Reproduce “AlignGS: Aligning Geometry and Semantics for Robust Indoor Reconstruction from Sparse Views” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yijie Gao, Houqiang Zhong, Tianchi Zhu, Zhengxue Cheng, Qiang Hu, Li Song
AlignGS, a new framework from Shanghai Jiao Tong University, achieves robust and semantically rich 3D indoor scene reconstruction from sparse views by synergistically optimizing geometry and semantics. It significantly improves novel view synthesis quality and geometric fidelity, for instance, doubling the F-score on ScanNet scenes compared to previous methods.
Key insights
- Semantic understanding can serve as an active, guiding force in geometric reconstruction, moving beyond passive semantic annotation to improve geometric fidelity.
- Robust initialization with an SfM-free method like VGGT is crucial for stable sparse-view 3D reconstruction, circumventing limitations of traditional pipelines.
- Leveraging both soft and hard semantic distillation effectively transfers comprehensive semantic knowledge from powerful 2D foundation models to 3D Gaussian primitives.
- Semantic boundaries provide a strong prior for regularizing surface normals, promoting sharper geometric features and more accurate surface reconstructions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.