Integrating a Physics Engine with Deep Neural Networks
This project was made during Jean Ponce's course "Object recognition and computer vision" (M2 MVA 2016 - 2017). We aimed at integrating a Physics Engine with Deep Neural Networks with weak supervision.
More precisely, we implemented a way to automatically annotate physics videos via baysian inference of the parameters of a known physical system, e.g a simple pendulum.
Used several tutorials:
http://www.petercollingridge.co.uk/pygame-physics-simulation/boundaries for Pygame physics simulation and: