Human Preference Evaluation on Long Videos 50 60s
30Win CountStream-R1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Stream-R1Dimension=Dynamic Reasonableness, Basline=Reward Forcing2026.05 | 30 | 3 | 17 | 63 | |
| Stream-R1Dimension=Visual Quality & Aesthetics, Basline=Reward Forcing2026.05 | 29 | 2 | 19 | 60 | |
| Stream-R1Dimension=Overall Preference, Basline=Reward Forcing2026.05 | 28 | 1 | 21 | 57 | |
| Stream-R1Dimension=Temporal Consistency, Basline=Reward Forcing2026.05 | 25 | 1 | 24 | 51 | |
| Stream-R1Dimension=Text-Video Alignment, Basline=Reward Forcing2026.05 | 22 | 9 | 18 | 54.1 |