www.dpreview.com/reviews/pa...

DPReview 발췌입니다.

AF improvements on the S5 fall into two general categories: better and faster recognition of subjects using deep learning technology and improvements to the way DFD calculations are made.

To start, the S5 includes a head-detection algorithm in addition to face, eye and body recognition, so that the camera is less likely to struggle when a subject turns away from the camera. Additionally, it can now detect a face that takes up as little as 2.5% of the frame, versus 5% on other S-series cameras. Other tweaks allow the S5 to make autofocus calculations 60 times a second for bodies and faces, whereas the company's other models do so at 12 and 30 calculations a second, respectively.

Secondly, the S5 relies more on DFD in continuous autofocus than contrast detection relative to previous models. We'll go into more detail on this later in the review, but in use, this means the S5 shows less hunting and refocusing 'flutter', which was distracting and made it hard to follow moving subjects.

Where DFD has historically been most challenged is video, since any imperfections in continuous autofocus, like hunting or pulsing, will show up in your final work product. However, the benefits of DFD are still reduced in video mode because of its frame-rate limitations. While shooting video, the camera simply cannot sample the sensor the same way or at the same rate, as it can when shooting stills.

That's not to say there aren’t AF improvements in video; thanks to head detection and improved deep learning algorithms, the camera should still do a better job of identifying and sticking to a subject.