Research map / Adversarial robustness and security / Attacks on text and graphs
DIA: The Adversarial Exposure of Deterministic Inversion in Diffusion Models
Seunghoo Hong, Geonho Son, Juhun Lee, Simon S. Woo
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 DDIM trajectory-targeting objective (DIA-PT) vs PhotoGuard | DDIM inversion + MasaCtrl · PIE-Bench | CLIP similarity: 22.8856 -> 18.3076 | Improved |
| DIA-PT vs Glaze | DDIM inversion + MasaCtrl · PIE-Bench | CLIP similarity: 23.8529 -> 18.3076 | Improved |
| DIA-PT vs PID | DDIM-to-DDIM · PIE-Bench | CLIP similarity: 21.2091 -> 23.4614 | Hurt |
| DIA-PT/R vs natural edit with Direct-to-P2P | Direct Inversion + P2P · PIE-Bench | CLIP similarity: 25.8333 -> 26.0563 | No clear effect |
| add x0 scaling to the trajectory attack (DIA-PT instead of DIA-MT) (ablation) | DIA · PIE-Bench DDIM-to-DDIM | CLIP similarity: 23.7177 -> 23.4614 | Improved |
| add x0 scaling (DIA-PT instead of DIA-MT) (ablation) | DIA · PIE-Bench | structure distance: 0.0514 -> 0.1059 | Improved |
| increase DDIM steps from 20 to 1000 | DIA-PT · PIE-Bench 140 images | CLIP similarity: 20.2686 -> 24.2379 | Hurt |
| increase DDIM steps from 20 to 1000 | DIA-R · PIE-Bench 140 images | CLIP similarity: 21.2578 -> 24.2615 | Hurt |
| increase attack iterations from 5 to 20 (ablation) | DIA-PT · PIE-Bench | CLIP similarity: 25.6048 -> 23.4575 | Improved |
| increase attack iterations from 5 to 20 (ablation) | DIA-R · PIE-Bench | CLIP similarity: 24.6790 -> 23.4670 | Improved |
| increase trajectory length from 10 to 20 (ablation) | DIA-PT · PIE-Bench | CLIP similarity: 23.4575 -> 24.1361 | Hurt |
| increase trajectory length from 10 to 20 (ablation) | DIA-R · PIE-Bench | CLIP similarity: 23.4670 -> 22.0941 | Improved |
Closest papers by meaning
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.89)
- Pixel Is Not a Barrier: An Effective Evasion Attack for Pixel-Domain Diffusion Models (AAAI 2025 · similarity 0.89)
- MimicDiffusion: Purifying Adversarial Perturbation via Mimicking Clean Diffusion Model (CVPR 2024 · similarity 0.88)
- An Edit Friendly DDPM Noise Space: Inversion and Manipulations (CVPR 2024 · similarity 0.88)
- Intriguing Properties of Diffusion Models: An Empirical Study of the Natural Attack Capability in Text-to-Image Generative Models (CVPR 2024 · similarity 0.87)
- Hiding Images in Diffusion Models by Editing Learned Score Functions (CVPR 2025 · 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.
