Olaf Dünkel

I am a third-year Ph.D. student at the Max Planck Institute for Informatics and Saarland University advised by Dr. Adam Kortylewski and Prof. Christian Theobalt. I am part of ELLIS, co-advised by Prof. Christian Rupprecht (VGG, University of Oxford).

I am interested in questions at the intersection of representation learning, generative models, and embodied AI. My current research investigates how generative models can improve robustness in computer vision by leveraging learned priors and controllable data acquisition for training and evaluation. I also work on representation learning methods that better capture real-world correspondences. More broadly, I am interested in foundational questions that make intelligent agents reliable under distribution shift. Previously, I worked on safe ML, autonomous driving, and robotics.

Prior to my PhD, I interned at Porsche and VITA (EPFL), working on autonomous driving, and I pursued my master thesis at KIT, advised by F. Pfaff and T. Salzmann. I obtained my Bachelor's in EECS and my Master's in Signal Processing and Robotics from KIT.

Email  /  Scholar  /  LinkedIn  /  Bluesky  /  X  /  GitHub

profile photo

Updates

  • November 2025: C3PO was accepted to 3DV 2026.
  • October 2025: Happy to be a "Top Reviewer" at NeurIPS 2025!
  • August 2025: Received the best poster award at the ELLIS Doctoral Symposium 2025 in Warsaw.
  • June 2025: Two papers accepted to ICCV 2025 (DIY-SC and CNS-Bench). See you in Honolulu!
  • Feburary 2025: Our work Common3D was accepted to CVPR 2025.
  • July 2024: Invited talk at the 3rd Virtual Symposium on Directional Statistics.
  • Feburary 2024: My master's thesis paper HuProSO3 was accepted to CVPR 2024.
  • November 2023: I started my PhD as an ELLIS student at the Max Planck Institute for Informatics.

Research
C3PO C3PO: Canonicalization of 3D Pose from Partial Views with Generalizable Correspondence Features
Y. Chi, L. Sommer, O. Dünkel, D. Muhle, D. Cremers, C. Theobalt, A. Kortylewski
3DV, 2026.
Paper
AttentionChains Attention (as Discrete-Time Markov) Chains
Y. Erel*, O. Dünkel*, R. Dabral, V. Golyanik, C. Theobalt, A. H. Bermano
NeurIPS, 2025.
Project Page / Paper / Code
CNS-Bench CNS-Bench: Benchmarking Image Classifier Robustness Under Continuous Nuisance Shifts
O. Dünkel, A. Jesslen*, J. Xie*, C. Theobalt, C. Rupprecht, A. Kortylewski
ICCV, 2025.
Project Page / Paper / GitHub
DIY-SC Do It Yourself: Learning Semantic Correspondence from Pseudo-Labels
O. Dünkel, T. Wimmer, C. Theobalt, C. Rupprecht, A. Kortylewski
ICCV, 2025.
Project Page / Paper / GitHub
Common3D Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space
L. Sommer, O. Dünkel, C. Theobalt, A. Kortylewski
CVPR, 2025.
Project Page / Paper / GitHub
Normalizing Flows Normalizing Flows on the Product Space of SO(3) Manifolds for Probabilistic Human Pose Modeling
O. Dünkel, T. Salzmann, F. Pfaff
CVPR, 2024.
Paper / GitHub
Joint Vehicle Trajectory Joint Vehicle Trajectory and Cut-In Prediction on Highways using Output Constrained Neural Networks
M. Brosowsky, P. Orschau, O. Dünkel, P. Elspas, D. Slieter, M. Zöllner
IEEE Symposium Series on Computational Intelligence, 2021.
Paper
Output Constraints Sample-Specific Output Constraints for Neural Networks
M. Brosowsky, F. Keck, O. Dünkel, M. Zöllner
AAAI, 2021.
Paper

Recent Positions

  • July 2022 - October 2022:
    Research Internship Vehicle Trajectory Prediction at VITA (EPFL)
  • November 2019 - April 2020:
    Bachelor's Thesis on Uncertainty Estimation in Vehicle Trajetory Prediction at Porsche AG
  • April 2019 - August 2019:
    Intern for Autonomous Driving and Deep Learning at Porsche AG

Miscellaneous

I am a passionate French horn player and I am regularly playing in various symphony orchestras. Music not only brought me to various places across the world but it keeps being a great regular joy of my life.
Other than that I do like sports, especially long distance running, swimming, hiking, and volleyball.

This website is based on Jon Barron's website, whose code is available on Jon Barron's GitHub page.