AF test (2): How accurate is Auto Focus?

0
DSC0011-voor-web
In a previous article about the speed of the auto focus, we discussed how the various AF systems work. And why one system is faster than another. We looked at how well the AF continues to work when there is (very) little light.
The differences between the different camera types (SLR, compact, mirrorless) and the lenses used appear to be very big in practice! The assumption that an SLR with phase detection AF focuses faster than a system camera with contrast detection appears to be outdated. Of all the cameras that we have tested, the Panasonic GX8—which makes use of an advanced form of contrast detection—focuses the fastest in our tests. With an attractively priced Panasonic 14-140mm kit lens, the Panasonic GX8 focused from infinity to 1 meter in 0.05 seconds. The slowest, just as modern, cameras need a second or more to focus under the same conditions as the Panasonic GX8.
What do you get out of the AF when it is fast, but not accurate? That is the questions we want to thoroughly examine now. We were surprised again…

damhert web

Fallow deer, Nikon D810 with 500 mm Nikkor lens @ 1/3200 sec f/5.6, ISO 3200. With a telephoto set-up like this, the focal depth is very limited and AF accuracy—also in low light—is crucial!

What is the best?

The temptation is great to immediately label the fastest AF as the best AF. That is a misconception! What do you get out of the AF when it is fast, but not accurate? First, the AF speed of a camera depends on the amount of light. The less light there is, the longer it takes to focus. In low light, there are enormous differences in AF speed. Some cameras also give up on focusing when it becomes dark, while you can still focus with other cameras under the same circumstances. Some camera users find the number of AF fields important. We do not share that opinion, since 95 of 100 shots are normally made with the central focus field. A moveable AF field is primarily useful when working on a tripod or if the subject is not in the center and you do not have much focal depth. A camera that makes it possible to decrease the size of the AF point also decreases the chance of focusing on the wrong point.
We certainly think that AF accuracy is just as important as the speed. What use is a quick photo if it is not in focus. CameraStuffReview naturally investigated that as well. Then it became apparent that the fastest AF is not automatically the most accurate.

Always focusing wrong:
too far forward (front focus) or too far back (back focus)

If a camera/lens combination always focuses too far forward or too far back, then that is a systematic error. There are solutions for that.

With SLR cameras, the focus module is at a different place in the camera than the sensor with which the photo is taken. What happens if the AF module is not perfectly aligned relative to the sensor? Then the AF system thinks that a sharp photo can be made. As soon as the mirror pops up, a shot can be taken, but the AF can no longer be adjusted, because no light is then falling on the AF sensor. The result is that a specific combination of camera and lens at a certain distance and with a certain focal length will always focus too far forward (front focus) or always focus too far back (back focus). This is a complex subject: the same lens on one camera has trouble with front focus and can focus perfectly on another camera. These are in particular the lenses with a very small focal depth/high brightness, so that it becomes visible that the AF has not focused precisely. And because modern cameras have sensors with increasingly high resolution, it also becomes more clearly visible if the camera has not focused completely right.

For the more expensive SLR cameras, there are options for correcting this per lens, but usually not per focal length or distance that is focused on. For the Sigma Contemporary, Art and Sports lenses, you can buy a USB dock, with which you can adjust the focus of the lens on your camera, such that the focus will always land on the right spot, independent of distance or focal length. That resolves the systematic error of the AF system. If you do not have a Sigma C, S or A lens, then you can have the AF of your lens adjusted on your camera by the technical department of the importer. That does not, however, resolve all focus errors.

If a camera/lens combination sometimes focuses too far forward and sometimes too far back, then that is an incidental error. That is one you have to learn to live with.

No camera focuses on exactly the same spot if focused repeatedly. Fortunately, focusing is so accurate, or the focal depth is so large, that most photographers will not notice that their camera is not (quite) focusing properly. In the past months, we have measured the range in sharpness for dozens of cameras and lenses, whereby we have intentionally not considered the systematic error (front focus/back focus). Just as for the AF speed, there appear to be enormous differences in AF accuracy as well.

Misses, an underexposed problem

InscherpSpinnenweb

And then we have the phenomenon of the “total misses”. Both in some practical situations and during testing in the studio, the camera sometimes beeps to indicate that the focus is good, when you can already see that this is totally not the case. In our testing procedure, we determine the AF speed by having the camera, on a tripod and always with the same image ratio, focus from infinity to ca. 1 meter on a high-contrast subject and then measuring how long the camera takes to do that. Because the focus time is never exactly the same, we do that test five to ten times and calculate the average (the median, to be precise). We do that again with variable light levels (“from candlelight to daylight”) in order to determine whether the AF speed depends on the light value. Altogether, it quickly adds up to 50 shots.

You do not even want to know how often a totally blurred shot appears in a series like that! And we are not just talking about consumer cameras that cost a couple hundred euros, but also about SLRs in the professional segment. The reason that this problem usually goes unnoticed in daily shooting practice is that when you look through the viewfinder, you see even before taking the shot that it is not in focus. You instinctively move the AF to another point and usually that resolves the issue. In the worst case, you manually overrule the AF to get the focus to land in the right place. In our laboratory tests, we expressly do not use the viewfinder again after we have checked at the start that the right AF point has been selected, and then we just click away, according to a fixed testing protocol. And then you only discover the misses later.

We track this percentage of “misses” and hope to be able to report it at some point per camera or per lens. Our experience so far is that “misses” happen more often with wide angles than with regular lenses. But also that a lens with a great bokeh (long focal length and/or a large aperture) on an SLR camera set to infinity sometimes refuses to focus on a subject that is very nearby (and vice versa). At the moment, our database is still too small to make any further statements about that. Stick with us…

Focal depth camouflages a lot

onscherptevoorgrond2 web
onscherptevoorgrond web
Because there is such a thing as focal depth, you can capture a subject sharply with multiple set distances. You will only see that the sharpness in front of and behind that subject varies. You would expect that with a camera on a tripod, when making use of the same focal field and exactly the same image ratio, the autofocus would always choose the same distance setting. That does not appear to be the case! Once you recover from the shock, you might expect in searching for an explanation in contrast detection, where the camera “wanders” until the area is found where the contrast no longer changes with a change to the distance setting; you can imagine that you would get different results when you approach that point from “far way” or “close up”. But with phase detection, that should not be needed: the camera “calculates” how far the lens should be turned, and then it doesn’t change anymore. But things do not always happen in practice in the way you would expect!
See the shot above, made with a good SLR and a good (24 mm) lens of a subject at about 500 m distance (call it infinity). All on a sturdy tripod, focused with 1/3200 sec at f/2.8, with the middle AF field on the farm. We repeated this about 10 times. The shot at the bottom is a detail (200%) of the grass in the foreground from two different shots in the series. The farm is in focus in both cases, but from the sharpness (or lack thereof) in the foreground, you see that there is a big difference in the set distance! The focal depth camouflages the differences in the focus on the farm in the distance, but they are definitely there.

EXIF information about focal distance useless

In order to explore this phenomenon a bit further, we decided to look at the EXIF information. Most photo editing programs do not show the set distance in the EXIF list of settings, but with practically all cameras, that information is stored in the RAW or jpg file. You just have to search with a special program like EXIF viewer. Unfortunately, that EXIF information is 4-bits, and thus only gives 16 steps, and those are primarily in the nearby field. That is understandable: the distance information is primarily used for controlling the flash intensity, and you always use the flash for up-close.
Although the EXIF information for the focal distance cannot give an answer about the exact distance that was used for the focus, it can be an indication: if it is apparent from the EXIF information that two shots that you thought were focused to the same distance have different focal distances, then you know that the camera has not focused on the same distance.

Shifting apertures

Focusing is always done at full aperture, even with contrast detection. If you are not photographing at full aperture, then the camera closes the aperture after focusing. That shifts the focal point a bit! It is caused by spherical aberrations. The degree to which that is visible differs per lens. You only see that with very bright lenses; due to stopping down, the focal depth increases so that you do not see the focus shift.

Live View

bloemen web
Focusing with Live View works with contrast detection. As we indicated previously, this is not the fastest, but it is a very reliable focusing method. You namely use the information from the recording sensor, so you eliminate front focus/back focus problems. Some cameras have the ability to zoom in on the Live View image; that of course does not make the AF better, but you can see it better. For a shot like the one above, the blur is part of the creative process. You can do something like that very nicely in LV.

 The hyperfocal distance

The hyperfocal distance is the set distance whereby you have the maximum focal depth range. That depends on the focal length of the lens used and on the aperture. The “furthest” distance that is still rendered sharply is then infinity; the “shortest” distance that is still sharp is exactly half of the hyperfocal distance. (We will spare you the mathematical formula that shows this.) Especially landscape photographers are interested in the hyperfocal distance; they want to have the landscape sharp from front to back and adjust the aperture accordingly. We have seen these photographers out working with awkward tables. We have wondered for years when a camera would appear that would calculate the hyperfocal distance for me and show it on the display, and that would preferably also set the distance on the lens as well. All the data for such a calculation is, after all, already in the camera’s processor! We actually find the aperture button that appears on many bodies rather useless for assessing the focal depth. The image is much too dark and only loans itself to an overall impression; we really never use it.
 

Conclusion

The deeper we dive into the phenomenon of auto focus and the more critically we look at it, the more we come to the conclusion that the AF speed has improved enormously in recent years. And that there is still a great deal of work to be done on the optimization of the accuracy of AF. Sometimes the algorithms just make a mess of it, even with the “high-quality” cameras. When you stop down, the irregularities often remain under the surface. But when working at full aperture, and with the enormous resolutions of modern sensors, you see every defect. You can bet that now that the very bright lenses (f/1.4, f/1.8) lie within the reach of the regular hobbyist, this problem will get more attention.

We now have a good-sized dataset with focusing data (speed and accuracy) from dozens of cameras and lenses. We are going to expand and work that out further and come back to that in part 3 of this AF series. 

LEAVE A REPLY

Please enter your comment!
Please enter your name here