Direct Malicious on MM-SafetyBench OOD
0.71ASRVLMShield
Evaluation Results
| Method | Links | |
|---|---|---|
| VLMShieldBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 0.71 | |
| VLMShieldBase Model=LLaVA-1.5-13B2026.04 | 0.71 | |
| ASTRABase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=Internal Defense2026.04 | 4.12 | |
| ASTRABase Model=LLaVA-1.5-13B2026.04 | 8.62 | |
| VLMGuardBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=Internal Defense2026.04 | 9.05 | |
| VLMGuardBase Model=LLaVA-1.5-13B2026.04 | 12.9 | |
| ECSOBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 13.28 | |
| SelfReminderBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 15.2 | |
| ECSOBase Model=LLaVA-1.5-13B2026.04 | 18.39 | |
| JailGuardBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 24.35 | |
| MirrorCheckBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 25.08 | |
| MirrorCheckBase Model=LLaVA-1.5-13B2026.04 | 25.08 | |
| JailGuardBase Model=LLaVA-1.5-13B2026.04 | 36.22 | |
| CIDERBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 46.91 | |
| CIDERBase Model=LLaVA-1.5-13B2026.04 | 46.91 | |
| SelfReminderBase Model=LLaVA-1.5-13B2026.04 | 51.27 |