Stride Convolution

For $n\times n * f\times f$ with $p$ and strid $s$ then the output is

$\lfloor\frac{n+2p-f}{s}+1\rfloor\times \lfloor\frac{n+2p-f}{s}+1 \rfloor$

so in the above example, $\frac{7+0-3}{2} +1 = 3$

Technical note on cross-correlation vs. Convolution

In mathmatics, the filter is supposed to be rotated 90 degrees and flipped as shown below for convolution operation.

Technically, what we’re actually doing in the convolution neural network, the operation we’ve been doing is called cross-correlation($\star$) instead of convolution($*$). But in the deep learning literature by convention, we just call this a convolutional operation.