NotaryMark multi-bit watermarking for T2I dataset auditing
Original title: NotaryMark: Multi-Bit Watermarking for Black-Box DatasetUsage Auditing in Text-to-Image Diffusion Models
NotaryMark aligns watermarks with diffusion latent denoising for robust black-box auditing across datasets and models.
NotaryMark is a multi-bit watermarking method for auditing unauthorized text-to-image diffusion model training dataset use. It overcomes signal decay issues in existing techniques by aligning with the model's noising-denoising process. Evaluations on three datasets and models show bit accuracies of 0.85-0.90 below 32 bits and 0.67-0.80 above, outperforming prior methods. A voting aggregation scheme with error-correcting codes enables reliable decoding, with robustness tested against 15 evasion attacks.
- Watermark capacities tested from 8 to 1024 bits
- Bit accuracy 0.85-0.90 for capacities below 32 bits
- Voting-based aggregation boosts accuracy with 30 queries
- Error-correcting codes enable reliable decoding
- Robust to most of 15 evasion attacks
From the original article
1 min read at arxiv.orgAs generative artificial intelligence (AI) models increasingly require data, more publicly available yet licensed data is being harvested and used without authorization. Although several previous studies have proposed watermarking methods for dataset copyright auditing, their applicability to the text-to-image (T2I) setting remains notably limited.
By Xinyi Yu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.