Canon 50mm f/1.4 and AF Reliability

[EOBeav]

Although I'm not in the market for a 5DmkIII, I've been reading with interest about the recent issues with light leaks and that body. That got me thinking about my 50mm f/1.4, a lens that I bought fairly soon after getting my first DSLR a few years ago. At the time, I was perplexed why I would carefully aim the AF, but at wide apertures, the area in focus was slightly different than where I had placed that little red dot. Not always, but sometimes. I thought I was doing something wrong. So, my remedy at the time was to take 10-12 clicks of the same exact comp whenever I was shooting that wide, particularly portraits of my two boys. I found that I was able to get some keepers in there somewhere.

Fast forward to 2012. Now that I have my 5DmkII, and have vastly improved my knowledge of the problem, I'm able to remedy the situation. And not with microfocusing, either. The problem I've come across in this lens isn't inaccurate AF, but inconsistently accurate AF. Had I the knowledge that I do now back when it was still under warranty, I might have sent it back asking for a different one. That may or may not have made a difference; what I'm reading is that this problem is likely model wide. How I've remedied the situation is to turn off the AF, turn on the image preview, and zoom in where I was to focus. From there, I can just spin that little black ring into position. The last time I did kid portraits, the results were amazing. I didn't have one shot where I said, "Dang, that's a nice pose. Too bad I can't use it because his eyes are out of focus." And this was a fairly active 4 y/o boy. It's not a perfect solution, but one that works for me.

So, all of that to say, these companies sometimes put out really good products that either have quirks or design flaws. I guess it's up to you to decide which one yours is. I've learned to live with the problems in my 50mm f/1.4. It is otherwise outstanding glass, especially when you consider the price. Since I can't return it, and I'm not willing to fork out money for a 50mm f/1.2 L, I've decided to just stick with it. It's become a lens I have on there all the time when I'm just out shooting. Like anything else, you have to know your gear, with all of the strengths and limitations it might have.

 

[TrumpetPower!]

While I wouldn't rule out the possibility of what you're experiencing being a hardware problem, what you're describing sounds much more like improper technique. Specifically, if you're using focus-and-recompose at f/1.4, or if you're not properly keeping the AF point on a moving subject in AI servo, the results would be much as you describe.

A good test would be to set the camera up on a tripod, focus on something that's not moving, make sure the AF point is smack-dab centered on whatever it is you're focussing on...and then manually focussing at infinity and engaging autofocus. Repeat that a few times with different subjects at different distances. If the hardware is inconsistent, it'll show up in this test -- and, if it is, you should consider getting it fixed.

But I'll bet you a cup of coffee that it won't be inconsistent and that what you really need is to improve your technique. The good news? That means spending more time taking pictures of kids....

Cheers,

b&

 

[EOBeav]

Been there, done that.

I've done a lot of tests, so at this point I'm ruling out technique. I definitely ruled out focus-and-recompose awhile back. I've learned better than to do that. I did utilize some microfocusing adjustments in my 5DmkII by about 3 steps. Even during that process, I pulled out a few stinkers.

I probably led you (and others) to believe that I was getting poor focusing results every single time. That's not the case. Out of every 10 shots wider than f/2.2 or so, 2-4 would show up with a misplaced focal point. Enough to where I couldn't say with 100% certainty that I was going to nail every shot. But when I manually focus and use live view, my keeper ratio jumps up dramatically.

 

[cpsico]

Yeah mine hunts like a blind dog on anything but the center point focus

 

[!Xabbu]

Yeah mine hunts like a blind dog on anything but the center point focus

What camera are you using? I think there is a significant difference between xxxD bodies and for example a 7D. I guess the 7D should focus pretty good with any focus point, but the xxxD might need the center point to get good focus.

More on topic - I have a 50mm f/1.4 as well and it is by far my least favorite lens. I seem to have focusing issues with it as well (could be technique, could be of center focus point of my 450D), but on top of that I don't feel like it gives me any better IQ than for example my 70-200mm f4.

 

[Marsu42]

What camera are you using? I think there is a significant difference between xxxD bodies and for example a 7D. I guess the 7D should focus pretty good with any focus point, but the xxxD might need the center point to get good focus.

The xxxd like the 600d only have af points with a sensitivity up to f2.8, while the 60d's and 7d's center point (and only this one, on both bodies!) is extra sensitive and can adjust the focus more precisely. This sure made a difference with using my 50/1.8 on my 60d, but for the even thinner dof of f1.4 focus and recompose might be just as bad as using an outer less-sensitive af point. Time to get a 5d3

The worst thing is that with the standard focussing screen the dof appears to be larger than it really is, so using the viewfinder you can never tell if the af has got it right. You can either exchange the screen or of course use live view with magic lantern's focus peaking which exactly shows what is sharp and what is not.

The only way I've found around this on non-moving objects: focus, shoot, re-focus, shoot again - some picture is bound to be sharp, but of course this eats shutter cycles.

 

[!Xabbu]

What camera are you using? I think there is a significant difference between xxxD bodies and for example a 7D. I guess the 7D should focus pretty good with any focus point, but the xxxD might need the center point to get good focus.

The xxxd like the 600d only have af points with a sensitivity up to f2.8, while the 60d's and 7d's center point (and only this one, on both bodies!) is extra sensitive and can adjust the focus more precisely. This sure made a difference with using my 50/1.8 on my 60d, but for the even thinner dof of f1.4 focus and recompose might be just as bad as using an outer less-sensitive af point. Time to get a 5d3

The worst thing is that with the standard focussing screen the dof appears to be larger than it really is, so using the viewfinder you can never tell if the af has got it right. You can either exchange the screen or of course use live view with magic lantern's focus peaking which exactly shows what is sharp and what is not.

The only way I've found around this on non-moving objects: focus, shoot, re-focus, shoot again - some picture is bound to be sharp, but of course this eats shutter cycles.

I thought there is a significant difference between even 60D and 600D AF - Chuck Westfall: "The EOS 60D has the same AF sensor as the EOS 40D and 50D, so all 9 points are standard-precision cross-type effective to maximum apertures as small as f/5.6. There is also an X-shaped high-precision cross-type focusing sensor on the center focusing point that is used with coupled lenses and lens/extender combinations that have maximum apertures of f/2.8 or larger."
The-Digital-Picture reports the 600D to have horizontal-line-only sensitive AF points apart from the center point which is cross type and gives extra precision for f/2.8 or faster aperture lenses.
All 7D AF points are cross-type as well, which should give them a considerable edge over xxxD cameras. However, you're right the center point delivers extra precision for f/2.8 or faster aperture lenses.

Is my assumption that off-center focus points for the 60D will work pretty well not correct? I was hoping that the off-center points would be at least as good as the xxxD center point (apart from the f/2.8 advantage).

 

[Mt Spokane Photography]

All lenses have inconsistent autofocus, and the lower cost ones are the worst. One easy way to observe this is to use Reikan Focal to AFMA your lens and observe the analysis chart. Extreme care is needed to make sure that movement or vibration is not affecting the results, just someone walking on a wooden house floor will cause noticible AF issues.

FoCal sets the lens to infinity and does a phase detect AF for each shot. The variation is normally small, but ocassionally, there is a large difference that will be out of focus.

In the chart below, the blue diamonds are the shots taken at each AFMA setting. Look at the ones stacked vertically, they represent identical shots with no external variation, (except vibration).

I have found the longer telephoto "L" lenses to have very little variation, but all others, including the "L" lenses do. Its a good idea to take several shots when you can, about one in 3 or 5 can be out enough to where its really bad.

In this example, I think the lens was at 400mm and I was walking around, I'm going to redo it on concrete for better stability.




This one is one of the better results.

 

[Marsu42]

Is my assumption that off-center focus points for the 60D will work pretty well not correct? I was hoping that the off-center points would be at least as good as the xxxD center point (apart from the f/2.8 advantage).

There's two criteria for af points that you seem to be mixing up when talking about a "good" af point: sensitivity and pattern detection. The sensitivity of the points is how precise the body tell the lens to af - this is needed when using a lens with a large open aperture because of the thin depth of field. The type of the detection (cross etc) just determines if the af point is able to focus at all or how likely it is to hunt (try to focus at a venetian blind without a cross type af...).

So yes, the 60d/7d have better pattern detection on the outer points, but that doesn't help when using a f2.8+ lens with open aperture because they are not sensitive enough.

 

[!Xabbu]

Is my assumption that off-center focus points for the 60D will work pretty well not correct? I was hoping that the off-center points would be at least as good as the xxxD center point (apart from the f/2.8 advantage).

There's two criteria for af points that you seem to be mixing up when talking about a "good" af point: sensitivity and pattern detection. The sensitivity of the points is how precise the body tell the lens to af - this is needed when using a lens with a large open aperture because of the thin depth of field. The type of the detection (cross etc) just determines if the af point is able to focus at all or how likely it is to hunt (try to focus at a venetian blind without a cross type af...).

So yes, the 60d/7d have better pattern detection on the outer points, but that doesn't help when using a f2.8+ lens with open aperture because they are not sensitive enough.

Thanks, I wasn't aware of that - I thought that a cross sensor would be also more precise.

 

[TrumpetPower!]

Thanks, I wasn't aware of that - I thought that a cross sensor would be also more precise.

My high school chemistry teacher beat into us the difference between accuracy and precision. Accuracy is how far off the mark your measurement is. Precision is how fine the gradations are on your ruler.

Imagine you're measuring the proverbial football field. If you use the yard markers painted on the grass, you'll get a very accurate but imprecise measurement. If you use a 36" fabric tape measure with 1/8" markings on the soggy grass, marking the end of the tape and picking it up and moving it a hundred or so times, you'll wind up with a very precise measurement (down to the eighth of an inch) that won't be anywhere near as accurate (I'd guess you'd be off by at least several yards).

On the other hand, if your accuracy is good, increasing precision will also increase accuracy. Measure the field not with the inch-thick every-three-feet paint stripes but with a laser rangefinder, and you might well find that the field isn't actually 100 yards, but 100 yards plus 7 319/256 inches.

In a similar vein, the cross-type sensors, though they don't direcly improve precision, may well enable the camera to focus definitively on a spot that it couldn't quite latch on to otherwise, resulting in a better overall measurement and a better in-focus picture.

Cheers,

b&

 

[steves]

Thanks, I wasn't aware of that - I thought that a cross sensor would be also more precise.

My high school chemistry teacher beat into us the difference between accuracy and precision. Accuracy is how far off the mark your measurement is. Precision is how fine the gradations are on your ruler.

Imagine you're measuring the proverbial football field. If you use the yard markers painted on the grass, you'll get a very accurate but imprecise measurement. If you use a 36" fabric tape measure with 1/8" markings on the soggy grass, marking the end of the tape and picking it up and moving it a hundred or so times, you'll wind up with a very precise measurement (down to the eighth of an inch) that won't be anywhere near as accurate (I'd guess you'd be off by at least several yards).

On the other hand, if your accuracy is good, increasing precision will also increase accuracy. Measure the field not with the inch-thick every-three-feet paint stripes but with a laser rangefinder, and you might well find that the field isn't actually 100 yards, but 100 yards plus 7 319/256 inches.

In a similar vein, the cross-type sensors, though they don't direcly improve precision, may well enable the camera to focus definitively on a spot that it couldn't quite latch on to otherwise, resulting in a better overall measurement and a better in-focus picture.

Cheers,

b&

Perhaps a better way to distinguish accuracy and precision is statistically. The focus point is really a random quantity due to various real-world imperfections. If you repeatedly focus on a target point (focusing elsewhere between each focus attempt), the actual focus points (as compared to the desired point) will have a random distribution. Accuracy is the bias of this distribution---the difference between the mean (average) of the distribution and the desired value. Precision is captured by the variance of the distribution. So an accurate focusing system will have small bias (the average value is close to the true value), but may still be imprecise (have high variance). Or vice-versa. What one usually cares about is minimizing the total error in some sense. The mean-squared-error (mse) is the sum of the variance and the square of the bias. So a "good" system, i.e. one with low error, must be both accurate and precise.

The classic example is in firing artillery at a target. Precision refers to the spread (dispersion) of the pattern of rounds. Accuracy refers to the deviation of the center of the pattern from the target. An accurate weapon will center the pattern on the target. A precise weapon will have a tight pattern. An "effective" weapon will do both.

But back to the 50 1.4. I got mine for low-light shooting, but mostly stopped using it when I discovered that it is soft wide open and doesn't really get maximally sharp until about 2.8---at which point I may as well use my EF-S 17-55 2.8. This softness is easy to see if your camera body supports live-view. Put the camera/lens on a tripod in AV mode and focus on something close with texture (e.g. a curtain). Put the camera in live-view and zoom the rear LCD view to 10x. Then as you adjust the aperture from 1.4 to 2.8 you can easily see the image get noticeably sharper.

Of course one can over-obsess over sharpness and the softness at 1.4 may not be a problem, particularly for images seen on a screen at 1 or 2 MP. But I keep waiting for Canon to produce a successor to the venerable 50 1.4....

 

[Marsu42]

Perhaps a better way to distinguish accuracy and precision is statistically. The focus point is really a random quantity due to various real-world imperfections. If you repeatedly focus on a target point (focusing elsewhere between each focus attempt), the actual focus points (as compared to the desired point) will have a random distribution. Accuracy is the bias of this distribution---the difference between the mean (average) of the distribution and the desired value. Precision is captured by the variance of the distribution. So an accurate focusing system will have small bias (the average value is close to the true value), but may still be imprecise (have high variance). Or vice-versa. What one usually cares about is minimizing the total error in some sense. The mean-squared-error (mse) is the sum of the variance and the square of the bias. So a "good" system, i.e. one with low error, must be both accurate and precise.

Thanks for the good descriptions, steves & TrumpetPower! And after reading this, I guess the af of the 60d is not that bad after all because it's seldom that it misses the focus completely, but it has some variance because the af square covers a large dof when trying to focus at a tilted surface like a butterfly. When shooting handheld and there is some shake, the af cannot possibly know where exactly I want the focus to be...

 

[EOBeav]

All lenses have inconsistent autofocus, and the lower cost ones are the worst.

In a round-about way, that was probably the point I was getting at. I know enough about my lens flaws to be able to capitalize on it's strong features.

Extreme care is needed to make sure that movement or vibration is not affecting the results, just someone walking on a wooden house floor will cause noticible AF issues.

Another excellent point. Whenever I see a focusing issue, I try to determine whether it's not sharp where I aimed it (misplaced focal point), or not sharp anywhere (movement/vibration). After 1000's of images taken between my two dslr's, I've finally figured out the difference.

Its a good idea to take several shots when you can, about one in 3 or 5 can be out enough to where its really bad.

Yes, "when you can." When I shoot small children, that plan goes out the window.

Maybe we've come to expect too much of AF?