Lawrence Ripsher

Autofocus

This post will focus on the important, often misunderstood and almost always badly reviewed aspect of Autofocus. The reason I say it’s usually badly reviewed is because at this time of writing (and very likely for quite some time), there is no easy, accurate or realistic way to test or rate Autofocus from a scientific perspective. Rob Galbraith probably came the closest with his 1D Mark III work, and I followed in those footsteps by shooting thousands of frames for my review of the same camera. However, I feel we both only really scratched the surface on the subject, and certainly were not able to come up with an empirical / foolproof measurement for AF performance in all situations.

The lack of a good review framework doesn’t stop the reviews however. Most of the time, AF often gets discussed in two ways… First, the number of AF points, whether each AF point is a cross hair, etc. This is interesting and I cover it below. The second thing that often gets a mention is the time taken to acquire a focus lock in a lab or studio. This gives an indicator but it is truly representative only if you want to measure the speed of locking AF onto lens charts – it says nothing for the cameras ability to hold focus on a moving subject, track a subject through a crowd, distinguish a foreground from a background and so on. As I said, I’ve yet to see a good scientific approach to this kind of testing…

So I’m going to try to tackle this area as I do with most photographic subjects – with a mostly qualitative assessment based on real world situations / experience, with a few relevant numbers and examples sprinkled on top.

AF points

So first of all, lets talk about the AF points. Criticism has been occasionally pointed at Canon due to the relatively low number of AF points it puts in its entry level, prosumer and mid range (e.g. 5D) cameras. The 5D Mark II ships with 9 AF points and 6 hidden assist points which can be activated through the menu system. The Nikon’s from the D300 up, in contrast, have 51 AF points which on paper is very impressive. I’ve heard very good things about the new Nikon AF systems but unfortunately I can’t comment from personal use as I’ve never used them for sports. I will say the 1D Mark III had the most heavily scrutinized AF system of all time and I rate it as Canon’s best ever. In fact, I’ve never seen any compelling evidence that to suggest that any other camera in the world is its peer for all round sports photography. I’d love to see the same scrutiny applied to Nikon’s D3 but anyway, that’s off topic.

So other thing to bear in mind about the 5D Mark II center AF point is that you need to use f/2.8 lenses and faster to get the best out of it. The truth is that, with the exception of perhaps some long zooms, the best lenses for the 5D are all f/2.8 and better anyway as anything slower is often not an L and will suffer with softer images, etc.

The center AF point is the cross sensor. This is the fastest, most accurate and the one I use 95%+ of the time. Some photographers do not like the 5D’s arrangement of AF points which cluster near the center of the image. I’m unconcerned with this for three reasons:

- I’d rather use the most sensitive AF sensor in the vast majority of cases than a sensor at the edge which might be closer to the subject but slower to acquire focus

- I find (and recommend) the technique of focus lock –> recompose to be the most effective way of getting good photographs for the vast majority of cases. I set my camera up specifically for this purpose. Even with a much more dispersed arrangement of AF points, I still tend to use the center focus point the majority of the time (I do this on my 1D Mark III also which has 19 cross sensors and 26 assist)

- Due to the huge resolution of the 5D Mark II the need to compose accurately in camera is beginning to become less important (which I wrote about here). Purists and people who shoot thousands of images per week may not like this but it’s the reality and learning how to use these advantages yields better pictures

So, in terms of AF points, sure I’d like to see a few more cross sensors but even if they existed, I am not sure I would use them very much – except for some sports in AI Servo mode. But AI servo mode is much more than just the number of AF points and we get to that later.

Focus acquiring – still subjects

This is what most people need – a camera that locks onto a subject quickly and accurately. Looking back in time a bit, with the original 5D, I found its Single Shot AF mode to be excellent. I used the 5D as my primary camera during my shooting of the hit TV show, the Contender Asia, and I never had any trouble. Of course, that is not to say it was the best available – the 1D Mark series have always been noticeably faster at focus acquisition. However, this doesn’t detract from the 5D’s AF system and I recommend taking a look at a sampling of my Contender Asia images – this will give an idea of what was possible with the old 5D, its AF system and 3 fps. If 5D owners complain about not being able to focus quickly on subjects, then I suggest a problem with the photographer, not the tool.

So, lets forward wind now to the new 5D. Somewhat unsurprisingly, the new version for focus acquisition is much of the same really. That is, using the center focus point, it is fast, accurate and a pleasure to use in almost all situations. It may be slightly improved over the 5D but not so much that I can tell in real world situations. It is still obviously slower than the 1D series but this does not cause a major issue until you get into very demanding situations which most people never encounter (by which point a 1D would probably be the better choice anyway).

It’s hard to illustrate focus acquisition in samples as the subjects are often static and when subjects are moving, you would typically be in AI Servo mode anyway. My Contender images are the best examples of focus acquisition with a 5D like system but for the new body I do have two samples from the past week though where fast focus acquisition was needed. The first is a picture of my dog (which I have posted before).

To take a shot like this, you need to first have a dog that likes to jump and mash snowballs in mid air :). Once you have one of those, make a snowball, put it in your left hand while holding the camera to your eye in your right hand. Then lob the snowball up in the air and squeeze off a couple of shots as your maniac dog attempts to slam dunk it. It’s impossible to prefocus as you don’t know where the “reception” will take place so the camera has to quickly focus lock, as it did here. I have many examples like this – and when I got it right, the camera was right there with me.

The second example was when I was out in the snow and noticed some birds flying overhead. Heavy snow was  falling and I quickly aimed upwards, snapped off a single shot and then forgot about the image till I looked on my camera later (this is cropped btw – I had a short 24-70mm lens on at the time). Despite the falling snow the AF system was able to easily pick out the contrast of the birds and produce a perfectly focused image.

Unfortunately I don’t have more examples as I’m trying to keep these review entries from becoming too much work – and besides as I say, they’re difficult to simulate (particularly in this weather). You’ll just have to take my word on it for now but if more opportunities crop up later, I’ll add the pictures here.

Oh, total side point – the Canon 5D Mark II has a microadjustment feature which I found invaluable on the 1D Mark III. I haven’t had to use it with any of my lenses yet on the 5D however.

AI Servo

So, the 5D Mark II can quickly focus on static subjects and also lock onto moving subjects. Great. How about subjects that keep moving? Well, that’s where AI Servo comes in.

I’m going to make this section easy and start with a conclusion. The 5D Mark II is not, and should not be expected to be a pro sports camera where AI servo is required. From what I’ve seen so far, the camera produces world class image quality, but it does not have Canon’s 1D AF system and will not be able to focus track as fast or as accurately. If you shoot sports where you can use Single Shot focus mode, then I can already tell you the 5D2 will do that superbly. If you need AI Servo in challenging situations, think again.

OK now we’ve got that out of the way, lets turn to a practical example of what this means in the real world. Last year I shot the Federer vs Sampras exhibition match. I sat in the press box next to the Reuters, Associated Press photographers, etc with a great view. Unfortunately I didn’t have a 1D Mk 3 with me (as it was pre AF problem fix at the time) so I shot the match with a 40D and fast telephoto primes. Through a combination of Single Shot focus and AI Servo I was able to get some great photos from the match, but the hit rate was low. The reason for that was because the AI Servo on the 40D was unable to track a fast moving player in relatively low light – it simply couldn’t do it. So instead I adapted my style and focused on shots when the player was positioned waiting for the ball, leaving me to focus on timing. This worked fine but it restricted me to a certain style.

Fast forward a few months when I was in the same position shooting Sharapova at an exhibition match. Here I had my fixed 1D Mark III and saw a hit rate many times better than at the Federer match while shooting the entire match in AI servo mode. You can’t necessarily tell from my samples from that match, but I can tell you, the difference was night and day.

The takeaway from all that? Well the 5D Mark II from my experience so far seems to have similar AI Servo capabilities to the 40D. Or to put it another way, don’t take the 5D Mark II to a professional tennis match if you want to shoot AI Servo all day.

Now, that is not to say that the 5D Mark II can’t AI servo at all. It can and thanks to my trusty action subject, I have a sequence to share.

AI Servo Sequence

To test AI Servo properly, you have to spend weeks doing it. I did that for my Canon 1D Mark III review but am not going to repeat it here – as I already know the 5D Mark II won’t be my primary camera for this kind of work. What I was interested in however were some real world examples of whether it could track a fast moving subject at all and some anecdotal evidence of what kind of accuracy I could achieve.

To do this, I got my dog to run at me at full(ish) speed so I could test the AI servo on a subject moving closer towards the camera. I ran this simulation about 6 times – enough for about a hundred images.

These were all shot with a Canon 200mm L f/2.8 (a superb and underrated piece of glass – better than the 70-200L in some respects) with camera settings of ISO 1600, f/2.8 and 1/800s shutter speed. You can see from the ISO that the light wasn’t very high – this was early in the morning. The images were in JPG. Shooting in AI Servo takes a fair bit of skill (as well as the right setup on the camera) – it is something that must be practiced to obtain good results.

With the 5D Mark II, I found the accuracy rating to be variable – but somewhere in the 50% – 75% range. On some bursts, I saw up to 3 images out of focus as the camera was trying to catch up. Other times I had a series of 7 or 8 sharp images in a row which was great. Overall, the above percentage seemed to be ballpark. Remember, this is only one set of conditions so take it with a big pinch of salt.

To compare with the 1D in this kind of situation (at f/2.8 on the same lens), I regularly achieve an accuracy rating of 90% and above. That is, more than 90% of frames are acceptably sharp. You can see evidence of that in my 1D review here.

So coming back to the 5D Mark II, lets talk real world again. What does a 50% – 75% accuracy rating give you? Well, here are a random sampling of the good images.

 

 

Below is an example 100% crop of one of the sharp images.

For parents, dog owners, occasional sports shooters who want to take advantage of this feature – I think it’s great.

The images are not ultra sharp – losing a bit of accuracy compared to the 1D series, but it is still good and would result in a very nice print up to 20” or so. Also remember that this is AI Servo mode at ISO 1600 with the subject moving at fast speed towards the camera. Shooting at lower ISO, in brighter conditions, at a smaller aperture would all have yielded a sharper image.

Overall, for a prosumer AF system, I think this performed quite well in AI Servo mode. It’s not on par with the 1D which shoots at 2.5x the speed yet still achieves a more accuracy rate, but then the 1D (mark III) can’t shoot at 21MP.

 

Conclusion

From my experience thus far, the 5D Mark II looks to have a good AF system which performs well at the tasks it was expected to be used for. I do not find the number of AF points, nor their arrangement an issue for real photography. When in single shot mode (and particularly when using the center focus point), focus lock is fast, accurate and a pleasure to use. A lot of sports can (and should) be shot with Single Shot mode with the right skills and understanding of the subjects (i.e. anticipation). So with the 5D Mark II’s excellent resolving power and high ISO performance, I do not rule out the 5D Mark II as a sports camera by any means. In some situations it would even be superior to the 1D Mark III. However, that should only be said with the very strong caveat that the AI Servo mode on the 5D Mark II is nowhere near the 1D Mark III. It is decent / usable but does not have the same degree of accuracy as Canon’s top of the line bodies. So it depends on the subject really. As for my area of expertise in sports photography – i.e. boxing / fight images – I can only wish I had this on the set of The Contender Season 4 during our filming in Oct 08, as this would have undoubtedly have been my primary camera.

Note – as I said above, AF is one of the most difficult areas to review in photography so these are my conclusions thus far. They could change as I encounter more situations with the camera and if they do, I’ll write another installment to Autofocus.