Proactive Video Understanding on StreamPro-Bench
7.5PU TimeQwen2.5-VL-7B
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen2.5-VL-7BParams=7B, Proactive Prompt=true2026.05 | 7.5 | 7.7 | 3.4 | 2.4 | 0.4 | 0.1 | 6.5 | 2.2 | 0.9 | 1.5 | 1.6 | |
| VideoLLM-OnlineParams=8B2026.05 | 17 | 1.6 | 0.4 | 0.5 | 0.1 | 0 | 21.8 | 7 | 5.7 | 2 | 0.6 | |
| StreamoParams=3B2026.05 | 20.3 | 17.2 | 7.8 | 27.2 | 19.1 | 14 | 6 | 2.6 | 2.7 | 8.2 | 10.4 | |
| MiniCPM-o-4.5Params=9B2026.05 | 20.6 | 16.9 | 13.5 | 5.2 | 0.8 | 0.6 | 6.6 | 4.9 | 3.8 | 6 | 6.4 | |
| MMDuet2Params=3B2026.05 | 37.2 | 28.5 | 9.8 | 47.2 | 30.1 | 3.1 | 22.3 | 7.4 | 2.7 | 5.2 | 5.9 | |
| MMDuetParams=7B2026.05 | 47.4 | 33.1 | 9.3 | 71.8 | 20.3 | 1.5 | 34.8 | 15.7 | 4 | 4.9 | 5 | |
| StreamPro-SFTParams=3B, Backbone=Qwen2.5-VL-3B, Sampling FPS=1, Sliding Window Size=2002026.05 | 47.7 | 40.4 | 13.3 | 44.7 | 26.1 | 21.1 | 41 | 17.8 | 3.2 | 12.5 | 16.3 | |
| StreamPro-GRPOParams=3B, Backbone=Qwen2.5-VL-3B, Sampling FPS=1, Sliding Window Size=2002026.05 | 48.1 | 44.6 | 27.3 | 59.7 | 47.3 | 32.9 | 13.7 | 7.8 | 4.2 | 21.5 | 28.1 | |
| Qwen3-VL-8BParams=8B, Proactive Prompt=true2026.05 | 60.1 | 51.4 | 6.6 | 27.6 | 16.6 | 0.8 | 23 | 12.2 | 2.6 | 3.3 | 3.4 | |
| StreamPro-GRPOParams=4B, Backbone=Qwen3-VL-4B, Sampling FPS=1, Sliding Window Size=2002026.05 | 66 | 61.6 | 45 | 67.5 | 60.4 | 44.4 | 41.2 | 52.7 | 7.6 | 32.3 | 41.5 | |
| StreamPro-SFTParams=4B, Backbone=Qwen3-VL-4B, Sampling FPS=1, Sliding Window Size=2002026.05 | 66.6 | 61.7 | 24.7 | 45.1 | 32.5 | 27.9 | 46.1 | 20.5 | 5.7 | 19.4 | 24.7 |