Research map / 3D vision / Gaussian splatting
Flash-Refine: Frustum-Guided Local Incremental Learning for Efficient 3D Gaussian Splatting Completion
Nuocheng Ji, Hanyang Zhuang, Chunxiang Wang, Ming Yang
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 point-wise consensus masking with gradient lock (ablation) | Flash-Refine · Counter (Kept images) | PSNR (dB): 20.33 -> 27.47 | Improved |
| add adaptive depth-bounded prior to frustum masking (ablation) | Flash-Refine · Counter (Kept images) | PSNR (dB): 24.04 -> 27.47 | Improved |
| add anti-deletion shield (ablation) | Flash-Refine · Counter (Kept images) | PSNR (dB): 20.87 -> 27.47 | Improved |
| add anti-deletion shield (ablation) | Flash-Refine · Counter (Dropped images) | PSNR (dB): 26.22 -> 29.66 | Improved |
| add mask (gradient lock) vs naive fine-tuning | Flash-Refine · Bedroom (Kept images) | PSNR (dB): 24.30 -> 28.71 | Improved |
| Flash-Refine vs naive fine-tuning on dropped views | Flash-Refine · Bedroom (Dropped images) | PSNR (dB): 26.08 -> 27.17 | Improved |
| Flash-Refine vs geometric stitching | Flash-Refine · DrJohnson (Kept images) | PSNR (dB): 25.82 -> 27.57 | Improved |
| Flash-Refine vs full retrain on Counter dropped views | Flash-Refine · Counter (Dropped images) | PSNR (dB): 31.03 -> 29.66 | Hurt |
| Flash-Refine repair vs base model on kept views | Flash-Refine · Counter (Kept images) | PSNR (dB): 31.91 -> 27.47 | Hurt |
| Flash-Refine vs base model on dropped views | Flash-Refine · Bicycle (Dropped images) | PSNR (dB): 15.19 -> 16.91 | Improved |
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.
- SplaTAM — SplaTAM: Splat Track & Map 3D Gaussians for Dense RGB-D SLAM (CVPR 2024) · better on 2 of 2 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
- RI3D: Few-Shot Gaussian Splatting With Repair and Inpainting Diffusion Priors (ICCV 2025 · similarity 0.91)
- EDGS: Eliminating Densification for Efficient Convergence of 3DGS (CVPR 2026 · similarity 0.91)
- Learning Efficient Fuse-and-Refine for Feed-Forward 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- ReorgGS: Equivalent Distribution Reorganization for 3D Gaussian Splatting (NeurIPS 2026 · similarity 0.90)
- PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · 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.
