Continuous learning in computer vision

In this thesis we focus on continuous learning, and specifically on continuous learning in the context of computer vision. Computer vision aims at interpreting the world from its visual dimension, in an automatic manner. The world in general is characterized by continuity, and so is the visual world in particular. Therefore, this thesis delegates the task of coping with the continuous aspects of the computer vision problems approached, to the learning algorithms. Rather than discretizing the problems in our pursuit to find a solution, we focus on the continuity and ways in which we can learn it or integrate it in our proposed models. Each one of the chapters of this thesis aims at learning continuous aspects of the world such as motion, or relative entities such as “goodness” and “importance”, or concentrates on the problem of learning continuous functions, on the whole.