When applying cross-correlation with padding, the spatial dimensions of the output increase. Specifically, if an input tensor of size $n_ extrm{h} imes n_ extrm{w}$ is padded with a total of $p_ extrm{h}$ rows (divided roughly equally between top and bottom) and a total of $p_ extrm{w}$ columns (divided roughly equally between left and right), and then convolved with a kernel of size $k_ extrm{h} imes k_ extrm{w}$, the resulting output shape is given by the formula: $(n_ extrm{h}-k_ extrm{h}+p_ extrm{h}+1) imes(n_ extrm{w}-k_ extrm{w}+p_ extrm{w}+1)$.