List your company
ResearchFormerly WizWand

Malicious Prompt Detection on MM-Vet OOD

3.67FPRVLMShield
3.37845.34677.3159.2833Apr 7, 2026
Updated 6mo ago

Evaluation results

MethodLinks
VLMShieldTarget Model=LLaVA-1.5-13B2026.043.67
VLMShieldTarget Model=Qwen2.5-VL-7B-Instruct2026.043.67
VLMGuardTarget Model=Qwen2.5-VL-7B-Instruct2026.043.92
ASTRATarget Model=Qwen2.5-VL-7B-Instruct2026.044.2
VLMGuardTarget Model=LLaVA-1.5-13B2026.045
JailGuardTarget Model=Qwen2.5-VL-7B-Instruct2026.045.62
ASTRATarget Model=LLaVA-1.5-13B2026.046.46
CIDERTarget Model=LLaVA-1.5-13B2026.046.72
CIDERTarget Model=Qwen2.5-VL-7B-Instruct2026.046.72
ECSOTarget Model=Qwen2.5-VL-7B-Instruct2026.046.77
JailGuardTarget Model=LLaVA-1.5-13B2026.0410.55
MirrorCheckTarget Model=LLaVA-1.5-13B2026.0410.59
MirrorCheckTarget Model=Qwen2.5-VL-7B-Instruct2026.0410.59
ECSOTarget Model=LLaVA-1.5-13B2026.0410.96