Research map / 3D vision / Gaussian splatting
Write Where It Matters: Policy-Guided Watermarks for 3D Gaussian Splatting
Nan Li, Yike Zeng, Qian Zhang, Qi Zhang, Zhiyi Pan, Wei Feng, Liang Wan
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 robustness loss L_rob to the invisibility-only reward (ablation) | W2M · LLFF 48-bit | bit accuracy: 54.17 -> 99.17 | Improved |
| add Gaussian-scaling regularizer R_scale (ablation) | W2M reward · LLFF 48-bit | PSNR: 36.02 -> 38.10 (bit acc 99.17 -> 98.75) | Improved |
| use per-Gaussian policy instead of per-anchor policy (ablation) | W2M · Blender 48-bit | bit accuracy / cost: bit acc 96.76 -> 98.97; time 9 -> 42 mins; memory 9.52 -> 16.04 GB | Mixed |
| increase number of anchors (ablation) | W2M · LLFF 48-bit | bit accuracy / PSNR | Improved |
| use W2M instead of GaussianMarker under rotation | GaussianMarker · 32-bit image-space distortion | bit accuracy: 79.76 -> 91.86 | Improved |
| use W2M instead of 3D-GSW under Gaussian noise on rendering | 3D-GSW · 32-bit image-space distortion | bit accuracy: 83.84 -> 97.20 | Improved |
| use W2M instead of baselines under JPEG compression | 3D-GSW · 32-bit image-space distortion | bit accuracy: 92.65 -> 92.77 | No clear effect |
| use W2M instead of baselines at 64 bits | GaussianMarker · 64-bit | bit accuracy: 90.84 -> 91.17 | No clear effect |
| use W2M instead of baselines under model-space Gaussian noise | GaussianMarker · 32-bit model-space distortion | bit accuracy: 87.12 -> 91.57 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- 3D-GSW — 3D-GSW: 3D Gaussian Splatting for Robust Watermarking (CVPR 2025) · better on 9 of 9 shared results
- GuardSplat — GuardSplat: Efficient and Robust Watermarking for 3D Gaussian Splatting (CVPR 2025) · better on 5 of 5 shared results
- GaussianMarker — GaussianMarker: Uncertainty-Aware Copyright Protection of 3D Gaussian Splatting (NeurIPS 2024) · better on 5 of 5 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
- Where, What, Why: Toward Explainable 3D-GS Watermarking (CVPR 2026 · similarity 0.92)
- GaussianMarker: Uncertainty-Aware Copyright Protection of 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.91)
- Can Protective Watermarking Safeguard the Copyright of 3D Gaussian Splatting? (AAAI 2026 · similarity 0.91)
- GuardSplat: Efficient and Robust Watermarking for 3D Gaussian Splatting (CVPR 2025 · similarity 0.89)
- 3D-GSW: 3D Gaussian Splatting for Robust Watermarking (CVPR 2025 · similarity 0.88)
- SecureGS: Boosting the Security and Fidelity of 3D Gaussian Splatting Steganography (ICLR 2025 · similarity 0.87)
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.
