Just goes to show when you're doing science you need to test and validate your experimental methodology, including the tools you use. In computer vision, its co

Just goes to show when you're doing science you need to test and validate your experimental methodology, including the tools you use. In computer vision, its common to need to do some kind of calibration for many algorithms which can usually reveal some kind of statistical error or problem. I wonder why none of the researchers thought to do some very simple validation of the data? And I wonder if the software was at one point correct and then this bug was introduced at a later point? Many times it feels like after a company does a formal scientific validation they never do it again despite the fact they have engineers constantly working on the code...