Source of the 5D Mark II AF Angst
There are a lot of people around the Internet that aren’t happy with the specifications of the autofocus on the new Canon EOS 5D Mark II. There are also a lot of people that have chimed in saying that it’s a bit too early to pass judgement. After all, the camera hasn’t even shipped yet. And there are those who question why people care so much about this topic. Having been a critical voice on this front with my previous post about the 5D Mark II, I’ve certainly had the opportunity to engage in more than my fair share of discussions in the past week on the topic. Though these discussions, I’ve had the chance to think—maybe even overthink—what’s behind the criticism.
To start, isn’t it a bit too early to pass judgement? Especially considering the camera looks to have absolutely stellar image quality, loads of resolution, and the ability to make HD movies that look amazing, such as Vincent Laforet’s Reverie? Isn’t it unfair to level criticism towards the camera at this point in time? Maybe. Probably. After all, the only real way to know what a camera can do is to use it. On the other hand, we do have quite a bit of information already at our disposal from both Canon sources as well as those that have their hands on pre-release cameras.
Here’s what we know for sure. The new 5D Mark II has the same 9-point autofocus layout as the previous generation with the same specifications of sensitivity for the points. Only the center focus point is a high-precision cross-type sensor. Like the first generation camera, the new one has an additional six focus points ganged up around the center focus point to help tracking in servo mode.
What’s less clear is what impact any improvements to the RISC chip that runs the autofocus system might have on performance. It seems that this processor is unchanged, but it might run faster in the new camera. Also unclear is what aid the new DiG!C IV chip might be able to provide. And, we have no information on any possible improvements to those hidden six sensors that aid servo tracking, other than that two of them are cross-type sensors. Because of these points, it is indeed too early to pass judgement on the ultimate autofocus performance of the new camera. Certainly, the speed and precision of the system are unknown at this point. Maybe it’ll be better, at least evolutionary so.
Regardless of any performance improvements that we may see, after shooting in the neighborhood of a hundred thousand frames on my 5D, I can tell you that ultimate speed of the focus system isn’t the big problem. Certainly not for a camera that isn’t targeted for sports shooters. No, the real issues with the autofocus system on the original 5D are found in the layout and sensitivity of the focus points. And, this is something that I think is more than fair to criticize, even before the camera’s release.
The focus point layout is the same exact layout in absolute size as that on the Canon 20D, a 1.6 crop frame camera. This layout has continued through the successors to the 20D, including the 30D, 40D, and 50D. On a crop frame, the layout of the focus points covered a pretty good amount of the viewfinder. On the full frame 5D, the 9 focus points are relatively bunched in towards the center of the frame, leaving most of the frame empty. Here’s an illustration of this comparison (with thanks to Stef for pointing to these viewfinder images in the comments of my previous post about the 5D Mark II).
I’ve scaled the viewfinders to illustrate their approximate relative size. As you can see, the result of this arrangement on the 5D is that you don’t have an autofocus sensor in many places where you might want one. In fact, there’s gobs of empty space in the viewfinder surrounding the small cluster of focus points. If you want to use the rule of thirds, sorry, you’re pretty much out of luck. The diamond pattern of the 20-50D doesn’t help much, but the relatively smaller pattern on the 5D really doesn’t help at all. Your only option is to focus-and-recompose. This works well enough when you are shooting at f/11 or f/16 and if you’re being careful. At the wide apertures of f/1.2 through f/4—ranges where you get the most out of prime and L lenses—the focus-and-recompose technique is a recipe for sad results.
The sensitivity, or lack thereof, of the non-center points only reinforces the problem. Because the outer 8 points don’t work very well in low light conditions, the camera might as well only have a center focus point when you are working in the dimly illuminated conditions where you can really get the most out of the sensor and Canon’s lenses.
Even so, as the first mainstream full-frame digital SLR, the image quality of the original 5D made it easy to overlook, or at least forgive, the faults of its autofocus system layout. Heck, I purchased my 5D when it first came out fully aware of these issues and accepted them for what they were. After all, it was understandable that Canon would leverage as much as it could from the 20D (and 30D which was then in development) to bring the camera to market at its price point. And, there are workarounds, such as using the ST-E2 Speedlight Transmitter to project an AF assist pattern. Besides, they could always fix it in the next model, right?
Right?
Well, three years later, we have the next model and our answer. At least enough of an answer. Many people, including myself, would have loved to have seen a 45-point autofocus system of the same caliber as the EOS 1D series cameras. After all, this would give us a close digital version of the Canon EOS 3, a reasonably sized reasonably rugged camera with a pro-autofocus system that was released in 1998. Ten years ago.
Even if Canon’s current market segmentation strategy precludes putting in a pro-level autofocus system into the new model, it is still perfectly reasonable to expect that at least some improvement to the sensor layout would be made. Maybe even some of those outer focus points could have been high-precision cross-type sensors so that those that wanted to compose using the rule-of-thirds in dark environments could have some help from the camera. Maybe the layout could have been improved with points near the rule of thirds area. Heck, maybe make all of the focus points high-precision cross-type like they did on the 50D.
Instead, the new 5D Mark II has none of that. Any of the improvements that we might see from the autofocus system in the new camera don’t help with the core problem that the layout is too small and only the center focus point is usable in poorly lit environments. It’s really too bad, because the rest of the camera is pretty sweet. Yah, you’ll be lens-limited with most of your lenses and will have to update your computer and storage to handle twice as much data as you’ve shot up to this point. But, really being lens-limited isn’t bad and as long as you don’t shoot thousands of frames at a time, the storage requirements can be overcome. The ultimate image quality you’ll get will make it worth dealing with these other issues.
It’s precisely because of ultimate image quality that this camera will probably be a success for Canon. After all, if you’re looking for resolution, you’ve got it. Lower noise at high ISO levels, check. Movies, oh yah. If you’re invested in Canon and want to shoot great landscapes, studio portraits, or the like, then you’ll probably want to evaluate the camera for your needs. On the other hand, if you shoot events or action, you’ll want to think about this a bit harder. And, if you’re not invested heavily in Canon lenses, well, there are now other full-frame alternatives to consider. Unlike three years ago, you can find other cameras at around the same price point that make a different set of trade-offs that might agree more with you.
Finally, for those of you that will inevitably wonder why on earth you’d want more than a center autofocus point, well... if it works for you and you are producing the images you want to be producing, great! Otherwise, you should consider giving some of those other focus points a try. You might find that you like having them at your disposal.
Related Posts:
30 Comments
Just curious, if one cropped in on the 5D (or Mark II) viewfinder, would the area be the same in terms of megapixels, or close? Wondering if Canon just expected folks to crop down from 21 Mpixels.
It is positive to hear the "speed" is not your biggest complaint with the 5D. The frame rate is a little slow for some folks, but very curious to see how the Mark II will perform for sports (for a non-professional shooter). The cluster of focus points in the middle would be a nice benefit for that.
Reply to this comment
Well, there are indeed pixels to burn, so cropping is definitely on the table when needed. But for those that really want to get every big out of their images, they’ll want to try to make it work without cropping. Luckily, most of those people are landscape types and will have the time to do what they need to do to get a sharp image.
The overall frame rate of the 5D Mark II is still a bit slow for my taste. For me, 5fps is about where I want to be for events. The new camera’s 3.9fps should be a bit better than the original 5D for non-pro sports shooting. And it will be interesting to see how it handles in servo mode off the center point.
Reply to this comment
Since the 5dmk2 is not due for release until late November, is it possible that Canon could still change the 9-point autofocus layout to something more along the line of the 21 point or, better yet, a 45-point autofocus system with 19 cross-type sensors; OR later produce a 5dmk2n with a better auto focus sensor?
Odd, but after reading the first Rob Galbraith review of the 5dmk2, I emailed him asking if he thought that Canon's failure to upgrade the autofocus on the 5dmk2 would prove to be Canon's downfall for it. It was not at all surprising that his response was "no. used for stills."
He is correct, of course, regarding the main use of the 5dmk2 is for "stills" photography, but I was taken aback at his response or lack thereof.
That he felt completely opposite from you, Mr. Duncan, as well as many others, will have me move to a love-hate relationship with this camera.
The other item I could not understand was why Canon did not increase the speed (fps) with the new Digic 4 processor to at least 4.9/5 or even 6 fps. I can't imagine that an increase from 3 to 3.9 fps would make that much of a difference; but Canon can at least boast they increased it by 33.3 percent. And, if Nikon can make use of 2 batteries to boost the fps to 8, can't Canon be more aggressive if not to equal, but at least come closer to Nikon's generous 8 fps?
I was on the cusp of switching to the Nikon 700, but with the advent of the 5dmk2, 9-point autofocus and 3.9 fps be dam..." I'll take Canon's technology over that of Nikon any day.
Simply put, if Vincent Laforet's beautiful Reverie shows us that low-light photography can be achieved with 1080p HD, then something must be different in the "backstage" end of the autofocus system. Let's, for all our sakes and all Canon enthusiasts, hope so.
Reply to this comment
It'd be highly doubtful that Canon would change the AF at this point.
I don’t think the camera will be a bomb for Canon. I think it probably will be a success. And Rob Galbraith is right on in that stills are where this camera will be best used and that the AF system will not doom the camera's success. But whether or not its a success is independent of the fact that they totally overlooked doing anything about something they should have.
On frame rate, the computing pipeline and battery power should not be the restricting factor. Two guesses as to why the frame rate didn't go up significantly: One, a faster mirror mechanism would have been more expensive. Two, it might be a market segmentation issue to keep the 5D Mark II away from the capabilities of the 1Ds Mark III.
From watching the behind the scenes video, Vincent was using Live View to focus at least some of his shots. In addition, the autofocus system in video mode is using contrast detection off the sensor, not the 9 point phase detection system. You may be onto something there tho in that using the contrast detection system in Live View mode is a way to focus on things outside of the center of the frame. It’ll be slower than the phase detection sensors, but...
Reply to this comment
Why is it bad to focus-and-recompose at wide apertures? Are you talking about moving subjects?
Reply to this comment
Do not see much contentious, the autofocus system looks outdated and silly. I have ordered the camera because of all the other goodies. But I am not looking forward to the autofocus. Opportunity missed for Canon, it is going to expose them to a lot of competition .... and there maybe a lot more at stake than the value of the 1D market segment.
Reply to this comment
I can only imagine it's a tough marketing decision for Canon to set the specs for the 5D. It is a bit of an odd camera - it's image quality is essentially the same as the top-of-the-line cameras, but the price point is less than half. If they upped the specs to be much closer to that of the 1Ds, there's really only the better environmental sealing of the 1Ds to differentiate, and I don't think most people would be willing to spend double the price for just that (well, unless you regularly shoot on a boat or in dust storms).
With the slow-ish frame rate, and so-so auto focus, the 5D doesn't make for the best "action" camera on the market. Fast moving sports events, and children hopped-up on sugar at a birthday party would probably be better off shot with different camera. Although the 5D can certainly take these photos as well, the 1D series would have a higher success rate for the reasons you point out.
It does seem like a shame that, with most likely a small incremental cost, Canon could indeed improve the AF sensors a little. Specifically, I completely agree that the outer AF points should be as sensitive as the centre point. I'd think that would be a reasonable compromise for Canon to make - the 1D series would still have superior AF performance to help differentiate that model line, but the 5D wouldn't be stuck with "focus/recompose" in low light. Personally, that technique doesn't bother me that much, as I usually shoot that way - old habits die hard - but also, I'm not usually shooting fast moving subjects in low light, so I have the luxury of taking a bit more time to focus and compose. Well, that and I like using a shallow DOF quite a bit, and that needs a single AF point most of the time anyway - using multiple AF points leads to amusingly random results. That said, I do wish all the AF points were as sensitive as the centre point - less re-composing if I could pick an outer point in some circumstances.
Thank you for writing this informative article. I've read quite a few comments from people unhappy with the underwhelming AF on the 5Dmk2, but most of these people aren't explaining why they are unhappy with it - you did explain why, and your reasons make sense to me.
Myself, I'm satisfied with the AF on my 5D, so it wasn't high on my wish list for the 5Dmk2. Actually, my biggest wish for the 5Dmk2 was for more dynamic range, and from what I've been reading, it seems that Canon has improved that somewhat - I'll have to wait for a bit and see if it's a dramatic enough improvement to warrant an upgrade. The rest of the improvements (higher resolution, video, etc.) is gravy - but what a tasty gravy...
Reply to this comment
the autofocus system is definitely not as good as nikon's. but what i'm really curious about is the frame rate.
according to imaging resource http://www.imaging-resource.com/PRODS/D300/D300A.HTM , frame rate on the nikon d300 (i know, i'm comparing a different camera) "slows down from 6.13 frames per second in 12-bit mode to 2.66 seconds in 14-bit mode".
so the quoted frame rate for the 5d-m2 is just under 4 per second. is that 12-bit or 14-bit? if 12-bit, will 14-bit really slow it down like the nikon d300?
anyone know?
Reply to this comment
C - I don't believe you can change the bit depth on the 5DMII. It's always recording at 14-bit.
I wrote a bit of a response here. I'd have been happy with five AF points, positioned better, and all cross-type.
Reply to this comment
Dan, to answer the question of how focus and recompose can go wrong, one which has come up more than a few times, I wrote this post: http://duncandavidson.com/2008/09/focus-and-recompose-exposed.html.
Reply to this comment
C, I’ve heard of the D300 frame rate slowing down when using higher bit depths. However, the D3 and D700 apparently don’t have this issue. Maybe somebody who is more familiar with what Nikon has been up to all this time can shed some light. As far as the 5D goes, I can’t imagine that they would shoot slower at full bit depth.
Reply to this comment
I have to say, I've never used anything except for the center focus point on my 20D in the last 2.5 years that I've owned it. Never felt the need to change, as I can focus/recompose or use some manual focus tweaks. I guess it depends what you shoot. It seems that the 5D MkII is clearly not aimed at action shooters and people looking for fast AF with larger frame coverage should look elsewhere.
Reply to this comment
Horses for Courses.
If focus flexibility is important remember that Nikon offer 15 cross type sensors as part of a 51 sensor system that is similar if not identical to the cheaper D300 and top of the line D3
(It seems Nikon are more about giving users what they need rather than keeping back as much as they can get away with.)
The clustering is tighter on full frame Nikon cameras much as it is with Canon between their different chip sizes.
IMHO the only things the new 5d can offer to 'beat' the D300 and D700 are;
Video. See my first line about horses.
Full Frame. Debatable advantage since full frame both giveth and taketh away
Megapixels. Really the only big advantage for many users. Admittedly a pretty big advantage, but you'd better have a need for it since this file size and bit depth requires some really gutsy computer specs.
For me the 5d is the first Canon I would consider purely for the video feature, but the verdict is out until I test it and see whether the jelly wobble concern is real and whether editing will be possible with final cut express (I believe it is used for the demo clip)
I may even use it photographically since the IQ will go a long way toward compensating for the limited specs
For those of you who need a camera that outperforms Nikon look to the 1D series since they do have many of the features required to compete with the D300, D700, and D3 (and for that matter, the new Sony A900)
Reply to this comment
When the D700/D3 are set to 14-bit RAW, there's no performance penalty. The 14-bit RAWs are larger, which in theory should fill up the buffer faster, but tests don't seem to show the buffer clearing any slower in 14-bit than in 12-bit mode.
The D3 is more responsive (it's got 2 EXPEED processors where the D700 has just one, and the battery is more powerful than the D700's), but for the most part they're equal. Canon for whatever reason seem to go out of their way to cripple the 5D Mk II vs the 1D series.
I was using a mixture of Olympus and Pentax gear, all of which I loved. But low-light work was a problem. Olympus's E-3 AF system is as fast as the pro Nikon and Canon bodies, but they don't make any fast lenses. Pentax have brilliant lenses but consumer-grade AF.
Low-light photography was something I did a lot of - I'd keep trying when everyone else had put away their cameras. Sometimes I'd get a good shot. But when people then hire you to do that work, you need to always get a good shot, not just sometimes.
So I waited for the 5D Mk II announcement, and when they pretty much left the AF system alone, I sold off all my gear and got a D700.
Low-light performance is brilliant. I can go as high as ISO 9000 for colour, 12800 for B&W. And the AF is as fast as I could want in conditions where my Pentax didn't stand a chance. (Oly's fastest wide angle is f2.8.)
Those of you upgrading to the 5D MK II are in for a low-light treat. High ISOs look to be as good as or slightly better than the D3/D700. And you've got considerably more resolution.
But is frustrating that Canon have a camera that could be the king of low light except they crippled the AF. (And still no MLU button.)
Reply to this comment
Note that the 5DmkII does have an important additional focusing option vs. the original 5D: Contrast detection.
I'm pretty sure it'll have lousy speed. We'll see if it can handle low light. It certainly should have wider coverage.
Reply to this comment
Small typo:
"No, the real issues with the autofocus system on the original 5D found in the layout and sensitivity of the focus points." You probably meant are found?
Reply to this comment
D3's is a better camera in terms of focussing and build quality just a little piglet to walk around with - I think it will be good for a few years yet too.
Who wants to shoot movies on a 24MP still camera anyways !!!?
Reply to this comment
Luke, thanks for chiming in with information about the D3/D700 in 14-bit RAW mode. And, you're absolutely right that one reason some of us care so much about this point is that need to get the shot because one is getting paid for it. Missing a shot as an amateur is disappointing. Missing it as a pro can cost.
Congrats on your new kit. :) I've been really impressed with the D3 and D700 after playing with them for a bit myself.
Reply to this comment
David, you're absolutely right in pointing out the contrast detection AF in Live View mode. I'm sure as well it'll be slow, but it should be a decent tool to use as long as you've got a bit of time to manage it. I've already found that Live View on my 1D mk3—even without contrast detection autofocus—is nice for spot checking focus on landscape shots.
Reply to this comment
Yep. I mean exactly that. Thanks for the catch.
Reply to this comment
dpreview's D3 and D700 pages clearly show how the 51 focus points are bunched in towards the center of the frame, just like the 5DmkII.
I'm still leaning towards ditching my EOS 30D with 15 years worth of lenses to get a D700, but it looks like we'll have to wait for the next rev of both Canon and Nikon to get an AF as drool-worthy as the sensor.
Reply to this comment
Canon briefly mentions what one trade is between adding a 50d vs. 5d AF sensors. 50d sensor is "very big".
http://www.dpreview.com/news/0810/08100302_canoninterview.asp
Reply to this comment
Just wanted to point out a couple of corrections.
The original 5D already has high precision mode with f/2.8 or faster lenses with the center point (1/3 DoF compared to standard 1 DoF).
Both the new and original 5D do not have any cross-type AF in the 6 inivisble assist points - 2 of them are high precision with f/2.8 or faster lenses, but non cross-type.
Reply to this comment
James,
Yea the Auto-focus could be a real problem for a lot of people. I don't do a lot of sports or action photography so maybe the FPS and AF are not of importance to me.
But I gotta think sports photographers from the 60s and 70s, before autofocus, had to get by some how right? Right? Well they did, so maybe its time to step up and be a better photographer with what you've got. I'm sure next year they'll launch a 1D MKIV or some other big hitter with more AF points that will hopefully satisfy your needs.
I think your Nikon D700 is a good solution though. Not really sure if there would be any need for you to even get a 5D Mark II, MP count probably isn't too important, and HD Video isn't everyone's cup of tea. But with that said, I'm excited to get my 5D Mark II :).
Reply to this comment
I imagine there is a reason for the 5D MkII framerate which I don't see mentioned usually: bandwidth. In other words: It takes time to move a 21 Mpix image from the sensor to the processor and into the memory. A camera with less megapixels has less demand for communication bandwidth - so it does not come as a complete surprise that Nikon reaches higher framerates.
I imagine that in the 1D this problem is solved not only by using 2 DIGIC processors, but equipping it with a double communication channel to the sensor as well.
Thus, "just" improving the mechanics of the mirror and shutter is not enough... And there is more than just a marketing reason to not have a higher frame rate in the 5D: cost!
Reply to this comment
I must say I've been surprised at the amount of vitriol thrown at the 5D markII after everyone got a look at its spec sheet (on other sites, as well as this one). Why is everyone so worried they won't be able to focus their shots?
From a personal standpoint, I have never had a beef with autofocus on the two Canons I've used - the original digital Rebel and now the 40D. The 40D has only 9 points, too, and if something's ever out of focus, it's my fault. I've shot sports, weddings, and wild birds with it, no problem at all with focus. I don't believe the number of autofocus points alone is enough to sway a purchasing decision.
21 megapixels, however...that threw me. I thought we were over the megapixel wars. That one bugs me, just gonna slow down processing time for me unless I upgrade my computer. I noticed a slowdown when I went from 6MP to 10MP.
Reply to this comment
> 21 megapixels, however...that threw me.
but you always have an option of using the "small raw" mode, and have the choice of 10MP as well as 5MP, if you absolutely detest the high pixel count. By the way, these modes should (if implemented properly in camera software) average the light from the adjacent pixels, which would theoretically result in higher quality images, like the 5MP small raw image would be equivalent to shooting with a matrix having very large (12.8nm!) pixels.
By the way, it is in fact an interesting question -- do the small raw images have better quality in low light ? theoretically, they should be much better, as 4 pixels, well, catch 4 times more light than one. Did anyone test that ?
Reply to this comment
From what I've been able to see so far on the mk3, sRAW doesn't seem to have any benefit from a noise perspective over just resizing a full size image down. It seems that the camera's software is just averaging bits after read out and not pixel binning on the hardware. From what I understand, to get a better noise threshold, you'd need to do it on the hardware before A/D conversion. But, sRAW might be implemented differently on the 5Dmk2.
Reply to this comment
Hi James,
Thanks for your article(s) on Canon and the upcoming 5DMkII. I fully share and support your view. I think it is deplorable that Canon has this "I'll give you only as much as I think you need" approach to camera design. The resurgence of Nikon may put an end to that philosophy. I am also seriously considering a Nikon D700, which means ditching a considerable investment in Canon lenses.. The process will take a while though, so I may actually go for the 5DII in the mean time.. as I reduce my inventory, and when I'm down to just the 5D + my wonderful 50mm f1.2.. sell them both and jump two feet into Nikon.
My ONLY problem with this approach is that the Nikon image quality on the D3 and D700 don't match the 5D (various posters have commented on this).. However the camera ergonomics, their recent announcement of serious lenses, and the whole philosophy of their backward compatibility (see D700 which uses D300 vertical grip.. and the vertical grip that doesn't have the stupid stalk, both simply genius). If I had to compare the two camera makers as software companies, one would be Microsoft and the other Apple. What I don't understand is why Nikon is courting Microsoft so much?! Opposites attract perhaps. Well, sorry for rambling here, but yes, I think Nikon have "got it" and Canon is still on the megapixel race bandwagon.. and have forgotten some of the niceties of photography.
Keep your articles coming, they are balanced and a good read.
M.
Reply to this comment
> From what I've been able to see so far on the mk3, sRAW doesn't seem to have any benefit from a noise perspective over just resizing a full size image down.
well, that's sad. But if you have a chance to test 5Dmk2, I'd be very curious to know if that's also the case with 5Dmk2.
Reply to this comment
Leave a comment