Master Thesis in 3D Scene Representation and Understanding with 3D Scene Graphs

During your master thesis, you will work on learning-based, open-vocabulary 3D scene understanding using 3D scene graphs, specifically on the incremental construction and fusion of 3D scene graphs. You will contribute to 3D scene graph prediction in one or more of the following areas: architecture design, training data and loss design, model efficiency, or model application in a real-world environment. Furthermore, you will evaluate your model on open-source datasets, benchmark it against state-of-the-art approaches, and demonstrate its capabilities on real-world data on a suitable robotics platform. Finally, you will have the opportunity to publish your results at peer-reviewed conferences.