Should you normalize RGB values by 255 or 256?

The article examines two methods for converting 8-bit integer pixel values to floating-point for image processing: dividing by 255 (the standard approach used by GPUs) versus dividing by 256 with a 0.5 bias offset. While the division-by-256 method theoretically distributes values more evenly across the [0,1] range, the standard division-by-255 approach is more practical because it maps black pixels to exactly 0.0, allowing algorithms to work with universal color values rather than being tied to 8-bit-specific constants.