5D3: AFMA variations between different AF points

[s2kdriver80]

Before a recent wedding shoot (wasn't the main photographer), I had calibrated my lenses and bodies, including a 5D3 and the 24-105 lens with the center point. In most of my shots, there is a slight front focus in each of them. In upper body and head shots, the focus is good or the OOF is very close to the intended point of focus so the photos are acceptable. For full body shots (where longer camera to subject distances exaggerate AF misfocus errors), most of the photos are unacceptable. AF'ing on someone's face rendered focus on the hands resting in front of them, or their shoulder if they were standing at an angle.

Realizing that I used the outer cross points most of the time, thought had occurred to me to try comparing the AFMA values for the center (which I had already calibrated correctly) and the outer cross points I'd used. Using Reikan's FoCal tool, I consistently get around a -3 AFMA setting difference when comparing the center to a couple of outer cross points, which makes sense since they were causing front focus. Then using FoCal's multi point focus analysis tool, the results show those outer points around 75% of the focus quality of the center point (when using an AFMA value calibrated for the center point).

Since AFMA adjustments are global and can't be adjusted per AF point by us, is this something that a Canon repair facility can fix? OOF from an f/4 lens was noticeable, it would have been even worse had I used a faster lens. I don't mind if all of the AF points as a whole were off 5 or 10 AFMA settings to a lens, since I can adjust. But if there are AFMA variations between different AF points, how can some points be trusted? Has anyone had a similar experience where different AF points behaved differently? The AF systems in the 1DX and the 5D3 are supposed to be Canon's best and the costs reflect, am I unreasonable in expecting a tighter tolerance?

[nicksan]

The outer points on my 5D3 aren't as accurate as the center cluster points. I only use crosstype points BTW. I found this to be the same on Nikon bodies like the D3, D3s, and D700.

[s2kdriver80]

Originally Posted by nicksan

The outer points on my 5D3 aren't as accurate as the center cluster points. I only use crosstype points BTW. I found this to be the same on Nikon bodies like the D3, D3s, and D700.

Perhaps, but I was using the outer cross types (the second to last columns).

And also, the single cross and the linear types ideally should have similar accuracy to the double crosses, just that they aren't as "precise", meaning higher standard deviations around the same value, that is, wider variations, but similar averages over the long run. The outer crosses I've used pretty much front focus all the time.

[mco_970]

I am really interested in this topic. I had a similar experience with my new 5D3. I ma'd my 35L using center point, then took it to a night theater event (Rocky Horror, YEAH!). Many of my pics using the outer points were consistently and badly front focused.

I guess I need to figure out how to test the outer points using FoCal. Can you elaborate on how you tested them?

After the event, I did re-calibrate and I ran the focus consistency tool for several MA values, and had FoCal take a LOT of photos at each point. Results were +2 MA points ahead of my initial FoCal test results, so that certainly may be part of the issue that I had.

ETA: I am set to only use cross type points on my 5D3 but I do use those outside of the center cluster.

[nicksan]

Just in case I wasn't clear, I only have crosstype points enabled and I find that the outer most cluster is definitely not as accurate as the center cluster points. This may not matter on a lens like the 70-200 f2.8 MKII, but for finicky lenses like the 50L and 24L MKII, I usually try to avoid the outer most crosstype cluster. Oddly enough, they do better in AI Servo mode.

[s2kdriver80]

Originally Posted by mco_970

I am really interested in this topic. I had a similar experience with my new 5D3. I ma'd my 35L using center point, then took it to a night theater event (Rocky Horror, YEAH!). Many of my pics using the outer points were consistently and badly front focused.

I guess I need to figure out how to test the outer points using FoCal. Can you elaborate on how you tested them?

After the event, I did re-calibrate and I ran the focus consistency tool for several MA values, and had FoCal take a LOT of photos at each point. Results were +2 MA points ahead of my initial FoCal test results, so that certainly may be part of the issue that I had.

ETA: I am set to only use cross type points on my 5D3 but I do use those outside of the center cluster.

Not sure what version of FoCal you have, but I have Pro. In the settings, you need to enable the feature to allow non-center AF points. Then, just go to the fully automated test (not really fully automated for our 5D3s since we have to manually change the AFMA settings during the test), do the target setup (enable focus points and choose an outer point and do target search), and run the test. I would do the test 2-3 times to make sure the results are consistent.

I also ran the "multi point focus test" under the tools menu. I had the body preset to the AFMA value that was calibrated for the center point, left the first AFMA box unchecked, but checked the next check box set to 3%. Do the target search to make sure all of the red areas are within the target area (this target is different than the one used for the fully automated test). Then run. I did the test twice and each iteration takes a while. What I've found is that the middle third (the double crosses) are all green meaning good, the left third are not as good but still good, like in the .8x scores, and the right third are slightly worse, like in the .7x or .8x scores. The second to last columns (the double crosses) are not better than the last columns (the linear dashes).

For the fully automated test, ran it from 50x focal length distance. For the multi point test, ran it from about 4 feet. Any farther, and I wouldn't be able to fit the AF points onto the target. Camera set on tripod, perpendicular to target, 12 EV brightness of constant halogen lighting. As for your 35mm, don't know if the field curvature issue would come into play and distort results. I did my tests at 105mm wide open.

[mco_970]

Here's a sample I put together - center point versus outer point. ALL of my outer point shots from that sequence were that far off (pretty much useless, in my book). From all the praise for the 5D3, I was expecting better...

[s2kdriver80]

Originally Posted by nicksan

Just in case I wasn't clear, I only have crosstype points enabled and I find that the outer most cluster is definitely not as accurate as the center cluster points. This may not matter on a lens like the 70-200 f2.8 MKII, but for finicky lenses like the 50L and 24L MKII, I usually try to avoid the outer most crosstype cluster. Oddly enough, they do better in AI Servo mode.

Yes, according to my FoCal testing using the multi focus point tool, the middle third was rated the best, followed by the outer thirds (the left did a little better).

[mco_970]

Thanks for the info. I do have the Pro version. I need to do an errand, but hopefully I can set this up in an hour or so and test it.

I bought my camera specifically for outer point AF performance, I think I would have done just as well or better at this event with manual focus & my good old 5D2 with the matte screen... I shot one night of the event with my NEX camera & 24/1.8 lens for it, and while it was slow, almost all the images were tack sharp. By comparison, I shot more images with the 5D3, but had a much lower keep rate.

[s2kdriver80]

Originally Posted by mco_970

Thanks for the info. I do have the Pro version. I need to do an errand, but hopefully I can set this up in an hour or so and test it.

I bought my camera specifically for outer point AF performance, I think I would have done just as well or better at this event with manual focus & my good old 5D2 with the matte screen... I shot one night of the event with my NEX camera & 24/1.8 lens for it, and while it was slow, almost all the images were tack sharp. By comparison, I shot more images with the 5D3, but had a much lower keep rate.

No problem. Just wondering, did you discover these outer AF point problems with longer lenses as well, or just with the 35L?

[mco_970]

Originally Posted by s2kdriver80

No problem. Just wondering, did you discover these outer AF point problems with longer lenses as well, or just with the 35L?

The only long lens I have used with 5D3 so far is a new 70-300L, which is actually pretty outstanding with the outer points. However, it's also one of the new lenses that is supposed to have more accurate focusing with the 5D3 (according to LensRental's testing).

As such, I'm really interested to do more testing with the 40 STM lens, which is supposed to also be super accurate on 5D3 according to LensRentals testing. I wondered if the outer points are really good with it.

I have used 50/1.4 a bit on 5D3 and am not impressed with the outer points, but I don't think highly of that lens anyway, so... I can't blame the camera for anything the 50/1.4 does wrong.

Next low light event, I was thinking of taking the pancake, cranking up the ISO a bit, and seeing how it fares.

I should be able to setup FoCal testing here soon. Interested to see the results.

[mco_970]

Ok, you just knocked the wind out of me.

Center point MA for my 5D3 + 35L is +6
Outer point MA is +13

Tested with a good 10 EV's on a heavy duty Gitzo w/ ballhead.

How will that ever work with a f/1.4 lens Damndamndamndamndamn.

ETA running the consistency test w/ MA = 6, 10, and 14, taking 10 shots per point. Interested to see what the graph curve looks like here.

[mco_970]

Yup, the AF Consistency test totally backs it up. If I want to use outer points, my MA will be off by about +8 points from the center point! FWIW, I am using the nearest AF point to center in the outer two columns of AF points.

[s2kdriver80]

http://photography-on-the.net/forum/....php?t=1172486
http://www.fredmiranda.com/forum/topic/1104180

Your case could be a field curvature issue most apparent using wide angles and outer points. Can you meticulously run the center-outer test using a longer lens and let us know what you find?

[mco_970]

Yeah, I can do 100L. I have already MA'd it using centerpoint, so I should be able to just test an outer point for comparison.

I am going to PM Invertalon and see if there are any updates.