Research map / Adversarial robustness and security / Black-box and transfer attacks
CLIP-Guided Generative Networks for Transferable Targeted Adversarial Attacks
Hao Fang, Jiawei Kong, Bin Chen, Tao Dai, Hao Wu, Shu-Tao Xia
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 |
|---|---|---|---|
| use CLIP text embeddings to condition a multi-target generator instead of one-hot conditioning (C-GSP) | CGNC · ImageNet-NeurIPS 1k multi-target | ASR: average +17.88% (Res-152 surrogate), +10.08% (Inc-v3 surrogate) over C-GSP | Improved |
| add cross-attention modules in the decoder (ablation) | CGNC · Res-152 surrogate, VGG-16 victim | ASR: 56.64 -> 63.36 | Improved |
| replace one-hot labels with CLIP text embeddings (without cross-attention) (ablation) | CGNC-CA · Res-152 surrogate, GoogleNet victim | ASR: 51.09 -> 54.29 | Improved |
| add VL-Purifier on CLIP text embedding (ablation) | CGNC · Res-152 surrogate, VGG-16 victim | ASR: 49.84 -> 63.36 | Improved |
| add feature fusion (concatenation) in F-Encoder (ablation) | CGNC · Res-152 surrogate, VGG-16 victim | ASR: 56.85 -> 63.36 | Improved |
| add feature fusion (concatenation) in F-Encoder (ablation) | CGNC · Res-152 surrogate, white-box Res-152 | ASR: 96.45 -> 95.85 | No clear effect |
| apply masked fine-tuning to get single-target generators | CGNC · Inc-v3 surrogate, single-target | black-box ASR: +15.36% average over previous single-target methods | Improved |
| single-target CGNC-dagger on Res-152 surrogate vs DGTA-PI | CGNC · Res-152 surrogate, Inc-Res-v2 victim | ASR: 47.61 -> 45.71 | Hurt |
| CGNC vs C-GSP under JPEG and smoothing defenses | CGNC · Res-152 surrogate input-processing defenses | average ASR: 56.16% -> 80.34% (JPEG), 49.05% -> 71.18% (smoothing) | Improved |
| smaller perturbation budget 8/255 vs 16/255 for CGNC (ablation) | CGNC · Res-152 surrogate, VGG-16 victim | ASR: 63.36 -> 21.46 | Hurt |
Closest papers by meaning
- AIM: Additional Image Guided Generation of Transferable Adversarial Attacks (AAAI 2025 · similarity 0.89)
- X-Transfer Attacks: Towards Super Transferable Adversarial Attacks on CLIP (ICML 2025 · similarity 0.86)
- Dual-Flow: Transferable Multi-Target, Instance-Agnostic Attacks via $\textit{In-the-wild}$ Cascading Flow Optimization (NeurIPS 2025 · similarity 0.86)
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.85)
- Image Hijacks: Adversarial Images can Control Generative Models at Runtime (ICML 2024 · similarity 0.85)
- A Frustratingly Simple Yet Highly Effective Attack Baseline: Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1 (NeurIPS 2025 · similarity 0.85)
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.
