
A lightweight and scalable regional point-language contrastive learning framework, RegionPLC, is proposed to enable robust and effective 3D learning from dense regional language supervision for open-world 3D scene understanding.


coremsg

regional-point-language-contrastive-learning-for-open-world-3d-scene-understanding


Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding


title_rewrite


A scalable and incremental approach for constructing open-set 3D scene representations by leveraging 2D foundation models, enabling open vocabulary instance retrieval, segmentation, annotation, and spatial reasoning.


opensu3d-scalable-and-incremental-3d-scene-understanding-using-foundation-models


OpenSU3D: Scalable and Incremental 3D Scene Understanding using Foundation Models
