基于CLIP模型的军事领域图片资源多模态搜索工具研究  

Research on multimodal search tool for domain image resources based on CLIP

作　　者：赵晋巍 刘晓鹏 罗威 程瑾 毛彬 宋宇 ZHAO Jin-wei;LIU Xiao-peng;LUO Wei;CHENG Jin;MAO Bin;SONG Yu(Information Research Center of Military Sciences,Academy of Military Sciences,Beijing 100142,China)

机构地区：[1]军事科学院军事科学信息研究中心,北京100142

出　　处：《中华医学图书情报杂志》2022年第8期14-20,共7页Chinese Journal of Medical Library and Information Science

摘　　要：深入调研梳理了OpenAI CLIP多模态模型和相关应用研究。借助CLIP模型,通过大规模军事相关图片数据集开展实验,设计开发了具有以文搜图和以图搜图功能的多模态搜索工具,且在实际测试中表现良好,可以为后续围绕军事相关图片的事件分类、目标检测、任务轨迹跟踪等方面的分析研究奠定基础。The Open AI′ multimodal model Contrastive Language-image Pre-training(CLIP) and relevant application research were thoroughly analyzed and sorted out. With the help of CLIP, a multimodal search tool with the functions of text search and image search was designed and developed through experiments with military-related image datasets on a large scale. The tool performed well in practical tests, which can lay a foundation for further analyses and research on event classification, target detection, task trajectory tracking of military-related images.

关 键 词：CLIP模型 多模态 图文检索 以图搜图 

分 类 号：G254.37[文化科学—图书馆学] G254.92