视觉世界搜索:持久视觉记忆、分层索引与源证据
原标题:Search over the Visual World: Persistent Visual Memory, Layered Indexes, and Source-Grounded Evidence
AI 摘要
本文提出了一种面向持续视觉数据流的搜索基础设施模型,包括分析器定义的场景、持久理解工件、视觉记忆和分层索引。作者开发了VideoDB数据格式(VDB)并在生产中实现,通过类型化搜索界面支持规划检索、状态调查、直接访问和基于证据的综合。在四个公共数据集上的9800多个查询对比中,通用组件流水线在Recall@1/@3/@10上优于商业视频原生引擎,表明视觉世界检索质量更多取决于系统设计而非视频专用预训练。
正文节选
Computer Science > Information Retrieval Title:Search over the Visual World: Persistent Visual Memory, Layered Indexes, and Source-Grounded Evidence View PDF HTML (experimental) Abstract:Most video-retrieval systems assume a bounded corpus and return ranked files or timestamps. Agents operating over cameras, screens, streams, and archives face a different systems problem: observations arrive continuously; models interpret them at different temporal granularities; context must be sele