VLSplat uses vision-language for object-centric 3D Gaussian Splatting
Original title: VLSplat: Vision-Language Guided Object-Centric 3D Gaussian Splatting via Scene Graph
VLSplat augments 2D-to-3D semantic lifting with vision-language model priors and scene graph refinement to improve object-level semantic consistency and structural completeness.
Existing 2D-to-3D semantic lifting approaches rely on 2D segmentation and lack object-level priors. VLSplat augments mask-based lifting with object-level semantic priors from vision-language models. Refinement uses intra-group suppression of semantic noise and inter-group expansion via scene graph. Experiments on indoor scenes demonstrate improved semantic consistency and structural completeness of object categories.
- Augments mask-based lifting with vision-language priors
- Uses object-group scene graph for refinement
- Intra-group refinement suppresses semantic noise
- Inter-group acquisition expands support regions
- Improves object-level categories on indoor datasets
From the original article
1 min read at arxiv.orgLearning semantic representations for 3D Gaussian Splatting has recently emerged as a promising direction for 3D scene understanding. However, existing 2D-to-3D semantic lifting approaches rely primarily on 2D segmentation supervision and lack object-level priors, causing the learned representation to be driven by local mask evidence rather than holistic object structure.
By Yuntae Jeon · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.