MultiCAM 20K 3D tracking lock-on redux

Hi all. A while back I posted a thread talking about this, and I've done some experiments since. I'm trying to get that list of feature requests together again, and wondered whether there are now enough D500/D850/D5 owners on the forum that I can confirm what I'm seeing (and if anyone knows a workaround). It'll take a few posts to show images, so bear with me. These are all obviously cell phone images through the finder. They were shot with a 70-200 clamped to a tripod under reasonably bright kitchen spotlights, and the results have been repeatable with different subjects (including outside). The D850 shots were set up with different buttons set to different AF area modes, so I could toggle between them.

 

TL;DR: On a MultiCAM 3500 (D700/D800/D810 etc.) AF system you can select 3D tracking and the camera will initially focus exactly on the selected AF point, moving off it only if the in-focus subject moves to another AF point. With a static subject, it behaves exactly like single-point AF. On a MultiCAM 20K (D850, D500, D5) 3D tracking appears to be free to pick a different AF point for initial focus acquisition, even with a static subject.

 

I'll show some images, then comment on what I think is going on.

 

Let's start old. D810, group AF:

 

 

The camera uses the points around the selected (central) AF point and picks the closest. Here, it's focussed on the handle of the front water jug.

And D810 with either single-point AF or 3D tracking, since both are identical in the viewfinder;

 

The camera can ignore the foreground subject and focus on the background (at the selected AF point), as intended. In 3D tracking, nothing moves, and as a static scene, 3D and single-point are identical.

D850 in single-point AF next. Again, the camera ignores the foreground jug and focusses on the requested AF point in the background.

On to the D850. Group mode again looks at the surrounding AF points (the middle point is again selected) and finds the closest one (again, on the handle of the front water jug).

Now, for comparison, D9 dynamic area. Note that although the helper points cover the foreground jug, the camera ignores them - it locks on the central AF point, and it would only be if the subject moved off that point that the surrounding points would be used:

Finally, 3D tracking set to "wide" area. The camera can look farther afield (I'd suggest as far as "group" looks), and wanders off around the front jug.

Note that the shown AF point is now off-centre - but neither camera nor subject moved.

On to the point of the thread. Here's 3D tracking on the D850, again starting at the centre point. Nothing moves, everything on a tripod, but the camera locks onto the side of the water jug (under, I believe, the "helper" AF points):

In this case, 3D tracking is set to "normal" area. It's clearly ignoring the background subject.

So what's going on? I think:

• In "normal" area 3D tracking, the camera looks for the closest point in the D9 area and locks on that for initial acquisition.
  
• In "wide" area 3D tracking, the camera looks for the closest point in the "group" area and locks on that for initial acquisition.
  

Why do I care? Three cases have come up:

• FTrying to focus on the eye of a subject who might move around the frame (hence I want 3D tracking). In 3D mode, if the subject is leaning, the D850 will consistently lock on eyebrows or glasses rather than the actual eye.
  
• Trying to shoot through branches (either birds in trees or for artistic framing of a background). There can be quite a detectable gap between where you think you're going to focus and what the camera tries to lock onto instead. 3D tracking should allow focus and recompose, and handles wind blowing subjects around - but only if you can focus on the actual subject in the first place.
  
• If you're trying to focus on a small subject on a ground plane, the camera will very likely try to focus on the ground in front of the desired subject. I first saw this trying to focus on distant sea otters with a hired D500, where the camera would consistently focus on some kelp in front of the otter. I assumed it was an AF fine tuning issue at the time, but since came to new conclusions.
  

Face tracking doesn't affect this behaviour in my experiments, for what it's worth.

 

For those with the affected cameras (D850/D500/D5), I'd love it if you could confirm this behaviour - it's pretty obvious in the viewfinder if you try to "shoot past" something. I bought Thom Hogan's D850 guide to check, but I don't believe this is covered, at least explicitly. If there's a workaround (other than live view or my current approach of assigning a button to select and initiate single-point AF, which fortunately you can do on the D850) I'd obviously love to hear it - otherwise I consider it a regression, and I'm definitely going to nag Nikon about it.

 

Thanks all. Sorry to give you homework.

Belatedly, and just to be clear: Those images show where AF settled using AF-C, after pressing a button to initiate focus. In all cases, the initial focal plane didn't matter - I can toggle between the above results just by pressing different buttons, so it's not (for example) that 3D tracking got confused because I initiated it after using Group AF.

The 3rd dimension in Nikon's "3D" is color. 3D tracking looks for being able to identify some sort of subject based on color contrast with the surroundings and tracks that as it moves in the frame.

 

If you have branches swaying about in front of the subject, the camera will be distracted in the multipoint modes. Use single point in this situation, or shoot through a clear foreground.

Edited February 5, 2020 by ilkka_nissila

I believe you just proved my assumption from the old thread that 3D tracking has a "close-focus bias" build in. Though one would need to eliminate any color influence to it to make sure.

 

The D500 does not have the d9 AF area mode - only d21. IF it behaves the same way as you describe above for the D850, then it might switch to a foreground subject even more readily given that the helper area is substantially larger for d21 than for d9 (assuming it doesn't use the d9 area internally even though it is not provided as a selectable option). I never use 3D tracking - my D500 is either in group AF mode or in single point, rarely nowadays in d21 (I probably would use d9 a lot more if Nikon provided it). I have single AF mode on the AF ON button and group AF on the center press of the joystick (with the occasional frustrating accidental moving of the entire group when I fail to press STRAIGHT down on the joystick).

 

Off topic - the behavior you describe is what irritates me so much on my Sony mirrorless bodies where I am much more relying on Auto AF modes but have to switch to single point to avoid the issues that you describe. Though for your type of photography, Sony's eye AF might be quite advantageous. Or, for that matter, the so far reportedly slightly less capable Nikon version in their mirrorless offerings.

 

They were shot with a 70-200 clamped to a tripod under reasonably bright kitchen spotlights, and the results have been repeatable with different subjects (including outside).

I see several potential issues with your selection of subjects here. For one, the white background subject is quite featureless and depending on the AF area mode, that single dark line might not be picked up at all. Secondly, the transparent handle of the foreground subject could also lead to focusing issues. Thirdly, as Ilkka pointed out - 3D includes color information - so to make sure that color isn't influencing the behavior, the foreground and background subject should be of the same color. For that purpose, having two high-contrast and well-lit "focus targets" at different distances to the camera might be the best option.

Are you sure the overlay in the viewfinder accurately aligns with the actual AF sensor array in the bottom of the dark-chamber Andrew?

 

I know there's a slight discrepancy in my Nikon DSLRs, which can be seen by moving the visible rectangle relative to a small feature and seeing if AF lock is still achieved.

The D500 does not have the d9 AF area mode - only d21. IF it behaves the same way as you describe above for the D850, then it might switch to a foreground subject even more readily

 

But the D9 (or the other dynamic area modes in Multi-CAM 20k cameras) doesn't switch to a foreground or background subject in the dynamic area if there is something focusable under the primary point. That's what Andrew is showing, it's focusing using the primary point in this situation. For it to use the other points, it has to lose focus under the primary point, basically, if you move the primary point to a featureless area, then it is likely that it'll find something under other points in the dynamic area. However, as soon as you put the primary point on the subject and there are adequate features to focus on, it'll quickly return to using the primary point. However, if there is a cloud of out-of-focus features in front of and around the subject, it can sometimes hesitate and wonder where to go. This is why it's probably best to use single point in such situations.

 

Since the D9 and single point focus on the point under the selected point in Andrew's examples, it doesn't seem likely that the markings are substantially in the wrong place, either. However, the actual AF sensors are sensitive to features in a somewhat larger area than shown in the viewfinder.

Edited February 5, 2020 by ilkka_nissila

Thanks for getting back to me, all.

 

Yes, colour could affect it - but the meter resolution is substantially higher than the AF sensors on a D850, so I'd not expect this to put it off. I've seen the same behaviour in the field (and just confirmed it) with bright colour differences (red mountains in the background, green backlit leaves in the foreground) as well as obvious luma differences. If there's a big enough gap between foreground and background then the AF may have trouble identifying anything in the OOF region, but in this case it hops quickly and accurately between them - I picked this example because there was a clear shadow line in the foreground and background, and I had a 1000lm spotlight pointing right at it in an otherwise well-lit (~6000lm total) kitchen. And I've seen this with relatively small AF misses, such as the "focus on the ground in front of the subject" and "focus on the glasses or eyebrows not the eyes" problems, and with larger misses, indoors in bright sunlight and under other lighting conditions. All cases where my D810 works fine.

 

As Ilkka says, while I'm sure the viewfinder alignment isn't perfect, it's absolutely accurate on my body to within the AF point. Indeed, by wiggling a bit I can get D9 to hop between points, and it's pretty much as indicated. Single point certainly works as expected. This is both on my D850 and on the D500 I hired a while back.

 

I also have problems pushing straight down on the joystick, and have been using fn1 to trigger single point AF - but this is annoying because I quite like having the front buttons available to mess with the metering (since Nikon's highlight-weighting doesn't actually preserve highlights). It's also annoying that you can't turn the joystick off and just use it as a button, leaving the multi-controller for AF positioning (which it does perfectly well).

 

So I can, and do, use single-point. But basically this means "3D isn't always fit for purpose because I can't get it to initiate on the subject" - and I lose 3D's ability to follow any movement. What I've seen so far is a clear regression compared with how I could use 3D on my D810 (or historically D800, or D700), where I could lock AF exactly on an eyeball or a bird in a tree and tracking would work fine if (and only if) the subject moves - and obviously the manual doesn't indicate that there would be a change. The joystick (which I don't actually find any easier to use than the multi-controller) and "track it yourself" isn't really a substitute.

 

Has anyone been able to repeat my results? I'm possibly doing something wrong (and there can be a little jumping around if there's a contrast issue or everything is too out of focus - but what I've seen is repeatable for me). I'd love to have an option that wasn't "give up on 3D". I'm finally putting my feature document together, and I'd like to know whether this is a non-issue, because it's pretty high on the list of things that actually affect how I use my D850.

But the D9 (or the other dynamic area modes in Multi-CAM 20k cameras) doesn't switch to a foreground or background subject in the dynamic area if there is something focusable under the primary point. That's what Andrew is showing, it's focusing using the primary point in this situation.

I was imprecise in what I wrote. As I understand Andrew, he showed that the d9 area influences the behavior when in 3D tracking mode and hence I assumed that the D500 would do the same - but with the d21 area (since it doesn't have a user-selectable d9 area mode - but could have one internally). If I can make the time, I may set up the D500 to try things out - but given that I have almost zero experience with 3D tracking mode, operator error looms large.

:-) Thanks, Ilkka. Any kind of second opinion would be much appreciated. I've always found 3D tracking useful (not least because it lets you focus and recompose without introducing errors from rotating the focal plane), and I'm surprised more don't seem to use it. Possibly including the Nikon engineers...

I sometimes use 3D tracking when photographing people. More often, I am using 9-point dynamic, though, as it is less unpredictable and gives me a slight room for error in the positioning of the primary point. 3D tracking doesn't work well if there are obstructions in the line of sight, and sometimes I'll just slip off the subject. However, when it works, it can indeed make life easier by allowing continuous recomposing. I don't have enough experience with it to know when it will fail and when it'll work, so I only use it for casual/non-critical situations. If I have to get the shot, I won't turn to 3D tracking.

Ilkka (or Andrew or Dieter perhaps),

A little of topic, but I've been trying to get my head around this for a while. From what you wrote upthread is the following a fair description of 9-point dynamic on the CAM3500 with AF-C.

1. The chosen focus point is used and focus is adjusted until such time as the detector is unable to find anything it can focus on.

2. At this point the camera decides to try an alternative.

3. My guess is that out of the surrounding 8 points it chooses the one with the least focus error. Is that your understanding?

4. The new focus point is retained until such time as:-

a) The original chosen point sees something to focus on at which point the camera switches to it. Goto 1.

b) If not a) and the present point becomes unable to see something to focus on then another one of the remaining 7 is chosen using the criterion in 3. above. Goto 4.

Richard, basically that is how it seems to work with Multi-CAM 20k bodies, but the 3500 series is a little different (I will explain the difference in the next paragraph). With regards to point 3 is that it also depends on how much detail there is that is contrasty enough to focus using, i.e. rather than select the auxiliary point which has the least focus error, it can prefer an auxiliary point with a lot of contrasty detail that is reliable to focus on.

 

The 3500 series modules seem to not return to focus using the primary selected point if any of the auxiliary points is in focus and corresponds to a closer distance than the subject under the primary point. So imagine you have a person who has a large mexican hat on their head, and you start by pointing the primary point on the subject's eye. the camera focuses on it. But if for a moment you lose the eye from under the primary point and there isn't enough detail on the skin around it, so the camera can then choose to focus on the hat using an auxiliary point and in this case, if you return the primary point on the eye, if the hat is still within the dynamic area, it may stick to retaining the hat in focus and not focus on the eye. This was fixed in the Multi-CAM 20k and it will return to the eye quickly after the primary point is returned to cover the eye.

 

The manual and sports AF guides always explained dynamic area in the same way, and the Multi-CAM 20k behaves as the manual description says, whereas the 3500 series dynamic area implementation doesn't quite do what is said. I think both systems function in the way the designers intended. The old system would help the photographer who is following a bird in front of a complex background and if the primary point slipped to the background, it would hold and be not too eager to focus on the background. But the new system follows the primary point selection quickly and only gives a moment's grace period where it will hold the subject in focus using the auxiliary points while the primary point is lost (e.g. pointed to the background or a smooth area of the subject where there isn't enough contrasty detail to focus on). So you can get some shots focused on the background with the new system if using dynamic area. Nikon do offer a solution, in the form of the group-area AF, which in AF-C focuses on the closest point within the dynamic area, so that can be used effectively to focus on subjects at longer distances, and you're reasonably safe from the camera focusing on the background. In the first implementations of dynamic area in the Multi-CAM 3500 bodies, there was no group area AF, so the dynamic area AF was implemented in a way that it held onto the subject and was protecting the photographer from slipups where the focus would go on the background too easily. However, the drawback in the older system is that in the mexican hat case it will hold onto the hat and not return to the eye. So in practice with Multi-CAM 3500 one has to use single point more often in such situations to ensure that the focus does go where requested, but they can use dynamic area more easily to photograph bird in flight in more complex backgrounds.

 

Finally, there are parameters where the photographer can describe how long the camera should wait when an object occludes the main subject before focusing on the new subject, and how erratic the subject movement can be etc. These tweak the system's behavior.

 

I hope this rather complex explanation is helpful.

Edited February 6, 2020 by ilkka_nissila

Thanks for that Ilkka. I was just able to confirm what you say using a picture frame on a blank wall (simulated eye) and a table lamp in the foreground just below (simulated Mexican hat). I focus on the edge of the frame and it's sharp. I shift to the blank wall to the right of the frame and the lamp underneath, in the foreground, is now in focus. I shift back to the picture frame but the lamp stays in focus with the frame out of focus.

 

It seems therefore that after initial focus the selected starting focus point is treated just the same as the other 8.

 

Interestingly on my D7200 if I turn display on of focus points I can see the expected behaviour as the focus point changes even though this isn't shown in the viewfinder.

 

Can the D850 show the 'helper' focus points if you turn on this display? If so would that help Andrew discover what's going on? Capture NX-D can show the focus points too - I don't know about non-Nikon programs.

The camera can show you the chosen focus point in image review (which I'm embarrassed to say I've just realised you can trigger right-handed; shame you can't zoom, other than toggling). In the viewfinder, though, the dynamic modes always show the centre point you selected and an outline of the surrounds of the area - not which point is actually being used, which means I have no way to tell that I've nailed focus when I'm in relatively dim conditions and the degree of focus miss is small. It also doesn't (I believe) use colour information like 3D tracking does, and without information about the focus location I'm wary of the wider dynamic modes.

 

To give the (common for me) situation why I care, and why 3D as implemented on the MultiCAM 3500 was ideal for me: I shoot photos of people playing tiddlywinks - see the splash image on www.cutwc.org. This means there's a table (covered in white felt in direct lighting) with small, dark but coloured plastic discs on it, often in a room with iffy lighting and piles of dark ugly chairs in the background that may have reflective bits of chrome on them - so I depend on wide apertures, hate LoCA, and wish there was a Nikkor with a decent variable apodization element that would stop reflected highlights turning into bright circles with sharp edges. But I digress, other than to point out that my depth of field is under 1cm.

 

People play shots by leaning over the table and stretching out (typically you play towards yourself), meaning the face is in shadow and I'm typically at an angle to the face, making the eye farther from me than eyebrows and/or glasses (and the AF point indication in the finder typically blocks some of the important bit of the image). I usually want the eye in focus. I want to frame the shot so you can see what's happening on the table, and often the reactions of the other players around the table. As people line up the shot, they settle in slightly different positions, and they move their heads. I have enough to do running to the side of the table where an arm isn't in the way, where there aren't people standing between me and the table, and where I'm not running into any other table in the room; there's no indication of exactly when a shot will be played (I usually take the photo just after so the shutter noise isn't distracting, which gives me a fraction of a second before the player stands up again) so the chances that I can get an AF point manually locked on the eye are small. I've been doing it with the D850 by framing wider, but I'm wasting pixels and losing depth of field that way, and it's still only as good as my ability to follow the subject. With a D810 I can 3D tracking lock on the eye and then worry about framing separately, knowing the camera will lock the focus until I'm ready to shoot.

 

So a 3D tracking mode that I can't get to lock exactly where I want it to is annoying. D9 is a start (I should try it), but any significant head movement is going to cause the camera to get lost, and I can't be sure whether focus is in the right place.

 

For what it's worth, highlight metering also does a terrible job of preserving the mat colour when the focal point is on something darker. Basically the thing I shoot most is pathological for every design decision Nikon has made. I'm not yet quite willing to switch to Sony and see if that fixes it, though.

chances that I can get an AF point manually locked on the eye are small

How about hire a Z6/7 for the weekend and see how Eye AF works here?

How about hire a Z6/7 for the weekend and see how Eye AF works here?

 

I could, although I couldn't necessarily afford the trade in right now. Annoying that the D850 hardware is presumably perfectly capable, since the D810 is. For so long as the D850 can do 14-bit full resolution images at 9fps with autofocus and a useful finder and the Z7 can't, it's not like I plan to switch yet anyway.

The camera can show you the chosen focus point in image review (which I'm embarrassed to say I've just realised you can trigger right-handed; shame you can't zoom, other than toggling)

Does this image review confirm your hypothesis that it's the 'helper' AF points that are responsible for the unexpected behaviour? Not having a D850 I don't know just what it shows. On my D7200 (CAM 3500) image review shows the focus point used for 9 point dynamic even though the viewfinder only ever show the initial selected point.

the D850 can do 14-bit full resolution images at 9fps with autofocus and a useful finder and the Z7 can't

Yup, if I go Z, it's the Z6 for me at the mo' mainly because of the video and fps.