Jailbreak Attack Success Evaluation on MM-SafetyBench TYPO
79.6Attack Success Rate (ASR)LLaVA-1.5-7B (Baseline)
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-1.5-7B (Baseline)VLM=LLaVA-1.5-7B2026.03 | 79.6 | |
| InternVL-Chat-19B (Baseline)VLM=InternVL-Chat-19B2026.03 | 67.9 | |
| ShareGPT4V-7B (Baseline)VLM=ShareGPT4V-7B2026.03 | 65.1 | |
| CMRMVLM=LLaVA-1.5-7B2026.03 | 47 | |
| ShiftDCVLM=LLaVA-1.5-7B2026.03 | 42.2 | |
| ShiftDCVLM=InternVL-Chat-19B2026.03 | 39.5 | |
| AdaShieldVLM=InternVL-Chat-19B2026.03 | 36.2 | |
| AdaShieldVLM=LLaVA-1.5-7B2026.03 | 32.6 | |
| CMRMVLM=InternVL-Chat-19B2026.03 | 22.5 | |
| ECSOVLM=InternVL-Chat-19B2026.03 | 20.7 | |
| ECSOVLM=LLaVA-1.5-7B2026.03 | 14 | |
| ShiftDCVLM=ShareGPT4V-7B2026.03 | 13.6 | |
| CMRMVLM=ShareGPT4V-7B2026.03 | 12.6 | |
| AdaShieldVLM=ShareGPT4V-7B2026.03 | 11.8 | |
| JRS-RemVLM=InternVL-Chat-19B2026.03 | 10.7 | |
| ECSOVLM=ShareGPT4V-7B2026.03 | 10 | |
| JRS-RemVLM=LLaVA-1.5-7B2026.03 | 8 | |
| JRS-RemVLM=ShareGPT4V-7B2026.03 | 4.2 |