Malicious Prompt Detection on MMBench OOD
0.16FPRVLMShield
Evaluation Results
| Method | Links | |
|---|---|---|
| VLMShieldTarget Model=LLaVA-1.5-13B2026.04 | 0.16 | |
| VLMShieldTarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 0.16 | |
| VLMGuardTarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 2 | |
| ASTRATarget Model=LLaVA-1.5-13B2026.04 | 2.34 | |
| CIDERTarget Model=LLaVA-1.5-13B2026.04 | 2.54 | |
| CIDERTarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 2.54 | |
| VLMGuardTarget Model=LLaVA-1.5-13B2026.04 | 3.08 | |
| ECSOTarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 4.93 | |
| JailGuardTarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 5 | |
| ASTRATarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 5.36 | |
| ECSOTarget Model=LLaVA-1.5-13B2026.04 | 7.2 | |
| JailGuardTarget Model=LLaVA-1.5-13B2026.04 | 8.75 | |
| MirrorCheckTarget Model=LLaVA-1.5-13B2026.04 | 9.83 | |
| MirrorCheckTarget Model=Qwen2.5-VL-7B-Instruct2026.04 | 9.83 |