Offline Video Understanding on Video-MME (Subtitle Dependency)
67.6Accuracy (with subtitle)Qwen2.5-VL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5-VLModel Scale=3B, Implementation Source=Original2025.12 | 67.6 | 61.5 | |
| MMDuet2Training Stage=RL, Implementation Source=Our implementation2025.12 | 67.5 | 58.1 | |
| MMDuet2Training Stage=SFT, Implementation Source=Our implementation2025.12 | 67.1 | 57.7 | |
| Qwen2.5-VLModel Scale=3B, Implementation Source=Our implementation2025.12 | 66.5 | 57.3 |