Diving between depth prediction and depth completion

Depth perception matters in many different industrial and scientific applications as self-driving cars and augmented reality. The raise of artificial intelligence usage in the last decade has also opened new horizons leading to effective techniques to predict depth from cheap and robust single cameras moreover active depth sensing techniques as LiDARs are becoming cheaper and miniaturized to the point that they can be installed into smartphones leading to robust depth completion methods. This thesis aims to increase the knowledge in the field analyzing the relationship between deep learning applied to depth prediction and depth completion by means of the proposal of architectures capable to perform both tasks and more.