Making machines see and think in 3D

On Friday 13 December at 9 a.m. in the AULA BLU (CNR headquarters, Via Moruzzi), the seminar ‘Making machines see and think in 3D’ by Dr. Benjamin Busam (Technical University of Munich - TUM) is scheduled.

Abstract

Our world is 3D. Machines, however, often rely on 2D pixel data, facing significant challenges in perceiving and interacting with our spatial environment. This talk explores recent advancements in 3D perception, focusing on photometric challenges and trends in object pose estimation. It highlights most recent methodological advances to generalize 6D object pose estimation towards unseen objects in sight of 3D perception obstacles. New datasets addressing limitations in accuracy and scalability for indoor 3D spaces are discussed. Multi-modal depth data combining LiDAR, ToF, stereo, and RGB+P sensing is mentioned, alongside methods for generating high-accuracy ground-truth depth maps. Potential solutions are proposed to objectively benchmark dense 3D vision, 6D pose estimation, and robotic grasping tasks involving everyday objects of varying complexity - and we try to show potential future directions for the field.

Short Bio

Benjamin Busam leads the Computer Vision Group, I16 at the Technical University of Munich (TUM). Formerly Head of Research at FRAMOS and 3D Vision Lead at Huawei Research London, he studied Mathematics and Physics at ParisTech, Melbourne, and TUM, graduating with distinction in 2014. His research focuses on 2D/3D vision for pose estimation, depth mapping, and multi-modal sensor fusion. Ben received the EMVA Young Professional Award 2015 and Noah's Ark Innovation Pioneer of the Year 2019, alongside multiple Outstanding Reviewer Awards at top computer vision conferences. He is an active program committee member for the leading international vision and robotics meetings.