Hallucination Detection on VQA-Med 2019 (All)
78.1AUCVIHD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VIHDBackbone=Hulu-4B2026.05 | 78.1 | 60.39 | |
| VASE + V-LoopBackbone=InternVL3-8B [20]2026.01 | 76.1 | 61.76 | |
| RadFlag + V-LoopBackbone=InternVL3-8B [20]2026.01 | 75.92 | 60.79 | |
| VIHDBackbone=Lingshu-7B2026.05 | 74.18 | 64.54 | |
| SE + V-LoopBackbone=InternVL3-8B [20]2026.01 | 74.14 | 57.67 | |
| VASEBackbone=Hulu-4B2026.05 | 72.99 | 57.41 | |
| VASEBackbone=InternVL3-8B [20]2026.01 | 72.07 | 54.18 | |
| V-LoopBackbone=InternVL3-8B2026.01 | 70.55 | 55.82 | |
| RadFlagBackbone=InternVL3-8B, Uncertainty-based=true2026.01 | 70.25 | 55.28 | |
| RadFlagBackbone=InternVL3-8B [20]2026.01 | 70.25 | 55.28 | |
| MaxEntBackbone=Hulu-4B2026.05 | 69.64 | 55.45 | |
| VASE + V-LoopBackbone=MedGemma-4B-it [34]2026.01 | 69.56 | 54.9 | |
| MaxProbBackbone=Hulu-4B2026.05 | 69.45 | 55.39 | |
| VASEBackbone=Lingshu-7B2026.05 | 68.85 | 60.8 | |
| VASE + V-LoopBackbone=Lingshu-7B [40]2026.01 | 68.8 | 68.43 | |
| VASEBackbone=MedGemma-4B-it [34]2026.01 | 68.19 | 44.71 | |
| AvgProbBackbone=Hulu-4B2026.05 | 67.9 | 55 | |
| AvgEntBackbone=Hulu-4B2026.05 | 67.72 | 55 | |
| RadFlag + V-LoopBackbone=Lingshu-7B [40]2026.01 | 67.23 | 65.22 | |
| VASEBackbone=Lingshu-7B [40]2026.01 | 67.09 | 57.33 | |
| RadFlag + V-LoopBackbone=MedGemma-4B-it [34]2026.01 | 67.02 | 53.91 | |
| SEBackbone=InternVL3-8B, Uncertainty-based=true2026.01 | 66.35 | 46.71 | |
| SEBackbone=InternVL3-8B [20]2026.01 | 66.35 | 46.71 | |
| SE + V-LoopBackbone=MedGemma-4B-it [34]2026.01 | 66.07 | 53.77 | |
| SE + V-LoopBackbone=Lingshu-7B [40]2026.01 | 66.01 | 65.62 | |
| V-LoopBackbone=Lingshu-7B2026.01 | 65.66 | 65.03 | |
| V-LoopBackbone=MedGemma-4B-it2026.01 | 63.28 | 52.15 | |
| RadFlagBackbone=MedGemma-4B-it, Uncertainty-based=true2026.01 | 62.95 | 50.14 | |
| RadFlagBackbone=MedGemma-4B-it [34]2026.01 | 62.95 | 50.14 | |
| MaxEntBackbone=InternVL3-8B2026.01 | 62.58 | 54.87 | |
| SEBackbone=MedGemma-4B-it, Uncertainty-based=true2026.01 | 60.91 | 48.78 | |
| SEBackbone=MedGemma-4B-it [34]2026.01 | 60.91 | 48.78 | |
| AvgEntBackbone=InternVL3-8B2026.01 | 60.17 | 47.9 | |
| MaxEntBackbone=MedGemma-4B-it2026.01 | 56.42 | 34.84 | |
| RadFlagBackbone=Lingshu-7B2026.05 | 56.04 | 37.85 | |
| RadFlagBackbone=Lingshu-7B, Uncertainty-based=true2026.01 | 55.91 | 42.77 | |
| RadFlagBackbone=Lingshu-7B [40]2026.01 | 55.91 | 42.77 | |
| AvgEntBackbone=MedGemma-4B-it2026.01 | 55.42 | 35.4 | |
| MaxEntBackbone=Lingshu-7B2026.05 | 55.15 | 36.82 | |
| MaxProbBackbone=Lingshu-7B2026.05 | 55.12 | 36.78 | |
| AvgProbBackbone=Lingshu-7B2026.05 | 55.11 | 36.81 | |
| AvgEntBackbone=Lingshu-7B2026.05 | 55.11 | 36.9 | |
| MaxEntBackbone=Lingshu-7B2026.01 | 54.4 | 41.3 | |
| AvgEntBackbone=Lingshu-7B2026.01 | 54.36 | 41.31 | |
| SEBackbone=Lingshu-7B2026.05 | 53.96 | 35.96 | |
| SEBackbone=Lingshu-7B, Uncertainty-based=true2026.01 | 51.9 | 38.72 | |
| SEBackbone=Lingshu-7B [40]2026.01 | 51.9 | 38.72 | |
| RadFlagBackbone=Hulu-4B2026.05 | 50 | 22.84 | |
| SEBackbone=Hulu-4B2026.05 | 47.42 | 21.43 | |
| AvgProbBackbone=Lingshu-7B2026.01 | 45.64 | 48.61 | |
| MaxProbBackbone=Lingshu-7B2026.01 | 45.6 | 47.82 | |
| AvgProbBackbone=MedGemma-4B-it2026.01 | 44.41 | 38.33 | |
| MaxProbBackbone=MedGemma-4B-it2026.01 | 43.64 | 35.69 | |
| AvgProbBackbone=InternVL3-8B2026.01 | 39.74 | 37.45 | |
| MaxProbBackbone=InternVL3-8B2026.01 | 38.09 | 32.09 |