The K-3 multisegment meter is too smart for me

asp1880 · 09-02-2014, 01:26 PM

Since getting a K-3 I've been giving the multisegment meter another chance. I didn't like the K-5 multisegment meter, I found it too unpredictable and just used centerweighted and exposure compensation instead.

The K-3 meter is better than the K-5 one, no doubt about that. It chooses brighter exposures that push further to the right of the histogram and it's clever about dark backgrounds and uneven lighting.

But then this happened the other day that left me scratching my head.

The following two crappy snapshots are taken within three seconds of each other, on shutter priority and auto ISO. I used the viewfinder, not live view.
Metering was multisegment, exposure compensation was 0 for both shots.

The framing, light and everything are essentially identical, yet the latter shot is 2 stops brighter than the first!

1/125, f/2.8, ISO 160

1/125, f/2.8, ISO 640

What is going on here? Did the multisegment metering see the face in shot 2 and decided to expose for it? Does the K-3 do that, like the Nikon D800?

If this behaviour is commonplace (I've seen it twice now), I think I'm going back to centerweighted.

Regards,
--Anders.

Last edited by asp1880; 09-02-2014 at 01:28 PM. Reason: clarified viewfinder use.

Adam · 09-02-2014, 01:38 PM

The K-3 doesn't have face detection through the viewfinder, so I think it's just the scene as a whole that affected the metering result. After all, the camera has no way of knowing what you want to ensure is properly-lit, so it basically has to guess based on some simple heuristic.

rawr · 09-02-2014, 02:40 PM

Originally posted by Adam

The K-3 doesn't have face detection through the viewfinder

While it may not have proper face detection via PDAF, there is a possibility that the metering system could be programmed to detect skin tones.

In these pictures, I think the issue is that the subject is mobile and dynamic, as is the lighting.

The histograms for both photos would be interesting.

stevebrot · 09-02-2014, 02:53 PM

Originally posted by Adam

The K-3 doesn't have face detection through the viewfinder, so I think it's just the scene as a whole that affected the metering result.

I agree, though I too believe that there is logic in place based on color mix and also on small-scale contrast mix. I have found it to be very sensitive to small changes in framing with some subjects, though the result is usually nicely acceptable.

Steve

aleonx3 · 09-02-2014, 03:02 PM

If you have this setting "Link AE to AF point" set to ON for multi-segment metering, then the exposure may change due to the focus point changed (in this case, perhaps, the face).

stevebrot · 09-02-2014, 03:08 PM

Originally posted by asp1880

What is going on here? Did the multisegment metering see the face in shot 2 and decided to expose for it?

According to the exif the image is an in-camera JPEG. Face Detect was turned ON with one face detected at a location appropriate to the subject. That might explain it.

Steve

Unregistered User · 09-02-2014, 03:28 PM

Originally posted by stevebrot

According to the exif the image is an in-camera JPEG. Face Detect was turned ON with one face detected at a location appropriate to the subject. That might explain it.

Steve

May I ask what program you are using for reading the exif?

stevebrot · 09-02-2014, 05:09 PM

Originally posted by Tjompen1968

May I ask what program you are using for reading the exif?

Phil Harvey's ExifTool.

Steve

---------- Post added 09-02-14 at 05:39 PM ----------

Hmmm...

I have been able to replicate "face detect on with live view off", but don't see that it makes any difference in exposure.

Steve

**southlander** · 09-02-2014, 06:30 PM

Originally posted by stevebrot

Hmmm...

I have been able to replicate "face detect on with live view off", but don't see that it makes any difference in exposure.

Steve

It would be sensible programming that, if face detect was switched on, metering would give priority for correctly exposing the area it detected as representing the face. Afterall, if focus was asked to be prioritised to a face, then it would be logical Captain for the user to also want correct exposure for the face to a priority. In shot 1, with only one eye visible, this would have been unlikely to have been identified as a face so the metering most likely defaulted to prioritising general protection of highlights and gave a shorter exposure.

gmans · 09-02-2014, 06:59 PM

T

Originally posted by asp1880

The following two crappy snapshots are taken within three seconds of each other,

The first shot looks good to me, the exposure is very good, the soft window light on the boys face is a great effect. The second shot is really blown and I have had similar experiences with my K3 overexposing. I thought it was something to do with fast lens and centre weight compensation but only guessing.

stevebrot · 09-02-2014, 07:35 PM

Originally posted by southlander

It would be sensible programming that, if face detect was switched on, metering would give priority for correctly exposing the area it detected as representing the face. Afterall, if focus was asked to be prioritised to a face, then it would be logical Captain for the user to also want correct exposure for the face to a priority. In shot 1, with only one eye visible, this would have been unlikely to have been identified as a face so the metering most likely defaulted to prioritising general protection of highlights and gave a shorter exposure.

That is what I would have thought, but that is not what I am seeing. It makes no difference unless shot in live-view. In live view with face recognition enabled, exposure is tailored to the face. I am not seeing that when live view is turned off.

Steve

**southlander** · 09-02-2014, 09:30 PM

Steve, on reflection, I must admit that I don't know how a dslr would be able to perform face detection with the mirror down and phase detect focus in use. Phase detect has nowhere near enough measurement points to work out whether there was a face in the frame. The 77000 point exposure module might still be struggling with enough sample points. Liveview on the other hand has the full readout of the sensor to work with. I've not thought too much about face detection algorithms - does it seek patterns that suggest eyes and/or significant contiguous areas of certain colors that might suggest skin, I don't know. Curious now and off to do some research on this question.

But back to the OP's question, if face detection is not involved, I'd be inclined to think about technique, specifically whether the OP was focusing and then recomposing while the shutter button was half pressed. This could lead to unexpected exposure results from shot to shot if the camera is set up to lock exposure on the half press.

**StigVidar** · 09-03-2014, 01:19 AM - **1 Like**

Originally posted by southlander

Phase detect has nowhere near enough measurement points to work out whether there was a face in the frame.

This is the image reduced to what the lightmeter in K-3 see. It is 86000 pixels.

**southlander** · 09-03-2014, 05:32 AM

Reading the K-3 manual, it looks as though Face Detect autofocus is only available in Live View (as Adam indicated in his response). Thus it would seem we can discount face detect as having any influence on the OP's experience with the two posted images. I'm back to pondering the AE-L or Link AE with AF Point settings on Custom settings area of the camera's menu. Or whether the differences in lighting between shot 1 and 2 was simply enough to trigger the camera to go down a different logic path (eg in shot 2 the proportion of very dark to very light was enough for the camera to assume a backlit scene where as in shot 2, it didn't see enough of this to trigger the back lighting logic. Overall, I'd say the lighting found in the OP's situation was not easy for a camera to determine what was required and would be the sort of scene where exposure bracketing could be helpful..

**savoche** · 09-03-2014, 05:46 AM

Originally posted by StigVidar

This is the image reduced to what the lightmeter in K-3 see. It is 86000 pixels.

Yes, I was wondering, too, if there could be a metering face detection algorithm - not at all related to the live view focus face detection.

Pure speculation, of course.

09-02-2014, 01:26 PM	#1
asp1880 Veteran Member Join Date: Jul 2010 Location: Lyngby, Copenhagen Photos: Albums Posts: 742	The K-3 multisegment meter is too smart for me Since getting a K-3 I've been giving the multisegment meter another chance. I didn't like the K-5 multisegment meter, I found it too unpredictable and just used centerweighted and exposure compensation instead. The K-3 meter is better than the K-5 one, no doubt about that. It chooses brighter exposures that push further to the right of the histogram and it's clever about dark backgrounds and uneven lighting. But then this happened the other day that left me scratching my head. The following two crappy snapshots are taken within three seconds of each other, on shutter priority and auto ISO. I used the viewfinder, not live view. Metering was multisegment, exposure compensation was 0 for both shots. The framing, light and everything are essentially identical, yet the latter shot is 2 stops brighter than the first! 1/125, f/2.8, ISO 160 1/125, f/2.8, ISO 640 What is going on here? Did the multisegment metering see the face in shot 2 and decided to expose for it? Does the K-3 do that, like the Nikon D800? If this behaviour is commonplace (I've seen it twice now), I think I'm going back to centerweighted. Regards, --Anders. Last edited by asp1880; 09-02-2014 at 01:28 PM. Reason: clarified viewfinder use.

09-02-2014, 01:38 PM	#2
Adam Administrator Site Webmaster Join Date: Sep 2006 Location: Arizona Photos: Gallery \| Albums Posts: 51,595	The K-3 doesn't have face detection through the viewfinder, so I think it's just the scene as a whole that affected the metering result. After all, the camera has no way of knowing what you want to ensure is properly-lit, so it basically has to guess based on some simple heuristic. Adam PentaxForums.com Webmaster (Site Usage Guide \| Site Help \| My Photography) PentaxForums.com server and development costs are user-supported. You can help cover these costs by donating or purchasing one of our Pentax eBooks. Or, buy your photo gear from our affiliates, Adorama, B&H Photo, KEH, or Topaz Labs, and get FREE Marketplace access - click here to see how! Trusted Pentax retailers:

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
For Sale - Sold: Hahnel Smart Charger for K-10D or K-20D	Sigmoid	Sold Items	2	04-04-2014 06:52 PM
Is the k511s too much camera for the 16-45 abd the 55-300?	Flyfisher22	Pentax K-5 & K-5 II	4	02-10-2013 11:59 PM
How Good is the Meter in the ME?	zx-m	Film SLRs and Compact Film Cameras	4	05-27-2010 11:28 AM
Is my equipment too good for me	RHW698	Troubleshooting and Beginner Help	27	05-17-2010 02:23 PM
Are we too odd as well as too few or is it just me?	jptreen	Pentax DSLR Discussion	72	10-09-2009 02:59 AM

09-02-2014, 02:40 PM	#3
rawr Veteran Member Join Date: Jun 2009 Posts: 11,913	Originally posted by Adam The K-3 doesn't have face detection through the viewfinder While it may not have proper face detection via PDAF, there is a possibility that the metering system could be programmed to detect skin tones. In these pictures, I think the issue is that the subject is mobile and dynamic, as is the lighting. The histograms for both photos would be interesting.

09-02-2014, 02:53 PM	#4
stevebrot Otis Memorial Pentaxian Join Date: Mar 2007 Location: Vancouver (USA) Photos: Gallery \| Albums Posts: 42,007	Originally posted by Adam The K-3 doesn't have face detection through the viewfinder, so I think it's just the scene as a whole that affected the metering result. I agree, though I too believe that there is logic in place based on color mix and also on small-scale contrast mix. I have found it to be very sensitive to small changes in framing with some subjects, though the result is usually nicely acceptable. Steve

09-02-2014, 03:02 PM	#5
aleonx3 Veteran Member Join Date: Jul 2007 Location: Brampton, Ontario Photos: Gallery \| Albums Posts: 3,996	If you have this setting "Link AE to AF point" set to ON for multi-segment metering, then the exposure may change due to the focus point changed (in this case, perhaps, the face).

09-02-2014, 03:08 PM	#6
stevebrot Otis Memorial Pentaxian Join Date: Mar 2007 Location: Vancouver (USA) Photos: Gallery \| Albums Posts: 42,007	Originally posted by asp1880 What is going on here? Did the multisegment metering see the face in shot 2 and decided to expose for it? According to the exif the image is an in-camera JPEG. Face Detect was turned ON with one face detected at a location appropriate to the subject. That might explain it. Steve

09-02-2014, 03:28 PM	#7
Unregistered User Guest	Originally posted by stevebrot According to the exif the image is an in-camera JPEG. Face Detect was turned ON with one face detected at a location appropriate to the subject. That might explain it. Steve May I ask what program you are using for reading the exif?

09-02-2014, 05:09 PM	#8
stevebrot Otis Memorial Pentaxian Join Date: Mar 2007 Location: Vancouver (USA) Photos: Gallery \| Albums Posts: 42,007	Originally posted by Tjompen1968 May I ask what program you are using for reading the exif? Phil Harvey's ExifTool. Steve ---------- Post added 09-02-14 at 05:39 PM ---------- Hmmm... I have been able to replicate "face detect on with live view off", but don't see that it makes any difference in exposure. Steve

09-02-2014, 06:30 PM	#9
southlander Site Supporter Join Date: May 2012 Location: Adelaide Photos: Gallery \| Albums Posts: 1,364	Originally posted by stevebrot Hmmm... I have been able to replicate "face detect on with live view off", but don't see that it makes any difference in exposure. Steve It would be sensible programming that, if face detect was switched on, metering would give priority for correctly exposing the area it detected as representing the face. Afterall, if focus was asked to be prioritised to a face, then it would be logical Captain for the user to also want correct exposure for the face to a priority. In shot 1, with only one eye visible, this would have been unlikely to have been identified as a face so the metering most likely defaulted to prioritising general protection of highlights and gave a shorter exposure.

09-02-2014, 06:59 PM	#10
gmans Veteran Member Join Date: Dec 2012 Location: Hunter Valley,NSW, Australia Photos: Gallery \| Albums Posts: 2,466	T Originally posted by asp1880 The following two crappy snapshots are taken within three seconds of each other, The first shot looks good to me, the exposure is very good, the soft window light on the boys face is a great effect. The second shot is really blown and I have had similar experiences with my K3 overexposing. I thought it was something to do with fast lens and centre weight compensation but only guessing.

09-02-2014, 07:35 PM	#11
stevebrot Otis Memorial Pentaxian Join Date: Mar 2007 Location: Vancouver (USA) Photos: Gallery \| Albums Posts: 42,007	Originally posted by southlander It would be sensible programming that, if face detect was switched on, metering would give priority for correctly exposing the area it detected as representing the face. Afterall, if focus was asked to be prioritised to a face, then it would be logical Captain for the user to also want correct exposure for the face to a priority. In shot 1, with only one eye visible, this would have been unlikely to have been identified as a face so the metering most likely defaulted to prioritising general protection of highlights and gave a shorter exposure. That is what I would have thought, but that is not what I am seeing. It makes no difference unless shot in live-view. In live view with face recognition enabled, exposure is tailored to the face. I am not seeing that when live view is turned off. Steve

09-02-2014, 09:30 PM	#12
southlander Site Supporter Join Date: May 2012 Location: Adelaide Photos: Gallery \| Albums Posts: 1,364	Steve, on reflection, I must admit that I don't know how a dslr would be able to perform face detection with the mirror down and phase detect focus in use. Phase detect has nowhere near enough measurement points to work out whether there was a face in the frame. The 77000 point exposure module might still be struggling with enough sample points. Liveview on the other hand has the full readout of the sensor to work with. I've not thought too much about face detection algorithms - does it seek patterns that suggest eyes and/or significant contiguous areas of certain colors that might suggest skin, I don't know. Curious now and off to do some research on this question. But back to the OP's question, if face detection is not involved, I'd be inclined to think about technique, specifically whether the OP was focusing and then recomposing while the shutter button was half pressed. This could lead to unexpected exposure results from shot to shot if the camera is set up to lock exposure on the half press.

09-03-2014, 01:19 AM - 1 Like	#13
StigVidar Site Supporter Join Date: Nov 2006 Location: Norway Photos: Gallery \| Albums Posts: 705	Originally posted by southlander Phase detect has nowhere near enough measurement points to work out whether there was a face in the frame. This is the image reduced to what the lightmeter in K-3 see. It is 86000 pixels.

09-03-2014, 05:32 AM	#14
southlander Site Supporter Join Date: May 2012 Location: Adelaide Photos: Gallery \| Albums Posts: 1,364	Reading the K-3 manual, it looks as though Face Detect autofocus is only available in Live View (as Adam indicated in his response). Thus it would seem we can discount face detect as having any influence on the OP's experience with the two posted images. I'm back to pondering the AE-L or Link AE with AF Point settings on Custom settings area of the camera's menu. Or whether the differences in lighting between shot 1 and 2 was simply enough to trigger the camera to go down a different logic path (eg in shot 2 the proportion of very dark to very light was enough for the camera to assume a backlit scene where as in shot 2, it didn't see enough of this to trigger the back lighting logic. Overall, I'd say the lighting found in the OP's situation was not easy for a camera to determine what was required and would be the sort of scene where exposure bracketing could be helpful..

09-03-2014, 05:46 AM	#15
savoche Master of the obvious Join Date: Jun 2011 Location: Lowlands of Norway Posts: 18,311	Originally posted by StigVidar This is the image reduced to what the lightmeter in K-3 see. It is 86000 pixels. Yes, I was wondering, too, if there could be a metering face detection algorithm - not at all related to the live view focus face detection. Pure speculation, of course.