Is AVCapture perform face tracking even if metadata output has not been added

I've noticed that my Catalyst app uses around 65% CPU when capturing from the camera, even when everything else is idle. Disabling camera capture drops the CPU usage to 0 (middle of graph in screenshot). I profiled using Instruments, and found that the most heavy stack trace involved face tracking (see attached screenshot), even though no metadata output was added for the camera. Is this a bug in AVFoundation?