Adversarial Attack on VQA v2 (test)
89.6Classification AccuracyMMAS
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MMASTarget Model=MiniGPT-4, config=Full2026.05 | 89.6 | 88.5 | 89.8 | 93.4 | 90.3 | |
| MMASTarget Model=Flamingo, config=Full2026.05 | 86.9 | 87.5 | 85.2 | 89.9 | 87.4 | |
| TA-UAPTarget Model=MiniGPT-42026.05 | 84.6 | 83.5 | 84.8 | 88.4 | 85.3 | |
| MMASTarget Model=LLaVA, config=Full2026.05 | 84.3 | 81.2 | 85.9 | 89.7 | 85.3 | |
| TA-UAPTarget Model=Flamingo2026.05 | 82.9 | 78.6 | 82.7 | 86.5 | 82.7 | |
| MMASTarget Model=BLIP-2, config=Full2026.05 | 82.6 | 87.5 | 89.6 | 92.4 | 88 | |
| TC-UAPTarget Model=MiniGPT-42026.05 | 81.6 | 80.5 | 81.8 | 85.4 | 82.3 | |
| TC-UAPTarget Model=Flamingo2026.05 | 79.9 | 76.8 | 80.3 | 84.1 | 80.3 | |
| TA-UAPTarget Model=LLaVA2026.05 | 79.3 | 76.2 | 80.9 | 84.7 | 80.3 | |
| TA-UAPTarget Model=BLIP-22026.05 | 77.6 | 82.5 | 84.6 | 87.4 | 83 | |
| TC-UAPTarget Model=LLaVA2026.05 | 76.3 | 73.2 | 77.9 | 81.7 | 77.3 | |
| TC-UAPTarget Model=BLIP-22026.05 | 74.6 | 79.5 | 81.6 | 84.4 | 80 | |
| CleanTarget Model=MiniGPT-42026.05 | 40.2 | 41.4 | 51.9 | 54.3 | 47 | |
| CleanTarget Model=LLaVA2026.05 | 40.1 | 42.3 | 47.5 | 48.7 | 44.7 | |
| CleanTarget Model=Flamingo2026.05 | 34.9 | 38.2 | 39.6 | 25 | 34.4 | |
| CleanTarget Model=BLIP-22026.05 | 31.2 | 34 | 35.7 | 39.8 | 35.2 |