Focus Testing

10D, EF28-135 IS, 135mm @ f5.6, close focus

Sample from the 28-135 IS, 135mm at f5.6, closest focus

I think slight focusing in front of subject is correct, as depth of field is not evenly distributed along focal plane with more depth of field behind the focal plane. Usually picture is taken with lens stopped down, which will cause the DOF to distribute along focus plane more evenly.
When you consider focusing accuracy of 3-5mm on 0.5m, this is amazing. In reality a handheld photo of subject can be out of focus easily, as in time between focus lock and shutter fire you easily move this 5mm (less movement is enough to defocus picture if you get away from subject). So please judge your focus accuracy only when shooting from sturdy tripod. Another fact with DSLR is anybody can now judge focus down to each pixel. With film nobody enlarged each frame to size 10x15’ to judge focus accuracy. But now when you find let’s say 10% of your frames out of focus, you start thinking of focus problems with the camera.

10D Stories, Episode 744: "Uncle Bob Saves Canon's b.tt". Sorry I couldn't resist :)

10D, Sigma 70-200mm f/2.8 EX APO HSM, 200mm @ f2.8 (1/90 s, ISO 100), approx 2.4 m away

The camera was on a tripod, self timer in use. Camera settings were reset just prior to taking the photo.

AF started from infinity, although when started from 1.8 m (the minimum focus distance for the lens), the results are pretty much the same. My printer added some dithering in the white area, which can be seen in the picture. With another test chart I have used, the results have been the same even though there is no dithering on that chart, so most likely the dithering didn't affect the shot.

At the local camera store today, I tested various other third party lenses with the 10D, and focus was also soft with these (we didn't use a test chart so I can't really tell whether we had front or back focus). We also got soft focus with the 10D that was in the store, but we of course forgot to test it with the Sigma.

However, none of the Canon lenses seem to suffer from this issue, at least according to my tests thus far. The store didn't have any fast Canon lenses so we didn't test them there, and thus I have only tested with my (cheapo) Canon 28-80 f3.5-5.6 USM and Canon 100-300 f5.6 lenses, which both AF relatively accurately (the focus point is easily inside the DOF, usually almost spot-on).

An interesting thing is that when I hooked my 380EX flash to the 10D, AF worked better and the focus point was usually within the DOF. The flash unit died though, so I only have 4 shots to prove it. :-)

The Sigma also focuses extremely slowly on my 10D, but works fast and accurately with my EOS 5.

A few weeks ago I did similar testing with my Rebel 2000 and 100/2 EF lens. The results were very much the same - the camera focuses a bit in front of the subject, but the point on which it is supposed to focus on is still quite sharp, probably within the specified DOF (I haven't done the calculations to verify it). The front focus was consitent, so I'm guessing the focus sensor is actually quite sensitive and precise, but the alignment and positioning of it (or the film gate) may not be. And why did I do the tests? Because I had a few shots where I felt the camera misfocused by quite a bit, but now it seems it was my own fault.

DOF distribution around the focus point is pretty uniform for macro shots (i.e. you get the same DOF in front of the subject as behind), but as you move out of the macro range DOF behind the subject increases faster than DOF in fornt of the subject. So if Canon were picking focus to get a slightly more uniform DOF distribution they'd have to focus slightly BEHIND the subject, not in front of it - and the general consensus seems to be that when focus is off, it's because it's in front of the subject. That wouldn't be a good thing to do anyway. It's best to put focus exactly where it should be!

As for 3rd party lenses, well, remember that they are all reverse engineered. Canon doesn't licence their lens technology to anyone else. Sigma/Tamron/Tokina have taken apart Canon lenses and looked at the data exchange protocol and tried to figure out how to make a lens work just like a Canon lens. Sometimes they get it right, sometimes they don't. Some Sigma lenses seem to need "rechiping" every time a new Canon body comes out so clearly they haven't quite got it right.

I've seen focus errors with Sigma lenses myself, on EOS film bodies. Not huge errors you'd notice right away but subtle errors you have to test to find. Buying 3rd party lenses for an EOS body is always a bit of a gamble. Optically they may be fine, but I always worry a little about the compatibility of the electronics. Some 3rd party lenses may be perfect, some may not be. It's figuring out which is which that's the problem.

Hello, Bob, just for to compare your results with mine: has there been any sharpening on your images? Thank you very much for your very helpful articles, Gottfried

I can't vouch for the validity...but I thought this would be of interest. It was posted on another forum as a Babblefish translation of an 'official statement' from the Canon Germany web site (according to another post the German blendenoeffnungen means aperture):

"Due to physical conditions the range of tolerance of an autofocus is larger when working with similar filmmaterial in principle, than this is when holding picture information on a chip the case. That leads to the fact that with similar pictures themselves still sharp picture results are then obtained, if the point of sharpness lies easily before or behind the motive.

This range of tolerance is smaller with digital pictures. Will this effect was really visibly taken up however exclusively with photographs, those with large blendenoeffnungen (starting from for instance F2,8 and more largely, dependent also on the focal length). With the possibilities of the digital representation and enlargement at the screen, this phenomenon becomes actually recognizable, where within the similar range so far no Unschaerfe became visible. With some few models of the EOS 10D this effect seems to become more strongly remarkable.

Also the assigned objective has sometimes influence on the sharpness result, since it is subject to similar tolerance values. Canon recommends therefore the following test to customers, who determined this,: Photograph with stand a flat surface (e.g. a magazine side, flat attached to a wall, well lit up) with greatest possible blendenoeffnung. Use thereby the autofocus. For a second picture you select the manual sharpness attitude (dioptrien-Einstellung of the Suchers do not forget).

Orient yourselves during the definition of the sharpness not at the announcement of the green light emitting diode, but at the picture in the Sucher. Open both pictures in a picture working on program next to each other with 100 per cent of enlargement and compare you the sharpness.

In the case of a clear sharpness waste in the autofocus picture Canon offers the free examination and if necessary adjustment of camera and/or objective. This applies naturally if no repair damage is present, which was caused e.g. by impact, case, humidity o. ae.. Camera and objective (E) should be returned in above case on: Canon Germany GmbH service center Siemens ring 90-92 47877 Willich"

Howard

From past statements by Canon it seems that the spec for focus on "consumer" bodies (and the 10D is a $1500 "consumer" body) is that focus is within the DOF. On the "pro" bodies focus spec is within 1/3 of the DOF because they use higher precision AF sensors which require faster lenses. With slower lenses (slower than f2.8 or f4, depending on the body), AF accuracy of the "pro" bodies reverts to the same as that of "consumer" bodies which don't have the high precision sensors. I don't know a reference where Canon state this in print, but I'll add one if anyone can provide one.

I've never seen this in print from a Canon source either, yet I continuously find it written and repeated. Soon enough, it becomes fact. When trusted people such as yourself repeat this information without quoting the original source it starts to blur fact from fiction.

I think you're correct about user error being a major contributor. My recent experiences have shown that to be true, but they also showed that there are many variables that affect the autofocus performance. What I've learned from my own tests are the following:

1. When I did tests anywhere near the minimum focus distance of my lens then the results were not as expected. This can be classified as an operator error.

2. The amount of light falling on the subject you're focusing on has great impact on the autofocus performance.

3. Just because the camera tells you it "got" focus, doesn't mean it got "perfect" focus.

4. The camera will "get" focus in low light scenarios but the results will be less repeatable than those taken in good lighting. Try taking five autofocus images in low light (inside the house somewhere) and five images outside in bright sunlight using the same subject or test chart. Evaluate both sets of results and you might notice that the low light shots have more accuracy uncertainty.

5. The information given in the last section (4) means that you shouldn't just taken one autofocus shot to evaluate your results. Take numerous (5 to 10) to see how repeatable and consistent your results are.

What I would have also liked to see with this article is a helpful list of the typical operator errors. I think of what we're classifying as operator errors can actually be split into two more distinct categories; incorrect assumptions about the expected behavior of the equipment, and actual equipment usage/setup errors.

Thanks for the test charts and instructions. A standard is needed here and yours ought to be a great start.

Well I know for sure that Canon say the "pro" high precision AF sensors are 3x as accurate as the "consumer" normal AF sensors, so that part's fact and comes straight out of Canon advertising literature for their pro series bodies.

I'm uncertain of the source of the claim that AF accuracy for "consumer" bodies is "within the DOF". I suspect it comes from discussions I've had with Canon (especially with Chuck Westfall) over the years. I'm pretty sure it's true but I don't know of a print source I can cite! That doesn't mean there isn't one, just that if there is I don't remember where it is...

The images in the article were not sharpened, though they were cropped and downsized from the original image. As such direct comparisons of absolute sharpness may not be valid - however the range of sharpness and its distribution about the focus point can be compared between images. This is really a focus test, not a resolution test. Small differences in camera to target distance can have a fairly large effect on DOF in the macro range since a small shift can be a significant change in magnification. So if you don't see the same DOF that I do when using the same lens at the same aperture it probably just means we used slightly different target to camera distances.

From what I read in the Canon statement they recommend photogaphing a flat object (e.g. page from a newspaper) at maximum aperture first with AF and then with MF (using the screen image for focus, NOT the focus confirmation light). If the MF image is significantly sharper then they offer to checkout the camera. That's attributed to Canon Germany though, not Canon USA. I still haven't seen a statement from Canon USA, though I presume they'll checkout any user complaint about any camera malfunction, just as they always have done.

AF with the EOS-3, written by Bob Atkins with input from Chuck Westfall, says:

A "normal-precision" AF sensor produces focusing accuracy within the depth of focus for the lens' maximum aperture. A high-precision" AF sensor produces focusing accuracy within 1/3 the depth of focus for the lens' maximum aperture.

Unfortunately, since two people were involved in putting that article together, only Chuck and Bob, memories permitting, can be sure who said exactly what. Still, Chuck has presumably read over the finished product, and would have commented had it been wrong. I had a quick look at the EOS 1N and EOS 3 brochures, and neither one gives details of how precise AF is.

BTW, minor note on the lens test chart - it should be in GIF format, not JPEG. JPEG compression is not well suited for line art; GIF is, and the same image as a GIF is just over 100 kB rather than close to 400 kB. JPEG also introduces artifacts, particularly around the edges of lines, though this image seems reasonably clean.

Steve - Thanks for finding that reference on my own website! Looking at the article I can see it was written by Chuck, not by me, so you can take that as an authoritative statement for a Senior technical source at Canon. I knew there was one, I just couldn't remember where it was!

A agree GIF is better for line art. In this case it doesn't really matter since it's not a resolution test chart, but the smaller file size would be good. I'll see about generating a GIF version from my original CAD file.

Its not surprice that focusing errer happen in high end camera. A magazine in China have design some interestings test about focusing missing half year ago. The result is the 1. The mid-end models from Canon and Nikon is better than the high-end models. 2. Overall speaking, Minolta and Pentax not worst than Canon, Nikon.

How big is the autofocus sensor in the EOS cameras? I had tested 3 EOS 10D camera bodies last week in a store and all three seemed to have front focus bias (one of these less than the others). I was unable to do the rigorous testing Bob describes above in the store - I simply focused on a page of magazine writing from an oblique angle. Text in front of the line I aimed at was sharper than text behind the line. If the focus sensor is LARGER than indicated by the red LED frame, that might explain front focus: the camera would tend to autofocus on the sharpest, most contrasty subject WITHIN the sensor area, which in many such cases would lie indeed in front (and outside) of the area pointed at by the center of the focus LED.

If this is the case - if someone could confirm that the sensor area is larger than the LED frame - than this, in my opinion, would explain 99% of the "autofocus errors" reported by 10D users.

I am saying this because I have a hard time to believe that 2 out of 3 new Canon 10D bodies should be "defective". Also, the camera I bought (which performed best in the store...) still has front bias on certain subjects on certain angles, even though it performs flawlessly on Bob's test above. To me this suggests, that we are dealing here with the limitations of autofocus (and the ease to detect them though instant digital feedback), rather than an inherent flaw or defect in Canon's 10D series cameras.

Based on test results from earlier EOS bodies it's quite possible that the actual AF area of the center AF sensor is larger than the square marked in the viewfinder.

You can do a rough test of that by seeing where the focus line has to be in oder to get focus lock. If the focus will lock on the line when it's outside the marked AF zone, then that gives you the answer!

10D focus test with 50mm @ f/1.4

I performed the test just to prove how silly this whole thing is. Used my Canon 10D with a Canon 50mm f/1.4 lens wide open. This was the result I got. I took delivery of my 10D on March 15th, the first week they were available (SN 0220105430).

I've repeated the test using different target types and I get similar results. I was also surprised to find that I get the same results when I repeat the test using manual focus and ignore the assist light in the viewfinder.. Anybody care to explain?

Guess I'll go see Canon on Monday.

I'm not quite sure how the adjective "silly" applies here.

There are clearly examples where novice users are misunderstanding the camera and attributing "soft" images to the camera that are not the result of any camera defect. These range from a lack of understanding about in camera and post-exposure sharpening to a lack of understanding of how the AF system works and what it will lock onto as the focus point.

Most 10Ds seem to be well within spec. Some seem to be out of spec. To find out which yours is, you need to do a simple test - like this one.

So what's so silly?

Canon 10D, love the camera. But there is a focusing problem, and I don't use cheap len's with it. My lens are L series lens, and two other friend had the same problem.

I took it to canon uk, and they kindly fixed it on the spot. They said it was out by 2cm.

It seems to be a lot sharper now,I may even get another one.

Ciao 4 now.

Thanks for the article Bob. I did the test and I'm happy to report that my 10D is focusing excellent! Good luck everybody!

Bob,

I think you are being too sensative. I assumed, like you, that a lot of people new to digital were not understanding the tendencey of digital images to look a bit soft when they come right from the camera . But when I performed your test I found that my camera was not doing a very good job with autofocus. So, I'm greatful for your article and test chart. I think a $1499 camera ought to do better than mine just did on the test.

I also repeated the test using five boxes of film (Fuji Provia 100F that has been sitting in the fridge since I got my first D-SLR in October) each 3/8" farther from the camera. I set the focus point on the center box and found that the sharpest point was not the box that I focused on but the one next to it.

Zap

I printed out the image and set it up so it was about 4 feet away from my 10D on a tripod. It was inside with an open window yielding pretty good light - exposures were around 1/3200 - 1/1600th at f1.4.

The results:

35mm f1.4. Focused all over the place but always in front. Typically about 5 cm in front. In 24 tests, it ranged from perfect (once) to 6 shots where the focus seemed to be at or beyond the front of the edge of the page. 5 cm seems to be the average. This is my "standard" lens so I tested it a lot - and tested it more and more in disbelief. I even tried things like changing the axis of the chart relative to lens but without any change.

100 mm f2.8 macro. Back focused 0.3cm. Tried 4 shots - all were exactly the same. Nice change.

50mm f1.8. In 4 shots taken, it was perfect twice and in front by 0.8cm and 1cm.

28-105 f3.5-4.5. At 28mm it focused around 5 cm in front. (exact point hard to read with such a large DOF). At 70 mm focused at 2-2.5 cm in front. I only tested 2-3 times at these two focal lengths.

I've just gone from a manual focus (FD) to EF/digital and I don't have any experience with this. It seems like two of my lenses front focus by a heck of a lot or is that my 10D misfocuses wider lenses? Does one of those make more sense to experienced EOS/EF users? Any recommended next steps?

Dave

If you really are 4ft away and your target is the same size as mine, you cannot isolate focus to only the focus line at 28mm or 35mm, you'll have some of the bar patterns in the AF area, so all bets are off on where you get focus. Even at 50mm it's on the edge of isolating focus. At 100mm you'll be OK

To use the chart properly you must have ONLY the single focus line within the AF frame (in fact it's best if the AF zone covers only the center part of the focus line as AF sensitivity may extend outside the marked zone).

AF zone and focus line

Focus should look like this with the AF zone over ONLY the focus line.

Hi Bob, I should have mentioned that I tried several modifications to the chart - the first mod was to space the test bars further to keep the them away from the center focus area. I've also tried adding a little detail and/or making the center line thicker in an attempt to make it easier. Here's an example of one modified chart:

To eliminate any chance of angle-bias - I also tried straight on shots but then nothing on the page was in focus.

Dave - That looks fine. As long as the camera has no choice but to focus (or misfocus) where you want it to.

I tried the test with a 24/2.8 and focus looked fine so I don't think it's a wideangle issue.

Something to be aware of (probably mostly when using wide angle lenses?) is that sometimes the camera will seem to focus then stop and the image will look pretty much in focus on the screen - but the green focus confirmation light may be blinking indicating that the camera hasn't actually confirmed focus. Maybe it's making a "best guess" at focus when the subject detail isn't stong enough to get a positive focus lock, or maybe it's just stopping somewher in the middle of the focus range and with a wideangle lens the DOF is enough that things look in focus. I don't know, but it's something to be aware of.

Thanks Bob. The green light is coming on solid for me. Playing around with my 28-105 has been interesting (shooting from a spot about 4.7 feet away - via window light - around 1/125th at f 3.5).

If I just stand over it and watch the focusing scale as I press the remote release: At 105mm it focuses at what appears to be 4.7 ft and it never wavers. Most successive focuses don't even emit a "tick" from the focus motor and if they do, it doesn't move perceptibly. At 28mm, the focus lands on a range from about 3.1 ft to 4.7 ft. and it focuses more often at the lower end. Sometimes it will stay in a good or bad place for quite a few successive touches and then head off in the right or wrong direction and stay there for a while. Other times it will dance around more.

HOWEVER, then I moved the tripod so the camera is within 2 ft. The 28-105 lens (at 28) focused to a spot in its "macro" range and hasn't budged on dozens of focuses - ie focus motor usually doesn't even tick. Shots taken look fine. Leaving the camera there, I mounted the 35 f/1.4. It too now takes fine shots consistently.

By the way, all tests were done with a version of your chart with just the single center line - but lengthened to 4.5 inches to avoid contamination from the focus test lines when shooting at 4.7 ft. .

So.... My interpretation is that my previous wide(ish) angle shots at 4.7 feet were exceeding the ability of my 10D's focus sensor to resolve that central line. (Sound reasonable?) I'm wondering though if I'm at the same limit as an average/normal 10D?. You said that your 24mm is working fine. Could you tell me the distance you were shooting from?

Oh here's a data point - I just borrowed an Elan IIe (turned eye control off and locked center focus). With the 28-105 at 28mm and at 4.7 feet, it dances around FAR more than the 10D. However, the focus indicator is always flashing too so it can't do it at all but lets you know.

Sounds like fairly normal behavior. I tested the 24/2.8 close focused to minimize DOF.

I also see different behavior with the camera horizontal and vertical. Sometimes one or other will give more positive focus, even using the center cross sensor.

I think we're just running up against the limits of autofocus technology and we're looking at prttey abnormal targets.

BTW I do not trust the distance scales much on zoom lenses. Very, very few "zoom" lenses are parfocal, i.e. don't show at least a little focus shift when zoomed. Most are varifocal and need some degree of refocusing after zooming. Obviously if that's the case the distance scale can't be correct at all focal lengths.

The bottom line is whether or not you get consistantly badly focused shots when using the lens to shoot "normal" subjects. If so then you may have a faulty camera. If not then you probably shouldn't worry.

Check out this improved focus target here. It consists of the diagonal scheme above, but the target of focus is a cut-out target that coincides with the plane of focus. Source, Target, Target in Use.

Ahh sorry - Peter Phan was referring to a target I created, that I've since taken down as I needed some space on Pbase for some other test pics. Sorry about that.

What I can say though, is that the strictly diagonal charts do present a challenge to the AF system and that results, as a consequence can be inconsistent. Rather than me repost my modified target, I'll describe an alternative, more consistent test that can be used with the one published here. Take the test chart as provided here, lay it flat on a table, with the grid lines running away from you, and place a vertical object, of similar size say, to a lens filter box, cigarette packet, matchbox etc on the centerline.

Place camera on tripod, and shoot the target from as low an angle as possible, whilst not being so low that you can't see the grid lines. (remember, the chart is lying on its back on the table, with target standing up at centerline). This gives a much more definite and real world type target at the focal point than the thin line used originally. One can still ascertain the focus accuracy, as you still have the grid lines to review the point of focus.

Whilst my 10D gives variable results, depending on ambient light etc with the unmodified test, it's bang on using my modified version, regardless of available light levels. I suspect the variances we are seeing reported are down to differences in test conditions. A simple change can eliminate those inconsistencies, and hopefully many who thought they might have a problem, won't send their camera's back unnecessarily.

For what it's worth, I've seen people reporting front focus using the unmodified test, sending their cameras back to Canon, and getting them back rear focussing. A theory I have, is that Canon, putting the camera on their test rig, find nothing significantly wrong. An assumption may be being made that cumulative tolerances of the lens and body are the cause of the incorrect focus, and to satisfy the owner, the camera is adjusted to focus rearward slightly. Probably still within specs, but they might just be altering the bias to the other end of the tolerance, thinking that the front focussing lens, will mate with the rear focussing camera, and lead to a satisfied customer.

However, what really happened was that the customer used a slightly flawed test chart/lighting, got an anomalous front focus result, and Canon gave them exactly what they asked for, even if that isn't exactly what they wanted.

My advice then, is be careful folks - if you use the chart as it stands, it may give you a false result. Test using a real world object at the centre line of the chart to get the focus point well away from those sloping grid lines. Then if your results indicate problem, send it back to Canon, but don't send it back on the basis of the unmodified test alone, or you could be wasting your time.

Simon

I have tested my 10D focussing using the chart, and remain confident that my particular camera is working very well (I have tested the focussing before using other methods).

I have what may be a useful tip for anyone looking at their results (from using the chart), trying to determine the exact point of focus: I have found that, if you load the resultant image into Photoshop or Photoshop Elements, you can use the "Image->Adjustments->Threshold..." menu item to have Photoshop 'display' the in-focus lines on the chart. Since the more 'in-focus' areas will be darker in the image, then, as the Threshold Level slider is dragged right and left, the 'displayed depth of field' is increased or decreased respectively. (Make sure that 'Preview' is checked.) So, as you carefully drag the slider to the left, the 'displayed depth of field' is reduced until only the sharpest, most in-focus lines and numbers are visible on an otherwise white image.

I found this rather useful. I don't know if anyone else will, but there it is!

Im new to canon (just switched from nikon)I performed the test and I found that my new 10D with the new 70200 2.8L focused perfectly. I also bought a 28-90 f/4-5.6, this one performed terrebly in the same test. my question...

is that normal?I mean, I know one is pro and the other one is "cheapo" but is that what I should expect from non L lenses. (sadly I couldn´t afford another 28 - 80 range L lens). I was considering returning this one and shitch to sigma 24-70 f2.

I decided to try this test with my new 10D and 24-85mm lens. I considered the exercise just another part of my learning curve as I continue to master this camera. I put the camera on a tripod, ~ 19" (minimum focusing distance) from the center of the chart, at the prescribed 45 degree angle, and used the self-timer to trip the shutter. My results:

@ 24mm, F/3.5; auto & manual focus: DOF normal

@ 62mm, F/4.0; auto focus: DOF slightly forward

@ 62mm, F/4.0; manual focus: DOF normal

@ 85mm, F/3.5; auto focus: DOF forward

@ 85mm, F/3.5; manual focus DOF slightly forward

I chose 62mm as the midpoint, as this was the closest I could zoom in and still obtain an F/4.0 aperture. In all cases, the target center line was sharply in focus, so I'm content. I should qualify myself by stating that, until last week, I haven't touched an SLR since the 70's; and have been using AF compact 35mm and digitals for the last 12 years. Also, I've gleaned a ton of current information from this site in the weeks preceeding my purchase, and I happily linked through to one of Photo.net's featured vendors to make my purchase!

the center FP of the 10D appears to have a broader "reach" on this test target than it does in regular use. I could consistently get all of my lenses to front-focus using a test target like the one shown here, even though there have been no issues whatsoever with my photos with any of the lenses (I get one "flier" OoF shot in about 200). If you are going to test your camera, use the test Canon uses; something like this. http://members.cox.net/phototest/testrig.jpg

Sending your camera away for weeks at a time so Canon can waste their time and money checking it holds up your own work, and other cameras with legitimate warranty issues. So take some time and effort and make SURE there is something wrong with your camera or camera/lens combo before sending it away.

Hello, Could anybody perform the test in tht dark or weak light?

It seems this test measures a camera's ability to focus in the center of the frame. Yet most focus problems, it seems, are due to one trying to auto-focus in a low light environment, where the flash provides the light for the film/ccd, post focus. To measure this "problem", one could do the same thing you are doing now, yet with varying light intensities on your 45 degree board, where the flash is needed to get the correct exposure. Then, the question might be, "At what light level does a camera loose focus by 1cm on your board?". Since the camera throws some light to auto-focus, one might want to try this at different camera to 45degree board distances.

I performed this test on my 10 day old 10D. I discovered severe back focusing with my 70-200 f/4 L and significant front focusing with my 28-135 IS. I returned the camera and got another one. I retested and got the exact same results. Could it be a problem with certain serial number cameras, or is it a case of two bad lenses? I am in a quandary if I should just return the camera or send everything in to Canon for troubleshooting and repair. When I called Canon tech support I sensed denial about the issue.

Like so very many others consumed by curiosity - or paranoia - or fear, I downloaded the calibration test, slapped on the Canon 50mm / f1.8, taped the chart to my wall and clicked the shutter to my heart's content.

I shot wide open and stopped down, manual focus and autofocus, at 30 degree and 45 degree angles and at distances from 1.5 - 3 feet.

And the results were all over the map. Sometimes the shot was dead on. Sometimes the lens focused in front, sometimes behind - but never more than 2cm. I had the same results regardless of whether I was using MF or AF.

So what did I learn? The autofocus sensor has limitations. And so do I. An aperture of f1.8 or larger leaves very little room for error for both man and machine. Which is why I stop down. There is no question that there are 10D's out there that need to be recalibrated. But for most of us, it is simply a matter of accepting the limitations of the eye and the autofocus - nothing more.

Those of us who have a history with film SLRs have boxes full of the ones that "got away." And I can guarantee you will have the same results with a digital camera - any digital camera. That's what the erase button is for. Remember - digital does not mean dead on. It's just a different type of "film".

Items of note:

If I set my 50mm to MF, turned the focus ring clockwise to its limit, switched to AF, refocused and took the shot, my 10D focused BEHIND (to the right of) the center line.

If I set my 50mm to MF, turned to focus ring counterclockwise to its limit, switched to AF, refocused and took the shot, my 10D focused in FRONT (to the left) of the center line.

On MF and diopter adjustment:

Many have asked how set up their 10D to achieve precise MF. Most of the board posts I've seen recommend focusing the eye on the focus points in the viewfinder and turning the diopter adjustment until the focus points themselves are tack sharp. In my experience, this is a very imprecise method.

The method I use is to press the shutter button half way, activating the viewfinder display. I then turn the diopter adjustment while focusing the eye on the display, and repeat until the display is tack sharp. This seems to yield much better results.

In addition, I have found that the calibration test chart is an excellent tool to dial in the diopter adjustment. Set the diopter, and using MF, take a test shot of the chart as Bob describes. Check the image. If focus is in front of the center line, turn the diopter 1 click clockwise. Repeat until center line MF is achieved. Reverse if focus is behind the center line.

I read with interest all of the comments made about the focussing of the 10D. I don't currently own one but do plan on purchasing soon. It seems to me that some of the cameras may indeed have a problem with the autofocus but, and this is a big but, why would anyone shoot wide open in the first place? You're not getting the best from your optics (consumer or pro) shooting at the widest f-stop. If you are shooting at any reasonable distance, DOF should take care of any slight focus errors and if you're shooting macro, should you be relying on auto-focus at all? Most if not all digital cameras are notorious for being slightly soft. So the question should be, how soft is too soft? That's why there is an Unsharpen Mask tool. If you really want to go nuts, shoot the same test with ISO 100 B&W film and see what you get. Canon builds nice gear but perfection is unattainable, especially when one is talking about a device with the complexity that this camera has. If perfection was possible there would be no need to ever offer a warrantee. If you consistently get bad photos when using a good tripod, then you may indeed have a problem. Otherwise, use your optics to their best advantage.

Just a short comment on these results. Actually DOF must be asymmetric around the focussing point, i.e. you have to expect a larger in-focus area behind the point and smaller in front of it. I think, this is very well shown in the manuals to many EOS cameras in the picture with a funny penguin ;)

I think in part of this discussion we should be asking how many steps can the lens motor really achieve? All the lenses use stepping motors, whether they are core motors or USM.

I did this test once at about .45 meters, as the article, and found a slight forward bias, maybe half a centimeter. I did it again, around .8 meters (was about the middle of the focus range on the 50mm lens) and found a slight rear bias, again, maybe half a centimeter. I wonder what would happen if we do a test, find the bias, and then move the camera by the offset, so that a "step" of the motor is now on the focus point?

Another thing I'm curious about... since the focus is through the wide open lens, once the focus point is within the DOF, can the focus sensor even tell that it's not the precise focus point? In other words, once it achieves any focus where the 'circle of confusion' is smaller than the pixels in the sensor, how can it know to try harder to focus?

Some thoughts.... curious if anyone has any perspective on those issues?

I bought the Canon 10D only fews month ago. Just last week I went to the Canon Digital School and some students asking the question about the problem with the 10D having problem with the focus. What Canon comment on this question was is true that the 10D have this problem but not all 10D had this problem. Just that will you be the lucky one to pick up a 10D to have this problem.

David

This test method is flawed: the focus target (thin line) covers only 1 pixel on AF sensor (it's 3 pixels in 100% crops, thus about 20u wide (7.4u*3), which is the size of the pixel on 10D's AF sensor, which has 50 pixels per mm (total length 2.5mm, 128 pixels)).

1 pixel of data is not good data at all.

Besides, some texture of paper inclined at 45 degrees may affect data on AF sensor and make a contribution to misfocus.

Try this test:

http://www.canondslr.com/articles/m-dream

Each thick bar covers 10 pixels on 10D's AF sensor, altogether 30 pixels out of 128 engaged by two thick bars. That's very bold data for AF sensor, and the target is parallel to sensors.

In its simpliest variant, this chart is simply a single center page and a regular ruler with 1 inch marks inclined at 75-80 degrees. 1 inch will correspond to 10 microns of AF error.

I had something that's not exactly the same, but similar, happen to me on my Rebel 2000. I had my camera cleaned up by a Canon technician, there was some dust on the focusing screen. After a while, I noticed that the autofocusing was wrong, and I had to override it to get the right results (pretty obvious and easy to do on my 100mm/f2, thanks to FTM).

I did some more precise testing, and it turns out that clearly enough, the focusing screen was off, but the autofocusing was correct! So when I thought that I corrected the focus, I actually messed it up! I sent it back to Canon, a few weeks later it came back working perfectly, no question asked.

I also must recommend my local camera shop, L L Lozeau in Montreal, Canada. I was registered for a full day studio portrait workshop when I sent my camera away for repair. They lent me one of their rental A2 for free, not only for the day of the workshop, but until I got my Rebel back! That was very nice of them.

sigma vs canon

I have a Canon 300mm F4.0L and a Sigma 70-200 f2.8 APO EX. The canon lens focuses in the centre of the expected dof. The sigma prefocuses 2-3 cm infront of the dof... way out. The shop I bought the camera from said it was the sigma lens. When I spoke ot Sigma, they were cagey and said it the flange fitting on the 10D. Apparently the tolerances for Canon lenses are less than for the Sigmas. The hinted I should take the camera back. Meanwhile the shop has "tested" the lens on a 300D (just buy taking pictures across the street!!...) Unfortuantely they are a small dealership and have very limited stock of anything. Anyway, I also tested my Sigma 15-30. Not so bad as the 70-200, but has anyone tried hyperfocal focusing with it... WAY WAY WAY out... I was distraught when I came back from Iceland. I couldnt figure out what was wrong. I thought it was my technique... If anyone else has this problem or has known solutions, please do let me know...

Cheers!

JR

Another experiment

Here is the image.

Test with Canon EF 50/1.8 at F1.8

My Eos 10D seems to be correct in focus, right? The slight difference to the right may result from not having a 100% mathematically measured setup (maybe 44.5 degree and not 45, etc.). Everything is ok and I shouldn't be concerned, right?

MF

Bob; First let me tell you, thank you for so much gret information, detailed information. After doing your test this is the result I got. I took many pictures, but they were somewhat the same. One question, should I take a MF and AF picture with this diagram of yours with every lens I have with the 10D or one lens is enough? The lenses I have are 24-70mm 2.8L USM, 27-70mm 2.8L USM, 70-200mm 2.8L IS USM, 100mm 2.8 Macro and also the 2X Teleconverter for the zoom.

Here is the test results: Picture one names MF for manual focus was at 1\80 at 2.8, and the secon picture was AF for auto focus at 1\60 at 2.8 taken with Center weight meetering and the center square was the active focusing point on the center line of your diagram. Also the center of the square was placed on thecenter line for more detailed informarmation.

Please let me know how the result is in your openion. I think it is good. Also let me know if I should take the same shots with the other lenses please.

AF

My AF picture

There was some conjecture among the comments that the focus sensors actually do act over an area larger than the delineated box in the viewfinder.

This is, I believe a true statement I have at times achieved focus lock from a sensor that didn't actually appear to be on any thing yet there was an edge that was in focus just outside the perimeter of the red box. I have noted it most commonly when shooting detailed macro shots where focus is easy to see, DOF is very shallow, and the work is done from a very stable tripod.

I haven't done a test yet myself but I suspect that my 10D has very slight front focus.

There was some conjecture among the comments that the focus sensors actually do act over an area larger than the delineated box in the viewfinder.

That is why I recommend the domino test, lining up the dominos to the sensors, and selecting the center sensor.

What is interesting about mine- I got one last week to play with, with the 28mm f/1.8 and the 50mm f/1.8. I had heard that the 28mm was a bit soft, and it didn't let me down, soft wide open, and then...

I had installed the various software that comes with the camera on my laptop, and set up the camera to take pictures and download directly to my laptop. I set up the camera and tripod for testing exactly HOW soft the 28 is, and shot the first photo...

First photo was of a scene I used for all of the tests, objects at various distances, between 2 and three meters away. No flash used for auto focus. I was using the silkscreened channels on the UHF dial of my television- they had proven to be soft in the past, and are great detail for the test.

The first shot was sharp. It wasn't soft, at all. The lens was obviously resolving things just fine. I proceeded through the test, and the lens indeed did sharpen slightly as I stopped down, but the numbers were perfectly readable when I first ran the test wide open. Then... I tried again, wide open, and shot. Out of focus. The numbers were out, no doubt. Re-shot, and again, out of focus. Manually focus it, and things... Were out of focus? Hmmm? Try again? Auto- it's fine.

In other words, my camera is rather unreliable with auto-focus. I am going to spend some time figuring exactly *how* it's unreliable, and perhaps I can work around it. I'm also going to repeat the test with the 50mm, and see if it's an interaction with the 28mm that's causing the trouble. My 50mm shots I remember as sharp, though, so I don't think so.

Interesting. ALL cameras come to me needing to be sent back to the various repair depots for focusing issues. My G2? Kyocera. My Leica? DAG. My Voigtlander Prominents? I had to set back-focus (collimated) and then rangefinder adjustment. Grrr.

My 20D is front focusing about 1 cm on the same chart, as well as on other charts.

How prevelant is the problem?

Did you all send back for service?