AI RESEARCH
RAG-3DSG: Enhancing 3D Scene Graphs with Re-Shot Guided Retrieval-Augmented Generation
arXiv CS.AI
•
ArXi:2601.10168v2 Announce Type: replace-cross Open-vocabulary 3D Scene Graph (3DSG) can enhance various downstream tasks in robotics by leveraging structured semantic representations, yet current 3DSG construction methods suffer from semantic inconsistencies caused by noisy cross-image aggregation under occlusions and constrained viewpoints. To mitigate the impact of such inconsistency, we propose RAG-3DSG, which