Deep 3D mesh watermarking with self-adaptive robustness

Cybersecurity

Table 2 Quantitative results on remeshed Hand and remeshed Asiadragon dataset

	Method	HD	MRMS	\(l_{cur}\)	Accuracy (%)
Asiadragon	Model	0.059	0.014	0.002	93.25
	Model\(\dag\)	0.042	0.010	0.002	84.63
	Pre-Model	0.069	0.016	0.006	83.63
	Pre-Model\(\dag\)	0.058	0.012	0.004	76.51
Hand	Model	0.112	0.018	0.005	91.78
	Model\(\dag\)	0.101	0.012	0.003	85.09
	Pre-Model	0.113	0.019	0.004	83.00
	Pre-Model\(\dag\)	0.099	0.012	0.003	77.34

Distance between original meshes and watermarked meshes (second to fourth column), bit accuracy under attacks from the attack layers (\(\%\), last column). \(\dag\) denotes training without the attack layers, Model represents the model trained on the remeshed dataset, and Pre-Model represents the model trained on the original dataset
The bold represents the ones with best preformance