PDA

View Full Version : Auto-focus point accuracy


Tim Dolan (Longwatcher)
December 8th, 2006, 07:46 AM
Reference: 1DsMKII, which I have had for over 2-years now.

So lately I have been using the AF system more and especially using the off-center focus points.

Anyway I noticed yeserday especially that I was getting a lot of shots focusing on the wrong thing. I mean almost every shot of one particular pose set. At first I assumed it was my lens, because when I switched from my 28-70L to my 85/1.2L I hit the focus on the money. And later when I swapped back on a different pose, most of the shots went bad with the 28-70L. Same thing later that day.

Luckily it was quick poses for a class project (Trying to improve my lighting skills) but I had to figure it out. So I decide to do the ruler test. 2-meters to a slightly angled ruler and then I added some batteries for targets.

The results were that the center AF point was right on the money and the outside AF point was also right on the money (or battery in this case), but at least three of the AF points were off to the left from where they were depicted. I also noted that it seemed to be worse on my 28-70L for some reason, and in all cases it back focused when it missed aiming for something at the center point.

So my question is.
Is this something I just never noticed before or is it something going wrong and needing fixing?

And I was not putting my big thumb on the wrong button this time, for the ruler test I used a remote switch and tripod.

Up until recently I was getting 95+ percent in focus shots, but have noticed a significant increase in out-of-focus shots.

Given that they are fewer on the 85/1.2 is it possible the 28-70L may also need some work?

Although I believe it possible the 85/1.2 is also experiencing this; if it is then it is something related to aperture opening and the amount of back focus is thus less on the 85/1.2.

In any case I am aware of it now and can take corrective action, but its anoying me because it didn't use to happen.

I was hoping to not have to send my 1DsMkII in for repair until the replacement came out, but if this can be fixed then I may have to just so it doesn't annoy me. At least I have some time coming up in early January where I can part with it. And it wouldn't be a problem but I like shooting a shallow DoF and any off focus is bad.

thanks for any advice or assist.

Sean DeMerchant
December 8th, 2006, 03:18 PM
Hi Tim,

The issue here is that many people do not understand what accuracy and precision are and how they relate to autofocus.

Accuracy (http://www.m-w.com/cgi-bin/netdict?accuracy) (use definition 2a) is the measure of the error in your measurement from the ideal. Inaccurate autofocus leads to front focus and back focus. Accurate focus is on target like shooting an arrow and hitting the bullseye on a target. Accuracy can be attributed to a single measurement or averaged over a group of measurements.

Precision (http://www.m-w.com/cgi-bin/netdict?precision) (use definition 2a) is the measure of how consistent a measurement is from measurement to measurement. Precise autofocus will be very close from one measurement to the next. Shooting arrows at a target you may miss the bullseye with 3 shots but still get a tight grouping. Precision cannot be applied to a single measurement and only exists as a property of a group of measurements.

Or visually:

http://www.envisagement.com/opf/Accuracy_and_Precision.gif


Precision Compared To Accuracy


Here the distance from a green dot (arrow) to the red dot (bullseye) is a measurement of the accuracy. Roughly speaking, the radius of the blue circle is the measurement of the precision.

Now, putting away the bow and arrows we have autofocus to consider. While these concepts are clearer in 2-D, autofocus is only a 1-D measurement.

The accuracy of you autofocus is determined by the autofocus sensor. Canon has two major types of autofocus sensors in terms of accuracy. Standard autofocus points that measure contrast* in one direction (vertical or horizontal) and cross type sensors that measure contrast* in both vertical and horizontal directions to select the distance to focus at. The cross type sensors are more accurate and exist even the lowest of Canon's DSLRs on the central autofocus point.

Canon also has high precision and low precision autofocus points. The low precision autofocus points place the focus with a DoF (Depth of Field) of the point focussed upon. The high precision autofocus points place the focus within 1/3 of a DoF of the point focussed upon. These high precision autofocus points are now in even the lowest of Canon's DSLRs for lenses f/2.8 and faster on the central autofocus point.

In most Canon DSLRs, except for the central autofocus point, they use less accurate and and less precise autofocus sensors (I am not sure what the pattern is on the 1 series bodies so I may be inaccurate here for some bodies).

So stepping back to the problem at hand. As you are probably aware, DoF is a linear measurement And DoF is smaller with larger apertures. Hence, at 2m at 70 mm an f/2.8 lens will have a DoF of 13 cm on a 1Ds Mk II. While at 2m at 85 mm an f/1.8 lense will have a DoF of 6 cm. I am using an 85/1.8 rather than 85/1.2 because this DoF calculator (http://www.dofmaster.com/dofjs.html) does not support f/1.2.

Now using the central autofocus point with its higher precision the 24-70/2.8 should focus within about 2 cm of your subject while an 85/1.8 would focus within 1 cm. This is because autofocus is performed wide open and your maximum aperture determines the DoF the autofocus is working with.

So as we step back, we can see that the precision of your autofocus is determined by both the autofocus sensor type and the maximum aperture of your lens.

Now, we also have to consider the accuracy. Here the autofocus sensor type is a factor and the central autofocus point is often (if not always) superior to the outlying autofocus point in terms of accuracy and precision.

So your observed inaccuracy with 85/1.2 versus the 24-70/2.8 should always have the same relative relationship between these two lenses regardless of the body you use and no matter how often Canon recalibrates the body.

That said, the question still left is the placement of the autofocus points in the viewfinder versus the actual measured locale for the non-central autofocus points. And there I lack any helpful input. But hopefully this helps you think about this more accurately and you can look at the problem with a more informed sense of what is actually occuring.

enjoy,

Sean



* They do not really measure contrast but use a phase detection method if my understanding is correct. This has little or nothing to do with the subject at hand.

Ken Tanaka
December 8th, 2006, 03:35 PM
Well, even as a 1DsII owner myself, there's little I can really add to Sean's very nice treatise.

Chuck Westfall, of Canon, prepared a wonderful reference paper on the 1D series that encompassed many such matters. I thought it was posted on photoworkshop.com free for public download. But I cannot relocate it. If you also cannot find it there send me a pm with your email address and I'll send you a copy of the PDF (approx. 1.1Mb).

Antonio Correia
December 8th, 2006, 03:44 PM
Hi Tim, The issue here is ...
So as we step back, we can see that the precision of your autofocus is determined by both the autofocus sensor type and the maximum aperture of your lens.
... with the subject at hand.[/COLOR]

So, the wider the lens, the better.

Thank you. Quite instructive.

Tim Dolan (Longwatcher)
December 8th, 2006, 08:22 PM
Well I would say the AF points are all precise, but not all of them seem to be accurate.

The cross type AF points are the 7 in the center in landscape orientation.

Except for the center AF point I rarely use these (kind of useless when shooting portraits)

The Center is precise and accurate and within the box depicted (P&A&T)
BY accurate I mean that at 2 meters it is focused well within the DoF if not right on the money.
BY precise I mean as you listed above in that they tend to hit the same mark every time if aimed at the same spot for ever shot.
And within the box means the red box depicted in the viewfinder. (I use T for on Target)

Middle right 1 AF point is P&A&T
Middle right 2 AF point is Precise and accurate and not in the box (left of the box almost on the right 1 point)
Middle right 3 AF point is Precise and accurate and not in the box (also left half way to middle right 2 box)
Middle right 4 and 5 are P&A&T

Middle Left 1 is off in the same direction as right 2 and 3
The rest of middle left appear to be accurate, but I only did a couple, because I had confirmed that some of the right side points were off and outside the box.

As mentioned, I have had the camera for 2 years. And although I have used the non-center AF points before, I have not used them as much as I am lately. Also I have always gotten a very high percentage of shots in focus (except wider then f2.0 then I miss a few because the model or I move ever so slightly)

If I were to assess my 1DsMkII, it starting to act like my D60 did as far as number of shots in focus. However, with the D60 that was WITH the center AF point selected.

So we go back to my question.

I would guess that the AF point position is off on some of the AF points, but

Is this because I just never noticed it or is it possible something is going on the camera?



After typing all of the above, While glancing at the manual to fact check something. I remembered a custom function that might be causing my problem (maybe). I have CF 17 set to 1 (expanded points). Could this be contributing to the problem in conjunction with recent use of AI Servo.

I don't think I was in AI servo mode for the second set of shots, but I can't remember when I switched it back, but I remember switching from AI Servo to One Shot within this past week.

It is obviously annoying me because it did not used to happen, but 90% of my problems are usually me doing something I didn't do before.

Don Lashier
December 8th, 2006, 08:27 PM
And within the box means the red box depicted in the viewfinder.
My experiments indicate that (for the center point anyway) the area of influence extends beyond the red box in the viewfinder, perhaps nearly to the mid-point between it and adjacent boxes.

- DL

Tim Dolan (Longwatcher)
December 10th, 2006, 08:32 PM
I am expecting some beyond the box, but not completely out the box.
On the bad AF points I have to be completely out of the box to find a correct focus point. I expect the focus point to extend beyond the box, but not be out of the box completely. And if it were all in one direction I could compensate, but it is only a few.

John_Schwaller
December 12th, 2006, 03:41 AM
Well, even as a 1DsII owner myself, there's little I can really add to Sean's very nice treatise.

Chuck Westfall, of Canon, prepared a wonderful reference paper on the 1D series that encompassed many such matters. I thought it was posted on photoworkshop.com free for public download. But I cannot relocate it. If you also cannot find it there send me a pm with your email address and I'll send you a copy of the PDF (approx. 1.1Mb).

Is this the article?

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

John

Tim Dolan (Longwatcher)
December 12th, 2006, 07:46 AM
Thanks for the link,

first for the stupid part, I already had downloaded that document, but it was the only file I had downloaded about my Canon equipment that I had not put 'Canon' in front of (plus I had changed the name to "how to get the most of" so did not see it when originally looking for solution. I have corrected that error.

And second, based on reading the manual, experimentation and confirmation from the how to get the most document, I am switching my CF 17 back to 17-0.

I seem to remember I switched it originally to 17-1 because I was missing some shots and it improved my success rate with how I was taking shots at that time. I think my style has changed slightly over the past couple years and it no longer works for me as it did, so time to change.

I appreciate the help in working through the problem.

Until my next shoot I won't know if it actually helped though and currently my next one won't be until January (unless something comes up).

Nill Toulme
December 12th, 2006, 08:20 AM
Tim I think CF17-1 could very easily contribute to, if not cause entirely, the problems you're describing. Even for action sports, after moderately extensive experimentation I ended up going back to CF17-0 and staying there.

Nill
~~
www.toulme.net

Tim Dolan (Longwatcher)
December 19th, 2006, 08:38 AM
Based on a quick kid Christmas pictures shoot for a friend. Switching to 17-0 seems to be helping.

still need a full up shoot to see if that has completely fixed the problem though.

BobSmith
December 20th, 2006, 05:47 AM
Years ago, when having some focusing issues on a Kodak 760, I devised a simple test. It can be a real eye opener with a lot of cameras in regards to how accurately those focusing points in the viewfinder are picking up actual scene detail to determine focus.

Take a fairly large sheet of white paper and place a small black dot on the paper. Choose one of your focusing points in the viewfinder and see how close you have to place that focusing point on the black dot in order for your camera to focus. Ideally it should be unable to focus (can't focus on plain white paper) until the black dot is within the indicated focusing point in the viewfinder. You may well find that the camera is consistently finding the dot when it's just outside of the focusing point rather than within it. That's what I found on the outer points on my 760 (and on some other cameras). The center point was always dead on accurate. I just quit using the outer points rather than trying to remember to place the critical focus area in the image just outside of the indicated point in the appropriate direction. If you're having focusing issues and you're relying on the outer points, this is one of the first tests that I would look at.