Belief Prediction on EgoToM (test)
33.8True AccuracyQwen2.5-VL-7B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen2.5-VL-7BIntervention Setting=+αΔ2026.03 | 33.8 | 90.6 | 27.9 | |
| LLaVA-Next-Video-7BIntervention Setting=+αΔ2026.03 | 32.9 | 99.8 | 30.8 | |
| Qwen2.5-VL-7BIntervention Setting=Baseline2026.03 | 29 | 92.5 | 23.7 | |
| Gemini-2.5-FlashIntervention Setting=Baseline2026.03 | 28.1 | 99.7 | 28.1 | |
| LLaVA-Next-Video-7BIntervention Setting=Baseline2026.03 | 19.5 | 99.7 | 19.2 |