Immunization on SHHQ Target: Grok
94Immunization RateMIRAGE
Evaluation results
| Method | Links | |
|---|---|---|
| MIRAGEEvaluation Protocol=Explicit API refusal, Perturbation budget=16/2552026.06 | 94 | |
| TDAEEvaluation Protocol=VLM-as-a-judge, Perturbation budget=16/2552026.06 | 7.4 | |
| PhotoGuardEvaluation Protocol=VLM-as-a-judge, Perturbation budget=16/2552026.06 | 2 | |
| PhotoGuard-VAEEvaluation Protocol=VLM-as-a-judge, Perturbation budget=16/2552026.06 | 2 | |
| BlurGuardEvaluation Protocol=VLM-as-a-judge, Perturbation budget=16/2552026.06 | 2 | |
| SquareAttackEvaluation Protocol=Explicit API refusal, Perturbation budget=16/2552026.06 | 0.7 |