SHield watermarks 4D Gaussian Splatting files
Original title: SHield: Watermarking 4D Gaussian Splatting via Spherical Harmonic Coefficient Perturbation
SHield embeds payloads in 4DGS via SH coefficient perturbation for decoder-free ownership verification.
4D Gaussian Splatting models are distributed as parameter files vulnerable to copying. SHield tiles zero-mean payloads across spherical harmonic coefficients of selected Gaussians. Geometric re-association using canonical centers survives tampering. Experiments on 30 scenes confirm near-perfect recovery with <=0.13 dB rendering penalty.
- Payload recovery with Pearson r >= 0.991 across 150 models
- Survives INT8 quantization and 30% opacity pruning
- Geometric re-association for tamper resistance
- Held-out evaluation shows zero false positives
- Applies brief anchored optimization for fidelity
From the original article
1 min read at arxiv.org4D Gaussian Splatting (4DGS) models are computationally expensive to train but distributed as easily copied parameter files. Existing watermarks protect these assets by fine-tuning the scene against a neural decoder to extract bit strings from rendered video, a heavy process vulnerable to the temporal dynamics of dynamic scenes.
By Kunal Bhosikar · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.