Object Hallucination Detection on MSCOCO Average performance across VLMs (test)
87.33AUCOverthinking Score
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Overthinking ScoreClassifier=MLP2026.03 | 87.33 | 58.12 | 72.86 | |
| Overthinking ScoreClassifier=GB2026.03 | 87.3 | 61.54 | 75.97 | |
| MetaTokenClassifier=MLP2026.03 | 83.83 | 45.45 | 67.32 | |
| MetaTokenClassifier=GB2026.03 | 83.46 | 44.61 | 72.51 | |
| HalLoc2026.03 | 81.17 | 54.36 | 71.85 | |
| Overthinking ScoreClassifier=LR2026.03 | 80.46 | 42.56 | 65.44 | |
| MetaTokenClassifier=LR2026.03 | 79.6 | 36.32 | 56.14 | |
| SVAR2026.03 | 78.26 | 39.71 | 55.8 |