Autofocus

The EOS-1D X Mark III is likely to be the camera of choice for many of the world's best sports photographers, so it should come as no surprise that a lot of Canon's efforts have gone in to making the autofocus system even more capable.

  • Pixel grid AF module provides 191 AF points (155 of which are cross-type)
  • 400,000 pixel RGB+IR metering sensor paired with new algorithms improves subject tracking through the viewfinder
  • Machine Learning underpins Face and Body recognition/priority modes
  • Dual-pixel AF available in live view at up to 20 frames per second

OVF shooting

There were rumors that the next 1D series camera would be a DSLR/mirrorless hybrid. For reasons we'll come to, that's arguably true. It certainly seems to be a DSLR that's learning a lot of lessons from mirrorless cameras.

Captured using the OVF, with iTR Tracking and Case Auto. Out-of-camera JPEG.
Canon EF 24-70mm F2.8L II | ISO 16000 | 1/1000 sec | F2.8

As a 1D series DSLR, it's likely to primarily be shot using the optical viewfinder and a secondary, dedicated phase-detection autofocus sensor. Unlike any DSLR we've seen before, this sensor uses a grid of pixels, rather than a series of line-shaped sensor strips, making it similar to an image sensor. That means the autofocus system has more scene information to work with when trying to determine how offset the stereo pair of images from 'left-looking' and 'right-looking' (or up-and-down looking) AF detectors are, making the system faster and more accurate at determining how far to drive the lens to achieve focus.

The smaller, denser pixels essentially increase the resolution of the AF system, helping lift the number of autofocus points from 61 AF points to an impressive 191 points. 155 of these AF points are cross-type, sensitive to both horizontal and vertical detail. The increased resolution of the grid-type AF detectors also improves focus on low contrast subjects, and additionally allows for only slight variations in distance to the subject to be accurately detected.

The most sensitive AF points are sensitive down to -4 EV (a stop lower than the Mark II) and up to 21 EV (three stops higher than the previous model). If you mount a lens or lens/teleconverter combination with an F8 maximum aperture, you still maintain 191 AF points (65 of which remain cross-type).

Live view shooting

The 1D X III more closely resembles a mirrorless camera when you engage live view mode. It can shoot at up to 20 frames per second with full autofocus and autoexposure, using either fully mechanical or fully electronic shutter. This makes it nearly comparable with the best sports-shooting mirrorless model available, but despite Canon's promises of the lowest level of rolling shutter they've yet achieved, Canon have not yet matched the fast readout, and non-existent levels of rolling shutter, nor banding under artificial light, that the Sony a9 has achieved.

Image captured using eye detection in Live View.
Rokinon / Samyang AF 85mm F1.4 EF | ISO 100 | 1/1000 sec | F1.4

In live view mode, the 1D X Mark III's Dual Pixel AF system covers 90% of the scene's width and 100% of its height. The camera itself will chose any of 525 segments across the screen (compared to 143 on the EOS R), while you can manually select any of 3869 positions.

Autofocus in live view works with lenses with a maximum aperture of F11, allowing for combinations like an 800mm F5.6 lens with a 2x teleconverter. AF using the center point (AF-S) is rated down to -6 EV with an F1.2 lens, which likely means slightly lower than -4 EV with an F2.8 lens. This suggests similar low light AF performance as with OVF shooting when using an F2.8 lens, but better performance with faster lenses. In our limited experience thus far, we find that Dual Pixel AF typically fares better in low light.


Subject tracking and machine learning

Historically, Canon DSLRs have often fallen behind the competition when it came to subject tracking - the ability of the camera to automatically shift AF points as necessary to follow your subject around the frame. The 1D X Mark III offers significant improvements that aims to change that.

Canon's latest autofocus algorithms (used in both OVF and live view mode) have been developed using machine learning: 'teaching' the system to recognize subjects by showing it thousands of images. The camera itself doesn't learn, but its AF system has been taught to identify certain types of subject.

The AF system has been trained to identify human heads and eyes

Specifically, the camera has face and body detection and has a 'People Priority' mode that can tell the camera to focus on humans in preference to non-human subjects. Specifically, the AF system has been trained to identify human heads and eyes, and the company says the algorithms were trained using appropriate data sets that took into account the AF system in-use: optimizing for either the 400,000 pixel resolution of the RGB + IR metering sensor for OVF shooting or the image sensor for live view shooting.

Dual Pixel AF in live view has been significantly updated, which speaks well for Canon's future mirrorless offerings. The 1D X Mark II used only luminance, color, and face-detection data to understand and track a subject around the frame in live view, but the Mark III adds depth-of-field information as well as eye and head detection to better understand, and track, your subject. And when it comes to distant subjects that are difficult to isolate against their backgrounds using distance information alone, the AF system prioritizes luminance, color and pattern detection to understand your subject.


AF Configuration

Autofocus configuration has been simplified, compared to the 1D X Mark II. There are now four presets (which can then be fine-tuned to match your precise shooting requirements), rather than six. Canon says the default 'Case 1' mode will do a better job of handling a wide range of subject and subject movement.

The bigger change, though, is the addition of an 'Auto' setting. This analyses the movement of your subject and any other movement in your scene (other potential subjects arriving and leaving) and tries to adjust its settings to suit the situation.

Finally, AF area modes are now largely consistent between OVF and Live View shooting, save for 'Face+Tracking' mode, which is missing when shooting with the OVF.