Direct Malicious on VLSafe OOD
90.67ASRSelfReminder
Evaluation Results
| Method | Links | |
|---|---|---|
| SelfReminderBase Model=LLaVA-1.5-13B2026.04 | 90.67 | |
| JailGuardBase Model=LLaVA-1.5-13B2026.04 | 72.43 | |
| CIDERBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 50 | |
| CIDERBase Model=LLaVA-1.5-13B2026.04 | 50 | |
| JailGuardBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 31.35 | |
| MirrorCheckBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 26.33 | |
| MirrorCheckBase Model=LLaVA-1.5-13B2026.04 | 26.33 | |
| ECSOBase Model=LLaVA-1.5-13B2026.04 | 24 | |
| ECSOBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 19.63 | |
| VLMGuardBase Model=LLaVA-1.5-13B2026.04 | 15.27 | |
| VLMGuardBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=Internal Defense2026.04 | 10.32 | |
| SelfReminderBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 8.55 | |
| ASTRABase Model=LLaVA-1.5-13B2026.04 | 8.03 | |
| ASTRABase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=Internal Defense2026.04 | 3.7 | |
| VLMShieldBase VLM=Qwen2.5-VL-7B-Instruct, Defense Category=External Defense2026.04 | 1.62 | |
| VLMShieldBase Model=LLaVA-1.5-13B2026.04 | 1.62 |