BobAtkins.com - click to return to home page

Contrast Detection AF - Can it miss focus? Sigma 150-600 Contemporary

It's a widely held belief, backed up with experimental evidence, that Contrast detection AF (which is what is used by most cameras in Live View mode) is more accurate and more consistent then Phase detection AF (the normal AF mode for DSLRs). The downside of Live view contrast detection AF is that it's usually much slower than phase detection AF.

Here's what Roger Cicala of lensrental.com concluded after testing phase and contrast sensitive AF accuracy and repeatability using a Canon 50mm f/1.4 lens mounted to a 5D Mk II camera:

LiveView (contrast-detection) AF on a still target is more accurate than phase-detection AF. It should be so. Contrast detection is using the actual sensor to determine focus; phase detection is not. Overall we found about one shot in 10 was out of focus with phase detection.

LiveView AF is as accurate as Roger View MF. You may be better than this, or you may not.

The theory that contrast detection AF might be expected to be at least as good as phase detection AF (and maybe better) can be explained by looking at how they work. The figure below is a simplified illustration of how each operates.

The above figure shows a plot of focus distance vs. time. The red vertical line is the position of the subject (which doesn't change with time). The lens starts out at position "A", which is closer than the desired focus distance.

The green line represents the focus action of phase detection AF. Phase detection AF is a sort of electronic rangefinder. Special sensors (usually on the floor of the SLR mirror box) look at the image from the left and right (or top and bottom) of the lens. By analyzing the difference between these images, the camera can calculate two things. It can tell in which direction the lens needs to be focused and it can tell how far the focus must be moved to bring the subject into focus. So the camera drives the lens along the green line to the predicted focus point. Exact details aren't published by anyone, but it's pretty well established that when the focus gets close to the predicted point, the camera lows down the speed of focus and checks again to see how close it is to focus and makes appropriate adjustments. When it thinks the focus is good (within some fraction of the depth of field), focus is locked. In "one shot" AF mode, no further check of focus is made.

In contrast detection, the digital sensor itself is used and the camera measures the contrast of the image. The better focus is, the higher the contrast will be. So starting again from point A the focus must be changes to see how contrast changes. The system is looking for a peak in contrast and the only way to find a peak is to pass though it. Usually (at least for Canon systems), the focus is driven towards a shorter distance until contreast either passes though a pear or focus gets to the minimum focus distance (MFD) at point B in the figure above. Then it changes direction and drives focus towards infinity, looking for when contrast peaks. It can't tell it's peaked until it's actually gone past the focus point so it drives focus to point C. Then it steps focus back to where it thinks contrast was at a maximum and passes it once again, but only by a single step (point D). For the Canon system it them appears to make two more steps, one in the reverse direction followed by a smaller step back. At this point focus locks. In "one shot" AF mode no further check of focus is made.

An important difference between phase and contrast detection is that phase detection doesn't look at the signal from the sensor. It determines focus at the focus sensor, which should be at the exact same distance from the lens as the sensor is - but it might not be. If it isn't there will be a systematic front or back focus error. With Contrast detection, the actual signal from the sensor is used, so there should be no possibility of front of back focusing.

With methods of focus depend somewhat on data supplied by the lens to the camera. The camera needs to know what the focal length is and what the characteristics of the focusing motor in the lens are, so that it knows how to shift focus by a given amount to find focus accurately. If any of this data is inaccurate, there can be focus errors.

Here's a high speed video of the final movements of a lens as one shot contest detection AF operates.

The fist clip shows the lens steeping the focus distance, then backing up. The second part shows the lens passing through best focus, but not backing up to the right distance for best focus when AF locks.

The Sigma 150-600/5-6.3 DG OS HSM Contemporary lens

I recently reviewed this lens and I thought it was pretty good. Good enough that I actually spent my own money to buy one! When I reviewed it I looked at image sharpness at various focal lengths, mainly using phase detection AF since that's what 90% of users will be using 90% of the time. I did look at Contrast detection AF also and at the time I noted some issues but didn't fully investigate exactly what was happening.

However, once I got the lens myself and had plenty of time to look at it in detail, I found some issues with contrast detection AF in come situations. Phase detection AF is very good and it's what I'd use pretty all the time. There's typically no real need or reason to use contrast detection one shot AF. However, once you mount a Canon 1.4x or 2x TC on it, you lose the option of phase detection AF. Your only options are manual focus or (if it works) contrast detection AF. I say "if it works" because when I tried contest detection AF using a Tamron 1.4x TC it would never lock focus. I think the "search for focus" speed was just too fast for the camera (EOS 70D, 7D or 6D) to keep up with it. Canon TCs slow down the focus speed so don't have that issue.

At 600mm using contrast detection AF in one shot mode with a Canon 1.4x or 2x TC, focus was consistently off. Using an EOS 70D it would consistently front focus when focus started out from a distance closer then the subject and it would slightly back focus when starting out from a distance more distant than the subject. It would do this every time. Occasionally, starting out from a longer distance it would get pretty close to acceptable focus, but even then it was still back focused. Typically the focus error was large enough to be a problem. For example, at a distance of 10m, with a 2x TC attached and the lens at 600mm, the focus point could be up to about 6" in front of the target. as shown below:

Crop of image from slanted focus test. The front end of the slanted bar is about 4.5" in front of where the focus should have been
Sigma 150-600 Contemporary with Canon 2x TC at 1200mm f13. Canon EOS 70D using one shot contrast detection AF in Live View mode

I tried using an EOS 6D and 7D with basically similar result. So it doesn't seem to an issue related to a specific camera body or camera model.

I tried changing the focal length of the lens. At 500mm with a TC the results were similar though perhaps a little better. At 400mm with a TC focus was starting to get pretty good. Good enough to say that there was no significant repeatable error. So was it the TCs that were throwing focus off. Well, no really. When I looked at the lens itself with no TCs attached, I saw the same behavior at 600mm (front/back focusing), though not to the same extent as with the TCs.

In this second set of images I show the difference in resolution between normal phase detection AF one-shot and contrast detection one-shot. The lens was at 600mm (no TC added) and an EOS 6D was used. The 70D has a hybris sensor with phase sensitive pixels, so it's not 100% clear exactly what parts of the Live View AF process use phase sensitive AF and what parts use contrast detection AF when the lens is used without a TC. With a TC the aperture is too small to use the phase sensitive pixels. The image crops below are all 100% crops from near the center of the image.

As you can see, even without a TC, one-shot contrast detection AF isn't hitting focus well. With the 6D and 70D the error is greatest when the lens starts out from the MFD (minimum focus distance). With the 7D, the error is greatest when starting out from infinity. Just looking at the way the focus distance changes during focusing in contrast detection AF, the algorithm for the 7D looks a little different with an additional "zig-zig" as the lens approaches the focus point. It was generally the case, with or without TCs that the 6D and 70D showed most error when starting out from the MFD (2.8m), but the 7D showed most error when starting out from infinity. The error was largest with the TCs added and the lens at 600mm.

Just to confirm the TCs weren't an issue I tried one shot contrast detection AF using a Canon 500/4.5L with 1.4x and 2x TCs. Focus was good, with accurate focus being archived and no front/back focusing.

So did I have a bad lens? Was it just a bit out of spec? Well, I obtained a second sample. It was the same as the first one, perhaps slightly worse. So it really doesn't look like a quality control issue or a lens that just happened to be slightly out of spec.

So what's the problem here? Why isn't the lens focusing correctly? I don't know. I don't have details of what the camera is telling the lens to do and I don't know what data the lens is telling the camera. I don't know why the camera doesn't acknowledge the problem and move the lens. I assume once the camera thinks it's moved the lens to the right spot it stops. It doesn't do a final check to see it things look good. Normally it doesn't need to since it gets things perfectly right with the Canon 500/4.5L lens. My guess - and it's purely a guess - is that it's a lens issue. Perhaps the Sigma 150-600/5-6.3 contemporary isn't passing the right data to the camera for contrast detection focus to work at 600mm. Why phase detection AF isn't affected I don't know, but Phase Detection AF and Contrast detection AF use different algorithms to achieve focus and maybe they use different data supplied by the lens. Again, this is just a guess. I don't know exactly what's going on, and probably nobody outside the Canons and Sigma engineering labs does either.

Is it a problem

Well, yes and no. It's clearly a problem if you want to use one-shot contrast detection AF at 600mm with a TC. On the other hand, how often do you want to do this? [BTW I can't speak for Nikon systems since I have neither a Nikon DSLR or the lens in a Nikon mount]. How often would you want to use a TC on a lens that starts out at f6.3? With a 1.4x it's 840mm f9 and you'd probably want to stop down to f10 or f11 for sharpest results. With a 2x it's 1200mm f13 and again you might want to stop down to f16 for slightly better sharpness. Those apertures are slow enough to make you think twice about TC usage. Sharpness with TCs is actually pretty good with a static subject, a tripod mounted camera and precise manual focus. TCs are usable and better than enlarging an image taken with out them.

Can you correct this using microfocus adjustment in the lens or camera

Many cameras have microfocus adjustment and some lenses (like the Sigma 150-600 Contemporary) can be microfocus adjusted via a USB connection. However there are two reasons why neither of those systems are likely to be able to correct the observed behavior. (1) As far as I know they only correct phase sensitive AF operation. (2), even if (1) isn't correct, phase sensitive AF is fine, so any change in microfocus adjustment which affected both phase and contrast AF would make phase sensitive AF accuracy worse!

Could I be wrong?

I guess my methodology could be flawed, but I've been testing lenses for a long time. Also, I usded exactly the same methodology, test targets, illumination, camera bodies and test range for the Canon 500mm lens as the Sigma 150-600mm lens and I didn't see an issue with the Canon. That pretty much excludes things like me not focusing on the right part of the target or misinterpreting the test images.

Would the Tamron 150-600 be better?

The is also a very good lens. However it won't accept Canon TCs and with the Tamron 1.4x TC, one-shot contrast detection AF in live view doesn't work at all. The AF never finds focus lock. So the answer is no.

Is the Sigma 150-600 Sports better?

I don't know. I haven't looked closely at contrast detection AF in live with the . I did find that it would actually AF in phase detection mode with a Tamron 1.4x TC, which was very surprising. Contrast detection worked and looked OK, but I did not extensively test the accuracy with TCs. So they answer is maybe.

Is there a Work-around?

Yes, there is a work-around for this issue. It's most practical on cameras that have a custom exposure mode (which the EOS 70D, 7D and 6D do). These cameras have a position on the mode dial marked "C" (or C1, C2, C3 in come cases). You can use this position to store a particular camera setup.

The trick is to use continuous focus Contest detection AF rather than One-shot and to move the control of autofocus initiation from the shutter release "1/2 press" to another button (which you can so with these EOS models). You then store that configuration (along with things like exposure mode - Av for example - white balance, ISO setting etc.) in the custom dial position. When you attach a TC to the lens you shift to the custom settings mode, elect Live View and allow the camera to continuously focus the lens. When you press the shutter, the focus freezes and a shot is taken.

Problems? Sometimes. In the continuous AF Contrast detection live view mode with TCs attached to the lens, you can get into an oscillating focus situation at times where the focus never settles down, just swings back and forth through focus. Doesn't happen all the time but it can.

Last resort is to use manual focus with Live view and 10x screen magnification. This is 100% accurate, but doing it with a 1200mm f13 lens isn't easy since every time you touch the focus ring, the camera will shake, even with a sturdy tripod and head.

Bottom Line

I think there's an issue with the Sigma 150-600/5-6.3 DG OS lens when it comes to Live View contrast detection AF on (some?) Canon EOS bodies. Is it something that would stop me from buying the lens? No. In fact I did buy a myself. It's a very good lens which works very well under the conditions that I (or anyone else) will normally use. However if you try to add a TC and are shooting with the lens set to 600mm (which will be the normal condition when using a TC), you can't depend on one-shot Live View AF, at least not with the lenses I've looked at with the 70D, 7D and 6D bodies. Maybe you might have better luck than I did, but certainly test the system before depending on it. Tracking (continuous) contrast detection AF is accurate (though sometimes can oscillate about focus).

Could the issue be fixed with a firmware update of the lens? I don't know. Only the Sigma engineers can sat if there really is a problem, and if so if it can be corrected. It's unlikely to be a Canon problem so I wouldn't expect Canon to get involved. If their camera work just fine with their lenses and their TCs, that's all the responsibility they have. They aren't obliged to support 3rd party lenses or supply any information to 3rd party manufacturers. As far as I know they don't license their technology to anyone, so third party lens makers have to "reverse engineer" the AF system to determine what formation to send to the camera and how to interpret the signals from the camera to focus the lens.

Feedback

I'd be interested to hear from anyone who repeats these tests, whether or not they see the same behavior as I do.

[Return to Top of Page]