Efficient Machine Learning for Augmented Reality on Microsoft HoloLens
Many of the current and upcoming mixed reality devices have features that rely on computer vision and machine learning. Since all of those devices operate at high framerates and the content itself requires a lot of resources, latency and power consumption are critical factors. During my presentation, I will talk about how novel sensors and synthetic training data allow us to use very small and efficient neural networks to perform tasks that often require much more resources. The main focus will be on hand tracking for the next generation HoloLens device.
Bio
Marek Kowalski is a Scientist in the Microsoft’s Human Understanding team in Cambridge, where he works on HoloLens. In 2018 he completed his PhD at the Warsaw University of Technology. Marek’s dissertation was about robust methods for facial landmark localization (face alignment), his main research interests are in computer vision and machine learning applied to computer vision. In 2016 and 2017 Marek was an intern at Microsoft Research in Redmond, where he worked on 3D reconstruction for telepresence on Microsoft HoloLens (Holoportation). In the past he also worked on 3D reconstruction using multiple Kinect sensors and on face recognition. To see some of the projects Marek worked on, visit his GitHub page: https://github.com/MarekKowalski/