- Ed Dozier
Sigma Focus Algorithms: Speed versus Accuracy
Sigma lets you program their “global vision” series of lenses with their USB dock. This includes the Sport, Art, and Contemporary lenses. One of the things you can program is which autofocus algorithm to use. You get three algorithms to choose from: “Fast AF Priority”, “Standard AF”, or “Smooth AF Priority”.
By assigning a different algorithm to different custom switches on the lens (C1 and C2), you can change your mind on the fly and pick the appropriate focus algorithm to fit the shooting conditions. The “Smooth AF Priority” algorithm is primarily for video use, so I never use it (it’s the slowest focus algorithm). I’m interested in getting the fastest focus performance that I can get, so I want to use the “Fast AF Priority” whenever I can.
I have already measured the speed of the “Fast AF Priority” algorithm versus the “Standard AF” algorithm, and found that the Fast algorithm is about 20 percent quicker than the Standard algorithm. I had used a Nikon D500 and the Sigma 150-600 Contemporary for the speed test.
I thought I’d try to determine just how repeatable the focus algorithms are. If a camera/lens combination is super fast to focus but is totally unreliable at getting to the correct target distance, then you haven’t really gained anything.
I decided to use my Sigma 70-200 f/2.8 Sport lens for this test. I have programmed the C1 switch for the “Fast AF Priority” algorithm, and the C2 switch is programmed with “Standard AF” (“Standard” is also Sigma’s default algorithm if you don’t program the lens). I used a Nikon D850 for the tests.
All of the test shots were done at 190mm and f/2.8 from a distance of 1.88 meters. This is a fairly close subject distance, but I wanted to do a test where I could spot even tiny focus errors.
Sigma Custom Switch (C1) settings options
The screen above shows how to access the autofocus speed options, via the “AF Speed Setting” button. It also shows how my C1 lens switch is currently programmed with the “Fast AF Priority” and “Moderate View Mode” optical stabilization on my 70-200mm lens. All of the same options are available for the C2 lens switch.
Sigma’s available programmable AF Speed algorithms
The picture above shows you the three autofocus speed selections that are available for programming a lens with their Optimization Pro software and their USB dock. You can always change your mind and reprogram the lens later, if you’re not happy with a selection.
Sigma already upgraded the firmware in their 150-600 lenses, which vastly improved focus speed. If I hadn’t purchased their USB dock, I couldn’t have taken advantage of their improvements.
Focus Comparison Testing Procedures
To perform the tests, I would start by first selecting the desired (already-programmed) custom switch setting. I mounted the camera onto a sturdy tripod, because it’s critical to keep the camera at a fixed distance from the target. The camera was set to phase-detect autofocus, with all of the same settings I’d use for regular action photography (where I want fast autofocus). I only used unsharpened raw format for the testing, although jpeg can be used here if you aren’t concerned with accurate target edge resolution values.
I mounted a focus target that is designed for focus evaluation/calibration using the free MTFMapper software. The target is designed such that the (middle) camera focus sensor only sees a single high-contrast edge, and won’t be confused by neighboring details to focus on. The target is mounted at a 45-degree angle relative to the camera sensor. This makes it easy to determine what’s in focus and what isn’t. I focus on the middle of the target, where the big vertical trapezoid edge is located. When the target is rotated about the vertical, the trapezoid shape starts to look like a rectangle.
I focus with the lens wide open, so that there will be no room for doubt about where the plane of best focus ends up. This is, by the way, the same basic setup that I use to focus fine-tune my lenses at close distances. I have bigger targets for focus calibration at longer distances.
To spice up the test a little bit, I shot the photos at a light intensity of EV 7.3, which is typical indoor room lighting, and definitely more of a challenge for a focus system than sunlight.
The Focus Target
The photo above shows what the focus target looks like. The little blue numbers on each little slanted square are resolution measurements for each measured edge. These numbers are placed there via the MTFMapper program when the photo is analyzed. I’m using a small target, which has overall dimensions of just 8.5 inches tall by 9.5 inches wide, plus some whitespace around that. I want small little squares so that I can discern very small focus errors. The “large” vertical target edge I focus on is just 2 inches tall, and each little square is just a quarter inch on an edge (6.4mm).
Since the test shots are done with the target rotated by 45 degrees, the little squares in front and behind the large black target edge go quickly out of focus, and have a very low corresponding measured resolution number. Ideally, the highest resolution measurement would be the large vertical edge in the middle of the shot, since that’s where the focus sensor I’m using is aimed. The little squares that line up with that large vertical edge should have a similar resolution number (assuming the camera sensor edge is aligned parallel to the chart).
I start by manually shifting the focus well away from the middle of the target and press my “AF-ON” button to initiate autofocus. If all goes well, then the camera will of course focus perfectly on the large vertical edge in the middle of the field of view. The resolution reading (little blue number on the edge) should be highest on that same edge. I repeat this procedure over and over again; each time I de-focus the lens and press the AF-ON button to re-focus on the target edge and then take the shot.
Reality rears its ugly head, however. The resolution measurements will show where the lens actually ended up focusing. If you have quality equipment and have properly calibrated the focus “fine tune”, the best focus should at least be “near” to the desired focus distance. The camera’s phase-detect sensors will tell the camera when focus is “good enough”, and the camera then tells the lens to stop focusing. If you were to shoot in really dim lighting, then you may experience focus-hunting; use bright-enough lighting that your camera doesn't have to struggle with this test.
This test, then, is to evaluate the range of distances where focus ended up while using first the “fast” autofocus algorithm (C1 switch), and then using the “standard” autofocus algorithm (C2 switch).
Examining the focus target up close
In the shot above, I had turned the focus target upside-down, so that the right side of the target is rotated away from me. As you can see, the zone of sharp focus is really narrow. In this shot, the focus was perfect, and the little squares aligned above and below the large vertical edge have the highest resolution numbers (0.18 cycles per pixel).
You might notice that your camera will tend to focus too near if you start your focus distance setting in front of the target. As soon as the camera thinks focus is “good enough”, it stops the focus action. If you start from the far side of the target, the focus can tend to be too far (once again, it entered the “good enough” zone and stopped). Keep this in mind when performing focus fine-tune calibration; do a set of shots starting focus nearer and then a set of shots beyond the focus target to verify your camera’s focus behavior. My Nikon D850 doesn't suffer from the stop-focus-too-soon problem, no matter if I focus near-to-far or from far-to-near.
I couldn’t detect any difference in the tendency to miss focus with either the Standard or Fast autofocus algorithms. I did half of the tests starting focus too near the target and half starting focus beyond the target; it didn’t alter the results. I didn’t have a single focus miss of more than 7mm at 1.88 meters target distance, no matter which focus algorithm was chosen. I shot about 100 tests overall, to best determine “average” focus behavior. Never make a focus determination on the basis of a single shot; this is one of those “statistical” things. With either focus algorithm, the focus was on average within 3mm of perfect.
I had previously done this same testing procedure on my Sigma 150-600 Contemporary lens. I didn’t see any accuracy or repeatability problems by using the Fast algorithm instead of the Standard algorithm on that lens, either. This doesn’t, of course, guarantee that all of Sigma’s lenses behave this well. Always "trust but verify".
Here, then, is a case where you get it all: speed, repeatability, and accuracy.
If there aren’t any focus repeatability differences between the Fast and the Standard algorithms, then why would you choose the slower Standard algorithm? I have kept my C2 switch programmed with the Standard algorithm as a sort of insurance policy, but I haven’t needed it yet for general photography. It may be that in extremely dim lighting the Standard focus algorithm might be more reliable, but I haven’t tested it. I’ll leave that task to the reader, as they say.
I tried to describe my test procedures in painstaking detail, in case you want to verify your own Sigma lens/camera combination. The autofocus algorithm choices, not to mention all of the other programmable choices, are of course unavailable to you if you don’t get the Sigma USB dock. For me, the ability to customize my Sigma lenses using their dock has made all the difference.