🎬 TimeLens2-2B — Video Temporal Grounding

Upload a video and describe an event. The model returns the time spans (in seconds) where that event occurs.

Model: MCG-NJU/TimeLens2-2B — a compact 2B Qwen3-VL fine-tune that achieves SOTA at the 2B scale on seven temporal-grounding benchmarks (average mIoU 44.5), even surpassing TimeLens-8B.