Research map / 3D vision / NeRF and radiance fields
The NeRF Signature: Codebook-Aided Watermarking for Neural Radiance Fields
Ziyuan Luo, Anderson De Rezende Rocha, Boxin Shi, Qing Guo, Haoliang Li, Renjie 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 distortion layer during optimization (ablation) | NeRF Signature · Blender (16-bit, combined image transformations) | bit accuracy: 55.32% -> 97.11% | Improved |
| add distortion layer during optimization (ablation) | NeRF Signature · Blender (16-bit, rotation) | bit accuracy: 49.41% -> 99.90% | Improved |
| add pose key and patch key (ablation) | NeRF Signature · LLFF room (48-bit) | bit accuracy / PSNR | Improved |
| add complexity-aware key selection (CAKS) instead of random patches (ablation) | NeRF Signature · LLFF room (48-bit) | imperceptibility (PSNR) | Improved |
| embed signature in the finest-resolution grid instead of lowest, medium or all resolutions (ablation) | NeRF Signature · Blender lego (32-bit) | PSNR | Improved |
| use signature codebook instead of external-module embedding (CopyRNeRF) at 48 bits | NeRF Signature · Blender (48-bit) | bit accuracy: 62.15% -> 99.94% | Improved |
| use codebook embedding instead of per-signature fine-tuning at 48 bits | NeRF Signature · Blender (48-bit) | bit accuracy: 85.26% -> 99.94% | Improved |
| use codebook embedding instead of CopyRNeRF | NeRF Signature · Training time (48-bit) | training time: >70 hours -> ~6 minutes | Improved |
| attacker knows pose key and has clean images (500 epochs fine-tuning) | NeRF Signature · Fine-tuning attack (16-bit) | bit accuracy: 99.98% -> 52.13% | Hurt |
| attacker lacks pose key but has clean images (500 epochs fine-tuning) | NeRF Signature · Fine-tuning attack (16-bit) | bit accuracy: 99.98% -> 97.50% | No clear effect |
| PGD adversarial attack with actual key (delta 0.1) | NeRF Signature · Adversarial attack (16-bit) | bit accuracy: 99.98% -> 61.75% | Hurt |
| PGD adversarial attack with guessed keys, delta 1.0 (seed 2) | NeRF Signature · Adversarial attack (16-bit) | bit accuracy: 99.98% -> 92.28% | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CopyRNeRF — CopyRNeRF: Protecting the CopyRight of Neural Radiance Fields (ICCV 2023) · 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
- WateRF: Robust Watermarks in Radiance Fields for Protection of Copyrights (CVPR 2024 · similarity 0.94)
- DreaMark: Rooting Watermark in Score Distillation Sampling Generated Neural Radiance Fields (AAAI 2025 · similarity 0.91)
- CodecNeRF: Toward Fast Encoding and Decoding, Compact, and High-quality Novel-view Synthesis (AAAI 2025 · similarity 0.85)
- WMCopier: Forging Invisible Watermarks on Arbitrary Images (NeurIPS 2025 · similarity 0.85)
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · similarity 0.85)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.84)
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.
