3
\$\begingroup\$

I feel like I quite consistently get:

  • Blurrier autofocus
  • Lower exposure

when shooting with my T5i's viewfinder than its LCD (i.e., in phase-based vs. contrast-based mode).

But whenever I search online, people seem to hail phase-based autofocus and say that contrast-based is just what cameras do when they can't do phase-based autofocus... and nobody even mentions anything about the exposure difference, making me wonder if I'm just... hallucinating?

Here's an example (phase-based vs. contrast-based). (Yes, they were taken at the same settings: ISO 1600, aperture f/5.6, shutter speed 1/64s) with stock 18-55mm lens.

Both images cropped from original JPG. Original, uncropped images here

If I'm being honest, I find it quite frustrating. :)

Am I misunderstanding which one is actually better?

If I actually care about quality, would I ever use phase-based autofocus?

Or is that just something I use when some other concern overrides quality (e.g. speed)?

\$\endgroup\$
15
  • 1
    \$\begingroup\$ Both shots seem to be done in very low light/high ISO (there is lots of color noise). I doubt that any effect you see in such situations can be transferred to normal light. \$\endgroup\$
    – Aganju
    Commented Jul 18, 2016 at 10:33
  • \$\begingroup\$ Were the 2 shots really taken 3 seconds apart? So you took a photo via the viewfinder and then went to live-view and took another 3 seconds later? Can you post the shutter speeds for each? They both look underexposed. RAW/JPEG won't be an issue. \$\endgroup\$
    – Steve Ives
    Commented Jul 18, 2016 at 10:44
  • 1
    \$\begingroup\$ @Mehrdad: My point is that both Autofocus behaviors get very poor and unpredictable in low light. Any comparison in phase vs. contrast in low light will not give you a useful result for Autofocus method selection. \$\endgroup\$
    – Aganju
    Commented Jul 18, 2016 at 11:47
  • 2
    \$\begingroup\$ Possible duplicate of Is it normal for phase-detect autofocus to be inaccurate with a Canon EOS 750D and EF-S 18-55mm lens? \$\endgroup\$
    – Michael C
    Commented Jul 18, 2016 at 14:44
  • 1
    \$\begingroup\$ @mehrdad: phase detection is not a miracle and requires sufficient details in the scene, which is nothing more for a camera than contrast. \$\endgroup\$ Commented Jul 18, 2016 at 21:34

2 Answers 2

4
\$\begingroup\$

As a first-order approximation:

  • Contrast-detection autofocus is more accurate and more flexible
  • Phase-detection autofocus is much faster
  • Contrast-detection always requires "hunting" back and forth to find the best focus; in ideal conditions phase-detect moves certainly to the right point
  • Contrast-detection which uses the main sensor is likely to hurt battery life
  • And as a practical matter, it means using the LCD screen or an EVF instead of the optical viewfinder

If your phase-detect system (camera and lens) are carefully micro-adjusted for the aperture, focal length, and focus distance you are using, results will be perfect.* Otherwise, phase detect results are likely to not be perfect — but they're generally good enough for 99.9% of people in 99% of cases. And the big speed advantage is real. So, that's why people count it as desirable.

As algorithms improve, and faster hardware comes to newer cameras, the speed advantage will come down, but it's hard to get around the hunting. The state of the art right now is hybrid modes, which use on-sensor phase-detect areas to get into the ballpark very quickly and then contrast-detect to fine-tune. I expect that eventually, pure phase-detect will become rare. But then, ten years ago, I thought we'd all be using Lytro-based cameras by now. It's hard to make predictions, especially about the future. :)

There is no inherent reason for exposure to be different. However, it is likely that as a side effect, your camera is also metering differently, using the main sensor instead of separate metering sensors. That probably accounts for the difference you are seeing in exposure, but I'd count that as a quirk of your particular camera model rather than something inherent. (And in any case, which exposure is chosen is a matter of preference, not something necessarily better or worse.)


* In fact, using contrast-detection mode is my favorite way to do micro-adjustment — but unfortunately, Canon doesn't provide that as a user-accessible feature for your camera.

\$\endgroup\$
4
  • \$\begingroup\$ +1 for the bullets. When you say "micro-adjust", do you mean adjust the camera for the particular lens? Because I haven't done that (I'm not sure how to either), I just put in the lens and started shooting. However, the issue isn't that the phase autofocus is always bad. If I'm out shooting landscapes in the sun, it usually turns out fine. It's when the conditions become less than perfect that this is no longer the case -- and I get this behavior quite consistently. So is the problem really an adjustment issue? I can buy that the metering might be different though, I'd have to look at that. \$\endgroup\$
    – user541686
    Commented Jul 18, 2016 at 19:01
  • \$\begingroup\$ Yeah, micro-adjustment is aligning your camera to work with a particular lens. Unfortunately, Canon reserves user-accessible micro-adjust as a feature for differentiating their higher-end models. You'll have to send your lens and camera in for service for them to do it for you. (Personally, I wouldn't buy a camera with phase-detect AF without this feature — but, again, obviously many, many people do without.) \$\endgroup\$
    – mattdm
    Commented Jul 18, 2016 at 21:13
  • \$\begingroup\$ @mattdm Pse see Mehrdad and my comments on my answrer. I'm happy for you to comment on any areas where our answers may vary. I think they overlap quite closely but I'd expect your detail to be better than mine. \$\endgroup\$ Commented Jul 19, 2016 at 9:52
  • \$\begingroup\$ The difference in exposure can probably be chalked up to a flickering light source illuminating the scene. Dim lights tend to flicker the most with the alternating current powering them. Or maybe the light coming through a window was affected by rapidly moving clouds? \$\endgroup\$
    – Michael C
    Commented Mar 10, 2023 at 10:16
3
\$\begingroup\$

1600, aperture 5.6, shutter speed 1/64 is 'low light'
Not dark by any means, but roughly 300 lux if I didn't do something wrong (no guarantee). (1 x 1600/100 x (5.6/1)^2 x 64/100 ~= 320)

The 1/64th shutter speed may be marginal shake-wise depending on what your actual (unstated) focal length is and whether you are using an (unstated) tripod.

Phase focusing is generally far FASTER than contrast focusing and usually no worse than contrast focusing provided mecahnical effects which may affect the results have been adjusted out.
One effect that can occur with one and not the other is that in an SLR the focus point on the main sensor is inferred by the focus sensors which use a different in-camera optical path. Lens seating can adversely affect the phase focusing as the optical path length can vary with mounting whereas contrast focusing detects the actual focus condition at the main sensor.
Have you "micro-adjusted the AF setting for this camera/lens combination?


You seem surprisingly resistant to people's attempt to help.
People are trying to establish your complete environment as it is extremely common for much time to be spent on questions before the full relevant details are provided. Your rep on others SE / SO sites indicates you will be well experienced in how questions are often misleading.

\$\endgroup\$
3
  • 1
    \$\begingroup\$ "Have you micro-adjusted"... OP's stated the body is a T5i--the dRebels don't have AFMA capability, they have to be sent to Canon service for adjustment. \$\endgroup\$
    – inkista
    Commented Jul 18, 2016 at 16:46
  • \$\begingroup\$ "Usually no worse than contrast focusing" -- but mattdm below says contrast-detection is "more accurate and more flexible". Do you disagree with him? If not, how should I reconcile the two? As for the last comment -- I'm not resistant to help obviously, rather it's that some questions like whether I "really" took these photos 3 seconds apart (is the assumption that I'm trying to trick people?) are kind of silly and frustrating when it's not even obvious to me if people really think the results are so unbelievable at all. \$\endgroup\$
    – user541686
    Commented Jul 18, 2016 at 18:57
  • 1
    \$\begingroup\$ @Mehrdad I think that what Matt & I said are close to the same. If in doubt, choose his version :-) (really). He said that contrast detection is more accurate [in absolute terms], and also noted that in ideal circumstances phase detect moves 'certainly to the right point' and also noted generally good enough for 99.9% of people in 99% of cases.-> All that is ~= summarised in my "usually no worse than". ie most people most time do not ind any difference (statisticaly) significant. | I & he discussed microadjustment to make phase focusing work, and why it is required. \$\endgroup\$ Commented Jul 19, 2016 at 9:51

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.