Research map / Adversarial robustness and security / Backdoor and poisoning
Dynamic Attention Analysis for Backdoor Detection in Text-to-Image Diffusion Models
Zhongqi Wang, Jie Zhang, Shiguang Shan, Xilin Chen
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 spatial correlation among attention maps (DAA-S over DAA-I) (ablation) | DAA-I · average over 6 attacks | F1: 73.43 -> 79.27 | Improved |
| add spatial correlation among attention maps (DAA-S over DAA-I) (ablation) | DAA-I · average over 6 attacks | AUC: 84.14 -> 86.27 | Improved |
| use EOS token instead of BOS token in the evolve-rate feature (ablation) | DAA-S · avg AUC over 6 attacks | AUC: 64.79 -> 86.27 | Improved |
| use EOS token instead of averaging all tokens (ablation) | DAA-S · avg AUC over 6 attacks | AUC: 70.15 -> 86.27 | Improved |
| use F-norm instead of SSIM as attention similarity (ablation) | DAA-S · avg F1 over 6 attacks | F1: 62.22 -> 79.27 | Improved |
| use F-norm instead of 1-norm as attention similarity (ablation) | DAA-S · avg F1 over 6 attacks | F1: 75.94 -> 79.27 | Improved |
| vary stability parameter gamma_L from -15 to -0.5 (ablation) | DAA-S · avg AUC over 6 attacks | AUC: 83.33 -> 87.09 | No clear effect |
| vary coupling coefficient c from 0.5 to 10 (ablation) | DAA-S · avg AUC | AUC: 85.40 -> 86.27 | No clear effect |
| start at later time step t=3 with span s=2 vs t=0 with s=1 (ablation) | DAA-I · training F1 | F1: 69.72 -> 76.55 | Improved |
| start at t=3 with span s=2 instead of t=3, s=9 (ablation) | DAA-I · training F1 | F1: 73.70 -> 76.55 | Improved |
| start at t=2 or 3 with long span for DAA-S (ablation) | DAA-S · training F1 | F1: 79.72 -> 4.67 | Hurt |
| apply adaptive attack removing the EOS embedding to EvilEdit | DAA-S · EvilEdit | AUC: 85.95 -> 82.83 | Hurt |
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.
- UFID — UFID: A Unified Framework for Black-box Input-level Backdoor Detection on Diffusion Models (AAAI 2025) · better on 3 of 3 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
- A Closer Look at Backdoor Attacks on CLIP (ICML 2025 · similarity 0.88)
- UFID: A Unified Framework for Black-box Input-level Backdoor Detection on Diffusion Models (AAAI 2025 · similarity 0.88)
- TCAP: Tri-Component Attention Profiling for Unsupervised Backdoor Detection in MLLM Fine-Tuning (ICML 2026 · similarity 0.87)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.87)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.87)
- UIBDiffusion: Universal Imperceptible Backdoor Attack for Diffusion Models (CVPR 2025 · 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.
