Open bupt01 opened 1 week ago
Hello , I'm having some confusion while calculating the ROC curve for watermark detection and would appreciate your help. Let's say we have 500 watermarked texts to calculate the detection accuracy. Should I calculate the false positive rate by selecting 500 original texts from the dataset, or should I use 500 texts generated by a large model without any watermark?
Hello , I'm having some confusion while calculating the ROC curve for watermark detection and would appreciate your help. Let's say we have 500 watermarked texts to calculate the detection accuracy. Should I calculate the false positive rate by selecting 500 original texts from the dataset, or should I use 500 texts generated by a large model without any watermark?