As part of my Master’s thesis, I proposed a model for automatic editing of 3D movies, that can learn from examples. Using a 3D virtual environment (CineSys) and a description of a scenario (i.e. timed actor actions), it casts the problem of film editing as a path-planning problem in a space of available cameras (pre-defined positions or automatically computed from actors positions). I defined cost functions for shot quality (e.g. shot type preferences, composition, visibility of actions), for cutting between shots (e.g. jump cuts, do not break the line of action), and for pacing. Each rule is associated to a weight: these weights can be automatically learned from manually produced movie examples, effectively representing the editing style of a particular user.

Mathieu Chollet
Lecturer - Virtual Social Interactions

My research interests include interactive virtual humans and multimodal behaviour understanding through social signal processing. My application domains include social skills training, healthcare and wellbeing, and team collaboration.