Research map / Adversarial robustness and security / Backdoor and poisoning
CopyrightShield: Enhancing Diffusion Model Security Against Copyright Infringement Attacks
Zhixiang Guo, Siyuan Liang, Aishan Liu, 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 |
|---|---|---|---|
| add adaptive optimization training to detection (ablation) | CopyrightShield (detection only) · Pokemon (poison rate 0.1) | CIR: 0.517 -> 0.305 | Improved |
| add poisoned sample detection to adaptive optimization training (ablation) | CopyrightShield (training only) · Pokemon (poison rate 0.1) | CIR: 0.594 -> 0.305 | Improved |
| add adaptive optimization training to detection (ablation) | CopyrightShield (detection only) · COYO+Midjourney (poison rate 0.1) | FAE: 49.36 -> 86.42 | Improved |
| add poisoned sample detection to adaptive optimization training (ablation) | CopyrightShield (training only) · COYO+Midjourney (poison rate 0.1) | FAE: 35.44 -> 86.42 | Improved |
| use dynamic penalty instead of fixed lambda=0.1 (ablation) | CopyrightShield · Pokemon (poison rate 0.1) | CIR: 0.318 -> 0.305 | Improved |
| use dynamic penalty instead of fixed lambda=0.05 (ablation) | CopyrightShield · COYO+Midjourney (poison rate 0.1) | CIR: 0.237 -> 0.217 | Improved |
| use detection threshold 0.35 instead of 0.4 (ablation) | CopyrightShield detection · Pokemon (rate 0.1) | F1: 0.615 -> 0.671 | Improved |
| use detection threshold 0.35 instead of 0.3 (ablation) | CopyrightShield detection · Pokemon (rate 0.1) | F1: 0.601 -> 0.671 | Improved |
| use spatial-masked attribution instead of D-TRAK | CopyrightShield detection · Pokemon | F1: 0.534 -> 0.671 | Improved |
| use CopyrightShield instead of TERD | CopyrightShield · Pokemon (poison rate 0.1) | CIR: 0.452 -> 0.305 | Improved |
| use CopyrightShield instead of T2IShield | CopyrightShield · COYO+Midjourney (poison rate 0.1) | FAE: 37.85 -> 86.42 | Improved |
| defend with CopyrightShield versus attacked model | CopyrightShield · Pokemon (generation quality) | FID: 54.8 -> 54.1 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TERD — TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline (ICML 2024 · similarity 0.93)
- Anti-Diffusion: Preventing Abuse of Modifications of Diffusion-Based Models (AAAI 2025 · similarity 0.90)
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.89)
- Disguised Copyright Infringement of Latent Diffusion Models (ICML 2024 · similarity 0.89)
- Harnessing Frequency Spectrum Insights for Image Copyright Protection Against Diffusion Models (CVPR 2025 · similarity 0.89)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.88)
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.
