#smartmage

SmartMage targets adaptive modality use in 3D scene understanding

The proposed MLLM routes visual and geometric inputs based on query relevance.