Research map / Image generation / Text-to-image diffusion
SWAP: Towards Copyright Auditing of Soft Prompts via Sequential Watermarking
Wenyuan Yang, Yichen Sun, Changzheng Chen, Zhixuan Chu, Jiaheng Zhang, Yiming Li, Dacheng Tao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use sequential watermark SWAP instead of backdoor-style BWAP-BadNet | CoCoOp · ImageNet | WSR: 98.60 -> 99.98 | Improved |
| use SWAP instead of BWAP-BadNet | PromptSRC · OxfordPets | WSR: 98.00 -> 99.99 | Improved |
| use SWAP instead of BWAP-BadNet (relative harmlessness H) | CoCoOp · ImageNet | relative harmless degree: 0.69 -> 0.00 | Improved |
| use SWAP instead of BWAP-WaNet | MaPLe · Caltech101 | relative harmless degree: 0.98 -> 0.01 | Improved |
| use SWAP instead of BWAP-BadNet | MaPLe · ImageNet | novel-class accuracy: 66.80 -> 69.26 | Improved |
| adapt CLIP backdoor BadEncoder to prompt learning | MaPLe · ImageNet | attack success rate: 0.18 -> 0.00 | Hurt |
| adapt BadCLIP-D to prompt learning | MaPLe · Caltech101 | attack success rate: 7.42 | No clear effect |
| use non-intrusive auditing on independent soft prompts with similar data | MaPLe · Caltech101 | UAP similarity CDF: 100% | Hurt |
| verify SWAP watermark on independent prompt | CoCoOp · ImageNet | WSR: 4.27 vs 99.98 | No clear effect |
| apply SWAP to cross-dataset transfer | PromptSRC · cross-dataset average | WSR: 99.95 | Improved |
Closest papers by meaning
- Safeguarding Multimodal Knowledge Copyright in the RAG-as-a-Service Environment (ICLR 2026 · similarity 0.80)
- AudioMarkBench: Benchmarking Robustness of Audio Watermarking (NeurIPS 2024 · similarity 0.80)
- XAttnMark: Learning Robust Audio Watermarking with Cross-Attention (ICML 2025 · similarity 0.79)
- StableGuard: Towards Unified Copyright Protection and Tamper Localization in Latent Diffusion Models (NeurIPS 2025 · similarity 0.78)
- A Resilient and Accessible Distribution-Preserving Watermark for Large Language Models (ICML 2024 · similarity 0.78)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.78)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
