Uncertainty Quantification and Robustness in Recognition of Street Scenes using Neural Networks
- Lamarr
- RC Trust
Abstract: Neural networks set new state of the art in the recognition of street scenes, however they are easily fail when confronted with previously unseen out of distribution (OoD) data. OoD can have several meanings - like the same street scene at another time of the day or street scenes where OoD objects occur that belong to a semantic category that is disjoint from those that were present in the the training data. We consider methods that help to identify and localize such OoD objects in semantic segmentation. We discuss, how to robustify neural networks against the domain shift using vision language pre-training. We also consider retrival and unsupervised continual learning of previously unknown objects in street scenes.
Prof. Dr. Hanno Gottschalk
Bio - Hanno Gottschalk studied physics and mathematics at the universities of Freiburg and Bochum. In 1999, he obtained his PhD in mathematical physics at U Bochum under the supervision of Sergio Albeverio. After a stay as DAAD fellow at the university La Spienza in Rome, Hanno Gottschalk worked as a PostDoc and lecturer at U Bonn from 2000 to 2007, where he obtained the venia legendi in mathematics in 2003. In 2007, he joined Siemens Energy as core competency owner for probabilistic design in gas turbine design. In 2011 he was appointed as associate professor for stochastics at U Wuppertal, where in 2018 together with Anton Kummert he became founding speaker of the Interdisciplinary Center of Machine Learning and Data Analytics (IZMD). After 12 years of research and teaching at U Wuppertal, Hanno Gottschalk in April 2024 became full professor at TU Berlin. Since then he leads the working group of Mathematical Modeling of Industrial Life Cycles. Hanno Gottschalk is Member of the center of excellence Math+ and the Werner von Siemens Center and the NVIDIA advisory board for machine learning, physics and fluid dynamics . In his research, he combines applied mathematics and computer vision.