Research map / 3D vision / Gaussian splatting
Confidence-Based Mesh Extraction from 3D Gaussians
Lukas Radl, Felix Windisch, Andreas Kurz, Thomas Köhler, Michael Steiner, Markus Steinberger
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add improved decoupled appearance embedding to SOF (ablation) | SOF · Tanks & Temples | F1: 0.474 -> 0.493 | Improved |
| add confidence-weighted photometric loss L_conf (ablation) | SOF + improved appearance · ScanNet++ | F1: 0.625 -> 0.655 | Improved |
| add per-primitive color variance loss (ablation) | SOF + appearance + L_conf · Tanks & Temples | F1: 0.509 -> 0.519 | Improved |
| add per-primitive normal variance loss (ablation) | SOF + appearance + L_conf + color-var · ScanNet++ | F1: 0.658 -> 0.668 | Improved |
| add normal variance loss (ablation) | SOF + appearance + L_conf + color-var · Tanks & Temples | F1: 0.519 -> 0.521 | Improved |
| use appearance-corrected image only in the luminance term of D-SSIM (decoupling) (ablation) | improved appearance module · Tanks & Temples | F1: 0.490 -> 0.493 | Improved |
| drop the luminance term from D-SSIM instead of compensating it (ablation) | improved appearance module · Tanks & Temples | F1: 0.493 -> 0.483 | Hurt |
| architectural improvements to VastGaussian appearance (exp activation, padding, positional encoding) (ablation) | VastGaussian appearance · Tanks & Temples | F1: 0.475 -> 0.484 | Improved |
| add confidence-steered densification (ablation) | Ours · Tanks & Temples | F1 and Gaussian count: F1 0.516 -> 0.521; 1.53M -> 1.27M Gaussians | Improved |
| enable L_conf later in training (iteration 15K instead of 500) (ablation) | Ours · Tanks & Temples | F1: 0.521 -> 0.501 | Hurt |
| enable L_conf at iteration 5K instead of 500 (ablation) | Ours · Tanks & Temples | F1: 0.521 -> 0.517 | Hurt |
| increase normal variance weight tenfold (0.005 to 0.05) (ablation) | Ours · Tanks & Temples | F1: 0.521 -> 0.515 | Hurt |
Closest papers by meaning
- LP-3DGS: Learning to Prune 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.89)
- Effective Rank Analysis and Regularization for Enhanced 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.89)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.89)
- Sparse2DGS: Geometry-Prioritized Gaussian Splatting for Surface Reconstruction from Sparse Views (CVPR 2025 · similarity 0.89)
- SfM-Free 3D Gaussian Splatting via Hierarchical Training (CVPR 2025 · similarity 0.88)
- SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
