Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2511.19316 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914610572951552 |
|---|---|
| author | Wang, Xincheng Sun, Hanchi Sun, Wenjun Xue, Kejun Zhou, Wangqiu Zhang, Jianbo Sun, Wei Zhu, Dandan Min, Xiongkuo Jia, Jun Fang, Zhijun |
| author_facet | Wang, Xincheng Sun, Hanchi Sun, Wenjun Xue, Kejun Zhou, Wangqiu Zhang, Jianbo Sun, Wei Zhu, Dandan Min, Xiongkuo Jia, Jun Fang, Zhijun |
| contents | Recent fine-tuning techniques for diffusion models enable them to reproduce specific image sets, such as particular faces or artistic styles, but also introduce copyright and security risks. Dataset watermarking has been proposed to ensure traceability by embedding imperceptible watermarks into training images, which remain detectable in outputs even after fine-tuning. However, current methods lack a unified evaluation framework. To address this, this paper establishes a general threat model and introduces a comprehensive evaluation framework encompassing Universality, Transmissibility, and Robustness. Experiments show that existing methods perform well in universality and transmissibility, and exhibit some robustness against common image processing operations, yet still fall short under real-world threat scenarios. To reveal these vulnerabilities, the paper further proposes a practical watermark removal method that fully eliminates dataset watermarks without affecting fine-tuning, highlighting a key challenge for future research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_19316 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Evaluating Dataset Watermarking for Fine-tuning Traceability of Customized Diffusion Models: A Comprehensive Benchmark and Removal Approach Wang, Xincheng Sun, Hanchi Sun, Wenjun Xue, Kejun Zhou, Wangqiu Zhang, Jianbo Sun, Wei Zhu, Dandan Min, Xiongkuo Jia, Jun Fang, Zhijun Computer Vision and Pattern Recognition Artificial Intelligence Recent fine-tuning techniques for diffusion models enable them to reproduce specific image sets, such as particular faces or artistic styles, but also introduce copyright and security risks. Dataset watermarking has been proposed to ensure traceability by embedding imperceptible watermarks into training images, which remain detectable in outputs even after fine-tuning. However, current methods lack a unified evaluation framework. To address this, this paper establishes a general threat model and introduces a comprehensive evaluation framework encompassing Universality, Transmissibility, and Robustness. Experiments show that existing methods perform well in universality and transmissibility, and exhibit some robustness against common image processing operations, yet still fall short under real-world threat scenarios. To reveal these vulnerabilities, the paper further proposes a practical watermark removal method that fully eliminates dataset watermarks without affecting fine-tuning, highlighting a key challenge for future research. |
| title | Evaluating Dataset Watermarking for Fine-tuning Traceability of Customized Diffusion Models: A Comprehensive Benchmark and Removal Approach |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2511.19316 |