“Machine Learning for Event-cameras”
Amos Sironi
Chief Machine Learning Scientists
PROPHESEE
Event-based cameras encodes visual information in a sparse and asynchronous stream of events, corresponding to log-luminosity intensity changes in the scene. By transmitting only changes, event-based cameras uniquely combine high temporal resolution, power and data efficiency.
However, to apply conventional machine learning methods on event cameras, one has to turn the asynchronous stream of events into a frame-like representation. This results in a loss of the power efficiency and non-redundant representation of the event data.
In this talk we will first present current advances in event-based technology and machine learning applications. Then we focus on alternative machine learning architectures, designed to fully exploit the properties of event-based data.