[Class A]

... has any company done Eye/Face Detection through the eyepiece without using CDAF?

Yes, Canon and Nikon have done face detection.

As caliscouser mentioned, one can use a high-resolution metering chip to recognise features such as faces and then select one AF point that covers that area.

For this to work well, one needs
1. many AF points, and
2. good coverage of the frame.

Hence the K-new may bring something like PDAF face-AF to the table. We'll see.

[Apet-Sure]

I knew this limitation of PDAF, but had forgotten to remember it. Thanks for the reminder. Bruce, the four shots above do a good job of nailing down the concept. TFP

[clackers]

I knew this limitation of PDAF, but had forgotten to remember it.

PDAF is way ahead of CDAF, Apet-Sure. There may be lines on your subject that PDAF can focus on but CDAF can't because they don't offer significant contrast difference to what's around them.

Mirrorless manufacturers became painfully aware of this problem, because they lack a mirror and dedicated AF module. Most have decided to steal pixels from the sensor (perhaps 12% are affected, Beholder3 estimated) and put them into service as AF strips. An algorithm 'fakes' what goes into your picture instead, like Content-Aware in Photoshop. When it gets it badly wrong, banding and other artefacts result. Mjkoski has noticed banding on IIRC Canon R series when lifting shadows even at ISO100.

Some mirrorless manufacturers have preferred not to hurt image quality and have stuck with CDAF. That's like shooting your Pentax full time in Live View, by the way. Panasonic have chosen to do that. I've shot with a GH5 and the autofocus can't be trusted in low light situations, especially backlit.

And if anyone chooses a Fuji medium format instead of the Pentax 645Z, congratulations, you've bought yourself into slower autofocus with your new toy!

[BruceBanner]

PDAF is way ahead of CDAF, Apet-Sure. There may be lines on your subject that PDAF can focus on but CDAF can't because they don't offer significant contrast difference to what's around them.

Mirrorless manufacturers became painfully aware of this problem, because they lack a mirror and dedicated AF module. Most have decided to steal pixels from the sensor (perhaps 12% are affected, Beholder3 estimated) and put them into service as AF strips. An algorithm 'fakes' what goes into your picture instead, like Content-Aware in Photoshop. When it gets it badly wrong, banding and other artefacts result. Mjkoski has noticed banding on IIRC Canon R series when lifting shadows even at ISO100.

Some mirrorless manufacturers have preferred not to hurt image quality and have stuck with CDAF. That's like shooting your Pentax full time in Live View, by the way. Panasonic have chosen to do that. I've shot with a GH5 and the autofocus can't be trusted in low light situations, especially backlit.

And if anyone chooses a Fuji medium format instead of the Pentax 645Z, congratulations, you've bought yourself into slower autofocus with your new toy!

That's really interesting.

In my 2-3yrs of using K-1 I would say I have found CDAF more accurate, because when it finds focus, locks on, there is never any front or back focusing issues, it's just in focus. However, as you rightly say there are scenarios limiting it's use, poor light being one of them but also distance. You can set the CDAF to being Single Point or SEL but it's still 'too big' a point, so that where you set it to focus on is just picking up too much things. If it were a smaller dot it might be better but as it stands it's quite limiting. Thus I say often to use CDAF for closer work and PDAF for further away stuff.

I too watched a video of a Fuji GH5 user and he struggled to get in focus shots of something as simple as a person walking down the street, a bit worrying indeed.

[Paul the Sunman]

Better late than never to learn how to AF I guess

The eye in the first shot was pretty dark and featureless, so I'm not surprised it wasn't perfect. You've nailed it in the second.

[BruceBanner]

The eye in the first shot was pretty dark and featureless, so I'm not surprised it wasn't perfect. You're nailed it in the second.

Yeah, and that was kinda the point of this thread, how can we better improve our shots for PDAF 'black furry/feathered' things. It really never occurred to me to try shooting the head outline of the animal to get better eye focus than actually just having the point on the eye itself. I wonder if white furry/feathered things have better 'lines' for the PDAF to work when aiming for the eye (dark eye/bright feathers etc), but with black furry feathered creatures the lines are less distinguished.

I'd like to thank Clackers for putting me onto this path, despite he seems to have reached his 'Bruce threshold' currently and hates my guts

[Paul the Sunman]

Yeah, and that was kinda the point of this thread, how can we better improve our shots for PDAF 'black furry/feathered' things.

Depends on the eye and the light. Some eyes are brighter and more strongly featured than others, e.g. ducks, Lorikeets, Tawny frogmouths, or have a sharp contrast with the feathers, e.g., Sulphur Crested Cockies. I find kangaroos particularly difficult.

[Joetitch]

It is refreshing when someone is prepared to learn rather than blame the equipment 😀

[BruceBanner]

Depends on the eye and the light. Some eyes are brighter and more strongly featured than others, e.g. ducks, Lorikeets, Tawny frogmouths, or have a sharp contrast with the feathers, e.g., Sulphur Crested Cockies. I find kangaroos particularly difficult.

Yeah, that's what I'm thinking. Like I said at the beginning, I don't really do wildlife, just playing with a PLM and shooting subjects under lighting that I have less control over than normally, tricky genre indeed.

It is refreshing when someone is prepared to learn rather than blame the equipment 😀

Thanks. I have been well educated in other genres that are nothing related to photography and I can honestly say you can never stop learning or improving, photography is no different I'm sure.

[Class A]

In my 2-3yrs of using K-1 I would say I have found CDAF more accurate, ...

Perhaps you need to perform AF microadjustments for your lenses?

Coincidentally, I just tested three focusing methods today with my K-5 and my Sigma 70/2.8 Macro lens (normal distance, though):

1. PDAF.
2. CDAF.
3. Manual focus, using LiveView and 6x magnification.

The results in terms of sharpness in reverse order of best focus were:

CDAF, focus wasn't terrible but at 100% visibly worse than any of the other two.

Manual focus, very, very close to the best possible result.

PDAF, this method not only nailed the focus 100% but also was super quick.
I used back button focus and confirmed focus multiple times, until the camera wouldn't make any adjustments anymore.

This result is not entirely unexpected because under certain circumstances, PDAF can indeed be more precise than CDAF.

BTW, another disadvantage of the mirrorless on-sensor PDAF approach is that all focus points have a single linear orientation, i.e., they are blind with respect to the orthogonal direction. With such a camera you can be unlucky and not get focus on a horizontal or vertical line that is exactly aligned to the sensor.

Essentially all DSLRs have cross type AF points which are sensitive to two orthogonal directions.

If any manufacturer released a DSLR today without any cross type AF point, DPReview would rake it over lava hot coals.
If a mirrorless camera does the same thing it is "progress".

[caliscouser]

One other point here that gets missed is that the cross-type focus sensor can be bigger than the red focus point square in the view finder.

You can test this yourself at home with a piece of paper and some marker lines drawn on it.

When one goes to focus on something like a birds eye at a distance and the red focus square completely covers the birds head, it may actually focus on something higher contrast behind the birds head like a a branch.
With this realization it becomes much easier to pick places to focus on in a scene.

I've noticed this more on the KP than my K-5iis.

[BruceBanner]

Perhaps you need to perform AF microadjustments for your lenses?

Coincidentally, I just tested three focusing methods today with my K-5 and my Sigma 70/2.8 Macro lens (normal distance, though):

1. PDAF.
2. CDAF.
3. Manual focus, using LiveView and 6x magnification.

The results in terms of sharpness in reverse order of best focus were:

CDAF, focus wasn't terrible but at 100% visibly worse than any of the other two.

Manual focus, very, very close to the best possible result.

PDAF, this method not only nailed the focus 100% but also was super quick.
I used back button focus and confirmed focus multiple times, until the camera wouldn't make any adjustments anymore.

This result is not entirely unexpected because under certain circumstances, PDAF can indeed be more precise than CDAF.

BTW, another disadvantage of the mirrorless on-sensor PDAF approach is that all focus points have a single linear orientation, i.e., they are blind with respect to the orthogonal direction. With such a camera you can be unlucky and not get focus on a horizontal or vertical line that is exactly aligned to the sensor.

Essentially all DSLRs have cross type AF points which are sensitive to two orthogonal directions.

If any manufacturer released a DSLR today without any cross type AF point, DPReview would rake it over lava hot coals.
If a mirrorless camera does the same thing it is "progress".

I make a point of Fine Adjustment all my lenses, my FA77 is the 'worst' needing a whopping +8!

I am merely saying you need to choose the right focusing method for the right scene, there is just no one 'best' mode. We also need to mention technique also as this can assist with success/results. For modes like CDAF and Face Detection, for example I have Focus Priority selected and never bother with this half pressing shutter first shenanigans, I see the yellow box present itself I just fully press the shutter button, which means it will not take the photo till it has become a green box (found focus), and by doing this there is no delay (which increases the chance of oof shot because of the time delay between finding original focus and chances of subject moving slightly by the time you take the shot). This I found especially true for kids who move a lot compared to adults which will hold a pose longer.

When using ocf and needing to be aware of surroundings, I have found Face Detection nice and helpful, sometimes shoving the eye down a viewfinder is not advisable, such as shooting at a beer festival or wedding and needing to be aware of 'tipsy' patrons

I have also found CDAF AF Tracking to be of real value when taking shots where the camera is in an awkward position and its not possible to get your eye through the OVF, and where it isn't managing to find Face Detection. Examples;



It might not look like it here, but the KP with the Sigma 10-20 was very low down on the ground, around shin height, flipping the screen out helps hugely but FD was struggling so I used the AF Tracking Mode which worked quite well in this instance (other times it can wander a bit).

I probably know CDAF better than PDAF from the type of work I have done thus far, a lot of portraiture and event work where I have the camera at funny angles and can't get my eye down the view finder, another example;



I made a 'short' video sometime ago to talk through all the CDAF modes in greater detail and where they might be useful;

Youtu.be

Youtu.be

I find CDAF is quite good for those shots where the subject is not in centre of the frame, anytime the camera focuses on something to an extreme left or right you get better focus (and you have more AF Points to work with than PDAF). The fact the shots never seem to suffer front or back focus is a huge bonus imo.

As I said at the start of this video, I don't really shoot wildlife much, this year is the first time I have really had a AF lens over 77mm to work with, and so with the 55-300PLM I have definitely been shooting PDAF a lot more and studying the results of using that lens and those AF modes more. I think just the very nature of holding the camera out in front of you in LV mode at 300mm is a huge turnoff in and of itself, there's something natural about wanting to have it close to the eye.
I replaced my focus screen with a Canon S Type Matte because I am using manual focus a lot more. I feel quicker to nailing focus now with the FA77 and Takumar 50 through the eyepiece, if the subject focus is further away LV with magnification is still a necessity with MF glass. Manual Focus I think is always the best and most accurate but often the slowest way to focus. Ironically, when I have used the eye piece at events such as weddings I would often be using manual focus as I found I could obtain focus off centre point faster than moving an AF point with the directional pads as well as keeping the screw drive AF noise down. The noise of the lenses is a bit overrated however, it feels a moot point when you actually hear the sound of the shutter in the K-1! The KP is the one true silent camera I have, using that camera in OVF mode with ES selected, dead silent shooting (albeit at the cost of a temporary black screen as it takes the shot).

I just believe the camera is a tool, when you get to know it more you can choose the function that best suits your scenario. It's probably better to become proficient with all shooting modes and options than not (become a master of one mode such as OVF for example), because sometimes you might be placed in a position where using your favourite mode (OVF) is not an option, and fumbling around choosing a better AF mode quickly can come at the risk of losing the shot. But that's just my 2 cents.

On other point here that gets missed is that the cross-type focus sensor can be bigger than the red focus point square in the view finder.

You can test this yourself at home with a piece of paper and some market lines drawn on it.

When one goes to focus on something like a birds eye at a distance and the red focus square completely covers the birds head, it may actually focus on something higher contrast behind the birds head like a a branch.
With this realization it becomes much easier to pick places to focus on in a scene.

I've noticed this more on the KP than my K-5iis.

Good to know. I always wondered about the box sizes and whether they were accurate or not.

[gatorguy]

The noise of the lenses is a bit overrated however, it feels a moot point when you actually hear the sound of the shutter in the K-1! The KP is the one true silent camera I have, using that camera in OVF mode with ES selected, dead silent shooting (albeit at the cost of a temporary black screen as it takes the shot).

Your comment reminded me of a model shoot I did last year.

I like in-studio professional or semi-pro models to do what they do with minimal direction, more just positive feedback, after the initial pre-shoot discussions. That's why they are paid models, and I'm accustomed to keeping up with them rather than them waiting on me. In this particular shoot I was using my KP more so than K70's. Early on I had to direct this model much more than I had with previous shoots. After 20 minutes or so I got just a tad short with her and asked why she was holding poses so long. Her answer: She didn't know I had already taken my shots since she wasn't hearing a shutter. My bad.

IMO shutter sound is an aid in the studio and on-location, and models are more comfortable hearing them.

[BruceBanner]

Your comment reminded me of a model shoot I did last year.

I like in-studio professional or semi-pro models to do what they do with minimal direction, more just positive feedback, after the initial pre-shoot discussions. That's why they are paid models, and I'm accustomed to keeping up with them rather than them waiting on me. In this particular shoot I was using my KP more so than K70's. Early on I had to direct this model much more than I had with previous shoots. After 20 minutes or so I got just a tad short with her and asked why she was holding poses so long. Her answer: She didn't know I had already taken my shots since she wasn't hearing a shutter. My bad.

IMO shutter sound is an aid in the studio and on-location, and models are more comfortable hearing them.

Haha, that's funny and a good point!

[JensE]

(Edit: I wrote that without noticing that there's a second page - but the stuff on the Pentax page is still worth knowing.)
Now you're well-prepared to digest How to optimize focusing accuracy with large-aperture lenses / explore, Part 1| RICOH IMAGING, including the difference between cross- and line sensors. The note on the central horizontal f/2.8 line sensor, which detects horizontal changes, i.e. vertical edges, is noteworthy.