Research map / 3D vision / Gaussian splatting
3DGS$^2$-TR: A Scalable Second-Order Trust-Region Method for 3D Gaussian Splatting
Roger Hsiao, Yuchen Fang, Xiangru Huang, Ruilong Li, Hesam Rabeti, Zan Gojcic, Javad Lavaei, James Demmel, Sophia 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 Hellinger-distance trust region to ADAM update (ablation) | ADAM · MipNeRF360 @30k | PSNR: 25.08 -> 25.21 | Improved |
| add Hellinger-distance trust region to ADAM update (ablation) | ADAM · Deep Blending @30k | PSNR: 27.36 -> 27.58 | Improved |
| add Hellinger-distance trust region to ADAM update (ablation) | ADAM · Tanks & Temples @30k | PSNR: 21.66 -> 22.42 | Improved |
| add Sophia-style diagonal Hessian second-order update on top of trust region | ADAM-TR · MipNeRF360 @30k | PSNR: 25.21 -> 25.39 | Improved |
| add Sophia-style diagonal Hessian second-order update on top of trust region | ADAM-TR · Tanks & Temples @30k | PSNR: 22.42 -> 22.85 | Improved |
| add Sophia-style diagonal Hessian second-order update on top of trust region | ADAM-TR · Deep Blending @30k | SSIM: 0.869 -> 0.869 | No clear effect |
| use 3DGS2-TR instead of ADAM | ADAM · Tanks & Temples | PSNR: exceed the best ADAM PSNR by 0.56dB at 7k iterations and 1.19dB at 30k iterations | Improved |
| use 3DGS2-TR instead of ADAM | ADAM · MipNeRF360 | training time (s): 212 -> 652 | Hurt |
| use 3DGS2-TR instead of ADAM | ADAM · MipNeRF360 | peak GPU memory (GB): 5.60 -> 6.51 | Hurt |
| use 3DGS2-TR instead of 3DGS-LM | 3DGS-LM · MipNeRF360 @30k | PSNR: 21.88 -> 25.39 | Improved |
Closest papers by meaning
- GaussianPro: 3D Gaussian Splatting with Progressive Propagation (ICML 2024 · similarity 0.92)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.92)
- DOGS: Distributed-Oriented Gaussian Splatting for Large-Scale 3D Reconstruction Via Gaussian Consensus (NeurIPS 2024 · similarity 0.92)
- AdpSplit: Error-Driven Adaptive Splitting for Faster Geometry Discovery in 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.92)
- 3D Student Splatting and Scooping (CVPR 2025 · similarity 0.92)
- SfM-Free 3D Gaussian Splatting via Hierarchical Training (CVPR 2025 · 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.
