p.15 #1 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
Wahoowa wrote:
I have yet had any explanation for it, but I would like to suggest you to do one more experiment. Print out both targets and try them. I might have an explanation afterward. Cheers.
I already have. The results remain consistent. I think this (and the post I just made about horiz vs. vertical) is the main reason my results have been all over the place. It seems to depend on whether the AF sensor is picking up the horizontal or vertical info in any given target.
p.15 #2 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
jmvdigital wrote:
Alright, here is a good riddle and cautionary note for anyone working on tuning their lenses. I have tested all manner of print targets, lighting, and various setups from indoors to outdoors. The photos below are from a simple test I did using my laptop screen as the target. Both are targets I made up in Illustrator really fast, and both, in theory, should work equally as well on dual-cross center point AF sensor in the 5D3.
The problem is that these two targets consistently test at very different tune values (i.e., the AF system reacts very differently in reading them). In the top image (with the cross-hatch grid lines), the calculated tune value works out to be around a -2. In the bottom image (with the checker board), the calculated tune value work out to be around a +8.
I shot manually focused images of both targets (focus is correct in viewing the RAW files) as a control. Then I used the PDAF system to give me focus confirmation while I focused the lens from both sides of center (one shot focusing from near->far, one shot focusing from far->near). Both gave consistent results of the near focus confirmation being slight in front of the target, and the far focus confirmation being slightly behind the target... with the mid-point between the two being perfect focus. ...Show more →
Did you reacquire manual focus for each unique on-screen target? If so, try using the same established focus for both on-screen targets, to isolate the test to just the focus detection logic and eliminate focusing differences from the equation.
p.15 #4 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
jmvdigital wrote:
A little bit more testing using a sheet with half white, half black with a solid. With the vertical line, the tune value is +7. With a horizontal line, the tune value is a -1.5.
What gives? Obviously, the AF sensor is reacting differently to horizontal and vertical information. My question is, is this normal? Anyone else with a 5D3 experiencing this? My camera is new... wondering if I should exchange it.
EDIT: The focus difference appears to be the largest with my 24mm TS-E. With the 14L, 35L and 100L, the difference is about 2-3 points instead of 5-6.
I wouldn't rely on a single line of contrast as an AF target. Try repeating the same test with the same contrast but with multiple lines in each direction.
I have my D4 set up to use the AF-ON as the usual focus mode, so during the Dot-Tune testing I kept the camera on AF and used the half shutter to activate the focus light.
For my test I am using a Sigma 24mm f/1.8 placed roughly 4 feet from the camera, everything on sturdy tripods and at right angles to each other. I used live view and manual focus to achieve what I believe is the best focus, and didn't touch it after.
I tried this test with the body set to AF, and (both cases) of the lens set to MF and AF because it has one of those clutch mechanism focus rings.
For all -20 to +20 values I did not achieve a single solid green dot. Almost all values with the exception of a few scattered points produced equal blinking left and right arrows.
p.15 #6 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
Interference star charts are awesome for manually focusing, but are not ideal for AF. The center of that chart would likely just appear as a grey blur to your AF point. I have confirmed this previously with my Siemens star chart (it at least has a solid black circle with surround white ring that helps AF lock on a bit better, but still not ideal).
p.15 #8 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
snapsy wrote:
I wouldn't rely on a single line of contrast as an AF target. Try repeating the same test with the same contrast but with multiple lines in each direction.
I repeated the test with a multi-bar target for both horizontal only and vertical only. The results changed very little overall.
The 14L stayed within a 2.5-3 point difference between the single bar, multi-bar target.
The 35L stayed within 1 point difference.
The 24 TS-E remained the highest with a 6-6.5 point difference between horizontal and vertical bar targets.
The 70-200 2.8L IS II shows a 1.5 point difference between horiz. and vert. bars.
In almost all cases the vertical bars produce a +1.5 to +6.5 bigger tune value than a horizontal target.
p.15 #9 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
For anyone wanting to try this horiz vs. vertical test on their own, here are the targets:
Horizontal & Vertical Bars PDF (on opening, both are visible, use the layers panel in Acrobat to turn off/on either so you only see one set of bars): http://www.jmvdigital.com/temp/Horz_vert_bars.pdf
Lars Johnsson Offline • • • • • • • Upload & Sell: Off
p.15 #10 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
A single line is not good for testing the AF. Most AF points are either a single line or two lines making a cross. So if the single line is in the same direction as your test line, it will not be good at all. And a cross line will nearly only use one of the lines. And only have one point with contrast
p.15 #13 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
Lars Johnsson wrote:
If it's the screen or monitor I would not use those either.
Regardless though, if the screen method were inaccurate, it should be inaccurate to both targets. Why would there be a consistent and repeatable difference?
p.15 #15 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
MA is really easy to set if you have a split focus screen.
Apr 19, 2013 at 03:53 PM
Lars Johnsson Offline • • • • • • • Upload & Sell: Off
p.15 #16 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
jmvdigital wrote:
Regardless though, if the screen method were inaccurate, it should be inaccurate to both targets. Why would there be a consistent and repeatable difference?
I don't know why you are getting it. I and probably most other people here are not getting the same difference as you.
p.15 #17 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
Well, I went out and got a second 5D3 body, and I can definitely say that both are exhibiting the same thing. The second body tested almost exactly inline with my original body (slightly different exact numbers, but same overall performance).
More or less, vertical bars cause the AF system to focus significantly different than horizontal bars. The exact amount depends on the lens, distance, and target characteristics.
All of my lenses exhibit further focusing tendency with the vertical bars than horizontal. The only exception appears to be my 100L, in which the horizontal bars cause both bodies to focus further back than the vertical bars. No explanation for that, but it consistent across both bodies.
p.15 #18 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
jmvdigital wrote:
Regardless though, if the screen method were inaccurate, it should be inaccurate to both targets. Why would there be a consistent and repeatable difference?
On a monitor? It's because the moire or any signal interference causes the AF to behave differently. That's my thought.
p.15 #20 · Dot-Tune: Autofocus Fine Tuning in under 5 minutes
jmvdigital wrote:
Well, I went out and got a second 5D3 body, and I can definitely say that both are exhibiting the same thing. The second body tested almost exactly inline with my original body (slightly different exact numbers, but same overall performance).
I got around to testing this today. On my 5DM3 + 50mm f/1.4 at a focus distance of 1600mm (32x) using indirect outdoor daytime natural lighting, MF once in LV and not altered between targets I get the following results from the ML DotTune (3 runs with each target):
Bob Atkin's target, focus on the intersection of the 2.2 horz/vert lines: 3,3,4
jmvdigital's horz+vert target at intersection of cluster of lines: 4,4,4
jmvdigital's checkerboard target: 1,2,2
Of all the targets the best in terms of AF confirmation consistency and speed was jmvdigital's horz+vert target, followed closely by the Atkin's target. The worst by far was the checkerboard target - I saw some huge individual AF confirmation outliers with that, although they averaged out with the ML Dot Tune. I didn't try the vert vs horz because in practice I would never rely on only one direction of detail for tuning anyway, even on a cross-point AF sensor.
