The paper "RingID: Rethinking Tree-Ring Watermarking for Enhanced Multi-Key Identification" by Hai Ci, Pei Yang, Yiren Song, and Mike Zheng Shou from the National University of Singapore revisits the Tree-Ring Watermarking method, a diffusion model watermarking technique known for its robustness against various attacks. The authors conduct an in-depth study and reveal that the distribution shift introduced during the watermarking process, apart from pattern matching, contributes significantly to its robustness. However, this distribution shift does not aid in identifying multiple distinct keys, exposing inherent flaws in the original design. To address these limitations, the authors propose RingID, a novel multi-channel heterogeneous watermarking approach. RingID integrates the advantages of different types of watermarks to enhance robustness against various attacks. It includes a series of enhancements such as discretization and lossless imprinting methods, which significantly improve the watermark's distinguishing ability. Specifically, RingID demonstrates substantial advancements in multi-key identification, achieving an AUC of 0.995 and an identification accuracy of 0.82, compared to 0.975 and 0.07 for Tree-Ring, respectively. The paper also discusses the theoretical and empirical analysis of the distribution shift, its impact on verification and identification tasks, and the design flaws in Tree-Ring. It provides a comprehensive evaluation of RingID's performance under various image distortions and highlights its superior robustness and larger valid key capacity. The authors conclude by discussing the limitations of their approach, particularly its vulnerability to cropping and scaling attacks, and suggest future directions for further research.The paper "RingID: Rethinking Tree-Ring Watermarking for Enhanced Multi-Key Identification" by Hai Ci, Pei Yang, Yiren Song, and Mike Zheng Shou from the National University of Singapore revisits the Tree-Ring Watermarking method, a diffusion model watermarking technique known for its robustness against various attacks. The authors conduct an in-depth study and reveal that the distribution shift introduced during the watermarking process, apart from pattern matching, contributes significantly to its robustness. However, this distribution shift does not aid in identifying multiple distinct keys, exposing inherent flaws in the original design. To address these limitations, the authors propose RingID, a novel multi-channel heterogeneous watermarking approach. RingID integrates the advantages of different types of watermarks to enhance robustness against various attacks. It includes a series of enhancements such as discretization and lossless imprinting methods, which significantly improve the watermark's distinguishing ability. Specifically, RingID demonstrates substantial advancements in multi-key identification, achieving an AUC of 0.995 and an identification accuracy of 0.82, compared to 0.975 and 0.07 for Tree-Ring, respectively. The paper also discusses the theoretical and empirical analysis of the distribution shift, its impact on verification and identification tasks, and the design flaws in Tree-Ring. It provides a comprehensive evaluation of RingID's performance under various image distortions and highlights its superior robustness and larger valid key capacity. The authors conclude by discussing the limitations of their approach, particularly its vulnerability to cropping and scaling attacks, and suggest future directions for further research.