[mst]

Falsehood. Gets another one? They're using the metering sensor.

Using the metering sensor for something it isn't designed for (and for which way better alternatives exist) is a workaround. Not a falsehood.

[Klaus]

i

Fact #3: Eye detection in DSLRs doesn't make any sense - albeit it's surely possible. It does require the main sensor plus a secondary, relatively high-resolution sensor for AI analytics. So a camera design that already requires a couple of strange workarounds (mirror, prism, AF sensor, exposure meter) gets another one.

Falsehood. Gets another one? They're using the metering sensor.

This doesn't negate the fact that the extra real estate for the exposure meter isn't needed for the exposure function ;-)

[Kunzite]

Falsehood. Gets another one? They're using the metering sensor.

Using the metering sensor for something it isn't designed for (and for which way better alternatives exist) is a workaround. Not a falsehood.

That's a red herring, the claim Klaus made is that it "gets another one". That - not your red herring - is a falsehood.
And care to explain why the 307,000 RGBIr pixels sensor is not designed for subject recognition?

---

i

Fact #3: Eye detection in DSLRs doesn't make any sense - albeit it's surely possible. It does require the main sensor plus a secondary, relatively high-resolution sensor for AI analytics. So a camera design that already requires a couple of strange workarounds (mirror, prism, AF sensor, exposure meter) gets another one.

Falsehood. Gets another one? They're using the metering sensor.

This doesn't negate the fact that the extra real estate for the exposure meter isn't needed for the exposure function ;-)

First, your "gets another one" claim was false. Now you're making another claim about "extra real estate". Are you sure the new sensor is larger than it would have been just for metering? We're not talking about center weighted, by the way.

It's weird. I'm discussing with knowledgeable people - you should know these things better than me - but it doesn't feels that way.

[mst]

And care to explain why the 307,000 RGBIr pixels sensor is not designed for subject recognition?

Subject recognition, yes. Nikon has been doing this for ages, even with less resolution. Their 3D AF tracking is based on it (and works very well, btw.)

Face recognition: maybe.

Eye recognition: certainly not. The resolution to do that reliably (!) is simply not there. And there's one more glitch: even if the metering sensor properly detects an eye, there still needs to be an AF sensor at that spot to actually focus on it. With typical AF systems, chances are, there is none. And if if there happens to be one at exactly the desired spot, chances are it's too big to cover the eye only. Or, in other words, to reliably focus on the eye, not the eye lid or the lashes.

[Klaus]

I reckon this is what you can expect from a 180k pixel DSLR eye AF:

https://www.youtube.com/watch?v=WUK2APvbPuo

It locks on "late" - as to be expected at this resolution.

[Kunzite]

And care to explain why the 307,000 RGBIr pixels sensor is not designed for subject recognition?

Subject recognition, yes. Nikon has been doing this for ages, even with less resolution. Their 3D AF tracking is based on it (and works very well, btw.)

Face recognition: maybe.

Eye recognition: certainly not. The resolution to do that reliably (!) is simply not there. And there's one more glitch: even if the metering sensor properly detects an eye, there still needs to be an AF sensor at that spot to actually focus on it. With typical AF systems, chances are, there is none. And if if there happens to be one at exactly the desired spot, chances are it's too big to cover the eye only. Or, in other words, to reliably focus on the eye, not the eye lid or the lashes.

Nikon does face recognition with less pixels. Check the specs for e.g. the D850:
https://www.nikonusa.com/en/nikon-products/product/dslr-cameras/d850.html#tab-ProductDetail-ProductTabs-TechSpecs

Pentax does (in the K-3iii) eye recognition with 307,000 RGBIr pixels. Remember that thomass said "it's not possible"? It is. You're still debating around this fact.
How well it works in this implementation, it remains to be seen... but this isn't the best a DSLR can do! And not all mirrorless can reliably focus on the eye instead of eye lid/lashes....

We would make progress... if only you'd start admitting that some assumptions about DSLR's limits are false.
Then, there's the elephant in the room - that DSLRs can evolve beyond the current configuration. Shouldn't we wish them to do so?

---

I reckon this is what you can expect from a 180k pixel DSLR eye AF:

https://www.youtube.com/watch?v=WUK2APvbPuo

It locks on "late" - as to be expected at this resolution.

Congratulations. You just proved thomass' claim (that "it isn't possible") a falsehood.

[mst]

Pentax does (in the K-3iii) eye recognition with 307,000 RGBIr pixels. Remember that thomass said "it's not possible"? It is. You're still debating around this fact.

No, you are. I'm trying to tell you that, no matter what the manufacturers claim, even with a 300k sensor it will not be usable in real life. And that's not an assumption, it's my opinion based on more than 10 years of experience with Nikon 3D tracking.

You are clinging to the words Thomas used ('is not possible'). Feel free to insist that you believe it is possible, because some marketing department tells you so (Pentax and/or Nikon). Allow me to still disagree based on the experiences I made with the available systems. I am pretty sure that is what Thomas intended to say: with the current technology, it's not possible to do in a way that is actually working and/or useful.

How well it works in this implementation, it remains to be seen... but this isn't the best a DSLR can do! And not all mirrorless can reliably focus on the eye instead of eye lid/lashes....

The D6 AF is generally regarded as one of the best there has ever been. So, the youtube video gives you an impression what one of the best AF modules on the market can do: it can not reliably focus on a subject's eye. And no one actually using the camera would expect that, to be honest. That's not what 3D tracking was made for... it marvels in other regards, though.

Then, there's the elephant in the room - that DSLRs can evolve beyond the current configuration. Shouldn't we wish them to do so?

Yeah, I'd love to see that. The issue is: why would any of the manufacturers (all of which seeing a severe market decrease, some of them in severe trouble already) invest into that, when the whole market is moving in a completely different direction? That's why I wrote about the dead-end: not because there is no potential left to improve DSLRs. But because only a minority of potential buyers is actually interested in it. And because there is a much bigger market now (mirrorless) with bigger opportunities to earn money (at all, in some cases... in other words: to survive as a brand).

Spoiler: it's not just about the cameras, the new mounts allow for much better designs. Any of the Z mount primes or high end zooms I've used so far make their F-mount counterparts look really poor.

[davidmanze]

And care to explain why the 307,000 RGBIr pixels sensor is not designed for subject recognition?

Subject recognition, yes. Nikon has been doing this for ages, even with less resolution. Their 3D AF tracking is based on it (and works very well, btw.)

Face recognition: maybe.

Eye recognition: certainly not. The resolution to do that reliably (!) is simply not there. And there's one more glitch: even if the metering sensor properly detects an eye, there still needs to be an AF sensor at that spot to actually focus on it. With typical AF systems, chances are, there is none. And if if there happens to be one at exactly the desired spot, chances are it's too big to cover the eye only. Or, in other words, to reliably focus on the eye, not the eye lid or the lashes.

     Err ..... cuckoo surprise surprise !! ....... the D750 has "face" and "eye recognition" ..... and it works !! ...... Nikon never made claims for it ..... but it's there in "auto area mode" ....   


 Here's DPR's video from the D750 review:

 https://www.youtube.com/watch?v=wQ4rEvbCNuw&t=1s

  In fact I've just tried it again with the D750 ...... 
  .... "3D AF" uses the initial focus point to recognize the subjects eye; then locks and sticks to it as one would expect ....... 

........ Auto area mode finds the nearest eye "by itself" !!  ....... the video is using "auto area mode" btw. !!

..... the D500 also has it (just checked) straight to the nearest eye !!
  
..........  just like a Sony A9 ......

Check out your D850 ....... you may surprised, even delighted that you have what many thought was a MLC only feature !!

...... mst .......... now you have two DSLRs with eye detect ....... cutting edge stuff

[mst]

Nope, I know the feature. In fact, the video pretty much proves the points I made above and matches my experience: in many cases, the AF is not spot-on.

With a Z6 or Z7, it would.

[Kunzite]

Pentax does (in the K-3iii) eye recognition with 307,000 RGBIr pixels. Remember that thomass said "it's not possible"? It is. You're still debating around this fact.

No, you are. I'm trying to tell you that, no matter what the manufacturers claim, even with a 300k sensor it will not be usable in real life. And that's not an assumption, it's my opinion based on more than 10 years of experience with Nikon 3D tracking.

I'm debating the fact, you're debating around it. And... see davidmanze's post; that's the D750, with a third of the pixels available to the K-3iii.

You are clinging to the words Thomas used ('is not possible'). Feel free to insist that you believe it is possible, because some marketing department tells you so (Pentax and/or Nikon). Allow me to still disagree based on the experiences I made with the available systems. I am pretty sure that is what Thomas intended to say: with the current technology, it's not possible to do in a way that is actually working and/or useful.

That's reality denial. The feature is implemented and working.
And thomass didn't say that. If he did, my answer would be to see the K-3iii before dissing it (or something on that line).

Yeah, I'd love to see that. The issue is: why would any of the manufacturers (all of which seeing a severe market decrease, some of them in severe trouble already) invest into that, when the whole market is moving in a completely different direction? That's why I wrote about the dead-end: not because there is no potential left to improve DSLRs. But because only a minority of potential buyers is actually interested in it. And because there is a much bigger market now (mirrorless) with bigger opportunities to earn money (at all, in some cases... in other words: to survive as a brand).

Propaganda again.
Pentax believes in the future of the SLR photography. They decided that, instead of going mirrorless - first because there's a need for DSLRs, and even if it becomes a niche that's fine for Pentax. And second, because earning money with mirrorless is actually very very difficult.
Haven't you seen what happened with Olympus? Nikon's troubles? Samsung's exit? But the propaganda doesn't tell this; for it, mirrorless is a panacea.

Spoiler: it's not just about the cameras, the new mounts allow for much better designs. Any of the Z mount primes or high end zooms I've used so far make their F-mount counterparts look really poor.

And now you're changing the goalposts again, to a very different subject. You're on a mission.
By the way, both the D FA* primes are excellent.