Go to advertiser website.
Go to advertiser website.
An analysis of EOS-1D Mark III autofocus performance - Continued
  May 29, 2009: The EOS-1D Mark III with revised AF calibration and v1.2.5 firmware
Earlier this year, Canon released firmware v1.2.5 for the EOS-1D Mark III and, at the same time, revealed they had found and corrected a problem in the factory AF calibration procedure for this camera, one that could result in miscalibration of the camera's outer AF points.

They also outfitted Canon service facilities worldwide with a new version of the AF calibration software used with this EOS-1D Mark III. Most and probably all Canon sales organizations around the globe then began offering free AF calibration for any owner of this model concerned that their camera might not be autofocusing correctly.

This update describes our experience with the new firmware and revised AF calibration.

Hit the Dirt: Canon EOS-1D Mark III (firmware v1.2.5, revised AF calibration) + EF 300mm f/2.8L IS, ISO 400, 1/2000, f/2.8. Click to enlarge (Photo by Rob Galbraith/Little Guy Media)

Q. Prior to Canon's latest changes, what did you think of EOS-1D Mark III autofocus?

That it has come a long way since the early days back in mid-2007, but it still isn't what it should be for a 1-series camera. Over time, Canon's efforts morphed the EOS-1D Mark III into a camera capable of acceptably stable and accurate autofocus of stationary or slow moving subjects, with the lenses we use most often. When using long lenses wide open, and the subject-to-camera distance is changing more rapidly (as it does often when shooting sports), autofocus performance has been mixed at best and poor at worst, even after the hardware and firmware changes the company introduced in the first year and a bit of its life.

This is a brief summary of the conclusions in the August 1, 2008 update to this article, which you can read it in its entirety here.

Q. What was the goal of this round of EOS-1D Mark III autofocus evaluation?

To see if Canon's latest changes improved the camera's ability to track moving subjects with a wide aperture telephoto lens. Really, we set out to do enough testing to see if we needed to do a lot more testing. To that end, the parameters set out were narrow: use the camera at real sporting events, use the sharp, fast-focusing EF 300mm f/2.8L IS at or near f/2.8 and see what results.

Q. Were you able to test an older body that had been AF calibrated with the revised service department calibration software?

Yes, one, an early production body. It had previously been through the sub-mirror fix procedure as well and was loaded with firmware v1.2.5. For the rest of this article update, we'll call this one the old EOS-1D Mark III.

Q. Were you able to test a new body that had been through the correct AF calibration routine at the factory?

Yes, two of them, both new units and both certified to have been through the correct calibration procedure at the time of manufacture. Each had firmware v1.2.5 installed. One was on loan from Canon USA, the other obtained independently. We'll call these the new EOS-1D Mark IIIs.

Also, we shot with two different EF 300mm f/2.8L IS lenses, one that had been AF calibrated by Canon service, the other we don't know (though it appeared to be focusing correctly).

Butter Fingers: Canon EOS-1D Mark III (firmware v1.2.5, revised AF calibration), ISO 400, 1/3200, f/2.8. Click to enlarge (Photo by Rob Galbraith/Little Guy Media)
Q. What did you photograph?

Football (practice), rugby and track mainly, plus strobed volleyball (which involves almost no continuous tracking, but gave an opportunity to see if the camera's peppy initial AF acquisition was impacted by Canon's changes).

Site co-editor Mike Sturk and I did most of the shooting side by side, first to compare the autofocus performance of the old EOS-1D Mark III (prior to being put through a revised AF calibration) with the new EOS-1D Mark IIIs, and then to swap 300mm lenses between the bodies being tested to ensure that the glass wasn't skewing the results.

The lenses were set at maximum aperture (f/2.8) for almost all photos. Exceptions were strobed volleyball, where we closed down two or three stops as per typical indoor strobed sports photography, and a handful of outdoor sequences where the aperture slipped to f/3.5.

Q. Which autofocus points did you use?

Because the stated change in Canon's AF calibration procedure affected focus points other than the centre one, we spent a bit more time with the camera set to use the outer 44 (in addition to the centre point). Ultimately, we tried the following AF point configurations:
  • Manually-selected centre point + six AF Assist points
  • Several manually-selected points to the extreme left and right of centre, + three or more AF Assist points
  • Manually-selected centre point only
Q. Does the revised AF calibration or firmware v1.2.5 bring an improvement in tracking autofocus?

No. The EOS-1D Mark III with these changes does not capture a greater proportion of properly focused or nearly focused frames when continuously tracking a moving subject. As before, the camera has a tendency to spontaneously focus in front of where it should - though it will at times shift the focus behind for several frames too, especially when Assist AF points are enabled - and it happens often enough to produce an unacceptable number of missed peak action moments.

With a quick-focusing lens like the EF 300mm f/2.8L IS, the EOS-1D Mark III remains able to rapidly acquire focus when a subject is moving and the AF system is first engaged. If the picture is taken immediately after focus is acquired, focus is often accurate too: the EOS-1D Mark III remains a killer strobed volleyball camera for this reason. Tracking, however, is still the camera's Achilles heel. This is true regardless of the tested AF point configuration, and regardless of whether single frames, short bursts or long bursts are captured.

Bumpy: Canon EOS-1D Mark III (firmware v1.2.5, revised AF calibration) + EF 300mm f/2.8L IS, ISO 320, 1/500, f/5.6. Click to enlarge (Photo by Rob Galbraith/Little Guy Media)

In short, there's nothing in the over 4000 frames we've shot leading up to this article update to suggest that the revised AF calibration or v1.2.5 firmware have improved the camera's principal remaining autofocus weakness. If you shoot sports, or something equivalent to it, and therefore require strong autofocus tracking, the EOS-1D Mark III will continue to disappoint. It can capture some in-focus tracking sports pictures at wide open apertures, as demonstrated by the several example photos sprinkled up and down this page. But, it continues to flub the focus on more frames than it should for this level of camera. And it's still easily outclassed by its predecessor, the EOS-1D Mark II N (and the EOS-1D Mark II before that).

  • We've alluded previously to the seemingly odd behaviour of EOS-1D Mark III autofocus when C. Fn III-8 [AF Expansion with Selected Point] is enabled and there's more than one subject in the frame, but it's not something we've emphasized. That was probably a mistake, because this setting is a problem. Switching on the camera's Assist AF points can lead to more out of focus pictures in some situations, including those that this option is meant to help.

    Specifically, there's a miswiring in how the camera utilizes the Assist AF points when C. Fn III-8-1 or III-8-2 are chosen, the result sometimes being the camera's tracking of the wrong subject. A related culprit is C. Fn III-4 [AI Servo AF Tracking Method], which is supposed to control how the AF system priorizes when choosing an AF point in the active group, but in practice it doesn't work.

    For instance, with C. Fn III-4-1 selected, either C. Fn III-8 expansion option turn on and a manually-selected single AF point chosen, the camera should give priority to the manually-selected point, particularly when it's a closer subject. In reality, when photographing running subjects, the camera is as likely to attempt to track the more distant subject, even when the distant subject is all but covered up by the closer subject and the manually-selected point is aimed square at the closer subject also.

    It's easier to see than explain, and the downloadable sequences at the end of this page show the problem plainly (see pictures 208-229 in the ZIP archive of photos from the old EOS-1D Mark III with revised calibration). We'd all but stopped using the Assist AF points in this camera because of this quirky tracking behaviour, but chose to try again because of the possibility that the quirkiness could have been related to a miscalibration of certain outer AF points, which the revised calibration might have then fixed.

    This has not been the case, and so we're back to keeping C. Fn III-8 locked to its default of III-8-0.

  • As noted earlier, for this article update we tried three EOS-1D Mark III bodies and two EF 300mm f/2.8L IS lenses. All conclusions here are based only on two of those bodies, an older unit that has been through the service department's revised calibration procedure and a brand new one that came from the factory already certified as being calibrated properly.

    We put aside the third body - also brand new and also one that came with the factory's seal of approval on its AF calibration - because its autofocus performance was noticeably worse than the other two (or in fact any EOS-1D Mark III we've used since the sub-mirror fix was introduced). We didn't attempt to figure out why, other than to try this body with both 300mm f/2.8s here and confirm that it focused equally poorly with each. It did.
The Eyes Have It: Canon EOS-1D Mark III (firmware v1.2.5, revised AF calibration) + EF 300mm f/2.8L IS, ISO 640, 1/3200, f/2.8 (Photo by Rob Galbraith/Little Guy Media)

Q. I've heard of photographers whose EOS-1D Mark III focused much better after being sent in for the revised AF calibration routine. Why are they getting better autofocus and you aren't?

We've heard the same, and in fact have been contacted directly by a handful of shooters over the last couple of months who claim dramatic autofocus improvements upon getting their camera back from Canon service. We don't doubt what they're saying is real, and that the explanation probably lies in the following:
  • First, a different mechanical starting point. Several of the photographers who've been in touch, and included a description of the service work done, said that Canon replaced one or more bits of hardware inside the mirror box at the same time as the revised AF calibration was performed. One indicated that the entire mirror box itself was removed and a new one put in. This suggests that these photographers' EOS-1D Mark III bodies had mechanical problems instead of, or in addition to, an AF miscalibration, problems that would likely contribute to worse autofocus performance than we were seeing previously.

    The fact that Canon has also advised owners of this camera of possible abnormal main mirror operation is probably further evidence of a disproportionate number of EOS-1D Mark III bodies being mechanically out of kilter in a way that could have lead to seriously degraded autofocus performance.

  • Second, a different calibration starting point. Obviously, there are EOS-1D Mark IIIs out there that were miscalibrated at the time of manufacture, and perhaps by Canon service too, otherwise Canon wouldn't be offering to perform the revised AF calibration procedure with revised AF calibration software for EOS-1D Mark III owners.

    From what we've been able to learn from Canon representatives in the U.S. and Canada, however, not all bodies that shipped, prior to the discovery of the calibration error, were in fact miscalibrated, only some. So it was within Canon's capability previously to do the calibration correctly, even if it wasn't being done correctly all the time. It's likely, then, based on the number of calibrations the various EOS-1D Mark III bodies we've used have been through, that our gear has been well calibrated in the past, perhaps well beyond typical tolerances. Taking just the old EOS-1D Mark III we tested for this update, it has had its AF system checked and/or calibrated a total of five times in three countries on two continents, in addition to the initial factory calibration. So, it may well be that we've been shooting with better-calibrated cameras than some others out there.
We could go on, since there are various other possibilities too. Including different shooting requirements that don't stress the AF in the same way peak action sports does. Whatever the reason, if you're among those whose EOS-1D Mark III AF is working much better now and, more importantly, is also meeting your expectations and allowing you to make pictures you previously were missing owing to blown focus, then more power to you: your EOS-1D Mark III is now probably working as designed. While we continue to find fault in that design, if for what you shoot, how you shoot, the lenses you shoot with and so on, the camera's getting the job done, there's no downside to that.

Q. Were you surprised that the revised AF calibration and new firmware didn't improve autofocus performance for you?

Based on what Canon has revealed about these changes, no.

There are no stated tweaks to the camera's AF system in firmware v1.2.5, and the revised AF calibration was to correct a problem affecting the camera's outer 44 points, whereas we'd mostly used the centre one alone. Going only on Canon's announcements about the firmware and the calibration, then, there would be little reason to expect an improvement in tracking autofocus, given how we'd mostly shot with the camera, and therefore little reason to be surprised when there wasn't one.

We opted to come out of EOS-1D Mark III testing retirement because we'd had a lot of reader requests to, because of the small but very happy number of shooters who were telling us how much better their camera's AF was after going through Canon's service, because of the real, if slim, possibility that the revised calibration could shore up Assist AF shooting and because the opportunity presented itself, in the form of Canon USA being amenable to loaning gear to supplement what we already own, thereby enabling the possibility of a useful test.

We set out to do enough testing to see if more was warranted, and after several sporting events it was obvious that the the camera's middling tracking capabilities are not improved. With that, we're done once again with testing the AF system in this camera.

Q. Does all of the above apply to the EOS-1Ds Mark III too?

We haven't used an EOS-1Ds Mark III that has been through the revised calibration procedure, so we can't say.

Downloadable sequences

Below is a small batch of downloadable sequences from football, rugby and track.

Rush Hour: Canon EOS-1D Mark III (firmware v1.2.5, revised AF calibration) + EF 300mm f/2.8L IS, ISO 200, 1/6400, f/2.8. Click to enlarge (Photo by Mike Sturk/Little Guy Media)

The photos are divided into three fairly large bundles, based on which camera was used to shoot them:
  • Old EOS-1D Mark III, before the revised AF calibration. These are fresh pictures, shot a few weeks ago, just prior to sending the camera to Canon Canada service. Example filename:


  • Old EOS-1D Mark III, after the revised calibration. Example filename:


  • New EOS-1D Mark III, certified to have been through the correct calibration procedure at the factory. Example filename:

All photos are marked with the manually-selected active AF point and, where applicable, the associated Assist AF points. The manually-selected point is a bright red frame, the Assist AF points are deep red frames. When Assist AF points are enabled, the picture metadata doesn't store which point or points in the group the camera chose, and it may well have been a point other than the one with the bright red frame. But, there's no way to know that or show that.

All downloadable photos were shot with C. Fn III-4 [AI Servo AF Tracking Method] set to III-4-1, [Main Focus Point Priority]. When Assist AF points are switched on, this Custom Function is supposed to give priority to the manually-selected AF point as well as whatever's closest, but as you'll see it does neither of these things.

The photos mostly demonstrate the camera's tendency to misfocus while tracking a moving subject, but there are sequences that also show the camera can track well, at times. Generally speaking, this is true when the camera-to-subject distance is changing slowly and none of the things that trigger EOS-1D Mark III focus error are present. Last year, we'd found that if the lens was also stopped down that tracking improved, but we've not attempted to verify that this trick still works now (see the workaround section on this page for more information).

Click on a thumbnail to download a ZIP file containing full-resolution pictures from the particular EOS-1D Mark III described in the caption beneath (processing notes).

Download: Old Canon EOS-1D Mark III (firmware v1.2.5, old calibration) + EF 300mm f/2.8L IS, AI Servo AF. Click thumbnail to download, 93 pictures, 132.7MB (Photos by Rob Galbraith and Mike Sturk/Little Guy Media) Download: Old Canon EOS-1D Mark III (firmware v1.2.5, revised calibration) + EF 300mm f/2.8L IS, AI Servo AF. Click thumbnail to download, 248 pictures, 382.7MB (Photos by Rob Galbraith and Mike Sturk/Little Guy Media) Download: New Canon EOS-1D Mark III (firmware v1.2.5, revised calibration) + EF 300mm f/2.8L IS, AI Servo AF. Click thumbnail to download, 224 pictures, 366.9MB (Photos by Rob Galbraith and Mike Sturk/Little Guy Media)

Back to Top: An analysis of EOS-1D Mark III autofocus performance
Go to Page: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 
Go to advertiser website.
2000-2013 Little Guy Media. Not to be reproduced without written permission.