Does it ever make sense to increase the number of mipmap levels (e.g. maybe have 3x, 5x, 7x downsampling rather than just powers of 2) instead of interpolating between two of them?