Image Captioning on MS-COCO 1,000 images subset 2014 2017 (val)
0.722SentenceBERT ScoreDefense off
Evaluation Results
| Method | Links | |
|---|---|---|
| Defense offModel=LLaVA-1.5-7B, Attack=Clean2026.02 | 0.722 | |
| Improved SmoothVLMModel=LLaVA-1.5-7B, Attack=Clean2026.02 | 0.722 | |
| VALDModel=LLaVA-1.5-7B, Attack=Clean2026.02 | 0.722 | |
| VALDModel=LLaVA-1.5-7B, Attack=MF-ii2026.02 | 0.716 | |
| SmoothVLMModel=LLaVA-1.5-7B, Attack=Clean2026.02 | 0.715 | |
| VALDModel=MiniGPT-v2, Attack=MixAttack2026.02 | 0.711 | |
| VALDModel=MiniGPT-v2, Attack=MF-ii2026.02 | 0.708 | |
| Improved SmoothVLMModel=LLaVA-1.5-7B, Attack=MF-ii2026.02 | 0.704 | |
| Defense offModel=MiniGPT-v2, Attack=Clean2026.02 | 0.703 | |
| VALDModel=MiniGPT-v2, Attack=Clean2026.02 | 0.703 | |
| Improved SmoothVLMModel=MiniGPT-v2, Attack=Clean2026.02 | 0.701 | |
| VALDModel=LLaVA-1.5-7B, Attack=MixAttack2026.02 | 0.694 | |
| SmoothVLMModel=MiniGPT-v2, Attack=Clean2026.02 | 0.685 | |
| Improved SmoothVLMModel=MiniGPT-v2, Attack=MixAttack2026.02 | 0.683 | |
| Improved SmoothVLMModel=LLaVA-1.5-7B, Attack=MixAttack2026.02 | 0.682 | |
| Improved SmoothVLMModel=MiniGPT-v2, Attack=MF-ii2026.02 | 0.681 | |
| SmoothVLMModel=LLaVA-1.5-7B, Attack=MF-ii2026.02 | 0.677 | |
| SmoothVLMModel=MiniGPT-v2, Attack=MF-ii2026.02 | 0.666 | |
| SmoothVLMModel=MiniGPT-v2, Attack=MixAttack2026.02 | 0.657 | |
| SmoothVLMModel=LLaVA-1.5-7B, Attack=MixAttack2026.02 | 0.65 | |
| Defense offModel=LLaVA-1.5-7B, Attack=MixAttack2026.02 | 0.44 | |
| Defense offModel=MiniGPT-v2, Attack=MixAttack2026.02 | 0.297 | |
| Defense offModel=LLaVA-1.5-7B, Attack=MF-ii2026.02 | 0.268 | |
| Defense offModel=MiniGPT-v2, Attack=MF-ii2026.02 | 0.214 |