Research map / Adversarial robustness and security / Black-box and transfer attacks
Box-Free Model Watermarks are Prone to Black-Box Removal Attacks
Haonan An, Guang Hua, Zhiping Lin, Yuguang Michael Fang
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 |
|---|---|---|---|
| train and test with Blank watermark concatenation instead of no concatenation (ablation) | EGG remover · V2 (PASCAL VOC) | PSNR (dB): 33.24 -> 34.81 | Improved |
| train and test with Blank concatenation instead of no concatenation (ablation) | Transferable remover (CGAN+CEILNet) · V2 (PASCAL VOC) | PSNR (dB): 30.58 -> 30.62 | Improved |
| test with a_i as concatenated image instead of Blank (ablation) | EGG remover · V2 (PASCAL VOC) | PSNR (dB): 34.81 -> 9.27 | Hurt |
| use CEILNet instead of EEENet as proxy extractor ENet' (ablation) | Transferable remover (CGAN HNet') · V1 (PASCAL VOC) | SR_R: 0.20 -> 1.00 | Improved |
| use CEILNet instead of EEENet as proxy extractor ENet' (ablation) | Transferable remover (UNet HNet') · V2 (PASCAL VOC) | SR_R: 0.24 -> 1.00 | Improved |
| use CEILNet instead of EEENet as proxy extractor ENet' (ablation) | Transferable remover (UNet HNet') · V1 (PASCAL VOC) | SR_R: 0.04 -> 1.00 | Improved |
| attack V2 instead of V1 with CGAN+EEENet transferable remover | Transferable remover · PASCAL VOC deraining | SR_R: 0.20 -> 1.00 | Improved |
| JPEG compression factor 50% vs 90% on V2 extraction (previous removal attack) | V2 victim · PASCAL VOC deraining | SR_E: 0.00 -> 0.00 | No clear effect |
| increase JPEG quality factor from 50% to 80% on V1 | V1 victim · PASCAL VOC deraining | SR_E: 0.10 -> 1.00 | Improved |
| use transferable remover with UNet+EEENet on V3 | Transferable remover · V3 generalizability | AUC: 0.99 -> 0.60 | Improved |
| use transferable remover instead of EGG for overwriting | Overwriting attack · V1/V2 overwrite | watermark overwrite: 0.961 and 0.984 | Hurt |
Closest papers by meaning
- Decoder Gradient Shield: Provable and High-Fidelity Prevention of Gradient-Based Box-Free Watermark Removal (CVPR 2025 · similarity 0.91)
- Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks (ICLR 2024 · similarity 0.88)
- Safe and Robust Watermark Injection with a Single OoD Image (ICLR 2024 · similarity 0.87)
- Invisible Image Watermarks Are Provably Removable Using Generative AI (NeurIPS 2024 · similarity 0.87)
- Transferable Black-Box One-Shot Forging of Watermarks via Image Preference Models (NeurIPS 2025 · similarity 0.87)
- ROBIN: Robust and Invisible Watermarks for Diffusion Models with Adversarial Optimization (NeurIPS 2024 · similarity 0.86)
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.
