Long-video Understanding on HourVideo 1.0 (test)
37.3Overall ScoreGemini 1.5 Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 1.5 ProFrames=0.5fps2025.08 | 37.3 | |
| Video-EM (Qwen2.5-VL)LLM Size=7B, Frames=Avg 302025.08 | 35.1 | |
| Video-EM (Qwen2-VL)LLM Size=7B, Frames=Avg 302025.08 | 34.8 | |
| VAMBAVenue=CVPR’25, LLM Size=10B, Frames=1282025.08 | 33.6 | |
| Qwen2.5-VL‡Venue=ArXiv’25, LLM Size=7B, Frames=642025.08 | 32.8 | |
| Qwen2-VL‡Venue=ArXiv’24, LLM Size=7B, Frames=642025.08 | 32.4 | |
| Qwen2.5-VL‡Venue=ArXiv’25, LLM Size=7B, Frames=322025.08 | 31.1 | |
| VideoLLaMA3Venue=ArXiv’25, LLM Size=7B, Frames=322025.08 | 31 | |
| LongVUVenue=ICML’25, LLM Size=7B, Frames=1fps2025.08 | 30.8 | |
| Qwen2-VL‡Venue=ArXiv’24, LLM Size=7B, Frames=322025.08 | 30.7 | |
| InternVL3Venue=ArXiv’25, LLM Size=7B, Frames=322025.08 | 30.1 | |
| GPT-4Frames=0.5fps2025.08 | 25.7 |