[Edgar_in_Indy]

As you know I also own a Sigma 100-300mm f/4 and I have never had any serious AF issues with it when I used it on my K10, K7 and K5IIs.

Yeah, I don't think it had anything to do with using the lens on a K5 in general. I think it was just his camera in particular. It goes back to my point that calibration can vary from individual body to individual body. (This all supposes that he was being honest about his reason for wanting to return the lens.)

Phase detection is not an one time reading and calculation. At least in the case of Pentax K-r and Pentax K-30 that I know well, they make adjustments sometimes while focusing in single mode. (My other SLR is manual and for borrowed ones from other companies I cannot be sure).

Maybe, I'm not sure. The AF sensor could certainly take multiple readings and change its mind about where focus should be, but if the AF sensor is only consulting itself, then it can still miss focus due to calibration issues. If the actual image sensor is not involved to confirm focus, then it's probably not the kind of closed-loop system I'm envisioning. Multiple readings from an AF sensor that is out of calibration in relation to the image sensor will still give you bad focus. The closure I'm looking for comes from the image sensor.

Let's do a thought experiment to demonstrate my point. If you use micro-AF on a Pentax body to adjust your lens, then the results should be consistent. For example, let's say you have a lens on a K-30 that focuses perfectly accurately with no micro AF adjustment. And then, just for the sake of experimentation, you dial in -5 on the AF adjust. You will then end up with a lens that consistently misses focus, even when the camera body thinks focus has been achieved.

I don't know how exactly how the system in the Canon 5D III works (it doesn't seem to be something Canon talks about in detail, as the articles I've read are based on controlled and repeated experimentation of various bodies and lenses), but it seems to be more sophisticated in confirming that focus has been achieved on the actual focal plane. The pntrs.com/t/TUJGRktHSkJHRkpISUVCRkpOSkVN?url=lensrentals.com writer said that the 5D III's phase detect system is the first one he's seen that's approaches the accuracy of a mirrorless contrast-detect system.

This is what has made PDAF systems so fast....Panasonic has recently developed a CDAF system where it also can estimate final focus parameters using some algorithm that looks at how blurred the image is before focusing starts. So far, it only works with some of their newer lenses.

Yeah, Panasonic calls it Depth From Defocus (DFD), and it works by referencing a lens profile, so it can recognize the nature of the bokeh to determine which direction the lens needs to move, and how far. In a sense, it combines the advantages of PDAF and CDAF. It results in faster focusing and less hunting. It's great for stills photography, but it particularly benefits video by acquiring and maintaining moving subjects much better.

I know the Panasonic GH4 and the new GX8 have it. I have a GX7, so I'm hoping they could add it with a firmware update, but I'm not keeping my fingers crossed.

---

Last edited by Edgar_in_Indy; 10-03-2015 at 05:34 AM.

[Joshua A]

Yeah, I don't think it had anything to do with using the lens on a K5 in general. I think it was just his camera in particular. It goes back to my point that calibration can vary from individual body to individual body. (This all supposes that he was being honest about his reason for wanting to return the lens.)




Maybe, I'm not sure. The AF sensor could certainly take multiple readings and change its mind about where focus should be, but if the AF sensor is only consulting itself, then it can still miss focus due to calibration issues. If the actual image sensor is not involved to confirm focus, then it's probably not the kind of closed-loop system I'm envisioning. Multiple readings from an AF sensor that is out of calibration in relation to the image sensor will still give you bad focus. The closure I'm looking for comes from the image sensor.

Let's do a thought experiment to demonstrate my point. If you use micro-AF on a Pentax body to adjust your lens, then the results should be consistent. For example, let's say you have a lens on a K-30 that focuses perfectly accurately with no micro AF adjustment. And then, just for the sake of experimentation, you dial in -5 on the AF adjust. You will then end up with a lens that consistently misses focus, even when the camera body thinks focus has been achieved.

I don't know how exactly how the system in the Canon 5D III works (it doesn't seem to be something Canon talks about in detail, as the articles I've read are based on controlled and repeated experimentation of various bodies and lenses), but it seems to be more sophisticated in confirming that focus has been achieved on the actual focal plane. The pntrs.com/t/TUJGRktHSkJHRkpISUVCRkpOSkVN?url=lensrentals.com writer said that the 5D III's phase detect system is the first one he's seen that's approaches the accuracy of a mirrorless contrast-detect system.



y,
Yeah, Panasonic calls it Depth From Defocus (DFD), and it works by referencing a lens profile, so it can recognize the nature of the bokeh to determine which direction the lens needs to move, and how far. In a sense, it combines the advantages of PDAF and CDAF. It results in faster focusing and less hunting. It's great for stills photography, but it particularly benefits video by acquiring and maintaining moving subjects much better.

I know the Panasonic GH4 and the new GX8 have it. I have a GX7, so I'm hoping they could add it with a firmware update, but I'm not keeping my fingers crossed.

Just a quick little thought on the matter at hand. You'd think that with on sensor phase detection becoming prevalent on mirrorless cameras, that Pentax (or any other DSLR manufacturer) should go down the same route, and use the same sensors on their mirrored bodies. They could roughly estimate the focus with the off sensor phase detection, utilizing the larger (physically) points on the AF module to gather the info required to estimate the focusing distance, then set the lens on, raising the mirror and opening the shutter and utilizing the on sensor phase detection to make any final adjustments, eliminating any room for AF sensor errors (mis-adjustment). Obviously, this may be too slow for AF.C shooting, so I propose a similar solution.


In the same vein, the camera could slowly build up a database for each lens, recording the amount of adjustment required for each lens depending on all different factors that could cause a misfocused image. For example: focal length (for zooms), focus distance, aperture (focus shifting) and temperature (physical shrinking/expansion due to thermodynamic changes). All this data could then be used by the camera to automatically adjust the lens to perfectly match the camera, drastically increasing the focus accuracy in AF.S and AF.C mode, and eliminating any requirement to for manual microadjustment. Your thought on these ideas?

[Edgar_in_Indy]

In the same vein, the camera could slowly build up a database for each lens, recording the amount of adjustment required for each lens depending on all different factors that could cause a misfocused image. For example: focal length (for zooms), focus distance, aperture (focus shifting) and temperature (physical shrinking/expansion due to thermodynamic changes). All this data could then be used by the camera to automatically adjust the lens to perfectly match the camera, drastically increasing the focus accuracy in AF.S and AF.C mode, and eliminating any requirement to for manual microadjustment. Your thought on these ideas?

It sounds like you're proposing a kind of automatic self-calibration for lenses. I've heard other people propose a similar thing, and it sounds like a good idea to me. I would love to see DSLR makers integrate such a feature, but maybe there are concerns about too many variables, and end-users screwing it up.

[Joshua A]

It sounds like you're proposing a kind of automatic self-calibration for lenses. I've heard other people propose a similar thing, and it sounds like a good idea to me. I would love to see DSLR makers integrate such a feature, but maybe there are concerns about too many variables, and end-users screwing it up.

That is mainly the reason why I suggest the camera does this all behind the scenes (a menu option with on and off preferably). In my line of work, I find that even simple excel sheets are beyond the capability of most people to comprehend effectively.

---------- Post added 10-03-15 at 10:54 PM ----------

It sounds like you're proposing a kind of automatic self-calibration for lenses. I've heard other people propose a similar thing, and it sounds like a good idea to me. I would love to see DSLR makers integrate such a feature, but maybe there are concerns about too many variables, and end-users screwing it up.

Also, I believe that the weakest point in any interface is that between two user defined devices (think of routers to PC's via Ethernet). This way, the system is a fully contained closed loop system completely dependant of any interference from the user, and is not dependant on the lens equipped on the camera (aside from any design flaws or production intolerance from the lens end of things.)

[jsherm323]

So, maybe I'm misunderstanding, how would I calibrate a Sigma lens to work on my K5ii and KS1? Can sigma lens be tuned to more than one camera body (maybe with dock)? In body fine tune adjustment seems to be a benefit in this use case.

[cfraz]

Yeah, I don't think it had anything to do with using the lens on a K5 in general. I think it was just his camera in particular. It goes back to my point that calibration can vary from individual body to individual body. (This all supposes that he was being honest about his reason for wanting to return the lens.)




Maybe, I'm not sure. The AF sensor could certainly take multiple readings and change its mind about where focus should be, but if the AF sensor is only consulting itself, then it can still miss focus due to calibration issues. If the actual image sensor is not involved to confirm focus, then it's probably not the kind of closed-loop system I'm envisioning. Multiple readings from an AF sensor that is out of calibration in relation to the image sensor will still give you bad focus. The closure I'm looking for comes from the image sensor.

Let's do a thought experiment to demonstrate my point. If you use micro-AF on a Pentax body to adjust your lens, then the results should be consistent. For example, let's say you have a lens on a K-30 that focuses perfectly accurately with no micro AF adjustment. And then, just for the sake of experimentation, you dial in -5 on the AF adjust. You will then end up with a lens that consistently misses focus, even when the camera body thinks focus has been achieved.

I don't know how exactly how the system in the Canon 5D III works (it doesn't seem to be something Canon talks about in detail, as the articles I've read are based on controlled and repeated experimentation of various bodies and lenses), but it seems to be more sophisticated in confirming that focus has been achieved on the actual focal plane. The pntrs.com/t/TUJGRktHSkJHRkpISUVCRkpOSkVN?url=lensrentals.com writer said that the 5D III's phase detect system is the first one he's seen that's approaches the accuracy of a mirrorless contrast-detect system.




Yeah, Panasonic calls it Depth From Defocus (DFD), and it works by referencing a lens profile, so it can recognize the nature of the bokeh to determine which direction the lens needs to move, and how far. In a sense, it combines the advantages of PDAF and CDAF. It results in faster focusing and less hunting. It's great for stills photography, but it particularly benefits video by acquiring and maintaining moving subjects much better.

I know the Panasonic GH4 and the new GX8 have it. I have a GX7, so I'm hoping they could add it with a firmware update, but I'm not keeping my fingers crossed.

I think the subject of the LensRental article and PDAF calibration are separate issues. The article notes that both a newer camera and a newer lens (with a rotation sensor) are needed before the higher precision AF results are observed.

I think the condensed version is something like:

Open loop: PDAF calculate direction and rotation angle needed to achieve focus. Send data to motor (in lens or in camera). Motor runs according to instruction from camera. Done.

Closed loop: same as above, except add another step. Using the lens rotation sensor, determine if the actual rotation angle was the same as what the camera requested and adjust if necessary. It's probably even more sophisticated. Request was for 46.8 degrees clockwise. I've gone 42 degrees so start slowing, 45 degrees so slow more, 46.8 degrees. Done.

With either open or closed loop one can take another PDAF reading or two or three if needed to get a more accurate focus. One can hear that happening on Pentax screw drive lenses.

Roger mentioned he adjusted AF (to correct front/back focus) before his tests, not because it was required to determine focus precision but because he didn't want to answer 50 emails about why he didn't do it. Closed loop is about improving consistency. Fine AF calibration is still needed to assure accuracy (eliminating front/back focus bias).

The article doesn't go into the new PDAF-on-imaging sensor technology which Canon has now and which has the potential to eliminate the need for front/back focus adjustments.

[derekkite]

A fast accurate AF loop requires information.

The pdaf sensor detects the direction, further or closer. And a distance calculated through the light distorted by the lens. It needs two things from the lens, how many turns out the screw or motor to get there, and how much the distortion represents in distance. The lens is adjusted and the results verified.

It anything in the body alignment is off center the calculations will be off, and all lenses will be affected. If the lens is decentered, same thing and an offset can be entered to adjust things.

A fast accurate system requires accurate information in the lens, and possibly each copy having its characteristics written to its chip. A fast communication protocol or caching of the lens data, and a fast response by the motor and focus mechanism. In the body it needs to be able to get the light accurately to the pdaf sensors, and a fast reading, as well as quick updates from the metering sensor that can be used for selecting the pdaf sensor to use, calculate movement offsets, etc. Then a quick recycle to verify.

I think Pentax falls short on many levels here. The lens data and communication, a seemingly slow data communication from the metering sensor (the pttl delay as evidence) and not quite there algorithms.

Cdaf doesn't know which way to go, a 50/50 chance of adjusting the wrong way, and can't calculate how far.

I suspect we will see an new generation of mount on a coming body that supports a different lens communication and data protocol. It wouldn't look different, but with a supported lens would work in a different way.

[biz-engineer]

lens communication and data protocol

I think there is not need for any information coming from the lens. But the drive capability of the body, the lens AF motor gear (or screw drive gear), and the load of the actuator, and the speed of phase sensing can make the loop focus faster.

[Edgar_in_Indy]

I think there is not need for any information coming from the lens. .

But if the lens had a rotation sensor, as discussed above.

[biz-engineer]

But if the lens had a rotation sensor, as discussed above.

What lenses really need is a step motor.... the camera would do a calibration routine when the lens is mounted for the first time, then the AF servo would be able to control the position of the focus element in real time and the AF would be spot on every time. It looks like Canon implemented that in their STM lenses.

[pacu]

Yeah, I don't think it had anything to do with using the lens on a K5 in general. I think it was just his camera in particular. It goes back to my point that calibration can vary from individual body to individual body. (This all supposes that he was being honest about his reason for wanting to return the lens.)




Maybe, I'm not sure. The AF sensor could certainly take multiple readings and change its mind about where focus should be, but if the AF sensor is only consulting itself, then it can still miss focus due to calibration issues. If the actual image sensor is not involved to confirm focus, then it's probably not the kind of closed-loop system I'm envisioning. Multiple readings from an AF sensor that is out of calibration in relation to the image sensor will still give you bad focus. The closure I'm looking for comes from the image sensor.

Let's do a thought experiment to demonstrate my point. If you use micro-AF on a Pentax body to adjust your lens, then the results should be consistent. For example, let's say you have a lens on a K-30 that focuses perfectly accurately with no micro AF adjustment. And then, just for the sake of experimentation, you dial in -5 on the AF adjust. You will then end up with a lens that consistently misses focus, even when the camera body thinks focus has been achieved.

I don't know how exactly how the system in the Canon 5D III works (it doesn't seem to be something Canon talks about in detail, as the articles I've read are based on controlled and repeated experimentation of various bodies and lenses), but it seems to be more sophisticated in confirming that focus has been achieved on the actual focal plane. The pntrs.com/t/TUJGRktHSkJHRkpISUVCRkpOSkVN?url=lensrentals.com writer said that the 5D III's phase detect system is the first one he's seen that's approaches the accuracy of a mirrorless contrast-detect system.




Yeah, Panasonic calls it Depth From Defocus (DFD), and it works by referencing a lens profile, so it can recognize the nature of the bokeh to determine which direction the lens needs to move, and how far. In a sense, it combines the advantages of PDAF and CDAF. It results in faster focusing and less hunting. It's great for stills photography, but it particularly benefits video by acquiring and maintaining moving subjects much better.

I know the Panasonic GH4 and the new GX8 have it. I have a GX7, so I'm hoping they could add it with a firmware update, but I'm not keeping my fingers crossed.

Still I have my reasons to beleive it is a closed loop. But its true that it is not the image sensor that calls the shots but the phase sensor, the aligment of the latter to former is not always acceptable hence the micro-adjustments. There is always always the built in tolerances to autofocus mechanisms of the lenses.
The end resault is that for single focus the contrast detect method has almost perfect reability, with phase detect it depends on the combination of camera and lens and your mileage may vary.
That being said mirrorless last time I checked them 2 years ago almost none had perfect focus on moving subjects even with lenses like olympus 40-150 f2.8.
I would say that Pentax K-30 with the right lens was a little better than omd em1. But I believe that it was an achievement for olympus to built a usable mirrorless camera for moving subjects. But still a long way to achieve the level canon 7d had a long time ago.

Maybe in 4 years the only problem with mirrorless will be eyestrain and eatting batteries

[Edgar_in_Indy]

That being said mirrorless last time I checked them 2 years ago almost none had perfect focus on moving subjects even with lenses like olympus 40-150 f2.8.

That's an area where a lot of improvements have been, and continue to be, made. Part of the reason is the recent emphasis on video, and being able to track a moving subject during video recording. But the latest high-end mirrorless cameras are getting very good in this regard.

The new Samsung NX1 in particular is amazingly good at tracking moving subjects. If Samsung had included IBIS in the NX1, there is a good chance I would have bought one.

[derekkite]

I think there is not need for any information coming from the lens. But the drive capability of the body, the lens AF motor gear (or screw drive gear), and the load of the actuator, and the speed of phase sensing can make the loop focus faster.

But the lens knows what the offset in focus is. It knows it's current position. It also knows what the offset of the PDAF sensor means in distance. A fast AF system requires a smart lens as opposed to -move the lens - check focus - oops need more - move the lens - check focus, etc.

The PDAF sensor returns an offset from focus in or out. The lens knows where it is, it knows the curve of adjustment over the focal length. The calculation takes all of that and determines a focus point, and using the lens data knows how many turns or what signal to give to the lens to focus. Then it is verified, and if not precisely focused starts the loop again.

I think some of that is available in the KAF3 mount spec, but not enough. An poorly calibrated lens will overshoot or undershoot, and/or the pdaf offset bias is incorrect meaning it doesn't find focus, hunts or is slow.

[Kharan]

Just a quick little thought on the matter at hand. You'd think that with on sensor phase detection becoming prevalent on mirrorless cameras, that Pentax (or any other DSLR manufacturer) should go down the same route, and use the same sensors on their mirrored bodies. They could roughly estimate the focus with the off sensor phase detection, utilizing the larger (physically) points on the AF module to gather the info required to estimate the focusing distance, then set the lens on, raising the mirror and opening the shutter and utilizing the on sensor phase detection to make any final adjustments, eliminating any room for AF sensor errors (mis-adjustment). Obviously, this may be too slow for AF.C shooting, so I propose a similar solution.

What you describe is what happens with OSPDAF. The phase-detection photosites provide a rough estimate of subject position and distance, but the final tweaks are done by contrast detection, thus ensuring perfect focus accuracy at the AF point. In a DSLR, it's much easier to provide both phase detection systems independently, like Canon does with Dual Pixel AF, ensuring faster performance in live view. The great obstacle for this kind of thing, in DSLRs, is that the motors (body and in-lens) simply aren't geared for the small, back-and-forth movements that CDAF requires. This is what causes the classic 'racking' of focus back and forth in live view. Pentax has, however, a pretty good motor for this thing in DC, which allows small steps in the focus mechanism.

[derekkite]

Mirrorless got good when there were PDAF sensors mounted on the image sensor.

Some sigma lenses don't focus via CDAF. My sigma 500 isn't reliable, and the same thing happens on other bodies as well. I'm not sure why, but it indicates that the lens data is used in CDAF to make it barely palatable.