Nikon, better subject detection, please!

If you would like to post, you'll need to register. Note that if you have a BCG store account, you'll need a new, separate account here (we keep the two sites separate for security purposes).

Status
Not open for further replies.
@ajrmd here are a few images of Bald Eagles at the point of picking up fish in 3 different locations and much different environments/backgrounds. This is very typical results for me. These were shot with a mix of my Z800mm and my old 500PF adapted
Appreciate the photos though it would be helpful to see a series of the bird approaching the water, making the strike, and flying away, with all of the images in focus rather than individual images from different strikes. What I have experienced is that the camera follows the bird down to the water and at the moment of impact, or just before, the AF jumps off the bird for a few frames and then reacquires the subject sometime later. I had posted such a series in an older thread and these occurrences are routine regardless of the AF mode, lighting, etc.
 
Appreciate the photos though it would be helpful to see a series of the bird approaching the water, making the strike, and flying away, with all of the images in focus rather than individual images from different strikes. What I have experienced is that the camera follows the bird down to the water and at the moment of impact, or just before, the AF jumps off the bird for a few frames and then reacquires the subject sometime later. I had posted such a series in an older thread and these occurrences are routine regardless of the AF mode, lighting, etc.
I rarely edit full sequences so I would have to go though and see if I have many or I would have to go back to process quite a few images
 
Appreciate the photos though it would be helpful to see a series of the bird approaching the water, making the strike, and flying away, with all of the images in focus rather than individual images from different strikes. What I have experienced is that the camera follows the bird down to the water and at the moment of impact, or just before, the AF jumps off the bird for a few frames and then reacquires the subject sometime later. I had posted such a series in an older thread and these occurrences are routine regardless of the AF mode, lighting, etc.
Bald Eagle 2-26-23 1aa.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-26-23 1b.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-26-23 1c.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-26-23 1bb.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-26-23 1aaa.jpg
You can only see EXIF info for this image if you are logged in.

Here is one sequence I found. It might take me a bit. I have an average of about 3000-4000 images in folders of 2022, 2023 and 2024 to sift though of processes images before looking to maybe process some old raw images
 
@ajrmd Here is a couple more sequences with the images closer together at the water. I can only put 8 images per post so I will make a 2nd post after this one

Bald Eagle 2-28-23 4c.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-28-23 4f.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-28-23 4e.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 2-28-23 4d.jpg
You can only see EXIF info for this image if you are logged in.
 
@ajrmd
Bald Eagle 3-1-23 3d.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 3-1-23 3i.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 3-1-23 3h.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 3-1-23 3g.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 3-1-23 3f.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 3-1-23 3e.jpg
You can only see EXIF info for this image if you are logged in.
 
@ajrmd And here is one more with maybe the most challenging for any AF system of any camera brand. Lots of water splash and coming right at the camera.
Bald Eagle 4-12-23 1aa.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1n.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1hhh.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1g.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1ff.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1e.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1c.jpg
You can only see EXIF info for this image if you are logged in.
Bald Eagle 4-12-23 1b.jpg
You can only see EXIF info for this image if you are logged in.
 
I rarely edit full sequences so I would have to go though and see if I have many or I would have to go back to process quite a few images
Here's another example of what I experience on a routine basis with the Z8 with Osprey/Eagle strikes. These are not cropped images and the bird is fairly large in the frame with effective panning. This sequence had a total of 70 consecutive images. Initial focus was achieved before the Osprey swooped towards the fish, and as it approached the water with talons forward, the AF simply lost the subject for a total of 7 frames before regaining after the strike and the bird remains in focus for the rest of the images in the sequence. It doesn't appear to AF on a splash, a more "contrasty" hot spot, etc. I have tried changing modes from Auto Area to a custom narrow, to Wide S, L, etc. without any difference. If this were a one off, I could understand but it happens all of the time as though the AF just has a brain fart and it suddenly FF's.

I've included a screenshot of part of the sequence as the osprey nears the water (in focus), included larger screenshots of that image and the next four OOF images (skipped the next 3 OOF) and then the image when it regained AF. Simply no explanation in my book.

first0000.jpg
You can only see EXIF info for this image if you are logged in.
second.jpg
You can only see EXIF info for this image if you are logged in.
third.jpg
You can only see EXIF info for this image if you are logged in.
fourth.jpg
You can only see EXIF info for this image if you are logged in.
fifth.jpg
You can only see EXIF info for this image if you are logged in.
sixth.jpg
You can only see EXIF info for this image if you are logged in.
 
Here's another example of what I experience on a routine basis with the Z8 with Osprey/Eagle strikes. These are not cropped images and the bird is fairly large in the frame with effective panning. This sequence had a total of 70 consecutive images. Initial focus was achieved before the Osprey swooped towards the fish, and as it approached the water with talons forward, the AF simply lost the subject for a total of 7 frames before regaining after the strike and the bird remains in focus for the rest of the images in the sequence. It doesn't appear to AF on a splash, a more "contrasty" hot spot, etc. I have tried changing modes from Auto Area to a custom narrow, to Wide S, L, etc. without any difference. If this were a one off, I could understand but it happens all of the time as though the AF just has a brain fart and it suddenly FF's.

I've included a screenshot of part of the sequence as the osprey nears the water (in focus), included larger screenshots of that image and the next four OOF images (skipped the next 3 OOF) and then the image when it regained AF. Simply no explanation in my book.

View attachment 95723View attachment 95724View attachment 95725View attachment 95726View attachment 95727View attachment 95728
That would be unacceptable to me. No wonder I bought and subsequently sold the Z9 twice over.
 
@ajrmd hmm, I can't say that that hasn't happened to me but that is not at all common in my experience with my Z9+Z800mm or prior when I had the 500PF on my Z9 for the 13 months I shot with it (Jan 2022 to Feb 2023). I would say I have seen that less then 5% of the time.

All I can tell you is how I shoot my Z9

Auto Area AF mode (the only AF mode I use 95% of the time for recognition, acquisition, and tracking) using Bird Subject Detectoin with blocked shot set to 5 and Subject Motion to steady (both in Menu Item A3). I turn off Release Timing Indicator (D15) shoot in Flat PC. I actually only use the custom PC for the real time blown highlights in stills which is based on the Flat PC. I shoot in 120Hz in the EVF
 
View attachment 95668
This is one of the images I shot with a Z6III and a 28-400 lens during a recent vacation. As can be seen, AF grabbed the eye (Wide-L, animal, standard picture control). Out of this particular burst of 20 or so only two were locked on the eye, but I was on a boat and keeping the gull in the frame was a challenge, even using VR. It isn't my experience that Nikon AF never finds the eye of a bird in flight at sensible distances. I'd love the hit rate to be higher, but it isn't zero either.
Bearing in mind you were using an f8 lens with which AF is usually not as fast or accurate as using wider apertures, and you had trouble keeping the bird in the frame from a moving boat - your expectations seem realistic - though of course we (including me) could probably get a higher rate with the 400 f2.8 TC and a flat calm sea.
I'll add that, in my experience, for long necked birds in flight, getting AF to grab the eye is much more difficult, but again, not zero. I'm certain Nikon's engineers are working on improvements.
Like me - an attitude of "where there is a will there is a way" approach can achieve better results than simply blaming the tools at your disposal.

Quote from Harry Skeggs on the role of failure "On a basic level, every failure was an indispensable insight into what went wrong and how to improve."
 
Bearing in mind you were using an f8 lens with which AF is usually not as fast or accurate as using wider apertures, and you had trouble keeping the bird in the frame from a moving boat - your expectations seem realistic - though of course we (including me) could probably get a higher rate with the 400 f2.8 TC and a flat calm sea.
In my post, I expressed no expectations. I stated the gear I used and the conditions in which I used them. And the result was that AF did what was intended, albeit to a limited degree. This was in reply to a prior post which stated something I knew not to be true in my experience. And yes, shooting with higher quality glass and from stable ground would have produced significantly more images in which AF grabbed the eye. I know that from my experience, too.

Like me - an attitude of "where there is a will there is a way" approach can achieve better results than simply blaming the tools at your disposal.
Again, my post was meant to show that a prior comment from a member seemed overly harsh… and from my experience, wrong. Nothing I said should be construed as blaming the tools at my disposal.
 
Last edited:
This thread is primarily about the partially understood machinations of Nikon's Subject Recognition technology, but obviously extends into the encompassing Z9 AF system.

Birds in Flight have become the benchmark to evaluate AF performance, but here the human skills factor (@Steve 's 4" behind the camera) definitely remains extremely important, especially for small, faster, more erratic flying subjects. Photographing fast moving wildlife subjects is highly dependent on the size of the subject in the frame, and the photographer's reactions to track whilst switching settings on the fly (including muscle memory).
The last 2 pages of this thread discuss SR and settings, and also operator influencers


IME, the ability to capture unexpected very fast events are a very tough test for modern ILCs: hunting cats in my experience... The sheer acceleration of leopards or lions, let alone cheetahs (Temperate zone predators must be similarly challenging, e.g. mountain lions, wolves, lynx); these challenges apply especially to capturing sudden ambushes on prey, who also react in very unexpected ways! These rarer incidents are unforgiving tests of one's reactions and abilities to think on the fly, particularly with interfering vegetation, and other obstructions to tracking the subjects.

It also seems debates about Cameras-and-BIF tend to recur every few months, covering much the same topics. This is besides justifiable responses to either new Firmwares, or responding to questions by new members. Reading back through threads, it's interesting how often the same questions, challenges, criticisms etc are revisited. Some of this is the anticipated tyre-kicking chat - including comparing the features/abilities of the primary CaNiSony ILC systems.





 
Last edited:
That would be unacceptable to me. No wonder I bought and subsequently sold the Z9 twice over.
Unfortunately, I have many more such OOF sequences all of osprey and eagles who strike through the catches rather than pelicans, terns, and other vertical diving birds where the AF works reasonably well. Given that this AF "glitch" happens repeatedly and predictably is troubling. Likewise, I never encountered these difficulties when shooting the R3/R5 although I recall similar AF challenges with the Canon R7 (cropped sensor) under like conditions. The R7 AF "glitch" I seemed to attribute to the cropped sensor and the need for greater light for effective AF. Perhaps, that's what's occurring here where a f/6.3 lens with a narrow FOV simply doesn't let in enough light (contrast) for the sensor and logic to differentiate what is occurring as the subject/background are rapidly changing with respect to each other? If this occurs less frequently with a faster lens such as a 400 f/2.8 or 600 f/4, then it might support that assertion?
 
not trying to start anything here, but some random thoughts:

1) all the systems have strengths and weaknesses. even if we took it on the face that nikon was worse at BIF (or specific BIF scenarios) can't really be extrapolated to "nikon's subject detection is worse". it really just means that the specific mode and settings and technqiue and all-the-associated-gear in use isn't as good at a specific scenario vs another camera with all it's associated gear a specific mode and associated settings and technique.

2) it's problematic to compare cameras modes in general. we saw a lot of youtube folks doing af comparisons where they simply picked a mode like auto area on both cameras and claimed one better or worse. the problem with this is we need to leverage domain knowledge of the cameras to figure out how to best shoot a situation with a given camera.

but what i really came to say, and this is just from my own experience of being self taught (not by choice!) to do action photography is to do an analysis loop after each failure. basically look at what happened, try to think about the various possible things that might have gone wrong, try to make a plan how you can test whatever hypothois you come up with, then go try that test, rinse, repeat.

i looked and couldn't find anyone who taught canine action photography, and it's a fair amount different than human sport photography because the subjects are much faster, don't refract light as well, and come in lots of varied shapes and colorings. so basically i've had a lot of failure in my life. 😂 all i could do is try to understand as much as i could about how things worked (thanks Steve for the AF books and videos), then try to guess what might have gone wrong and try to change things up and try again.

and this journey continues. i've gotten a feel for how different things work, but i still continue to have situations i struggle with.
 
I find it interesting one person is having this consistently while others are not. If it was a definite software fault you would think that it would be repeatable across all cameras.

And if this is an issue with one subject and not others, isn’t it a simple thing to just AF handoff to 3d Track before the bird hits the water? I have a 3d track with no subject detection on FN1 specifically for issues with subject detect, I just override. If that fails my other FN2 key is set to dynamic small if I need to zone focus ahead of an action and burst capture it. I don’t think any of these cameras are at the 100% detect all known subjects flawlessly stage of AF quite yet so these situations will occur with different animals/birds that were not machine learned well, across all brands to various extent.

To me the question is by just handing off do you then get reliable results, if yes than it’s only an issue until Nikon machine learns the specific combo you have an issue with.
 
Last edited:
i also wanted to note some things about how complicated this whole subject is.

for example, it's been pointed out that subject detection and af are two different things, which i think is true, but obviously they are intended to work closely together.

as i understand it, we have af sensors built into the image sensor so from a basic perspective, we could assume that subject detection will basically tell the camera which af sensor to use.

but....

as i understand it, it also uses contrast detection.

so basically the camera is using either the af sensor, or contrast detection, or some combination of the two. does it vary how much (like "in this situation i'll use 60% af sensor and 30% contrast detection")?

so one big thing that comes to mind is apeture impacts both subject detection and, we would assume, contrast detection. we can see this when cameras struggle where everything is so out of focus the camera can't tell which way is the right way to get it into focus, and thus has to hunt back and forth.

another situation to think about. so if you have subject partially obscured by foreground foliage, if the way it's working is subject detection decides which af sensor to use and it relies primarily or mostly on the af sensor, there's a fair chance that the af sensor is going to pick up the branch, not the subject since it's going to just find whatever it first sees.

so the only way this works is that it has to switch to favoring contrast detection?

the other big thing i wanted to point out, and this is on my mind a lot based on my subject matter is that we can't forget a big part of af is PREDICTION. when something is coming directly at you, afaik, the camera is basically keeping track of how the subject is moving and moving the focus predictively to meet the subject where it thinks it will be. this seems to work well for items that have a relatively constaint velocity. however, i think this means that we should expect focus misses when things abruptly change velocity (like jumping dogs 🙃).

this has led me to try to keep af engaged as much as possible to let it get the best idea of how the subject is moving. i also suspect the longer you keep af engaged, it seems to get a better idea about what your subject is.

but basically, the more you understand about the af and subject detection systems, the more you realize how complicated it is and how much we don't know
 
Unfortunately, I have many more such OOF sequences all of osprey and eagles who strike through the catches rather than pelicans, terns, and other vertical diving birds where the AF works reasonably well. Given that this AF "glitch" happens repeatedly and predictably is troubling. Likewise, I never encountered these difficulties when shooting the R3/R5 although I recall similar AF challenges with the Canon R7 (cropped sensor) under like conditions. The R7 AF "glitch" I seemed to attribute to the cropped sensor and the need for greater light for effective AF. Perhaps, that's what's occurring here where a f/6.3 lens with a narrow FOV simply doesn't let in enough light (contrast) for the sensor and logic to differentiate what is occurring as the subject/background are rapidly changing with respect to each other? If this occurs less frequently with a faster lens such as a 400 f/2.8 or 600 f/4, then it might support that assertion?
Then why isn't everybody having the same issue? At least not consistently or pretty rarely? Also a couple of my friends shoot with a 600 f4tc and one sees issue and the other gets results like me. So I can't really blame it on the aperture
 
Last edited:
The other thoughts I have is perhaps it is an error in the predictive af algorithms as the subject is moving at a relative speed during the soar, swoop, and dive followed by an acute slowing as it moves through the strike. All of the OOF images appear to be FF'ed so that's what leads me to believe that the predictive af algorithms may be involved. One thing I do need to explore is whether changing the a3 setting from steady to erratic would be helpful or not. It would be interesting to hear from those who shoot large mammals hunting prey whether they use steady or erratic.
 
Then why isn't everybody having the same issue? At least not consistently or pretty rarely? Also a couple of my friends shoot with a 600 f4tc and one seizure issue and the other gets results like me. So I can't really blame it on the aperture
f/4 vs. f/6.3 lens. Alternatively see the comment regarding predictive af algorithms. Either way, this did not occur on my Sony/Canon gear (with the exception of the R7) for these subjects. Interestingly, Canon was known for having poorer tracking for subjects moving directly towards the camera. That was something that they improved and the MILC R5/R3 were much better than the Canon DSLR's in that regard. Again, all of these AF systems have opportunities for improvement.
 
I find it interesting one person is having this consistently while others are not. If it was a definite software fault you would think that it would be repeatable across all cameras.

And if this is an issue with one subject and not others, isn’t it a simple thing to just AF handoff to 3d Track before the bird hits the water? I have a 3d track with no subject detection on FN1 specifically for issues with subject detect, I just override. If that fails my other FN2 key is set to dynamic small if I need to zone focus ahead of an action and burst capture it. I don’t think any of these cameras are at the 100% detect all known subjects flawlessly stage of AF quite yet so these situations will occur with different animals/birds that were not machine learned well, across all brands to various extent.

To me the question is by just handing off do you then get reliable results, if yes than it’s only an issue until Nikon machine learns the specific combo you have an issue with.
I agree and have said as much, that none are perfect.

But I will say that with FW 4.10 with Bird subject detection I have absolutely zero need for the old handoff method. I had been using and doing that since February of 200.2 Since 4.10 I've only needed Auto Area AF mode for recognition, acquisition and tracking. No much so I've dialed all AF modes but Auto Area, 3D and single point. Only needed 3D probably less than 5% of the time
 
f/4 vs. f/6.3 lens. Alternatively see the comment regarding predictive af algorithms. Either way, this did not occur on my Sony/Canon gear (with the exception of the R7) for these subjects. Interestingly, Canon was known for having poorer tracking for subjects moving directly towards the camera. That was something that they improved and the MILC R5/R3 were much better than the Canon DSLR's in that regard. Again, all of these AF systems have opportunities for improvement.
I've have been recommending since FW 3.0 when people have had AF issues that makes a huge difference almost every time, resetting you camera and reinstalling the FW then stopping you camera back up manually(NOT from a saved .bin file). This has fixed many AF issues people were seeing.

Just something to consider since there isn't an issue with the SD it AF algorithms since many people do not have the issue you're seeing. At least more often than just a very random sequence here or there which all systems will do from time to time
 
The other thoughts I have is perhaps it is an error in the predictive af algorithms as the subject is moving at a relative speed during the soar, swoop, and dive followed by an acute slowing as it moves through the strike. All of the OOF images appear to be FF'ed so that's what leads me to believe that the predictive af algorithms may be involved. One thing I do need to explore is whether changing the a3 setting from steady to erratic would be helpful or not. It would be interesting to hear from those who shoot large mammals hunting prey whether they use steady or erratic.
So before even exploring the available settings in the camera you took to the internet to bash Nikon's AF algorithm?
 
i also wanted to note some things about how complicated this whole subject is.

for example, it's been pointed out that subject detection and af are two different things, which i think is true, but obviously they are intended to work closely together.

as i understand it, we have af sensors built into the image sensor so from a basic perspective, we could assume that subject detection will basically tell the camera which af sensor to use.

but....

as i understand it, it also uses contrast detection.

so basically the camera is using either the af sensor, or contrast detection, or some combination of the two. does it vary how much (like "in this situation i'll use 60% af sensor and 30% contrast detection")?

so one big thing that comes to mind is apeture impacts both subject detection and, we would assume, contrast detection. we can see this when cameras struggle where everything is so out of focus the camera can't tell which way is the right way to get it into focus, and thus has to hunt back and forth.

another situation to think about. so if you have subject partially obscured by foreground foliage, if the way it's working is subject detection decides which af sensor to use and it relies primarily or mostly on the af sensor, there's a fair chance that the af sensor is going to pick up the branch, not the subject since it's going to just find whatever it first sees.

so the only way this works is that it has to switch to favoring contrast detection?

the other big thing i wanted to point out, and this is on my mind a lot based on my subject matter is that we can't forget a big part of af is PREDICTION. when something is coming directly at you, afaik, the camera is basically keeping track of how the subject is moving and moving the focus predictively to meet the subject where it thinks it will be. this seems to work well for items that have a relatively constaint velocity. however, i think this means that we should expect focus misses when things abruptly change velocity (like jumping dogs 🙃).

this has led me to try to keep af engaged as much as possible to let it get the best idea of how the subject is moving. i also suspect the longer you keep af engaged, it seems to get a better idea about what your subject is

but basically, the more you understand about the af and subject detection systems, the more you realize how complicated it is and how much we don't know
If I recall correctly, the Z9 used phase detect for AF and contact is only used for confirmation of the AF lock
 
Status
Not open for further replies.
Back
Top