Image Captioning on MS COCO (CLIPScore)
0.865CLIPScoreNo Defense
Evaluation Results
| Method | Links | |
|---|---|---|
| No DefenseVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.865 | |
| PDAVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.862 | |
| SemanticSmooth-SummariseVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.861 | |
| PDAVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.86 | |
| No DefenseVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.858 | |
| SmoothLLM-InsertVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.855 | |
| SemanticSmooth-ParaphraseVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.855 | |
| SmoothLLM-InsertVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.854 | |
| SmoothLLM-PatchVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.851 | |
| SemanticSmooth-SummariseVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.851 | |
| SemanticSmooth-ParaphraseVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.85 | |
| SmoothLLM-SwapVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.849 | |
| SmoothLLM-SwapVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.849 | |
| SmoothLLM-PatchVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.843 | |
| FAREVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.835 | |
| FAREVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.833 | |
| TeCoAVictim Model=LLaVA-1.5-13B, Input Condition=Clean2026.04 | 0.823 | |
| TeCoAVictim Model=LLaVA-1.5-7B, Input Condition=Clean2026.04 | 0.811 | |
| FAREVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.806 | |
| FAREVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.805 | |
| TeCoAVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.791 | |
| TeCoAVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.783 | |
| PDAVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.778 | |
| PDAVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.774 | |
| No DefenseVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.757 | |
| SmoothLLM-SwapVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.756 | |
| SemanticSmooth-SummariseVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.754 | |
| No DefenseVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.753 | |
| SmoothLLM-PatchVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.751 | |
| SemanticSmooth-ParaphraseVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.751 | |
| SemanticSmooth-SummariseVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.75 | |
| SemanticSmooth-ParaphraseVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.748 | |
| SmoothLLM-InsertVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.746 | |
| SmoothLLM-SwapVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.744 | |
| SmoothLLM-InsertVictim Model=LLaVA-1.5-13B, Input Condition=Adversarial2026.04 | 0.743 | |
| SmoothLLM-PatchVictim Model=LLaVA-1.5-7B, Input Condition=Adversarial2026.04 | 0.741 |