Multimodal Prompt Injection Attack on MSCOCO
96.08ASRCrossMPI
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CrossMPITarget Model=BLIP-2, Best Source Model=MiniGPT4-llama22026.05 | 96.08 | 65.72 | |
| CrossMPITarget Model=BLIVA, Best Source Model=MiniGPT4-llama22026.05 | 91.18 | 53.03 | |
| CrossMPITarget Model=InstructBLIP, Best Source Model=MiniGPT4-llama22026.05 | 90.57 | 52.57 | |
| CrossMPITarget Model=MiniGPT4-vicuna, Best Source Model=MiniGPT4-llama22026.05 | 90.2 | 71.22 | |
| CITarget Model=BLIP-2, Best Source Model=MiniGPT4-llama22026.05 | 88.1 | 61.34 | |
| CITarget Model=BLIVA, Best Source Model=MiniGPT4-llama22026.05 | 66.67 | 45.12 | |
| ARE-BTarget Model=InstructBLIP, Best Source Model=MiniGPT4-llama22026.05 | 56.41 | 42.42 | |
| ARE-BTarget Model=BLIVA, Best Source Model=MiniGPT4-llama22026.05 | 56.41 | 42.95 | |
| ARE-BTarget Model=BLIP-2, Best Source Model=MiniGPT4-llama22026.05 | 53.85 | 46.73 | |
| CITarget Model=InstructBLIP, Best Source Model=MiniGPT4-llama22026.05 | 49.02 | 45.51 | |
| CITarget Model=MiniGPT4-vicuna, Best Source Model=MiniGPT4-llama22026.05 | 40.48 | 44.19 | |
| CrossMPITarget Model=MiniGPT4-llama2, Best Source Model=MiniGPT4-vicuna2026.05 | 28.43 | 43.17 | |
| ARE-BTarget Model=MiniGPT4-vicuna, Best Source Model=MiniGPT4-llama22026.05 | 28.21 | 48.65 | |
| CITarget Model=Qwen2.5-VL, Best Source Model=Qwen2.5-VL2026.05 | 26.19 | 37.3 | |
| ARE-BTarget Model=Qwen2.5-VL, Best Source Model=Qwen2.5-VL2026.05 | 20.51 | 39.3 | |
| CITarget Model=MiniGPT4-llama2, Best Source Model=MiniGPT4-vicuna2026.05 | 19.05 | 30.94 | |
| ARE-BTarget Model=MiniGPT4-llama2, Best Source Model=MiniGPT4-vicuna2026.05 | 15.38 | 40.1 | |
| CrossMPITarget Model=Qwen2.5-VL, Best Source Model=Qwen2.5-VL2026.05 | 12.75 | 38.74 | |
| ARE-WTarget Model=BLIP-2, Best Source Model=MiniGPT4-llama22026.05 | 10.2 | 20.33 | |
| ARE-WTarget Model=InstructBLIP, Best Source Model=MiniGPT4-llama22026.05 | 8.16 | 30.98 | |
| ARE-WTarget Model=BLIVA, Best Source Model=MiniGPT4-llama22026.05 | 6.12 | 30.72 | |
| ARE-WTarget Model=MiniGPT4-vicuna, Best Source Model=MiniGPT4-llama22026.05 | 2.04 | 37.24 | |
| ARE-WTarget Model=MiniGPT4-llama2, Best Source Model=MiniGPT4-vicuna2026.05 | 1.35 | 37.82 | |
| ATPITarget Model=MiniGPT4-llama2, Best Source Model=MiniGPT4-vicuna2026.05 | 0 | 29.05 | |
| ATPITarget Model=MiniGPT4-vicuna, Best Source Model=MiniGPT4-llama22026.05 | 0 | 28.98 | |
| ATPITarget Model=InstructBLIP, Best Source Model=MiniGPT4-llama22026.05 | 0 | 25.3 | |
| ATPITarget Model=BLIP-2, Best Source Model=MiniGPT4-llama22026.05 | 0 | 12.67 | |
| ATPITarget Model=BLIVA, Best Source Model=MiniGPT4-llama22026.05 | 0 | 28.76 | |
| ARE-WTarget Model=Qwen2.5-VL, Best Source Model=Qwen2.5-VL2026.05 | 0 | 33.8 | |
| ATPITarget Model=Qwen2.5-VL, Best Source Model=Qwen2.5-VL2026.05 | 0 | 30.24 |