Research map / 3D vision / Gaussian splatting
D$^2$GS: Depth-and-Density Guided Gaussian Splatting for Stable and Accurate Sparse-View Reconstruction
Meixi Song, Xin Lin, Dizhe Zhang, Haodong Li, Xiangtai Li, Bo Du, Lu Qi
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 |
|---|---|---|---|
| replace uniform dropout with depth-and-density guided dropout plus DAFE | 3DGS sparse-view (DropGaussian base) · LLFF (3-view, 1/8 resolution) | PSNR (dB): 20.76 -> 21.35 vs DropGaussian | Improved |
| apply D2GS on Mip-NeRF360 sparse views | D2GS · Mip-NeRF360 (3-view) | PSNR (dB): 19.74 -> 20.09 | Improved |
| add density score and depth-based layering dropout to baseline (ablation) | 3DGS · LLFF (3-view, 1/8 resolution) | PSNR (dB): 19.22 -> 21.02 | Improved |
| add depth-based layering to density+depth score (ablation) | DD-Drop · LLFF (3-view, 1/8 resolution) | PSNR (dB): 21.10 -> 21.17 | Improved |
| add DAFE far-field loss (ablation) | DD-Drop · LLFF (3-view, 1/8 resolution) | PSNR (dB): 21.17 -> 21.35 | Improved |
| add DAFE (ablation) | DD-Drop · LLFF (3-view), IMR | IMR: 3.088 -> 3.039 | Improved |
| set equal depth and density weights (0.5/0.5) instead of 0.8 depth (ablation) | DD-Drop · LLFF (3-view, 1/8 resolution) | PSNR (dB): 21.04 -> 21.16 | Improved |
| raise maximum dropout rate from 0.3 to 0.5 (ablation) | DD-Drop · LLFF (3-view, 1/8 resolution) | PSNR (dB): 21.16 -> 21.06 | Hurt |
| vary DAFE far-region ratio tau from 5% to 15% (ablation) | DAFE · LLFF (3-view, 1/8 resolution) | PSNR (dB): 21.25 / 21.26 / 21.20 | No clear effect |
| swap DepthAnything V2 for weaker depth estimators (ablation) | DAFE · LLFF (3-view, 1/8 resolution) | PSNR (dB): 21.35 -> 21.21 (MiDaS), 21.27 (DPT) | Hurt |
| use D2GS instead of DropGaussian for stability | D2GS · LLFF (3-view) | IMR: 3.205 -> 3.039 | Improved |
Later papers that compared with this method
1 later accepted paper reports D2GS as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- StereoGS: Sparse-View 3D Gaussian Splatting via Stereo Priors (ECCV 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DropGaussian — DropGaussian: Structural Regularization for Sparse-view Gaussian Splatting (CVPR 2025) · better on 7 of 7 shared results
- CoR-GS — CoR-GS: Sparse-View 3D Gaussian Splatting via Co-Regularization (ECCV 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Generative Sparse-View Gaussian Splatting (CVPR 2025 · similarity 0.94)
- DropGaussian: Structural Regularization for Sparse-view Gaussian Splatting (CVPR 2025 · similarity 0.93)
- Quantifying and Alleviating Co-Adaptation in Sparse-View 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.92)
- Sparse2DGS: Geometry-Prioritized Gaussian Splatting for Surface Reconstruction from Sparse Views (CVPR 2025 · similarity 0.92)
- DropoutGS: Dropping Out Gaussians for Better Sparse-view Rendering (CVPR 2025 · similarity 0.92)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.92)
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.
