The actual concept isn’t difficult to implement (z+1? Shift it over a pixel and fill with an appropriate noise/blur function, then shift back for z-1)

the hard part is building the depth map, especially when you’re using higher resolution. You’re essentially doing 3D rendering at that point