Video-to-text on VideoHallucer (test)
1Hallucination RateTIGER
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TIGERBackbone=Qwen2.5-Omni-7B2026.05 | 1 | 26 | 52.6 | 19.8 | |
| FrozenBackbone=Qwen2.5-Omni-7B2026.05 | 1.5 | 21 | 49.5 | 17 | |
| BoN+VideoCLIPBackbone=Qwen2.5-Omni-7B2026.05 | 1.5 | 21.9 | 50.1 | 17.5 | |
| VolcanoBackbone=Qwen2.5-Omni-7B2026.05 | 1.5 | 25 | 49 | 17.5 | |
| TCDBackbone=Qwen2.5-Omni-7B2026.05 | 1.6 | 22 | 50.5 | 18 | |
| BoN+CycRewBackbone=Qwen2.5-Omni-7B2026.05 | 1.7 | 22.7 | 50.8 | 18.9 |