Language-Embedded 3DGS
3DGS: 3D Gaussian Splatting
Overview
Language-embedded 3D Gaussian Splatting studies how to represent 3D scenes with Gaussian primitives that carry both visual geometry and language-aligned semantic features. Common approaches lift image-language features such as CLIP embeddings into 3D, then use multi-view alignment, instance grouping, or scene graphs to support open-vocabulary segmentation, object retrieval, and visual question answering. Recent work also extends these representations to dynamic scenes, sparse or unposed inputs, and robotic navigation and manipulation. A central tension is how to preserve reliable, fine-grained semantic evidence across views while keeping feature storage and reconstruction efficient.
Latest papers
27 papers6 papers in the last four weeks, up 50% on the four weeks before. 0.0% of all new papers.