to all edges as equally as possible. Since we are going to need this later we give the number of patch con- volutions we apply horizontally for Same padding: Tin/sh] (3.2)