Masterarbeit: Evaluation von Vision-Language Models für soziale Situationen

Master thesis at Fraunhofer IGCV in Augsburg on evaluating vision-language models for capturing socially situational contexts in industrial environments – benchmark development & model comparison. Activities include literature review on vision-language models and human–human interaction/human–object interaction, development of an annotation strategy and taxonomy for social scenes (interactions, salience, task context, social norms) with an annotation tool (e.g., CVAT), annotation of an industrial image dataset (4 sectors) with multiple annotators and inter-annotator agreement, systematic comparison of local VLMs (e.g., LLaVA, InternVL2, Qwen2-VL) on NVIDIA Jetson Orin AGX including prompt-engineering and analysis of prompt sensitivity, and documentation of results.

What they are looking for

Details

Compensation

Paid