Object Hallucination Detection on AMBER out-of-distribution (OOD)
0.8611AUCOverthinking Score
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Overthinking ScoreVLM=LLaVA-1.5, Classifier Type=GB2026.03 | 0.8611 | 0.3665 | 0.7158 | |
| Overthinking ScoreVLM=LLaVA-1.5, Classifier Type=MLP2026.03 | 0.8328 | 0.3091 | 0.5976 | |
| MetaTokenVLM=LLaVA-1.5, Classifier Type=GB2026.03 | 0.8215 | 0.2348 | 0.6554 | |
| SVARVLM=LLaVA-1.52026.03 | 0.7951 | 0.1278 | 0.5687 | |
| MetaTokenVLM=LLaVA-1.5, Classifier Type=LR2026.03 | 0.795 | 0.1155 | 0.5476 | |
| MetaTokenVLM=LLaVA-1.5, Classifier Type=MLP2026.03 | 0.7941 | 0.1404 | 0.5854 | |
| Overthinking ScoreVLM=LLaVA-1.5, Classifier Type=LR2026.03 | 0.7076 | 0.0797 | 0.5392 | |
| HalLocVLM=LLaVA-1.52026.03 | 0.5 | 0.0227 | 0.4944 |