D800 owner tests continue to pour in (again, URLs please, not attached images). As they do, one thing is clear: focus testing isn't easy and many of you seem to think you can improvise on my instructions (don't).
Let me reiterate why we're testing cameras and why we're testing them one particular way: Nikon programmed some cameras incorrectly. Thus, we want a test that checks for that single problem first and foremost, and that's what I proposed in my original article on the problem, which I've updated with a couple of additional notes. We are looking for a very simple pattern: out of six images (three live view, three phase detect), one may be very, very different than the rest. If you find that result using my test, your camera is potentially a candidate for having its internal focus sensor table reprogrammed. If you don't find that result, then your camera likely doesn't have that problem (it might have another problem, or it might have no problem).
What's happening is that many people are testing in ways different than I suggest, with different targets, at different distances, and with different techniques. The problem with that is simple: you're not testing for the thing we know is wrong with some number of cameras, and you're not isolating out that variable well enough to discern if your camera has the problem. As one reader pointed out to me, some of what we're seeing on the Internet related to the D800 focus issue must certainly be attributable to the Dunning-Kruger effect: a situation where an unskilled person makes poor decisions and reaches erroneous conclusions.
So let me enumerate for a moment. Any number of things can cause focus problems. Here are just a few:
- Misprogrammed tables. This is the thing we're looking for. Some number of cameras were misprogrammed at the factory. The problem manifests itself as a clear left side focus miss for phase detect autofocus. When tested for as I suggest the result is obvious if you have the problem: you'll have five "good" or "near good" results and one that is clearly "really bad." If you find that result, that's when you contact Nikon to see what they say and how to return your camera for correction if they agree you have the problem.
- Misaligned mirrors. There are two mirror adjustment points that position the main and secondary mirrors so that light falls in the correct path. Misalignment of these can cause side to side or top to bottom errors. But those errors look different than those from a misprogrammed focus position table. At the risk of oversimplification, you might have "bad", "less bad", and "good" results on the three phase detect tests (there are other possibilities).
- Misaligned mounts. The lens mount itself needs to be absolutely parallel to the image sensor. Again, you might get side to side issues, but the results again look different than from a misprogrammed table.
- Misaligned lens elements. Again you'll get side-to-side variation, and this variation may have sub-variations in it when zooming or using different focus distances.
- Misaligned sensor. Image sensors are shimmed to be parallel to the lens mount. This problem would look very much like a lens mount alignment problem.
- Field curvature. This one is tricky. If you focus on the left side, the center may not be in focus due to field curvature on some lenses. This is one reason why I asked for six different samples. The Live View samples should show the field curvature, so we'd have something to compare the phase detect samples to. In other words, if you focus at the left and the center phase detect and Live View samples match, I'm not overly worried about whether they are in perfect focus or not, only that they match.
- Focus shift. Even more tricky: if you use anything other than the maximum aperture (e.g. stop down), internal focus shift can produce results different than you'd expect and depth of field may or may not compensate for that (depends upon the lens, distance, aperture).
- Non-flat, improper target. Yep, even trickier. The reason is that the autofocus sensor you're testing is actually composed of multiple sub-sensors. If you have a curved target (real object instead of flat, parallel test target) it becomes a real gamble in terms of where within the overall sensor the camera decides to focus. It's a bit of "majority rules" but "closer is sometimes used instead" logic that Nikon has never explained. I proposed a flat target with detail only in the direction I wanted to test in order to eliminate this variable.
- AF Fine Tune. In theory, lenses ought to group in a bell curve around 0, thus my suggestion not to use AF Fine Tune for testing for the left sensor problem. If you saw different results between Live View and phase detect across the board, then perhaps doing an AF Fine Tune would be appropriate before doing the test I proposed. The problem, however, is that AF Fine Tune, especially not done correctly, can mask the problem we're looking for.
Yes, there are even more variables than those. I was specific in my suggested test in an attempt to isolate out the one thing we were looking for. Even that wasn't perfect, but it should be good enough to get to a simple answer: was my camera's focus sensor table programmed correctly at the factory?
One problem that is now obvious is that people are testing in ways they never have before. And they're seeing things they've never seen before, mainly because they never looked for them before ;~). Coupled with the increased sampling of the 36mp sensor, people are seeing things that may have always existed, but have been hidden by the lower resolution of the system. As I noted when the D800 was announced: you'll see just how well your lens performs. You'll see how well the focus system performs. You'll more easily see subtle mis-alignment.
With a few people I've had them perform the following (and often revealing test): photograph the same thing with your 12mp camera and your 36mp D800. Now downsize the D800 images to 12mp. Hey, wait a minute, the problems went away! No they didn't. They were always there; we just now have a high enough sampling to see them.
But this is a bit of a diversion. First and foremost: some D800 bodies apparently were programmed wrong. You probably want to know if yours is one of them. That's the whole (and only) reason for the test I suggested. The results on a misprogrammed camera are generally not ambiguous. If your camera tests "positive" following my directions, your next step is to contact Nikon to see about getting it fixed.
As I noted in the next story, at some point I need to write some articles and expand sections in my books that help you do other testing and what different results may mean and what you should do about them. But that's not going to happen today, so be patient.