Research map / Adversarial robustness and security / Attacks on text and graphs
Diff-NAT: Better Naturalistic and Aggressive Adversarial Attacks via Class-Optimized Diffusion for Object Detection
Qinglong Yan, Tong Zou, Xunpeng Yi, Xinyu Xiang, Xuying Wu, Hao Zhang, Jiayi Ma
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 instance-level latent refinement to prompt tuning (ablation) | Diff-NAT · INRIA Person, YOLOv3 | AP (%): 32.97 -> 22.87 | Improved |
| add semantic-level prompt tuning to latent refinement (ablation) | Diff-NAT · INRIA Person, YOLOv3 | AP (%): 56.53 -> 22.87 | Improved |
| increase patch resolution from 256 to 512 (latent-only optimization) (ablation) | Diff-NAT w/o prompt tuning · INRIA Person, YOLOv3 | AP (%): 56.53 -> 24.24 | Improved |
| vary the fixed text prompt (latent-only, 512x512) (ablation) | Diff-NAT w/o prompt tuning · INRIA Person, YOLOv3 | AP (%): 24.24 to 60.11 across prompts | Mixed |
| use Diff-NAT instead of DM-NAP | diffusion-based naturalistic patch · INRIA Person, YOLOv4 (white-box) | AP (%): 14.95 -> 13.71 | Improved |
| use Diff-NAT instead of PG-ECAP / LAP on YOLOv2 | naturalistic patch · INRIA Person, YOLOv2 | AP (%): 9.70 -> 4.37 | Improved |
| transfer Diff-NAT patch trained on YOLOv2 to YOLOv3 | Diff-NAT · INRIA Person, transfer | AP (%): PG-ECAP 36.04 vs Diff-NAT 50.97 | Hurt |
| use Diff-NAT on cross-dataset CCTV-Person with YOLOv4 | Diff-NAT · CCTV-Person | AP (%): 5.00 -> 3.45 | Improved |
| use Diff-NAT patch in human naturalness vote | Diff-NAT YOLOv3 patch · human study, 45 participants | naturalness vote (%): 75.56% | Improved |
Closest papers by meaning
- Intriguing Properties of Diffusion Models: An Empirical Study of the Natural Attack Capability in Text-to-Image Generative Models (CVPR 2024 · similarity 0.88)
- DiffAdvMAP: Flexible Diffusion-Based Framework for Generating Natural Unrestricted Adversarial Examples (ICML 2025 · similarity 0.88)
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.87)
- DiffHammer: Rethinking the Robustness of Diffusion-Based Adversarial Purification (NeurIPS 2024 · similarity 0.86)
- Variance as a Catalyst: Efficient and Transferable Semantic Erasure Adversarial Attack for Customized Diffusion Models (ICML 2025 · similarity 0.86)
- AdvI2I: Adversarial Image Attack on Image-to-Image Diffusion Models (ICML 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.
