Why is the 5D3's AF system so sensitive to the lens you use?

[yakim_peled1]

<p>This AF system separates all EF lenses to 8 groups and only when you use lenses in group A can you exploit its full potential. Unlike all other AF systems which changed behavior only when the max aperture changed this AF system is sensitive to particular lenses (6 pages in the manual). It's not a matter of price, max aperture, discontinued or still in production, AF motor type or what have you. <br /><br />Does any one have an idea why Canon designed this AF system this way?<br /><br />Happy shooting,<br />Yakim.</p>

[craigd]

<p>Interesting. I hadn't heard about this, but I downloaded the 5D Mark III manual from Canon to see what you were talking about. Pages 79-84 cover the eight lens types (Groups A through H, where Group A gets the best AF and Group H gets only the one point in the center). For each type, there is a graph showing what AF points are usable (and which ones are dual cross, single cross, or whatever -- and the same point maybe a dual cross in Group A, a single cross in Groups B and C, and completely unavailable in Group F). At first glance, I don't see an obvious pattern, though most of the really awful cheap zoom lenses Canon put out back in the '80s and '90s are in Groups F and H -- though oddly the EF 180mm f/3.5L Macro, which is an excellent lens, is in Group G.</p>

<p>I can only guess that Canon has really pulled out the stops in trying to make the 5D Mark III's AF system as good as possible, and in doing so they found that to take maximum advantage of their best lenses, they had to do things that didn't work well on some other lenses. They probably tested all the EF lenses to see how they behaved with the 5D Mark III and found empirically that they fell into these eight groups. But you'd have to understand the internal workings of the lenses and the AF system to know exactly why it works out this way.</p>

[yakim_peled1]

<blockquote>

<p>I can only guess that Canon has really pulled out the stops in trying to make the 5D Mark III's AF system as good as possible, and in doing so they found that to take maximum advantage of their best lenses, they had to do things that didn't work well on some other lenses.</p>

</blockquote>

<p>Best lenses? And how were these measured? Not by FL, not by IQ, not by max aperture, not even if they are still in production. How can the discontinued 28-70/2.8 be in group A while the newer 24-70/2.8 be in group B and the super-new 100/2.8 IS in group C?</p>

<p>It's not only that I don't understand the logic of the lenses in each group, I also don't understand why they designed it to be so lens sensitive in the first place. </p>

<p>Happy shooting,<br />Yakim.</p>

 

[craigd]

<p>For everyone's edification, here are a few examples (I'm not going to type in the whole six pages):</p>

<p>Group A (61 AF points, including 5 dual-cross points and 36 cross points): EF 24mm f/1.4L II, EF 50mm f/1.8 II, EF 135mm f/2.8 Soft Focus, EF 400mm f/2.8L, TS-E 90mm f/2.8, EF 70-200mm f/2.8L. (The inclusion of the 50mm f/1.8 II, Canon's cheapest lens, shows that it's not just pro-grade lenses that get the best AF. Also, all Group A lenses are f/2.8 or faster, though not all fast lenses are in this group.)</p>

<p>Group B (61 AF points, including 1 dual-cross point and 40 cross points): EF 15mm f/2.8 Fisheye, EF 28mm f/1.8, EF 24-70mm f/2.8L. (All lenses in Group B are f/2.8.)</p>

<p>Group C (61 AF points, including 41 cross points):EF 50mm f/2.5 Compact Macro, EF 100mm f/2.8L IS Macro, EF 300mm f/4L IS, TS-E 24mm f/3.5L II, EF 17-40mm f/4L, EF 70-200mm f/4L IS. (Group C has some f/2.8 lenses, but most of them are f/3.5 or f/4. No lens in Group C is slower than f/4.)</p>

<p>Group D (61 AF points, including 1 dual-cross point and 30 cross points): There is only one lens in this group, the EF 28mm f/2.8. Why this particular lens is singled out in this way is anyone's guess.</p>

<p>Group E (61 AF points, including 21 cross points): EF 400mm f/5.6L, EF 500mm f/4.5L, EF 20-35mm f/3.5-4.5, EF 28-90mm f/4-5.6, EF 75-300mm f/4-5.6. (Lenses slower than f/4 and variable-aperture zooms begin to appear with this group.)</p>

<p>Group F (47 AF points, including 21 cross points): EF 800mm f/5.6L IS, EF 28-80mm f/3.5-5.6, EF 35-70mm f/3.5-4.5, EF 35-80mm f/4-5.6, EF 80-200mm f/4.5-5.6 II.</p>

<p>Group G (33 AF points, including 15 cross points): EF 180mm f/3.5L Macro, EF 1200mm f/5.6L.</p>

<p>Group H (only the center AF point is available): EF 35-105mm f/4.5-5.6.</p>

[craigd]

<p>Like I said, Yakim, I think Canon found in testing that they could get superb AF performance from some lenses, but that other lenses didn't work as well, so based on their test results they divided their lenses into eight groups and enabled certain features only for the lenses that worked well with them. That's the only way I can see that they could have come up with an arrangement as seemingly arbitrary at times as this one.</p>

<p>It does make a certain amount of sense that better AF performance can be achieved with faster lenses, since the lens is wide open when AF is working. It also makes sense that optically mediocre lenses like the cheap variable-aperture kit lenses of the '90s would end up not working as well. But there are a number of surprises, such as the excellent EF 180mm f/3.5L Macro ending up in Group G with only 33 AF points, and the singling out of the EF 28mm f/2.8 as a group in itself. I can only explain this as the result of empirical testing.</p>

<p>However, this all leads me to wonder what the 5D Mark III does with third-party lenses. I doubt Canon bothered to test with Sigma, Tamron, or Tokina lenses, to say nothing of Peleng or Samyang. How many AF points do you get with these lenses? How many cross-type or dual-cross-type points?</p>

[sarah_fox]

<p>Maybe it has to do with the size of the rear element and/or its distance from the focal plane? Just a guess.</p>

[tom_burke3]

<p>It makes me wonder if this sort of thing is true of other EOS cameras as well, and this is just the first time it's been revealed.</p>

[michael.gregory]

<p>If I were to hazard a guess at the reasons for these discrepancies in focusing performance between lenses I would say that it has to do with the degree of phase variation introduced at the focal plane by the variation in construction of the various lenses. Since modern autofocus systems are measuring optical phase variations to achieve focus, and since differences in path length of fractions of a wave length of light at whichever frequency these sensors use can easily be accounted for by the number of elements and their locations within the optical path as well as the coatings and different lens materials, (crown, flint, flourite...), I would hazard that phase performance variations of fractions of a wavelength were never an issue before now. Moire patterns that substantially less than 180 degrees of phase apart would not affect contrast focusing sensors but might seriously degrade the performance of sensors capable of measuring phase differences of substantially less than 180 degrees. Odds are in my opinion fair that at frequencies to which the human eye is not sensitive every lens will create a moire that can be measured by sensors which are sensitive at that frequency, and likely there is a large degree of variation in that pattern from lens design to lens design having little or nothing to do with aperture but likely much to do with transmission characteristics of materials of or coated on the lenses of the placement of elements or groups of elements in the optical path. This type of issue would not occur in systems where the focus was measured by contrast at visible frequencies or where the sensors were large enough to be substantially unaffected by small variations in brightness caused by moire patterns at specific frequencies.</p>

[philip_wilson]

Michaels explaination is interesting but the fact that the beam splitting is done in the semi silvered mirror ( and thus after

the lens) and that the wavelength of light is so small makes it unlikely. Especially since in is mainly the cross sensors that

get removed. As I understand phase detection AF in essence a sensor is a pair of detectors. The light passes through

the lens and some through the mirror onto a second mirror which directs it to the AF sensor array. An AF element on the

AF array has a pair of sensors that look at light coming from two different paths. For example each one will have a micro

lens that selects the light it receives. So if the AF point is in the center of the image one of the pair of sensors (for the

horizontal pattern) will look at light coming from the center of the subject but traveling through the right hand side of the

aperture. The other sensor in the pair will look at light from the center of the subject but traveling on a path at the left hand

side of the sensor. The sensor will then have the lens adjusted until the cross correlation (or more complex) algorithms

determine that the two sensors are reporting the same thing.

 

My GUESS (I am interested in what the answer is) is that the micro lenses used in the new cross sensor on the 5DIII AF

array are much more selective than those used in the past. This would mean that the actual position of the lens aperture

and direction of the incident ray on the sensor become more important. Thus the physical lens aperture size and incident

direct of the light will possibly make the sensor unusable with some lenses as two little light will get through the micro lens

- even though the lens may have a fast aperture. This explaination makes sense to me as I can see that Canon might

improve the selectivity of the micro lenses on the AF array to improve accuracy.

 

As I say this is a guess and I would love to know the actual answer. I also wonder if this will be seen in the 1DX.

[jim_larson1]

<p>All I have to say is. . . this seems like utter <em>madness.</em></p>

<p>I wonder what the real world differences really are (i.e. how important are the dual cross sensors?)</p>

[peter_langfelder]

<p>Just to set the record straight, to the best of my knowledge the phase detection autofocus does NOT measure the phase (or phase difference) of the incoming light waves, in the theoretical sense of the word phase, which is the argument of the periodic function (sin or cos) that describes the light wave. Measuring phase diffence would be difficult to do, since the frequency of visible light is much higher, and the times involved much shorter, than the speed/time resolution of anything you could put in a camera. You can look at the Wikipedia page at http://en.wikipedia.org/wiki/Autofocus#Phase_detection to see an explanation of phase detection AF. <br /> <br />As for Yakim's question - it is possible that different lenses have different vignetting/effective aperture characteristics at various focusing distances and this makes a difference when using the secondary "crosses" of the dual cross-type sensors (which seems to be the main difference between the various f/2.8 and faster lenses). For example, the 100 macro at close focusing distances most likely isn't f/2.8, it's probably more like f/4, and this may impact some of the sensors. <br /> <br />Other than this I don't have any good explanations...</p>

[yakim_peled1]

<p>Michael, what exactly is phase variation? When looking for this, it appears to be linked to biology. <br>

<br>

Peter, this may be it. I wonder...<br>

<br>

Happy shooting,<br>

Yakim. </p>

[ed_avis2]

<p>I would be interested to see focusing using all 61 points with the 90mm tilt-shift lens. This lens is manual focus only, but you can get focus confirmation. For me with the 1Ds and 1Ds Mark II I have only ever seen the centre focus point work for this focus confirmation. If the 5D3 is able to provide focus confirmation using all of its AF points, it could make it an interesting choice for manual-focus lenses which have an AF confirm chip. Or it could be an oversight in Canon's manual.</p>

[steve_dunn2]

<p>My guess, and it's entirely that, is that it may have something to do with the geometry of the light rays coming out of the back end of the lens, perhaps related to the size of the exit pupil and/or how far it is from the sensor plane. This is related to Philip Wilson's suggestion regarding microlenses. A few years ago, there was some press/hype about how DSLRs worked best with "digitally optimized" lenses because the microlenses on the image sensor needed the light rays to come in within a fairly narrow range of angles for optimal performance, whereas film cares little or not about the angle at which an incoming light ray strikes the film, and so not all of the light coming from some film-era lenses was usable by a digital sensor. Perhaps something similar affects this AF sensor; they've had to put a lot of wizardry into this highly advanced sensor, and maybe that means it's more sensitive to the characteristics of the lens than previous sensors were.</p>

 

<p>There have been issues like this in the past, if I recall correctly, in which <em>some</em> combinations of camera, lens, and teleconverter end up disabling some peripheral AF points even though the lens+TC combo meets the f/5.6 minimum (i.e. an f/5.6 lens will work with all the AF points, yet a lens+TC that's f/5.6 may not). But that affected only certain specific combinations, and I don't believe it happened with just a lens, only with a lens and a TC.</p>

 

<p>As for how it affects third-party lenses, that's a very interesting question, and one I don't expect Canon to answer (other than perhaps to suggest that Canon bodies are designed for optimal performance with Canon lenses and all bets are off if you use third-party lenses). I wonder if the AF system is programmed to react specifically to which lens is connected (the specific lens model appears in the metadata, so clearly the body knows which lens you're using, not just its focal length and aperture) or if it simply checks to see which AF points seem to be receiving adequate light and disables any which aren't.</p>

 

<p>Perhaps we'll be lucky and Bob Atkins will have a crack at this with his knowledge of optics.</p>

[drew bedo]

<p>A lot of high quality informed speculation in this thread ( and I mean that to be a posative statement).</p>

<p>Is there any explanation from Canon? </p>

[scott_ferris]

<p><a href="http://www.lensrentals.com/blog/2012/06/5d-mk-iii-af-points-and-3rd-party-lenses">Third party lenses don't fare very well.....</a></p>

[drew bedo]

<p>Can't anyone from Canon explain this?</p>

[Marcus Ian]

<p>I'm sure they can, just don't expect to find that<em> published</em> anywhere.... ;-)</p>

<p>While I was surprised when I first read about this, I was never particularly shocked by it. The physical limitations of certain lenses have <em>always</em> limited AF performance to some degree or another - though we've never been privvy to exactly <em>how... </em> I suspect that the incredible sensitivity of the new array is 'why' we see this though. ie. as capable as it is, the performance difference is very noticeable (whereas before it was much harder to quantify), and instead of explaining to the world 3mo after the release (when somebody w/ a 35-105/4.5-5.6<em> noticed</em> that they could only get one AF point to function), they wrote an override into the firmware.</p>

<p>and thanks Scott for finding Roger's take! Now we know how many 3rd party lenses actually perform... </p>

<p>What that also makes clear is that instead of linking performance to an actual physical limitation for 3rd party lenses (or letting the camera evaluate and decide), they artificially handicapped 3rd party lenses by programming in a 'cap' for them at Group C. --- way to go Canon. So now, even when a 3rd party lens performs on par <em>or better</em> than an EF unit, from a functional or optical standpoint, it's AF performance will be artificially handicapped on the 5D3... Not a 'feature' in my book...</p>

<p> </p>

[steve_dunn2]

<cite>they artificially handicapped 3rd party lenses by programming in a 'cap' for them at Group C</cite>

 

<p>That's not what the article says:</p>

 

<blockquote>We have no way to tell with third party lenses if the central points are acting as single or dual-cross. (Other than knowing the dual-cross sensors are NOT active if the maximum aperture of the lens is smaller than f/2.8, because Canon has said that.) Anyway, for purposes of this post, we’ll call them all Group C, because that’s all we can tell.</blockquote>

[Marcus Ian]

<p>Steve, you are absolutely right. I totally misread that! Of course that leaves open the question of how the camera picks which points to activate, and,<em> if</em> the 'dual' cross types are active (or not) for 3rd party glass (since it is <em>possible</em> there is a cap... ;-0 j/k! )...</p>

[yakim_peled1]

<blockquote>

<p>Can't anyone from Canon explain this?</p>

</blockquote>

<p>I'm kicking myself for not thinking about it earlier. I've e-mailed Chuck Westfall and I'll report back once he replies.</p>

<p>Happy shooting,<br>

Yakim. </p>

[steve_dunn2]

<p>Yakim, thanks for thinking of asking Chuck! If anyone can give us whatever amount of technical information Canon is prepared to release, it would be him.</p>

[yakim_peled1]

<p>Hi, Yakim: <br /> <br />The short answer is that the new 61-point AF system on the EOS 5D Mark III and EOS-1D X is designed to achieve the best possible performance from every EF lens. The long answer is far too detailed to cover in an e-mail message, but the gist of it is that the size of exit pupils and their distance from the focal plane varies substantially according to the optical formula. These differences can cause the distribution or pattern of light hitting the AF sensor to vary as well, with the result that some AF points won't receive light from certain lenses even when their maximum aperture is the same as other lenses that autofocus with all 61 points. That is why EF lenses are classified into individual groups in terms of AF characteristics when used on the 5D Mark III and 1D X. <br /> <br />Keep in mind that when I say "best possible performance," I really mean it. As I've mentioned elsewhere, the new 61-point sensor has 21 standard-precision cross-type points that work with maximum apertures as small as f/5.6 versus none for the 45-point AF sensor on previous 1D-series cameras; Additionally, by designing the AF system on the 1D X with a longer base length for all AF points, even the 20 outer f/4 cross-type AF points rival the precision of f/2.8 cross-type AF points on the 1D Mark IV. In addition, the dual-cross f/2.8 AF points on the 1D X significantly surpass the precision of high-precision cross-type f/2.8 AF points on the 1D Mark IV. The hierarchy goes like this: <br /> <br />1D X (f/2.8 dual cross) is much better than 1D Mark IV (f/2.8 high precision cross-type) is slightly better than 1D X (f/5.6 standard precision cross-type) is much better than 1D Mark IV (f/5.6 single axis) <br /> <br />This is all in addition to other benefits such as: <br /> <br />· Greater low-light sensitivity (EV-2 compared to EV-1 for the 1D Mark IV) <br />· More focusing points (61 vs. 45) <br />· Greater width of focusing point distribution (19mm vs. 15mm) <br /> <br />And I haven't even mentioned the 1DX's 100,000 pixel RGB metering sensor, which adds another dimension to subject tracking performance with face detection and color tracking. <br /> <br />Best Regards, </p>

<table>

<tbody>

<tr valign="top">

<td rowspan="2"><img src="cid:_1_47E2862447E280A4004B092385257A25" alt="Canon U.S.A. Inc" /></td>

<td><strong>Chuck Westfall</strong><br /> Advisor, Technical Information</td>

</tr>

</tbody>

</table>