Computer Vision News - February 2023

18 Congrats, Doctor Larissa! Autonomous vehicles require a detailed understanding of their environment. Therefore, vehicles are equipped with a variety of sensors, such as cameras, LiDARs, and RADARs. They capture the surrounding and real-time neural networks then predict the location and type of objects in the scene, such as vehicles, persons, traffic lights [ Triess2020IV ] [ Triess2021NeurIPS ] . The training of the neural networks is usually done with large amounts of annotated data. Ideally, it covers the application domain as good as possible. However, there are various effects that can cause domain shifts. Figure 1 shows a selection of such domain changes that lead to impaired performance in perception models [ Triess2021IV ] . It is not feasible to acquire data and annotations for each domain, especially if it involves safety- Figure 1: Examples of domain shifts for autonomous vehicles. Larissa Triess has recently finished her PhD at the Karlsruhe Institute of Technology in cooperation with the Mercedes- Benz AG. Her research focused on domain adaptation and 3D LiDAR data generation to improve perception algorithms for autonomous driving. Larissa is now a Machine Learning Engineer in the autonomous driving department at Mercedes where she continues her work on 3D scene understanding. Congrats, Doctor Larissa!