Reproduce “OmniIndoor3D: Comprehensive Indoor 3D Reconstruction” 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
Xiaobao Wei, Xiaoan Zhang, Hao Wang, Qingpo Wuwu, Ming Lu, Wenzhao Zheng, Shanghang Zhang
OmniIndoor3D introduces a unified 3D Gaussian Splatting framework that jointly optimizes appearance, geometry, and panoptic understanding for comprehensive indoor 3D reconstruction from RGB-D data. This framework achieves state-of-the-art performance across all three tasks on the ScanNet and ScanNet++ datasets, providing precise geometry and semantic-instance awareness for robotic navigation.
Key insights
- Comprehensive 3D scene understanding for robotics necessitates a unified approach that integrates appearance, geometry, and panoptic segmentation, as these aspects can mutually reinforce each other.
- The inherent conflict between optimizing for appearance and precise geometry in 3D Gaussian Splatting can be mitigated by decoupling their parameter adjustments through a lightweight Multi-Layer Perceptron.
- Leveraging panoptic priors, specifically semantic confidence, can effectively guide the spatial distribution and densification of Gaussian primitives, leading to improved geometric completeness and semantic consistency, particularly on planar surfaces.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.