Darmstadt, Germany
Researcher in computer vision and biometrics at TU Darmstadt and Fraunhofer IGD.
I work on face recognition systems and how far they can be trusted: estimating how usable a face image is, separating a genuine face from a spoof or a morph, and getting vision transformers to explain the decisions they make.
Most of it sits where accuracy meets cost and accountability. Quality estimation that needs no training or that exits the network early, foundation models adapted to attack detection, and vision language alignment for explainability. The work has appeared at WACV, CVPR, FG, IJCB, ECCV and ACM MM, and the code is released with each paper.
Before Darmstadt I took an M.Sc. in Machine Learning at the University of Tübingen, and B.Sc. degrees in Electrical and Electronics Engineering and in Mathematics at METU in Ankara.
Available. Open to research collaborations, and to selected consulting work in computer vision and biometrics. What I can help with →
IEEE International Conference on Automatic Face and Gesture Recognition, 2026
Estimates face image quality from the intermediate layers of a vision transformer, so a full forward pass is not required.
IEEE International Conference on Automatic Face and Gesture Recognition, 2026
Uses attention maps as a quality signal, which also indicates which regions of a face led to a low score.
IEEE/CVF Winter Conference on Applications of Computer Vision Workshops, 2026
Estimates face image quality from a pretrained vision transformer with no training and no quality labels.
IEEE/CVF Winter Conference on Applications of Computer Vision Workshops, 2025
Adapts foundation models to presentation attack detection and reports how they behave with limited training data and across domains.
ACM International Conference on Multimedia Workshops, 2026
Trains presentation attack detection on images of vegetables rather than faces, and looks at what the models were relying on.
European Conference on Computer Vision Workshops, 2026
Aligns face embeddings with language so that a recognition decision can be described in words.
Early exit representations from vision transformers for face image quality assessment.
Interpretable face image quality assessment based on attention.
Face image quality assessment from a frozen ViT, with no training.
Foundation models adapted for face presentation attack detection.
Face morphing attack detection with foundation models.
Explaining face recognition decisions through vision language alignment.
I am open to research collaborations, and I take on a small amount of consulting where a team needs face recognition, quality assessment or attack detection that holds up outside a benchmark. The fastest way to reach me is email.
