[fnordpiglet]

Actually I would expect AI up scaling of SDTV in this case would perform better. It would assume semantically the props were real and would extrapolate them as such.