There has been a lot of talk on various forums regarding supposed problems with the performance of AI Servo focusing on the MKIII. I am posting this note based on my own experiments, in the hope it may help anybody who feels they are having problems. I have not experienced any problems with single shot AF as described in the Roger Galbraith article, but then it does not often get that hot here
On my first few attempts to photograph birds in flight with the MKIII I was not pleased with the results and started to believe there might be an actual problem with the camera. I set about experimenting and fully evaluated the nearly 5gb of flight photos I had taken in the previous few days to try and get some decent examples for Canon to describe the "problem" with AF. I carefully analysed the photos in Zoombrowser, which displays the selected focus point, and carefully read the manual (a foreign concept to me ). I have now concluded that very few of the photos in which the initial frame, and therefore the subsequent frames, are OOF cannot be put down to user error based on the usage rules described in the manual. As a result I am now comfortable that there is not anything inherently wrong with the AF, at least on my camera.
The camera works very differently to the 1d II with regards to AF in AI Servo mode when using the "ring of fire". On the MkII the "ring of fire" would select an initial focus point automatically, not necessarily the centre point. Clearly p83 of the MKIII manual states:
"When AF point selection is automatic, the camera first uses centre AF point to focus. During auto focusing, if the subject moves away from the centre AF point, focus tracking continues as long as the subject is covered by the Area AF"
I have tested this in a controlled situation and it does exactly what the manual says, whatever is under the CP (CP = centre point) is what it focuses on, having said that if the subject under the CP has no contrast, e.g. a clear blue sky, it will find a subject on another focus point. My experiments seem to show that any focus point expansion selected in CF III-8 have no effect on initial acquisition of focus, they are used to hand off focussing if necessary when tracking starts.
If a MKII user does not take account of this different MKIII behaviour and assumes initial focus point will be selected automatically it may well be the reason for the perception that the MKIII will not focus well in AI Servo mode. Maybe Canon did this because it was sometimes slow for the automatic point to be selected and it was a bit of a lottery what it actually ended up focussing on.
I experimented taking flight shots of subjects which were often small, many against deliberately confusing backgrounds, I was using the drive at a full 10fps. Here are my findings:
Centre point selected
---------------------
All the initial shots in a sequence where I had accurately positioned the CP on the subject, and the subject was large enough to cover the entire CP, were acceptably in focus. As long as the CP remained on the subject the tracking appeared to be spot on in the vast majority of cases. On initial focus if the subject is not fully covered by the CP (e.g. part of a wing and part background water, trees etc.) then it is a matter of luck which part of the CP the camera chooses to focus on but I think that is only to be expected and would not be accurate on any camera.
Automatic Focus point (Ring of fire)
-------------------------
All the initial shots in a sequence where I had accurately positioned the CP on the subject, and the subject was large enough to cover the entire CP, were acceptably in focus. If CP focus was correctly obtained then the tracking seemed spot on in the vast majority of cases as long as the subject remained within the "ring of fire", even against difficult backgrounds. My feeling is that once I had realised about needing to use the CP for initial acquisition my success rate was higher than I would have achieved with the MKII and I am impressed by how well it tracks a small object against a difficult background.
To sum up, the AF behaviour is obviously different to the MKII and you cannot just pick it up and use it in exactly the same way as you used the MKII. It may actually be slightly more difficult to obtain the initial focus using the CP than letting the camera automatically choose it, but it should be more accurate once you get used to it. Might be nice if there was another CF:
CF III – 18 0 – In AI Servo use centre point for initial focus with Automatic focus point selection.
CF III – 18 1 – – In AI Servo automatically select initial focus point with Automatic focus point selection.
Final conclusion, I wouldn’t give mine back or swap it for my old MKII for anything.
"When AF point selection is automatic, the camera first uses centre AF point to focus.
Although not explicitly stated in the Mk2 manuals, this has been the way in which Automatic focus point selection has always worked, at least with lenses with a maximum aperture of f4, and probably with faster lenses too. I definitely think that the central focusing point in the Mk2 cameras covered a larger area of the view finder than is covered by the central focusing point in the Mk3.
I have now experienced some AF issues, tracking running hares. The animal starts out in focus in each sequence, (Automatic focus point selection/AI servo/focus tracking sensitivity set at mid-point ) and the 45-point elipse never moves off the animal, but all subquent frames are out of focus. The AF just didn't track.
Phil Shaw said "Although not explicitly stated in the Mk2 manuals, this has been the way in which Automatic focus point selection has always worked, at least with lenses with a maximum aperture of f4...."
Maybe so but it is not what I observed when doing my tests. I did my testing mainly with a 300mm f2.8 and some with a 600mm f4. Testing in a controlled situation using my 1dsMkII and 300mm f2.8 on an object against a grass backgound showed that it would focus on the foreground object even if the object was at the edge of the elipse ( I no longer have a 1D MkII to test the same thing). When performing the same test on the 1D MK III the camera always focused on the grass covered by the CP.
Welcome to the world of the D2x owner highly complicated overly sensitive and infuriatingly difficult to get consistent results, but when you do boy is it accurate and extremely fast.
"When performing the same test on the 1D MK III the camera always focused on the grass covered by the CP."
Isn't this what we want, though? I want the camera to focus on the thing I point it at, not to guess, or to focus on the thing that may be closest to the camera.
I know that when I'm photographing badgers with my 1Ds2 and 300/f2.8 using 45-point auto focus selection, what I get in focus is whatever the centre focus point is over. It doesn't guess that I want the eye to be in focus and therefore defer to the focus points at the right or left edge of the elipse.
I'm not trying to start an argument here, just trying to generate some light to better illuminate the Mk3 AF issues.
"Isn't this what we want, though? I want the camera to focus on the thing I point it at, not to guess, or to focus on the thing that may be closest to the camera."
I agree and I do not normally use auto selection anyway. I am merely relating what my 1DS2 did on my test and it is repeatable. It may be there was insufficient contrast for the 1ds2 to focus on the grass area covered by the CP whereas the extra sensitivity of the MKIII was able to focus.
Discussion
5 repliesOn my first few attempts to photograph birds in flight with the MKIII I was not pleased with the results and started to believe there might be an actual problem with the camera. I set about experimenting and fully evaluated the nearly 5gb of flight photos I had taken in the previous few days to try and get some decent examples for Canon to describe the "problem" with AF. I carefully analysed the photos in Zoombrowser, which displays the selected focus point, and carefully read the manual (a foreign concept to me
The camera works very differently to the 1d II with regards to AF in AI Servo mode when using the "ring of fire". On the MkII the "ring of fire" would select an initial focus point automatically, not necessarily the centre point. Clearly p83 of the MKIII manual states:
"When AF point selection is automatic, the camera first uses centre AF point to focus. During auto focusing, if the subject moves away from the centre AF point, focus tracking continues as long as the subject is covered by the Area AF"
I have tested this in a controlled situation and it does exactly what the manual says, whatever is under the CP (CP = centre point) is what it focuses on, having said that if the subject under the CP has no contrast, e.g. a clear blue sky, it will find a subject on another focus point. My experiments seem to show that any focus point expansion selected in CF III-8 have no effect on initial acquisition of focus, they are used to hand off focussing if necessary when tracking starts.
If a MKII user does not take account of this different MKIII behaviour and assumes initial focus point will be selected automatically it may well be the reason for the perception that the MKIII will not focus well in AI Servo mode. Maybe Canon did this because it was sometimes slow for the automatic point to be selected and it was a bit of a lottery what it actually ended up focussing on.
I experimented taking flight shots of subjects which were often small, many against deliberately confusing backgrounds, I was using the drive at a full 10fps. Here are my findings:
Centre point selected
---------------------
All the initial shots in a sequence where I had accurately positioned the CP on the subject, and the subject was large enough to cover the entire CP, were acceptably in focus. As long as the CP remained on the subject the tracking appeared to be spot on in the vast majority of cases. On initial focus if the subject is not fully covered by the CP (e.g. part of a wing and part background water, trees etc.) then it is a matter of luck which part of the CP the camera chooses to focus on but I think that is only to be expected and would not be accurate on any camera.
Automatic Focus point (Ring of fire)
-------------------------
All the initial shots in a sequence where I had accurately positioned the CP on the subject, and the subject was large enough to cover the entire CP, were acceptably in focus. If CP focus was correctly obtained then the tracking seemed spot on in the vast majority of cases as long as the subject remained within the "ring of fire", even against difficult backgrounds. My feeling is that once I had realised about needing to use the CP for initial acquisition my success rate was higher than I would have achieved with the MKII and I am impressed by how well it tracks a small object against a difficult background.
To sum up, the AF behaviour is obviously different to the MKII and you cannot just pick it up and use it in exactly the same way as you used the MKII. It may actually be slightly more difficult to obtain the initial focus using the CP than letting the camera automatically choose it, but it should be more accurate once you get used to it. Might be nice if there was another CF:
CF III – 18 0 – In AI Servo use centre point for initial focus with Automatic focus point selection.
CF III – 18 1 – – In AI Servo automatically select initial focus point with Automatic focus point selection.
Final conclusion, I wouldn’t give mine back or swap it for my old MKII for anything.
Jack Levene
Suffolk
England
Although not explicitly stated in the Mk2 manuals, this has been the way in which Automatic focus point selection has always worked, at least with lenses with a maximum aperture of f4, and probably with faster lenses too. I definitely think that the central focusing point in the Mk2 cameras covered a larger area of the view finder than is covered by the central focusing point in the Mk3.
I have now experienced some AF issues, tracking running hares. The animal starts out in focus in each sequence, (Automatic focus point selection/AI servo/focus tracking sensitivity set at mid-point ) and the 45-point elipse never moves off the animal, but all subquent frames are out of focus. The AF just didn't track.
Maybe so but it is not what I observed when doing my tests. I did my testing mainly with a 300mm f2.8 and some with a 600mm f4. Testing in a controlled situation using my 1dsMkII and 300mm f2.8 on an object against a grass backgound showed that it would focus on the foreground object even if the object was at the edge of the elipse ( I no longer have a 1D MkII to test the same thing). When performing the same test on the 1D MK III the camera always focused on the grass covered by the CP.
Phillip.
"When performing the same test on the 1D MK III the camera always focused on the grass covered by the CP."
Isn't this what we want, though? I want the camera to focus on the thing I point it at, not to guess, or to focus on the thing that may be closest to the camera.
I know that when I'm photographing badgers with my 1Ds2 and 300/f2.8 using 45-point auto focus selection, what I get in focus is whatever the centre focus point is over. It doesn't guess that I want the eye to be in focus and therefore defer to the focus points at the right or left edge of the elipse.
I'm not trying to start an argument here, just trying to generate some light to better illuminate the Mk3 AF issues.
"Isn't this what we want, though? I want the camera to focus on the thing I point it at, not to guess, or to focus on the thing that may be closest to the camera."
I agree and I do not normally use auto selection anyway. I am merely relating what my 1DS2 did on my test and it is repeatable. It may be there was insufficient contrast for the 1ds2 to focus on the grass area covered by the CP whereas the extra sensitivity of the MKIII was able to focus.