Visual Jailbreak Defense on Visual Adversarial Attacks epsilon = 32/255
14.97ASRIMMUNE
Evaluation Results
| Method | Links | |
|---|---|---|
| IMMUNEModel=LLaVA-1.52024.11 | 14.97 | |
| IMMUNEModel=Qwen-VL2024.11 | 15.07 | |
| IMMUNEModel=LLaVA-1.62024.11 | 16.71 | |
| AdaShieldModel=Qwen-VL2024.11 | 18.2 | |
| IMMUNEModel=MiniGPT-4-7B2024.11 | 20.77 | |
| AdaShieldModel=LLaVA-1.62024.11 | 24.15 | |
| IMMUNEModel=MiniGPT-4-13B2024.11 | 25.82 | |
| FigStepModel=Qwen-VL2024.11 | 36.92 | |
| AdaShieldModel=LLaVA-1.52024.11 | 39.71 | |
| CoCAModel=LLaVA-1.62024.11 | 41.88 | |
| AdaShieldModel=MiniGPT-4-7B2024.11 | 42.45 | |
| CoCAModel=Qwen-VL2024.11 | 44.73 | |
| CoCAModel=LLaVA-1.52024.11 | 46.11 | |
| OriginalModel=Qwen-VL2024.11 | 48.03 | |
| FigStepModel=LLaVA-1.52024.11 | 50.3 | |
| AdaShieldModel=MiniGPT-4-13B2024.11 | 50.37 | |
| FigStepModel=LLaVA-1.62024.11 | 52.07 | |
| FigStepModel=MiniGPT-4-7B2024.11 | 52.43 | |
| OriginalModel=MiniGPT-4-7B2024.11 | 58.12 | |
| CoCAModel=MiniGPT-4-7B2024.11 | 60.09 | |
| FigStepModel=MiniGPT-4-13B2024.11 | 61.05 | |
| OriginalModel=LLaVA-1.52024.11 | 64.08 | |
| OriginalModel=MiniGPT-4-13B2024.11 | 66.13 | |
| OriginalModel=LLaVA-1.62024.11 | 67.19 | |
| CoCAModel=MiniGPT-4-13B2024.11 | 74.31 |