Points in Focus Photography

An Investigation into Canon AF Performance

0

This was originally published in 2 parts back in 2012. This piece is a consolidation of those two original articles, with updated performance data from more modern cameras, and new lenses (including the STM lenses that prompted the original articles). Most of the content prior to the 5D mark IV tests remains verbatim from those articles, though I have added some new insights to them in this update.

A bit of back story here, when Imaging Resources previewed the Canon EOS-M they complained that the AF speed (using the hybrid phase and contrast detect sensor) was slow. They claimed it was comparable to the EF 40mm f/2.8 STM they were using on the Rebel T4i they were testing.

Additionally, they provided some numbers for the speed, saying the STM lenses took 1.2-1.7 seconds to focus and lock on the T4i/EOS-M while other systems were doing contrast based focusing in 1/4 of that time.

When talking about AF performance, there’s two aspects that interplay. One is the camera’s processing and measuring capabilities, and the other is the lens’s ability to shift focus. In other words, if you put a slow moving lens on a fast camera, the overall performance will be slow. Similarly, a fast moving lens on a slow processing camera will likewise perform slowly.

However, this raises an interesting question, just how fast does the 40mm f/2.8 STM focus?

The story from Canon has been that the STM lenses are optimized for smooth, and silent, focusing when shooting video. So the question then becomes how fast are STM lenses when it comes to fast focus shifts?

Since I didn’t have a 40mm STM at the time I originally wrote this part of the article, I’ve started with the following bit from SLR Gear.

In practice the lens is indeed much quieter to focus than previous lenses, and is still very quick to focus, taking about one second to go from close focus to infinity. – SLR Gear

Now SLR gear isn’t completely transparent on their test conditions, so there’s a lot left up for assumption. In coming up with my numbers, I elected to simply test the hunting speed from infinity to close focus and solely the lens drive not the complete focus operation. That is, the time from when the lens starts moving to when the lens stops moving is all that I’m measuring.

Secondly since behavior will change with light, as the AF sensor is given more time to integrate sufficient signal in the dark I’ve provided 2 times. The dark slew time is taken with a lens cap on the lens and represents a situation where the camera is operating below the bottom extreme of the camera’s AF capabilities.

Test Methodology

The tests in being preformed here were done looking at the complete AF hunting sweep, from infinity focus to minimum focus and back to infinity. However, timing is provide from both half sweeps and the full sweep.

Hunting represented the worst case focusing performance as the camera will deliberately drive the lens slower in an effort to acquire better contrast/focusing information. In other words, in actual use, when there’s a subject capable of being focused on the performance shouldn’t be worse than the reported numbers. Moreover, since the hunting pattern covers the full focusing range, more normal shorter focusing throws, again should never be slower..

Tests were preformed under two conditions, darkness and a diffuse white background. The dark tests were done with the lens cap attached.

In the original test methodology, for the light tests, the lenses were placed close to a diffuse white target in front of a 100 W tungsten lamp. This was provided by a softbox, with a sheet of white acrylic placed immediately in front of the front most scrim.

For this new round of testing, the light test case is provided using a sheet of velum placed immediately over the front of the lens, and a continuous LED light shone on it.

Timing information in the original test was measured by counting frames of 60 FPS video recording the motion of the distance scale on the lens. At 60 FPS, frame times are approximately 17 ms.

For the more recent tests, timing was done with 240 FPS video (4.2 ms frame times). For lenses with distance scales the same approach as the previous test was used in examining the distance scale’s movement. In cases where there is no distance scale, another approach was used and will be detailed in that test.

In all cases, the center most AF point, of AF position was used. Additionally the camera and lens were kept level to insure that gravity wasn’t a factor in one focus direction or the other.

Testing with the EOS-1D mark III

Lens Light (seconds) Dark (seconds)
Canon EF 16-35mm f/2.8L II USM 0.27 0.27
Canon EF 24-70mm f/2.8L USM 0.41 0.81
Canon EF 24-105mm f/4L IS USM 0.33 0.34
Canon EF 70-200mm f/4L IS USM 0.72 0.98

Note: The numbers recorded in this test only cover the entire hunting operation (e.g., from infinity to mfd and back) and aren’t broken by direction.

Commenting on some interesting observations that came out of this test. There’s an interesting amount of variation between the light and dark tests. Some lenses, such as the 16-35 and 24-105 didn’t show any appreciable difference in hunting behavior between the two cases. However, other lenses, most notable the 24-70 showed a dramatic difference in hunting time with the dark case being twice as slow as the light case.

Additionally, the 16-35 and 24-105 didn’t slow down appreciably under dark conditions, instead they paused longer at the mid point (macro in the case of my tests) before returning to the start position.

While there’s clearly differences in lens performance, there are also clearly differences in the way the camera is processing the data as well. For example, the fast ultra-wide focused very quickly regardless of lighting, while the similar speed 24-70 wasn’t nearly as fast. These differences may be a product of focal length (the video shows the 16-35 being tested at 35mm, but I can’t tell what focal length the 24-70 was tested at) and the consequently different depths of field that the camera is working with.

The differences don’t appear to be related to AF sensor type though, as the 24-105/4 focusing was between the two f/2.8 lenses. If we were to pick a focus distance, and compare we see depths of field as show in the following table (10 ft. focusing distance).

Focal Length Aperture Depth of Field
16 2.8 31’2″ (9.51 m)
24 2.8 7’2.7″ (2.2 m)
35 2.8 3’1.2″ (945 mm)
70 2.8 9″ (229 mm)
105 4 5.6″ (143 mm)
200 4 1.5″ (38.2 mm)

Though I don’t have the data anymore, it’s likely the 24-105 was tested at 24mm, and the 24-70 was tested at 70 mm as those are the shortest/collapsed focal lengths for those lenses. In any of these cases it’s possible the camera was adjusting for allowable focusing errors to improve focusing speed.

Testing with the EOS 40D

In this round of testing (which was initially preformed a few weeks after the first sections, again back in 2012), I broke out performance from infinity to MFD as well as having the full time.

Lens Dark Conditions Light Conditions
Inf→Macro Slew Total Time Inf→Macro Slew Total Time
Canon EF-S 10-22mm f/3.5-4.5 USM 0.23 0.90 0.23 0.48
Canon EF 16-35mm f/2.8L II USM 0.27 0.94 0.26 0.54
Canon EF 24-70mm f/2.8L USM 0.81 2.14 0.41 0.85
Canon EF 24-105mm f/4L IS USM 0.34 0.98 0.34 0.68
Canon EF 70-200mm f/4L IS USM 1.05 2.51 0.38 0.77

In my previous article, I made the following comments.

In my initial tests with the 1D3, the light used for the AF light tests was a 100W incandescent lamp behind a Smith Victor DP10 diffuser. Only with the 16-35/2.8 was it necessary to provide a second diffusion layer—a piece of velum was used—to stop the camera from momentarily pausing while hunting. With the 40D it was necessary to double diffuse the front light for the entirety of the tests. Moreover, a sheet of velum was insufficient to stop the mid-hunt pauses, and it was necessary to use a sheet of opaque white acrylic.

I’m not entirely sure yet what this is telling me about the AF sensors, as it’s certainly not a product of the lenses. My only current working theory is that the 40D is momentarily seeing some sort of pattern in the out of focus image that is causing it to pause momentarily to re-evaluate the situation and then move on.

For the moment, I’m not really willing to start drawing any conclusions from the data, and there are quite a few more tests I’d like to try. However, didn’t want to sit on the data and possibly lose it either. So there it is, part 2 of my rather limited look at lens focus slew performance during autofocus “hunting” operations.

Test with the EOS 5D mark IV (PDAF)

In this round of testing, for zoom lenses I tested at both end of the focal range. In most cases this didn’t matter, at least not much, but in some instances there’s a marked difference in AF focusing rates.

Lens Focal Length Time ∞→MFD
(dark / light in s)
∞→MFD→∞
(dark / light in s)
EF 16-35mm f/4L IS USM 16 0.238 / 0.246 0.868 / 0.517
EF 16-35mm f/4L IS USM 35 0.517 / 0.234 0.859 / 0.505
EF 24-70mm f/2.8L II USM 24 0.292 / 0.321 0.913 / 0.663
EF 24-70mm f/2.8L II USM 70 0.300 / 0.296 0.947 / 0.626
EF 24-105mm f/4L IS USM 24 0.317 / 0.304 0.909 / 0.634
EF 24-105mm f/4L IS USM 105 0.392 / 0.317 1.047 / 0.626
EF 70-200mm f/4L IS USM 70 0.321 / 0.329 0.934 / 0.684
EF 70-200mm f/4L IS USM 200 0.947 / 0.334 2.507 / 0.734
EF 70-200mm f/4L IS USM + 1.4x II 98 0.321 / 0.321 0.851 / 0.667
EF 70-200mm f/4L IS USM + 1.4x II 280 1.226 / 0.571 2.707 / 1.205
EF 40mm f/2.8 STM 40 0.605 / 0.613 1.664 / 1.285
EF 50mm f/1.8 STM 50 0.672 / 0.630 1.706 / 1.376

In this round of testing, I had some interesting results fall out, namely Canon’s AF algorithm appears to adjust the focusing drive rate based on the focal length and aperture of the lens. This is most markedly seen with the 70-200 where the drive rate at 200 mm is more than twice as slow as the lens at 70 mm.

Another interesting observation in the above chart is that the STM lenses on a pro level AF system are substantially slower focusing than their USM counterparts. Even when you have nearly apples for apples counterparts like the EF 24-70mm f/2.8L II USM at the same aperture, or the EF 16-35mm f/4L IS USM at nearly the same focal length; both are about twice as fast as the STM lens.

Testing 5D mark IV Dual Pixel AF

These tests were done in live still photography live view mode.

Lens Focal Length Time ∞→MFD
(dark / light in s)
Time ∞→MFD→∞
(dark / light in s )
EF 24-70mm f/2.8L II USM 24 0.300 / 0.405 0.663 / 0.759
EF 24-70mm f/2.8L II USM 70 0.355 / 0.409 0.759 / 0.734
EF 40mm f/2.8 STM 40 1.397 / 1.089 2.598 / 2.261
EF 50mm f/1.8 STM 50 0.909 / 0.742 1.647 / 1.443

As an aside, I was goign to retest most if not all of the lenses I tested in the last section, and I may still do that, but I had time to work on a more limited test range and I wanted to get that data up.

What’s interesting to me, and reflects my experiences with Canon EF 600mm f/4L IS II USM in Yellowstone, is that Canon’s DPAF performance can be nearly indistinguishable form their standard PDAF performance.

That said, there’s some interesting behaviors that come up with DPAF that I don’t see in PDAF. One of these is that almost every test case, the lens hesitated at least once while focusing from infinity to the MFD, but didn’t do so when focusing the other way. The 24-70/2.8 was the least affected by this, with no hesitation in the dark testing and a very minimal one in the bright testing, but showed the stutter none the less.

This may indicate that the focusing algorithms are much more complicated than many suspect, as most lenses that hesitated seemed to do so around what might be expected to be reasonably normal focusing distances. In the case of the 24-70/2.8 it would hesitate with the focusing distance around 8 feet.

Since phase measurements provide direction and distance information, and are generally capable of determining focusing errors over a fairly substantial range, this may be a DPAF technique to aid in faster focusing. That is, in situations where it can’t ascertain focus, hesitate at a focus position to gather better data that’s close enough to where the lens is expected to be focused to shortcut the entire trip to MFD and back.

It would be interesting to see — I’ll have to make a note to test this — if the camera hesitates in the same position if the focusing sequence starts at MFD instead of infinity.

Conclusions

I’m going to leave this as is for the time being. I have some further work that I would like to investigate, namely the AF performance of Canon’s embedded phase pixel (as in the EOS M3) and Dual Pixel AF (as in the live view of the 5D mark IV) modes, as well as looking at the performance differences between the mark II and III teleconverters. However, it’s also clear to me right now that testing every lens I own at each of the extreme focal lengths, is a much larger undertaking than I’m willing to take on a the moment. Consequently, I have to consider just how far I want to take this investigation as I continue to update it.

I’m publishing this now in part so that the data is out there for others to build upon if they’re so inclined. I’ve never personally seen any comprehensive AF drive rates or other performance data, and I thought this might be interesting.

Comments

There are no comments on this article yet. Why don't you start the discussion?

Leave a Reply

Basic Rules:
  • All comments are moderated.
  • Abusive, inflamatory, and/or "troll" posts will not be published.
  • Extremely long comments (>1000 words) may be blocked by the spam filters automatically.
  • If your comment doesn't show up, it may have been eaten by the spam filters; sorry about that.
  • See the Terms of Use/Privacy Policy for more details.


Follow me on twitter for updates on when new comments and articles are posted.

Email Notice Details: By checking the above checkbox, you are agreeing to recieve one email at the email address provided with this comment, for the sole purpose of notifing you that the article author has been reseponded to your comment.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Our cookie and privacy policy. Dismiss