Title of the Report: Understanding point clouds through geometric decriptors and natural language

Presenter: Dr. Fabio Poiesi

Affiliation: The Italian FBK Research Institute

Date of the Report: December 26th, 2024 (Thursday) at 16:00

Location of the Report: Room 1104, Building A, Feicui Science and Education Building

Abstract: Three-dimensional artificial perception enables robots to comprehend their environments and interact with objects to accomplish human-directed tasks. This talk explores novel approaches to point cloud understanding through the integration of geometric descriptors and natural language processing. We examine applications in robotic navigation and manipulation, with particular emphasis on four key downstream tasks: point cloud registration, 6D object pose estimation, functional segmentation, and dense captioning.

Biography of the Presenter: Fabio Poiesi is the Head of the Technologies of Vision Lab at Fondazione Bruno Kessler in Trento, Italy. He completed his PhD and Postdoctoral research at Queen Mary University of London. He joined FBK in 2016 as a Research Scientist and became a tenured researcher in 2023. His work focuses on developing computer vision and deep learning solutions for both research and real-world industrial applications. Fabio Poiesi has been working on algorithms for multi-object tracking and target behavior understanding since 2010. At FBK, he works on 3D scene understanding techniques for point cloud registration and understanding, using geometric techniques and natural language processing, which he has particularly applied to robotic perception for object manipulation. He has developed world-leading solutions for object 6D pose estimation, winning the popular BOP Challenge multiple times against competitors like NaverLabs, NVIDIA, and META. He is a member of the ELLIS network.