Decoupling Semantics and Logic: A Training-Free Coarse-to-Fine Pipeline for Video Retrieval-Augmented Generation
解耦语义与逻辑:一种无需训练的从粗到精的视频检索增强生成流水线
机构 * School of Computer Science & Tech, Huazhong University of Science and Technology(华中科技大学计算机科学与技术学院) ; School of AI and Automation, Huazhong University of Science and Technology(华中科技大学人工智能与自动化学院)
专题命中 多模态RAG :RAG(summary_cn,abstract);retrieval-augmented generation(title);dense retrieval(abstract);分类 cs.CL、cs.AI
AI总结 提出一种无需训练的两阶段级联视频RAG流水线,通过解耦语义检索与逻辑推理,实现跨语言长视频理解、严格角色遵循和零幻觉时间定位。
Comments To be presented at ACL 2026 MAGMAR Workshop (Oral; Retrieval leaderboard No.1)