Powered by Sistema ARGO

Samuel Rota Bulò

Mapillary, Graz, Austria
Samuel Rota Bulò
Senior researcher

Samuel Rota Bulò is a senior researcher with Mapillary Research since 2017, working mainly in the fields of computer vision and machine learning with particular emphasis on deep learning methods. Additional research interests are in the fields of optimization, graph theory and game theory. He received his PhD in computer science from the University of Venice in 2009. He worked there as a PostDoc until 2013, holding also teaching positions. He then became a researcher for Fondazione Bruno Kessler in Trento until he switched to his current position in 2017. He regularly publishes works in top journals and conferences in the fields of computer vision and machine learning. He was awarded the prestigious Marr Prize in 2015, Best Student Paper Award at ICIAP 2017, Honorable Mention Award at TASK-CV 2018 and Best Paper Award at Valuetools 2011 . He got the National Academic Qualification as Full Professor in 2018. He won the Robust Vision Challenge (CVPR2018) and the Autonomous Navigation in Unconstrained Environments Challenge(ECCV2018) both on the semantic segmentation track. He serves on the editorial board for “Pattern Recognition” and “International Journal of Machine Learning and Cybernetics” and is regularly on the program committee of international conferences of his field. He participated to several EU projects (SIMBAD, VENTURI, REPLICATE).

Mapping on a Global Scale with Deep Learning and Computer Vision
Abstract: In this talk, I will introduce the challenges underlying the automatic creation of street-level map data at a global, world-wide scale level using millions of images that are collaboratively contributed from any capture device (different camera types, resolutions, image qualities, non-calibrated, etc.) to the Mapillary platform. I will discuss the way deep learning currently contributes to advancing the technology towards the grand goal of automatic mapping at a global scale. Specifically, I will give details about the state-of-the-art, computer vision and deep learning methods that we have developed and published at top conferences over the last couple of years and how those find practical application in the process of automatizing map data generation from street-levelimagery. I will conclude by sketching potential new directions that can be taken in the future.