Stefan Roth, PhD

Short info

My research focuses on machine learning approaches to understanding and analyzing digital images and videos. My lab and I develop new deep learning models and methods for scene understanding, motion estimation, image editing and synthesis, video analysis, image restoration, and more. We aim to make such approaches robust to a broad range of real-world conditions. To that end, we are incorporating inductive biases through combining deep learning with classical models, endowing deep networks with explicit representations of uncertainty, or adapting pre-trained models to changing test-time circumstances. We also aim to reduce the dependency on labeled data by developing semi-supervised and self-supervised learning pipelines.

Open Science

Hahn, O., Araslanov, N., Schaub-Meyer, S., & Roth, S. (2024).
Boosting Unsupervised Semantic Segmentation with Principal Mask Proposals.
arXiv preprint2404.16818.

Hesse, R., Bağcı, D., Schiele, B., Schaub-Meyer, S., & Roth, S. (2025).
Beyond Accuracy: What Matters in Designing Well-Behaved Models?.
arXiv preprint arXiv: 2503.17110

Zöngür, B., Hesse, R., & Roth, S. (2025).
Activation Subspaces for Out-of-Distribution Detection.
arXiv preprint arXiv: 2508.21695.

Articles

Bahmani, S., Hahn, O., Zamfir, E., Araslanov, N., Cremers, D., & Roth, S. (2022).
Semantic Self-adaptation: Enhancing Generalization with a Single Sample. Transactions on Machine Learning Research (TMLR).

Endres, J., Hahn, O., Corbière, C., Schaub-Meyer, S., Roth, S., & Alahi, A. (2025).
Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model.
Proc of the IEEE/RSJ International Conference on Intelligent Robots and Systems.

Endres, J., Hahn, O., Corbière, C., Schaub-Meyer, S., Roth, S., & Alahi, A. (2025).
Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model.
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).

Gong, X., Hahn, O., Reich, C., Singh, K., Schaub-Meyer, S., Cremers, D., & Roth, S. (2025).
Motion-Refined DINOSAUR for Unsupervised Multi-Object Discovery.
IEEE International Conference on Computer Vision (ICCV) Workshops.

Hahn, O., Reich, C., Araslanov, N., Cremers, D., Rupprecht, C., & Roth, S. (2025).
Scene-Centric Unsupervised Panoptic Segmentation.
Proceedings of the Computer Vision and Pattern Recognition Conference, 24485-24495.

Hesse, R., Fischer, J., Schaub-Meyer, S., & Roth, S. (2025).
Disentangling Polysemantic Channels in Convolutional Neural Networks.
Proceedings of the Computer Vision and Pattern Recognition Conference, 4799-4803.

Hesse, R., Schaub-Meyer, S., & Roth, S. (2021).
Fast axiomatic attribution for neural networks.
Advances in Neural Information Processing Systems, 34(2), 19513-19524.

Hesse, R., Schaub-Meyer, S., & Roth, S. (2023).
Funnybirds: A synthetic vision dataset for a part-based analysis of explainable ai methods.
Proceedings of the IEEE/CVF International Conference on Computer Vision, 3981-3991.

Hesse, R., Schaub-Meyer, S., & Roth, S. (2024).
Benchmarking the attribution quality of vision models.
Advances in Neural Information Processing Systems, 37, 97928-97947.

Hesse, R., Schaub-Meyer, S., & Roth, S. (2023) (2023).
Content-adaptive downsampling in convolutional neural networks.
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 4544-4553.

Jevtić, A., Reich, C., Wimbauer, F., Hahn, O., Rupprecht, C., Roth, S., & Cremers, D. (2025).
Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion.
Proc. of the Twentieth IEEE International Conference on Computer Vision.

Jevtić, A., Reich, C., Wimbauer, F., Hahn, O., Rupprecht, C., Roth, S., & Cremers, D. (2025).
Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion.
IEEE International Conference on Computer Vision (ICCV)