Textual Modal Attack on LAION-COCO subset, UnsafeDiff, and I2P NSFW prompts (test)
84.9Q16 ASR (Step 4)MMA-DIFFUSION
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MMA-DIFFUSIONTarget Model=SDv1.5 [5], Attack Setting=White-box, Prompt Type=Adversarial2023.11 | 84.9 | 73.23 | 84.8 | 75.1 | 80.4 | 54.2 | 83.37 | 67.54 | |
| MMA-DIFFUSIONTarget Model=SDXLv1.0 [24], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 76.3 | 49.28 | 71.7 | 44.87 | 73.1 | 40.38 | 73.7 | 44.84 | |
| MMA-DIFFUSIONTarget Model=SLD [34], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 75.6 | 53.05 | 78.7 | 61.33 | 75.9 | 45.72 | 76.73 | 53.37 | |
| I2PTarget Model=SDv1.5 [5], Attack Setting=White-box, Prompt Type=Human-written2023.11 | 69.68 | 46.05 | 52.04 | 31.42 | 61.9 | 32.28 | 61.27 | 36.58 | |
| I2PTarget Model=SLD [34], Attack Setting=Black-box, Prompt Type=Human-written2023.11 | 39.89 | 20.48 | 32.04 | 16.42 | 28.39 | 12.37 | 33.44 | 16.42 | |
| GENETICTarget Model=SDv1.5 [5], Attack Setting=White-box, Prompt Type=Adversarial2023.11 | 39 | 20.05 | 33.6 | 18 | 35.26 | 14.85 | 28.45 | 12.22 | |
| GREEDYTarget Model=SDv1.5 [5], Attack Setting=White-box, Prompt Type=Adversarial2023.11 | 37.89 | 18.23 | 35.9 | 18.65 | 36.9 | 16.9 | 29.1 | 13.48 | |
| QF-PGDTarget Model=SDv1.5 [5], Attack Setting=White-box, Prompt Type=Adversarial2023.11 | 27.4 | 11.35 | 20.7 | 7.75 | 26.26 | 9.7 | 21.02 | 7.57 | |
| GREEDYTarget Model=SLD [34], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 21.8 | 9.08 | 23.1 | 10.13 | 23.1 | 8.92 | 22.67 | 9.37 | |
| GENETICTarget Model=SLD [34], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 19.3 | 7.78 | 20.5 | 9.72 | 23.4 | 8.8 | 21.07 | 8.77 | |
| QF-PGDTarget Model=SLD [34], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 12.8 | 4.4 | 11.8 | 4.6 | 13.6 | 5.18 | 12.73 | 4.73 | |
| I2PTarget Model=SDXLv1.0 [24], Attack Setting=Black-box, Prompt Type=Human-written2023.11 | 9.6 | 8.24 | 5.97 | 4.48 | 6.31 | 3.3 | 7.29 | 5.34 | |
| GREEDYTarget Model=SDXLv1.0 [24], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 3.2 | 1.15 | 1.88 | 0.67 | 1.92 | 0.7 | 2.34 | 0.84 | |
| QF-PGDTarget Model=SDXLv1.0 [24], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 2.24 | 0.78 | 1.54 | 0.46 | 1.63 | 0.51 | 1.8 | 0.58 | |
| GENETICTarget Model=SDXLv1.0 [24], Attack Setting=Black-box, Prompt Type=Adversarial2023.11 | 1.57 | 0.57 | 3.44 | 1.26 | 2.08 | 0.75 | 2.36 | 0.86 |