How does EOS AF work? You think you know?

[Wojtek Garwol]

Hello everyone,

yeah, how does AF work in EOS cameras? I also thought I knew. It's supposed to be very simple: as soon as the camera sees sharp image, it stops adjusting focus. And in MF mode with focus confirmation, it confirms focus as soon as it sees sharp image. As simple as that.

BUT IT IS NOT THE CASE!!!!

It must work in some different way... 300D (DRebel) in MF mode with Sigma 24-70/2.8 confirms focus when in focus, but also when it is quite a bit off, both directions. Say, if subject distance is 2.5 m (8 feet) it confirms focus when it is set from 1.8 m (6 feet) to 3 m (10 feet). For border settings (1.8 m and 3 m) it is clearly NOT in focus, by f2.8 it doesn't even deserve the description of "soft", it just totally out of focus.

HAS ANYONE AN EXPLANATION how this AF works??? I don't understand it anymore (and probably never did )

--
Regards, Wojtek

 

[Lemming51]

I believe the AF on the DRebel and similar non-pro bodies indicate "in focus" when the subject is within the range of depth of field. At 24mm, f/2.8 and 8 ft subject distance, the DoF would be from about 6.2 ft to 11.3 ft. That seems to correspond to what your getting in your test.

With higher level bodies, Canon uses "high-precision" AF that focus to within 1/3 of the depth of field. The high-precision mode requires lenses with maximum apertures of f/2.8 or larger.

Hello everyone,

yeah, how does AF work in EOS cameras? I also thought I knew. It's
supposed to be very simple: as soon as the camera sees sharp image,
it stops adjusting focus. And in MF mode with focus confirmation,
it confirms focus as soon as it sees sharp image. As simple as that.

BUT IT IS NOT THE CASE!!!!

It must work in some different way... 300D (DRebel) in MF mode with
Sigma 24-70/2.8 confirms focus when in focus, but also when it is
quite a bit off, both directions. Say, if subject distance is 2.5 m
(8 feet) it confirms focus when it is set from 1.8 m (6 feet) to 3
m (10 feet). For border settings (1.8 m and 3 m) it is clearly NOT
in focus, by f2.8 it doesn't even deserve the description of
"soft", it just totally out of focus.

HAS ANYONE AN EXPLANATION how this AF works??? I don't understand
it anymore (and probably never did )

--
Regards, Wojtek

 

[Petteri Sulonen]

It's not a simple feedback loop.

The AF sensors can tell not only whether the image is out of focus, but how much out of focus it is and in which direction. They take the reading, then the camera tells the lens how much it needs to move in order to be in focus. Then the sensors take another reading. If it's close enough, beep. If not, they do another adjustment, and that's about it. They don't keep iterating until the cows come home.

If it's totally OOF, the lens hunts -- it scans back and forth until the AF sensors see something they can understand, after which it snaps into focus.

AI Servo is a bit different: in that, the camera records readings in sequence, computes how fast the subject is moving towards or away from the camera, and estimates where it will be in a split-second, and sets focus to that. And repeats, over and over. That is a loop.

More or less.

Petteri
--
Me on photography: [ http://www.prime-junta.tk/ ]
Me on politics: [ http://p-on-p.blogspot.com/ ]

 

[Wojtek Garwol]

It's not a simple feedback loop.
The AF sensors can tell not only whether the image is out of focus,
but how much out of focus it is and in which direction. They take
the reading, then the camera tells the lens how much it needs to
move in order to be in focus. Then the sensors take another
reading. If it's close enough, beep.

But in case of 24 mm f/2.8 on EOS 300D "close" enough seems to be "really out of focus". I can accept that the picture is not perfectly in focus, but why the camera in MF mode confirms focus when the image is not in focus at all, not even assuming some reasonable tolerance?

If not, they do another
adjustment, and that's about it. They don't keep iterating until
the cows come home.

Yeah, but that reasonable approach does explain confirming the focus in MF mode.

You better see samples, those are 100% crops from a center of the frame, focus point, MF mode, 300D + 24-70/2.8 EX DF DG:

in focus:



out of focus:



In both cases, in MF mode, correct focus was confirmed by the camera.
Is the tolerance really meant to be so tolerant?
--
Regards, Wojtek

 

[MrFoopy]

Hello everyone,

yeah, how does AF work in EOS cameras?

This is not very technical, but a good general introduction:
http://electronics.howstuffworks.com/autofocus.htm/printable

 

[Wojtek Garwol]

I believe the AF on the DRebel and similar non-pro bodies indicate
"in focus" when the subject is within the range of depth of field.
At 24mm, f/2.8 and 8 ft subject distance, the DoF would be from
about 6.2 ft to 11.3 ft. That seems to correspond to what your
getting in your test.

Please, take a look here:





In both cases, camere confirmed correct focus in MF mode. It may lie within DoF theoretically, but it is clearly out of focus when you take a look at 100% magnification.

With higher level bodies, Canon uses "high-precision" AF that focus
to within 1/3 of the depth of field. The high-precision mode
requires lenses with maximum apertures of f/2.8 or larger.

Do I really need 1D to get really sharp pictures?

--
Regards, Wojtek

 

[Petteri Sulonen]

First off, I just plain don't trust the focus confirmation. I don't know what it does. Manual focusing on one of these screens is an exercise in futility; I don't even bother anymore -- I shoots in AF and takes my chances. Luckily, it works -- my hit ratio is way better than it was with MF SLR's. It just feels different because with MF, I knew who was to blame and with AF, I don't know if it's me or the camera.

So my suggestion to you would be to learn to use AF as it's meant to be used, and forget about MF.

Petteri
--
Me on photography: [ http://www.prime-junta.tk/ ]
Me on politics: [ http://p-on-p.blogspot.com/ ]

 

[Wojtek Garwol]

First off, I just plain don't trust the focus confirmation. I don't
know what it does. Manual focusing on one of these screens is an
exercise in futility; I don't even bother anymore -- I shoots in AF
and takes my chances.
So my suggestion to you would be to learn to use AF as it's meant
to be used, and forget about MF.

But with 300D and Sigma at 24 mm and f/2.8, I get 75% or more of out-of-focus results, although I think I know how to use AF - only center AF sensor activated, use contrasty motives when focusing, reframing after locking focus etc etc. But it just gives random results, with really big percentage of what you could see in out-of-focus example, which is unacceptable even without magnifying it to 100%.

The example in MF mode just illustrates what I get in AF mode when applying AF as well as I can.

Best regards,
Wojtek

 

[Wojtek Garwol]

This is not very technical, but a good general introduction:
http://electronics.howstuffworks.com/autofocus.htm/printable

Well, this describes exactly how I thought it works

But it doesn't seem to be the case, take a look here, in both cases correct focus was confirmed by the camera:





--
Regards, Wojtek

 

[Petteri Sulonen]

Then it sounds like either your lens(es) or your body need calibration. It's been known to happen.

Petteri
--
Me on photography: [ http://www.prime-junta.tk/ ]
Me on politics: [ http://p-on-p.blogspot.com/ ]

 

[Wojtek Garwol]

Then it sounds like either your lens(es) or your body need
calibration. It's been known to happen.

My 300D with 17-40L, 70-200/4L and 50/1.8 is just at Canon Service Center in Germany, we'll see the results when it comes back, I'm really curious.

But from what I have read, Sigma lenses for EOS cannot be calibrated. Is that so, maybe you know?

Best regards
Wojtek

 

[Cesare Ferrari]

I had weird problems with my D30's autofocus, and concluded that the sensor was broken. It decided to hunt around and gave autofocus lock at random points.

Anyhow, I decided to give the camera a good clean, and suddenly it worked fine again. My conclusion was that a big lump of dust had got in the light path to the AF sensors, and i'd dislodged it with a blower.

So I suggest you do some systematic tests. Give the camera a good clean (blower in the body, and have a look at the mirror under the main mirror that directs light to the AF sensor.

Get yourself in good lighting conditions with a test target (you know, a piece of white paper with a big bold cross in the centre) and see how the camera performs. Try a different lens if you have one available. Decide if the problem follows the camera or a particular lens.

Of course the other thing to do which really helps is to find a friend with a similar camera, and try theirs as well to help you decide if this is a fault with your camera.

Just remember, the autofocus problems with the Canon 10D/300D style AF was overplayed. We are talking about AF missing by a few mm here and there, not a reading of 10ft for a test target 5ft away.

Cesare

 

[Wojtek Garwol]

Before you get your stuff calibrated try this...

The camera with all my Canon lenses is already at the service

I had weird problems with my D30's autofocus, and concluded that
the sensor was broken. It decided to hunt around and gave autofocus
lock at random points.

Well, that's exactly what I have.

Anyhow, I decided to give the camera a good clean, and suddenly it
worked fine again. My conclusion was that a big lump of dust had
got in the light path to the AF sensors, and i'd dislodged it with
a blower.
So I suggest you do some systematic tests. Give the camera a good
clean (blower in the body, and have a look at the mirror under the
main mirror that directs light to the AF sensor.

Your tip could be the answer... I have some dust inside indeed, clearly visible in the viewfinder.

Get yourself in good lighting conditions with a test target (you
know, a piece of white paper with a big bold cross in the centre)
and see how the camera performs. Try a different lens if you have
one available. Decide if the problem follows the camera or a
particular lens.

Well, f4 lenses perform quite well (17-40L and 70-200L), the faster ones do not (50/f1.8 and 24-70/f2.8).

Of course the other thing to do which really helps is to find a
friend with a similar camera, and try theirs as well to help you
decide if this is a fault with your camera.

I did, my friend's 300D was actually the one the samples were taken with, but it is exactly what I got with my camera

Just remember, the autofocus problems with the Canon 10D/300D style
AF was overplayed. We are talking about AF missing by a few mm here
and there, not a reading of 10ft for a test target 5ft away.

But there's probably something tricky in the way AF works in EOS cameras, vide 20D+24-70EX+420EX+focus assist - there is just no logical explanation why AF should be wrong so...

Anyway, many thanks, Cesare, as soon as my camera is back from service, I will check your tip.

--
Regards, Wojtek

 

[Fer]

Well, this describes exactly how I thought it works
But it doesn't seem to be the case, take a look here, in both cases
correct focus was confirmed by the camera:

This is my take:

1. The lens hunts for focus following the direction of the camera

2. The camera takes a last reading. If the local contrast (as measured by the linear / cross AF sensor) is "good enough":
2.1 Confirm focus / stop lens
3. Otherwhise:
3.1 Give the lens the last command and stop without reading another time

Now, I don't know how focus confirmation works, but I don't trust it at all. In my experience, it fails as badly as AF, expecially with some Sigma lenses (most notably my 18-125).

If someone is in good relationship with Chuck Westfall, he is the man who can answer this question once and for all..

Fernando
--
Portfolios: http://www.usefilm.com/photographer/18417.html
http://www.bytephoto.com/photopost/showgallery.php?ppuser=1468

 

[PhotoTom]

One of the things to watch out for in both AF and MF-confirmed modes is that the focus detector is larger than the indicator in the viewfinder. If a high contrast point is in the area covered by the sensor, it may "grab" focus to that point, even if it is not the spot that you intended to be your primary focus point.

This is not very technical, but a good general introduction:
http://electronics.howstuffworks.com/autofocus.htm/printable

Well, this describes exactly how I thought it works
But it doesn't seem to be the case, take a look here, in both cases
correct focus was confirmed by the camera:





--
Regards, Wojtek

--
Tom

 

[Wojtek Garwol]

This is my take:
1. The lens hunts for focus following the direction of the camera
2. The camera takes a last reading. If the local contrast (as
measured by the linear / cross AF sensor) is "good enough":
2.1 Confirm focus / stop lens
3. Otherwhise:
3.1 Give the lens the last command and stop without reading another
time

Well, that's what I believe too, but have you seen my samples a few posts above? This is not what I see as "good enough" focus, nevertheless it was confirmed as such by the camera.

3.1 Give the lens the last command and stop without reading another
time

Yes, but then focus should not be confirmed!!! How can the camera confirm focus without checking???

Now, I don't know how focus confirmation works, but I don't trust
it at all.

And it seems, not without a reason.

In my experience, it fails as badly as AF, expecially
with some Sigma lenses (most notably my 18-125).

Yeah! That's curious, why of all things with a Sigma lens? But it's somehow no surprise from Canon's side...

If someone is in good relationship with Chuck Westfall, he is the
man who can answer this question once and for all..

It would be great

Best regards
Wojtek

 

[Wojtek Garwol]

One of the things to watch out for in both AF and MF-confirmed
modes is that the focus detector is larger than the indicator in
the viewfinder. If a high contrast point is in the area covered by
the sensor, it may "grab" focus to that point, even if it is not

Yes, that's what I have heard too, and that, if true, could explain a lot.
How do you know that AF sensors are larger than the indicators?

Best regards
Wojtek

 

[PhotoTom]

Here's an image of the 10D's focus points with the indicators (black squares on the image) and the actual sensor size and configuration (Highlighted in red). Image is courtesy of Chuck Westfall, with the red color courtesy of me.



I suspect that the D-Rebel/300D is similar, though I don't think they use squares to show the focus points.

One of the things to watch out for in both AF and MF-confirmed
modes is that the focus detector is larger than the indicator in
the viewfinder. If a high contrast point is in the area covered by
the sensor, it may "grab" focus to that point, even if it is not

Yes, that's what I have heard too, and that, if true, could explain
a lot.
How do you know that AF sensors are larger than the indicators?

Best regards
Wojtek

--
Tom

 

[Wojtek Garwol]

Here's an image of the 10D's focus points with the indicators
(black squares on the image) and the actual sensor size and
configuration (Highlighted in red). Image is courtesy of Chuck
Westfall, with the red color courtesy of me.
http://home.comcast.net/~trwilk3/Images/Focus_Point_Size.jpg

Thanks a lot, Tom, I'll take a closer look at this fact.

I suspect that the D-Rebel/300D is similar, though I don't think
they use squares to show the focus points.

I think that I have read somewhere that 10D and 300D share AF system, but I'm not sure.
--
Regards, Wojtek

 

[RDKirk]

In both cases, camere confirmed correct focus in MF mode. It may
lie within DoF theoretically, but it is clearly out of focus when
you take a look at 100% magnification.

Depth of field varies inversely with the final viewing magnification. Canon designs its af systems around a final viewing magnification of A5 to A6. Viewing an image on a monitor at 100 percent of the pixel resolution is like burying your nose into a A1 print.

Check out the tips for judging AF accuracy in Canon's own words:

http://www.photoworkshop.com/canon/EOS_Digital.pdf

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'