Research map / 3D vision / Gaussian splatting
SplatPainter: Interactive Authoring of 3D Gaussians from 2D Edits via Test-Time Training
Yang Zheng, Hao Tan, Kai Zhang, Peng Wang, Leonidas Guibas, Gordon Wetzstein, Wang Yifan
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 stage II TTT refinement on top of stage I (ablation) | SplatPainter · Local refinement (TexVerse, 370 assets) | PSNR: 20.84 -> 21.07 | Improved |
| add stage II TTT refinement on top of stage I (ablation) | SplatPainter · Local refinement (TexVerse, 370 assets) | LPIPS: 0.3782 -> 0.3673 | Improved |
| add TTT refinement instead of cross-attention (ablation) | SplatPainter · Global relighting | PSNR: 24.05 -> 24.56 | Improved |
| provide more relit views (1 to 4) | SplatPainter · Global relighting | PSNR: 24.56 -> 25.03 | Improved |
| provide 2 relit views instead of 1 | SplatPainter · Global relighting | LPIPS: 0.0721 -> 0.0719 | No clear effect |
| use LocalAtt + feats query instead of VoxelAtt (ablation) | Voxel compression · Voxel compression (stage I) | PSNR: 30.65 -> 34.02 | Improved |
| use feats query instead of latent query (ablation) | Voxel compression · Voxel compression (stage I) | PSNR: 33.84 -> 34.02 | Improved |
| voxel compression versus uncompressed GS-LRM | SplatPainter · Voxel compression (stage I) | PSNR: 34.19 -> 34.02 | Hurt |
| use SplatPainter local TTT update instead of VoxHammer | Creative editing · Creative editing, non-edited region | PSNR: 13.86 -> 33.02 | Improved |
| use SplatPainter instead of 3DGS optimization | Local refinement · Local refinement (TexVerse, 370 assets) | time (s): 323 -> 1.54+0.64 | Improved |
| use SplatPainter instead of SRGS | Local refinement · Local refinement (TexVerse, 370 assets) | SSIM: 0.6759 -> 0.5780 | Hurt |
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.
- GS-LRM — GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (ECCV 2024) · better on 3 of 3 shared results
- RelitLRM — RelitLRM: Generative Relightable Radiance for Large Reconstruction Models (ICLR 2025) · 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
- SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis (CVPR 2025 · similarity 0.90)
- GaussianPainter: Painting Point Cloud into 3D Gaussians with Normal Guidance (AAAI 2025 · similarity 0.90)
- 3DitScene: Editing Any Scene via Language-guided Disentangled Gaussian Splatting (ICLR 2025 · similarity 0.90)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- GaussianEditor: Swift and Controllable 3D Editing with Gaussian Splatting (CVPR 2024 · similarity 0.90)
- Textured Gaussians for Enhanced 3D Scene Appearance Modeling (CVPR 2025 · similarity 0.89)
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.
