Visual Question Answering on MMStar (test)
68AccuracyFine-Tuning (Clean)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Fine-Tuning (Clean)Target Model=InternVL2026.05 | 68 | 0 | |
| Robust-U1Category=Ours, Intensity=clean2026.06 | 67.2 | — | |
| BAGELCategory=General MLLM, Intensity=clean2026.06 | 66.13 | — | |
| Robust-U1Category=Ours, Intensity=25%2026.06 | 65.8 | — | |
| MMGUARD-CRS-MaxTarget Model=InternVL2026.05 | 65.2 | 2.8 | |
| Robust-U1Category=Ours, Intensity=50%2026.06 | 64.87 | — | |
| BAGELCategory=General MLLM, Intensity=25%2026.06 | 64.67 | — | |
| Robust-U1Category=Ours, Intensity=100%2026.06 | 63.87 | — | |
| BAGELCategory=General MLLM, Intensity=50%2026.06 | 61.33 | — | |
| BAGELCategory=General MLLM, Intensity=100%2026.06 | 59.6 | — | |
| Robust-R1Category=Robust MLLM, Intensity=clean2026.06 | 56.86 | — | |
| Qwen2.5-VL-3BCategory=General MLLM, Intensity=clean2026.06 | 54.73 | — | |
| Robust-R1Category=Robust MLLM, Intensity=25%2026.06 | 54.4 | — | |
| Robust-R1Category=Robust MLLM, Intensity=50%2026.06 | 53.6 | — | |
| Qwen2.5-VL-3BCategory=General MLLM, Intensity=25%2026.06 | 52.9 | — | |
| Qwen2.5-VL-3BCategory=General MLLM, Intensity=50%2026.06 | 51.86 | — | |
| InternVL-4BCategory=General MLLM, Intensity=clean2026.06 | 51.53 | — | |
| InternVL-4BCategory=General MLLM, Intensity=25%2026.06 | 50.26 | — | |
| InternVL-4BCategory=General MLLM, Intensity=50%2026.06 | 49.6 | — | |
| Robust-R1Category=Robust MLLM, Intensity=100%2026.06 | 49.53 | — | |
| Qwen2.5-VL-3BCategory=General MLLM, Intensity=100%2026.06 | 48.66 | — | |
| InternVL-4BCategory=General MLLM, Intensity=100%2026.06 | 46.93 | — | |
| Gemma3-4BCategory=General MLLM, Intensity=clean2026.06 | 43.93 | — | |
| Gemma3-4BCategory=General MLLM, Intensity=25%2026.06 | 43.2 | — | |
| Gemma3-4BCategory=General MLLM, Intensity=50%2026.06 | 42.6 | — | |
| Gemma3-4BCategory=General MLLM, Intensity=100%2026.06 | 41.33 | — | |
| Robust CLIPCategory=Robust MLLM, Intensity=clean2026.06 | 33 | — | |
| Robust CLIPCategory=Robust MLLM, Intensity=25%2026.06 | 32.26 | — | |
| Robust CLIPCategory=Robust MLLM, Intensity=50%2026.06 | 31.8 | — | |
| TeCoACategory=Robust MLLM, Intensity=50%2026.06 | 30.73 | — | |
| TeCoACategory=Robust MLLM, Intensity=25%2026.06 | 30.6 | — | |
| TeCoACategory=Robust MLLM, Intensity=clean2026.06 | 30.46 | — | |
| Robust CLIPCategory=Robust MLLM, Intensity=100%2026.06 | 29.46 | — | |
| TeCoACategory=Robust MLLM, Intensity=100%2026.06 | 28.06 | — |