Research map / 3D vision / Gaussian splatting
Improving Sparse-View 3DGS Generalization via Flat Minima Optimization
Kangmin Seo, Sangeek Hyun, MinKyu Lee, Jae-Pil Heo
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 |
|---|---|---|---|
| use anisotropic scale-adaptive noise instead of isotropic noise (ablation) | flat-minima 3DGS · LLFF 3-view | PSNR: 20.54 -> 20.88 | Improved |
| perturb Gaussian position rather than scale, rotation or opacity (ablation) | flat-minima 3DGS · LLFF 3-view | PSNR: 20.21-20.23 -> 20.88 | Improved |
| also perturb scale in addition to position (ablation) | flat-minima 3DGS · LLFF 3-view | PSNR: 20.88 -> 20.33 | Hurt |
| apply perturbation stochastically per Gaussian instead of mixed loss (ablation) | flat-minima 3DGS · LLFF 3-view | LPIPS: 0.212 -> 0.184 | Improved |
| add linear perturbation magnitude scheduling (ablation) | flat-minima 3DGS · LLFF 3-view | PSNR: 20.69 -> 20.88 | Improved |
| add periodic reinitialization of scale, rotation, SH (ablation) | flat-minima 3DGS · LLFF 3-view | PSNR: 20.58 -> 20.88 | Improved |
| add flat-minima plug-in to FSGS | FSGS · LLFF 3-view | PSNR: 20.43 -> 21.03 | Improved |
| add flat-minima plug-in to DropGaussian | DropGaussian · LLFF 3-view | PSNR: 20.76 -> 21.05 | Improved |
| add flat-minima plug-in to Difix3D+ | Difix3D+ · Mip-NeRF360 12-view | PSNR/LPIPS: PSNR 19.25 -> 19.63; LPIPS 0.375 -> 0.386 | Mixed |
| replace 3DGS post-optimization with ours after AnySplat | AnySplat · Mip-NeRF360 12-view | PSNR: 16.97 -> 17.36 | Improved |
| apply method with more input views (36-view) | 3DGS · Mip-NeRF360 36-view | PSNR: 25.15 -> 25.01 | Hurt |
| apply method vs DropGaussian on Mip-NeRF360 12-view | 3DGS · Mip-NeRF360 12-view | PSNR: 19.74 -> 19.50 PSNR; LPIPS 0.364 -> 0.348 | Mixed |
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 8 shared results
- CoR-GS — CoR-GS: Sparse-View 3D Gaussian Splatting via Co-Regularization (ECCV 2024) · better on 1 of 1 shared result
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
- DropGaussian: Structural Regularization for Sparse-view Gaussian Splatting (CVPR 2025 · similarity 0.93)
- Gaussian Herding across Pens: An Optimal Transport Perspective on Global Gaussian Reduction for 3DGS (NeurIPS 2025 · similarity 0.91)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.91)
- Generative Sparse-View Gaussian Splatting (CVPR 2025 · similarity 0.91)
- LP-3DGS: Learning to Prune 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.91)
- 3DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with View-consistent 2D Diffusion Priors (NeurIPS 2024 · similarity 0.91)
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.
