David: SLTs have back focus "by design" ?

[MarcoC]

Over on DPR in the Sony SLR forum there is a debate raging about whether Sony SLTs have back focus "designed in". I wondered what your thoughts were? The relevant thread:-

http://forums.dpreview.com/forums/readf ... d=41695465

[David Kilpatrick]

No. They measure focus at f/5.6 (there are no f/2.8 sensors in the SLT AF modules as far as I know, just cross-type f/5.6). That means that with almost any lens used at f/2.8, f/2 or wider may appear to have back focus because of a combination of foremost depth of field limit and possible aperture related focus shift.

When any Sony Alpha is turned off, the lens parks at infinity. I'm fairly sure I have come across the same happening with sleep but may be wrong there and it takes too long to check.

This means the lens is nearly always focusing from infinity towards the near distance. If the lens is at f/5.6, depth of field at a portrait distance with say the 16-80mm at 80mm will extend around 4 inches in front of the point of perfect focus, at 8ft. This means the AF module will identify sufficient coincidence of the reading four inches before the lens reaches it ideal focus point, if the focus is initiated from a point behind the subject. In practice it will do better than this, but unlike CD-AF it does not 'loop' round the best reading and refine the position. Instead, it predicts the coincidence of waveforms and decelerates the lens with adjustment according to firmware.

At f/5.6 the picture will be just OK and at f8 to f22 it will look fine in a normal sized print (A4) or on a laptop. On a 27 inch screen, a 20 x 16 print, or 100% pixel-peep view you'll be able to see the focus discrepancy at any aperture.

At f/5 or wider it will be clear that the real point of focus is for example a person's temple or mouth, not their eyelashes even when you thought you targeted the eye.

If you focus on something closer than your subject, then make a single AF-S focus action to refocus on the subject, the reverse will happen and there's a chance than the image will be front focused - but since there is usually nothing in front of the subject. you just see it as slightly poor focus and less depth of field than you expect, as it is now mostly wasted on empty air.

For anyone who doubts this, my advice is always the same. Set the camera to manual focus, and watch the green AF confirm light of the A77. Now carefully adjust the focus ring both ways, checking when the light goes out, and when it comes on, in either direction. Observe that this is not a sudden brief point, but a wide range; for an 18mm focal length, aiming at my computer screen from about 55cm, the light is confirming focus at all distances from 50cm to 60cm. It's telling me I have 10cm depth of field (very generous!) and also reminding me that AF could confirm and lock at ANY point within this range.

David

[artington]

Interesting comments, David. By implication, you appear to be saying that absolute accurate focus may only be achieved through manual focusing on an SLT camera. Is this what you mean or do I misunderstand you? And would your comments apply to conventional (D)SLRs too?

[Chris Malcolm]

I've tested AF very carefully on both my A350 and A550, and found exactly the difference in focus point that David decribes depending on whether you approach focus from behind or in front.

If AF was good enough for the most critical work they wouldn't have to keep on bringing out improvements. Nor would the latest top cameras with the latest most precise AF also be offering improved ways of doing manual focus faster and more accurately.

[Dr. Harout]

David, is there a way to fix that bug? Any particular setting. I've noticed that happening with the Zeiss, checked frequently for B & F focusing... and yet a lot of times (maybe 30-40%) not nailed.
I have noticed the green light thing, also same goes for focus peeking (eventually I turned it off and using the magnifier instead).

[MarcoC]

David - are you suggesting that the DSLRs with f2.8 cross type AF sensors (does the a700 have a couple of those?) are therefore going to be more accurate at focusing than any of the SLTs ?

No. They measure focus at f/5.6 (there are no f/2.8 sensors in the SLT AF modules as far as I know, just cross-type f/5.6). That means that with almost any lens used at f/2.8, f/2 or wider may appear to have back focus because of a combination of foremost depth of field limit and possible aperture related focus shift.

When any Sony Alpha is turned off, the lens parks at infinity. I'm fairly sure I have come across the same happening with sleep but may be wrong there and it takes too long to check.

This means the lens is nearly always focusing from infinity towards the near distance. If the lens is at f/5.6, depth of field at a portrait distance with say the 16-80mm at 80mm will extend around 4 inches in front of the point of perfect focus, at 8ft. This means the AF module will identify sufficient coincidence of the reading four inches before the lens reaches it ideal focus point, if the focus is initiated from a point behind the subject. In practice it will do better than this, but unlike CD-AF it does not 'loop' round the best reading and refine the position. Instead, it predicts the coincidence of waveforms and decelerates the lens with adjustment according to firmware.

At f/5.6 the picture will be just OK and at f8 to f22 it will look fine in a normal sized print (A4) or on a laptop. On a 27 inch screen, a 20 x 16 print, or 100% pixel-peep view you'll be able to see the focus discrepancy at any aperture.

At f/5 or wider it will be clear that the real point of focus is for example a person's temple or mouth, not their eyelashes even when you thought you targeted the eye.

If you focus on something closer than your subject, then make a single AF-S focus action to refocus on the subject, the reverse will happen and there's a chance than the image will be front focused - but since there is usually nothing in front of the subject. you just see it as slightly poor focus and less depth of field than you expect, as it is now mostly wasted on empty air.

For anyone who doubts this, my advice is always the same. Set the camera to manual focus, and watch the green AF confirm light of the A77. Now carefully adjust the focus ring both ways, checking when the light goes out, and when it comes on, in either direction. Observe that this is not a sudden brief point, but a wide range; for an 18mm focal length, aiming at my computer screen from about 55cm, the light is confirming focus at all distances from 50cm to 60cm. It's telling me I have 10cm depth of field (very generous!) and also reminding me that AF could confirm and lock at ANY point within this range.

David

[OneGuyKs]

David - are you suggesting that the DSLRs with f2.8 cross type AF sensors (does the a700 have a couple of those?) are therefore going to be more accurate at focusing than any of the SLTs

He said nothing about SLT vs SLR. His comments apply both to DSLRs and SLTs. I guess that' the nature of PDAF system ...

I think A700 did have one F2.8 type in center, but check this post on DPR

http://forums.dpreview.com/forums/read. ... e=36416221

"Nikon doesn't use f/2.8 AF sensors on any of their cameras. At any price level. Never has. They only use f/5.6 sensors, and actually emphasize this fact as a benefit in their online literature. And yet, nobody beats Nikon's low light AF-ing ability."

[MarcoC]

I know, but I thought the cross-type AF sensors on the DSLRs like the a700 were f2.8 sensitivity - does that not suggest they might be more accurate than f5.6 sensors?

David - are you suggesting that the DSLRs with f2.8 cross type AF sensors (does the a700 have a couple of those?) are therefore going to be more accurate at focusing than any of the SLTs

He said nothing about SLT vs SLR. His comments apply both to DSLRs and SLTs. I guess that' the nature of PDAF system ...

[OneGuyKs]

I know, but I thought the cross-type AF sensors on the DSLRs like the a700 were f2.8 sensitivity - does that not suggest they might be more accurate than f5.6 sensors?

Yes, A700 and some Canon cameras did, but Nikon and Pentax never had F2.8 sensitivity ... See this post

http://forums.dpreview.com/forums/read. ... e=36416221

[artington]

David, is there a way to fix that bug? Any particular setting. I've noticed that happening with the Zeiss, checked frequently for B & F focusing... and yet a lot of times (maybe 30-40%) not nailed.
I have noticed the green light thing, also same goes for focus peeking (eventually I turned it off and using the magnifier instead).

I presume you're referring to the 24/2. Sounds like focus-shift to me. This is common with very fast (f2.0 and wider) lenses, most noticeably at close range. My understanding, before reading DK's comments above, had been that auto-focusing occurs with the lens wide open, but if the focusing sensor is 2.8 or 5.6 maybe this is not the case. When focusing manually, focus on many fast lenses will shift when the aperture is closed down and this is presumably also the case for AF. When focusing manually a working (but not perfect) solution is to use the DOF button and focus at f4 if you are going to use apertures smaller than this (which become progressively more difficult to focus because of lower light transmission and also broader DOF making point focus more difficult) and to focus at the working aperture for apertures wider than f4. Clearly this is not a workable solution for AF. The solution for the A-77 may be to use the above method with LiveView but since I am not personally familiar with this camera I cannot be certain of it. As far as I can tell there is no such problem when focusing manually with the NEX-5 or -7 because the LCD / EVF shows the image at the working aperture although, of course, deeper depth of field will still make it harder to point-focus accurately as apertures narrow, which is why I still focus at f4 before stopping down.

[bfitzgerald]

I'm surprised most cameras don't have the double cross f2.8 AF sensor in the Film Dynax 7. What a great AF Sensor I found it very accurate even with fast lenses.
Dynax 7 film was 12 years ago..so what gives it's 2012? I'd be a lot more impressed with a smattering of 9/11 double cross sensors, than the 30-50 AF sensors on some models.

[David Kilpatrick]

I am indeed saying that manual focusing may be the best way achieve good focus, using the AF confirm light and carefully setting to the midpoint of its active range. But if you refocus a few times on a subject, even with AF, it may improve matters.

Ref f/2.8 sensors - Sony's new modules, A580, A55, A77 etc no longer have any f/2.8 sensors they use f/5.6 sensors like Nikon. In the past they used f/7.1 line sensors and f/2.8 centre-only cross sensors. But the new Sony AF modules are very similar to Nikon.

They are also very good in low light and beat nearly all other makes, operating down to EV minus 2.

David

[Greg Beetham]

I get the impression that there are practical limits and compromises with AF from a function and design standpoint in two conflicting directions (speed and accuracy) and still have it remain useful to the user. One desirable attribute would be to lock focus in the shortest elapsed time possible with moving subjects and the camera mounted on a moving support (the user), most likely with predictive AF engaged and shutter release on focus confirm. But in order to achieve that they might have had to design in a small permissible DOF zone margin (plus or minus) so the system didn’t check and tinker with the focus accuracy endlessly (if it was confined to zero tolerance) and thus not allow the shutter to be released within a time span to actually even get a photo at all.
Maybe they could introduce a couple of levels of focus performance too choose from.
A fast mode or the fastest AF for hand held action and moving subjects but has a small ‘forgive zone’ of acceptable accuracy. This mode has selectable active AF sensors.
And a slow AF mode for static or stationary subjects with the camera also stationary (on a tripod) for the utmost accuracy with no time limit considerations and no tolerance zone built in. They most likely would make that mode only available with centre spot focus; no other focus point could be used. There is the potential with that mode though that the system might not be satisfied with the degree of accuracy obtained on subjects with uneven surfaces or three dimensional subjects that have foreground and background depth within the focus sensor area and frequently go off on focus excursions looking for a ‘better’ focus lock position.
Maybe Zeiss got it right after all with their MF lenses for Nikon and Canon, claiming that with the smooth MF action on those lenses one can achieve greater accuracy than with AF, but I doubt that would be true when it came to telephoto BIF or action type shots, I think I’d have to see it too believe it with those.
Greg

[bfitzgerald]

Greg one point overlooked by some is the actual AF sensor size. On the A200 it was a lot bigger than the indicator in the VF that on it's own could lead to it locking onto something else other than the intended target, the A200 was good in decent light..and quite poor for AF accuracy in low light v the Km5d. Pentax bodies again massive sized AF points esp the central one. The Km5d and other Minolta film bodies I had all have fairly small AF points and they are pretty close to the actual AF indicator in the VF.

[OneGuyKs]

Greg one point overlooked by some is the actual AF sensor size. On the A200 it was a lot bigger than the indicator in the VF that on it's own could lead to it locking onto something else other than the intended target, the A200 was good in decent light..and quite poor for AF accuracy in low light v the Km5d. Pentax bodies again massive sized AF points esp the central one. The Km5d and other Minolta film bodies I had all have fairly small AF points and they are pretty close to the actual AF indicator in the VF.

I think the newer Sony's AF is smaller