The Smart Embedded Systems Lab (SES Lab) at OTH Regensburg works on AI systems that hold up outside the lab. Our current focus is the detection of AI-generated content in text and images: detectors that generalise across datasets, generators and domains, that survive adversarial manipulation of their input, and whose decisions can be explained to the people who rely on them. We also study how well people themselves recognise generated faces, and how large language models perform at generating test scenarios for automated driving.

All code, model weights and demos released here are free to use for research. Each repository states its own licence.

A detector of AI-generated text built for deployment conditions: unknown domains, unknown generators, adversarially perturbed input and no labels for threshold calibration. The central finding is that Unicode normalisation acts in opposite directions depending on where it is applied. Normalising the training corpus deletes the adversarial supervision, while normalising at inference is an effective defence.

Code, the released checkpoint and a live demo are public; see the repository to reproduce every number in the paper.

AI-Generated Content Detection: A Cross-Modal Survey of Methods, Challenges, and Future Directions. Mohamed Mady, Yupei Li, Björn W. Schuller, Berrak Sisman, Johannes Reschke. Preprint, Research Square, 2026. Preprint · DOI

DeBERTa-ConPara: Attack-Aware and Deployment-Realistic Detection of AI-Generated Text. Mohamed Mady, Yupei Li, Johannes Reschke, Björn W. Schuller. AACL-IJCNLP 2026, main conference. Paper · Code · Model · Demo

Feature-Augmented Transformers for Robust AI-Text Detection Across Domains and Generators. Mohamed Mady, Johannes Reschke, Björn W. Schuller. arXiv preprint, 2026. Paper

David vs. Goliath: A comparative study of different-sized LLMs for code generation in the domain of automotive scenario generation. Philipp Bauerfeind, Amir Salarpour, David Fernandez, Pedram MohajerAnsari, Johannes Reschke, Mert D. Pesé. arXiv preprint, 2025. Paper

Tackling fake images in cybersecurity: Interpretation of a StyleGAN and lifting its black-box. Julia Laubmann, Johannes Reschke. arXiv preprint, 2025. Paper

Deepfake Detection of Face Images based on a Convolutional Neural Network. Lukas Kroiß, Johannes Reschke. arXiv preprint, 2025. Paper

Can you tell a real face from an AI-generated one? Our gamified study AIliens shows five faces to classify, then a short training on what gives generated faces away, then five more. It measures how well people spot AI-generated faces and whether a few minutes of training helps.

Play the game · works on phone and desktop, and can be installed as an app · Source

• Prof. Dr.-Ing. Johannes Reschke, professor of self-learning and adaptive systems
• Mohamed Mady, doctoral researcher (cooperative doctorate with the Technical University of Munich)
• Gerald Schickhuber, lab engineer

For questions about our code or models, please open an issue in the respective repository. For research collaboration, contact details are on the lab website.