Master Thesis – Enhanced Reasoning in Vision-Language Models Using Structured Visual Representations

This role involves two master's students investigating how structured visual information can be integrated into Vision-Language Models to support more detailed and interpretable visual reasoning. The work includes exploring existing surveillance metadata, fine-tuning open-source VLMs, and evaluating improvements in spatial, object-level, and temporal reasoning capabilities. The thesis aims to enhance model understanding of complex video scenes through structured visual representations.