Comment by techbio

I also observe this bias, I imagine it is an artifact of > versus >= on the pixel boundary.