[c.a.m]

For OVF shooting, there is no option specifically called face detection. It just doesn't exist. Yes, there's subject recognition but Pentax does not state anywhere what subjects are actually being recognized. It may be a face, but maybe it's not and it's specifically looking for an eye, who knows.

@Kobie, based on your experience, what is your view on the following statement from the K-3 Mark III Special Site (Challengers):

Image tracking performance The essential feature of a flagship model | Challengers | PENTAX K-3 Mark III | RICOH IMAGING

"The camera also features a new RGB-Ir AE sensor with around 307,000 pixels, in comparison to the RGB sensor with around 86,000 pixels installed on previous models. Thanks to the sensor’s improved resolving power, the camera automatically selects the optimum focus point, after the AE sensor detects the subject’s face and pupils."

I think you're correct about the Operating Manual, and the distinction between OVF and LV 'subject' recognition. There's scant insight into the specific operation of the subject detection/recognition functionality.


- Craig

[Kobie]

@Kobie, based on your experience, what is your view on the following statement from the K-3 Mark III Special Site (Challengers):

Image tracking performance The essential feature of a flagship model | Challengers | PENTAX K-3 Mark III | RICOH IMAGING

"The camera also features a new RGB-Ir AE sensor with around 307,000 pixels, in comparison to the RGB sensor with around 86,000 pixels installed on previous models. Thanks to the sensor’s improved resolving power, the camera automatically selects the optimum focus point, after the AE sensor detects the subject’s face and pupils."

I think you're correct about the Operating Manual, and the distinction between OVF and LV 'subject' recognition. There's scant insight into the specific operation of the subject detection/recognition functionality.


- Craig

Okay, here we go lol....

Pentax cameras can AE for faces from the RGB sensor but that's not regarding AF.
My thoughts on the K-3 III capabilities is that the new RGBir sensor is used for exposure but due to the ir element it can now more accurately find and track faces but then sends the position to the actual AF sensor. Coupled with the deep learning built into the software once it detects a face, it will try to find an eye. However, it is not actual "Face Detection" per se or there would be an option in OVF shooting to specifically turn on/off Face Detection. Subject recognition is basically...
1) Analyze the scene>Scan through programmed subjects to recognize>I see a flower>check database>found it=Set focus to track that subject.
2) Starts the same as above but this time... >I see the shape of a face>scan database for face positions>found them>Now look for eyes that go with face position>Not found>default to highest contrast/closest phase match within AF area.
3) Same as above but... Found eye>follow position of that subject (eye).
4) Same as above but.... Check database for subjects>None found>Default to highest contrast/closest phase match in AF area.

In the way that it works, if you want to stretch it out a bit, yes it does do face detect, but ONLY to find an eye. If it doesn't find an eye, it will default to regular AF parameters without recognizing the subject at all.
When I tested it, I'd get a shoulder, an eyebrow, my nose as primary focus points when it couldn't nail my eye even though my whole face was in the frame.
In Live View Face Detect puts an entire box over the face. OVF does not. It will do a few points seemingly randomized in AF.S comparing it to the Face Detect from Live View.

[licht96]

Good summery again of what I´ve written in different words ;-)

I was fighting too hard to explain it clackers to be precise in mentioning menu topics, so I lost my way.

Two things are interesting for me:

1. what´s new and interesting is, that the AF behavior could be even better in some situations without subject recognition as Eric Seavey mentioned...when I think with all here mentioned things about this new algorithm ... the database ... machine learning...it´s even more clear to me: sometimes normal Phase AF is more or less disturbed by some "information" from the AE sensor...hmmm.
But if it´s affecting only the 5 point Af in this way...?

Someone here (I was in discussion about the same issue) said the complete different thing: that this 5 point AF really NEEDs the AE information...otherwise it would only choose ..may be the nearest point of all 5...or the one with most "phase detect edges or contrast images"


2. as Kobie mentioned: the Live view is giving you this nice box. I use it very often now since I´ve had already mentioned problems with outer AF points in OVF (Phase AF). This problem doesn´t occur in Contrast AF, so I like it now a lot more for portraits.
Is this technology not possible only for the "eye only" like in newer mirrorless Cameras? I know: again some different AF (with sub pixels on the sensor or something like this?)


...But since I can choose a very small box in Live view, that I can move over the eye (and it´s working good as well)...that would be very nice to have this in Live view, too. With an f2,8 lens the box for the face is enough...but with f1,4 the eye in a smaller box would be better because the "face box" choose something in-between the tip of the nose and the ear) Hmmm...what do you think?