Canon 1D Mark III Review – AF Part 2, AI Servo mode, Bright Sunlight

OK so this is the area that has received the most attention.
It’s surprised many as well lit environments are usually the easiest for auto
focus. With the 1D however, initial reports showed that this was THE area that
was affecting a very large number of people. I’m going to a different approach
with this section and start with my summary first. That is:

• The Submirror fix does stop AI servo issues in bright
  sunlight from getting worse than compared to usage at lower temperatures.
• Despite the fix, the Canon 1D Mark III does continue to
  have problems sometimes when using AI Servo in bright sunlight

So the problems – what exactly are they? Well, essentially,
it’s more out of focus frames than you would expect. In a sequence of say 10
images on a subject that should not be that difficult to track, I’ll find
somewhere between zero and four that are grossly out of focus. The rest will
either be superbly accurate or acceptable. And this happens consistently.
Sometimes it’s 100% of the frames that are perfect, and sometimes it’s more
like 60%, I’ve yet to find it lower than that.

We all know that pictures are often worth more than a few words,
so I’ll cover this section by taking you through a number of sequences /
examples. Note that all the sequences below have the same criteria:

•  Shot with the 1D Mark III and a 200mm f/2.8 prime
• The lens / camera combination has been microadjusted
• No custom settings to the AF system were made (e.g. AF
  expansion, etc).

Sequence 1 – Illustration of AI servo issues in bright sunlight
This sequence was actually shot before the submirror fix but I don’t believe it to
be temperature related as I have other sequences which exhibit similar
properties. It’s shot with the 200mm wide open (f/2.8) using the center focus
point.

This marathon runner was coming towards me at regular
jogging speed. This was frame 2 in a sequence of 10 in a 1s burst. You can see
from the 100% crop below that it’s exceptionally sharp.

Moving targets basically don’t get sharper than this.

 

However, on the very next frame, I saw the following as a
100% crop.

 

I would call this frame grossly out of focus. I could
sharpen it in a larger version but quite honestly, I would expect it to be very
usable. The next frame after this was similar (OOF) and yet the next four
frames were exceptionally sharp like the first one. 

The target was not doing anything crazy to warrant the AF
system being fooled – this is an example of bright sunlight causing the AF
problem.

Note, out of the sequence of 10 images, 4 were grossly out
of focus and 6 were tack sharp. I’ve yet to find a worse example.

Sequence 2 – Illustration of no AI servo issues in bright sunlight

To show that the AI servo issue does not plague all the
time, take another example. Again, shot in bright sunlight at the same
marathon, I have sequences which exhibit NONE of the AF issues described above.
Below is a sequence of 100% crops which constitutes a 20 frame burst. Again,
it’s shot with the 200mm wide open (f/2.8) using the center focus point. The
first image in the sequence is this:

the last image in the sequence was this:

 

From this sequence, I consider 18 or 19 to be perfectly tack
sharp and 1 or 2 to be acceptably sharp. None are grossly out of focus.

This is the inconsistency that will sometimes fool users
into thinking they have an all-the-time good or bad camera.

For me, I can confidently say that in bright sunlight,
shooting wide open with a telephoto lens will gives me worst case 60% tack
sharp images and 100% tack sharp images.

However, can it be improved? That will now be the topic of
discussion.

Sequence 3 – Illustration of improved AI servo performance in bright sunlight when stopping down

For some people, the above results may already be enough.
For me at least, it is evidence to show that the camera is capable of producing
excellent results under bright conditions – it’s more a matter of the degree of
excellence that is in question. On the other hand, there is room for
improvements and something I have found to give me better results, is to stop
down (decrease aperture) to increase the chances of infocus shots.

If there is a fortunate thing about AI servo issues in
bright sunlight, it’s that there are options on how to deal with it. Because
you can shoot in low ISO to begin with (and with high shutter speed), at least
the option of stopping down is there – this would not be the case if AI Servo
issues existed in low light. So by shooting a moving subject at say f/5.6, how
do things fare then?

For this test, I took a faster, more eratic and harder to
shoot subject. This time, my dog, a 70lb Rottweiler running towards me at full
speed. The sequence was shot at a beach later in the afternoon when the sun was
both warm and bright. The sequence consisted of 24 images. The first and last
images are shown below. This time, I let the camera choose the focus point.

First image in sequence:

Last image in sequence:

So how were the results? Well, despite an erratic moving subject which is prone to user error (unlike tracking marathon runners), despite the fact that my dog is a smaller target and is covering twice as much area in half the time, the results were dramatically improved. Below is a series of the 100% crops of all 24 images. Remember, this is using the auto selection point method, which doesn’t know that a dog’s face is the subject – just that the dog itself is the target. So the focus point may move around slightly on within the subject itself – the key is that it’s on the subject (if this were a person whose frame is flat rather than long, this issue wouldnt’ be a problem). As a result, so included a slightly larger area for the crop and you may have to look within the frame to find the focus point. All images in the sequence are center crops of around 950 x 640 pixels (100% of resolution), with the exception of sequence numbers 3 and 6 which are I cropped slightly lower than center as the AF system’s selection point choose too low a point in the frame for it to be effectively covered in the crop size I chose. Note, each image has had a minor levels correction to improve the lack of contrast caused by the light and also a minor unsharp mask.

 

The this sequence was quite typical for the performance when stopping down so I consider these images to be representative. You can see the first couple of images, the sharpness is “OK” but not exceptionally. However, it’s certainly not grossly out of focus and I consider the resulting images to be usable. After the first couple of frames however, the camera gets an effective focus lock and is able to track my dog through the sand with excellent clarity. I caution also to remind that the camera is on automatic AF selection here. So sometimes it’s going to select the collar around his neck as the focus point rather than his face This is most noticeable in image IMG_8173 where the face looks out of focus but the collar is sharp. This is the AF system simply doing its job with a subject that is leaping in and out of the central AF points. Towards the end of the sequence, when the rate of movement is at its greatest (relatively speaking from the camera’s point of view) every frame looks tack sharp.

How does the camera do in this instance? Is there still room for improvement? I would say again, yes but it’s clear to me that in cases of bright sunlight, when it’s possible to stop down you massively increase your chances for a greater number of in focus shots. As I’ve said, I’ve found this sequence to be highly representative of my experience with the camera but just to make sure, lets take another example. By the way, I’ve included some final, processed shots from this test scene / location under the Image Quality section (where I actually try to produce a good image rather than just a well focused one).

Sample 4 – Second illustration of reduced AI servo issues in bright sunlight when stopping down

My second example of stopping down to improve AI servo performance was at the greyhound races in the UK. Here the temperature was very cold but the light was very bright. Here, still shooting with the 200mm f/2.8, I was trying to get multiple subjects in focus and so stopped down further, to f/8. If you’ve never seen greyhound racing before (and I’ll put my own moral objections of it as a sport to the side for one second), it’s highly impressive. The dogs, although small in frame and weight, run past at speeds of up to 45 mph (72 km/h).  They cover 400m on a curved track in under 25s.

For these shots, I was using the center focus point. If shooting purely for the quality of the image itself, I would have turned on the AF assist points in one of the custom settings but here I was attempting to challenge the ability the AF system under the difficult condition of a extremely difficult to track subject.

Below is a sample which was representative of the cameras performance in these conditions. Because the dogs move faster, the sequence is shorter – just 9 images. The first and last image are shown below. First image in the sequence:

Last image in the sequence

What follows is the ten 100% crops from the sequence.

If you study the images closely I encourage you to look at the center area. This is where the camera is being instructed to focus. You can see in image one that as I start the sequence I completely miss the subjects and get the erupting track behind them. Despite that, the image is sharp. The second image, I start to correct the shot but I’m still off target – but again, still sharp. For the next five or so shots I correctly track the dog I was intended to focus on and to my eyes at least, at 100%, the images are shown to be very sharp. If they could be sharper I would say it could have been through an even higher shutter speed than I employed perhaps but again, I’m of the opinion this is as good as its going to realistically get. With a little unsharp mask these images look tack sharp when printed.

AI Servo mode, Conclusion

The debate of the 1D Mark III’s AI Servo mode under bright / sunny conditions has caused intense debate and I don’t expect my own conclusions to put the subject to rest. However I have a clear view on how my camera performs in my hands, and to summarise, I get sharp shots when shooting wide open for 60 – 100% of the frames shot with 60% being more likely than 100%. To put it another way, I’d probably expect to find 1 or 2 (possibly 3) frames to be grossly out of focus for every ten when shooting in those conditions. The other conclusion I drew from these weeks of shooting was that when stopping down the likelihood of sharp frames increases significantly.

Now, to put the potential 60% in a bit more perspective, the other thing worth mentioning when comparing this performance to the previous standard bearer, the 1D Mark IIn is that you’re also getting 2 extra frames per second (10fps vs 8fps). So even if the Mk IIn was shooting at a perfect 100% all of the time, the worst case scenario I experienced with the Mark III was 20% fewer frames in focus per second. In a best case, it would be 20% extra frames. On average, it might even get close especially when the aperture is decreased a stop.

So what does this mean for me? Well, I can’t really judge the camera’s AF system on the performance of this set of scenarios alone. For example, I’d happily take "not quite as good" performance under bright sunlight when shooting wide open if I was able to get better performance indoors or in low light. Stopping down is an option, even a short term resolution, when in bright sunlight but that’s rarely the case when the light gets low. Quite honestly, for the amount of AI Servo work I do in bright sunlight (less than 5% of my total photography), I actually found the 1D’s AF performance to be acceptable.

Now does that mean things can’t be improved? Of course they can. Canon can most definitely still do better and I among many others hope for another firmware upgrade. But when I did my own analysis the sky was most definitely not falling.

I’ll provide a caveat at this point that although my camera performed regularly within the above parameters in my hands, I can make no broad generalizations about the rest of the 1D’s on the market as I simply have not used a large enough sample. However, it would not be surprising for me to find others have similar experiences and from what I read from qualitative reports, it also appears to be largely the case.