With the widespread adoption of 3D Gaussian Splatting (3DGS) in digital content creation, copyright protection for 3DGS assets has become increasingly important. Existing 3DGS watermarking methods mostly rely on rendered images for watermark decoding, making them ineffective against partial infringement, where local subsets of Gaussian primitives are extracted and integrated into new scenes. To address this problem, we propose GaussMark, a native watermarking method based on Key-Projected Quantization Index Modulation (KP-QIM) for primitive-level copyright protection of 3DGS. GaussMark uses the high-order spherical harmonic coefficients of each Gaussian primitive as the watermark carrier. It generates an orthogonal projection matrix from a secret key and performs QIM encoding in the resulting key-controlled subspace, enabling each individual Gaussian primitive to independently carry and decode a copyright codeword. To reduce the impact of quantization perturbations on rendering quality, we further design a multi-round progressive hard-projection strategy during training, which interleaves watermark embedding with 3DGS optimization. Experimental results show that GaussMark enables effective partial-infringement detection while maintaining low rendering-quality degradation and high embedding and extraction efficiency. In addition, GaussMark can be combined with existing image-domain watermarking methods and is compatible with common Gaussian variants such as 2DGS and 4DGS.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: