TimeLens2-27B is a video temporal grounding model based on Qwen/Qwen3.8-27B. Given a video and a text query, it predicts the relevant time intervals in seconds.
Paper · Code & Usage · Collection
| Benchmark | mIoU (%) |
|---|---|
| Charades-STA | 61.17 |
| ActivityNet Captions | 63.75 |
| QVHighlights | 71.10 |
| VUE-TR | 56.34 |
| VUE-TR-v2 | 53.74 |
| MomentSeeker | 31.77 |
| Ego4D-NLQ-v2 | 23.04 |
| Average | 51.56 |
The two VUE benchmarks use the available-video subsets. The average is the unweighted mean over the seven benchmarks.
For inference and evaluation, follow the official TimeLens2 repository and use ZhuYuhan/TimeLens2-27B as the model path.
model-00005-of-00013.safetensors 4.8 GBmodel-00012-of-00013.safetensors 4.8 GBmodel-00011-of-00013.safetensors 4.8 GBmodel-00004-of-00013.safetensors 4.7 GBmodel-00008-of-00013.safetensors 4.7 GBmodel-00003-of-00013.safetensors 4.7 GBmodel-00007-of-00013.safetensors 4.7 GBmodel-00010-of-00013.safetensors 4.7 GBmodel-00006-of-00013.safetensors 4.7 GBmodel-00002-of-00013.safetensors 4.7 GBmodel-00009-of-00013.safetensors 4.6 GBmodel-00001-of-00013.safetensors 2.5 GBmodel-00013-of-00013.safetensors 99 MBQwen/Qwen3.8-27B