MorganP
Well-known member
Z8 How well dose Eye subject Detection work in real life? I'm talking about Elk, Deer, Moose, Big Horn, Marmots..... Any insight you wish to share.
Z8 How well dose Eye subject Detection work in real life
- Thread starter MorganP
- Start date
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).
In my experience it works quite well as long as the subject is a decent size in the frame. I've had some mammals with relatively large round ears like Pika and Nutria sometimes confuse eye detection but that can usually be resolved by going to a very small custom C1/C2 area like a 1:1 focus area. If the animals are way off and tiny in the frame or worse if they're in tangled brush then it doesn't always nail the eye right away but in general I find subject and eye detection works very well with both the Z8 and Z9. The addition of dedicated Bird subject detection mode helped a lot but even when I sometimes forget to toggle when say a mammal shows up and I'm shooting in bird mode or vice versa it tends to work quite well.
For the mammals you mention: Elk, Deer, Moose, Big Horn Sheep, Marmots I haven't had any trouble with subject or eye detection at all even when they're on the run.
Morgan, I have wondered the same thing. My Z6ii does a great job on static animals. I have been tempted to add or upgrade to a Z8. When fall rolls around, I know that I will be presented some great opportunities.
DRwyoming, good info.
DRwyoming, good info.
In my experience, not as well as promoted. I rented a Sony A7rV a while back and it was unbelievably good, no matter how small the subject in the frame was. On my Nikon Z8, I have to be much closer to the subject. I still love my Z8 and since I often use single point focus and place the square myself, having eye detect focus is not as important to me. But if that is a crucial factor for you then maybe there are better options like the aforementioned Sony.
Dionigi Colombo
Well-known member
The eye recognition of the Z9, which is identical to the Z8, on mammals such as marmots, deer, chamois, ibex, and roe deer works rather well only if the subjects are large enough in the frame, otherwise better to rely on the single point without subject recognition.
For large mammals, I find the Z8 works very well. There are several ways to improve focus if you are having a problem.
- Try to have a reasonable size subject in the frame. A bear a half mile away is not large enough. The subject needs to be obvious or AF may seek a better subject. Using DX mode may help with a small subject because it makes the subject larger in the EVF.
- Use a smaller AF area if you are having problems with focus. It helps the camera to know where the subject is, and limits seeking other areas.
- Try to have backgrounds that let the subject stand out. Busy backgrounds provide more distractions.
- Properly expose your subject. If the subject is underexposed, the camera may focus on the body or another target rather than the head or eye.
- Pre-focus near your subject. It makes it easier to recognize the subject. If the subject is completely out of focus, the camera may not see your subject. It focuses with the image you see in the viewfinder.
- Be willing to use manual focus override if the camera can't find your subject or target.
- Have a Plan B for focus. I use Dynamic Small if the camera needs to focus through branches, grasses, etc.
- Practice - for fast moving subjects it takes practice to stay on the subject.
SCoombs
Well-known member
I have found that there are two layers to this question.
The first is how well the camera locates an animal's eye and attempts to focus on it, and I'd say that this works quite well. It isn't perfect, but it can usually identify the eye for me even at relatively small sizes.
The second is how well focus is actually acquired once it identifies the eye. This is where I find the system will miss at times, but usually for a specific reason. A key factor that I've found is that it tends to be much more hit or miss with eyes which are in darker areas of the face. For instance, on red winged blackbirds or sometimes geese I get hit or miss results since the eye area of the face is quite dark. This doesn't mean it misses horribly, just that focus can be slightly off the ideal point. Usually if I shoot a burst of photos in this situation I will get a number of photos where the focus is just how I'd like it and others where it's just off. With animals with a wider range of colors in the face this doesn't really happen very much.
The first is how well the camera locates an animal's eye and attempts to focus on it, and I'd say that this works quite well. It isn't perfect, but it can usually identify the eye for me even at relatively small sizes.
The second is how well focus is actually acquired once it identifies the eye. This is where I find the system will miss at times, but usually for a specific reason. A key factor that I've found is that it tends to be much more hit or miss with eyes which are in darker areas of the face. For instance, on red winged blackbirds or sometimes geese I get hit or miss results since the eye area of the face is quite dark. This doesn't mean it misses horribly, just that focus can be slightly off the ideal point. Usually if I shoot a burst of photos in this situation I will get a number of photos where the focus is just how I'd like it and others where it's just off. With animals with a wider range of colors in the face this doesn't really happen very much.
Sean
That's a really good description. I've also found that underexposure or low light can cause it to miss the eye and possibly the head, but it usually finds the body. With many subjects, finding the body is not good enough. There seems to be a bit of random variation as well - a burst or longer series may have some misses that are unexplained, but I suspect it is related to light level or contrast.
Cameron T
Well-known member
Yes, but the manual explictly calls out that af might struggle with low contrast situations. I'm not sure this should be a shock to anyone
They only promise dogs, cats, and birds in the manual. Quote:
If a dog, cat, or bird is detected when [Animal] is selected, the focus point will appear over the face of the animal in question (animal-detection AF). If the camera detects the subject’s eyes, the focus point will instead appear over one or the other of their eyes. If the camera can detect neither face nor eyes, it will display a focus point over the detected animal.
They follow up with this:
Cautions: Animal-Detection AF
Subject detection may not perform as expected if:
the subject’s face is too large or small relative to the frame,
the subject’s face is too brightly or dimly lit,
the subject’s face or eyes are obscured by fur or the like,
the subject’s face and eyes are of similar colors, or
the subject moves excessively during shooting.
The camera may display a border around subjects that are not dogs, cats, or birds but which resemble these animals.
The light from the AF-assist illuminator may adversely affect the eyes of some animals; when using animal-detection AF, select [OFF] for Custom Setting a12
To paraphrase the old song: I beg your pardon, I never promised you the moose focus....
If a dog, cat, or bird is detected when [Animal] is selected, the focus point will appear over the face of the animal in question (animal-detection AF). If the camera detects the subject’s eyes, the focus point will instead appear over one or the other of their eyes. If the camera can detect neither face nor eyes, it will display a focus point over the detected animal.
They follow up with this:
Cautions: Animal-Detection AF
Subject detection may not perform as expected if:
the subject’s face is too large or small relative to the frame,
the subject’s face is too brightly or dimly lit,
the subject’s face or eyes are obscured by fur or the like,
the subject’s face and eyes are of similar colors, or
the subject moves excessively during shooting.
The camera may display a border around subjects that are not dogs, cats, or birds but which resemble these animals.
The light from the AF-assist illuminator may adversely affect the eyes of some animals; when using animal-detection AF, select [OFF] for Custom Setting a12
To paraphrase the old song: I beg your pardon, I never promised you the moose focus....
Last edited:
True, but I have experienced continual improvement with each new firmware update. Yesterday I photographed ducks on a lake in which light levels were changing rapidly and there were ducks near to shore and quite far away. I had few problems with head/eye detection unless the ducks were in full shade with low contrast. Then I had to use single point.
ajrmd
Well-known member
True, though it's still not as agile or accurate as my Canon or Sony gear. They were able to "eye detect" at further distances and held onto the eye better, though as you observed, the Nikon has improved. Compared to my Canon gear, for example the Nikon has fewer AF points and lacks dual pixel. Obviously, this tradeoff is offset by the stacked sensor, lack of rolling shutter, EVF delay, and better buffer management.
MorganP
Well-known member
Thank you all for your time in replying.
TimC
New member
Here's my experience: it works well enough that I routinely use it. This was not the case with my z8ii. It always works superbly when the shot is great--close enough, well lit, not too much in the way. When it doesn't grab the eye, it generally is NOT going to be a great shot--neither the camera nor my viewers can adequately identify the eye. This is usually because my subject is positioned so their eyes are in the shade. (I think they don't want to look into the sun, either). There are always exceptions. In fact, I hit the shutter less and less when the subject ID is struggling or just ID's the body...because it will often be marginal and I have a computer disc full or marginal wildlife shots.
macwalter
Active member
I find that often when the subject is a bit far off but still close enough to photograph and the camera has located the eye, it will still not always give a tack sharp shot. I think often it is due to the fact that the actual autofocus point even on single point size allows some overlap with other things in the image. I do find if I shoot short bursts, usually some shots will give me what I want. Dark subjects tend to be less reliable but that is what you get in many cases. Manual override is essential in thick brush etc. I can’t speak for other platforms like Canon and Sony but for me having affordable high quality glass makes me stay with Nikon. I find the z8 is so vastly superior to the d850 which I loved for years.
In other words, I need to be using my D850!
macwalter
Active member
I can say having used my d850 a few weeks ago while the z8 was in the shop was a great reminder of why I have switched to mirrorless. I used the d850 exclusively since it came out and loved it but compared to the z8 it doesn’t even come close. Not sure what others are photographing but for Me the z8 is a vast improvement over dslr. i Mainly do wildlife and love fast action.
MorganP
Well-known member
Thanks for replying.
fcotterill
Well-known member
Official disclaimer no less, no more.
Outside of Nikon's Autofocus laboratories, no one has much clue about the specifics in how the engineers/software geeks are training the Deep-learning algorithms, which we use in compiled software in our Nikon's cameras.
However it's safe to say that the Z9 AF engine must be built upon a wide diversity of Bayesian Priors, these will be undergirding its abilities to recognize a wider variety of subjects in the wild.
They likely use at least one supercomputer cluster, "fed" with training materials comprising a diversity of imagery. Pure speculation, but to match the real world situations, images of the same subject must vary in orientation, tones, eye/headlight orientations etc. This is besides presenting a range of subject categories in each of the 9 classes. The Animal and Bird categories must have become very complicated and data intensive - the software has had to learn a variety of taxonomic categories ie many different species (body shapes, eye morphology etc).
And quite possibly, imagery is presented to replicated copies of a robotic controlled camera-lens apparatus. Moreover, the training will have to involve dynamic footage of different subjects to train the Z9 AF engine on moving subjects (athletes, vehicles, BIF etc, besides birds, dogs, cats etc etc).
Over several years, Nikon engineers would have trained the 3D AF mode in the D5 and D6 AF engines to track athletes. It follows the Z design engineers will have had the choice to build the Z subject detection algorithm(s) on a platform of Pro DSLR AF pattern recognition software; possibly expanding the subject databases to the other categories besides athletes.
Last edited:
fcotterill
Well-known member
Considering the incremental advances we've seen in firmware releases, the Z9 Subject Recognition is a work in progress....this only makes sense if the superclusters have continued to train it's Deep-learning algorithms. Nikon admits very little, but it's possibly similar to how Canon engineers refines their eye recognition Deep-learning algorithms.
Each Z9 firmware is a window into a relentless learning progress - in a way, each updated release of compiled software code is rather like a pupil's graduation through a succession of classes. And the Z8 is following it's older sibling - it graduated recently as Version 2.0
Photographers can expect these AF engines to continue to be improved.... More robust SR in challenging scenes, "knowing" a greater diversity of subjects, so it will get 'Stickier' - including on eyes etc.
Logically Nikon R&D will be prioritizing its Deep-learning investments to strengthen and advance this keystone asset of its Imaging Division. They are building Dynamic subject Recognition into other industrial applications, including manufacturing and biomedical technology.
www.microscope.healthcare.nikon.com
www.microscope.healthcare.nikon.com
Each Z9 firmware is a window into a relentless learning progress - in a way, each updated release of compiled software code is rather like a pupil's graduation through a succession of classes. And the Z8 is following it's older sibling - it graduated recently as Version 2.0
Photographers can expect these AF engines to continue to be improved.... More robust SR in challenging scenes, "knowing" a greater diversity of subjects, so it will get 'Stickier' - including on eyes etc.
Logically Nikon R&D will be prioritizing its Deep-learning investments to strengthen and advance this keystone asset of its Imaging Division. They are building Dynamic subject Recognition into other industrial applications, including manufacturing and biomedical technology.
'How deep learning is used within microscopy' From Science and Nikon Instruments Inc.
Our goal in this supplement is to help inform readers of the possibilities presented by deep learning–based imaging and analysis methods for microscopy. This e-book includes an educational primer on…
www.microscope.healthcare.nikon.com
Nikon Announces Artificial Intelligence (AI) for Predictive Imaging, Image segmentation and Processing
Nikon Instruments Inc., innovator of advanced microscope systems, today announced NIS.ai, a powerful image analysis and processing module for NIS-Elements that leverages Deep Learning and Artificial Intelligence. NIS.ai is a…
www.microscope.healthcare.nikon.com
Last edited: