10D's AF Accuracy - Statistical Analysis

[mishkin]

Finally, I did the test I promised some time ago: 10D's AF accuracy/consistency.


Test Procedure
============

The test was performed in the shade in good daylight, 1/1000s at f/2.8 ISO100, or 1/4000s at f/1.4.

4 lenses were tested: 16-35mm f/2.8L, 24-70mm f/2.8L, 70-200mm f/2.8L IS, and 50mm f/1.4. The camera was mounted on Gitzo 1227 tripod with Acratech ballhead. Remote release was used. Neither mirror lockup, nor IS were used due to sufficiently high shutter speed.

16-35 was tested at 16, 24, and 35mm. 24-70 was tested at 24, 50, and 70mm. 70-200 was tested at 70, 135, and 200mm.

Center AF point was used only. JPEG was used with default processing parameters.

For each focal length, 20 AF attempts were made: 10 shots AFing from defocused position at lens' minimum focus distance and 10 shots AFing from infinity. (I wanted to see if the direction of approach makes any difference.) Altogether, 200 shots were taken.

Here is what a typical shot looked like (50/1.4, 100% crop):
http://www.canondslr.com/articles/af-accuracy/images/sampleimage1.jpg
- an obvious ~80 micron backfocus.

The results were digitized and analyzed in Excel, here's an example of the test data:
http://www.canondslr.com/articles/af-accuracy/images/sampledata.gif


Results
=======

The Excel file with all the data and analyses can be downloaded here:
http://www.canondslr.com/download/data/10D-AF-Accuracy.xls

Here are the histograms of the distributions of AF offsets (in microns) for each of the four lenses:
http://www.canondslr.com/articles/af-accuracy/images/hist1635.gif
http://www.canondslr.com/articles/af-accuracy/images/hist2470.gif
http://www.canondslr.com/articles/af-accuracy/images/hist70200.gif
http://www.canondslr.com/articles/af-accuracy/images/hist50.gif

As you can see, 16-35 has a nice Gaussian-like distribution (although I'd prefer the delta-function ) with a standard deviation of 20 microns. The average AF offset is equal to -17 microns - this could be a miscalibration of the camera, lens, or both.

What surprised me, were the next two histograms. 24-70 and 70-200 turned out to be very inconsistent, with no distinct maximum and AF offset spread out everywhere in the range of 160-180 microns wide. As for 50/1.4, I should have taken more than 20 samples.

From the 24-70 and 70-200 histograms it's clear that I need to calibrate these two lenses and 10D. But even if the average after calibration becomes zero, the standard deviation will still be there, unless Canon can somehow "tighten up" AF system of both camera and lenses.

How big is a standard deviation of 40 microns? +/-40 microns fits in Canon's AF spec for D60 +/-75 microns (for 10D it's the same, I presume), even the most of the wings outside of the standard deviation will fit into +/-75 microns range. However, 40 micron AF offset creates a blur of CoC=40/2.8=14.3 microns, that is, 2 pixels. This blur effectively reduces the resolution by a factor of [very] approximately 3x (area of CoC = 3.14*22/4 = 3.14), down to about 2 megapixels. 75 micron offset creates a blur of CoC=75/2.8=26.8 microns, or 3.6 pixels - an effective resolution of only 0.6MP. 1/10th of camera's 6 megapixels... ouch!

Another interesting result is that for the "golden triad" (16-35, 24-70, 70-200IS zooms) it looks like the middle of the zoom range has the smallest miscalibration (and thus best resolution), with the long end having the largest miscalibration (worst resolution).
http://www.canondslr.com/articles/af-accuracy/images/summarydata.gif

However, the "side effect" of this is that the standard deviation - lens' AF "volatility", so to speak - is the largest in the middle. You can't eat the cake and have it, too!

So far, all calculations were carried out for the lens' maximum aperture. When lens is stopped down, the effects of misfocus are decreased until they're hidden "inside" pixels. Here are the graphs of the effective resolution at lens' average AF offset as a function of the aperture:
http://www.canondslr.com/articles/af-accuracy/images/resolavg1635.gif
http://www.canondslr.com/articles/af-accuracy/images/resolavg2470.gif
http://www.canondslr.com/articles/af-accuracy/images/resolavg70200.gif
http://www.canondslr.com/articles/af-accuracy/images/resolavg50.gif

As expected, the resolution plato'es out at 6.3MP at some F-stop. 16-35 seems to be good at all apertures at average AF offset (we shall not forget about inconsistency/volatility, though). 24-70 in its present miscalibration is only useable (resolution > 3MP) at f/4.5 or higher, with average AF offset completely hidden by f/6.7 or higher. 70-200 is useable at f/3.5 and up, with full 6.3MP at f/5.6 and up. Ultra-fast f/1.4 aperture revealing any misfocus is what makes 50/1.4 look bad wide open. As we can see, 50/1.4 should be stopped to f/2.8 to get decent results; f/4 kills the average misfocus completely.

Those were averages, how about random variations? Here are the effective resolutions for all 4 lenses at different apertures (60 samples for each zoom and 20 for the prime):
http://www.canondslr.com/articles/af-accuracy/images/resolrnd1635.gif
http://www.canondslr.com/articles/af-accuracy/images/resolrnd2470.gif
http://www.canondslr.com/articles/af-accuracy/images/resolrnd70200.gif
http://www.canondslr.com/articles/af-accuracy/images/resolrnd50.gif

16-35 can be tammed effectively with f/5.6 and completely at f/8 and up. But look at those bad guys 24-70 and 70-200... they're just wild animals! Even at f/8 you may expect to get an effectively 2MP picture from time to time. Wide open, 24-70, 70-200, and 50/1.4 can be completely miserable, with an effective resolution of a VGA screen. This inconsistency makes you feel insecure when you hold your finger on the shutter release button and you want/need to shoot wide open. "Will I get the shot?" I now shoot 5 times more frames than I'd like just to make sure I'll get at least one shot tack-sharp. Quite frustrating, huh?

By adjusting camera's and lenses' AF offsets by constant amounts, it's possible to nullify the average AF offset for each combination (assuming the offsets are additive):
http://www.canondslr.com/articles/af-accuracy/images/calibdata.gif

Calibration will improve things on average, especially with the last three lenses, but as can be see from these "before/after" graphs of CoC in microns, the inconsistency will still be there:
http://www.canondslr.com/articles/af-accuracy/images/calibgraphs.gif

Of course, all these results apply for my camera and lenses - your mileage may vary. However, in my opinion, the standard deviations are characteristic of the accuracy of 10D's AF system and thus are quite general.


Conclusion
=========

It took me 6 months to realize that I should have bought a 1D with its 3x (? - any 1D measurbators over there? ) more accurate AF system. However, 1D is too heavy a "brick" for someone who likes to hike in the mountains, bike and even run with the camera. D60's AF was way too slow (but accurate! but slow! but accurate! ... ) Looks like there was no real choice but to cope with 10D's mediocre AF accuracy.

As I argue in my "Flaw in 10D's AF Design" article, the Elan 7's AF system was adequate for the full-frame film camera, but is not adequate for a 1.6x crop digital camera with 7.4 micron pixels. The same inconsistency presented in this test would be less visible with Elan. 10D and Elan's AF sensor just don't get along well.

My personal conclusion out of all this - I would never buy a camera with such inconsistent AF again. And when I buy any camera, the first thing to do with it is to measurbate it to find out its real strengths and weaknesses. My next camera will be the rumored Canon 3D. But wait a minute... where's all the green stuff? ;)

Mishkin

[defordphoto]

Good grief. This is sad. Please sell this equipment before you go completely insane.

[ilya]

Having seen your measurbation on many occasions, my theory is that on a certain, perhaps subliminal level, like the fact that your 10D isn't perfect, and you get to solve the mystery of why. One also doesn't often see that kind of passion directed at inanimate objects. Good for you. I for one like reading your stuff.

Ilya

[bigsexy]

Thank you so much for taking the time to carry out these tests. I am currently preparing to upgrade from my D60 and trying to decide in what direction to go. You have deffinately provided me with a ton of food for thought.

[wallytwosheds]

Thank you for your report. You have saved me a lot of money! After all the focus disappointments with the G3 I have I thought I would get a real camera. But jumping out of the frying pan into the fire is not for me!

C'mon Canon get you focus act together - there are a lot of disappointed people reading these forums out there!

[jambo]

jesus.....does anyone actually go out into the real world and take photos of things and stuff?

[Belmondo]

Were all these tests run on the same 10D? I'm assuming that's the case as to do it any other way would diminish the value of your data as regards the specific body in question.

Assuming that's the case, have you run any similar tests on a different 10D body or better, a sampling of 10D bodies? It would be interesting to see if these types of results repeat on every supposedly nominal 10D that's been passed by Canon's QA department.

If you haven’t --- if you only did these tests on a single unit --- then it might require a small leap of faith to conclude that this is a design defect that will affect ALL 10D’s. It is not unreasonable, however, to accept your contention that YOUR 10D has a focusing problem of a nature you're laboriously working to describe.

Regardless, your efforts deserve recognition. You’ve gone to a lot of trouble. Getting everyone to come to your conclusion, however, looks like it’s going to require an even greater effort than what you’ve already expended.

Anyway, I’m a fan.

Tom

[mishkin]

Tom, I don't have the resources and time to test multiple cameras. The main purpose of my test was for me to understand MY camera and document its performance before I send it for servicing. I specifically mentioned that YMMV. It would be interesting to see how large are random deviations in others' cameras. Any takers? You can download the chart here:
http://www.canondslr.com/articles/m-dream/

The more data we have, the cleared the whole picture.

[Dans_D60]

One certainly cannot argue with your statistical and analytical results. One could argue with the protocol and usefulness of such a study, but, data is data and you have done a noble job of representing your points.

Nevertheless, I must weigh in on the purpose of such an intense study in the first place. Photography is both art and science. Being a E.E. myself I also got caught up in science and focusing issues concerning my 10D. Yes, I confess; I too carried out a focus test with my D30, D60, and 10D using several L and non-L lenses. And, yes results did indicate some minor front and rear focusing with all combinations of camera bodies and lenses. I went so far as to post the VISIO focus grid file in this forum that I made from scratch intended for this test so others could do a similar test themselves.

The question still exists if all resources, energy, and time to perform these studies and tests resulted in conclusions that these camera bodies and lenses do not perform satisfactorily in the “real world”. My experience is absolutely NO. I suggest my focus tests were a total waste of my time. I can confidently say that after taking 40,000+ images with these camera bodies over the last couple of years with outstanding results. I’m still amazed, over and over again, with the superb quality in all aspects these cameras deliver. From color reproduction, to dynamic range, to resolution, is simply amazing.

Does this mean I never have issues with my 10D? Of course not. But I also have issues with my Elan and Mamiya cameras including auto focusing. Photography is a hobby of mine for over 30 years and I find my “left” brain gets in the way of my “right” brain all the time including getting all wrapped up in these ridiculous focusing issues. Results speak much louder than statistics. Because photography is a hobby, I shoot everything from sporting events, portraits, commercial studio work, snapshots, etc. Is the 10D perfect. No. For example, I shoot surfing events and on several occasions a professional colleague joins me with his 1D. Does the 1D out focus the 10D? Yep. Do I get my fair share of perfectly focused actions shots with the 10D? Yep. And, when we compare large printouts, the “pro” agrees that my 10D prints just look “better”, especially with large 20X30 enlargements. In the studio I have never …. And I mean never had a problem.

I take a lot of model shots and have the models move around while I snap away trying to get that perfect look. The images are so sharp down to seeing the separation in individual eyelashes that even my Mamiya can’t manage. The 10D is currently the best in class. Perfect. No. Focus problems “in the real world” No. Would I buy it again. Absolute Yes.

Dan

[Belmondo]

I would think that the methodology would be fairly easy to duplicate, but the variable of the optics almost mandates that the same optics be used on all the bodies being tested. We know that not all lenses are necessarily created equal---not even L lenses. Where there are electronics and moving parts, there is likely to be deviation in performance, albeit slight.

It would be interesting to have several 10D owners get together at one place and time, and conduct your test in one consistent manner, all using the identical setup and lens(es).

Perhaps a group of people in one of the larger cities might be willing to get together and do just such a thing.

Tom

[Joytek]

I have just got my 10d back from focus adjustment service and have noticed a huge improvement in the AF performance. The percentage of super sharp pics has definitely risen by a great factor.

However, on the down side the AS center has been unable to clean my sensor efectively. It seem that the dust has gotten under the AA filter and I do not feel like sending my cam in again for an other two weeks.

I would recommend sending in a sharp lens and the body to get the focus adjusted. I have sent in my 70-200 f4 and it seemd to have worked really well.

w.

[Webster]

Mishkin,

I'm a bit confused about the units of measure you are using. Are these really in microns? In millionths of a meter? If so, I'm wondering 1) who cares? and 2) where did you get those pins?

I'm sure I'm just missing something obvious.

[mishkin]

webster wrote:
Are these really in microns? In millionths of a meter?
Yes, 1/1000000th of a meter.

If so, I'm wondering 1) who cares?
I do. If you don't care, you're in the wrong thread ;)

and 2) where did you get those pins?
Who cares? ;)

[defordphoto]

ROFL! At least you still have a sense of humor, if nothing else. :)

[KurtKuhn]

1st of all, mishkin, thanks for your endeavors. I appreciate the effort you put into the data collection.

This thread leaves me with the following question:

After using / exploring my S45 to it's fullest extent, should I keep my plans to upgrade to the 10D? I expect noticeably better photos as the result of shelling out large amounts of coin for a 10D body and L glass. I believe that I take some nice photos now with a modest S45, so you can image that I've set my expectations pretty high for the 10D.

[ilya]

Guys:

This is very simple. Mishkin is a self-proclaimed measurebator, which is not a slur on my part, its a classification he agrees with and looks for others of the same kind to interact with. Read this link, you'll love it.

Ken Rockwell Link (http://www.kenrockwell.com/tech/7.htm)

The only issue here I think is that the folks that are thinking about buying a 10D and doing all the research may be swayed by all this. Please do not be swayed. My 10D and the vast majority of 10Ds sold today are working perfectly. There is a percentage, I don't know whether its 1% or 20%, that have a known focusing problem. You test bodies before you buy them, as I did though I realize that's a luxury not all of us have.

To all those thinking of buying a 10D - its a terrific camera. Read Dan_D60's post, it says it very well.

Ilya

[Dale]

Well said Dan. If one looks hard enough fault can be found in anything.

[kfong]

There is something I don't understand about the AF result:
The AF is a servo feedback system. It drives the lens motor until the AF sensor determined that it is in focus. To prevent hunting there is probably a dead band in the feedback path. Now different lenses have different focal lengths, that corresponds to different gains. That explains the difference in the standard deviations for the different lenses. But the back focus is presumably due to the misalignment between the AF sensor and the main CMOS sensor. So why would the back focus point changes with different lenses? Is my understanding of how the AF works totally wrong?
Could some one please explain?

Ken

[mishkin]

Ken, AF on SLRs is not a closed loop. It depends on proper calibration of both the body and lens. You may want to read various bits about AF on my site:

http://www.canondslr.com/articles

Mishkin

[Belmondo]

mishkin wrote:
The main purpose of my test was for me to understand MY camera and document its performance before I send it for servicing.

Mishkin:
I guess I'm still not clear what the point of all this is. You said that you were just trying to understand what your camera is doing, and apparently you've proven it has a focusing problem.

I think there's a disconnect in your reasoning. I don't see how you go directly from point A, your camera malfunctions, to point B, all 10D's are deficient in design. I think it's been reasonably well established that the vast majority of 10D owners have no such problem with their cameras and are well pleased, I being one of them.

I have to go watch Arnold’s victory speech now.

Tom

[mishkin]

Tom, after the Arnold's speech is over, can you spend 15 minutes and shoot 20 frames with 50/1.4 wide open (defocus each time before AFing). Upload the shots somewhere. I'm curious to see how your 10D+50/1.4 performs wide open.

Mishkin

[CyberDyneSystems]

Mishkin,

This was an awesome undertaking! :)

Thank you for the time you put into this.

I hope that Canon can improve your 10Ds autofocus.

[Dans_D60]

Not 20 frames, but these images (which I have posted before) were taken with my 50 1.4 on the D60 and 10D both auto focus center point.
(Ohhh ... the engineer in me ... why and I still doing this?)
Dan...
http://www.pettusphoto.com


10D 50 1.4 WIDE OPEN:

http://www.pettusphoto.com/focus/10d-01.jpg



D60 50 1.4 WIDE OPEN:

http://www.pettusphoto.com/focus/d60-01.jpg

[mishkin]

Are these 100% crops?

[kfong]

mishkin wrote:
Ken, AF on SLRs is not a closed loop. It depends on proper calibration of both the body and lens. You may want to read various bits about AF on my site:

http://www.canondslr.com/articles

Mishkin

What? No closed loop? I'm SOOO disappointed :-(
But how does feed-forward work on teleconverter and extension tubes?

Ken

[RichardtheSane]

I accept that you wist to prove your 10D has a fault, and it would seem that it does.

I found a fault with my 10D by looking hard enough in various tests. I then decided to go and do some real photography and the performance is absolutely spot on - after I took a few pointers about usage from people here on this forum on the transition from SLR to DSLR.

I think my point to everyone is if you look hard enough two things could happen.
1: You could find a fault with your 10D that may not exist.
2: You could become a measurbator

[Dans_D60]

mishkin wrote:
Are these 100% crops?

Yes …. 100% crops. I sincerely believe these tests were interesting but indecisive and pretty much a waste of my time, I won’t bore you all with the details. But I did run multiple tests with several lenses and with both camera bodies (D60 and D10). The only conclusion based on my test protocol is that my D10 and D60 perform almost identical using the same test parameters.

Dan
http://www.pettusphoto.com


THE SETUP FOR THE TEST

http://www.pettusphoto.com/focus/1.jpg


THE VISIO TARGET FILE AND FOCUS POINT

http://www.pettusphoto.com/focus/2.jpg

[mishkin]

An AF test with slanted target is not a valid test. Also, one shot is one sample. The whole point of this thread is to study variances between multiple shots.

[Longwatcher]

While my experience is similiar to Mishkin's, and I did a ruler test just to confirm for myself whether or not there was a problem (but not to the extent Mishkin did). I found that in real picture taking the autofocus problem is a minor one to me. The autofocus does a far better job then I can do with my own eyes. Usually it hits the target right on, the majority of the rest of the time the subject is in good focus, just maybe not exactly where I wanted the focus point to be. And on about 1 in 20 shots (usually when it would have been a great shot - maybe) it completely misses the mark. I also get about 1 in 50 shots where the camera is just plain off.

My experience with film is similiar although I have a big advantage with the 10D, I can see some of the bad results immeadiately. Also, I have learned that I can correct some minor problems using photoshop, provided I took the picture in RAW. And sometimes, although rarely, the bad focus point actually helps the shot.

So I am happy with my 10D as I am still with my D60 (although I usually shoot with my 10D for obvious reasons to anyone having both). The only areas of disapointment are.
- The 10D had trouble focusing on dark objects/bodies if they fill more then about 20% of the shot.
- I would like a real focusing screen that works for me
- I can't get wide enough because of the crop factor.

Those are the only factors that would make me consider another camera other then of course cost. I would love to have the 1Ds, but I suspect the percentage of good pictures would not be significantly higher then with my 10D.

[JABACo]

I always knew most of you guys were above my head. But now, I'm not only stupid I'm not in the same hemishphere. I only thought I was anal.

Good job...........I think.

[Dans_D60]

mishkin wrote:
An AF test with slanted target is not a valid test. Also, one shot is one sample. The whole point of this thread is to study variances between multiple shots.

OK OK OK … you’re most likely correct with what you wanted to achieve from your test protocol. I’m officially done with this subject! Nevertheless I still proclaim that I have thousands of “keepers” from my 10D that (so far) surpass any other camera body I have owned. And with over the 30+ years in this hobby, I have dozens of different camera bodies all the way back to my Canon QL17 GIII … my favorite rangefinder.
Dan
http://www.pettusphoto.com

[kresner]

First ... ABSOLUTELY no intention to flame you personally or dismiss your obviously detailed and time consuming efforts to quantify your "focus problem".

But, the use of the term "statistical" is without merit. For this testing to have any validity statistically you would have to select "X" number of cameras, "Y" number of lenses and perform the experiment "Z" number of times.

With X, Y and Z dependent on the number of cameras produced (camera population), number of lenses produced (lense population) and lastly the "level of confidence" desired in the test results. Want to be 90% confident or 95% confident?

Major Complication 1: Once you get past a certain confidence level the number of samples and iterations required increases non-linearly ... 90% confidence = test 3 cameras, 92% = test 7 cameras, 94% = test 40 cameras (just as examples).

Major Complication 2: One must determine if the problem is a correlation between camera body and lense? I.e. is it a camera problem, a lense problem, or a combination of both problem?

Major Complication 3: Given the HUGE number of 10D's and Canon lenses produced, to achieve results with ANY meaningful level of confidence one would have to test hundreds of camera/lense combinations, then determine if any correlations exist.

Major Complication 4: EVERYTHING in engineering, design and pricing is a compromise. Been there, done it.
You want perfect focus out of 99.99% of cameras produced - no problem! $5000 each. For $1499 the 10D is a great camera compared to it's digital peers in the same price range. It is in NO way perfect or immune to manufacturing defects, nor are it's digital peers.

Equations to determine what X, Y and Z must be for a given confidence level, and correlation methods, can found in most introductory statistical analysis textbooks.

ps
I am the happy owner of a 10D, a G2, 70-300 USM IS, 28-135 USM IS. I can find "flaws" in all of them ... and any of my film bodies, scanners, monitors, cars, motorcycles and everything else with more than one moving part.

If you see "smiley" faces in my reply don't ask me how the $%# they got there. I saw them in the message preview, but didn't ask for em.

[mishkin]

You should re-read my analysis. I never said that I gathered statistics across multiple 10D's. I gathered statistics across multiple shots on a particular 10D. That's valid statistics for my 10D. My opinion that other 10D's probably have similar deviations is just that - an opinion, based on my understanding of how AF works, and I clearly stated that.

Do you have solid statistical data for your 10D? Share it with us.

[lightandlife]

To be a fair test, one needs to compare 10D's performance with that of 1D or any other comparable digital cameras.

If a lens has to move from A to B (the final target), it takes time. I would presume the shorter the distance between A and B, the more accurate the final focus C will be, i.e., C will be closer to B.

If one pushes the shutter hard, it will take pictures even before the focus has arrived at point C.

To take sharp pictures, it would be desirous to focus it twice, unless you have a moving target. The first focusing would have moved the lense from A to some place C, close to B, and the second focusing would have moved it almost there, in an epsilon (meaning very small) neighborhood of C. I think this would be a good practice.

Of course, with more or high powered chips, one can achieve the focus in a single focusing effort. 10D is still much smarter than B29's used for carpet bombing.

I would like Mishkin to do a similar test to 10D's rivals, and see how good they are relative to 10D. I don' think they are any better if their prices are with $1000 distance of 10D price.

Also, one swallow does not make a summer. But Mishkin has provided an interesting test, which should encourage more scientific tests.

[mishkin]

"I would like Mishkin to do a similar test to 10D's rivals."

I'd be happy to see 'em coming in the mail ;)

[Webster]

Not to flog a dead horse here, but the little gradations on the left in this image are millimeters. If the mm can be seen, it means millimeters. There are one thousand microns for every millimeter. Comparing this picture (taken with a 50mm 1.8 on a D60) with the picture linked to in the first message of this thread makes me think that someone here doesn't know the difference between millimeters and micrometers (a micron is a micrometer, according to dictionary.com). Maybe it's me, I've certainly made stoopider mistakes that that. But if it is me, I'd appreciate somebody telling me where I've gone wrong. Because it sure looks to me like we're dealing with a measurbator who doesn't know what he's measuring.

http://www.veelos.com/webster/mm.jpg

[Belmondo]

webster wrote:Because it sure looks to me like we're dealing with a measurbator who doesn't know what he's measuring.



I fear our friend Webster has just re-awakened the sleeping giant.

Everyone head for the hills.

[slin100]

Webster,

The way Mishkin has his test chart set up, the misfocus error of the subject in millimeters directly translates to misfocus error at the focal plane in microns.

Just be aware that the chart is not 100% accurate; there's at least 10% error due to some simplifications in the equation used to construct the test chart.

[John-M]

joytek wrote:
I would recommend sending in a sharp lens and the body to get the focus adjusted. I have sent in my 70-200 f4 and it seemd to have worked really well.

w.


No lens is required to get a focus calibration. The focus adjustment to the body is not set with your lens, it is set using a Canon "test" lens.

JohnM

[Webster]

The way Mishkin has his test chart set up, the misfocus error of the subject in millimeters directly translates to misfocus error at the focal plane in microns.


Aha! Now that makes sense. So I was missing something. Thank you, Steven. Now there's just more little thing to know and I just might get this whole thing through my thick skull. How does it happen that there is a precise 1:1000 ratio of the real-world focus to the focal plane focus? Is this a 35mm universal? Does it happen to be true at a certain distance from the lens? From the focal plane? Did Mishkin mention this somewhere that I (again) missed? Please help.

[okteh18]

Wow!! Thanks a million for your hardwork. At least now I felt consoled somewhat because my new 300D is absolutely useless when it came to focus accuracy. I've complained to Canon and they loaned me another unit to try but the results are the same. I am VERY disappointed with the focusing quality of the 300D. I did a concert shoot last nite and basically, 99% of the photos are not sharp. (with the original 18-55mm lens) I tried my Tamron 28-300mm lens on it and the results are totally crab, not a single photo is even close to being fair. It was out of focus so much that I'm ashamed that I actually took those photos. Now don't get me wrong. I've been taking photos for almost 25 years and I love my EOS 5, EOS 10 EOS 33 and G3 but 300D....I was actually thinking of upgrading to a 10D and after reading your test, well what can I say. My G3 actually gave me a better sharper pictures. What do you say Canon ? Disappointment, Disappointment, Disappointment!!!

[mishkin]

webster wrote:
it sure looks to me like we're dealing with a measurbator who doesn't know what he's measuring.

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

(read the "Theory" section carefully).

You're dealing with the King of Measurbators ;)

Mishkin

[mishkin]

slin100 wrote:
Just be aware that the chart is not 100% accurate; there's at least 10% error due to some simplifications in the equation used to construct the test chart.

Not at least 10%. The error depends on misfocus amount. If misfocus is 5% of the distance, then the error is 2*5=10% (2 is due to quadratic term). If misfocus is 1%, the error is 2%.

In other words, the chart is the first order approximation, which allowed to exclude focal length and make it universal accross vastly different focal lengths. Second order terms would make the ruler not universal, but fortunately, they're small.

Anyway, the accuracy of the chart is good enough for all practical purposes. It doesn't make much difference whether the AF offset is 50 or 55 microns, especially that the internal accuracy of measuring the misfocus is perhaps +/-5 micron for f/1.4 lens and +/-10 micron for f/2.8 - due to 1 pixel hiding 7.4u*1.4=10 microns at 1.4 and 20 microns at 2.8.

Finally, I was surprised and delighted to learn that Canon itself uses the dx=da/2500 formula in their D60 Service Manual. If the first order approximation is good enough for Canon, it's good enough for Mishkin, too.

Mishkin

[mishkin]

okteh18 wrote:
Wow!! Thanks a million for your hardwork. At least now I felt consoled somewhat because my new 300D is absolutely useless when it came to focus accuracy.

Use Measurbator's Dream to find out exactly what's going on. It's compatible with 300D.

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

Mishkin

[slin100]

Mishkin,

My apologies. You are correct. I was earlier convinced that your equation had too much error.

I now believe that I came to that conclusion because you used an approximation of the equation (an approximation of the approximation, if you will) in another forum. You used da=dx(a/f)² instead of da=dx(a/x)².

[kresner]

mishkin wrote:
You should re-read my analysis. I never said that I gathered statistics across multiple 10D's. I gathered statistics across multiple shots on a particular 10D. That's valid statistics for my 10D. My opinion that other 10D's probably have similar deviations is just that - an opinion, based on my understanding of how AF works, and I clearly stated that.

Do you have solid statistical data for your 10D? Share it with us.


Nope, totally pointless.

To generate "solid statistical data" that would possess any validity for my 10D I'd have to dig out my Statistical Analysis books to formulate a valid methodology, satisfying the critiera I mentioned in my previous post.

As for my Canon 10D and G2, ... the % of pic's I get OOF are no greater or no less than when using my EOS 3 or Nikon N6006 film bodies. I guess I'm super lucky and got a good 10D, G2, EOS 3, Nikon 6006, Canon 25-135 ,70-300 and 17-40 lenses! Hey I'm even happy with my Canon S9000 printer and don't have the fading problems reported by the zillons on every forum .... god, what are the odds of this happening!

So ... I spend my time shooting hundreds of nice pic's with my 10D and "lowly" G2 using an assortment of lenses (all lowly non-L glass) on a variety of subjects, static and moving. Allow though when I zoom in 400% in Photoshop I do sometimes see it's not "tack" sharp.

Let's home my friends, family and clients aren't examing my pics's with 400X microscopes.

[Webster]

Mishkin,

Thanks for the link to your theory page. I must have missed that earlier. The fog is slowly clearing from my brain as I slog through this stuff. I still have a few areas of confusion. The thing is, I really don't understand some of the equations on that page, most specically, the lens equation (1/a + 1/x = 1/f) and the equation for magnification (M = x/a). I do not understand how an equation for magnification can not have the focal length in it. It seems obvious to me that the size of an image on the film/sensor can vary dramatically, depending on the focal length of the lens (unfortunately, many untrue things seem obvious to me, so I don't put much store in that). And I don't comprehend the lens equation at all. Would it be possible to get a more detailed explanation of these, or a pointer to some reference (preferably online, of course) that provides such an explanation? Thanks in advance.


kresner,

I want you to come with me next time I buy a camera or lens. Dude, you have got the touch! Does your mojo work with cars? Mine is getting kinda high in the mileage.;)

[mishkin]

Webster, here's a good online resource:
http://hyperphysics.phy-astr.gsu.edu/hbase/hframe.html
E.g. lens equation and magnification:
http://hyperphysics.phy-astr.gsu.edu/hbase/geoopt/lenseq.html

Mishkin

[kresner]

Unfortunately my good luck mojo does NOT extend to 4 wheeled vehicles, at least new ones. I've had two lemon Ford 150's, both bought new ... there's a Toyota in my driveway now (time will tell).

Anyone willing to "exchange mojo's"?

Not 100% mind you! Just a portion of your new vehicle good mojo for my camera, lense, printer, computer, motorcycle good mojo. %'s are negotiable!

Having been on the design/engineering/marketing side of a major corporation (larger than Canon) I can only reiterate that the price - performance - production - quality equation is always a BIG compromise for any mass produced consumer product, as the technical complexity of a product increases so does finding an acceptable balance in this equation.

Therefore. Anytime I buy a high $$$ item I do some upfront research to understand what are it's known strengths and weaknesses. Upon purchase, IMMEDIATELY and thoroughly put it through it's paces. If performance or quality are sub-standard/defective (vs. what I know given pre-purchase research) ... exchange it ASAP. If the exchange exhibits same problems - maybe one more exchange. After that, I accept the fact there is a geniune design defect and/or I don't have good mojo for this product, so I either live with it or move on to choice #2.

While obviously not hip - for big $$$ items I often buy from one of them dinosaur, old fashioned, locally owned & operated, brick and mortar stores ... where the manager/owner knows my name, appreciates the generosity of my wallet and my repeat business. In a surprising number of cases I'm able to get a very competitive price vs on-line purchase. Do they match or come in cheaper ... rarely, but often close enough for me. And, when I walk in and say "I'm not satisfied with this $900 lense I bought here 2 weeks ago, it appears to be sub-standard/defective", they say "We're sorry Mr. Resner, let's exchange it with another one we have in stock. Give it a try, or better yet bring your camera and test it out first." ... no getting RMA numbers, shipping fees, down time, hassles.

[Webster]

Mishkin,

Thanks for the links, but seeing how I'm personally a few links short (actually quite a few) of being an astrophysicist, they were somewhat over my head. Anything that the average math challenged person could understand would be appreciated.


kresner,

I'm with you on the whole idea of shopping locally whenever possible. Online shopping is great, but for me it's only used when something I want is not available in my favorite local shops. I know the time is coming when there are no local shops, but I weep every time another one closes.

[Belmondo]

kresner wrote:
Unfortunately my good luck mojo does NOT extend to 4 wheeled vehicles, at least new ones. I've had two lemon Ford 150's, both bought new ... there's a Toyota in my driveway now (time will tell).

I hope the Toyota in your driveway is a Tundra. In almost 45 years of owning cars and trucks, the Tundra is the best vehicle I've ever owned----not just the best truck, but the best vehicle.

This is off subject, I know, but isn't the vehicle that delivers us to our photo location part of our essential equipment, too?

[mishkin]

Webster,

When camera misfocuses, it means that the image created by lens is slightly above or below the imaging sensor. This offset is usually very small, that's why it's measured in microns. Measuring this offset, as opposed to measuring the distance between the subject and the plane of sharpest focus, abstracts one from infinite number of combinations of subject distances and focal lengths and gives the universal measure of the misfocus.

Canon calculates the AF offset by dividing the misfocus amount by 2500, because the magnification in their setup is 1:50, and AF offset = misfocus amount x magnification squared.

Once AF offset is calculated, this number is entered in the Service Manual software Canon and the camera is calibrated.

My chart is designed to make it easy to set the magnification to 1:50. Every 25mm (1inch) of misfocus thus corresponds to 25/2500=0.01mm=10micron AF offset.

Hope this clears up things a bit.

Mishkin

[Webster]

Mishkin,

I can't tell you how much I appreciate your efforts to shed light on this issue. Unfortunately, my denseness is resisting the light.

What I was actually hoping for was a verbal explanation of the "lens equation" (1/a + 1/x = 1/f) and an explanation of how the equation for magnification (M = x/a) could not include a factor for focal length.

Your last response, while clearing up many questions, does bring a couple more to my mind.

When you say "the image created by lens is slightly above or below the imaging sensor" does that mean the same as "in front of or behind the imaging sensor"? If not, my confusion increases. I thought the imaging sensor sat with the light sensitive part facing forward.

When you say "magnification in [Canon's] setup is 1:50" does that setup include a lens of a particular focal length? (I'm clearly still stuck on that focal length/magnification thing.)

Finally (ok, more than a couple), how does one go about getting one of those Canon service manuals?

If you would continue your kindness in shedding light, I would be eternally grateful.

[mishkin]

webster wrote:
What I was actually hoping for was a verbal explanation of the "lens equation" (1/a + 1/x = 1/f) and an explanation of how the equation for magnification (M = x/a) could not include a factor for focal length.

Let's consider the lens equation in this form: x = f*a/(a-f). Here x is the distance between the lens and the image plane, a is subject distance from the lens, f is focal length.

Let's start from infinity: a=?‡. The formula gives x=f - the lens should be placed at distance f from the sensor for image to be sharp. As subject moves closer, x increases - lens should be moved farther away from the sensor for image to be sharp.

Magnification equation M=x/a is just a geometrical fact: the size of the subject's image is in the same proportion to the real size as distance from image to the lens to distance from lens to subject. f is included implicitly in this formula - if you substitute x from lens equation, you get M=f/(a-f). Note that when a is much greater than f, a good approximation is M=f/a, since x is very close to f.

When you say "the image created by lens is slightly above or below the imaging sensor" does that mean the same as "in front of or behind the imaging sensor"?

Yes.

When you say "magnification in [Canon's] setup is 1:50" does that setup include a lens of a particular focal length?

Yes. Canon uses 50/1.8 "tool" (well-calibrated) lens with the test chart at 2.5m. Thus, M=50mm/2500mm=1:50. I went one step further and said "let's test every lens at subject distance = 50 focal lengths, to put all lenses in the same situation, when M=1:50".

Finally (ok, more than a couple), how does one go about getting one of those Canon service manuals?

Canon stopped selling Service Manuals to public a few months ago. You may try searching one on ebay. Only manuals for D60 and 1D were available, AFAIK.

Hope this helps.

Mishkin

[Spinners]

I feel so left out.. how can i test my AF system? Just a simple test to see if it works properly.

I mean, my pictures look sharp, but how would i really know if they all look the same right? I would have nothing to compare them to.

So someone set me up with a test please!

[Spinners]

I feel so left out.. how can i test my AF system? Just a simple test to see if it works properly.

I mean, my pictures look sharp, but how would i really know if they all look the same right? I would have nothing to compare them to.

So someone set me up with a test please!

[Spinners]

OOPS forgot there was a second page and reposted.!!!

sorry all!

[mishkin]

Spinners, your pictures look sharp at what aperture? Do you have f/2.8 or faster lenses? Do you shoot with them wide open? If you don't have fast lenses, do you plan to buy them in the future?

If yes, it's worthwhile to check the accuracy of your AF, because fast apertures reveal AF errors with brutal honesty.

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

Mishkin

[Spinners]

Question,


is the speed of the lense solely determined by the lowest fstop? even if im never shooting at it?

i usually dont shoot less than 4.5, ever, however i do have a 2.8 24-85 (i believe thats right) canon lense.

i do mainly portrait and wedding work. sometimes my shots are right on, and other times close. it seems as though my main problem is shake or lighting, but i want to solve this issue once and for all!

[Spinners]

Question,


is the speed of the lense solely determined by the lowest fstop? even if im never shooting at it? I mean will a lense that can shoot 2.8 be faster @ f8 than a lense that can shoot 4.0 but used @ f*.. (make sense??)

i usually dont shoot less than 4.5, ever, however i do have a 2.8 24-85 (i believe thats right) canon lense.

i do mainly portrait and wedding work. sometimes my shots are right on, and other times close. it seems as though my main problem is shake or lighting, but i want to solve this issue once and for all!

[Spinners]

hehe my * is an 8.. man im having a bad typing day.

[Jesper]

Mishkin, did you ever think about bringing your 10D and lenses to Canon? They can calibrate them for you so that they will work perfectly together. That would save you a lot of measurebating trouble.

http://www.rogercavanagh.com/helpinfo/29_calibration.htm

[mishkin]

My data is the "Before" snapshot of my camera/lenses AF performance. After my camera returns from service in November, I'll take the "After" snapshot and publish it here.

My question to Chuck W. about why zooms are calibrated at the two ends only (my data shows that zooms can have totally different calibration in the middle) has been deleted. Folks over there are afraid of deep questions.

Mishkin

[DamienB]

My 10D (my third, first two being faulty) & 100-400 IS L has recently begun to produce a worryingly large number of very soft/out of focus images. Now as I've been doing a lot of low shutter speed/long zoom work (propeller driven aircraft) I wrote it off as simply the conditions. Until I started doing some high shutter speed work (jets) and found it was no better.

So I did a simple test - point camera with zoom out to 400mm at top of tree about 200 feet away, f5.6 and 1/500, shoot 25 shots in a row. On AI servo no less than 40% of shots are useless. Turn IS off - no real difference (43% duff one time, 39% another). Change to one shot AF - 100% of shots were useless. It didn't focus correctly first time. Try again. 100% crap. Try again. 100% perfection. Bingo - the AF is just getting it wrong too often.

Same test on D60. Between 3% and 7% of shots duff with AI servo. It never once got it wrong on one shot AF.

I'm not interested in taking pictures of calibration targets, rulers or nonsense like that - I judge on real life results. And my 10D has moved from being a stellar little performer to being practically useless for my purposes, which are primarily action photography. Were I to need it for non-moving targets it'd be 100% useless as I simply can't trust it to focus correctly any more.

This 10D goes back to Canon. If it had been like this when I bought it only 6 months ago it would have been refund time.

[mishkin]

DamienB wrote:
I'm not interested in taking pictures of calibration targets, rulers or nonsense like that - I judge on real life results.

I strongly disagree. You can spend hours examining real life results and never figure out what's going on. It takes just a few minutes with a good test setup to find out exactly what's the problem, and quantify it. Quantifying is important because if you know the numbers, you can compare them before/after the service. It's quite common that the camera returns from service in the same or even worse condition. Many users reported 2, 3 return trips to service center. That's quite frustrating.

If you present the technician with numbers instead of real life shots, he'll know you know what you're talking about and won't give him any slack, and he'll calibrate your camera well in first attempt.

Mishkin

[lightandlife]

I agree with Mishkin.

If 10D cannot focus well with still subjects, it cannot do any better with moving targets.

If it does well with still subjects, it is yet to be subject to moving target tests. Proof is in the pudding!

[Webster]

Mishkin,

Thank you for taking the time to answer my questions so well. It will take some time and some thought for me to understand it all, but I'm quite comfortable now that if I cannot, the fault is my own.

[DamienB]

After a phone call to Canon that didn't fill me with any confidence that they'd actually do anything to fix it, popped into Jessops and they are happy to simply swap it for another 10D, which is pretty good service I feel. Now I just have to wait for them to get one... which could be a while... damn I'm glad I kept that D60 too!