Research map / Adversarial robustness and security / Backdoor and poisoning
TooBad: Backdoor Diffusion Models with Ultra-Low Poison Rate and Imperceptible Trigger
Vu Truong, Long Bao Le
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 |
|---|---|---|---|
| optimize the trigger with the diffusion-tailored loss instead of using a fixed stop-sign trigger (VillanDiffusion) | backdoor injection on DDPM · CIFAR-10 (1% poison) | ASR: 0.00 -> 0.93 | Improved |
| use TooBad trigger optimization at 5% poison | DDPM backdoor · CIFAR-10 (5% poison) | ASR: 0.38 -> 0.98 (vs VillanDiffusion) | Improved |
| use TooBad on NCSN at lower poison rates | NCSN backdoor · CIFAR-10 NCSN | ASR: 0.70 -> 0.98 at 50% poison | Improved |
| use TooBad on LDM for high-res data | LDM backdoor · CelebA-HQ (5% poison) | ASR: 0.32 -> 0.98 (vs VillanDiffusion) | Improved |
| add the invisibility constraint (epsilon bound) to trigger optimization (ablation) | TooBad · CIFAR-10 TERD defense | detection TPR: 76.66 -> 0.00 | Improved |
| add the sparsity constraint to trigger optimization (ablation) | TooBad · CIFAR-10 TERD defense | detection TPR: 12.03 -> 0.00 | Improved |
| use different backdoor targets (cat, stop sign) instead of fedora hat (ablation) | TooBad · CIFAR-10 ASR | ASR: 0.66 and 0.63 at 0.2% poison; 0.99 at 5% | No clear effect |
| TooBad reaches high ASR in few injection epochs | TooBad vs baselines · CIFAR-10 5% poison | epochs to near-100% ASR: 3-5 epochs vs 30-50 epochs | Improved |
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.
- UIBDiffusion — UIBDiffusion: Universal Imperceptible Backdoor Attack for Diffusion Models (CVPR 2025) · 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
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.90)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.89)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.89)
- UIBDiffusion: Universal Imperceptible Backdoor Attack for Diffusion Models (CVPR 2025 · similarity 0.88)
- Elijah: Eliminating Backdoors Injected in Diffusion Models via Distribution Shift (AAAI 2024 · similarity 0.88)
- SampDetox: Black-box Backdoor Defense via Perturbation-based Sample Detoxification (NeurIPS 2024 · similarity 0.87)
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.
