Egocentric 4D Reasoning on HD-EPIC
91.1Average Multiple-Choice AccuracySample Human Baseline
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Sample Human Baseline2026.03 | 91.1 | 89.5 | 95.6 | 96.3 | 87.6 | 93.2 | 84.3 | |
| EgoReasoner Stage 2 - RFTParameters=3B, Stage=RFT2026.03 | 37.5 | 32.7 | 30.6 | 50.4 | 59.5 | 30.5 | 21 | |
| EgoReasoner Stage 1 - SFTParameters=3B, Stage=SFT2026.03 | 32.2 | 29.3 | 29.2 | 42 | 44 | 28 | 20.5 | |
| LongVA2026.03 | 28.1 | 32.3 | 26.6 | 41.2 | 34.5 | 10.2 | 23.5 | |
| Qwen-2.5-VL-3BParameters=3B2026.03 | 27 | 31 | 18.4 | 37.8 | 39 | 16.2 | 19.5 | |
| Qwen-2.5-VL-7BParameters=7B2026.03 | 25.7 | 22.7 | 22.6 | 34 | 33 | 20.2 | 22 | |
| VideoLlama 22026.03 | 25.5 | 17.7 | 18.8 | 31 | 44 | 11 | 30.5 | |
| LLaVA-Video2026.03 | 20.9 | 16.3 | 21.8 | 30.6 | 20 | 9.8 | 27 |