Research map / 3D vision / Gaussian splatting
X-SG2S: Safe and Generalizable Gaussian Splatting with X-dimensional Watermarks
Zihang Cheng, Wentao Bao, HUIPING ZHUANG, Chun Li, Xin Meng, Ziqian Zeng, Cen Chen, Ming Li, Fei Yu
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 |
|---|---|---|---|
| use interaction block between Gaussian points and watermark patches instead of normal MLP gate (ablation) | X-SG2S · ACID ablation (multimodal) | extraction accuracy: 0.958 -> 1.000 | Improved |
| use interaction block instead of random locations (ablation) | X-SG2S · ACID ablation (multimodal) | extraction accuracy: 0.667 -> 1.000 | Improved |
| use interaction block instead of normal MLP gate (ablation) | X-SG2S · ACID ablation (multimodal) | PSNR (org scene): 24.276 -> 27.804 | Improved |
| use interaction block instead of random locations (ablation) | X-SG2S · ACID ablation (multimodal) | PSNR (org scene): 21.534 -> 27.804 | Improved |
| add optional rendering MSE and LPIPS losses (ablation) | X-SG2S · ACID ablation (multimodal) | extraction accuracy: 0.979 -> 1.000 | Improved |
| add optional rendering MSE and LPIPS losses (ablation) | X-SG2S · ACID ablation (multimodal) | PSNR (image): 21.923 -> 22.791 | Improved |
| add optional rendering MSE and LPIPS losses (ablation) | X-SG2S · ACID ablation (multimodal) | PSNR (org scene): 27.757 -> 27.804 | Improved |
| inject multi-modal (XD) watermark instead of 1D only (ablation) | X-SG2S · ACID ablation (multimodal) | PSNR (org scene): 27.814 -> 27.804 | No clear effect |
| inject watermark only into higher-order SH coefficients instead of all SH (ablation) | X-SG2S · ACID | visual fidelity | Improved |
| increase cropout rate from 10% to 25% | X-SG2S · ACID | extraction accuracy (%): 100 -> 97.7 | No clear effect |
| add Gaussian noise to SH (sigma=0.05) | X-SG2S · ACID | extraction accuracy (%): 100 -> 96.22 | Hurt |
| X-SG2S versus GaussianMarker 3D dec under noise | X-SG2S · Blender+LLFF, noise to SH | bit accuracy (%): 99.90 -> 94.12 | 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.
- 3D-GSW — 3D-GSW: 3D Gaussian Splatting for Robust Watermarking (CVPR 2025) · better on 5 of 5 shared results
- GuardSplat — GuardSplat: Efficient and Robust Watermarking for 3D Gaussian Splatting (CVPR 2025) · 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
- GuardSplat: Efficient and Robust Watermarking for 3D Gaussian Splatting (CVPR 2025 · similarity 0.93)
- GaussianMarker: Uncertainty-Aware Copyright Protection of 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.93)
- 3D-GSW: 3D Gaussian Splatting for Robust Watermarking (CVPR 2025 · similarity 0.91)
- GS-Hider: Hiding Messages into 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.89)
- SecureGS: Boosting the Security and Fidelity of 3D Gaussian Splatting Steganography (ICLR 2025 · similarity 0.89)
- Geometry Cloak: Preventing TGS-based 3D Reconstruction from Copyrighted Images (NeurIPS 2024 · similarity 0.88)
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.
