The example code is storing encoded absolute UV positions in an R16G16_SNorm. It’s enough precision to handle full screen 4k outlines with some modest amount of anti-aliasing precision.
But if it was changed to store relative pixel offsets and limit yourself to less than 126 pixel wide outlines with no AA (encode with
(pixelOffset + 126.0) / 255.0), or maybe 31 pixel outlines with some AA (
(pixelOffset + 31.0) / 63.0), then you could easily use an 8 bit format to store the pixel offsets. Otherwise you could also encode a value with a greater precision / range across two 8 bit channels.