Research map / Adversarial robustness and security / Backdoor and poisoning
Semantic-level Backdoor Attack against Text-to-Image Diffusion Models
Tianxin Chen, Wenbo Jiang, Hongqiao Chen, Zhirun Zheng, Cheng Huang
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 semantic regularization (lambda_reg 0 -> 0.5) (ablation) | SemBD on SDv1.5 · trigger specificity | FTR (%): 77.78 -> 3.04 | Improved |
| add semantic regularization (lambda_reg 0 -> 0.5) (ablation) | SemBD on SDXL · trigger specificity | FTR (%): 80.85 -> 5.72 | Improved |
| increase lambda_reg from 0.5 to 0.9 (ablation) | SemBD on SDXL · trigger specificity | ASR (%): 100 -> 72.56 | Hurt |
| use multi-entity target instead of single-entity target (ablation) | SemBD · NaviT2I defense | DSR (%): 34.0 -> 12.00 | Improved |
| use multi-entity target instead of single-entity target (ablation) | SemBD · T2IShield CDA defense | DSR (%): 98.0 -> 2.00 | Improved |
| increase pruning ratio from 0.3 to 0.5 against SemBD | SemBD on SDv1.5 · pruning defense | ASR (%): 97 -> 0 | Hurt |
| apply pruning ratio up to 0.3 | SemBD on SDv1.5 · pruning defense | ASR (%): 100 -> 97 | No clear effect |
| increase number of semantic triggers (ablation) | SemBD · ASR | ASR (%) | Improved |
| fine-tune the backdoored model on clean data (full and LoRA) | SemBD on SDv1.5 · fine-tuning defense | ASR (%): ASR consistently above 90% | No clear effect |
| semantic-level trigger vs word-level triggers (SemBD vs Rickrolling) | SemBD · NaviT2I defense | DSR (%): 68.60 -> 12.00 | Improved |
Closest papers by meaning
- Personalization as a Shortcut for Few-Shot Backdoor Attack against Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.89)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.88)
- Distraction is All You Need: Memory-Efficient Image Immunization against Diffusion-Based Image Editing (CVPR 2024 · similarity 0.87)
- UFID: A Unified Framework for Black-box Input-level Backdoor Detection on Diffusion Models (AAAI 2025 · similarity 0.87)
- Semantic Shield: Defending Vision-Language Models Against Backdooring and Poisoning via Fine-grained Knowledge Alignment (CVPR 2024 · similarity 0.86)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 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.
