[beholder3]

On the net the myth of "CDAF is always precise" is often spread by the noobs.

Here is a good example of that not really being the case with even the most expensive and most recent gear:
Fuji 110/2 AF performance on GFX 50S

Unreliability of Fuji's CDAF implementation here gives a spead of anything between 1,200 and 1,500 cy/PH resolution at F2.8.
That 20% sharpness spread is just about the difference between the results of a lower average lens and the best ones, so even your best lens might create just an image like a much poorer one thanks to AF inconsistency (on a static subject).

[BigMackCam]

The take-away from this (and it's always been a key factor regardless of focusing system) is that the target we choose is critical for AF accuracy. We have to understand what our chosen focus system requires - be that high contrast, horizontal or vertical details etc. - then choose targets that meet those requirements. Many people simply don't know what to point at when they use AF, and why it did or didn't work properly as a result.

[Digitalis]

There will always be a spread when autofocus systems are involved Vs precise manual focus is used. Af systems are built with a certain error margin, what the variables that cause these errors are will be different depending upon the AF system used. The optical comparison Vs the two camera systems gives into another fallacy - all cameras are created equal, they aren't. The thick sensor stack used on the Sony AR7II camera coupled with a third party adapter could cause loss of contrast and sharpness. Examining the Images at 200+ zoom a little bit silly.

I suspect there are some processing shenanigans on Fuji's part regarding the level of CA and the appearance of what I suspect are USM effects in their images Vs a fully manual lens. The Zeiss Otus 85mm f/1.4 is one of the best performers I have seen regarding CA control [among Zeiss lenses]. Lens corrections and USM being applied to raw files is something we have seen before.

[Na Horuk]

When people say CD AF is precise, what they mean is that certain problems like consistent front and backfocusing that we are used to from PD AF will not occur wit CD AF. And they won't. That said, of course CD AF has its own parameters, algorithms, and thus weaknesses.
It is well known that low contrast or homogeneous targets are bad for CD AF. Moving targets are usually pretty bad, as well. But CD AF doesn't need AF adjustment - it just needs better targets or better overall contrast-finding algorithms. Hm, its just different types of functioning, so it will not have the same strengths and weaknesses as PD AF, though it will still have its own pros and cons

And when I answered questions about PD AF and CD AF in the past, I usually added that the photographer has to learn how to use it.

[photoptimist]

These results may say more about spherical and chromatic aberrations in the lens and how those common optical defects interact with particular CDAF algorithms especially if focusing is done with the lens wide-open prior to stopping down for the photo.

And unless the tester uses the identical algorithm for assessing sharpness that the camera uses for CDAF, the two will get slightly different results. Thus a test like this is just as likely to say something about the unreliability of the lab 's assessment of sharpness as it is the camera's. Who's definition of focus is the correct one?

That said, CDAF is not perfect for many reasons already mentioned. And if the camera does the AF with the lens wide open, it can even have consistent front-focus/back-focus issues although they won't be as severe as the ones seen with PDAF.

[beholder3]

That said, CDAF is not perfect for many reasons already mentioned. And if the camera does the AF with the lens wide open, it can even have consistent front-focus/back-focus issues although they won't be as severe as the ones seen with PDAF.

I think on the blog there is even an entry about one of the lenses showing this "focus shift" = different focal plane position depending on working aperture used.


The old Sigma 50mm 1.4 HSM (non Art) is known for that as well.

[Na Horuk]

Hm, I know focusing should be done wide open, but do all cameras do that? I think the K-01 sometimes does it with the lens slightly stopped down, which can cause errors if you shoot at wide open aperture (since DoF becomes thinner for the photo than it was for the focusing)

Also, Im sure live view CD AF is only going to get better and better. Modern mirrorless cameras are already way ahead of the models from only some years ago. So research like in the OP link will only improve CD AF in the next generation

[Macario]

These results may say more about spherical and chromatic aberrations in the lens and how those common optical defects interact with particular CDAF algorithms especially if focusing is done with the lens wide-open prior to stopping down for the photo.

And unless the tester uses the identical algorithm for assessing sharpness that the camera uses for CDAF, the two will get slightly different results. Thus a test like this is just as likely to say something about the unreliability of the lab 's assessment of sharpness as it is the camera's. Who's definition of focus is the correct one?

That said, CDAF is not perfect for many reasons already mentioned. And if the camera does the AF with the lens wide open, it can even have consistent front-focus/back-focus issues although they won't be as severe as the ones seen with PDAF.

as far as I know, AF is always done with the lens wide open. If it wasn't, from f6.3 most AF systems would just stop working because there is not enough light.

[Winder]

That guy has way too much free time.

[stevebrot]

And if the camera does the AF with the lens wide open, it can even have consistent front-focus/back-focus issues although they won't be as severe as the ones seen with PDAF.

I think on the blog there is even an entry about one of the lenses showing this "focus shift" = different focal plane position depending on working aperture used.

Absolutely! Focus shift on stop-down is a well-known concern going back over 50 years and is usually associated with true vintage lenses. It has nothing to do with front/back focus and occurs regardless of focus method. With most lenses having the fault, the shift is very slight (a millimeter or two at portrait distance) and is masked by increased depth of field at normal magnifications and viewing distance. In the old days that was fine. With current ability to view images at high magnification (pixel peeping) any shift is readily evident. Focus shift is more common with wide maximum aperture lenses and becomes more severe as focus distance decreases.

I have not done a full survey, but I did do a quick hand-held look-see using my Super Program with several lenses. At 2 feet with stopdown to f/5.6 using DOF preview, any shift in plane-of-focus is readily apparent in the split image finder.

• MC Rokkor 58/1.4 (checked on Minolta SRT 102) -- about 5mm with most of the shift happening between f/4 and f/5.6, DOF @ f/5.6 FF = 34mm
• Auto-Rikenon 55/1.4 -- (same as above), DOF @ f/5.6 FF = 38mm
• Pentax-K 55/1.8 -- almost undetectable
• Super-Tak 55/1.8 -- almost undetectable
• Pentax-M 50/1.7 -- none
• XR Rikenon 50/2 -- none
• KMZ Helios 44M 58/2 -- 2-3mm with most of the shift from f/4 to f/5.6, DOF @ f/5.6 FF = 34mm

I did not note whether the shift was to front or back. Note that the shift should be the same for APS-C. Lenses well-corrected for spherical aberrations should have little or no focus shift.


Steve

[stevebrot]

Precision: the degree of deviation over multiple samples/attempts. Focus can be precise without being accurate. Precision may be evaluated using statistical tools.

Accuracy: whether results overall conform to a standard or known value. Focus can be accurate without being precise. That being said, determining accuracy when precision is poor is an exercise in futility.

Acceptable: whether results overall meet expectations within the limits of capture medium, lens performance, and user skill. Experience fielding questions on this site has taught me that acceptable focus performance is unattainable by some users, even with the best hardware.

Real world performance is strongly influenced by intended point of focus, the nature of the subject, and the quality/quantity of light. This is true regardless of focus system. Within the full range of available options (both manual and AF) here are a few points to consider:

• All systems are severely challenged by receding, curved, or complex (high relief) surfaces
• All focus systems are hampered by low contrast, low light, and lack of target detail. Any of the those factors may result in inability to consistently detect the out-of-focus condition resulting in poor precision, hunting, missed focus, and/or inability to attain focus.
• Best all-round performance (speed, precision, and accuracy) likely goes to manual focus wide-baseline optical rangefinder systems (e.g. Leica M-series bodies) (I said "likely" because I am unaware of any comparison studies.)
• Calibration of an imprecise system is a waste of time (scatter is greater than the range available for calibration)*
• CDAF offers potential for absolute accuracy with precision limited by implementation, subject, light quality/quantity, and hardware. The size of the focus area being evaluated as well as the algorithms used may limit the ability to do critical or selective focus and increase the potential for missed focus with some subjects. The linked article in the original post illustrates these points nicely.
• CDAF is relatively slow depending on implementation, lens match to system, and what it is being compared to
• PDAF accuracy is potentially high, but may be hampered by implementation, system optics, and hardware
• PDAF precision has the potential to be abysmal** resulting in inconsistent accuracy
• PDAF can be very fast
• PDAF precision may be limited by lens maximum aperture with no system able to detect the out-of-focus condition at better than if an f/2.8 lens is mounted
• Manual focus using magnified output from the image sensor (either Live View or EVF) is the gold standard for digital camera accuracy/precision. Accuracy is absolute with precision for a given subject and lighting limited only by sensor resolution and mechanical precision of the focus mechanism.
• Manual focus using focus confirm or catch-in-focus is no more accurate or precise than the PDAF it uses
• Manual focus using the stock focus screens on most (all?) current model dSLRs has accuracy and precision only slightly little better than PDAF on consumer-level bodies.
• Manual focus using screens with split-image focus aides and/or optimized matte field rival the precision/accuracy of optical rangefinder systems

Rant over...returning to the original post...

I just read the linked article in detail and suggest that any extrapolation beyond the author's description and discussion would be risky. The setup was intended to characterize observations from an earlier session and is a lead-in to additional work to follow. It appears he found a fairly specific issue with the Fuji spot-mode CDAF algorithm. The Fuji engineers will be thrilled, I am sure

BTW...he was not measuring CDAF precision per se.


Steve

* Most of the questions on this site regarding PDAF calibration problems (e.g. "No calibration works") are related to poor AF precision for the lens and focus setup. The user makes an adjustment based on a single test image and experiences a rash of missed focus the next time out shooting. A series of shots of the slant scale often shows scattered points-of-focus on both sides of some center, often with "0" as the center. Even an accurate AF fine calibration value will not improve ratio of missed focus shots. The fault is usually traceable to the lens and may include optical issues, mechanical/structural deficiency, and/or a failure in the control loop. Some focus calibration tools assess precision as part of the process. The LensAlign FocusTune software does just that. A series of photos is required and if precision is poor the series is rejected.

** I know that is a strong statement, but some combinations of subject, light, and lens present an image to the detector with no unambiguous points for comparison (the PDAF sensor is essentially a split-image rangefinder). The result is that the detector will either fail to assign focus or will do so with low precision between attempts.

[Ian Stuart Forsyth]

Absolutely! Focus shift on stop-down is a well-known concern going back over 50 years and is usually associated with true vintage lenses.

I have found that focus shift also can change with the distance to the target