Keynote: Geometry Processing and Learning for 3D Modeling of Complex Scenes
Geometry Processing and Learning for 3D Modeling of Complex Scenes
Geometry Processing is a field of research aiming to design and implement efficient data structures and algorithms for the acquisition, reconstruction, analysis and transmission of complex 3D models. In the first part of my talk I will give my perspective to Geometry Processing. I will argue that streamlining the geometry processing pipeline cannot be achieved by direct adaptation of existing signal processing methods: a focused research phase is required to address such fundamental issues as the reconstruction and approximation of complex shapes, in order to develop techniques that are robust to defect-laden and heterogeneous inputs. The quest for robustness to these imperfect data has motivated variational methods and more recent approaches inspired by the theory of optimal transportation. In the second part of my talk I will review some recent work carried out in the Inria “Titane” project-team, on geometric modeling of large-scale scenes, from aerial or satellite imagery. I will show how we combine geometric computing, geometry processing and machine learning to reconstruct landscapes and cities in 3D, with semantic attributes and well-defined levels of details conforming to standard information models such as CityGML.