DFG Funds Research on Reliable AI for Open-World Object Detection

The German Research Foundation (DFG) has approved funding for a new research project led by Matthias Rottmann at Osnabrück University. The project, titled “Unified Uncertainty Estimation for Fine-tuned Open-Vocabulary Models in Image Classification and Object Detection,” aims to make modern AI systems more reliable and transparent when operating in open-world environments.

Building on recent advances in foundation models such as CLIP and Grounding DINO, the research will develop new methods to quantify and interpret uncertainty in AI-based image understanding. This will help AI systems recognize when they are unsure — a key capability for safety-critical applications such as autonomous driving or medical imaging.

Over the next three years, the project will design and evaluate novel techniques that combine deep learning with generative models like stable diffusion to create realistic “out-of-context” examples (see Figure). These synthetic samples will be used to train AI systems that can better handle unknown objects and situations. A particular focus lies on transferring these capabilities to smaller, efficient object detectors that can operate on limited hardware, extending the applicability of reliable AI to real-world settings.

Starting on January 1, 2026, the project will open new avenues toward safer, more transparent, and more responsible applications of machine learning in the years to come.