Research map / 3D vision / Gaussian splatting
Tile-wise vs. Image-wise: Random-Tile Loss and Training Paradigm for Gaussian Splatting
Xiaoyu Zhang, Weihong Pan, Xiaojun Xiang, Hongjia Zhai, Liyang Zhou, Hanqing Jiang, Guofeng Zhang
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 image-wise loss with random-tile loss (RT-Loss) in 3DGS | 3D-GS · Mip-NeRF360 | PSNR: 28.69 -> 29.66 | Improved |
| replace image-wise loss with RT-Loss | 3D-GS · Deep Blending | PSNR: Scaffold-GS 30.21 vs Ours 30.15; 3D-GS 29.41 -> 30.15 | Mixed |
| apply random-tile training to Scaffold-GS | Scaffold-GS · Mip-NeRF360 | PSNR: 28.84 -> 29.86 | Improved |
| use RT-Loss on scenes with more training views (SMERF) | 3D-GS · SMERF Alameda at 4k iters | PSNR: 17.62 -> 20.11 | Improved |
| apply RT-Loss to dynamic scene reconstruction | 4D-GS · Neural 3D Video | PSNR: 31.15 -> 31.74 | Improved |
| add RT-SSIM structure loss on random tiles (ablation) | RT-Loss · KITCHEN | PSNR: 31.65 -> 32.35 | Improved |
| use RT-SSIM instead of image-wise SSIM during random tile training (ablation) | RT-Loss · COURTHOUSE | PSNR: 23.50 -> 24.15 | Improved |
| use Iteration-Count rather than Tile-Count densification (ablation) | RT-Loss · KITCHEN | PSNR: 31.95 -> 32.35 | Improved |
| use image-wise batch training (B=4) instead of random tiles (ablation) | 3D-GS · KITCHEN | PSNR: 31.67 -> 31.22 | Hurt |
| increase number of views V for tile sampling (ablation) | RT-Loss · COURTHOUSE | PSNR: 23.76 -> 24.29 (V=2 to V=6) | Improved |
| uniform instead of non-uniform tile sampling (ablation) | RT-Loss · KITCHEN | PSNR: 32.30 -> 32.38 | No clear effect |
| use RT-Loss with random point initialization instead of SfM (ablation) | 3D-GS · KITCHEN | PSNR: 29.89 -> 31.22 | Improved |
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.
- Scaffold-GS — Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering (CVPR 2024) · better on 3 of 3 shared results
- Mip-NeRF360 — Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fields (CVPR 2022) · 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
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.92)
- DashGaussian: Optimizing 3D Gaussian Splatting in 200 Seconds (CVPR 2025 · similarity 0.92)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.91)
- GaussianPro: 3D Gaussian Splatting with Progressive Propagation (ICML 2024 · similarity 0.91)
- Optimized Minimal 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
- IBGS: Image-Based Gaussian Splatting (NeurIPS 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.
