Temporal Hallucination Detection on QVHighlights (test)
80AUROCMTLA
Evaluation Results
| Method | Links | |
|---|---|---|
| MTLABackbone=Qwen3-VL-8B2026.07 | 80 | |
| InternalConfBackbone=Qwen3-VL-8B2026.07 | 59.5 | |
| last_logpBackbone=Qwen3-VL-8B2026.07 | 58.1 | |
| SVARBackbone=Qwen3-VL-8B2026.07 | 41.5 |