Research map / Adversarial robustness and security / Black-box and transfer attacks
Diffusion Models for Imperceptible and Transferable Adversarial Attack
Jianqi Chen, Hao Chen, Keyan Chen, Yilan Zhang, Zhengxia Zou, Zhenwei Shi
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 |
|---|---|---|---|
| perturb diffusion latent instead of text embedding (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | AVG(w/o self) accuracy: 80.1 -> 65.4 | Improved |
| perturb latent instead of text (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | FID: 58.8 -> 62.3 | Hurt |
| add hard pseudo mask restricting perturbation to foreground (nearest) (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | AVG(w/o self) accuracy: 65.4 -> 74.2 | Hurt |
| add hard pseudo mask restricting perturbation to foreground (nearest) (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | FID: 62.3 -> 59.1 | Improved |
| add soft mask with bilinear upsampling (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | AVG(w/o self) accuracy: 65.4 -> 76.5 | Hurt |
| use top-2 category names as guided text instead of top-1 (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | AVG(w/o self) accuracy: 65.4 -> 70.4 | Hurt |
| use top-5 category names as guided text instead of top-1 (ablation) | DiffAttack · ImageNet-Compatible, surrogate Inc-v3 | AVG(w/o self) accuracy: 65.4 -> 69.1 | Hurt |
| add structure controls (inversion strength and self-attention) (ablation) | DiffAttack · ImageNet-Compatible, surrogate VGG-19 | FID: 96.2 -> 63.9 | Improved |
| add structure controls (inversion strength and self-attention) (ablation) | DiffAttack · ImageNet-Compatible, surrogate VGG-19 | AVG(w/o self) accuracy: 25.5 -> 38.2 | Hurt |
| use DiffAttack versus S2I-FGSM on VGG-19 surrogate | transfer attack · ImageNet-Compatible, surrogate VGG-19 | AVG(w/o self) accuracy: 49.0 -> 38.2 | Improved |
| use DiffAttack versus MI-FGSM on Mob-v2 surrogate | transfer attack · ImageNet-Compatible, surrogate Mob-v2 | AVG(w/o self) accuracy: 49.9 -> 40.5 | Improved |
| use DiffAttack versus MI-FGSM on Inc-v3 surrogate | transfer attack · ImageNet-Compatible, surrogate Inc-v3 | AVG(w/o self) accuracy: 60.6 -> 65.4 | Hurt |
Closest papers by meaning
- Adv-Diffusion: Imperceptible Adversarial Face Identity Attack via Latent Diffusion Model (AAAI 2024 · similarity 0.92)
- AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks (NeurIPS 2024 · similarity 0.91)
- Pixel Is Not a Barrier: An Effective Evasion Attack for Pixel-Domain Diffusion Models (AAAI 2025 · similarity 0.90)
- CAT: Contrastive Adversarial Training for Evaluating the Robustness of Protective Perturbations in Latent Diffusion Models (ICML 2025 · similarity 0.89)
- Perturbing Attention Gives You More Bang for the Buck: Subtle Imaging Perturbations That Efficiently Fool Customized Diffusion Models (CVPR 2024 · similarity 0.89)
- Variance as a Catalyst: Efficient and Transferable Semantic Erasure Adversarial Attack for Customized Diffusion Models (ICML 2025 · 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.
