Friday, August 30, 2013

Canon EOS 70D Video Autofocusing Review: Stress Test


As pointed out in the ISO samples review, judging the Canon EOS 70D by its still image quality is a bit of an academic exercise. The advancements made in the sensor department have nothing to do with improving resolution or dynamic range. In fact that the 70D shows any improvement over the 60D in these areas is largely despite of the new Dual Pixel architecture, not because of it.

However, in playing with the live view and video autofocusing, the promise of faster and more reliable does come through. That doesn't mean that the 70D is prodigious as a video rig; it's more a case that the new focusing technology finally makes the traditional DSLR work as a usable videography machine....period.

Thanks once again to the kind folks at Broadway Camera

To date, the majority of the videos posted on the web have established that the 70D is fairly quick and accurate for live view focusing when it comes to still situations. In truth, the modern m4/3 cameras like the Olympus PEN E-P5 are actually faster, but the real promise of Canon's system is that phase detection means that the autofocus system can perform tracking autofocus... but does it do it well?

Here are two examples of the 70D in Movie AF Servo mode. Both videos are horrid by any videography standards, but I wanted to see how well the 70D could perform against unrealistic expectations.(You know, like how camera brand x will make noise about ISO improvements and the next thing you have is people comparing samples at ISO 25600) The first videos is a straight-up (pun intended, you can see that it wasn't shot level) test of the the system's ability to acquire and re-acquire focus once camera motion settles down.


As you can see, there is a bit of a delay before focus is acquired, but the behaviour is more like how the phase detection works for stills, and less like how contrast-detection works. It's not necessarily that it's blindingly fast (though it is fast enough for a more polished approach to camera panning) so much that the focus is reliable. The lens moves from out-of-focus to focus in one steady motion, and locks on without hunting back and forth.

Also interesting to note is that the camera takes a longer time to focus on the wall of boxes  in front than it does on the display case below. It's certainly something to explore further, but I would speculate that the the difference is because there is more detail for the camera to pick through on the wall, slowing down the AF processing time. One thing I did notice about the 70D's live-view AF is that it also suffers from the same maladies that conventional PDAF suffers from... main that it can front or back-focus relative to where you want

The second test is below: It's been affectionately labelled "the jerk" test around here... as in "jerking the camera around" or "shoot like a jerk"... or possibly a commentary on my videography skills. This is much more challenging than the first test as it involves more than just simple distance-to-subject calculations on th cameras part. The side-to-side and fore-aft motion mean that the camera has to continuously recognize if what it first locked on focus is still in focus, and adjust accordingly.


As you can see, even if I was overly aggressive with my panning motions, the Dual Pixel sensor isn't going to turn you into an action videography hero. Yes, it does track focus continuously, but in this case, it never holds focus during aggressive motions. However, it does re-lock focus once the motion settles down, which is more than what you could say for a conventional DSLR. If you had attempted this with a Nikon D7100 or the older Canon 60D, you would have ended up with a blurry unusable mess.

However, there was a rhyme and  reason for this test. The modern Nikon's with 3D tracking can track objects moving this speed fore/aft and across the frame, but for still photography. This was essentially a test to see if Canon could replicate that performance with video shooting. I would say that the 70D can't, but would concede that such performance might be too early to expect from this technology. Motion tracking is computationally intensive. Nikon's 3D tracking system makes use of both its PD autofocus arrays and the exposure meter array in order to track moving objects. That is because the Nikon system is looking at not only movement/distance, but also colour. Theoretically, Canon's system would be able to do this as well, since the entirety of the imaging sensor is available. However, for now, I would say that Nikon's 3D tracking is more successful at keeping up with moving objects for still photography than Canon's on-chip system is for videos. Not really comparable, but there's nothing else to compare them to.

All in all, I'm fairly enthusiastic about the 70D. Canon has been conspicuously understated about the performance of the new system since it was announced, and with these samples, you can see why. For an experienced video shooter, it's a substantial improvement, but it can also be a let down if you have unrealistically high expectations of what the new technology can do. However, there is a lot of creative potential, especially for non-big-budget content creators who would like to be freed from the confines of always relying on static shots, or  from having to do manual focus or use focus pullers. In fact, the 70D is almost like a big iPhone; with the touch-screen LCD, you can do focus-fulls simply by tapping on the display screen. All in all, the new PDAF system is a fairly good "Version 1.0" implementation, and shows a lot of promise for future iterations.

No comments:

Post a Comment