p.19 #1 · Z8 extremely unreliable/inconsistent AF in mildly low light
snapsy wrote:
In the previous video I demonstrated how the Z8's eye AF is constantly hunting in low-light (3.5EV) conditions. In this next video I demonstrate the reason - it's the phase evaluation from the PDAF pixels that's driving the hunting. If we switch the camera to MF we can see the camera's phase evaluation via the onscreen rangefinder. I demonstrate how the rangefinder is constantly flickering in 3.5EV conditions, indicating how the phase detection is vacillating between "in focus" and "oof", even though the camera and subject are locked down and the lens is idle. I then open the lens up to a larger aperture and demonstrate how the phase detection stabilizes. This showcases how sensor noise from low-light conditions makes phase evaluation tougher for the camera to focus. It also demonstrates how Nikon's decision to focus stopped-down creates a handicap for Z cameras vs other MILCs that focus wide-open.
Note in this video I've switched to the left eye, to make it easier on the AF system.
While viewing move your mouse cursor away from YT's embedded player so that the overlay at the bottom goes away - otherwise it'll obstruct the rangefinder display at the bottom left of the LV feed.
I have little doubt that these findings are likely reproducible and accurate. Nikon's engineers have designed an AF system which performs reasonably well though like nearly every design, it has apparent shortcomings. All of these models exhibit compromises involving technology, cost, manufacturing, etc. An interesting comparison would involve repeating these experiments with Sony and Canon to see if their AF systems demonstrate similar or alternative AF limitations. For example, I recall quite well that the Canon 7d and mkii suffered af inconsistencies and my experience with the R7 demonstrated af variability or pulsations (perhaps due to crop factor design?). Canon uses a dual pixel AF system so some of their advanced cameras might not demonstrate this phenomenon. Interestingly, since Nikon uses Sony sensors, it might be worthwhile exploring if and/or how Sony avoided this low light AF issue?
p.19 #2 · Z8 extremely unreliable/inconsistent AF in mildly low light
armd wrote:
I have little doubt that these findings are likely reproducible and accurate. Nikon's engineers have designed an AF system which performs reasonably well though like nearly every design, it has apparent shortcomings. All of these models exhibit compromises involving technology, cost, manufacturing, etc. An interesting comparison would involve repeating these experiments with Sony and Canon to see if their AF systems demonstrate similar or alternative AF limitations. For example, I recall quite well that the Canon 7d and mkii suffered af inconsistencies and my experience with the R7 demonstrated af variability or pulsations (perhaps due to crop factor design?). Canon uses a dual pixel AF system so some of their advanced cameras might not demonstrate this phenomenon. Interestingly, since Nikon uses Sony sensors, it might be worthwhile exploring if and/or how Sony avoided this low light AF issue? ...Show more →
I compared to a Sony A7r IV and Canon R3 a few days back:
p.19 #3 · Z8 extremely unreliable/inconsistent AF in mildly low light
Received new mannequin today with glass-type eye and eyelashes.
Here is an informal eye AF vs eyelash test on both the Z8 and R3, using each system's respective 50mm f1/.8 lens. This test was under ~7EV conditions. Both cameras had significant trouble locking on the eye of this new darker-skinned mannequin. As for focusing on the eye rather than eyelash, most of the Z8 shots had the eyelash in focus whereas the R3 had the eye in focus.
p.19 #4 · Z8 extremely unreliable/inconsistent AF in mildly low light
I will say that when shooting in adequate light that there is no problem with acquiring consistent focus, I have not generally found the Z8 to focus on the eyelashes more than the iris. It has happened sometimes, yes, but not really more often than I've found the actual eye in focus.
p.19 #6 · Z8 extremely unreliable/inconsistent AF in mildly low light
snapsy wrote:
Received new mannequin today with glass-type eye and eyelashes.
Here is an informal eye AF vs eyelash test on both the Z8 and R3, using each system's respective 50mm f1/.8 lens. This test was under ~7EV conditions. Both cameras had significant trouble locking on the eye of this new darker-skinned mannequin. As for focusing on the eye rather than eyelash, most of the Z8 shots had the eyelash in focus whereas the R3 had the eye in focus.
Thanks for demonstrating what has been known and discussed for some time, namely the tendency of the Nikon cameras to focus on parts of the human eye other than the iris. In birds and most animals, the difficulty doesn't exist.
With respect to your test on the Canon and Sony, it doesn't appear that you specifically looked for pulsation in AF-C mode.
p.19 #7 · Z8 extremely unreliable/inconsistent AF in mildly low light
snapsy wrote:
Received new mannequin today with glass-type eye and eyelashes.
Here is an informal eye AF vs eyelash test on both the Z8 and R3, using each system's respective 50mm f1/.8 lens. This test was under ~7EV conditions. Both cameras had significant trouble locking on the eye of this new darker-skinned mannequin. As for focusing on the eye rather than eyelash, most of the Z8 shots had the eyelash in focus whereas the R3 had the eye in focus.
Im not seeing the value in this. Neither of these cameras function like that with real people.
The scientific method nerd in me noticed all of the Nikon shots were at a higher angle basically shooting down through the lashes, while canon was more eye level. I suggest a real woman and a bit more consistency (tripod).
p.19 #8 · Z8 extremely unreliable/inconsistent AF in mildly low light
I'm a nerd for the scientific method as well.
To be fair, nobody's ever proven that the camera reacts the same way to a mannequin as it reacts to a real person... but then nobody's ever proven that the camera reacts differently. Nobody's ever bothered to do that experiment. Until the we have proof that the models are interchangeable and there is no difference to the camera between them, I'm going to have trouble putting much faith into the mannequin studies as being applicable to real people. To trust the mannequin tests assumptions have to be made that the substitution of one model vs. another does not interfere with the testing.
But that's not even the question that anyone should be spending their time and effort trying to answer. The question that we're really interested in answering is not whether the camera behaves differently on mannequins vs. people. The question remains whether the camera can accurately eye-focus on people in dark conditions such as EV 4.
I have to confess, I haven't been paying attention to the mannequin tests. Instead I've been paying attention to the SCoombs tests with human subjects. If there's an interest in performing detailed tests, I think the highest payback for their effort would be to replicate those conditions in which SCoombs has demonstrated that there is a problem -- with people in a dim but evenly lit room, where the AF software has to sort through a busy scene to identify the subject, then identify the eyes, then focus properly. If we could get several people to perform similar tests in controlled conditions like that, then we might get an interesting answer.
That said, it is good to know whether the Z cameras tend to preferentially focus on eyelashes, though there seem to be competing test objectives in this thread.
p.19 #9 · Z8 extremely unreliable/inconsistent AF in mildly low light
Buckeye2604 wrote:
Im not seeing the value in this. Neither of these cameras function like that with real people.
The scientific method nerd in me noticed all of the Nikon shots were at a higher angle basically shooting down through the lashes, while canon was more eye level. I suggest a real woman and a bit more consistency (tripod).
Some have reported the Z's regularly focus on eyelashes rather than the eye in real-world situations, noticeable when shooting at very shallow DOF. I was able to recreate that using the mannequin, and I did so many times in varying experiments prior to recording the video demonstration I posted.
p.19 #10 · Z8 extremely unreliable/inconsistent AF in mildly low light
I though of a potential workaround to allow Z's to focus at larger apertures.
Yesterday I posted a video showing why the Z8's Eye AF is struggling in low-light - it's because the camera focuses with the aperture stopped down to the shooting aperture (up to f/5.6), which handicaps the AF system vs other MILC system by reducing the amount of light reaching the AF system.
Today I thought of a workaround: Assign a control to the Z8's "Recall Shooting Functions" feature to temporarily switch to a larger aperture while also engaging the focus system. Then use that control to focus, releasing just before pressing the shutter to return the aperture back to the shooting aperture.
Setup
Menu -> Custom Settings -> fControls -> Custom Controls (shooting)
Pick a control to assign. In my demonstration below I chose AF-ON but you can use a different control if you prefer to keep AF-ON operating in a standard function.
Scroll down to Recall Shooting Functions and click into it
Put a check next to "Aperture" and select a large aperture you want to focus at. In my demonstration I chose f/2
Put a check next to "AF-ON"
If you'd like to make subject detection and tracking unique to this method you can also put checks next to those as well and configure them accordingly.
Accept the changes
Shooting
Set your desired shooting aperture. In my demonstration I chose f/5
To focus, hold down the control you assigned. This will open the aperture to f/2 and focus.
To take a photo, release the control and press the shutter button immediately
after. This will return the aperture back to the shooting aperture f/5 and take the photo
Naturally this requires a bit of coordination but I was able to get the hang of it rather quickly. How well this works in practice will depend on how fast you can release the assigned control and pressing the shutter in quick succession. But the limit wont be the camera because it seems to respond quite quickly to these control presses.
p.19 #11 · Z8 extremely unreliable/inconsistent AF in mildly low light
coralnut wrote:
I'm a nerd for the scientific method as well.
To be fair, nobody's ever proven that the camera reacts the same way to a mannequin as it reacts to a real person... but then nobody's ever proven that the camera reacts differently. Nobody's ever bothered to do that experiment. Until the we have proof that the models are interchangeable and there is no difference to the camera between them, I'm going to have trouble putting much faith into the mannequin studies as being applicable to real people. To trust the mannequin tests assumptions have to be made that the substitution of one model vs. another does not interfere with the testing.
But that's not even the question that anyone should be spending their time and effort trying to answer. The question that we're really interested in answering is not whether the camera behaves differently on mannequins vs. people. The question remains whether the camera can accurately eye-focus on people in dark conditions such as EV 4.
I have to confess, I haven't been paying attention to the mannequin tests. Instead I've been paying attention to the SCoombs tests with human subjects. If there's an interest in performing detailed tests, I think the highest payback for their effort would be to replicate those conditions in which SCoombs has demonstrated that there is a problem -- with people in a dim but evenly lit room, where the AF software has to sort through a busy scene to identify the subject, then identify the eyes, then focus properly. If we could get several people to perform similar tests in controlled conditions like that, then we might get an interesting answer.
That said, it is good to know whether the Z cameras tend to preferentially focus on eyelashes, though there seem to be competing test objectives in this thread....Show more →
Agreed. Hopefully those two can establish dialogue and present the videos to Nikon support for evaluation.
p.19 #12 · Z8 extremely unreliable/inconsistent AF in mildly low light
Buckeye2604 wrote:
Im not seeing the value in this. Neither of these cameras function like that with real people.
The scientific method nerd in me noticed all of the Nikon shots were at a higher angle basically shooting down through the lashes, while canon was more eye level. I suggest a real woman and a bit more consistency (tripod).
Hi Buckeye2604, etal,
While I agree that measurably repeatable test setup scenes need to be attained here moving forward, I disagree on the need for an actual human...even my ancient and lowly DfD Lumix S1 "knows" how to throw up the eye focus/recognition crosshairs on a "picture of a face" hanging on the wall.
Also, depth-of-field is both subject-distance and lens focal-length dependent at any aperture and I would suggest here that providing focus tests at various (selected) measured subject distances and varying apertures might provide some additional insights into the Eye-AF tests being performed.
Also, the lens-focusing technology will factor in here as well, I suspect that the performance of Nikon's stepper motor lenses will perform differently than a lens that does not use steppers in that there might not be enough steps available for the camera's AF system to accommodate fine-focusing accuracy at close distances.
Also, in tying these two together with the scene-EV-brightness tests one may well discover that there are distance limits to Eye-AF accuracy per, er, scene-Candela.
Of course, enough data would all lead to the ability to create spreadsheets and graphs that would illustrate the boundaries for successful Eye-AF, but that would fall on the interests and desires and energy of the OP and those providing additional testing to pursue. It has only been because of their kindness and determination and persistence here that this conversation has developed as far as it has!
In my humble opinion at this point, methinks that enough demonstration has been shown that as illumination decreases human eye-AF performance (for those cameras so far tested) reach a point of non-accurate performance (perameters-dependent for each camera-lens combo). I should think this would be of interest to all manufacturers since shooting in EV 4 conditions is certainly a reasonable expectation, heck, to stay on topic, the Z8 is advertised to AF to...
"Detection range (in photo mode, AF-S, ISO 100, f/1.2 lens, 20°C/68°F) -7 to +19 EV (-9 to +19 EV with starlight view)"
...the above tested a7R IV is advertised to AF to...
"Focus Sensitivity Range EV-3 to EV 20 (ISO 100 equivalent with F2.0 lens attached)"
...and the above tested R3 is advertised to AF to...
"AF Working Range EV -7.5 – 20 (at 23°C & ISO100) ^3 The lowest light AF is available with an f/1.2 lens, centre AF point, One-Shot AF, at 23°C / 73°F, ISO 100 Except RF lenses with a Defocus Smoothing (DS) coating".
At the very least, the OP and other Z8 shooters will hopefully bring Nikon's Tech Support into the conversation here otherwise all of this testing and conversation will all just be one big time waster...well, IMO.
p.19 #13 · Z8 extremely unreliable/inconsistent AF in mildly low light
snapsy wrote:
I though of a potential workaround to allow Z's to focus at larger apertures.
Yesterday I posted a video showing why the Z8's Eye AF is struggling in low-light - it's because the camera focuses with the aperture stopped down to the shooting aperture (up to f/5.6), which handicaps the AF system vs other MILC system by reducing the amount of light reaching the AF system.
Today I thought of a workaround: Assign a control to the Z8's "Recall Shooting Functions" feature to temporarily switch to a larger aperture while also engaging the focus system. Then use that control to focus, releasing just before pressing the shutter to return the aperture back to the shooting aperture....Show more →
Thank you for taking the conversation off the bench and toward real world solutions
In addition to what you're suggesting, if people are struggling to get focus under certain conditions, they can:
1) Set "recall shooting function hold" to one of the non-eye based AF modes, like 3D tracking, and then attempt to grab the leading eye as an object to track, or...
2) As has always been the case when cameras struggle in low contrast & low light scenarios, use AF to grab a high(er) contrast area ahead of - or in line with - the eye. I'm typically taking a step back to shoot a longer FL, looking for leading shoulder, hairline, waistband, something on the shirt...
3) Lastly, when appropriate, I'd personally set myself up with a super compact hotshoe LED at its lowest power - IMO not much more distracting or intrusive than the red AF assist beams, and certainly about the same as the regular AF assist, but of course in this setup you'd pick up AF-C. Ulanzi makes one that's $40 and fits the bill for me...
My bet is even at 25% power it's probably a touch too bright, and I'd put an added layer of diffusion over it. 4 hours running time, I'd buy two or research if it can be on while charging from a cage & NPF/V/Gold mount setup (super common now that we're expected to switch from stills to video).
I haven't shot more than maybe 15 really low light events but I have, ages ago, assisted some of the top tier wedding shooters, and they routinely had an assistant with a low powered LED on a stick roving the dance floor with them, off-axis. I think for a lot of event stuff, doubling down on the absurdity of being "stealthy" with a massive camera setup, engaging with people, embracing and making it part of the fun, does a lot to overcome gear shortcomings.
p.19 #15 · Z8 extremely unreliable/inconsistent AF in mildly low light
Nice work again snapsy.
It would be nice if Nikon were to let people choose whether they want the camera to focus wide-open or stop down… I think Sony does that.
And Canon seem to focus wide-open all the time… I’ve never had an issue with focus shift due to this.
This workaround obviously won’t help when using an F/4 or slower lens - but looks like it will be helpful in some instances.
p.19 #16 · Z8 extremely unreliable/inconsistent AF in mildly low light
snapsy wrote:
Here is a new video I created that fully describes the AF aperture workaround I posted about earlier.
Thanks for diving deep into this. Great information and something I was completely unaware of! I think that it would make a lot of sense if Nikon offered an option to focus wide open as Sony does.
Looking at your settings here, I am assuming this is mainly applicable to flash use as I feel at ISO 50,000 most of us would be tempted to just open up the lens. On the plus side, it was reassuring to see how fast and accurate the AF is in this lowish light once the lens is opened up.
p.19 #17 · Z8 extremely unreliable/inconsistent AF in mildly low light
I've been messing around a bit more and have started to come up with an alternative, or at least a complementary theory, as to what is going on here.
After some fortuitous observations and then a lot of followup in which I tried to replicate what I noticed, I'm starting to think that one factor here is that the AF system is just in lower light having a hard time adjusting focus quickly enough to keep up with subject movement.
If I have a person in lower light walk towards me, I get a nearly 0% hit rate with the focus almost always lying behind the place where the focus point is supposed to be. If I have the same person in the same conditions walk across the frame, maybe at a slight angle so that they are changing distance but not nearly so much as when walking towards me, I get much better results. Very importantly, note that while this is slightly worse with subject detect, it's really almost as bad with the single point and dynamic modes which we had formerly found tended to produce significantly better results. This is proving for me not to be the case when the subject is moving - and I'm not asking for anything extreme, but just having them walk at a normal pace.
I am also finding that when I take care to shoot people who are not moving, even in low light, the success rate is imperfect but it's a lot better than I had initially reported three of four threads ago.
Factors in favor of this explanation are that it would explain why so many of the issues with focus have been back focused, rather than randomly missing, and would also make sense in the light of the way snapsy did his tests by moving around the mannequin an so constantly changing focal distance. It would also make sense with the fact that reducing the focal length of the lens seems to decrease the problems we've observed, as longer focal lengths always make it much more difficult to maintain focus on moving subjects. Additionally, a lot of people who are regularly shooting in lower light may not necessarily be shooting a lot of moving subjects in lower light.
It may also explain why some people have reported not noticing these sorts of problems and also use "focus priority" for their release priority. I did some experimenting with disabling back button focus and using only the shutter button so that I would press it and acquire focus all in one swift motion and this improved results a fair bit. In other words, when focus and shot occur at almost the same time, before the subject can move much, results are better.
p.19 #18 · Z8 extremely unreliable/inconsistent AF in mildly low light
Snapsy - do you know whom to contact at Nikon with your findings? I have some findings with video focusing in broad daylight, which I want to share with whomever may use it to improve focus with this camera. You may PM me with a response if you like.
p.19 #19 · Z8 extremely unreliable/inconsistent AF in mildly low light
ronno wrote:
Snapsy - do you know whom to contact at Nikon with your findings? I have some findings with video focusing in broad daylight, which I want to share with whomever may use it to improve focus with this camera. You may PM me with a response if you like.
I don't have any contacts. Sometimes Thom Hogan forwards my stuff to his Nikon contacts so you might try reaching out to him.
p.19 #20 · Z8 extremely unreliable/inconsistent AF in mildly low light
It's probably worth mentioning that Thom Hogan has now mentioned, both in an article and in some followup discussion over on DPReview that he considers AF in low ambient lighting to be a problem with the Z8. He hasn't mentioned the specific issue that has been discussed here, but commented on issues with AF in low light in general.
"Trial and error coupled with focus misses in strobe managed light is not what we want. Thing is, a modern Nikon camera knows when a trigger or flash is in the hot shoe and active. Why doesn't that automatically reconfigure the camera to be optimally set for that situation (with user customization, of course)? We know that low ambient studio light might make the focus system struggle, so where's Nikon's solution for that?"
Speaking of in studio solutions like modeling lamps,
"it's what I do when I'm in my studio. However, it's another thing to juggle extra gear like that when you're trying to do a spontaneous event, such as a wedding reception. It is an issue. Perhaps not for you, but it is for many, and for many situations. Given that I can think of at least two potential solutions helps off the top of my head, I believe Nikon needs to address this."
Jimmy G