One Ranking, Any Budget: Matryoshka Evidence-to-Context Frame Selection for Long-Video Understanding

作者：Wang Chen, Yu Chen, Xiang Wang, Shuai Li, Jinfa Huang, Xiawu Zheng · 单位：Media Analytics and Computing Lab, Xiamen University, Xiamen, China, Department of Computer Science, University of Rochester, Rochester, NY, USA · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-08-07