Torsten Sattler

About

Torsten Sattler is a prominent computer vision researcher whose work centers on visual localization, camera pose estimation, and autonomous perception systems. He has made foundational contributions to understanding how cameras can determine their precise position and orientation within known environments — a capability critical for self-driving vehicles, augmented reality, and robotic navigation. Sattler's most influential work, "Understanding the Limitations of CNN-Based Absolute Camera Pose Regression" (2019, 401 citations), rigorously interrogated the reliability of deep learning approaches to localization, revealing critical shortcomings that reshaped how the community evaluates pose estimation methods. His research spans both classical geometric techniques and modern learned approaches, including essential matrix-based localization and mesh-based representations, reflecting a nuanced perspective on when learning is — and isn't — appropriate. Beyond localization, Sattler has contributed to obstacle detection for autonomous vehicles using monocular cameras, fisheye-stereo visual odometry, and embedded real-time stereo systems, demonstrating a broad command of practical perception challenges. His work on semantic consistency for long-term localization further highlights his attention to robustness in real-world, changing environments. With hundreds of citations across a decade of research, Sattler has established himself as a rigorous and influential voice in geometric computer vision.

Research Focus

Key Achievements

Top Papers

1. 1
   Understanding the Limitations of CNN-Based Absolute Camera Pose Regression

   401 citations · 2019
2. 2
   Semantic Match Consistency for Long-Term Visual Localization

   150 citations · 2018
3. 3
   Obstacle detection for self-driving cars using only monocular cameras and wheel odometry

   93 citations · 2015
4. 4
   To Learn or Not to Learn: Visual Localization from Essential Matrices

   89 citations · 2020
5. 5
   Hybrid Scene Compression for Visual Localization

   65 citations · 2019
6. 6
   MeshLoc: Mesh-Based Visual Localization

   53 citations · 2022
7. 7
   Direct visual odometry for a fisheye-stereo camera

   49 citations · 2017
8. 8
   Minimal Solvers for Generalized Pose and Scale Estimation from Two Rays and One Point

   45 citations · 2016
9. 9
   Is This the Right Place? Geometric-Semantic Pose Verification for Indoor Visual Localization

   45 citations · 2019
10. 10
    Embedded real-time multi-baseline stereo

    28 citations · 2017

Key Collaborators

Contact & Links