Video Captioning on Dream-1K re-annotated
45.4RecallCapRL++
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CapRL++Backbone=Qwen3-VL-4B2026.06 | 45.4 | 82.3 | 58.5 | |
| Qwen3-VL-235B-A22BVariant=Instruct2026.06 | 39.5 | 89.5 | 54.8 | |
| Qwen3-VL-8BVariant=Instruct2026.06 | 35.8 | 88.9 | 51.1 | |
| Qwen3-VL-4BType=baseline2026.06 | 33.4 | 89.4 | 48.6 |