However, I'm wondering what other people use for the error-metric for selecting levels? I currently have hard-coded values for the distance from the camera where each level switches at. This is pretty much the fastest method as the distance is an approximation and all it needs is one or two compares per object. However, I'm wondering if I should scale the cut-off distances based on the object's size? Obviously this would introduce more calculations into every object draw, but might make the choice of LOD level more accurate.
My current hunch is that the hard-coded values are fine and I can just have a "geometry detail" setting in the game whereby people with nicer/older hardware can move these distances closer/further away. Plus I'm not keen on introducing too many calculations into each draw call as I'd ideally like to have *many* objects floating around in view and too much fiddling will end up costing me FPS.