SOTA Watermark Detection Robustness benchmarks and papers with code

Benchmarks

Dataset Name	SOTA Method	Metric
C4	PMARK (Online)	TP@FP=1%96.97	12	3mo ago
Gemma-2 9B Pre-trained (PT) (test)	KGW	TPR (Baseline)100	7	25d ago
Gemma-2 2B Pre-trained (PT) (test)	KGW	TPR (None)100	7	25d ago
Llama-3-8B GPT-4o Paraphrase, 150 Tokens	KGW	Mean P0.26	6	3mo ago
Llama-3-8B GPT-4o Paraphrase, 30 Tokens	KGW	Mean P29	6	3mo ago
Llama-3-8B Delete 50%, 150 Tokens	PRC	Mean P0.36	6	3mo ago
Llama-3-8B Delete 50%, 30 Tokens	PRC	Mean P0.38	6	3mo ago
Llama-3-8B Delete 30%, 150 Tokens	PRC	Mean P0.31	6	3mo ago
Llama-3-8B Delete 30%, 30 Tokens	PRC	Mean P0.26	6	3mo ago
Llama-3-8B Swap 50%, 150 Tokens	KGW	Mean P0.12	6	3mo ago
Llama-3-8B Swap 50%, 30 Tokens	KGW	Mean P25	6	3mo ago
Llama-3-8B Swap 30%, 150 Tokens	KGW	Mean P0.03	6	3mo ago
Llama-3-8B Swap 30%, 30 Tokens	KGW	Mean P0.16	6	3mo ago
C4 subset (n=512)	MC2MARK	Detection Rate (RTR 10%)84.81	4	3mo ago
C4 subset (n=256)	MC2MARK	RTR (10%) Score92.35	4	3mo ago
C4 subset (n=128)	MC2MARK	RTR (10%) Score97.85	4	3mo ago
C4 (n=64)	MC2MARK	RTR Accuracy (10%)99.09	4	3mo ago
C4 subset (n=32)	MC2MARK	Robustness - Random Token Replacement (10%) - Score100	4	3mo ago
C4 subset (n=16)	MC2MARK	RTR (10%) Score100	4	3mo ago

Showing 19 of 19 rows