Dr. Florent Lafarge
In this talk, I will present some recent works to extract geometric structures from images and 3D data. These geometric structures are, for instance, planar graphs, polygons or polyhedral meshes that can be used for a variety of vision tasks. I will first introduce Delaunay Point Process, a stochastic framework that simultaneously locates and groups geometric primitives (line-segments, triangles) to form extended structures. Applications of Delaunay point processes include line-network extraction, object contouring or image compression. I will then present a kinetic data-structure for partitioning images into meaningful polygons, a much more compact representation than traditional superpixels. Finally, I ll propose a plane slicing mechanism that allows to efficiently reconstruct a 3D object as a piecewise planar surface given a set of detected planar primitives. Bio: Florent Lafarge is a research scientist at Inria Sophia Antipolis. He received his Ph.D. in applied mathematics from Ecole des Mines ParisTech in 2007 and his habilitation from University of Nice in 2014. His research topics include surface reconstruction and approximation, urban scene analysis, object recognition and probabilistic modeling in Vision and Graphics. He is a developer of the CGAL library.