Variations in AF performance: any reason why?

[MM1]

30D TTL-SIR really has nothing to do with this thread and nor really can you make a straightforward jump between XXD series AF and 7D AF.

Why can't I? 7D AF looks more like xxD AF on steroids, than 1D AF lite.

two entirely different AF designs? also later designs on the XXD series had major defocus - something the 1D may have calculated out based upon multiple cross types.

Compare the sensors... On xxD, 5D and 7D, the AF points are formed out of single baselines, either one or two for a cross sensor. 1D AF sensor is completely different, where AF points are just segments of several long baselines of the AREA-SIR sensor.

Even if I make 7D sensor a new design, it is more similar to xxD sensor than to 1D sensor.

BTW, what's "major defocus", please?

undoubtably .. the 1D's high precision sensor would be more accurate than the 30D's -

Why? They both have the same "paper" specs.

actually no they don't, and no they weren't. the 30D wasn't a 2.8 cross type at the center. it had "additionally sensitive" sensor regions, but werent' 2.8 cross types as the 1D series had. Slightly different. also I'm completely assuming here that you are only talking about centerpoint accuracy.

Of course I talk about the center point. But someone is wrong here... Either you or (for example) the guys who describe 400D AF on this page (couldn't find 30D description, but 400D should have the same AF):

http://www.imaging-resource.com/PRODS/XTI/XTIA4.HTM

however again that's neither here nor there when compared to a far more tighter level of precision required as the 7D requires.

Do you say that DOF (depth of field) is somehow tighter on 7D because of pixel pitch?

yes. depth of field is a measure of image size magnification to observer distance. if you view from a 30D and a 7D both at 100% magnification, you are very much increasing the image magnification, thus decreasing the appearance of DOF.

where what you experienced may be true with the 30D to 1D .. however that's 3 generations older AF on even the XXD series.

I was merely pointing out to my previous poster that blaming the lens' stepper motors is not a way to go. It's all about the AF system of the camera.

Also, I believe that on 7D, the seemingly inaccurate lens would focus well in CD AF mode in live view. Although the mechanics of the lens is the same and it receives the same commands over the same communication protocol.

not even close to being a good hypothesis.

It is. The post I was replying to was blaming lens mechanics for poor AF outcome. But poor lens mechanics would hamper the CD method also.

CDAF is hill climbing so any PEC / backlash errors can be handled by the AF.

Of course. But even CD AF wouldn't be able to find good focus if the stepper motors were soooo terrible.

on TTL-SIR they are not - it's entirely possible for CD-AF to settle at the "best focus" through ititeration .. on a fire and forget AF such as canon TTL-SIR it is not.

Yet, 1D can focus that lowly 50/1.8 better than any xxD. And 1D Mk2 N has also focus priority.

stepper precision can lead to more "imprecise" focus - especially on fast lenses - this would be a fact. the older stepper designs were based upon film tolerances.

Gosh, my 50/1.8 is metal mount Mk1, made 13 years before EOS D30. Yet, it can be focused very well by my 1D Mk2 N which is 18 years younger.

a 7D requires 7 times more accuracy than film tolerances.

How many times more accuracy does APS-H with 8MP need? I'm genuinely curious to know.

compound that by the accuracy of the backlash calculations - can lead to focus being good / bad or indifferent from shot to shot.

then we have something called major defocus events - which are by nature far more imprecise .. and we carry imprecision from the AF calculations .. all the way through .. with each step adding more % of error into the mix.

where it may not be the entire reason - it's very much understandable that it's not helping the scenario.

--
Cheers,
Martin

 

[MM1]

In other words, if the camera can command the lens to attain accurate focus using CD AF, then the lens stepper motor must be capable of the step resolution required to attain accurate focus.

Nope, the lens still is still a possible culprit - if on stopping down the focus shifts, which is an all too common ailment of fast lenses...

Is the lens stopped down when doing CD AF?

--
regards
Karl Günter Wünsch

--
Cheers,
Martin

 

[schmegg]

nope.

because both AF systems don't focus the same. one checks to see if it achieved focus, the other one doesn't.

CDAF is ititerative .. meaning .. it does this.

in focus? nope .. half the distance .. in focus? nope .. half the distance .. rinse and repeat.

stepper motor inaccuray isn't just the finite amount of precision between steps - it's also backlash (the mechanical settling of the gears, etc) .. and pec (periodic error in the drive assembly itself).

all of this is 100% handled by CD-AF without the lens being involved in the process.

with TTL_SIR .. phase detection. it's just going .. move the lens to x distance.

everything else is depending on the lens - if the lens isn't manufacturered with high enough tolerances from stepper, drive chain, gearing - it can be greatly inaccurate as the pixel density increases. in mechanical gearing there's always a certain amount of "slop" .. CD-AF counters that by continually checking the actual image on the sensor .. PD-AF just assumes the damned lens knows what's it's doing

Thanks for sticking with me - I get your point now (finally!, I hear you mutter ;-) )

I was aware of the difference between how the two modes worked but didn't think hard enough about the implications of the chain of tolerances and there effect on the PD method.

Thanks again for getting it through my thick skull

 

[rrcphoto]

30D TTL-SIR really has nothing to do with this thread and nor really can you make a straightforward jump between XXD series AF and 7D AF.

Why can't I? 7D AF looks more like xxD AF on steroids, than 1D AF lite.

two entirely different AF designs? also later designs on the XXD series had major defocus - something the 1D may have calculated out based upon multiple cross types.

Compare the sensors... On xxD, 5D and 7D, the AF points are formed out of single baselines, either one or two for a cross sensor. 1D AF sensor is completely different, where AF points are just segments of several long baselines of the AREA-SIR sensor.

Even if I make 7D sensor a new design, it is more similar to xxD sensor than to 1D sensor.

BTW, what's "major defocus", please?

undoubtably .. the 1D's high precision sensor would be more accurate than the 30D's -

Why? They both have the same "paper" specs.

actually no they don't, and no they weren't. the 30D wasn't a 2.8 cross type at the center. it had "additionally sensitive" sensor regions, but werent' 2.8 cross types as the 1D series had. Slightly different. also I'm completely assuming here that you are only talking about centerpoint accuracy.

Of course I talk about the center point. But someone is wrong here... Either you or (for example) the guys who describe 400D AF on this page (couldn't find 30D description, but 400D should have the same AF):

http://www.imaging-resource.com/PRODS/XTI/XTIA4.HTM

nope, it simply states "higher accuracy with 2.8 or faster" .. the 1 series is 2.8 cross types period. also just because they are the same cross types does not mean the granularity of the sensor is the same - even in between models of the XXD series.

the 40D I believe stated it was "more accurate" than the 2.8 cross of the 30D for instance.

not even close to being a good hypothesis.

It is. The post I was replying to was blaming lens mechanics for poor AF outcome. But poor lens mechanics would hamper the CD method also.

no, and it was explained to you proceeding.

mechanical systems always have error - and slop.

on contrast detect - the AF on the camera body checks that .. and handles it by nature of how the focus works.

on PD, the camera assumes the lens knows what it's doing.

so how can you say that one precludes the other?

 

[MM1]

30D TTL-SIR really has nothing to do with this thread and nor really can you make a straightforward jump between XXD series AF and 7D AF.

Why can't I? 7D AF looks more like xxD AF on steroids, than 1D AF lite.

two entirely different AF designs? also later designs on the XXD series had major defocus - something the 1D may have calculated out based upon multiple cross types.

Compare the sensors... On xxD, 5D and 7D, the AF points are formed out of single baselines, either one or two for a cross sensor. 1D AF sensor is completely different, where AF points are just segments of several long baselines of the AREA-SIR sensor.

Even if I make 7D sensor a new design, it is more similar to xxD sensor than to 1D sensor.

BTW, what's "major defocus", please?

undoubtably .. the 1D's high precision sensor would be more accurate than the 30D's -

Why? They both have the same "paper" specs.

actually no they don't, and no they weren't. the 30D wasn't a 2.8 cross type at the center. it had "additionally sensitive" sensor regions, but werent' 2.8 cross types as the 1D series had. Slightly different. also I'm completely assuming here that you are only talking about centerpoint accuracy.

Of course I talk about the center point. But someone is wrong here... Either you or (for example) the guys who describe 400D AF on this page (couldn't find 30D description, but 400D should have the same AF):

http://www.imaging-resource.com/PRODS/XTI/XTIA4.HTM

nope, it simply states "higher accuracy with 2.8 or faster" .. the 1 series is 2.8 cross types period. also just because they are the same cross types does not mean the granularity of the sensor is the same - even in between models of the XXD series.

the 40D I believe stated it was "more accurate" than the 2.8 cross of the 30D for instance.

Of course, but because of additional X positioned baselines on the central sensor. That is indeed more accurate than 30D's cross type which is high precision in one direction only (but as are the 1D sensors, too).

not even close to being a good hypothesis.

It is. The post I was replying to was blaming lens mechanics for poor AF outcome. But poor lens mechanics would hamper the CD method also.

no, and it was explained to you proceeding.

mechanical systems always have error - and slop.

Just like the lens which correct their position if they overshoot it in the PD AF. Camera AF - open loop, lens motors - closed loop.

on contrast detect - the AF on the camera body checks that .. and handles it by nature of how the focus works.

on PD, the camera assumes the lens knows what it's doing.

I agree with this. Surely, you can expect CD AF operating miscalibrated lenses well, while PD AF fails all the time. But we are talking about design, right? So I assume lens in good shape and calibrated, too.

so how can you say that one precludes the other?

All I said was - the mechanics of the lens is fine enough so that the lens design should not be blamed for AF failures. I am emphasising the word design again.

Worn out or miscalibrated lens could be blamed for AF failures. But saying that 50/1.8 is designed for film era and therefore unable to focus reliably on digital bodies is nonsense.
--
Cheers,
Martin

 

[rrcphoto]

mechanical systems always have error - and slop.

Just like the lens which correct their position if they overshoot it in the PD AF. Camera AF - open loop, lens motors - closed loop.

right .. but do they correct it well enough? CDAF insures they do. PDAF does not.

on contrast detect - the AF on the camera body checks that .. and handles it by nature of how the focus works.

on PD, the camera assumes the lens knows what it's doing.

I agree with this. Surely, you can expect CD AF operating miscalibrated lenses well, while PD AF fails all the time. But we are talking about design, right? So I assume lens in good shape and calibrated, too.

calibrated against what standards? film?

if the lens is told to move to 10.32 feet .. does it move to 10 or 10.3 or 10.32? what's the margin or tolerance of error?

at the time these lenses were developed - plane of focus accuracy was around 1/3 DOF on a 5x7 print.

we expect 1/3 DOF on a 60x40 print with the same mechanical engineering .. see the problem there?

 

[rwbaron]

Surely this imprecision comes from the AF algorithm though, not the ability of the stepper motor itself.

on major defocus? no - and re-read that as well please.

Nope - sorry. I've re-read it but still don't get your drift. I'd like to understand but your not making sense to me.

The way I see it - if the camera is capable of accurately focusing the image using one AF mode with a given lens, but not capable of accurately focusing the image using a different AF mode and the same lens, then the lens is removed from suspicion of imprecision and the AF mode is implicated.

In other words, if the camera can command the lens to attain accurate focus using CD AF, then the lens stepper motor must be capable of the step resolution required to attain accurate focus.

From there, if the same camera and lens is unable to attain accurate focus using PD AF, then the AF algorithm is the only variable and it must be the algorithm, not the lens, that prevents accurate focus acquisition.

I agree with Schmegg. I know someone will tell me I'm wrong but to me there is something very basic that explains the difference. In PD AF the camera drives the lens to a focus point and expects it hits that point. There is no final adjustment or fine tuning. This is what might be referred to as an "open loop" control system. With CD AF the lens is racked back and forth using "feedback" to obtain the best possible result for contrast which is similar to a "closed loop" control system. IMO the difference in the basic operation of the two systems is what accounts for the inaccuracy of the PD AF. The speed at which the PD AF must operate for AI Servo and 8 FPS does not allow for a closed loop system with "feedback". In other words with current technology you can have either speed or accuracy but not both.

Bob
--
http://www.pbase.com/rwbaron

 

[joharis]

Do you say that DOF (depth of field) is somehow tighter on 7D because of pixel pitch?

No, but Depth Of Focus is, and that's what AF performance specs refer to.

We discussed depth of focus earlier. I understand that the focal plane is an optical image inside the camera that matches more or less with the position of the sensor. What has this to do with the pixel pitch of the sensor?

how do you determine what's in focus or not after you click the shutter button? most likely like the rest of us, view the image.

as you increase the image "size" from the sensor size up .. you decrease the circle of confusion. changing the circle of confusion's size .. changes the depth of focus depth as well as the depth of field depth.

so if you view at 100% on a 8.2um pixel pitch and 100% on a 4.2um pixel pitch .. the depth of focus and the depth of field on the 4.2um pixel pitch sensor "appears" narrower.

Circle of confusion is not a physical thing related to the size of a pixel, it's something about the human eye. Sensors can't be confused, human eyses can.

Circle of confusion says that at a given print size and a given viewing distance the human eye considers a circle as a point and so considers the picture as sharp. That circle is an optical phenomenon and for the human eye it really doesn't matter if that circle is represented by 10 pixels or by 100 pixels (at that print size and viewing distance). As long as the magnification factor from sensor size to print size is the same, the circles of confusion in the prints are also the same, there is no relation with the number of pixels and the size of the pixels. That is of course completely different from viewing a picture at 100%. At 100% you see pixels and no depth of field or depth of focus.

Not intended to offense anyone, just trying to get things clear.
--
http://www.flickr.com/photos/joharis/

 

[rrcphoto]

Do you say that DOF (depth of field) is somehow tighter on 7D because of pixel pitch?

No, but Depth Of Focus is, and that's what AF performance specs refer to.

We discussed depth of focus earlier. I understand that the focal plane is an optical image inside the camera that matches more or less with the position of the sensor. What has this to do with the pixel pitch of the sensor?

how do you determine what's in focus or not after you click the shutter button? most likely like the rest of us, view the image.

as you increase the image "size" from the sensor size up .. you decrease the circle of confusion. changing the circle of confusion's size .. changes the depth of focus depth as well as the depth of field depth.

so if you view at 100% on a 8.2um pixel pitch and 100% on a 4.2um pixel pitch .. the depth of focus and the depth of field on the 4.2um pixel pitch sensor "appears" narrower.

Circle of confusion is not a physical thing related to the size of a pixel, it's something about the human eye. Sensors can't be confused, human eyses can.

Circle of confusion says that at a given print size and a given viewing distance the human eye considers a circle as a point and so considers the picture as sharp. That circle is an optical phenomenon and for the human eye it really doesn't matter if that circle is represented by 10 pixels or by 100 pixels (at that print size and viewing distance). As long as the magnification factor from sensor size to print size is the same, the circles of confusion in the prints are also the same, there is no relation with the number of pixels and the size of the pixels. That is of course completely different from viewing a picture at 100%. At 100% you see pixels and no depth of field or depth of focus.

right so what happens to the "print size" when you view an image from a 8mp 1.6 cropped sensor and a 18mp cropped sensor both at 100%?

the magnification changes WRT to final image size .. the more dense sensor at the same sensor size will inheriently change the circle of confusion at 100% magnification with respects to a less dense sensor.

 

[tko]

Spoken from one who grew up in the analog age with knobs and meters and all.

 

[joharis]

Do you say that DOF (depth of field) is somehow tighter on 7D because of pixel pitch?

No, but Depth Of Focus is, and that's what AF performance specs refer to.

We discussed depth of focus earlier. I understand that the focal plane is an optical image inside the camera that matches more or less with the position of the sensor. What has this to do with the pixel pitch of the sensor?

how do you determine what's in focus or not after you click the shutter button? most likely like the rest of us, view the image.

as you increase the image "size" from the sensor size up .. you decrease the circle of confusion. changing the circle of confusion's size .. changes the depth of focus depth as well as the depth of field depth.

so if you view at 100% on a 8.2um pixel pitch and 100% on a 4.2um pixel pitch .. the depth of focus and the depth of field on the 4.2um pixel pitch sensor "appears" narrower.

Circle of confusion is not a physical thing related to the size of a pixel, it's something about the human eye. Sensors can't be confused, human eyses can.

Circle of confusion says that at a given print size and a given viewing distance the human eye considers a circle as a point and so considers the picture as sharp. That circle is an optical phenomenon and for the human eye it really doesn't matter if that circle is represented by 10 pixels or by 100 pixels (at that print size and viewing distance). As long as the magnification factor from sensor size to print size is the same, the circles of confusion in the prints are also the same, there is no relation with the number of pixels and the size of the pixels. That is of course completely different from viewing a picture at 100%. At 100% you see pixels and no depth of field or depth of focus.

right so what happens to the "print size" when you view an image from a 8mp 1.6 cropped sensor and a 18mp cropped sensor both at 100%?

100% of what?

the magnification changes WRT to final image size .. the more dense sensor at the same sensor size will inheriently change the circle of confusion at 100% magnification with respects to a less dense sensor.

--
http://www.flickr.com/photos/joharis/

 

[MM1]

Surely this imprecision comes from the AF algorithm though, not the ability of the stepper motor itself.

on major defocus? no - and re-read that as well please.

Nope - sorry. I've re-read it but still don't get your drift. I'd like to understand but your not making sense to me.

The way I see it - if the camera is capable of accurately focusing the image using one AF mode with a given lens, but not capable of accurately focusing the image using a different AF mode and the same lens, then the lens is removed from suspicion of imprecision and the AF mode is implicated.

In other words, if the camera can command the lens to attain accurate focus using CD AF, then the lens stepper motor must be capable of the step resolution required to attain accurate focus.

From there, if the same camera and lens is unable to attain accurate focus using PD AF, then the AF algorithm is the only variable and it must be the algorithm, not the lens, that prevents accurate focus acquisition.

I agree with Schmegg. I know someone will tell me I'm wrong but to me there is something very basic that explains the difference. In PD AF the camera drives the lens to a focus point and expects it hits that point. There is no final adjustment or fine tuning. This is what might be referred to as an "open loop" control system. With CD AF the lens is racked back and forth using "feedback" to obtain the best possible result for contrast which is similar to a "closed loop" control system. IMO the difference in the basic operation of the two systems is what accounts for the inaccuracy of the PD AF. The speed at which the PD AF must operate for AI Servo and 8 FPS does not allow for a closed loop system with "feedback". In other words with current technology you can have either speed or accuracy but not both.

The problem is you cannot choose. On lower end bodies at least. On my 1D Mk2 N, I can choose focus/release priority and the camera won't release shutter unless it feels it has focus. Yep, it is slower, but I get more keepers.

Bob
--
http://www.pbase.com/rwbaron

--
Cheers,
Martin

 

[Phil214757]

Digital is of course just an agreed reality the where the circuit physics is still playing out in analog.
--
Phil Agur

7D - 5DMKII - Dual Body Bag - Go 'L' and don't look back!

Got 7D? See http://www.usa.canon.com/dlc/controller?act=GetArticleAct&articleID=3167

 

[photo nuts]

the 7D AF sensor is far less sensitive than its older counterparts (such as the one on the 400D/450D/500D/550D/20D/30D. It requires a lot more contrast and light. Sadly, in this regard, it is quite a bit more inferior.

Will upload the photos and expound on this later.

 

[rwbaron]

the 7D AF sensor is far less sensitive than its older counterparts (such as the one on the 400D/450D/500D/550D/20D/30D. It requires a lot more contrast and light. Sadly, in this regard, it is quite a bit more inferior.

Will upload the photos and expound on this later.

This agrees with the results of a simple test I did against my 40 and 50D. The 7D AF system is excellent and highly configurable but it appears that Canon took a step backward in low light sensitivity.

Bob
--
http://www.pbase.com/rwbaron

 

[MM1]

--
Cheers,
Martin

 

[GammyKnee]

the 7D AF sensor is far less sensitive than its older counterparts (such as the one on the 400D/450D/500D/550D/20D/30D. It requires a lot more contrast and light. Sadly, in this regard, it is quite a bit more inferior.

Others have noted the same, e.g.
http://www.imaging-resource.com/PRODS/E7D/E7DIMAGING.HTM
(search for section titled "low light")

In other forums I've also seen the same criticism leveled at the 1D IV.

In real use though I feel (totally unscientific) that my 7D performs a little better in low light than my 50D used to, but this may have more to do with compensation for particular types of indoor lighting etc. Or it may just be wishful thinking on my part!

 

[photo nuts]

The following test shows how sensitive the 7D AF sensor is to color differences.

Camera mounted on tripod. Canon 17-55 f/2.8 IS lens used. IS turned off. Single shot. Center normal AF point, f=35 mm, ISO 200, f/2.8, 1/800 sec

10 shots are fired at the target. The lens is intentionally defocused between shots by turning the lens towards the minimum focusing distance or infinity.

When the AF point rests on the demarcation between the red tiles and yellow walls,



7 out of 10 shots are NOT in focus.

When the shot is out of focus, both the red tiles and trees are out of focus.



so no one can argue the camera focused on the trees. Anyway, at f=35mm and this target distance, the DOF should cover both the trees and red tiles.

The same test when administered on a lowly XSi/450D throws up 10 out of 10 shots which are in PERFECT focus.

Such AF results from an advance camera like the 7D are pathetic.

Some may argue the lens/camera combo is not properly calibrated. To disprove this, the same test is repeated with the AF point resting on the demarcation between the green and yellow portions of the same building.



Result: 10 out of 10 shots in focus (see left portion of above pic).

The test (10 shots fired each time) is repeated several times and the results are always the same.

I only carried out extensive testing 'cos I experienced a number of OOF shots from the 7D during an event shooting.

For all its advance features and superb handling, a camera that cannot AF reliably for simple single shot photography is completely useless.

 

[Hans vdC]

Seems the exact same experience I have with my 7D. Maybe Canon can fix this with a firmware update?

 

[jm67]

So the 7D is colour blind to red as a source for focus? This thread was going in the direction of poor focusing ability in low light but these photos are in fine light. So I suppose in low light with red as your only option to focus on, you'd just be out of luck altogether?