[biz-engineer]

Not matter how many times I tried, I could never fully understand (in depth) how PDAF works.
More specifically, the thing that I can't understand is why some points are said to be sensitive to f2.8 and other points sensitive to f8.
Also, I would tend to believe that AF accuracy is improved when shooting with faster lenses , larger aperture wide open, e.g f/2.8 vs f4 or f5.6 wide open, but that may not be true if some points are only sensitive to f8.

For example, I shot a series of vintage cars (old-timer race at slow speed), with my DFA 70-210 f/4, most shots are at f5.6 and f8 due to the depth of field required for the cars, and I chose to use the DFA70-210 f4 because it is light weight, compact, easy to shoot hand-held all day. But while I managed to get a fair number of keepers, there are still some images that aren't quite focused at the right place, the camera triggered the shutter a bit too early or a bit too late. Based on how I use AF (AFC, 1 shot focus priority, AF hold disabled), I just wonder if I would have had better focused images if I shot those vintage cars with my DFA 70-200 f2.8, thanks to its f/2.8 wide open aperture.

I would tend to believe (hypothesis) that a lens with larger aperture (camera focusing with lens wide open) will create more phase shift on the PDAF sensor, so that the focus error will have to be reduced further to achieve lock focus, but I may be wrong. I've tried with the DFA70-200 2.8, but focus accuracy didn't seem to be better. So... I don't know.

---

Last edited by biz-engineer; 05-06-2023 at 06:18 AM.

[UncleVanya]

It's a mystery to me also. I find it interesting that Sony uses PDAF but as I understand it they also use contrast detection I'm a little unclear which operates when. I do know that PDAF is harder to calibrate with a DSLR since the sensors are not located in the same exact plane/path as the sensor. Apparently focus shift due to aperture is a thing that mirrorless has less issue with since it can focus stopped down with limited impact to the user. Ultimately I just want it to work and I don't get too hung up on how/why.

[Apet-Sure]

I'm going to follow this thread with interest. Usually I don't use live view in order to save the KP's battery (and my K-5IIs live view is pretty awful). However, I've been thinking I may get more focus hits if I did use it. I have calibrated my AF lenses using the 'dot-tune' method, and usually get pretty good results with CDAF, but not as often as I would like. (Of course the misses could be some kind of operator error, and the focus point on the subject has to have enough contrast for CDAF to work.) I'm old enough to have cut my photography teeth with manual focus lenses. I often have time to manually focus shots (the flowers don't run away). Anyway, I can always stand to refine and improve my technique with all 3 focusing methods. Looking forward to reading people's thoughts on PDAF.

[ramseybuckeye]

I'm going to follow this thread with interest. Usually I don't use live view in order to save the KP's battery (and my K-5IIs live view is pretty awful). However, I've been thinking I may get more focus hits if I did use it. I have calibrated my AF lenses using the 'dot-tune' method, and usually get pretty good results with CDAF, but not as often as I would like. (Of course the misses could be some kind of operator error, and the focus point on the subject has to have enough contrast for CDAF to work.) I'm old enough to have cut my photography teeth with manual focus lenses. I often have time to manually focus shots (the flowers don't run away). Anyway, I can always stand to refine and improve my technique with all 3 focusing methods. Looking forward to reading people's thoughts on PDAF.

I use live view on the KP a lot because of using the tilt screen and a tripod, the autofocus does seem pretty accurate, but usually I will manually focus with magnification and focus peaking. I'm not sure that autofocus in live view is as precise as you can get it through the viewfinder. I do feel that the manual focus with magnification and focus peaking is better than either autofocus, but of course that is limited to shooting non moving scenes.

[gatorguy]

While this doesn't answer Biz questions, it does give readers a primer on what PDAF is, and how it works, thus a good starting place.
https://photographylife.com/how-phase-detection-autofocus-works

I'll also be watching for the answers on fast lenses and whether they are more focus-accurate on our cameras. Good topic Biz.

EDIT: I stumbled on a very similar discussion linked here:
https://photo.stackexchange.com/questions/47122/does-autofocus-work-better-w...-f-4-or-slower

[Ian Stuart Forsyth]

Not matter how many times I tried, I could never fully understand (in depth) how PDAF works.

This might help
https://www.photonstophotos.net/MarianneOelund/AutoFocus_System_Design/AutoF...tem_Design.htm

There can also be some more additional benefits to having lets say a 70-200 2.8 to using a 70-200 f4, sometimes that more expensive lens can be constructed with better AF motors and including better calibration for action photography. there is also the other side of the coin if that faster lens introduces flare it might impede how accurate and quick that lens can focus.

[biz-engineer]

There can also be some more additional benefits to having lets say a 70-200 2.8 to using a 70-200 f4

The problem is, in the tests I've done yesterday, comparing the DFA70-200 2.8 and DFA 70-210 4, and DFA 28-105 @ 105, the DFA 28-105 (f5.6) deliver more repeatable results, and the DFA 70-200 2.8 give the worst max focus error (AFC, 1st shot in focus, button full press, take 10 shots).

While this doesn't answer Biz questions, it does give readers a primer on what PDAF is, and how it works, thus a good starting place.https://photographylife.com/how-phase-detection-autofocus-works

This might helphttps://www.photonstophotos.net/MarianneOelund/AutoFocus_System_Design/AutoF...tem_Design.htm

Both points to Marianne Oelund info.


Which can be summed up in one statement: only the center AF points receive the extra light rays that the f2.8 aperture let pass, while the outer AF points receive light as if the lens was stopped down to f4, f5,6, f8 etc... depending how far away from the center.
Which means, if the subject was focused on an outer AF point receiving only light as if the lens was stopped down to f/8, there is no difference in AF accuracy between using an f2.8 (wide open), f4 (wide open) or f5.6 (wide open) lens.

[Ian Stuart Forsyth]

If I understand your problems is that this is one of AF precision, my understanding is that this is done by the spacing of the AF sensors. If the AF sensors that are tuned for 2.8 coverage they are spaced wider and will not work for 5.6 and vise versa. The separator masks limits what portion of the lens is used in the collection of the light. So even those AF sensors that detect for 2.8 they only receive what the separator mask allows thru.

https://www.photonstophotos.net/MarianneOelund/AutoFocus_System_Design/TS560...Separators.jpg

many people would be shocked to know that their faster lens wider than f7.5 don't increase the amount of light reaching the sensor for lower light AF speeds

---------- Post added 05-07-2023 at 01:27 AM ----------

The problem is, in the tests I've done yesterday, comparing the DFA70-200 2.8

Sounds like the lens is not really setup for this kind of work.

I know with several of the lens that I use that are designed around AF speed and accuracy the 2.8 have an advantage over the F4
the 24-70 2.8 is very good the 24-85 is rather slow and less accurate
The price of the lens is more of the deciding factor than the F stop I know with the 70-200 f4 is no comparison to the the 200-400 f4 but then the lens has the bells and whistles of a $7000 lens

[gatorguy]

https://www.photonstophotos.net/MarianneOelund/AutoFocus_System_Design/TS560...Separators.jpg

many people would be shocked to know that their faster lens wider than f7.5 don't increase the amount of light reaching the sensor for lower light AF speeds

From reading further explanations Marianne Oelunds was referring to very specific Nikon camera bodies at the time and not all DSLR's

[biz-engineer]

Looking to understand between how PDAF is done optically is one thing, but other things can have an impact on focus accuracy.
My most interesting discovery about the AF-C mode, is that I get approximately 80% shots in acceptable focus when I use the shutter button. But when I use the IR remote to focus+take picture (same AF-C mode), I get 100% shots in focus.
There is also an AFC.A mode but I haven't noticed any difference with focus priority. If Pentax could add, via firmware update, a third AF-C mode that works the same way as with the IR remote, that would be great.

[TheSwede]

Using AFC on still subjects will get me porer results than with AFS. I use AFS and focus 2 or 3 times before taking the shot. Using K-3 III AFC has improved so that there are more keepers but the K-1 really is not good when using AFC on still subjects, it hunts even if nothing moves. Actually I rarly use AFC with the K-1, I use AFS and BBF on BIFs with better result than AFC.

I would say that the f/4 will get you more shots in focus since it has a deeper DOF. The 70-200/2.8 is a bit tricky to get a good focus hit with for some reason I do not get. Using AFS and BBF, focusing 2-3 times and shoot, then repeat, will get me more keepers...

[biz-engineer]

Using AFC on still subjects will get me porer results than with AFS.

That's a good point and you are right, AFS is the right choice for still subjects. I'm using a still subject to test my AFC mode as proxy because I don't have a moving vintage car driving in a loop for testing various settings and various approaches.

Just for better understand, I have to describe my workflow.

- I shot multiple images for later stitching as panaramas, which include a vintage car on the road and the landscape in the background
- I shot the car first, that when auto-focus is required, the goal is to have the (moving) car in good focus
- soon after I shot the car, car disappeared from the scene, I block AF using the back button AF option "Cancel AF", so that the lens focus doesn't change, then I shot a 2 to 6 additional frames to include the scenery around the car
- I put images together for stitching, tell the stitching software to include the car (using a mask)
- Final result is a "large format" panorama, between ~70Mpixels (3 images stitch) and 200Mpixels (20 images stitch)

I don't want to use fast burst or AFC tracking, because the focus is changing from shot to shot, and changing focus is a problem for stitching and final rendering of the panoramas.
Best case: the car in well focused, also allow to have the car (main subject) in a corner (not possible with a single DSLR shot due to AF point located in the center) , distant landscape is slightly out of focus, plenty of resolution printed big.
Worse case: the plane of focus isn't on the car, or the lens focus shifter between frame (e.g I didn't press well enough to cancel AF after the first picture was taken), in that case the whole panorama is ruined.

Observation with AFC, single shot, 1st frame focus priority: around 8/10 or 9/10 are reasonably well focused, but 2/10 or 1/10 can be complete out of focus (unusable). So I am trying to understand what happened.

Other than that, I've been working on this technique for the last two years, and my workflow improved a lot from two years ago. I'm now able to produce "dynamic" panoramas with some "wow" effect (looking like shot with 4x5" with detailed shiny vintage car in the foreground and scenic landscape in the middle and background.

---------- Post added 08-05-23 at 08:19 ----------

The 70-200/2.8 is a bit tricky to get a good focus hit with for some reason I do not get.

Surprisingly, I get more keepers with the DFA70-210 f4 (the cheaper and smaller one), but, they are different.
With the DFA 70-210 4, I get a lot of shots that are acceptably well focused but not perfectly focused, still usable.
With the DFA 70-200 2.8, some shots are perfectly focused (100% nailed focus), and some shots are way out of focus, much more dispersion of focus from shot to shot.

[UncleVanya]

Using AFC on still subjects will get me porer results than with AFS. I use AFS and focus 2 or 3 times before taking the shot. Using K-3 III AFC has improved so that there are more keepers but the K-1 really is not good when using AFC on still subjects, it hunts even if nothing moves. Actually I rarly use AFC with the K-1, I use AFS and BBF on BIFs with better result than AFC.

I would say that the f/4 will get you more shots in focus since it has a deeper DOF. The 70-200/2.8 is a bit tricky to get a good focus hit with for some reason I do not get. Using AFS and BBF, focusing 2-3 times and shoot, then repeat, will get me more keepers...

Even my Sony A7R III works this way. AFS is more accurate than AFC for a still subject. Newer Sony cameras are a bit more forgiving in that respect.

Some people disagree and say afc is good all the time but the data suggests that afc has more variability.

[biz-engineer]

Some people disagree and say afc is good all the time but the data suggests that afc has more variability.

It's complicated. AF.C is a tradeoff between accuracy and timing (from what I could observe) since it's supposed to trigger an exposure on a moving target. I tried AF.S, single shot, on moving targets, but sometimes the camera will refuse to take the shot because focus error doesn't get below AF.S accuracy threshold.

After opening this thread, using K1 and DFA70-200 2.8, I recorded lens AF window on the lens barrel with my phone in slow motion mode, in order to see what lens is doing.

So, from the video recordings at 120FPS, I could observe that AF.C focus priority proceeds in three steps:

1) one continuous AF run up to a point close to focus. I guess it's a first AF run calculated from the phase value given by in camera AF sensor
2) second small step to close the gap between first run and focus
3) after second small step, the shutter opens immediately, regardless if it's fully in focus or slightly out off focus.
If step two is out of focus completely, shutter won't open (= focus priority).

There are two other modes for AF.C: Release priority and Auto.

For Release priority, it's the same 1,2 steps , but there is no condition that prevent to open the shutter, lens AF continues to adjust focus indefinitely after step 2 , as long as activated.
For Auto, if the step 1 of the AF cycle can't achieve enough focus, AF switches to Release priority, and shutter opens.

When using AF.C focus priority with the IR remote, shutter open after step 2, but not immediately, shutter open only after AF confirm (bip), which give time to close the gap if step 2 isn't fully in focus.
The difference between AF.C focus prio and AF.S , is just the threshold of OOF error is relaxed for AF.C , hence not every shot is completely focused, and I'm not aware of anywhere in camera menus where this threshold could be changed, the threshold (accuracy) was , presumably, decided by Pentax engineers.

[biz-engineer]

I'll also be watching for the answers on fast lenses and whether they are more focus-accurate on our cameras. Good topic Biz.

As it turns out, PDAF sensor related error is only one contribution to the focus error. There is also how fine is the step of the AF motor and "gear box" between the motor and optical elements.
Practically, there is a run to run dispersion of focus, due to the propagation of errors from sensors, to electronics, actuation, mechanical to optical elements, and it is futile to expect better.

There were a couple of articles about that from Roger Cicala of Lensrentals :
Lens Rentals | Blog
Lens Rentals | Blog

With a 70-200 2.8 lens, sensitivity to error increases with focal length and with wider aperture:
- At 135 - 200mm focal length @ f4, 10 feet, 0.5 degree position error on AF barrel can throw things slightly out of focus
- At 70mm f5.6, 30 feet, the same 0.5 degree of position error on AF doesn't create a significant out of focus blur.