(commentary)
Yesterday I wrote that the real issue that needs addressing in cameras is focus. Here are some specifics on where I believe Nikon is missing the boat with focus on DSLRs:
- Closest Subject Priority. Remember this? It was on all our film SLRs and a few of the early DSLRs. Curiously, “back focus” complaints didn’t really start becoming common until after Nikon started removing this function from the cameras ;~). The issue for the focus system is simple: what’s the subject? If the camera knew what (and where) the subject was, it would have a far better chance at focusing on it. More often than not, the subject we’re shooting is the closest thing in the focus area. Not always, so we need ways of dealing with it when it isn’t. But we need CSP back. Obviously, the Focus Tracking With Lock-On still needs to be there so we don’t get sudden unwanted shifts, but that function needs to not be buried in menus (see #8). Moreover, why can’t I have Close, Far, and No Subject Priority? The number one problem in focus is identifying where the photographer wants the focus, why not let us help?
- AF Sensor Shape and Size. Right now, our sole way of telling the camera where to focus is via 51 small indicators in the viewfinder. Only one problem, those indicators are a complete and total lie. The actual area being used is neither the size nor the shape of the indicator. Heck, the viewing screen with the indicators can be misaligned, too. Thus, the photographer can think that they have selected one thing and the camera think something else. This needs to be fixed.
- AF Fine Tune is a mess. I’ll get to one of the reasons why in a moment, but Nikon has neither documented how to use this function to its fullest, nor do we have all the correction we need. Zoom lenses are a real problem, as they often need different parameters at different focal lengths. Nikon needs to document what’s going on, how to optimize it, and provide additional capabilities if this is going to really work to fix the manufacturing tolerance issues. And dare I point out that we have no way of actually getting to the focus sensor position tables, which was at the root of the D800 focus problem?
- Lenses and cameras aren’t consistent. After testing hundreds upon hundreds of cameras and lenses during the D800 focus fiasco, what I found was this: not a single Nikon item I tested fit into a predictable bell curve other than the f/4 zooms. The 24-70mm results were as random as it can get. This indicates that manufacturing consistency isn’t there. The further we get from the “expected” position for everything (AF Fine Tune value of 0 is a stand-in for that) the fewer samples we should see. But in the case of the 24-70mm, I saw as many samples that were +20 as 0 as -20 as any number in between. Random. As I noted, the 16-35mm, 24-120mm, and 70-200mm f/4 all fell into a decent bell curve centered on 0. As did every Canon camera/lens data set I’ve seen. So it can be done. Just that Nikon didn’t do it with every lens. So Nikon needs to nip this in the bud right now and make sure that all subsequent camera and lens production does exhibit a predictable bell curve response.
- Repeatability. I’m getting more and more sure that most Nikkors aren’t particularly repeatable in focus. In other words, if I were to focus on the same thing over and over, I wouldn’t get the same exact focus position twice. That’s actually true of Canon DSLRs and lenses too, but not nearly to the same level as I’m seeing with Nikon lenses. The DSLR focus systems have a “tolerance” to focus. They’ll report “focus achieved” for a small range of distances. That’s been true since the early days of film SLRs with autofocus, as was documented by the late Herbert Keppler in Popular Photography decades ago. My sense is that Canon has achieved better repeatability than Nikon in recent years. My suspicion is that Canon’s manufacturing tolerances are tighter. We shouldn’t be satisfied with the current tolerances: we want more precision, especially since we have far more resolution now to see what is actually happening to the focus plane.
- Repairs shouldn’t move focus positions. Well, okay, the D800 repair should have changed the table that contains sensor positions. But the most common result of a Nikon D800 focus repair was that the camera shifted to +10 overall on AF Fine Tune. See #4. If the lenses aren’t in a bell curve, moving the camera so far means that you probably can’t tune all your lenses any more. My theory is that the camera/lens that Nikon used to create the repair target was not centered on the bell curve, but +10. Which means that they’re not paying attention to the bell curves.
- Tracking needs to be better. Thing is, all 51 AF sensors could be reporting tracking data continuously without shifting the focus. Moreover, because we know which direction and distance those sensors are reporting, we should be able to form conclusions about subject just from that data. Take the case of bird in flight directly or diagonally towards the camera. At very long distances the bird would be triggering “coming towards” tracking data on only one or a few sensors. As it got nearer, the we’d get more and more sensors reporting the same thing. At some point, just from that data I’d be pretty sure what the subject was ;~). Just as Nikon built a data set of 300k samples for its matrix meter, I believe that you could build a similar data set for focus. Indeed, with the color information from the secondary sensor, this should be a remarkably rich data set. I see three options here: use tracking intelligence, force tracking to item detected on current/adjacent sensor, ignore tracking. Again, this need to be under user control. We kind of have this with the current AF-C settings, but there’s something slightly wrong with Nikon’s tracking assumptions, IMHO.
- We don’t have good real time controls. I’m actually a fan of the change from the rear switch for AF options to the front button (plus command dials). Only problem is that the darned button is a bit awkwardly located and too small. I almost always have to “feel” to find it. That needs to be fixed. Moreover, some cameras with a great focus system—I’m looking at you Nikon 1—have pathetic user control over focus buried deep in menus. Most of us serious shooters use the AF-On button for a reason: it allows us to tell the camera “follow the subject I’ve selected” or “don’t follow a subject”. Most of the controls we’re given are geeky and obviously designed by engineers and don’t directly solve user problems. Users want to enable/disable focus, jump focus to another spot, change the manner in which focus is detected, and so on. Heck, we could even tell the camera what we’re shooting (human movement, birds in flight, trains/planes/automobiles, etc.). We could even subcategorize that: linear movement, mostly linear movement, erratic movement. But we have no way of telling the camera what we know about the subject. None. Zero. No interface detected. And then we have very little fast direct control to reconfigure the system when it changes.
- Nikon’s documentation and accessibility sucks. Back in the film days I was able to talk to Nikon engineers about how the focus system worked. When I did, I learned things about the system that weren’t in any Nikon literature. The F5 “special” custom setting focus mode, which prioritized left-to-right detection order was actually quite useful for some subjects. But you wouldn’t have learned about it from Nikon’s manuals. At least not learned anything useful about it. Give a half dozen of us third-party Nikon documenters a day with the autofocus engineers to ask questions and get answers and I can guarantee you that your camera will focus better the next day ;~).
- Focus Shift and Field Curvature need to be addressed. Is there a table in the camera that addresses focus shift (common on fast lenses)? I don’t think so based upon experience. If there is, it isn’t accurate. Likewise, what do we do about lenses with field curvature? The focus system can’t perform optimally if the lenses then take that optimization away. Nikon hasn’t rationalized this across the DSLR camera and lens offerings. Switching to Live View to confirm or dial in focus isn’t a solution (and especially true on the highest resolving Nikon DSLR, the D800 ;~).
When I wrote yesterday about not needing more megapixels on sensors, one of the things that prompted that comment was focus. D800 users know what I’m about to write: if you can’t accurately place the focus plane to exactly where you want it, you don’t get the resolution that 36mp suggests. You may have even seen comments on the Internet (or DxO numbers ;~) where people claim that the 36mp camera is “resolving more like a 24mp one.” Focus precision and accuracy is more necessary the higher the resolution of the camera.
The thing about most mirrorless systems is that they use contrast detect to do focus or to verify a phase detect focus position. Contrast detection is highly accurate (assuming the right subject is in the autofocus sensor position). Phase detection is very fast and understands subject movement directly. The two together can be used to create both a fast and accurate system. Separately, they each have their weaknesses. The X-T1 and E-M1 I was using for the past two weeks had different weaknesses in that respect. My sense was that the E-M1 was faster to initial focus, less reliable on moving subjects, while the X-T1 was slower to initial focus, more reliable on moving subjects once focus was achieved. Neither is better than the other, they just have different optimal uses.
Phase detect on imaging sensor focus systems have one big weakness, though: the math precision needed is getting out of hand. I’ve written before that basic phase detection put on the imaging sensor is variable with sensor size. Why? Because the “depth” between the splitter optics and the detector is very small in the image sensor type of phase detect system, which gives you very small discrimination numbers. So small, that as the DOF decreases due to larger sensor size (CoC increases), you don’t get a lot of precision. At least for the current implementations. This is one reason why the Nikon 1 does so well with phase detect on sensor and the Sony A7 doesn’t. Certainly there will be attempts to address this issue—incremental engineering is a way of life in the hardware world—but at the moment we’re in that “not quite there” world for most mirrorless systems using phase detect on sensor. Canon has a unique design that helps here, but we’ll see more of that soon, I think.
Which is one of the reasons why I wrote that focus is the thing the camera makers need to be spending more time and effort improving, not sensor quality.
Put simply: I don’t care if I have 10mp or 100mp, if the system didn’t focus right I don’t have a picture.
