>>The methodology is not about testing the efficiency of >the >>whole AF system in real life situations. It is about >testing >>wether or not, on the simplest possible target, phase >>detection produces front or back focusing. > >I understand your approach. What I'm suggesting is that I >believe the simplification is largely counterproductive >because I think it inherently prevents the chosen sensor >(which is part of a larger system) from functioning most >effectively (as a fundamental part of the larger system) for >the application dictated by your methodology. Some AF system >experts should chime in here I think. > >>If the camera is not successful at properly focusing such >a >>target, it's a clear sign there is a problem. You cannot >hope >>to get good results in real life situations like you >describe >>if the camera cannot even handle ideal conditions. It's >just >>step zero in the troubleshooting strategy. > >My point is that the condition you set up in your methodology >is not in fact ideal. I believe that isolating a target to >provide, effectively, a single data registration point for the >AF system, does not reflect the engineering or software design >approach of the AF system. The AF system is designed to work >best, as I understand it, when provided with significantly >more target and proximal data than is provided when using your >method.
This is only true in the multi-point AF modes. In single-servo, single-point AF the only data that is important is that which is under the selected focus point. If the camera can not perform well in this mode on a target with good contrast (in the proper plane), there is no reason to believe it would perform well in "real-world" situations.
While the camera may achieve better results on still subjects using single-servo, dynamic-area or continuous-servo, dynamc-area or 3D-tracking when the single-servo, single-point results are poor, this is completely beside the point.
If all of the focus points did work well in single-servo, single-point, I would bet a lot of money that the performance in continous-servo and dynamic-area modes would improve greatly.
IMO it's a ignoratio elenchi to say "yes AF-S single-point doesn't work well in test scenarios, use AF-C dynamic-area in the real-world."