notesum.ai

Published at November 29

GuardSplat: Robust and Efficient Watermarking for 3D Gaussian Splatting

cs.CV

cs.CR

Released Date: November 29, 2024

Authors: Zixuan Chen¹, Guangcong Wang², Jiahao Zhu¹, Jianhuang Lai³, Xiaohua Xie³

Aff.: ¹School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China; ²School of Computing and Information Technology, Great Bay University, Dongguan, China; ³School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China; Guangdong Province Key Laboratory of Information Security Technology, Guangzhou, China; Key Laboratory of Machine Intelligence and Advanced Computing, Ministry of Education, China

Arxiv: http://arxiv.org/pdf/2411.19895v1

Methods	16 bits				32 bits				48 bits
Methods	Bit Acc	PSNR	SSIM	LPIPS	Bit Acc	PSNR	SSIM	LPIPS	Bit Acc	PSNR	SSIM	LPIPS
CopyRNeRF [24]	91.16	26.29	0.9100	0.0380	78.08	26.13	0.8960	0.0410	60.06	27.56	0.8950	0.0660
WateRF [13]	95.67	32.79	0.9480	0.0330	88.58	31.19	0.9360	0.0400	85.82	30.86	0.9300	0.0400
WateRF [13] + 3DGS	92.89	31.01	0.9678	0.0475	90.15	29.56	0.9611	0.0492	87.30	29.13	0.9562	0.0534
HiDDeN [60] + 3DGS [17]	63.07	31.59	0.9890	0.0171	52.46	34.51	0.9882	0.0209	53.08	33.42	0.9787	0.0299
GuardSplat (Ours) + HiDDeN [60]	98.75	40.48	0.9909	0.0025	95.58	38.32	0.9897	0.0025	93.29	38.56	0.9886	0.0032
GuardSplat (Ours)	99.64	41.55	0.9957	0.0017	99.04	39.40	0.9939	0.0022	98.29	38.90	0.9923	0.0028