Research map / 3D vision / Gaussian splatting
ReInGS: Re-Initializing 3D Gaussians against Sparsity Discrepancy in Few-Shot Novel View Synthesis
Junao Shen, Tian Feng, Haojie Dong, Jinkang Ji, Tianjia Shao
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 EHPS (coarse-level expansion-based hybrid point sampling) (ablation) | DNGaussian baseline · LLFF 3-view (ablation) | PSNR: +0.49 dB | Improved |
| add IVPS (intra-view point sampling) after EHPS (ablation) | ReInGS · LLFF 3-view (ablation) | PSNR: +0.41 dB | Improved |
| add CVPS (cross-view point sampling) (ablation) | ReInGS · LLFF 3-view (ablation) | PSNR: +1.09 dB | Improved |
| add EHPS on top of VDPS alone (ablation) | ReInGS · LLFF 3-view (ablation) | PSNR: 19.66 -> 20.88 | Improved |
| add all strategies to the DNGaussian baseline (ablation) | ReInGS · LLFF 3-view (ablation) | PSNR: 18.87 -> 20.88 | Improved |
| add EHPS alone (ablation) | ReInGS · LLFF 3-view (ablation) | LPIPS: 0.313 -> 0.348 | Hurt |
| increase random initial points from 2000 to 8000 | DNGaussian · LLFF 3-view pilot | PSNR: 18.65 -> 18.94 | Improved |
| increase random initial points from 2000 to 8000 | FSGS · LLFF 3-view pilot | PSNR: 18.15 -> 18.33 | Improved |
| use dense fused stereo points from extended views instead of random points | CoR-GS · LLFF 3-view | PSNR: 16.32 -> 20.45 | Improved |
| use dense fused stereo points instead of random points | FSGS · LLFF 3-view | PSNR: 18.28 -> 20.31 | Improved |
| use ReInGS re-initialization instead of DNGaussian (random init) | DNGaussian · LLFF 2-view | PSNR: 16.47 -> 19.39 | Improved |
| use ReInGS re-initialization instead of second best (random init) | LLFF 4-view · LLFF 4-view | PSNR: 20.86 -> 22.24 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- DNGaussian — DNGaussian: Optimizing Sparse-View 3D Gaussian Radiance Fields with Global-Local Depth Normalization (CVPR 2024) · better on 9 of 9 shared results
- FewViewGS — FewViewGS: Gaussian Splatting with Few View Matching and Multi-stage Training (NeurIPS 2024) · better on 4 of 4 shared results
- CoR-GS — CoR-GS: Sparse-View 3D Gaussian Splatting via Co-Regularization (ECCV 2024) · better on 3 of 3 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
- FewViewGS: Gaussian Splatting with Few View Matching and Multi-stage Training (NeurIPS 2024 · similarity 0.91)
- Quantifying and Alleviating Co-Adaptation in Sparse-View 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
- 3DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with View-consistent 2D Diffusion Priors (NeurIPS 2024 · similarity 0.91)
- NexusGS: Sparse View Synthesis with Epipolar Depth Priors in 3D Gaussian Splatting (CVPR 2025 · similarity 0.91)
- Metropolis-Hastings Sampling for 3D Gaussian Reconstruction (NeurIPS 2025 · similarity 0.90)
- LP-3DGS: Learning to Prune 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.90)
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.
