What trade-offs do you consider between accuracy and speed in vision systems?