Adversarial Attack on Mantis Eval
84.57Attack Success RateLAMP
Evaluation results
| Method | Links | |
|---|---|---|
| LAMPTarget Model=Qwen-VL-Chat, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 84.57 | |
| LAMPTarget Model=Qwen-2.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 82.71 | |
| LAMPTarget Model=LLaVa-v1.6, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 79.65 | |
| LAMPTarget Model=Mantis-Idefics2, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 75.44 | |
| LAMPTarget Model=VILA-1.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 75.21 | |
| LAMPTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 73.43 | |
| LAMPTarget Model=Mantis-SIGLIP, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 73.21 | |
| UAP-VLPTarget Model=Qwen-VL-Chat, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 66.68 | |
| UAP-VLPTarget Model=Qwen-2.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 65.45 | |
| CPGC-UAPTarget Model=Qwen-VL-Chat, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 64.68 | |
| CPGC-UAPTarget Model=Qwen-2.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 63.84 | |
| UAP-VLPTarget Model=LLaVa-v1.6, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 60.79 | |
| CPGC-UAPTarget Model=LLaVa-v1.6, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 59.76 | |
| Pandora's boxTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 57.58 | |
| Doubly-UAPTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 56.87 | |
| Doubly-UAPTarget Model=Mantis-SIGLIP, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 56.74 | |
| UAP-VLPTarget Model=Mantis-SIGLIP, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 55.55 | |
| UAP-VLPTarget Model=VILA-1.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 54.79 | |
| UAP-VLPTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 54.68 | |
| Jailbreak-MLLMTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 54.68 | |
| UAP-VLPTarget Model=Mantis-Idefics2, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 54.22 | |
| Jailbreak-MLLMTarget Model=Mantis-SIGLIP, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 53.68 | |
| X-transferTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 52.82 | |
| Doubly-UAPTarget Model=Mantis-Idefics2, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 52.74 | |
| CPGC-UAPTarget Model=VILA-1.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 52.68 | |
| CPGC-UAPTarget Model=Mantis-SIGLIP, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 52.13 | |
| CPGC-UAPTarget Model=Mantis-Idefics2, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 51.79 | |
| CPGC-UAPTarget Model=Mantis-CLIP, Evaluation Protocol=Standard2026.01 | 51.66 | |
| Doubly-UAPTarget Model=LLaVa-v1.6, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 51.43 | |
| Doubly-UAPTarget Model=VILA-1.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 50.74 | |
| Doubly-UAPTarget Model=Qwen-VL-Chat, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 49.42 | |
| Doubly-UAPTarget Model=Qwen-2.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 48.41 | |
| Jailbreak-MLLMTarget Model=Mantis-Idefics2, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 40.81 | |
| Jailbreak-MLLMTarget Model=VILA-1.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 36.15 | |
| Jailbreak-MLLMTarget Model=LLaVa-v1.6, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 36.15 | |
| Jailbreak-MLLMTarget Model=Qwen-VL-Chat, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 33.15 | |
| Jailbreak-MLLMTarget Model=Qwen-2.5, Evaluation Protocol=Zero-shot Cross-Model Evaluation2026.01 | 30.76 |