[Gimbal]

My K5 has the problem so bad (obviously a lot worse than some of you or you wouldn't be trying to play it down) that it has to be fixed or exchanged for a better one, because there is no way that I can resell it later down the road and claim that it is a fully working camera.

Unfortunately Pentax has a roomier conscience then I

[FH_le]

Thanks for the elaborate work... let's see what future udpates will bring us

[dlacouture]

Okay, I've found another interesting fact about the K5's AF: the front-focus problem is indeed target-dependent...

I've made some focusing tests (light is the same power and distance in both cases):
- black rod in front of a white wall : camera metered the scene at 2.8EV, focus is good.
- white rod alone with dark empty space around : metered at 1.4EV, severe front-focus.

This is quite strange, as in both cases there is the same ample contrast between the target and its surroundings, so the AF should theoretically behave identically... But in the second case, it seems like the camera metering system provoked the FF... This sounds to me as something fishy in the AF algorithm...

I'm pretty sure this could be proved again by using two "negative" targets : one white with a black line in the middle, and the other black with a white line...

EDIT: okay, another test done...
Using a laptop LCD (daylight lighting!), I drew a slanted colored line against a black background... I lower the LCD brightness down so I'm below EV3-4...
- A green line gets a good focus (I can see individual pixels).
- Blue is slightly OOF.
- Red is totally OOF.

[jolepp]

Different estimate on the light wavelength and hence different compensation?

[dlacouture]

Different estimate on the light wavelength and hence different compensation?

Or blindness to red?

Well, thinking about it, it's not the case, as the camera did lock unto the red line, just like it did for the green and the blue...

Got another chance to do some more tests, and in the same settings as my black/white rods, I tried to focus on a face... Complete OOF...

[sjwaldron]

Pentax should just hire this guy and let him "attack" their cameras before they are released to work out any issues like this. I'm sure it would save the company money in the long run.

[jbinpg]

@dlacouture: A lot of variables, for sure. I wonder if the "face recognition" AF mode would do any better at low light on a person's face?

---

Last edited by jbinpg; 02-28-2011 at 05:55 PM.

[Christine Tham]

Okay, I've found another interesting fact about the K5's AF: the front-focus problem is indeed target-dependent...

I've made some focusing tests (light is the same power and distance in both cases):
- black rod in front of a white wall : camera metered the scene at 2.8EV, focus is good.
- white rod alone with dark empty space around : metered at 1.4EV, severe front-focus.

This is quite strange, as in both cases there is the same ample contrast between the target and its surroundings, so the AF should theoretically behave identically... But in the second case, it seems like the camera metering system provoked the FF... This sounds to me as something fishy in the AF algorithm...

I'm pretty sure this could be proved again by using two "negative" targets : one white with a black line in the middle, and the other black with a white line...

EDIT: okay, another test done...
Using a laptop LCD (daylight lighting!), I drew a slanted colored line against a black background... I lower the LCD brightness down so I'm below EV3-4...
- A green line gets a good focus (I can see individual pixels).
- Blue is slightly OOF.
- Red is totally OOF.

Thanks for your tests.

Interesting results indeed, and confirm my earlier suspicions about the focus being affected by the brightness of the target as well as the brightness of additional objects in the background (for reference: I discovered the camera actually has a slight FF when there is a really bright object in the background - even in bright daylight conditions, and also the camera FF trying to focus on an incandescent bulb).

It seems the camera's AF may well depend on metering.

I discovered I have metering issues yesterday (the camera tends to over-expose) due to an installation of a 3rd party focusing screen - this may explain the differences in behaviour of my AF assist light for example.

Falk - do you have any access to people in Pentax and can you comment on dlacouture's discovery?

[falconeye]

Okay, I've found another interesting fact about the K5's AF: the front-focus problem is indeed target-dependent...

I've made some focusing tests (light is the same power and distance in both cases):
- black rod in front of a white wall : camera metered the scene at 2.8EV, focus is good.
- white rod alone with dark empty space around : metered at 1.4EV, severe front-focus.

This is quite strange, as in both cases there is the same ample contrast between the target and its surroundings, so the AF should theoretically behave identically... But in the second case, it seems like the camera metering system provoked the FF... This sounds to me as something fishy in the AF algorithm...

I'm pretty sure this could be proved again by using two "negative" targets : one white with a black line in the middle, and the other black with a white line...

EDIT: okay, another test done...
Using a laptop LCD (daylight lighting!), I drew a slanted colored line against a black background... I lower the LCD brightness down so I'm below EV3-4...
- A green line gets a good focus (I can see individual pixels).
- Blue is slightly OOF.
- Red is totally OOF.

Christine, dlacouture made a good test and it is one on a list of additional tests I figured could help to further investigate.

However, I won't ask Pentax about it before 1.03 is out. A Hamburg technician returning from Tokyo told a German forum member that Japan is rather confident they found a way to fix it. Let's see.

dlacouture's test is compatible with my max. aperture hypothesis: The signal to determine focus phase does not depend on max. aperture nor on the difference if black on white or white on black is used. Still, the focus shift threshold does.

So, this is a second hint that the total amount of intermediate wavelength light entering the lens does matter.

Because the phase sensor cannot use it, i've the idea that another sensor (+sensor or whatever) is at play. And because this sensor may actually be outside the AF module ( unlikely but possible), it may be tested differently too.

So, besides dlacouture's test ( a light source pointing towards the camera beneath the feature in focus was another variant ), I imagined that pointing an LED flashlight from behind into the eyepiece when focussing may be another interesting test. Just in case a +sensor is combined into the 77 zone metering. I realized that the +feature is always combined with the 77zone meter feature: There are 3 cameras with the AF+ feature (k5 k7 645d) and they all use 77 zone metering. And no camera with 77 zone metering but without the AF+ feature exists...

So, there is a chance that the isdue is related to a hypothetical 77zone AF+ sensor which fails with too little light, as emphasized by a black target or a slow lens. Why I imagined the LED flashlight into eyepiece test which seems to be an odd idea otherwise...

[Christine Tham]

A Hamburg technician returning from Tokyo told a German forum member that Japan is rather confident they found a way to fix it. Let's see.

That is good news, if it is true.

Let's hope they truly understand the cause - I would hate for them to release a "fix" which hides the problem under certain conditions only for it to reappear under different conditions.

Your theory about the extra sensor is an interesting one - are you suggesting this sensor is located in the upper part of the camera near the viewfinder?

[Ray Pulley]

Christine, dlacouture made a good test and it is one on a list of additional tests I figured could help to further investigate.

However, I won't ask Pentax about it before 1.03 is out. A Hamburg technician returning from Tokyo told a German forum member that Japan is rather confident they found a way to fix it. Let's see.

dlacouture's test is compatible with my max. aperture hypothesis: The signal to determine focus phase does not depend on max. aperture nor on the difference if black on white or white on black is used. Still, the focus shift threshold does.

So, this is a second hint that the total amount of intermediate wavelength light entering the lens does matter.

Because the phase sensor cannot use it, i've the idea that another sensor (+sensor or whatever) is at play. And because this sensor may actually be outside the AF module ( unlikely but possible), it may be tested differently too.

So, besides dlacouture's test ( a light source pointing towards the camera beneath the feature in focus was another variant ), I imagined that pointing an LED flashlight from behind into the eyepiece when focussing may be another interesting test. Just in case a +sensor is combined into the 77 zone metering. I realized that the +feature is always combined with the 77zone meter feature: There are 3 cameras with the AF+ feature (k5 k7 645d) and they all use 77 zone metering. And no camera with 77 zone metering but without the AF+ feature exists...

So, there is a chance that the isdue is related to a hypothetical 77zone AF+ sensor which fails with too little light, as emphasized by a black target or a slow lens. Why I imagined the LED flashlight into eyepiece test which seems to be an odd idea otherwise...

Hey Falk,

I think the + color sensor is located in the prism area and is not part of the normal metering sensor and is also not part of the AF sensor optical path.

I had seen some diagrams or cutaways of this, but I cannot find them.

There is a Hoya patent that describes a similar system and diagrams it as being in the prism area. I think you can probably locate it in your K5 cutaway picture, especially if you look at the patent diagrams, as the cutaway has very similar looking optics in almost exactly the location this patent calls for. The patent describes the problem of AF color temperature (more accurately wavelength) dependent AF shift as being due to an effective focal length change in the optics of the AF module due to different wavelength content.

The AF optics are also described as having an IR cut filter in the AF optical path, which makes sense and explains why an IR cut filter has never solved this problem even though it is a popular theory (and not a bad one, actually).

I think that the CA theory can be laid to rest also by the problem that is described here and which is allegedly solved by this sensor. I am sure that the optical folks at Pentax and Hoya know why this shift happens and are not wrong about the cause, and as I have said before, most certainly did not include extra hardware if it does not address the problem.

Regardless, there is still something wrong somewhere in the K5 since it clearly does not work properly and by most accounts, the K7 does.

Here's a link to the Patent:

Digital camera

The IX+ AF system might not be exactly as described here, but the color sensing element seems to fit this description pretty closely.

Ray

[dlacouture]

The AF module given in this patent is really interesting!!!

Its direction is reversed compared to what is in the K5! In the K5, the secondary mirror sends the light back toward the bottom-front of the camera, where a mirror will send it toward the back. The light path makes a 'Z', if you want...
In this patent, the secondary mirror sends the light straight down, where another mirror will send it toward the front of the camera... More like a ']'...

EDIT: well, this patent seems to be the next SAFOX '+' implementation, where the object color will also be taken into account...

[the swede]

That patent is for the next (full frame) camera. Bigger prism and housing being able to fit the different AF sensors in a better place, thus better light paths. New AF brackets leds to.

Only joking!!!

Im sorry Just had to!

[merengue]

Thanks for your tests.

I discovered I have metering issues yesterday (the camera tends to over-expose) due to an installation of a 3rd party focusing screen - this may explain the differences in behaviour of my AF assist light for example.

Falk - do you have any access to people in Pentax and can you comment on dlacouture's discovery?

Christine:

Is the focusing screen a Katz-eye? Which metering used?

Earl

[merengue]

Christine:

Is the focusing screen a Katz-eye? Which metering used?

Earl

Sorry, wrong thread. Please reply on K-5 Forum-Katzeye Split Screen & K5 Anyone?

Earl