Anything you can do, I can do better (No you can't)

Many of the comments in this paper apply to any scientific domain and are not unique to computer vision, but some other research domains have well-defined methods for evaluating research (e.g., in medical research, does it help the patient?). I make several assumptions about the nature of computer vision research: it has a lot of small pieces; by itself each piece is meaningless (by itself computer vision is meaningless); no one has solved all the problems; many real problems are solvable with current techniques; a perfect system will never exist (human vision is not perfect); and people normally see what they “want” to see rather than what is there. I will assume that everyone is honest, but I also assume that people work under some pressure for results or publications and do not want any additional nonresearch work. First I will present a fictional story to ilhtstrate the nature of the problem, followed by an outline of the problems that arise. Then I will give the results of an informal survey of published papers to see how they contribute to the problem. The paper will close with some suggestions for improving the current situation.