Why can the inputs in computer vision problems get huge? Explain it with an example.