Reproduce “VCR-GauS: View Consistent Depth-Normal Regularizer for Gaussian Surface 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
Hanlin Chen, Fangyin Wei, Chen Li, Tianxin Huang, Yunsong Wang, Gim Hee Lee
VCR-GauS, developed primarily by researchers from the National University of Singapore, introduces a view-consistent depth-normal regularizer for 3D Gaussian Splatting, enabling high-quality 3D surface reconstruction from multi-view images while maintaining efficient rendering. The method achieves state-of-the-art F1-scores on large-scale datasets like Tanks and Temples, matching the geometric accuracy of implicit methods at significantly faster optimization times.
Key insights
- Direct supervision of rendered Gaussian normals is insufficient to effectively update Gaussian positions for surface reconstruction, requiring a depth-normal coupling for full geometric optimization.
- Robustly handling noisy and inconsistent monocular normal priors through a confidence-aware weighting mechanism is crucial for preventing artifacts and achieving smooth, accurate surface geometry.
- Optimizing Gaussian distribution and size through targeted densification of surface Gaussians and principal-axis splitting significantly enhances surface smoothness and precision beyond mere view synthesis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.