aim for an eye, using an outer AF point... it misses focus most of the time. When I do the same thing, but use the center AF point instead, I get everything in focus most of the time.

I am obviously no technician... but this can't be (entirely) explained by the difference in single axis vs cross type. Because an eye contains no straight lines and has both horizontal and vertical contrast.

Canon explains why in this quote from the 5D (I, not II) white paper, page 9:

Additionally, the center AF point’s f5.6-sensitive, horizontal line-sensitive sensors each have two lines with individual pixels arranged in a zigzag pattern, making a total of four lines for horizontal line-sensitive focusing at the center. With this improvement, the center AF point can now do a better job of detecting difficult-to-read, low-contrast subjects, thus improving the overall performance of the autofocusing system.

brainiac wrote:
noise floor has a large scale criss-cross pattern visible in shadows at low isos


i actually did some testing on this and combined with that done by a few others it seems that the 1dmkiii, 1dsmkiii, 40D generation were the kings of low banding pattern read noise at low ISO and that the 50D, 5DMkII generation is a step backward in that one regard, having as much as 2x the amount of ISO100 banding pattern read noise.

The new CFA filter array also make the chroma noise get a bit of a penalty on the new generation and the very low frequency chroma noise is worse than the prior generation.

The total read noise adjusted for MP difference though does seem a little bit less for the 50D than the 40D at ISO100, but the eye noticed pattern noise a lot more readily.

The 5DMkII, adjusted for MP difference, actually less total read noise at ISO100 than the 1DMkIII too, but almost 2x the pattern banding read noise.

Of course, a more real world comparison would be the DR at ISO100 and for that you also need to take into account the max level they can reach before saturation, which if not the same for the various cameras, would mean the read measurements alone won't tell the entire story)


anyway the 5DMkII is pretty fine camera and it's nothing to get too worried about, although it is a shame they didn't keep the prior generations level of low ISO banding pattern noise, but again nothing to get too worried about and I hear it has less than the old 5D.

in fact is more than pretty fine, it is a truly awesome camera (so long as you don't have to have fast mirror/fps)

skibum5 wrote:
brainiac wrote:
noise floor has a large scale criss-cross pattern visible in shadows at low isos


i actually did some testing on this and combined with that done by a few others it seems that the 1dmkiii, 1dsmkiii, 40D generation were the kings of low banding pattern read noise at low ISO and that the 50D, 5DMkII generation is a step backward in that one regard, having as much as 2x the amount of ISO100 banding pattern read noise.

The new CFA filter array also make the chroma noise get a bit of a penalty on the new generation and the very low frequency chroma noise is worse than the prior generation.

The total read noise adjusted for MP difference though does seem a little bit less for the 50D than the 40D at ISO100, but the eye noticed pattern noise a lot more readily.



The T1i / 500D is just as good as the 40D regarding pattern noise and MP adjusted read noise, and seems a little better than the 50D regarding low frequency chroma noise.

I am obviously no technician... but this can't be (entirely) explained by the difference in single axis vs cross type. Because an eye contains no straight lines and has both horizontal and vertical contrast.

Au contraire. If you draw an eye on a piece of paper, you will find yourself drawing mostly horizontal lines--eyebrow and eyelids--and only very short sections of vertical lines. The eye is predominantly a subject of horizontal linear contrasts.

When you activate one of the far end peripheral sensors and turn the camera vertically for a headshot, then try to focus on an eye, you will have turned the camera to the very orientation that makes it most difficult to focus on the eye.

Turn that camera so that the rectangle is perpendicular to the eyelids and eyebrows, and you'll find the peripheral point is just as accurate as the center point. Sit down with the camera and try it.

The well stated fact that the focus points are too clustered.

They are as far from the center axis of the lens in absolute millimeters as any other DSLR camera. This is a limitation of optical geometry to which all of the cameras are subject.

RDKirk:

Excellently stated!

RDKirk wrote:
I am obviously no technician... but this can't be (entirely) explained by the difference in single axis vs cross type. Because an eye contains no straight lines and has both horizontal and vertical contrast.

Au contraire. If you draw an eye on a piece of paper, you will find yourself drawing mostly horizontal lines--eyebrow and eyelids--and only very short sections of vertical lines. The eye is predominantly a subject of horizontal linear contrasts.

When you activate one of the far end peripheral sensors and turn the camera vertically for a headshot, then try to focus on an eye, you will have turned the camera to the very orientation that makes it most difficult to focus on the eye.

Turn that camera so that the rectangle is perpendicular to the eyelids and eyebrows, and you'll find the peripheral point is just as accurate as the center point. Sit down with the camera and try it.

Yep, I tried it... countless of times... didn't help much.

Look, I appreciate your technical explanations. And I do agree that at least some part of the reduced accuracy of the outer AF points can be explained by the single axis vs cross type argument. But to me it is not the whole story.

The times when the outer AF points are really so much worse than the center one, is:
1) in low light
2) while shooting with wide apertures (in good and worse light)
3) with AI Servo (especially combined with large apertures and in low light)

In these cases the difference between the outer AF points and the center one is too great to be entirely explained by your primairy argument (single axis/cross type).

Also, when focus misses, it ends up at the side/back of the head. Because the outer AF points read more contrast there then at the actual eye? Could be... But when this happens, most of the time the focus ended up under another outer AF point that I didn't use to aim with. Maybe the outer AF points are connected in some way and always search for the area with most contrast (?).

BTW The performance of the outer AF points becomes noticebly better (in comparison to the center one), when there is good light/contrast, when I use smaller apertures and stay away from AI Servo.

Make of this what you will... but I ain't going to use the outer AF points, because they don't work for me in a reliable fashion (like the center AF point does). Whatever the cause...

RDKirk wrote:
The well stated fact that the focus points are too clustered.

They are as far from the center axis of the lens in absolute millimeters as any other DSLR camera. This is a limitation of optical geometry to which all of the cameras are subject.

If you compare the 5D2 AF sensor positions with the one in the 1Ds3... it is obvious that the ones in the 5D2 are akwardly placed... and that there is still some room to spread them out a little further towards the edges of the frame.

Daan B wrote:
RDKirk wrote:
The well stated fact that the focus points are too clustered.

They are as far from the center axis of the lens in absolute millimeters as any other DSLR camera. This is a limitation of optical geometry to which all of the cameras are subject.

If you compare the 5D2 AF sensor positions with the one in the 1Ds3... it is obvious that the ones in the 5D2 are akwardly placed... and that there is still some room to spread them out a little further towards the edges of the frame.

I am pretty sure the 5DMK2 sensors are as far to the edge as the 1DsMK3. The 1DMK3 sensors are closer to the edge since they use the same sensor array in the cropped format. There is a comparison overlay floating around.

jerrykur wrote:
Daan B wrote:
RDKirk wrote:
The well stated fact that the focus points are too clustered.

They are as far from the center axis of the lens in absolute millimeters as any other DSLR camera. This is a limitation of optical geometry to which all of the cameras are subject.

If you compare the 5D2 AF sensor positions with the one in the 1Ds3... it is obvious that the ones in the 5D2 are akwardly placed... and that there is still some room to spread them out a little further towards the edges of the frame.

I am pretty sure the 5DMK2 sensors are as far to the edge as the 1DsMK3. The 1DMK3 sensors are closer to the edge since they use the same sensor array in the cropped format. There is a comparison overlay floating around.


1Ds3 vs 5D sensor AF positions (copied from their respective white papers). Keep in mind that the 5D2 has an even larger VF than the 5D (although these sample pictures probably represent the focus screen at 100%):

Daan B wrote:The most right and left AF sensors are similairly placed in the 1Ds3 and 5D. But the center top and bottom positions aren't. The ones in the 1Ds3 are placed further towards the edges of the frame. The other 5D AF sensors could have been placed more towards the corners as well.

Yes, exactly. It's not the number of points but rather the placement of said points that is problematic, at least for my use. The corner points are especially poorly placed if, for example, you want to pick out an eye shooting a headshot in portrait orientation.

I feel that in landscape mode the left-most and right-most points are well placed, while the others are too close to the center. I would love to see a rectangular layout in a future iteration, which would almost surely move the outer points further out than they are now.

Hi Daan,

So far so good with my 5DmkII. About 12.000 shutters and still counting....

Only production flaws so far:

- exposure metering with flash tends to underexpose
- loose/noisy cf card door

Daan B wrote:

1Ds3 vs 5D sensor AF positions (copied from their respective white papers). Keep in mind that the 5D2 has an even larger VF than the 5D (although these sample pictures probably represent the focus screen at 100%):

Daan B wrote:
Ok, besides missing light shield plates and broken LCD's...

Are there any other problems / issues (minor or big) that I should watch out for when buying a new 5D2?

I am fully recovered from my previous 5D2 experiences (bad ones ) and am ready to try it one more time

Seriously...


Didn't get to read all 4 pages yet but have you pick up another 5D2 to give it another go? if so how do you like it so far? Other than the squeaky CF door after 5000+ clicks I'm still loving mine, Center AF is dead on accurate and nail every single shot with 100% hit rate so far with 85 1.8, 70-200 4.0 IS, and 24-105L.

jerrykur wrote:
If focusing in these corners is an issue for you than I think you should forget a 5DMK2. No matter what else it does right this will negatively color your perception of the camera. Time to move on to a 1DsMK3 or even a 1DMK3 since the focusing points cover even more with the 1.3 crop.

I appreciate your concerns... but I already have a 1Ds3. The 5D2 will perform back-up duty in the studio and on location. And it will be one of the primaries for shooting weddings and events (low light high ISO stuff with little flash involved).

Like I said earlier... I will only be using the 5D2's center AF point. I will either focus-recompose or crop afterwards to compensate.

yauyi wrote:
Didn't get to read all 4 pages yet but have you pick up another 5D2 to give it another go? if so how do you like it so far? Other than the squeaky CF door after 5000+ clicks I'm still loving mine, Center AF is dead on accurate and nail every single shot with 100% hit rate so far with 85 1.8, 70-200 4.0 IS, and 24-105L.

Placed the order yesterday...

Not too thrilled about paying 2250 euros for something that is worth much less IMHO... But it is the only suitable option in the Canon camp right now.

When I had a try with a few 5D2's some months back, I also noticed the very good accuracy of the center AF point. Especially with my 35L... Which I will be using a lot on it

Daan B wrote:
jerrykur wrote:
If focusing in these corners is an issue for you than I think you should forget a 5DMK2. No matter what else it does right this will negatively color your perception of the camera. Time to move on to a 1DsMK3 or even a 1DMK3 since the focusing points cover even more with the 1.3 crop.

I appreciate your concerns... but I already have a 1Ds3. The 5D2 will perform back-up duty in the studio and on location. And it will be one of the primaries for shooting weddings and events (low light high ISO stuff with little flash involved).

Like I said earlier... I will only be using the 5D2's center AF point. I will either focus-recompose or crop afterwards to compensate.

Sorry, I missed your willingness to use focus and recompose. I hope you get a better camera this time. You have had the worse luck.

Daan B wrote:
Not too thrilled about paying 2250 euros for something that is worth much less IMHO... But it is the only suitable option in the Canon camp right now.


I would wait til xmas

Daan B wrote:
Not too thrilled about paying 2250 euros for something that is worth much less IMHO...

There you go, I share that view......and because of that very view I have not bought the 5D MkII. Hell, I have never bought the 5D MkI either.......and still consider it overpriced these days.

(For FF, I've gone to 1Ds MkII only very recently.)

Daan B wrote:
dhphoto wrote:
No problems with my 5DII at all. Love it.

Obviously it has an old and rather inferior focusing system, but you know that.

Canon quality control is just not good enough though.

David

Good to hear that David

Yeah, to by-pass the inferior AF I will only use the center AF point and crop to my liking afterwards. That should do the trick... especially since the 5Ds has 21MP's to start with

I will be using the 5D2 mainly for high ISO reportage/wedding work.

I think you will be glad you gave it another chance. It does high iso quite well.

Daan B wrote:
yauyi wrote:
Didn't get to read all 4 pages yet but have you pick up another 5D2 to give it another go? if so how do you like it so far? Other than the squeaky CF door after 5000+ clicks I'm still loving mine, Center AF is dead on accurate and nail every single shot with 100% hit rate so far with 85 1.8, 70-200 4.0 IS, and 24-105L.

Placed the order yesterday...

Not too thrilled about paying 2250 euros for something that is worth much less IMHO... But it is the only suitable option in the Canon camp right now.

When I had a try with a few 5D2's some months back, I also noticed the very good accuracy of the center AF point. Especially with my 35L... Which I will be using a lot on it


I paid less for my 5DII than I did for my original 5D and couldn't be happier. I think it is a fantastic camera

David

Daan B wrote:

The times when the outer AF points are really so much worse than the center one, is:
1) in low light
2) while shooting with wide apertures (in good and worse light)
3) with AI Servo (especially combined with large apertures and in low light)

Maybe you already know this, but I came across this post by Chuck Westfall in his Tech Tips column on TheDigitalJournalist site a couple days ago. It discussed the Mark III but seems it should also apply to other Canon cameras:

http://digitaljournalist.org/issue0905/tech-tips.html

Q: In poor lighting, say an exposure of f/1.2, 1/15 to 1/30 second, at ISO 800 in evaluative metering with an EOS-1D Mark III, selecting the center AF point manually, I'm observing a fairly significant difference in the AF sensitivity between One-Shot AF and AI Servo AF. In AI Servo mode, on an object with relatively decent contrast, the 1D Mark III is unable to lock focus. When I switch to One-Shot AF, the center AF point quite accurately focuses on the subject and gives a confirmation beep and I am able to capture the frame with extremely sharp accuracy handheld. It is almost as if in AI Servo mode, the AF sensitivity has suddenly switched "profiles" to a rather less sensitive "mode," so as to not be able to lock onto the subject in a fairly low-light situation, whereas One-Shot AF mode has absolutely no issues with that same situation and subject. Does my camera need to go back to Service or is the camera performing according to its design?

A: The light level you describe (ISO 800, 1/15 at f/1.2) is close to the threshold of the EOS-1D Mark III's low-light AF sensitivity. Without the use of flash under these lighting conditions, it would be very difficult to obtain sharp photos at f/1.2 unless the camera was steadied through use of a tripod, and also the camera's reflex mirror should be locked prior to exposure. Assuming that you're prepared to take those considerations into account, then your observation about the difference in focusing capability between One-Shot AF and AI Servo AF for the EOS-1D Mark III is correct. Focus will lock on a subject with readable contrast in One-Shot AF, but AF will fail under the same lighting conditions in AI Servo AF. That is completely normal and to be expected. It's due to differences in the amount of time that light is allowed to accumulate on each pixel in the AF sensor. That period of time, which Canon does not disclose, is longer for One-Shot AF than it is for AI Servo AF, and the result is superior low-light performance for One-Shot AF. This is essentially the performance level that the EOS-1D Mark III is designed to offer in extremely low light, so there would be no benefit in sending the camera in for service on this issue.