Real-time Visual Perception and Backward Tracing on OVO-Bench
93.2Real-Time ScoreHuman
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Human#Frames=-2026.01 | 93.2 | 92.33 | 92.83 | |
| Qwen2.5-VL-32B + HERMES (4K tokens)#Frames=1 fps2026.01 | 72.37 | 55.42 | 63.9 | |
| Qwen2.5-VL-32B + HERMES (6K tokens)#Frames=1 fps2026.01 | 71.93 | 57.71 | 64.82 | |
| Gemini 1.5 pro#Frames=1 fps2026.01 | 69.32 | 62.54 | 66.41 | |
| Qwen2.5-VL-7B + HERMES (4K tokens)#Frames=1 fps2026.01 | 68.98 | 49.43 | 59.21 | |
| Qwen2.5-VL-7B + HERMES (6K tokens)#Frames=1 fps2026.01 | 68.42 | 48.1 | 58.26 | |
| LLaVA-OV-7B + HERMES (4K tokens)#Frames=0.5 fps2026.01 | 66.34 | 50.2 | 58.27 | |
| LLaVA-OV-7B + HERMES (6K tokens)#Frames=0.5 fps2026.01 | 65.07 | 48.8 | 56.94 | |
| GPT-4o#Frames=642026.01 | 64.46 | 60.75 | 62.87 | |
| Qwen2.5-VL-32B#Frames=1 fps2026.01 | 64.4 | 50.33 | 57.37 | |
| LLaVA-OV-7B#Frames=642026.01 | 63.06 | 43.64 | 53.35 | |
| TimeChat-Online-7B#Frames=1 fps2026.01 | 61.9 | 41.7 | 51.8 | |
| StreamForest-7B#Frames=1 fps2026.01 | 61.2 | 52.02 | 56.61 | |
| InternVL-V2-8B#Frames=162026.01 | 60.73 | 44 | 52.37 | |
| Qwen2-VL-7B#Frames=642026.01 | 60.65 | 48.58 | 54.62 | |
| Qwen2.5-VL-7B#Frames=1 fps2026.01 | 59.9 | 44.65 | 52.28 | |
| LLaVA-OV-7B + ReKV#Frames=0.5 fps2026.01 | 57.33 | 44.16 | 50.75 | |
| Dispider-7B#Frames=1 fps2026.01 | 54.55 | 36.06 | 45.31 | |
| LLaVA-OV-0.5B + HERMES (4K tokens)#Frames=0.5 fps2026.01 | 50.72 | 34.8 | 42.76 | |
| LLaVA-OV-0.5B + HERMES (6K tokens)#Frames=0.5 fps2026.01 | 50.34 | 34.75 | 42.55 | |
| LLaVA-OV-0.5B#Frames=642026.01 | 49.7 | 34.59 | 42.15 | |
| LLaVA-OV-0.5B + ReKV#Frames=0.5 fps2026.01 | 43.77 | 33.06 | 38.42 | |
| Flash-VStream-7B#Frames=-2026.01 | 29.86 | 25.35 | 27.61 | |
| VideoLLM-online-8B#Frames=2 fps2026.01 | 20.79 | 17.73 | 19.26 |