2
\$\begingroup\$

I'm using an EOS 750D with an EF-S 18-55mm IS STM lens. I often have trouble getting accurate focus using the viewfinder autofocus. Live view autofocus works fine.

Here are two crops of the same scene, the first is taken with live view autofocus, the second one using viewfinder autofocus. In both cases, I used point focusing on the Nine Men's Morris board on the right, which is the center of the full image. The images were taken on a tripod in manual mode with the exact same settings (250ms exposure, f/3.5, 18mm, ISO 100). I manually defocused the lens before taking each picture to make sure the autofocus system actually did something. Live View Viewfinder

You can see that the focus in the second image is significantly in front of the target. The curtain on the left is in much better focus than the board, which was the actual autofocus target. The curtain is about 1.5m from the camera, the board is about 3m from the camera, so the focus error seems quite significant to me, relatively speaking. Sometimes the focus is a lot better, but it's never quite perfect, and as this example shows, it varies quite a lot. About 2/3 of the shots are this severely out of focus.

This seems to happen with all sorts of targets and also with other autofocus points than the center one. However, it only happens at 18mm. If I zoom to 55mm, the viewfinder autofocus is spot on. It also doesn't happen with any of my other lenses (EF-S 55-250mm IS STM, EF 50mm STM), although the EF-S 18-55mm IS STM lens doesn't have issues at these focal lengths either.

Any ideas what could be causing this? Am I doing something wrong? Should I send back my camera and/or lens? Or am I simply expecting too much from the autofocus system (although 1.5m focus offset on a 3m target seems quite a lot to me)?

At this point, I'm ready to send my camera kit back, unless this is the sort of performance that is expected from viewfinder autofocus.

The thread Why is my camera focusing fine in liveview but getting it wrong with the viewfinder? does not really answer my question. I already know the possible reasons for why PDAF might be less accurate than CDAF. What I would like to know is if the severity in my example shots is normal and why it only happens at 18mm. It would be nice if someone with a similar camera could report on their experiences.

\$\endgroup\$
8
  • \$\begingroup\$ Just to check: could you confirm that you are manually selecting the focus point you want to use, and are in one of P, Tv, Av or M mode? \$\endgroup\$
    – Philip Kendall
    Dec 30, 2015 at 12:53
  • 3
    \$\begingroup\$ One more thought: which focus mode are you using (One Shot, AI Focus or AI Servo)? \$\endgroup\$
    – Philip Kendall
    Dec 30, 2015 at 13:01
  • \$\begingroup\$ @PhilipKendall he already says he's using manual mode: The images were taken on a tripod in manual mode with the exact same settings (250ms exposure, f/3.5, 18mm, ISO 100) \$\endgroup\$
    – Dragos
    Dec 30, 2015 at 13:05
  • 1
    \$\begingroup\$ Possible duplicate of Why is my camera focusing fine in liveview but getting it wrong with the viewfinder? \$\endgroup\$
    – dpollitt
    Dec 30, 2015 at 13:35
  • 1
    \$\begingroup\$ @PhilipKendall I'm using One Shot AF and the focus point was selected manually. \$\endgroup\$
    – Jules
    Dec 30, 2015 at 14:35

3 Answers 3

2
\$\begingroup\$

PDAF systems work based on the amount of contrast that they can detect. Your test scene includes a poorly lit and fairly low contrast target in the same frame with a much brighter and much higher contrast area at a different distance from your target. Even if both areas were equally in focus the area to the left would appear, to our eyes, to be better focused. So your test, as constructed, is pretty much designed from the start to cause PDAF to fail.

Is it normal for phase-detect autofocus to be inaccurate...

Yes. Even the best systems aren't perfect. There's no such thing as a perfect AF system, and certainly not a perfect PDAF system.

Roger Cicala at lensrentals.com recently wrote a series about focus performance that is pretty detailed and touches on several of these issues. It is a lot of material to go through, but I found it interesting reading.

Autofocus Reality Part 1: Center-Point, Single-Shot Accuracy
Autofocus Reality Part 2: One vs. Two, Old vs. New
Autofocus Reality Part 3A: Canon Lenses
Autofocus Reality Part 3B: Canon Cameras
Autofocus Reality Part 4: Nikon Full Frame

What I would like to know is if the severity in my example shots is normal and why it only happens at 18mm.

The reason it is more pronounced at 18mm is simply because the area of sensitivity for each AF point on the PDAF array buried in the floor of your camera covers a greater area of coverage of your target as you increase the angle of view of your lens. And those areas of sensitivity are often much larger that those little squares in the viewfinder!

The PDAF system in your 750D/Rebel T6i is pretty much a clone of the PDAF system introduced in the EOS 7D back in 2009. A few features available in the original 7D are not available in the 750D. This is not due to a hardware difference but rather a limitation in the firmware that reflects a choice by Canon to only include certain features in their higher tier cameras. As such, this answer to another question contains information highly relevant to your question. Pay particular attention to Roger Cicala's AF Reality, part 3b blog entry. The 7D (and 70D, 750D) PDAF system, though very configurable, is not exactly Canon's most consistently accurate one.

\$\endgroup\$
1
  • 1
    \$\begingroup\$ Thank you for your answer, this information is definitely helpful. I will read up, do some more tests, and report back in a few days. \$\endgroup\$
    – Jules
    Dec 31, 2015 at 0:09
1
\$\begingroup\$

It is going to be either extremely difficult or incorrect for anyone to make assumptions based in a single example image in less than perfect testing conditions.

My advice would be to send the pair to Canon for adjustment and fine tuning if you desire the utmost in AF accuracy. Since your model does not offer AF microadjustment, having Canon do it is the only way to my knowledge.

Generally speaking, that much out of focus is not desirable but we all have different levels of tolerance to this type of accuracy.

See this question for a very similar situation: Why is my camera focusing fine in liveview but getting it wrong with the viewfinder?

Maybe something in Rogers lengthy posts will help you, very good info here: http://www.lensrentals.com/blog/2012/07/autofocus-reality-part-1-center-point-single-shot-accuracy

\$\endgroup\$
2
  • \$\begingroup\$ Thanks for your reply (I can't upvote anything yet due to lack of reputation, sorry). As being said, the large variance in focus accuracy seems to happen in all sorts of shooting conditions and on all sorts of target surfaces. Would it help if I added more example photos? \$\endgroup\$
    – Jules
    Dec 30, 2015 at 14:59
  • \$\begingroup\$ I'm not sure AF microadjustment would help, since that would only shift the focusing distance, but my main problem is that the variance is so large (1.5m back and forth on a target at 3m distance). So instead of focusing between 1.5m and 0.1m front focus, the focus would be somewhere between 0.7m front and 0.7m back focus, which still sounds like a lot to me. \$\endgroup\$
    – Jules
    Dec 30, 2015 at 15:02
-1
\$\begingroup\$

After some more testing, it seems that high contrast areas can affect the PDAF, even if they are far away from the selected autofocus point. However, even in the very best conditions (flat, high contrast target on a flat, uniformly colored surface), some inaccuracy remains.

I also tried a different EOS 750D with the same results, so it seems unlikely that it is just an issue with my particular bodies.

I guess I'll just have to live with these inaccuracies of the PDAF and use the CDAF when I need perfect focus.

\$\endgroup\$
3
  • \$\begingroup\$ I appreciate that you came back to follow up, but saying that phase detect AF is significantly inaccurate in real use just because of high contrast areas? Do you have proof of this? \$\endgroup\$
    – dpollitt
    Jan 10, 2016 at 23:54
  • \$\begingroup\$ I don't have hard proof. When shooting through an open window, it looked to me like the AF sometimes tried to focus on the window frame, which went through one of the outer AF points and therefore was quite far away from the center AF point that was selected. This is why I qualified the statement with "it seems". I can also remove the sentence if everyone thinks that this is unlikely. \$\endgroup\$
    – Jules
    Jan 11, 2016 at 1:17
  • \$\begingroup\$ @dpollitt It's been fairly common knowledge for a long time. Bryan at The-Digital-Picture finally showed it in acton in his review of the Canon EOS 7D Mark II. It's about 25% down the page in the graphic with the "No Contrast | Low Contrast" mouseovers beneath the chart. \$\endgroup\$
    – Michael C
    Aug 2, 2017 at 9:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.