Choosing Local Model for Vision