Shattering the Rings: Reproducibility and Vulnerability Analysis of the ZoDiac Watermarking Framework

TMLR Paper6050 Authors

Abstract:

This paper presents a reproducibility study and robustness evaluation of the paper ‘Attack-
Resilient Image Watermarking Using Stable Diffusion’ by Zhang et al. (2024), which pro-
poses ZoDiac, a Stable Diffusion-based framework for attack-resilient image watermarking.
While verifying the original method’s core claims achieving >90% watermark detection rate
(WDR) against diffusion-based regeneration attacks and across MS-COCO, DiffusionDB,
and WikiArt datasets, we identify critical vulnerabilities under adversarial and geometrically
asymmetric attack paradigms. Our vulnerability analysis demonstrates that gradient-based
adversarial perturbations reduce ZoDiac’s WDR, a threat model absent in prior evaluation.
We also investigate rotationally asymmetric attacks achieving WDR below 65%. We discover
composite attacks combining adversarial noise with other methods reducing WDR to near-
zero, exposing vulnerabilities through multi-stage offensive pipelines. Our implementation
can be found on Anonymous Github.

Submission Length: Regular submission (no more than 12 pages of main content)

Previous TMLR Submission Url: https://openreview.net/forum?id=xoQV6kdTqG

Changes Since Last Submission:

Summary of Revisions
- **Enhanced Contextualization and Literature Review**: The manuscript has been revised to provide a more thorough historical context and to more clearly position our work within the existing body of related literature.

- **Strengthened Methodological Justification**: We have expanded the manuscript to better substantiate the paper's primary claims and to provide a more robust justification for the specific assessment methods utilized in our analysis.

- **Improved Clarity of Methods and Results**: Significant improvements have been made to enhance the clarity regarding the selection of our research methods. Furthermore, the presentation and interpretation of the results have been refined for greater lucidity and impact.

Assigned Action Editor: ~Sheng_Li3

Submission Number: 6050