This is the 12th post in a series of Nikon D850 tests. The series starts here.
The D850 has a tuning feature for the phase detection autofocus (PDAF). For the contrast detection autofocus (CDAF) no tuning is necessary or possible. The PDAF tuning works like this:
- Manually focus on a target at your chosen aperture and distance.
- Tell the camera to autofocus on that target, then recommend a correction
- Store the correction in memory for that lens (this is done automatically)
I’ve had a D5 since they first shipped, and thus I have experience with a similar system. Sounds neat, right? Nobody likes to tune the autofocus manually, and many don’t bother. But Nikon recommends that you don’t use the feature unless you’re having problems. Why is that?
It turns out that there are two sets of random variations that can cause all kinds of trouble.
- Errors in manual focusing — your errors
- Errors in autofocusing, which the camera needs to do to estimate the correction — camera errors
The combination of the two sets of errors means that you need to perform the procedure many times and average the readings for best results. If you do the auto-tune just once or twice, you are likely to make things worse, not better. I believe that’s what Nikon wrote their cautionary note.
Let’s talk a bit about the errors in turn.
The D850 is easier to focus manually than its predecessors, the D800 (which was awful), and the D810 (which wasn’t bad). In any of the three cameras, you need to use live view to focus critically; using the viewfinder is at best approximate, especially since all the focusing aids have been removed in the transition from manual-focus-only film cameras to autofocus-primarily digital ones. I for one am very sad to have experienced the demise of by far the most accurate reflex focusing aid, the aerial image ground glass. For those too young to remember, in the olden days when dinosaurs roamed the earth, SLRs had interchangeable ground glass finder inserts, and one of the options had a clear center with a very fine cross embedded in it. You got the image approximately in focus with the ground glass on the periphery, then moved your eye up and down or from side to side. If the image was out of focus in one direction, it moved one way with respect to the center of the cross. If it was OOF the other way, it moved the other way. RIP, aerial focusing. It was great, even though it only worked on-axis and was slow at best.
But I digress. Where was I? Oh, that’s right: liveview. The D800 was immensely handicapped by line skipping sensor readout. The D810 improved that. Neither offered focus peaking. The D850 has still better liveview clarity, and adds peaking. What’s not to like? Two things:
- The magnification doesn’t go high enough
- The peaking is too weak when used at full magnification.
Insufficient live view magnification is a common complaint. I don’t think I’ve ever heard anyone say their camera had too much magnification for critical focusing. Too-weak peaking is not a problem I’ve ever had with a camera; most have peaking that’s too sensitive. But even with the peaking set to its most-sensitive position on the D850, with the finder zoomed in all the way on a Siemens Star with a sharp lens like the 105/1.4, I can’t get the peaking to light up. It looks like Nikon tuned the peaking for low-magnification use. So, on balance, I find that I can’t manually focus the D850 as consistently as either the a7RII or the GFX-50S. Both of those cameras have their manual focussing issues, but I’ve found ways to get good results from them. It’s kind of strange that the D850, with a better LCD display at magnification than either camera, doesn’t deliver the godds as well. And of course the D850 can’t be focused through the finder at all in live view, which is the best way to use both the GFX and the a7RII.
Oh, and don’t be tempted to use the misfocusing arrows in the D850 finder when you’re dong the manual focusing part of auto-tune. They are driven by the same information that drives the PDAF itself.
Now we get to the variations in PD autofocus. In the D850, they are profound:
There has to be a better way to set up your PDAF than running endless variations of Nikon’s auto-tune procedure. I have some ideas, and I’ll get to them in a future post.
Hi Jim,
Very interested in what your ideas are for AFT. I’m currently about to go through that “exciting” exercise of setting up PDAF on a new D850 body (I had a D850 from day of release but it had to be returned due to dust in the optical viewfinder path). I concur (from field experience) that PD autofocus is just a crapshoot. Well, maybe I’m overstating a bit. You’re nicer and correctly call it variations. My practice when I shoot handheld and close or wide open is to take 2 or 3 shots while manually defocussing in between each autofocus activation. Looking forward to more on the D850 and the A7R3 when you get one in your hands!
I used Steve Perry’s method where you throw out the high and low outliers and avg the rest. My solution to avoiding focus error is to run the HDMI out to a large monitor and then look for small defects of my test target near the ROI and flat to the target. Like a pencil line, a speck of dust or a kind of orange peel when the target is critically focused. I defocus in both directions between shots and recompose.
I am very satisfied with the results I am getting from the D850.
Yes, a big monitor certainly helps. I’ve ordered a tray for my camera stand. Using the D850 has caused me to completely rethink focusing strategies, though. More on that in a future post.
A magnifying optical loupe pressed against the LCD works well and is a much more portable solution.
Right you are. I like the Hoodman one.
Jim,
The camera doesn’t autofocus for the auto fine tuning feature, at least not in full sense of the word. It’s only measuring the phase differential for the user-supplied manual focus and storing the delta as the AF tune value. The measurement of the phase differential is subject to error, which is why the procedure must be repeated several times and averaged for a given instance of a user-supplied manual focus. This is also why my DotTune procedure requires the user to engage the viewfinder dot confirmation for several seconds (and multiple times).
Regarding user focusing error for the procedure, this can be eliminated if a 3D target is used, such as a Lens Align target. This has the added benefit of allowing a specific front/rear DOF balance to be baked into the AF tune value. For example, if the user wants more rear DOF than front for portraits to make sure the eye is in focus when focusing on other parts of the face such as the nose.
That’s what I meant, but I was saying it in too compressed a way. Now that you mention it, the way the camera does the AFFT eliminates only the lens positioning error. If it uses the same mechanisms that it uses for Focus Shift Shooting, that error is small.
AF tune (be it manually determined by the user or with AFFT) eliminates phase differential sensing errors that arise from misalignment along the optical path to the AF sensors relative to the path to the imaging plane on the image sensor. Potential errors on this path include the lens (decentered/tilted elements), camera mount (tilted), mirror subassembly misalignment, and finally misalignment of the AF sensors themselves. AF tune (on cameras anyway) isn’t designed to account for lens positioning errors – the semi closed-loop feedback mechanism of PDAF is supposed to handle that, although the speed demands of the system may allow last-step errors to go uncorrected.
I understand how you could do this manually tuning the PDAF, but I don’t see how this would work with the automatic feature.
The automatic tuning feature operates on the same principle as manually tuning. The AF tune value adjusts what phase differential the camera interprets as “in focus”. This allows the user to not only correct base focusing errors but also bake-in a user preferences such as DOF balance and also coarse focus shift correction.
For Canon and Nikon DSLR’s, there is the FoCal software that automates the fine-tuning process. It actually analyses pictures and averages over a good many focusing events. The newest version doesn’t yet support the D850. I’ve had mixed results with the D810. The recommended correction values bounce around a fair bit between sessions.