Gabriele Goletto

With the advent of machines capable, in principle, to cooperate with humans in performing tasks of daily living, it becomes crucial the development of methods to quickly and easily infer human actions from common data. Egocentric vision is rapidly becoming one of the most promising source of information for this purpose, demonstrating impressing capabilities in pose estimation, action recognition, anticipation, retrieval, and many more. However, often these models suppose very controlled scenarios, while the major challenges for bringing egocentric vision into the wild are seldom considered overall. The aim of this project is to work on the boundaries that egocentric vision models should consider for realistic applications, namely the dimensionality of the model, the real-time of inference, the potential to easily generalize across conditions (i.e. new spaces, users, tasks), and the capability to work on continuous stream of data.