IMHO it's quite clear from RG's images provided today that the camera still suffers some sort of inherent design flaw in the system, obviously affecting more than just the sub-mirror assembly. There are two things I found particularly disturbing about RG's report:

1) The little ditty about setting AF Microadjustment to -1 fixing some of the issues, and all the caveats thrown in. If you haven't read that part, you need to see it, it's really a rather odd response from Canon as to how the function should be used, and not used.

2) And, more importantly, I must point out a very interesting phenomenon in the 'fast runner ' series, something I've occasionally noticed about my MkIII's soft images. The examples show it very clearly and consistently, and I've not heard RG or anyone else comment on it with respect to actual example images. I've provided a screengrab at the bottom of the post to clearly show what I'm talking about.

[In regard to RG's 'fast runner', I would agree that about 50% is closer to the mark for the MkIII in the 'fast' running sequence (the MkIIn did indeed perform significantly better). For some reason the hurdle sequence was better, perhaps illustrating the somewhat random nature of the beast, although the subject in that one was wearing a white shirt.]

Note in the screen grab below that virtually every time the MkIII has an out-of-focus image, it almost ALWAYS exhibits some sort of secondary 'ghost' image, suggestive of some sort of movement (which at 1/8000th, I'd think is unlikely on RG's part). This is very clearly and consistently exhibited on the 'Brooks' logo the runner's blue shirt, since it has a high-contrast area (dark background with white lettering). With the camera in the vertical orientation, the 'Brooks' logo has a ghosted image above it, indicating some sort of movement somewhere in the system (side-to-side in relation to a landscape orientation). This would obviously contribute to an image's overall softness regardless of the actual focal plane, though it appears to me to be consistenly manifested mostly when the camera ALSO has focused either slightly in front of (or more typically) slightly behind the subject. Note that the MkIIn's OOF images do not exhibit this odd characteristic (at least that I could find), suggesting the lens has nothing to do with it.

Whether the cause of this apparent 'movement' in the system is also causing the focus errors becomes the question in my mind (there is also a chicken-and-egg component here). I suppose there is always the slight chance that it could be lens/IS related, but I would doubt that he was using IS for such sequences, and one would expect it would have been duplicated in the MkIIn images. I have no idea what component in the system could be moving, whether it is a result of the new anti-dust mechanism, due to the sensor moving (which seems unlikely), or some sort of overall 'looseness' in the parts of the camera that is exacerbated by the MkIII's extremely fast shooting rate.

Like I said, I have noticed these 'ghosted' secondary images in several of my OOF MkIII images, and as you all have probably seen, there is a fair amount of internet chatter about overall image softness (not the all-over OOF 'misses' that we've seen so much of), though image softness can often be attributable to user error of a variety of sorts.

====> I would suggest that any of you interested take a close look into this series of images and see if you agree with what I'm seeing. It's quite obvious once you start really looking at the images at 100%.

[ As an aside, I'm currently using the Trial version of CS3, and Bridge3 makes it really, really easy to quickly evaluate these images, especially with the 100% loupe feature (a la Aperture). Nice! ]

I love this camera, the files from it are exceptional, but I bought it to shoot outdoor sports (in addition to serving the purpose that my 1Ds served, which it admirably can). This issue clearly has not been resolved, and it's writing on the wall for the 1DsMkIII (which is, I'm certain, why Rob took so long to make certain of his results). It's also clear by reading in between the lines that it affects every single MkIII produced prior to October 2007, despite the fact that some people have not seen any AF inconsistencies in their specific usage(s). For example, notice that they are no longer talking about 'adjustment', but 'replacement'.

Anyway, more food for thought. Mine will be getting the submirror fix, but to what end remains unclear for my shooting.

Regards,

Jeff

PS: I'd be very curious to hear others' thoughts on a potential reason for the results noted in the screengrab below. I'm getting tired of talking about this too, but RG's tests are significant points of data in this ongoing puzzle, and today's installment provides a compelling illustration of the camera's ongoing AF quirks.


Edited by Jeff on Dec 10, 2007 at 07:11 PM GMT (Reason: Moved wide image below to re-wrap text properly)

Edited by Jeff on Jan 03, 2008 at 02:37 PM GMT (Reason: Title change for update)

Edited by Fred Miranda on Jan 07, 2008 at 03:23 PM GMT

Edited by Jeff on Jan 23, 2008 at 03:55 PM GMT (Reason: Title change for final update)

Edited by Jeff on Jan 26, 2008 at 12:37 PM GMT (Reason: yet another title update to reflect current status)

Screengrab (from RG's "Fast1" sequences):

I concur with your findings re: the ghosting issue... Typically for my Mk III it's the same problem as well - the out of focus images tend to be a little off, and sometimes I can't tell whether it's movement or out of focus or both. Perhaps it is because of the vibrating filter?

Mine is in for the fix right now, and will be back shortly. I've not run into huge problems so far with it, but there's been enough of a problem to require sending in.

Nice analysis Jeff. I have a feeling that 2007 will be remembered as the year canon laid 3 eggs*. Oddly enough, it seems that the 40d will probably be the best of the 3 new cams, and it is woefully outclassed by the d300. It's a shame, but canon got complacent in the midrange and suffered from an engineering mistake in the high end.
The good news is that 2008 will (hopefully) see the resurgence of innovation and putting everything into each product.

* OK, the 40D will prove to be a great midlevel camera so it doesn't actually qualify as an "egg".

hey, what about resonance? perhaps the 10fps hits some natural frequency within the camera body. this could effectively shake certain components and create this.

london bridge is falling down, falling down, falling down...

My N is on the way.....low clicks, $2300...yeah baby. Hope you "3" owners get this one worked out.

Thinking more about this, I'm now wondering if if the AF 'problem' is worse when shooting vertically, compared to horizontally...

It also seems the soft pics are in a series and then the camera makes some adjustment then goes blurry again. The more i think about it, it seems like some kind of processing problem rather than hardware

I'm a bit confused. Look at a shot like #12 (of Fast #1). The face is definitely just out of focus (or motion blur, which seems impossible). And yet the foot is in focus (though perhaps just within the DOF (when viewed at an enormous size, like the equivalent of a 26" x 39" print).

To me, the entire series looks just a wee bit front-focused.

But honestly, when viewed at something like 10" x 15" on the screen, almost all shots are "acceptable", and perhaps even within Canon's specs (1/3 DOF at 8x12?). #12 is one that would be just past "acceptable" at that size, to me.

Another thing . . . I'm not sure my 1D-2 would give a series of shots any better than this (even though RobG's seems to). Does he just have some spectacular 1D-2's?

And one last comment . . given this much light, I'd simply stop down to f/4 for shots like this, which would probably make ALL of them acceptable. If you look at the DOF in these shots at f/2.8, it leaves little margin for error. I'd take a bit less background blur at f/4 in a situation like this to get a crisper-looking subject. Wouldn't you?

Jeff wrote:
IMHO it's quite clear from RG's images provided today that the camera still suffers some sort of inherent design flaw in the system, obviously affecting more than just the sub-mirror assembly.

I agree with you on that one, based on a series of symptoms we have seen in the past 6 months or so. It looks as if Canon have been trying to mitigate the problem with iterative design band-aiding . That approach may yield incremental improvements which should tie them over until the next model camera is ready for release.

I'm not sure if its related or not, but I do alot of aerial photography, and when I use a gyro-stabilizer on the camera and we're banking tightly around a site I sometime have a situation where the resulting images are sharp in the center but very fuzzy at the horizontal ends of the image. I always felt that it was caused by some sort of torsional vibration. Is there something going on inside the MkIII's that might cause a similar effect.

I can also tell you from working closely with japanese engineers at a previous job that even if they know what the problem is they are very unlikely to ever release that information to anyone here in the United States. That would be admitting a mistake and losing face.

That ghosting is just like mine only the ones pre fix were worse. I described it as looking through a blur filter

Jeff, you should email this to RobG to get his comments on it.

I second DavidP on that Jeff.

R Longenbach wrote:
I second DavidP on that Jeff.

Thanks.

Just MAYBE Canon could use these image to pick up on that "something else" that seems to be causing issues and figure out how to fix it.

I sort of wish the two camera (IIN and III) had been set up and fired simultaneously. Would have been a more interesting comparison and perhas more useful comparison.

Good catch Jeff.

I think the III will be remebered as a cursed body and will have the shortest production run time of any of the 1D digital bodies. Very unfortunate, but not entirely unexpected when you consider how many envelopes a body like this pushes.

I fully anticipate the production run of 1D3's to be the shortest in the 1D series. We will find out soon enough. I wonder if the next body will be a 1D3N, 1D4, or possibly skipping an unlucky "4" go right to the 1D5?


Paul

Great analysis Jeff! Mine goes back tomorrow...will see how it works out.

DavidP wrote:
Jeff, you should email this to RobG to get his comments on it.

I did exactly that this morning as soon as I took a look at his images. Something tells me Canon will be a little less 'proactive' in running this one down. I think the submirror certainly fixed some issues with the jumpy/errant AF points, but this is clearly something different. And odd.

nathanlake wrote:
I sort of wish the two camera (IIN and III) had been set up and fired simultaneously. Would have been a more interesting comparison and perhas more useful comparison.


Maybe, but frankly I was impressed with the consistency of Rob's images between the two series. Gives credence to his shooting skills and methods, to be sure.

Jeff wrote:
Thinking more about this, I'm now wondering if if the AF 'problem' is worse when shooting vertically, compared to horizontally...

I just took a look at the 'static' AI Servo series, which was shot horizontally. Interestingly, the 'motion blur' (for lack of a better descriptive term) tends to be in the same axis, in the same direction, e.g. it shows as ghosting to the right of the 'Brooks' logo.

And, the last thing I took a look at tonight was prompted by DavidP's thread, and relates to accuracy of comparison sequences. In the 'slow' series, I was able to find the runner in virtually identical positions, in nearly the exact same spot on the track. The bottom 'loupe' view shows two things: that the focal plane was in a nearly identical position, and the runner was no more than 1 to 2 inches fore/aft of the exact same position in both images (note the slight softness of even the specular highlights in the MkIII image).

More importantly, the upper loupe view shows a significant discrepancy in the sharpness of the runner's face, even giving some wiggle room for the runner being up to 2" off in distance to the camera. It's this kind of result that MkIII owners familiar with the MkIII's AF problems readily admit to being anything but 'normal' for any camera. There is simply no logical explanation for it, other than some technically complex hardware issue.

Enough for the night, and adios!

Jeff wrote:
Note in the screen grab below that virtually every time the MkIII has an out-of-focus image, it almost ALWAYS exhibits some sort of secondary 'ghost' image, suggestive of some sort of movement (which at 1/8000th, I'd think is unlikely on RG's part).

I have no idea what component in the system could be moving, whether it is a result of the new anti-dust mechanism, due to the sensor moving (which seems unlikely), or some sort of overall 'looseness' in the parts of the camera that is exacerbated by the MkIII's extremely fast shooting rate.


Here are some more ideas, but I cannot check their validity as I don't own a 1D III. I hope, though, there is some idea in them that might give some spark to people who have a lot of experience with the camera.

1) try reducing the frame rate and check whether the problem still persists,

2) maybe the camera expects the lenses to be (slightly) faster in focusing but when the camera is taking the next frame, the lens is still settling for the latest focus change? After all, with the higher frame rate, there is less time for focus changes. Maybe not all lenses are fast enough to change their focus and freeze all movement before the next image is taken (being still in slight oscillation, which, with the ultrasound motors, would be of high frequency).

3) or maybe it is simply a timing issue in the firmware (sending the focus change command too late or opening the shutter too early)?

4) short exposures might actually make the problem more visible. If the unstable period is of a fixed length, it will be a much larger proportion of the whole exposure time. Therefore it would appear more often in bright conditions (as you get a shorter exposure time).

Can anyone confirm if the anti-dust filter is bonded to, or part of the the same assembly as the bayer filter?

My thinking is that if the bayer filter is indeed part of the anti-dust assembly and there is an electronic glitch with the anti-dust filter control electronics, (or the assembly is mechanically loose to where high frame rates cause it to oscillate), maybe that is causing the odd "motion blur" that appears in a particular direction in relation to camera orientation.

Jeff's conclusions are very interesting. I have no idea why ghosting appears in the out of focus shots but I too am concerned why -1 is dialled in on a perfectly calibration matched body and lens. I mean I have heard of athletics photographers having their bodies intentionally front focussing to catch the face in focus for the lean over the finish line but having to have -1 microadjust dialled in for every action shot in AI-Servo seems frankly ridiculous!

I also see these strangely unsharp pictures.

In some cases like the logo on the t-shirt maybe there can sometimes be other explanations though, the edges of the stitching/printing is perhaps not all that sharp, maybe there's lens CA at the high contrast edge between blue and white, maybe there's sensor blooming because of local overexposure (though CA and blooming usually have a color to it). Just some ideas, I'm not saying this explains anything or everything of course.