maxim_me wrote:
if this is accurate, why is it that when i tested with "larger" AF area ( those protroding from the markings, i found that the AF did not respond
ie, as i slowly move towards a contrasty object, the AF did not latch on until the boundary is within the markings
Try testing this "larger" AF area, and you will know what i mean
Test for its effect on AF
If I'm not mistaken (and I often am), the idea of having the indicator boxes smaller than the actual focus sensors is to allow for tolerances - that is, the actual sensor may not be perfectly aligned with the focus indicator. This makes some sense given that the viewfinder and AF light paths are separated at the main mirror. Any small angular difference between actual and ideal mirror alignment with either the main or secondary mirror will have a rather significant effect on where the actual focus sensors line up relative to the indicators.
Also note that this image is based on the 10D - I'm not sure how well it correlates to different models like the 20D and 1-series.
Finally, as you've already read, RDKirk tests the sensor positioning/size by allowing only one focusable point in "a sea of unfocusable blandness". Haven't tried the test myself, but I may give it a little testing if I get the time.
RDKirk wrote:
When you attach a lens with a maxium aperture of f2.8 or greater, the camera turns on a second vertical line of phase detection pixels in the center sensor pixel array.
This is what Canon calls the "high precision" mode and supposedly will place the actual plane of focus within 1/3 of the depth of focus (the level of accuracy at the sensor plane).
It appears to me that switching on the additional sensor is a function of the identification information the lens provides to the camera ("My maximum aperture is f2.8") rather than the amount of light being transmitted by the lens. I believe it applies to any shooting aperture used while that lens is mounted--the camera always uses the maximum aperture depth of focus factor for all its focus solutions, not the shooting aperture.
I don't have a document stating this outright, but looking at the chain of events depicted in Canon's patent application, the camera calculates the focus solution for a shot before it calculates the exposure solution (this is definitely true when using evaluative mode autoexposure), so when it sends the movement command to the lens, it doesn't yet know what the shooting aperture will be....Show more →
Just to put a point on this - its worth spending the extra money for f/2, f/1.8, f/1.4, and even f/1.2 max aperture lenses for use with the 20D - not just for low light situations, but even when I shoot at, say, f/8.
Just to put a point on this - its worth spending the extra money for f/2, f/1.8, f/1.4, and even f/1.2 max aperture lenses for use with the 20D - not just for low light situations, but even when I shoot at, say, f/8.
Is that right?
Logically it would seem so.
However, perusing the complaints at DPReview, a possible problem has occured to me. A number of people are complaining that their 20D cameras focus perfectly well with lenses that have a maximum aperture of 2.8, but poorly with faster lenses.
Now, it could be possible that the 20D computer simply isn't good enough to keep within the tolerances, but some of these people have said that their 300D cameras do better.
I'm wondering if a possible firmware bug might involve a camera not "getting the message right" when it first interrogates the lens for maximum aperture. If the lens says "My maximum aperture is f1.4," but the camera erroneously fails to poke that info into the right memory location (thus, not overwriting the maximum aperture recorded from the previously mounted lens), then it's doing its calculations with the wrong number. But put an f2.8 lens on it, and it's fine...it would also work fine with a slower lens.
If the Canon techs are using an f2.8 tool lens (I don't know what kind of lens they use), then they won't see that problem.
However, perusing the complaints at DPReview, a possible problem has occured to me. A number of people are complaining that their 20D cameras focus perfectly well with lenses that have a maximum aperture of 2.8, but poorly with faster lenses.
Now, it could be possible that the 20D computer simply isn't good enough to keep within the tolerances, but some of these people have said that their 300D cameras do better.
I'm wondering if a possible firmware bug might involve a camera not "getting the message right" when it first interrogates the lens for maximum aperture. If the lens says "My maximum aperture is f1.4," but the camera erroneously fails to poke that info into the right memory location (thus, not overwriting the maximum aperture recorded from the previously mounted lens), then it's doing its calculations with the wrong number. But put an f2.8 lens on it, and it's fine...it would also work fine with a slower lens.
If the Canon techs are using an f2.8 tool lens (I don't know what kind of lens they use), then they won't see that problem....Show more →
You make a good point, but i highly doubt this is the case, because if the camera wasnt communicating the lenses maximum aperture correctly, then you could assume the camera also wouldnt let you dial the aperture down past 2.8, to say, 1.4, or whatever it would be.
I read somewhere that it was a good idea to turn the camera off before changing lenses so that the the CPU's could reset to the new data presented by the new lens. I'm not sure that makes any sense, but it seems like a good idea anyway.
Wonotch wrote:
You make a good point, but i highly doubt this is the case, because if the camera wasnt communicating the lenses maximum aperture correctly, then you could assume the camera also wouldnt let you dial the aperture down past 2.8, to say, 1.4, or whatever it would be.
You could be right, however the communication of lens characteristics does appear to continue beyond the initial interrogation, as noted by the fact that variable aperture zoom lenses appear to communicate their changing apertures to the camera (or they communicate the changing focal length and the camera calculates the changing aperture, which still amounts to continuing communication).
Wonotch wrote:
You make a good point, but i highly doubt this is the case, because if the camera wasnt communicating the lenses maximum aperture correctly, then you could assume the camera also wouldnt let you dial the aperture down past 2.8, to say, 1.4, or whatever it would be.
Wonotch,
RDKirk was referring to a firmware bug in the AF code section, and not necessarily to a camera to lens communication problem. His example [not overwriting the previous lens data, as it ought], is just one possible bug scenario for the AF code section not getting the message.
RDKirk,
My take on the perception of service techs to focus accuracy issues is that mechanical shot-to-shot variance in the USM system is well outside of 1/3 DOF. Since system is open loop, techs would not interpret small excursions past 1/3 DOF as camera calculation error, but lens shot-to-shot mechanical variance.
My take on the perception of service techs to focus accuracy issues is that mechanical shot-to-shot variance in the USM system is well outside of 1/3 DOF. Since system is open loop, techs would not interpret small excursions past 1/3 DOF as camera calculation error, but lens shot-to-shot mechanical variance.
If I understand you correctly, I think there should be minimal shot-to-shot mechanical variance because the lens drive process is closed-loop to the extent that the lens monitors itself to ensure it followed the command from the camera and makes corrections as necessary.
That is also my understanding, that the lens focus group position is monitored after completion of the move command. I assume the tolerances of that system is what results in shot-to-shot focus variance. The only real info from Canon techs that I got was that they didn't expect 1/3 DOF due to mechanical variations. The rest is my inferrences.
jmaio wrote:
I read somewhere that it was a good idea to turn the camera off before changing lenses so that the the CPU's could reset to the new data presented by the new lens. I'm not sure that makes any sense, but it seems like a good idea anyway.
It's not necessary to turn the camera off (which would be intolerable to most professionals).
The camera interrogates the lens whenever a new lens is mounted. In fact, I believe a consistent problem I have with a Tokina f2.8 80-200 zoom is specifically because the 20D interrogates the lens even when it's "asleep"--apparently the Tokina is programmed to react first to some standing signal sent when the camera is awake--lacking that signal when the camera is asleep, the lens won't respond to interrogation.
i use the 1dmII,, and have often wondered and asked other users what mode they use as far as moving subjects,, i.e.,,, say a bear that is just moseying around, no particuluar speed, or an animal that is stil now, but could bolt or move off at any time,,i know my non moving focus is fine, but i seem to get a way more oof shots on small movements, as types of movement menttioned above,, the oof does not appear to be user movemtn, just softness,,while it appears to get caught between focus lock hunting movements,,
You make a good point, but i highly doubt this is the case, because if the camera wasnt communicating the lenses maximum aperture correctly, then you could assume the camera also wouldnt let you dial the aperture down past 2.8, to say, 1.4, or whatever it would be.
Different memory registers. It stores the maximum aperture somewhere and uses that constantly for CoC calculations. As far as I can tell, the working aperture is only used for exposure calculation.
jmaio wrote:
I read somewhere that it was a good idea to turn the camera off before changing lenses so that the the CPU's could reset to the new data presented by the new lens. I'm not sure that makes any sense, but it seems like a good idea anyway.
I don't think that makes a difference with the 20D (and probably not with the 350XT, either). I have a Tokina 80-200 that is balky about communicating with the camera if I attach it when the camera is in sleep mode. Once it's attached, no manipulation of the on/off switch will re-interrogate the lens. I have to either remove and remount the lens or remove the battery from the camera. As far as I can tell, the camera interrogates the lens whether it's on or off, just as it reads a new CF card whether it's on or off.
maxim_me wrote:
if this is accurate, why is it that when i tested with "larger" AF area ( those protroding from the markings, i found that the AF did not respond
ie, as i slowly move towards a contrasty object, the AF did not latch on until the boundary is within the markings
Try testing this "larger" AF area, and you will know what i mean
Test for its effect on AF
As the subject goes out of focus, it's blurred image on the AF sensor expands (just as any small blurred subject would on a print).
It seems plausible to me that these AF sensors might be three times the dimensions of the AF box so that they can capture the complete blurred image of a feature which would resolve to the size of the box when properly in focus.
Does anyone know more about the "phase-shift" detection algorithm used by Canon? Does it require a full line-image of the convolved (blurred) feature in order to accurately compute the phase-shift?
If you look at the rectangles in the viewfinder, you'll see that the top and bottom are horizontal rectangles, the center is a square and the rest (the three on the left and the three on the right) are vertical rectanges.
The shape of the rectangle represents the alignment of the sensor. The horizontal rectangles (top and bottom) represent a horizontal sensors which focus best on vertical lines.
The vertical rectangles (three on left and three on right) represent vertical sensors which focus best on horizontal lines.
The center square uses both sensors (depending on the aperature of the lens, as described earlier in the thread).
This arrangement makes sense to me, as the sensors along the center horizontal line are aligned to focus on the horizon or other horizontal features and the sensors along the center vertical line are aligned to focus on vertical features (people, poles, trees, tall buildings, etc.)
You can easily test each AF sensor by setting it as the active AF point and then aiming the camera at a horizontal or vertical boundary between two solid evenly-lit colors and trying to focus. You can even use the windows on your computer monitor.
That Sigma said that Canons AF stop at "the first part of focus" does not correspond with what Ive seen:
I recently got a Sigma 28-70mm 2.8 EX DG and a Sigma 70-300mm APO DG for my 350D. I jumped right in doing the ol' AF-test with chart, tripod and IR-remote. I took shots at apertures 2.8 up to 8.0 at 28, 38 and 70mm. Manual "defocusing" between each shot, both to infitity and "near-focus".
Distance to chart about 40cm (33cm is the minimum for the 28-70).
I pretty much instantly saw that the lenses didnt focus properly with AF on my camera. The focus locked on quite fast and without chasing (but with some noise as expected). But it is as if the AF-motor didn't stop fast enough and focus ran past the correct point. Eg, when starting from infinity I always ended up frontfocused about 1.5-2cm at ALL apertures. When starting from "near-focus" I always ended up back-focused about 1.5cm at ALL apertures.
This clearly results in a soft image in the plane where the focus shold be.
I havent tested the 70-300 as thoroughly (yet) but it seems to exhibit the same problems but, help me with the math here, 1.5-2cm error from 2m distance, at ~200mm 4.5 maybe isn't very much?
What do you think I should do? Is there a fix for this, some sort of calibration? Im thinking I should do this test with the kit-lens as well, as it might be the camera that's faulty.
Im grateful for any thoughts and suggestions.
Aha! That image from Japan fills in a puzzle piece. I'll have to fool with the babelfish translation later, but that image gives an important datum.
Notice the color coding of the sensors and the legend.
At f5.6 (blue colored on the diagram), we see the two linear pixel sensors of each array (a pair of sensors make up an array to sense the phase difference for a line of contrast). So we see two sensors arranged horizontally (the array that senses vertical lines) and two sensors arranged vertically (the array that senses horizontal lines).
At f2.8 (both green and red on the diagram) we see the same two vertically aligned sensors active, but a different set of horiztonally aligned sensors have been activated. These are farther apart, making the baseline wider, I'd suspect to provide the greater precision Canon claims with f2.8 lenses.
This is different from the way I understood it, but it does correspond to Rob Galbraith's description in his review of the 20D.
The photgraph of the sensor itself creates a lot more questions, though. I'm wondering about which sensors are which (I sure picked a lousy time to give up learning Japanese).
This is my guess at the sensors that aren't marked in the drawing:
Haven't tried the test myself, but I may give it a little testing if I get the time.