By Carey Rose - Published 1/24/19 - Experience based on a pre-production camera


Out-of-camera JPEG.
ISO 1600 | 1/1000 sec | F4.5 | Olympus 40-150mm F2.8 + 1.4x teleconverter

As water pooled on the camera, I held my composition. A motorcyclist rocketed into the frame and I watched as the camera immediately drew a box around the rider's helmet and began focusing as I half-pressed the shutter. As he screamed through the scene, leaving chunks of flying red mud in his wake, I fired off a burst at 15 frames per second.

Thanks to the Olympus E-M1X's new 'Deep Learning' autofocus, well, I didn't need to think about autofocus. I didn't need to follow the motorcyclist with a group of points, or time a half-press just right as he entered a pre-selected zone. I didn't even have to place a focus point over him and initiate tracking.

The camera immediately drew a box around the rider's helmet and began focusing

With so much less to think about, I found myself shooting with tighter framing than I normally would. That's crucial for sports and action, because tighter framing means less cropping in 'post' and better overall image quality. And considering that I was shooting at a fairly high ISO value on a Four Thirds sensor, I wanted to maintain the best image quality I could get.

There's no doubt that between Nikon, Canon and Panasonic's product announcements, 2018 was the year of the full-frame mirrorless camera. But with the E-M1X and its 20MP Four Thirds sensor, Olympus is taking a different approach. Many will cry foul, saying that this sizable, double-grip sports camera is evidence of Olympus losing their way.

Image processed in-camera from Raw.
ISO 200 | 1/80 sec | F5.6 | Olympus M.Zuiko 40-150mm F2.8 + MC-14 1.4x teleconverter

On the contrary, I think the E-M1 X is evidence of Olympus charting a new path in this ever-more-competitive world occupied by computationally capable smartphones and the inescapable allure of full-frame sensors and optics. And I think that new path shows a lot of promise.

Full disclosure: Olympus flew me down to Orlando, put me up in a nice hotel, fed me some tasty meals and handed me a pre-production E-M1X to test out. All images and impressions here are based on that pre-production camera, and performance and image quality may change with final production firmware.

What is the E-M1X?

The E-M1X is aimed straight at sports and action-shooting professionals, particularly those looking for lots of reach but not wanting to haul huge full-frame telephoto lenses around. In addition, it offers lots of direct control, durability, and highly configurable autofocus.

But there's no question: that Four Thirds sensor, inside this large camera body, looks a little silly. And though it offers great image quality for its size, you'll have noisier images and deeper depth-of-field than you would with a larger APS-C or full-frame sensor given the same exposure parameters. That's just the way it is.

On the other hand, between its new hand-held high-resolution mode, crazy effective image stabilization and new autofocus tricks, the E-MX is an exciting piece of innovation.

You see, the E-M1X is among the first 'traditional' cameras from a 'traditional' camera manufacturer to make use of 'deep learning' and true computational photography techniques like we've been seeing on high-end smartphones (Sony's latest Real-time Tracking AF is similar...but different). And that is exactly what traditional camera manufacturers need to stay relevant in today's shifting market.

Deep learning autofocus

Out-of-camera JPEG.
ISO 200 | 1/60 sec | F5 | Olympus M.Zuiko 40-150mm F2.8 Pro

Does this new intelligent autofocus actually work? Yes and no. Let's start with the 'yes' bit.

With the 'Motorsport' setting enabled, the camera would reliably detect both cars and motorcycles in a given scene. If you set the camera up to have a single autofocus area, you can place that area over a particular subject to manually choose which car or motorcycle you want the camera to focus on.

As I said earlier, this has the potential to be an incredibly powerful tool. I am by no means an experienced sports shooter, but this new autofocus system really did leave my mind free to focus more on composition, anticipating subject movement, and improving my panning technique.

Happy with the pan, less happy with the rain on the lens. Processed in-camera from Raw.
ISO 200 | 1/50 sec | F16 | Olympus M.Zuiko 12-100mm F4 Pro

Now for the 'no' bit. This is a pre-production camera, of course, but there are bursts where the camera positively identified a subject, indicated it was tracking, and then I ended up with soft or out-of-focus shots. Also, I found when shooting NASCAR that if the car is temporarily blocked by an object, the camera effectively gives up and you have to re-engage tracking when it appears again. Engineers told me that adjusting my C-AF Sensitivity may help combat this, but we'll have to confirm that when we get final firmware.

But my biggest gripe concerns usability. The function lever to the right of the viewfinder allows you to switch between AF modes (single, continuous, etc) and/or AF areas with one flick. But it doesn't allow you to switch between face detection or deep-learning settings.

While you can at least change face detection if you assign a button to AF area, you must enter the menus to enable or disable the deep learning autofocus. This strikes me as something of an oversight, particularly since Face Detect overrides all other settings and the deep learning modes sometimes 'see' motorcycles and cars in everyday objects. In any case, I hope this gets fixed in subsequent firmware revisions.

The rest

Out-of-camera JPEG, hand-held high-res shot. Notice how the palms lose some detail due to motion, and the ghosting of the individual in the lower right.
ISO 200 | 1/250 sec | F5.6 | Olympus M.Zuiko 12-100mm F4 Pro

The other feature I'd like to touch on is the hand-held high-res shot. It really, actually, works. It aligns and stacks a total of 16 images, giving you more resolution and lower noise levels. And unfortunately, you get ghosting and / or a loss of detail on moving subjects.

Because you can't use it on moving subjects, it's less robust than, say, the Google Pixel 3, which stacks up to 15 images each time you press the shutter and has no shutter lag. But hey, this is Olympus taking a step in the right direction, and it's only going to improve.

Olympus is thinking of computational applications for 'real' cameras

And this is the big takeaway. Not only is the hand-held high-res shot showing Olympus moving toward offering you bigger-sensor image quality, but the deep-learning autofocus shows that they are thinking of computational applications for experienced users of 'real' cameras. They're actively looking at ways to improve the experience of using a traditional camera through software.

Olympus could have jumped on the full-frame bandwagon and called it good (the marketing would likely have been easier). But I honestly think this path they've chosen is the braver, if riskier, one, and I'm looking forward to seeing how it pushes the rest of the market forward.