List your company

Visual Question Answering on MMStar (test)

68AccuracyFine-Tuning (Clean)
26.462437.246248.0358.8138May 14, 2026May 17, 2026May 21, 2026May 25, 2026May 29, 2026Jun 2, 2026Jun 6, 2026

Evaluation results

MethodLinks
Fine-Tuning (Clean)Target Model=InternVL2026.05680
Robust-U1Category=Ours, Intensity=clean2026.0667.2—
BAGELCategory=General MLLM, Intensity=clean2026.0666.13—
Robust-U1Category=Ours, Intensity=25%2026.0665.8—
MMGUARD-CRS-MaxTarget Model=InternVL2026.0565.22.8
Robust-U1Category=Ours, Intensity=50%2026.0664.87—
BAGELCategory=General MLLM, Intensity=25%2026.0664.67—
Robust-U1Category=Ours, Intensity=100%2026.0663.87—
BAGELCategory=General MLLM, Intensity=50%2026.0661.33—
BAGELCategory=General MLLM, Intensity=100%2026.0659.6—
Robust-R1Category=Robust MLLM, Intensity=clean2026.0656.86—
Qwen2.5-VL-3BCategory=General MLLM, Intensity=clean2026.0654.73—
Robust-R1Category=Robust MLLM, Intensity=25%2026.0654.4—
Robust-R1Category=Robust MLLM, Intensity=50%2026.0653.6—
Qwen2.5-VL-3BCategory=General MLLM, Intensity=25%2026.0652.9—
Qwen2.5-VL-3BCategory=General MLLM, Intensity=50%2026.0651.86—
InternVL-4BCategory=General MLLM, Intensity=clean2026.0651.53—
InternVL-4BCategory=General MLLM, Intensity=25%2026.0650.26—
InternVL-4BCategory=General MLLM, Intensity=50%2026.0649.6—
Robust-R1Category=Robust MLLM, Intensity=100%2026.0649.53—
Qwen2.5-VL-3BCategory=General MLLM, Intensity=100%2026.0648.66—
InternVL-4BCategory=General MLLM, Intensity=100%2026.0646.93—
Gemma3-4BCategory=General MLLM, Intensity=clean2026.0643.93—
Gemma3-4BCategory=General MLLM, Intensity=25%2026.0643.2—
Gemma3-4BCategory=General MLLM, Intensity=50%2026.0642.6—
Gemma3-4BCategory=General MLLM, Intensity=100%2026.0641.33—
Robust CLIPCategory=Robust MLLM, Intensity=clean2026.0633—
Robust CLIPCategory=Robust MLLM, Intensity=25%2026.0632.26—
Robust CLIPCategory=Robust MLLM, Intensity=50%2026.0631.8—
TeCoACategory=Robust MLLM, Intensity=50%2026.0630.73—
TeCoACategory=Robust MLLM, Intensity=25%2026.0630.6—
TeCoACategory=Robust MLLM, Intensity=clean2026.0630.46—
Robust CLIPCategory=Robust MLLM, Intensity=100%2026.0629.46—
TeCoACategory=Robust MLLM, Intensity=100%2026.0628.06—