jlafferty wrote:
I’m sorry but mannequin “tests” are utter BS IMO. Get a person under the same lighting scenario and show it side by side and if the results are the same, I’ll take my words back, but anything based on a mannequin is silly. Mannequins offer near zero that’s instructive about people photography, even something as base as lighting technique.
What about the mannequins do you believe is different from a human subject that would be significant to a camera in terms of AF?
suteetat wrote:
Problem is that you pick on issues that may happen in small numbers of cases, either in small niche use or for whatever reason is an issue for some users but is not really the normal experience by majority of users and claim that to be the benchmark. There are plenty of people who shoot portraits on this forum. Yes, I noticed some people share similar observation. Many also don't share that observation.
If I go out and shoot a portrait session and found that I get more eyelashes then eyes in focus, I certainly would complain, not just on this forum but directly to Nikon and to someone who has more direct line into figuring out and solving the problem as well. But what can I say, I don't have that problem and it is not surprising that many other people don't have the same problem and are quite sick and tired of some people insisting that this is because of Nikon's inferior AF.
I had an opportunity to spend a few hours in Himba village last week and I was shooting lots of portrait there. To my amusement, my Z8 was struggle once in awhile to detect face and eyes of those people with all their head dress, decorations, hairdos and reddish skin color from their mud/herb stuffs they put on their skin. Was there an easy work around for that, yes. Does this mean that Z8 face/eye detection is really poor? I don't think so. Too bad, for once, except for a Q3, everybody else were shooting Nikon on the trip so no idea how other cameras would handle these people. ...Show more →
I usually pick issues that I find people are complaining about the most, both out of my technical curiosity as an engineer and also a desire to help others. I only came to look at the eye AF after seeing quite a few posts about it. It certainly doesn't mean it fails to perform in all scenarios. But it did so in the ones I tested.
Would you ever mistake a mannequin face for a real human? Does it seem absurd to think anyone would? So… if the AF is trained on a data set that’s specific to real human faces, why would any of the dozen things that make a mannequin look comparatively fake not impact AF reliability?
Again, prove me wrong and I’ll reverse my position: put a relatively similar skin/eye toned mannequin and human face under the same conditions, run a few trial tests, and then compare hit rate and show there’s little to no deviation. Then I’ll reconsider my position.
But until then my hunch is that basing an AF “test” on a mannequin face is absurd at arm’s length.
snapsy wrote:
What about the mannequins do you believe is different from a human subject that would be significant to a camera in terms of AF?
Curious if the super bright EVF will throw off exposure in post. On my Z6, the EVF was brighter than the images came out on the monitor and mostly showed in post.
jlafferty wrote:
Would you ever mistake a mannequin face for a real human? Does it seem absurd to think anyone would? So… if the AF is trained on a data set that’s specific to real human faces, why would any of the dozen things that make a mannequin look comparatively fake not impact AF reliability?
Again, prove me wrong and I’ll reverse my position: put a relatively similar skin/eye toned mannequin and human face under the same conditions, run a few trial tests, and then compare hit rate and show there’s little to no deviation. Then I’ll reconsider my position.
But until then my hunch is that basing an AF “test” on a mannequin face is absurd at arm’s length.
The fact that cameras happily detect faces and eyes on not just 3D mannequins but also 2D photographs of faces argues against the significance of human's being able to distinguish real faces from mannequins having relevance to how and what cameras detect. Everyone is free to their opinions but to call a test BS based on a hunch or intuition isn't very persuasive.
ArizonaImage wrote:
Curious if the super bright EVF will throw off exposure in post. On my Z6, the EVF was brighter than the images came out on the monitor and mostly showed in post.
Interesting question. When I looked at Z6 iii EVF over the weekend and I thought the EVF looked brighter, it is not so much
the absolute brightness of the EVF but the ability to see various shades of black and detail in the shadow that stood out.
I was testing the camera in in door location with a small stage with live music and lots of variable lighting, spot lights etc on stage.
I dont have problem with Z7/ Z7 ii EVF being brighter than actual pictures though. Much will depend on your monitor setup as well. I calibrated my monitor using Calman program and lighting at my computer desk is pretty stable. However, I find image that looks perfectly exposed on my monitor can look too dim on certain phones sometimes.
Is there a real world use case where one would shoot portraits with a DOF so shallow that either the pupils or eyelashes are in focus? Wouldn't that mean that nothing else is in focus - nose, ears, eyebrows, hair, lips? Eyelash tests feel completely contrived to me. What am I missing?
GroovyGeek wrote:
Is there a real world use case where one would shoot portraits with a DOF so shallow that either the pupils or eyelashes are in focus? Wouldn't that mean that nothing else is in focus - nose, ears, eyebrows, hair, lips? Eyelash tests feel completely contrived to me. What am I missing?
You are not missing anything.
But since we have super bright lenses that we pay big money for, we shoot at f1.2 and this results in these kind of discussions.
But... i do also, and I am getting great consistency on people's pupil both in my darkish studio and outside.
So at least I am not able to reproduce the issues that our friend Snapsy has been flooding several threads with for months now. I see overall many people very happy with the AF of the Z8/Z9... and a tiny numbers of unhappy individuals. As far as i recall Snapsy never even had his camera checked by Nikon to see whether it could be defective. He seems to have decided that whatever he experiences with his single sample is representative of how all Nikon bodies behave.
But I have never tried to shoot mannequins. The camera certainly is able to identify a mannequin as a face, but whether the AF fine tuning works identically on a mannequin as it does on a person is anybody's guess until someone does a scientific comparison of both in a single environment with a single camera.
And of course, this is all unrelated to the Z6III.
That’s fine, but claiming a mannequin equates 1:1 with a human face is also “not very persuasive” I also push back against this idea that a camera detecting faces on a 2D surface proves anything apart from an incidental convenience - it’s presumptuous to say it equates with practical insights into how the AF behaves with actual faces.
If anything we’re both guessing as to how the AF is designed and making some speculative assumptions. I would just never count on photographing a mannequin and think I’ve proven anything definitive. Again, until you’ve demonstrated the same behavior carrying over to an actual face under similar conditions IMO you’ve proven nothing of value (unless your job is taking photos of mannequins).
As an aside I’ve got no trouble getting pupils in focus for maybe 85% of the images I take on the comparatively simple Z6. My hit rate with the Z8/Z9 is even higher 🤷🏻♂️
And in real world practical uses, nobody would notice the difference.
snapsy wrote:
The fact that cameras happily detect faces and eyes on not just 3D mannequins but also 2D photographs of faces argues against the significance of human's being able to distinguish real faces from mannequins having relevance to how and what cameras detect. Everyone is free to their opinions but to call a test BS based on a hunch or intuition isn't very persuasive.
jlafferty wrote:
That’s fine, but claiming a mannequin equates 1:1 with a human face is also “not very persuasive” I also push back against this idea that a camera detecting faces on a 2D surface proves anything apart from an incidental convenience - it’s presumptuous to say it equates with practical insights into how the AF behaves with actual faces.
If anything we’re both guessing as to how the AF is designed and making some speculative assumptions. I would just never count on photographing a mannequin and think I’ve proven anything definitive. Again, until you’ve demonstrated the same behavior carrying over to an actual face under similar conditions IMO you’ve proven nothing of value (unless your job is taking photos of mannequins).
As an aside I’ve got no trouble getting pupils in focus for maybe 85% of the images I take on the comparatively simple Z6. My hit rate with the Z8/Z9 is even higher 🤷🏻♂️
And in real world practical uses, nobody would notice the difference.
I'm not claiming a mannequin equates 1:1 with a human face. I'm claiming it equates in all the aspects that would be relevant to a camera's AF system. Again, if you can articulate a theory as to why that might not be the case then I would be genuinely interested in hearing it. When thinking that over, give some consideration to how much diversity there is to the human face, and consider how you believe what's different about a mannequin might fit into that spectrum of diversity.
“I’m not saying they’re the same, I’m saying all relevant aspects are the same” is pedantic.
It’s not my job to convince you that a mannequin isn’t a relevant tool in testing human face AF. It’s your job, and a really low barrier, to prove to us that a mannequin offers any practical insight into human derived AF, and demonstrate that the two track closely. You have only done half a test at best.
Apart from this I find it really strange that you can’t find an actual human to lend 20 minutes of their time to your test, and yet you claim to have proven something useful for the majority of us who photograph… actual people. Nothing you’ve demonstrated on a mannequin is anything I’d count on to inform my workflow or behavior.
snapsy wrote:
I'm not claiming a mannequin equates 1:1 with a human face. I'm claiming it equates in all the aspects that would be relevant to a camera's AF system. Again, if you can articulate a theory as to why that might not be the case then I would be genuinely interested in hearing it. When thinking that over, give some consideration to how much diversity there is to the human face, and consider how you believe what's different about a mannequin might fit into that spectrum of diversity.
snapsy wrote:
I'm not claiming a mannequin equates 1:1 with a human face. I'm claiming it equates in all the aspects that would be relevant to a camera's AF system. Again, if you can articulate a theory as to why that might not be the case then I would be genuinely interested in hearing it. When thinking that over, give some consideration to how much diversity there is to the human face, and consider how you believe what's different about a mannequin might fit into that spectrum of diversity.
RustyRus wrote:
What does any of this have to do with the Z6III ?
The topic of how Nikon's AF compares to other systems came up during the discussion of Jared Polin's Z6 III AF test video. That lead to discussions of people making general claims about AF performance across systems, with the notion that specific elements of performance should be discussed instead. That lead to a discussion of Eye AF, which in turn lead to my tests for Nikon's Eye AF performance.
jlafferty wrote:
Would you ever mistake a mannequin face for a real human? Does it seem absurd to think anyone would? So… if the AF is trained on a data set that’s specific to real human faces, why would any of the dozen things that make a mannequin look comparatively fake not impact AF reliability?
Again, prove me wrong and I’ll reverse my position: put a relatively similar skin/eye toned mannequin and human face under the same conditions, run a few trial tests, and then compare hit rate and show there’s little to no deviation. Then I’ll reconsider my position.
But until then my hunch is that basing an AF “test” on a mannequin face is absurd at arm’s length.
The reason a mannequin was employed was because people were complaining that it was happening on real humans. Mostly Scoombs and astirusty (on DPR). It was a reasonable test, given the circumstances. Regardless, as Groovy stated, stop shooting razor thin depths of field and the problem goes away. There were three 150 posts threads on DPR and a 7 page one here, all because one guy pounds on his hand with a hammer and wonders why it hurts. If I were a Nikon engineer I would be on suicide watch.
Max Power wrote:
The reason a mannequin was employed was because people were complaining that it was happening on real humans.
A leap only an engineer would make 🤦🏻♂️
There were three 150 posts threads on DPR and a 7 page one here, all because one guy pounds on his hand with a hammer and wonders why it hurts. If I were a Nikon engineer I would be on suicide watch.
jlafferty wrote:
“I’m not saying they’re the same, I’m saying all relevant aspects are the same” is pedantic.
It’s not my job to convince you that a mannequin isn’t a relevant tool in testing human face AF. It’s your job, and a really low barrier, to prove to us that a mannequin offers any practical insight into human derived AF, and demonstrate that the two track closely. You have only done half a test at best.
Apart from this I find it really strange that you can’t find an actual human to lend 20 minutes of their time to your test, and yet you claim to have proven something useful for the majority of us who photograph… actual people. Nothing you’ve demonstrated on a mannequin is anything I’d count on to inform my workflow or behavior.
The real challenge would be to use crash test dummies! They have no eyelashes! Just a thought. I’m also considering asking all my talent and customers to cut off their eyelashes in order to take better photos of them. Hopefully, by the time you get all this sorted out, my Z6III will have arrived.
story_teller wrote:
The real challenge would be to use crash test dummies! They have no eyelashes! Just a thought. I’m also considering asking all my talent and customers to cut off their eyelashes in order to take better photos of them. Hopefully, by the time you get all this sorted out, my Z6III will have arrived.
If you can't take the heat get out of the kitchen... and move to the adult film industry where, with a few exceptions, they have had a "no lashes" policy for decades. Now I understand that this came about to help mitigate autofocus problems for the second rate photographers who operate there and tend to take zoomed in pictures of the "eyes". Ahem...🤭
I also push back against this idea that a camera detecting faces on a 2D surface proves anything apart from an incidental convenience - it’s presumptuous to say it equates with practical insights into how the AF behaves with actual faces.
I’m also considering asking all my talent and customers to cut off their eyelashes in order to take better photos of them. Hopefully, by the time you get all this sorted out, my Z6III will have arrived.