About me

Hi there! 👋 I’m Stefan, an ML Engineer working on autonomous driving at Mercedes-Benz.

I have obtained my PhD as part of the Autonomous Vision Group at the University of Tübingen, supervised by Andreas Geiger. You can find my thesis here.

My research is at the intersection of machine learning and 3D Vision, especially motion-based self-supervised learning for 3D LiDAR sensors.

I strongly believe in the Bitter Lesson: In the long run, methods that can make better use of data and compute tend to outperform all the clever tricks we come up with as computer vision researchers. That said, I think motion-based self-supervised learning in 3D is a powerful shortcut for models to obtain a deeper world understanding: It scales with data and compute, while taking advantage of how things actually move and exist in 3D space without requiring any human annotations.

Please check out my related publications below:

Publications

• PhD Thesis: Learning 3D LiDAR Object Detection without Human Annotations
  • Stefan Baur 2025, Faculty of Science, University of Tubingen
• LISO: Lidar-only Self-Supervised 3D Object Detection
  • Stefan Baur, Frank Moosmann and Andreas Geiger
  • 2024 European Conference on Computer Vision (ECCV)
  • Paper
• SLIM: Self-Supervised LiDAR Scene Flow and Motion Segmentation
  • Stefan Baur^*, David Emmerichs^*, Frank Moosmann, Peter Pinggera, Bjorn Ommer and Andreas Geiger (^*: equal conribution)
  • 2021 International Conference on Computer Vision (ICCV), Oral
  • Paper
• Quantifying point cloud realism through adversarially learned latent representations
  • Larissa T. Triess, David Peter, Stefan Baur and J. Marius Zöllner
  • 2021 Proc. of the German Conference on Pattern Recognition (GCPR)
• PillarFlowNet: A Real-time Deep Multitask Network for LiDAR-based 3D Object Detection and Scene Flow Estimation
  • Fabian Duffhaus; Stefan Baur
  • 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
• Real-time 3D LiDAR Flow for Autonomous Vehicles (Oral)
  • Stefan A. Baur; Frank Moosmann; Sascha Wirges; Christoph B. Rist
  • 2019 IEEE Intelligent Vehicles Symposium (IV)

Misc

• Co-chair of the session “Range Sensing and Deep Learning” @ 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Software Projects

• RBRT: A lightweight Ray Tracer that I wrote to dive into raytracing and Rust.
  • fast: supports Intel AVX & SSE SIMD instructions
  • meshes can be loaded from .obj files
• Sudoku Solver:
  • uses constraint propagation and backtracking
  • Rust compiled to WebAssembly with React Frontend
  • try it out (runs completely in the browser)
• Snake: A terminal based snake game
  • no window manager or GUI required
  • cross platform: Latest Release (Windows & Linux).