Jeff wrote:
New firmware did nothing for my camera, although I suppose it is possible (based upon recent flaky results with the 24-70) that it made it worse. I hadn't thought of that...

Firmware 1.2.3. made all the difference in the world in my blue dot. I went from not trusting the MKIII to using it for everything. I may be in the minority, but it really changed how my MKIII works....for the better!

Bob

Well, for all of you that have kept up: I sent a second letter to Canon's president on Saturday, I think. I got a call from a Canon rep this afternoon, to tell me that they are exchanging my Mark III. There is already a shipping label on the way, and once I can prove that the old body has shipped, they will send me a new (hopefully, probably refurbished) body...Too bad this process has been close to a year!

Jeff wrote:
Yakim Peled wrote:
So, situation is still that you have better AF than the first body but still not 100% satisfactory?

Happy shooting,
Yakim.


Correct; definitely better in AI Servo than the first one, but still has erratic One-Shot focusing (mostly at f/2.8).

Ouch.

Happy shooting,
Yakim.

Jeff wrote:
Yakim Peled wrote:
So, situation is still that you have better AF than the first body but still not 100% satisfactory?

Happy shooting,
Yakim.


Correct; definitely better in AI Servo than the first one, but still has erratic One-Shot focusing (mostly at f/2.8).



EXACTLY the same experience I am having (505XXX). Camera with Canon CPS right now.

lermite wrote:

- Center AF point only, pointing on a stationary subject, stable support (tripod or bean bag) --> the AF is continuously adjusting back and forth by a significant amount even if camera and subject aren’t moving at all. This results in a lot of out of focus images (even at f8), as with wildlife you haven’t always the time to switch to One Shot mode. (The MkIII white paper states: " Even if AI Servo AF is used to focus on a still subject, stable control is enabled so that the lens drive isn’t constantly fine-tuning. If the subject starts to move, the focusing responds immediately to detect the subject constantly.”).

My photography (sports, corporate, events) isn't quite the same as yours but I found that after installing firmware 1.2.3 the issue of AI Servo twitch dropped off considerably, though not completely gone. Certain of my lenses, particularly the faster ones and the 50 f/1.2L in particular, still cause substantial AI Servo twitch. But with lenses like the 70-200 f/4L, 400 2.8 IS or 600 IS, the camera is much calmer and doesn't feel like it's trying to focus on every molecule of air between it and the intended subject.

One note that might interest you: You can quickly switch between AI Servo and One Shot by setting CF III-6-4 and using the AF stop buttons on the big IS tele lenses. Personally I prefer III-6-5 to enable IS start and remove AF start from the shutter release so I can lock AF by letting go of the AF-ON button (I actually reverted back to the * button).

- The slower a subject moves the less reliable is the AF tracking. I had rather good results at 10fps testing on cars driving towards me on a motorway at 120km/h, but with 4fps and birds slowly flying, walking or swimming towards me, more than half of the images are out of focus.

Yes, I found this too and Rob Galbraith has also reported it. I noticed it in photos of people walking or running at a slow pace. Those tend to have a higher number of out of focus images in a sequence. I can't say I've found a solution for it...

Ron

I've actually seen that as well. AI Servo of people merely walking toward the camera befuddles the AF system to an abnormal extent (regardless of FPS). It's like it's expecting the subject to be a certain speed or something...

Jeff wrote:
I've actually seen that as well. AI Servo of people merely walking toward the camera befuddles the AF system to an abnormal extent (regardless of FPS). It's like it's expecting the subject to be a certain speed or something...

if the AF algorithm in ai-servo is "predictive", then it's reasonable to assume that it extrapolates the subject's distance and/ or speed toward/away from the camera based on previous data points. that assumption, coupled with the 1d III AF servo's aggressively fast tuning = error prone.

the following factors also kill things more, for any ai-servo system:
- fast subject speed
- close subject to camera distance
- non-constant subject speed
- wide lens aperture

i don't think you can get accurate and fast at the same time. i'm sure they (canon engineers) strove for a good balance, but didn't quite get there. i have fought these challenges with robotic servo control before (and lost), with better servo motors.