Research map / 3D vision / Gaussian splatting
SGI: Structured 2D Gaussians for Efficient and Compact Large Image Representation
Zixuan Pan, Kaiyuan Tang, Jun Xia, Yifan Qin, Lin Gu, Chaoli Wang, Jianxu Chen, Yiyu Shi
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 multi-scale (coarse-to-fine) fitting with 3 pyramid levels (ablation) | SGI · FGF2 | optimization time (min): 71.59 -> 48.43 | Improved |
| add multi-scale (coarse-to-fine) fitting with 3 pyramid levels (ablation) | SGI · FGF2 | PSNR: 30.58 -> 31.24 | Improved |
| increase pyramid levels from 3 to 4 (ablation) | SGI · ICB | PSNR: 35.27 -> 33.78 | Hurt |
| add entropy coding with context model (lambda 0.001 vs 0) (ablation) | SGI · FGF2 | model size (MB): 104.08 -> 16.33 | Improved |
| add entropy coding with context model (lambda 0.001 vs 0) (ablation) | SGI · FGF2 | PSNR: 32.36 -> 31.24 | Hurt |
| increase lambda from 0.001 to 0.003 (ablation) | SGI · ICB | PSNR: 35.27 -> 33.51 | Hurt |
| increase Gaussians per seed K from 10 to 20 (ablation) | SGI · FGF2 | model size (MB): 16.33 -> 14.83 | Improved |
| increase Gaussians per seed K from 10 to 20 (ablation) | SGI · FGF2 | PSNR: 31.24 -> 30.62 | Hurt |
| decrease Gaussians per seed K from 10 to 5 (ablation) | SGI · FGF2 | model size (MB): 16.33 -> 18.48 | Hurt |
| use seed-based 2D neural Gaussians alone without entropy coding (ablation) | unstructured 2D Gaussians (LIG) · FGF2 / ICB | model size: about 3% reduction | No clear effect |
| use SGI instead of GaussianImage at 3.5M Gaussians | 2D Gaussian image representation · FGF2 | PSNR: 27.30 -> 31.24 | Improved |
| use SGI low-rate instead of I-NGP on STimage | image representation · STimage | PSNR: 35.57 -> 33.96 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GaussianImage — GaussianImage: 1000 FPS Image Representation and Compression by 2D Gaussian Splatting (ECCV 2024) · better on 4 of 4 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
- GaussianSR: High Fidelity 2D Gaussian Splatting for Arbitrary-Scale Image Super-Resolution (AAAI 2025 · similarity 0.91)
- Large Images Are Gaussians: High-Quality Large Image Representation with Levels of 2D Gaussian Splatting (AAAI 2025 · similarity 0.90)
- Structure Consistent Gaussian Splatting with Matching Prior for Few-shot Novel View Synthesis (NeurIPS 2024 · similarity 0.89)
- FewViewGS: Gaussian Splatting with Few View Matching and Multi-stage Training (NeurIPS 2024 · similarity 0.88)
- LOD-GS: Achieving Levels of Detail using Scalable Gaussian Soup (CVPR 2025 · similarity 0.88)
- Compressed 3D Gaussian Splatting for Accelerated Novel View Synthesis (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.
