Reproduce “GaussianSeal: Rooting Adaptive Watermarks for 3D Gaussian Generation Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Runyi Li, Xuanyu Zhang, Chuhan Tong, Zhipei Xu, Jian Zhang
GaussianSeal introduces the first bit watermarking framework for 3D Gaussian Splatting (3DGS) generative models, embedding copyright information directly within the generation process. The method achieves a 97.93% bit decoding accuracy for 16-bit messages while preserving generated object quality and significantly improving efficiency by processing 3DGS objects in 0.18 seconds.
Key insights
- Adaptive bit modulation with learnable coefficients enables in-generation watermarking of 3DGS models without noticeable degradation of output quality.
- Embedding watermark information at multiple stages within the generative network significantly improves decoding accuracy and robustness.
- The DWT low-pass subband of rendered 2D views serves as a highly effective and robust target for watermark decoding from 3DGS objects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.