Hugging Face Papers ↗ · 2026-11-16
VideoNexus:超越推理上下文的视频感知
Media reports Hugging Face Papers Source language: Chinese
Paper abstract (author-reported, not re-tested by this site): Understanding long videos requires locating evidence, resolving visual details, and integrating observations into an answer.
Full text
VideoNexus separates detailed video perception from evidence-driven planning and reasoning, using perception sub-agents to return compact temporally grounded evidence and achieving higher accuracy at lower cost on long-video benchmarks.