Jailbreak Attack Success Evaluation on MM-SafetyBench SD+TYPO
81.4ASRLLaVA-1.5-7B (Baseline)
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-1.5-7B (Baseline)VLM=LLaVA-1.5-7B2026.03 | 81.4 | |
| ShareGPT4V-7B (Baseline)VLM=ShareGPT4V-7B2026.03 | 73.1 | |
| InternVL-Chat-19B (Baseline)VLM=InternVL-Chat-19B2026.03 | 73.1 | |
| CMRMVLM=LLaVA-1.5-7B2026.03 | 48.7 | |
| ShiftDCVLM=LLaVA-1.5-7B2026.03 | 45.7 | |
| ShiftDCVLM=InternVL-Chat-19B2026.03 | 42.5 | |
| AdaShieldVLM=LLaVA-1.5-7B2026.03 | 35.3 | |
| AdaShieldVLM=InternVL-Chat-19B2026.03 | 33.3 | |
| ECSOVLM=InternVL-Chat-19B2026.03 | 28.3 | |
| ECSOVLM=LLaVA-1.5-7B2026.03 | 23.1 | |
| CMRMVLM=InternVL-Chat-19B2026.03 | 20.8 | |
| ECSOVLM=ShareGPT4V-7B2026.03 | 17.3 | |
| ShiftDCVLM=ShareGPT4V-7B2026.03 | 16.6 | |
| AdaShieldVLM=ShareGPT4V-7B2026.03 | 13.1 | |
| CMRMVLM=ShareGPT4V-7B2026.03 | 12.3 | |
| JRS-RemVLM=LLaVA-1.5-7B2026.03 | 11.7 | |
| JRS-RemVLM=InternVL-Chat-19B2026.03 | 5.7 | |
| JRS-RemVLM=ShareGPT4V-7B2026.03 | 5.1 |