Autofocus

Key Takeaways:

  • AF system is relatively straightforward, though face/eye detection is prone to false positives and could be better implemented
  • Subject tracking is 'stickier' than the X-T3 and shows clearly what's being focused on, letting you set your focus point and recompose.
  • Subject tracking appears to be quite heavily weighted towards color so performance varies with subject and background
  • Even with fine-tuning of the system response, the X-T4's autofocus finds subjects with unpredictable acceleration/deceleration challenging, again giving very situation-dependent results.

Autofocus system

The X-T4 has an updated version of the one from the X-T3, with the biggest change being the promise of improved subject tracking, including in low light.

Beyond that it'll be familiar to existing Fujfilm users: you can select from a single AF point, a larger zone or the entire AF region, with a combined 'All' setting that lets you scroll through all these modes as you adjust your AF point size. In the largest of these modes, S-AF gives you 'Wide' mode, where the camera chooses a subject based on proximity and centrality within the frame, while in C-AF you get 'Tracking' mode, which will follow whatever is under your initial AF point around the frame.

Fujifilm's face detection system is rather prone to false-positives. The camera found a face in this scene, which initially over-rode the chosen AF point.

ISO 160 | 1/750 sec | F5.6 | Fujinon 16-55mm F2.8 lens @ 16mm
Photo: Richard Butler

Separate and on top of this there's a Face/Eye detection system, that can be toggled if you've assigned it to a button. This ignores your chosen AF point and prioritizes anything it thinks is a face (which in some instances isn't actually a face: it's a system rather prone to false positives). Depressing or nudging the AF joystick overrides the Face-Detection system.

There's also a Face Selection system, which needs to be assigned to a button and re-engaged each time you turn the camera on. This lets you choose between faces if there's more than one in your scene. It ignores your AF point and prioritizes the face nearest the camera, initially, letting you use the joystick to swap between the detected faces. If your subject looks away or the camera briefly stops identifying their face, it will jump to other faces in the scene.

This too can be over-ridden by pressing the joystick inwards, making it easy enough to switch between Face Selection and your chosen AF mode. It's not clear to us why Face Selection needs to be an awkward-to-engage extra mode, rather than being the default behavior of face detection.

Autofocus performance

The first part of our autofocus assessment looks at how well the camera can refocus on an approaching subject, using a single, central AF point.

At its default settings the camera did a go job of refocusing once the rider had reached an even speed but would miss focus by a little as the target initially accelerated up to that speed. We shot the sequence again with the camera set to be more responsive to changes in acceleration and explicitly told to focus on the nearest subject. If anything this made the results during the acceleration phase a little worse.

Close-up shows a magnified view of the selected central AF point

Subject tracking

We then look at the camera's subject tracking to see how well it can identify a subject and automatically choose focus points to follow and focus on it. We don't use face detection in this test and selected an AF point over the subject's body (the face being too small to select, at these distances).

One thing that became immediately apparent in our testing was that the latest iteration of Fujifilm's AF Tracking system is heavily weighted towards subject color. The focus point would tend to lag a little behind the subject and, if the rider was wearing a fluorescent yellow top, would then lock onto a light green/yellow point in the background. This would happen even when the AF was told to prioritize the front-most subject, and stopped happening as soon as we changed the rider's outfit.

This essentially rules-out the use of AF tracking for team sports, where members of the same team are likely to be dressed similarly, but also adds a degree of uncertainty with any subject: performance will be unpredictable if the background contains similar colors to your subject.

Test run shot in C-AF mode 3 'For accelerating/decelerating subject.' We also conducted test runs with the camera adjusted to be more responsive and more sensitive to changes in speed, but got similar results.

Close-up shows the a representative region of the body, which was the originally selected target.

When shooting a subject that couldn't be confused with the background, the camera did a better job of subject recognition. However, as with the first part of the test, we couldn't find a combination of settings that could cope with the inherent acceleration and deceleration of the subject through the turns, meaning all our sequences pulsed between perfect focus and slightly back-focused throughout every run. This remained true when we dropped the camera to 10 fps.

However, it's worth noting that most of the misfocused shots are only a fraction out of focus (the distinction only becomes really apparent when you directly compare the near-misses to the perfectly in-focus shots).

At close quarters

We also tested how well the X-T4's AF performed at closer-range and in lower light (dimmed LEDs in this instance). To a great extent low light appears to exaggerate the existing weaknesses of the face/eye detection: that the system appears to have a slightly simplistic understanding of what a face looks like.

In good light, the camera will occasionally briefly 'lose' the face it's looking at – especially if your subject looks down or away – and will jump between focusing on the near and more distant eye. In low light this becomes more extreme, especially if the face isn't lit from behind the photographer. Face/eye detection still finds its subject and can position focus correctly, but the reliability drops considerably: the boxes indicating a face and eye flicker on and off and dance around as the camera loses and regains its subject.

We found around 1/3 of the images we shot at around 2 EV were not using a focus point near the eye by the time the shutter had been pressed. In addition, Continuous AF mode pulsed the AF constantly as it was trying to find its target. We found we got the best results in single AF.

Overall

We were a little disappointed by how the X-T4 performed in our AF tracking tests, given how good the very best rivals are. However, this should be put in context: the camera's subject recognition is better than its predecessor, meaning it's now possible to casually point the AF target at your subject, half-press and recompose, rather than having to mess around manually moving your AF point.

Even for shooting action the system can be very effective, and for subjects moving at predictable speeds (e.g. sprinting, motorsports) it can do an excellent job. Even when it's challenged by more complex movement, the results tend to be pretty good and the 15fps bursts mean you'll get a good number of well focused shots (albeit not every shot, such that you the best expression or decisive moment might not be pin sharp).

Unfortunately, the X-T4's tracking ability isn't well-integrated with its face detection system

Unfortunately this ability isn't well integrated with the face detection system: you can't use your chosen AF point to specify which face you want to focus on. And, given the system's habit of finding faces that don't exist, and losing ones that do, you will need to be ready to override the camera's decisions.

Overall, then, our findings are that the X-T4's subject tracking performance and, more generally, continuous AF performance as a whole, are distinctly subject-dependent. It works well in a wide range of situations but isn't as simple and dependable as the best of its peers.