Multi-CAM 20K and 3D tracking initialisation

Hi all. I think I mentioned this one before, but I wanted to ensure I had a definitive answer before I start putting it in a feature request list. Apologies if it's known information, but it felt like a Friday afternoon question...

 

On my D810, I normally shot in AF-C (with AF-On focus activation) with 3D tracking. I could position the AF point where I wanted to start, then if I or the subject moved, I could usually rely on the camera tracking the subject. I could also focus and recompose within the AF area, and have a reasonable chance that the AF point would stick to the subject.

 

I've found a problem doing the same thing with my D850: if there's something nearer to me, and near but not under the selected AF point, the camera front-focuses. For example, if I shoot through a gap in some branches, even if the AF point is clear, the branches tend to be in focus. Likewise if I try to focus on someone's eye and they're at an angle, I tend to hit focus on their eyebrows (in circumstances like this).

 

My suspicion is that, instead of just being used to help with tracking a moving subject, the unselectable "helper" AF points of the Multi-CAM 20K are being used to acquire initial focus and overriding the actual AF point I've selected. I've been working around it by assigning a front button to single-point AF (which does focus in the selected location), but obviously I then lose the advantages of 3D tracking. I'd quite like to be able to tell the AF system to focus exactly where I told it to, before tracking starts.

 

Using D850 terminology, option a4 (face tracking) doesn't seem to be relevant, and a5 (the 3D tracking watch area) is set to "normal" - set to "wide" even the adjacent visible AF points can get selected; I guess I want a "narrow" option under a5 that just uses the original AF point for initial acquisition.

 

Experimentally, if I place the selected AF point over a background subject with a foreground object nearby (not under the AF point rectangle, but closer than the adjacent AF points), I can toggle between the foreground and background objects according to whether I'm using single-point AF or 3D tracking, without anything moving.

 

Are others with relevant cameras (D5, D500, D850) seeing this behaviour? Am I doing it wrong? Obviously the D810(/D800/D700) never had this problem, because they didn't have "helper" AF points.

Sorry Andrew, I can't help with this one - I have never even tried the 3D tracking option on any of my cameras. I did compare the auto-area AF on the D500 with that of the D810 - the former was delivering a passable performance, the latter was just simply unusable.

Andrew. I think you have to remember that there's a total disconnect between the prism viewing screen and the actual AF 'rectangles' (that's if they're even properly rectangular).

 

The rectangles that light up in the optical viewfinder are on an LCD membrane in the prism-housing that's only approximately aligned with the AF module stuck in the base of the mirror box.

 

Of course, if the same thing happens in LiveView, that's a different matter.

I first assumed there was a systematic error in viewfinder alignment - but single point AF is perfectly aligned. That I can point at a static subject with single point AF and hit it consistently, but not a nearby closer subject with 3D tracking suggests the neighborhood issue. I've got Pv set to single point AF and AF-On to 3d tracking, so I can toggle focus distances without moving the camera. It seems consistent. Unless there's something weird about the camera or I'm misdiagnosing. It's basically the difference between being able to eye track and getting nose focus, which is annoying.

 

Agreed that auto area on the D850 seems nearer to useful than on older bodies.

Which is the Nikon focus 'mode' that selects the closest target?

Which is the Nikon focus 'mode' that selects the closest target?

Auto-area AF and group-AF. I wouldn't be surprised if 3D has a "close-subject" bias build in as well.

I wouldn't be surprised if 3D has a "close-subject" bias build in as well

 

That was what I was wondering too.....:-)

Ironically, in my experiments, auto-area seems just as happy to find a background subject. Group AF is a bit front-happy. D9 seems to lock on the centre point and only use the surrounding points to retain focus, as it says on the tin. 3d seems to use the same auxiliary points as d9, but for the initial focus acquisition, not just tracking. I can see why you might want this (if 3d is for hard to track subjects like "tennis players"), but it breaks the way I've been using 3D on the D700/800/810.

 

Looks like it's going on my feature request list unless I'm missing a way to change the initial acquisition behaviour.

I believe that people typically want images where there there is a clean line-of-sight between the camera and the main subject. 3D tracking assesses color patterns to identify the subject but I believe the subject needs to be fairly large for it to work correctly. (I have not studied what size it can identify.) If within that larger subject there is a nearer subject (such as a branch) it would not surprise me if it tried to focus on that instead. I would therefore make sure that the color pattern your subject is identified based on is in plain sight and constitutes a large part of the overall image.

The meter on the D800 and D810 is 91K pixels, which works out as roughly RGB x 213 x 142 (although I've no idea if this is the arrangement). The D700 has the same 1005-segment meter as the F5, which is a bit bigger than a tenth of that. The D850's meter is 180K pixels, which if they're RGB could be a 300x200 grid. That's a lot smaller than an AF point, although it could certainly be checking a larger area.

 

Still, if there's a consistent colour under the AF point, jumping to a nearby AF point seems wrong. I could actually cope with this - it's the identification of the initial subject that's giving me trouble. On the D810 and D800, and even D700, 3D tracking can be initiated on an eye, and it sticks to the eye - or on a background, and sticks to the background (which I've got to say may be a very different colour from the foreground, in my experiments). If nothing moves, the AF point doesn't either. The D850 just doesn't seem willing to use the central point to initiate 3D tracking - it always uses the extended 9-zone area including helper points.

 

Summary: I used to be able to eye-track, but now I can't because I can't get the AF point to lock on the eye rather than the surroundings; there's always something near to the eye that's closer (lashes, eyebrows...) which stops the AF hitting the eye. I think. Single-point works fine, but then I have to use the joystick manually and ensure my subject tracking is perfect (and independent of desired framing).

 

I'd love someone else to experiment and tell me whether they see the same behaviour, though - even if it just makes everyone swear off 3D.

My experience with 3D tracking is that on the D850 and D5 it works much better than on the D810. But I'm not trying to track an eye in such a way that eyelashes are to be ignored, this just doesn't seem something that I would consider realistic for 3D tracking. I'm fine if the eyelashes are sharp. If one wants the iris sharp and have the camera ignore the eyelashes, then I think one needs a mirrorless camera with eye AF, basically. The eyelashes are such a source of contrasty detail that I think it would be difficult to make the camera ignore them unless the face is filling the frame. I'm typically tracking faces where at least half the person's body is in the frame (sometimes the full body).

 

In fact with the D850 and 105/1.4, I am not able to get people's eyes (including eyelashes) in focus in a walking subject scenario consistently using 9-point dynamic area AF but 3D tracking does a phenomenal job at it. But it seems that Andrew is being much more demanding on the results. It could be also a fine tune issue or something to do with the different shapes of the sensitive areas for each AF point, or the fact that so many are cross-type in the Multi-CAM 20k.

have the camera ignore the eyelashes

Maybe deliberately AF Fine Tune to ~6mm further away....;)

 

Won't work in Video/LV though.....:D

To be fair, eyelashes don't bother me so much, it's focus on eyebrows when the head is at a tilt - and as someone spending a disproportionate amount of time photographing people playing tiddlywinks, I have a lot of shots of people's faces from an oblique angle.

 

Still, it's not just eyes. 3D tracking used to behave identically to single-point (except it was easier to focus-and-recompose) when shooting through branches, but with the D850 having anything vaguely close to the AF point causes a jump to the nearer object (and the difference can be a lot for a landscape framed by a tree). Live view fixes it, but that single-point AF does as well tells me that it's not the hardware of the AF unit that I should distrust.

 

Oh well, I'll definitely report it to Nikon, but if anyone feels up to confirming (or failing to reproduce) the behaviour with a contrived test, I'd love to know. I've found it quite intrusive - at least enough that I need to have one of the front control points set to activate single-point AF, which I never needed before.

Which lens are you using? I find newer lenses much more precise in AF and what you are seeing could be also related to the lens. You could try 105/1.4 or 70-200/2.8 FL if you know someone with those lenses.

I believe I've seen it with the 70-200 FL (and I agree that it seems to focus more precisely than the 70-200 VR2, although that may be the D850 rather than the D810 - my old 80-200 AF-D certainly had distance problems, but I think that was known behaviour). I was most recently experimenting with the Sigma 85mm f/1.4, which is at least reasonably precise. Again, I can toggle the AF distance (and hear the lens move) by toggling between single-shot and 3D tracking mode without moving anything, and the distance difference can be substantial - not just a slight miss. Possibly I should take some photos through the viewfinder so the behaviour is documented.