CopyRNeRF: Protecting the CopyRight of Neural Radiance Fields

About

Neural Radiance Fields (NeRF) have the potential to be a major representation of media. Since training a NeRF has never been an easy task, the protection of its model copyright should be a priority. In this paper, by analyzing the pros and cons of possible copyright protection solutions, we propose to protect the copyright of NeRF models by replacing the original color representation in NeRF with a watermarked color representation. Then, a distortion-resistant rendering scheme is designed to guarantee robust message extraction in 2D renderings of NeRF. Our proposed method can directly protect the copyright of NeRF models while maintaining high rendering quality and bit accuracy when compared among optional solutions.

Ziyuan Luo, Qing Guo, Ka Chun Cheung, Simon See, Renjie Wan · 2023

Related benchmarks

Task	Dataset	Result
Digital Watermarking	Blender and LLFF (test)	Bit Accuracy (No Attack)91.16	39
3D Scene Watermarking	Blender and LLFF 16 bits	Bit Accuracy91.16	14
3D Scene Watermarking	Blender and LLFF 32 bits	Bit Accuracy78.08	14
3D Scene Watermarking	Blender and LLFF 48 bits	Bit Acc60.06	14
3D Watermarking Robustness against Diffusion Attacks	Blender and LLFF (test)	Bit Accuracy (Deterministic)0.512	6
Image Quality Assessment	Blender and LLFF views (test)	SSIM0.747	6
3D Watermarking	LLFF and Blender (train)	Training Time (min)85	6
3D Scene Watermarking and Reconstruction	Blender	PSNR30.29	5
3D Scene Watermarking and Reconstruction	LLFF	PSNR24.03	5
3D Scene Watermarking and Reconstruction	MipNeRF360	PSNR22.47	5