Z8 Subject Detection Inaccurate in Low (but not that low) light

[hubschko]

Would be interesting if someone would do the test with a Sony or Canon!

 

[Leonard Shepherd]

Would be interesting if someone would do the test with a Sony or Canon!

I believe it would be interesting to do it side-by-side system comparisons but with real world subjects.

I would expect every system to perform with less efficiency in low light because most subjects have lower contrast in lower light and AF works less efficiently in lower light.

With any system, a faster aperture used could improve AF efficiency in challenging low light low contrast situations.

Nikon (and I expect Canon and Sony) caution AF efficiency reduces in low light and lower contrast situations.

I find I need to occasionally resort to manual focus much less frequently with the Z8 and Z9 than with the now sold D850 and to a lesser extent the now sold Z7.

I expect low light issues will reduce with the Z8 and Z9 when Nikon add eye focus confirmation in manual focus as on the ZF, though so far I have had none using eye AF on the Z8 and 9.

 

[JustSomeone]

My advise after shooting with the Z 8 quite a bit.

• in low light, use the 'Adjust for ease of viewing' setting in the custom menu, d8: View Mode (Photo Lv)

• if you do not want to do the above, make sure the image you see in the evf is as bright as it can be, that really helps with the keepers in low light.

• enable d19: High FPS Viewfinder Display

The first point though, is a game changer.

Try it and let us know how the results improve.

 

[SCoombs]

My advise after shooting with the Z 8 quite a bit.

• in low light, use the 'Adjust for ease of viewing' setting in the custom menu, d8: View Mode (Photo Lv)

• if you do not want to do the above, make sure the image you see in the evf is as bright as it can be, that really helps with the keepers in low light.

• enable d19: High FPS Viewfinder Display

The first point though, is a game changer.

Try it and let us know how the results improve.

I appreciate the suggestion. As noted several times above, I am already using this option and have experimented with just about every possible option. Most of the people reporting similar problems are already using this and a few other settings to try to get something reliable.

In fact, I'd even say in my more recent testing that the results mat have been better with this setting off. This surprised me, but my theory as to why it made a positive difference is that by making it harder on the AF it's forcing the subject detection to operate more slowly and it seems as though part of the problem is the subject detection making frequent adjustments which in lower light have a higher probability of error. By making the system work harder, it might very that it's staying on any good focus it finds for longer stretches before trying to move the focus again. That's just a theory, but regardless I have found recently that adjust settings for ease of viewing seems to yield slightly more inconsistent results for me.

 

[jlafferty]

It's more than a little disturbing that these Nikon MLCs are still struggling to focus on the closest eye in low light, the light range where professional studio photographers are shooting ...

I don’t see many “professional studio photographers” having this problem, frankly. If it was an issue that couldn’t be resolved with some simple tried and true techniques, nobody would be buying the Z8/Z9. Even the OP admits that adjusting their settings yields 75-100% accuracy… take a guess what “professional studio photographers” do in a situation where they find one setting works and another works… less.

 

[JustSomeone]

My advise after shooting with the Z 8 quite a bit.

• in low light, use the 'Adjust for ease of viewing' setting in the custom menu, d8: View Mode (Photo Lv)

• if you do not want to do the above, make sure the image you see in the evf is as bright as it can be, that really helps with the keepers in low light.

• enable d19: High FPS Viewfinder Display

The first point though, is a game changer.

Try it and let us know how the results improve.

I appreciate the suggestion. As noted several times above, I am already using this option and have experimented with just about every possible option. Most of the people reporting similar problems are already using this and a few other settings to try to get something reliable.

In fact, I'd even say in my more recent testing that the results mat have been better with this setting off. This surprised me, but my theory as to why it made a positive difference is that by making it harder on the AF it's forcing the subject detection to operate more slowly and it seems as though part of the problem is the subject detection making frequent adjustments which in lower light have a higher probability of error. By making the system work harder, it might very that it's staying on any good focus it finds for longer stretches before trying to move the focus again. That's just a theory, but regardless I have found recently that adjust settings for ease of viewing seems to yield slightly more inconsistent results for me.

Interesting. Probably our shooting environment is different and that matters in terms of end results.

I'm sorry this has been discussed before, I just didn't (and won't) read all the comments on such a long thread, haha!

You mention: "it's staying on any good focus it finds for longer stretches before trying to move the focus again."

Another setting you must look at is 'a3: Focus Tracking with Lock-On' (I'm sure you have tried!!) which does exactly what you said. Maybe worth trying both ends of the scale.

 

[HyperCypher]

Read about 50% of the thread. Did not see any mention of VR status.

When I got my Z6 the first thing I noticed was that AF C never seemed to “settle”. It does if you are on tripod. I would like to see a similar test where the camera is on tripod and only the subject moves.

 

[SCoombs]

It's more than a little disturbing that these Nikon MLCs are still struggling to focus on the closest eye in low light, the light range where professional studio photographers are shooting ...

I don’t see many “professional studio photographers” having this problem, frankly. If it was an issue that couldn’t be resolved with some simple tried and true techniques, nobody would be buying the Z8/Z9. Even the OP admits that adjusting their settings yields 75-100% accuracy… take a guess what “professional studio photographers” do in a situation where they find one setting works and another works… less.

This is not really what I have said. I said that limiting one's choices to a small selection of focus modes seemed to yield a 75% hit rate with easier subjects. However, these focus modes are not really ideal for a lot of the photography that takes place in conditions where this problem occurs and their use causes some shots to be missed.

Second, in my update I added that when going into a real hands on situation at an event I found these modes not to be as reliable as initial testing suggested. They missed a lot and I had to check every shot and take every photo three times to try to ensure I got the results I needed.

 

[SCoombs]

Read about 50% of the thread. Did not see any mention of VR status.

When I got my Z6 the first thing I noticed was that AF C never seemed to “settle”. It does if you are on tripod. I would like to see a similar test where the camera is on tripod and only the subject moves.

I have wondered a lot about whether the VR was a factor and experimented with the VR. One problem trying to reach any conclusions is that when shooting in conditions dark enough that this problem arises you are of necessity shooting with lower shutter speeds.

Of course withba speedlight that shouldn't matter too much I suppose. Either way, I haven't noticed a difference when trying to switch the VR off.

 

[Rodenmg]

This business of not focusing where the indicator box says it should be definitely looks like a firmware glitch that should be fixed.

But, do you have a lens faster than f4? In low light situations, much darker than your first post examples, I can put on the 50 f1.2 and it’s amazing. No indicator jumping from eye to eye. It just locks on. Even a f1.8 lens will make a huge difference.

 

[SKersting66]

I think I can guess the answer to this, but what does 3D-tracking do? We already know that the dynamic AF modes behave better than the qualitative subject ID and tracking modes. 3D-tracking is a subject designation - and - pattern tracking mode that makes no qualitative judgements about what it's viewing. My suspicion is that 3D-tracking would be more consistent than any of the subject identification and tracking modes. Unfortunately, 3D-tracking wouldn't be able to handle a subject turning partially away from the lens, but that requires qualitative processing.

3D adds distance information from the lens for predictive AF (depth, the third dimension). It also samples the color/contrast under the focus point and tracks that... this information used to come from the 3D matrix meter (which uses lens focus distance); now it's all sensor based.
However, 3D doesn't work when subject recognition is active... e.g. place the selected focus point on her red lips and it will still jump to the eyes. 3D will work if the camera doesn't find a subject, or if subject detection is disabled.
--
https://www.flickr.com/skersting

The PDAF pixels themselves are what provides distance information and that is used for all focus operations other than pin-point AF.

Phase detection has no concept of focus distance; it only knows phase offset (amount/direction). That correlates to distance as the direction of phase offset dictates the direction of focus drive; but that is not the same thing as reported focus distance (with D and later lenses).

Unfortunately, the mirrorless have a very small baseline for phase offset/detection.

 

[SCoombs]

This business of not focusing where the indicator box says it should be definitely looks like a firmware glitch that should be fixed.

But, do you have a lens faster than f4? In low light situations, much darker than your first post examples, I can put on the 50 f1.2 and it’s amazing. No indicator jumping from eye to eye. It just locks on. Even a f1.8 lens will make a huge difference.

I've seen this with the 70-200 2.8, the 40 f2, and the 85 1.8. The 85 at 1.8 is slightly improved.

The tradeoff or issue is this: if you search around there are a lot of discussions of Z9 users complaining about the same or very similar issues and one thing they discuss a lot is specifically that they have the problem the worst with their 1.8s and 1.2s. What they were saying is that the depth of field with these was shallow enough that the camera couldn't in lower light consistently get focus accurate enough for the eyes to be in focus at those low apertures. The conclusion many of them reached was that they needed to NOT use fast lenses in low light. Ironic, but it shows how even with fast lenses this can be a problem.

From what I can read, before firmware 2.0 on the Z9 they couldn't even use single point with these fast lenses *and even in good light*, but as of 2.0 (Z9) they could now use fast lenses in all light but subject detect in low light was still a problem.

 

[HRS]

It's more than a little disturbing that these Nikon MLCs are still struggling to focus on the closest eye in low light, the light range where professional studio photographers are shooting ...

I don’t see many “professional studio photographers” having this problem, frankly.

Anyone take a survey on this?

If it was an issue that couldn’t be resolved with some simple tried and true techniques, nobody would be buying the Z8/Z9. Even the OP admits that adjusting their settings yields 75-100% accuracy… take a guess what “professional studio photographers” do in a situation where they find one setting works and another works… less.

Imagine a professional wedding photographer delivering proofs with 25% out of focus shots. It's not exactly the case that you could redo the wedding.

--
"Ad hominem" somebody if you can't offer facts to refute theirs.

 

[HRS]

It's more than a little disturbing that these Nikon MLCs are still struggling to focus on the closest eye in low light, the light range where professional studio photographers are shooting .... it is an ongoing issue that does not seem to go away in spite of many FW updates .......

If it was me shooting a model and I was having these issues it would drive me "absolutely nuts" .... it seems you have to check every image to make sure you've got the shot.

A perfect two-paragraph summary of Nikon M/L since their introduction in 2018.

I've been "there" three times (Z6, 2x Z6ii) and have reached the conclusion that Nikon doesn't have the ability to fix this.

 

[Dinosaur dave]

It's more than a little disturbing that these Nikon MLCs are still struggling to focus on the closest eye in low light, the light range where professional studio photographers are shooting ...

I don’t see many “professional studio photographers” having this problem, frankly. If it was an issue that couldn’t be resolved with some simple tried and true techniques, nobody would be buying the Z8/Z9. Even the OP admits that adjusting their settings yields 75-100% accuracy… take a guess what “professional studio photographers” do in a situation where they find one setting works and another works… less.

Ah, this thread must be in the fictional portrait photography section I suppose .... but then I'm just an observer .

maybe your POV would be better aimed at those who are having the issues rather than those looking on ..... or is it that I'm easy meat?

 

[Horshack]

Here's a test with the Sony A7r IV. I don't have a native midrange lens handy so I had to use a Canon 24-70 f/4 IS with the Sigma MC-11 adapter. It works pretty well but you'll notice a few times the combo goes out to lunch and goes completely OOF needs a second or two to get back.

Note unlike Nikon, the Sony lets you specify an eye preference - I specified the model's right eye.

One observation on the Sony is that it wont detect an eye at anything approaching an oblique angle, as opposed to the Z8 which shows eye indication at pretty steep angles. I would say the Z8 performs equal to the Sony at the shallow angles. Perhaps the issue is the camera is indicating eye AF beyond the actual capability of the system and so it's more a reporting/display issue than a fundamental AF issue.

Note my original Z8 video was at 120mm so I shot the Nikon again today at 70mm for a fair comparison. I also enabled the red AF point display for image review/playback, to make it obvious what the camera focused on for each exposure. The Z8 does noticeably better at 70mm than it did at 120mm, and today I noticed it does better if I step back a bit. It seems to have the most trouble at very close focus distances, where it appears to put the wrong eye in focus (model's left eye) and leave it there even though the eye AF indication is on the right eye

Sony @ 70mm (adapted lens):



Nikon @ 70mm:

 

[SCoombs]

Here's a test with the Sony A7r IV. I don't have a native midrange lens handy so I had to use a Canon 24-70 f/4 IS with the Sigma MC-11 adapter. It works pretty well but you'll notice a few times the combo goes out to lunch and goes completely OOF needs a second or two to get back.

Note unlike Nikon, the Sony lets you specify an eye preference - I specified the model's right eye.

One observation on the Sony is that it wont detect an eye at anything approaching an oblique angle, as opposed to the Z8 which shows eye indication at pretty steep angles. I would say the Z8 performs equal to the Sony at the shallow angles. Perhaps the issue is the camera is indicating eye AF beyond the actual capability of the system and so it's more a reporting/display issue than a fundamental AF issue.

Note my original Z8 video was at 120mm so I shot the Nikon again today at 70mm for a fair comparison. I also enabled the red AF point display for image review/playback, to make it obvious what the camera focused on for each exposure. The Z8 does noticeably better at 70mm than it did at 120mm, and today I noticed it does better if I step back a bit. It seems to have the most trouble at very close focus distances, where it appears to put the wrong eye in focus (model's left eye) and leave it there even though the eye AF indication is on the right eye

Sony @ 70mm (adapted lens):
Nikon @ 70mm:

How much of this is the system performing better at 70mm vs greater depth of field.

 

[Horshack]

Here's a test with the Sony A7r IV. I don't have a native midrange lens handy so I had to use a Canon 24-70 f/4 IS with the Sigma MC-11 adapter. It works pretty well but you'll notice a few times the combo goes out to lunch and goes completely OOF needs a second or two to get back.

Note unlike Nikon, the Sony lets you specify an eye preference - I specified the model's right eye.

One observation on the Sony is that it wont detect an eye at anything approaching an oblique angle, as opposed to the Z8 which shows eye indication at pretty steep angles. I would say the Z8 performs equal to the Sony at the shallow angles. Perhaps the issue is the camera is indicating eye AF beyond the actual capability of the system and so it's more a reporting/display issue than a fundamental AF issue.

Note my original Z8 video was at 120mm so I shot the Nikon again today at 70mm for a fair comparison. I also enabled the red AF point display for image review/playback, to make it obvious what the camera focused on for each exposure. The Z8 does noticeably better at 70mm than it did at 120mm, and today I noticed it does better if I step back a bit. It seems to have the most trouble at very close focus distances, where it appears to put the wrong eye in focus (model's left eye) and leave it there even though the eye AF indication is on the right eye

Sony @ 70mm (adapted lens):
Nikon @ 70mm:

How much of this is the system performing better at 70mm vs greater depth of field.

The OOF is well outside the DOF difference of either focal length so it's the system performing better. Plus I can tell when the camera is failing to focus on the right eye - I don't even need to take a photo - but I do so anyway for demonstration purposes.

I don't think it's the focal length that matters but instead the size of the face. At 70mm the camera performs noticeably worse when I'm close, roughly equal to the performance of 120mm a few steps back, ie equal framing.

 

[Horshack]

I think I can guess the answer to this, but what does 3D-tracking do? We already know that the dynamic AF modes behave better than the qualitative subject ID and tracking modes. 3D-tracking is a subject designation - and - pattern tracking mode that makes no qualitative judgements about what it's viewing. My suspicion is that 3D-tracking would be more consistent than any of the subject identification and tracking modes. Unfortunately, 3D-tracking wouldn't be able to handle a subject turning partially away from the lens, but that requires qualitative processing.

3D adds distance information from the lens for predictive AF (depth, the third dimension). It also samples the color/contrast under the focus point and tracks that... this information used to come from the 3D matrix meter (which uses lens focus distance); now it's all sensor based.
However, 3D doesn't work when subject recognition is active... e.g. place the selected focus point on her red lips and it will still jump to the eyes. 3D will work if the camera doesn't find a subject, or if subject detection is disabled.
--
https://www.flickr.com/skersting

The PDAF pixels themselves are what provides distance information and that is used for all focus operations other than pin-point AF.

Phase detection has no concept of focus distance; it only knows phase offset (amount/direction). That correlates to distance as the direction of phase offset dictates the direction of focus drive; but that is not the same thing as reported focus distance (with D and later lenses).
Unfortunately, the mirrorless have a very small baseline for phase offset/detection.

If that were true the camera wouldn't know how far to drive the lens to correct the phase differential. We know that it does. And we know the camera can translate lens positions to distances, so the math is all there for the camera to use the phase differential to calculate distance.

 

[Andre Yew]

From what I can read, before firmware 2.0 on the Z9 they couldn't even use single point with these fast lenses *and even in good light*, but as of 2.0 (Z9) they could now use fast lenses in all light but subject detect in low light was still a problem.

That is not my experience, and it was the opposite for me: 2.0 felt like the peak of the Z9's AF abilities. I have studio shots (so dark ambient) that were critically sharp on her eyes for almost every photo for a moving subject except for a few when she wasn't already in a frame and entered it (eg. a dancer getting a running start on a jump). I was shooting a 24-120/4 between wide open and f/8 for the shots. With faster lenses in demanding situations (eg. a dark theater), the Z9 then had a huge hit rate, feeling like it almost couldn't miss.

Whereas more recently, shooting against the sun at golden hour with subject detect on, with lenses like the 85/1.2S and the 135/1.8, the hit rate was surprisingly bad: not even bridge of nose vs. eye bad, but a couple feet off. I have to wonder how Nikon tests their AF: these AI heuristics can be very difficult to test because they can be pretty sensitive to changes from what they were trained on. For me, it feels like human subject detect at least has regressed over the last 2 years on the Z9.

And then there was that odd behavior with the half covered face I described above.

--
https://www.instagram.com/lolcar/