Alan, thank you for that information.
In my opinion, you really could not have had a completely rigid geometry between the AF sensor and your target because of handholding. However miniscule movement and contrast/phase change might have been there, one would expect Servo AF to pick it up and respond to it. Right... respond, but not in the way of throwing the focus completely out of whack.
Alan321 wrote:
Then focus got back onto the couple for another 4 frames until the sequence ended.
These subjects are not large in the frame but are certainly bigger than the central AF sensor. There remains enough contrast on the subjects for the focus not to drift off to the far distance another 100m away.
- Alan
Did the camera have something better to focus on during the last 4 frames compared to the few where it missed?
It looks like the camera is pointed in the dark brown sweatshirt in the frame you show. It probably decided to find something else to focus.
The first frame at least has the arm/sweatshirt combo to work with.
This thread really reminds me of dpreview somehow :P
I don't really belive handholding a stationary target with AI Servo can show the problem. Moving objects should. Could you shoot for example a model train or another moving object that moves at a constant speed and direction?
Being that Alan has also taken a sequence of shots with a couple walking in a park. Showing us AF drift. Then it appears there is a problem with some copies of the 40D. Is any one else getting this? We know the 40D has no problems taking great pics of stationary subjects while in one-shot AF mode.
Looking forwards to hearing from Alan after he gets his 40D back. Was it really the camera? Hmmm....
Alan321 wrote:
Secondly, what if we want to focus on something close-ish such as a wildflower without using a tripod ? Your own movement can upset the focus unless you use AI Servo mode.
I use that regularily for macrophotography. Not only do I move - the flower may also move due to wind. However, this type of movement seems to be somewhat irregular, and the depth of field is so shallow, that only 1 in 10 images is sharp exactly where I want it to be (with my 30D). Some are even unfocused everywhere, so this is at least a tough job for AI Servo.
Last Sunday I was photographing steeplechase (=horses over fences), using the middle AF point and AI Servo. When I checked them at home, 80% of the images were in focus where I wanted them to be. Some of the rest were user errors (AF point not on target or shot taken too quickly for AI Servo to stabilize on the target) and some camera issues. In any case, I am happy, as 30D is no 1D series camera (and did not cost anywhere near as much), so I don't expect a perfect performance.
Alan321 wrote:
The next two frames were focusing behind and then the next three looked like this...
The problem here might be with the diagonal focus sensors. It is a known fact that the focus sensors are clearly larger than the indicator square in the viewfinder (check the bottom of the mirror box in your camera and you'll get an idea of how big they are). With the 40D, the diagonal sensors may make this discrepancy even larger.
What I see in your example images is that a strong contrast (that of the hill in the background) approaches the focus point, and at the point of the misfocused image may get grabbed by one of the diagonal sensors. I've understood that the diagonal sensors are very sensitive.
According to some (BIF) reports here, the 40D seems to be faster than 30D in changing the subject. 30D has a perceptible delay (about 0.5 seconds in my estimation) before it starts tracking a completely new subject. 1D series cameras have that delay adjustable via a custom setting but x0D camera owners need to accept how it is set.
So far it seems that for BIFs, 30D, for other purposes, 40D has better AF. Maybe flower macros fall in the domain of 30D as well?
Because I was using an f/3.5 lens the diagonal AF sensors should not have come into play at all.
Although I was focusing on the woman's dark top there was a distinct bright stripe offering hirizontal lines of contrast. There was also reasonable contrast between her top and the background, or so I reckon.
CarpeyBiggs wrote:
It can't just "freeze," otherwise it will fail to find the new focus point and pattern. The camera can't find focus without the focusing mechanisms actually moving. That is how focus is detected. If it "freezes" it is no longer capable of finding focus again.
Focus is actually detected by the camera looking at the defocused contrast pattern being presented to it and deciding which way and how far to move the lens to make the focus correct. It generally does not search for correct focus because it already knows how to get there. It then issues the command to the lens and leaves it up to the lens to get there in a reasonable amount of time. In AI Servo mode it will then continually reassess focus.
Therefore the camera can safely suspend focus changes and regain focus without initiating a lens search, should it lose the plot for a moment. Any subject that it has been successfully tracking successfully for some seconds should not be causing it to give up and initiate a lens search, in my opinion.
If the defocus is too great then the camera can't see enough of a contrast pattern to determine the required focus and it initiates a lens search. During the search the camera is trying to measure focus again but this time it has a moving target. Faster, better AF systems handle this better, especially if contrast is low and if the lens aperture is large. Sometimes in lower light a large aperture lens will not focus once a search has been initiated - you can see focus snap in and out before the camera has responded. In my examples a lens search during the shooting sequence might have been done but it was not needed. If it did do a search then it quickly regained focus long before the lens had a chance to do the full search pattern and so that search was detected by me. Normally I recognise when it is doing a lens search by the sound and the viewfinder image.
During the lens focus search the predictive focus is not being performed - that stopped as soon as the camera decided that it had lost focus and will take a while to resume after it regains focus.
The 40D is better than most at grabbing focus from a more significantly defocused position, and that helps it avoid initiating a lens search, In many cases that alone can speed up AF greatly. In the Canon range only the 5D and 1D3 have similar capability.
If more than one AF sensor is active then the camera will use the other sensors in preference to initiating a lens search.
I do not want to hijack this, but did you try it with one-shot, too?
I was expecting probs with my 1D MK III, where in one-shot I got different results on a tripod with remote etc. on a stationary object. One shot after the other the focus hunts a bit and results in different places.
Focus as shown in zoomBrowser is every-time center, whereas other areas are sharp. Seems that AF seems to be easily mislead if other "interesting" areas are covered by other f-points. If I aim at an edge, where no other areas are in front or beside, than AF is a win-win.
Will wait for firmware-update, before I will contact CANON. Sending expensive stuff is not my favourite
Thnx for your time and regards,
Oliver.
Edited by Oliver Francke on Sep 19, 2007 at 10:35 AM GMT
Beni wrote:
Canon cameras cannot be used with AI Servo on a static subject, the AF starts to hunt. This was detailed on the old CPS FAQ site which I don't think still exists? This has been long known.
dcains wrote:
Call me crazy, but I'd have to say that AI Servo isn't for potted plants.
Well if you're crazy then so am I and so is the techo who told me not to shoot test shots in AI Servo mode. AI Servo is for moving subjects and while I accept that Alan has had the same problem with tracking focus on moving subjects, it might have been worthwhile checking it's performance on pot plants in Single Frame mode before sending it off.
Sorry to put it this way Alan but I was having similar problems with inconsistent focus on my 300mm F2.8 until the techo pointed this out to me.
Was the focus confirmation light blinking or was it off. If it was blinking then it wasn't tracking. Anyway when I use AI servo I use the * button for AF so I can just release it it to lock focus if the subject stops moving; I would not keep the * button pressed for a static object as it's well known you can get misfocus.
I don't know if the new Zoom Browser will display RAW files. In fact, I can't remember if the old one does (I use Adobe Bridge and don't shoot RAW that often). Even so, if you can convert to JPEG you will still be able to display the screen cap with focus points.
Thanks for posting the link. I'm considering buying the 40D. Or was. Not so sure now. Is a 58% OOF rate "common" with all DSLRs when shooting movement? Of is 58% considered very high?
I asked Drew to test the upcoming Nikon D300 and Sony A700. Hope he does. Would be interesting to see how they compare.
Edited by Andrew Chase on Sep 20, 2007 at 03:26 PM GMT
I don't think testing a third party tele-macro lens at close to or at the minimum focusing distance and hand-held (?) & on Al Servo (?) is a fair test. Macro lenses are really designed to be on a tripod, and manually focused. There is no margin of error. When a centimeter here or there determines critical focus and you don't intervene to override the camera's AF then something is wrong with your shooting technique as it applies here.
From the above link showing "running man" and the resulting 40D slightly deficient keeper rate compared to the 1 series... what do you expect? If a better keeper rate is needed use a movie camera.
Andrew Chase wrote:
Thanks for posting the link. I'm considering buying the 40D. Or was. Not so sure now. Is a 58% OOF rate "common" with all DSLRs when shooting movement? Of is 58% considered very high?
I asked Drew to test the upcoming Nikon D300 and Sony A700. Hope he does. Would be interesting to see how they compare.
I'm sorry Andrew that your relying on Drew's evaluation of the 40D's AI Servo performance because his test is questionable at best. The BG's are different and it doesn't even look like he ever had the subject in focus at the start of the test.
The AI Servo drift that is being proposed by the OP is another example of bad information based on a flawed test. Make sure you get good information before you make a decision about buying the 40D.
In addition if your expecting the perfromance of a 1 series camera then you will not be satisfied. It perfroms better then any xxD camera before it and is a good improvement in AI Servo performance.
Let's assume that the tests between the 40D and 1 series cameras were done the same, although flawed. While the number of oof shots for the 40D looks to be alarming, it is only 14% behind a 1 series body. I don't think that you can ask for much more.
