This chapter deals with video applications. While methods applied to images can be applied to single frames of videos, this usually comes with a loss of temporal consistency. We will try to strike a balance between what's possible on consumer hardware and what's interesting enough to show and implement.

Quite a few applications should come to mind when talking about video, such as object tracking, event detection (surveillance), deep fake, 3D scene reconstruction, and navigation (self-driving cars).

A lot of them require many hours or days of computation. We'll try to strike a sensible compromise between what's possible and what's interesting. This compromise might be felt more than in other chapters, where computations are not as demanding as for video. As part of this compromise, we'll work on videos frame by frame, rather...