Abstract: In this talk I will give an overview about the state of non-rigid correspondence methods in the era of deep learning, where current methods shine and what limitations still exist today. These range from the question whether a found solution is actually the global optimum, down to how fine-grained details in shapes from different classes can be matched to each other. Finally, we will look at the case of partial-to-partial correspondence in which no full mapping exists from one shape to another and the additional problem of estimating the overlaps poses significant challenges.

Short-Bio: Zorah Lähner is an assistant professor (tenure track) and head of the “Geometry in Machine Learning” group at the University of Bonn, and part of the Chair for Visual Computing and the Lamarr Institute. She received her PhD from the Technical University of Munich and was a postdoctoral researcher at the University of Siegen. She is interested in how geometric properties can be used to form effective priors and guide optimization processes in geometric deep learning and 3D computer vision applications.

--
Comunicaciones DCC