暂无图片
暂无图片
暂无图片
暂无图片
暂无图片
【松鼠会】论文分享:《基于向量量化的最大内积搜索》- 戴新颜(香港中文大学博士)
792
30页
8次
2020-12-10
免费下载
Xinyan DAI
xinyan.dai@outlook.com
DBSys Group@SUSTech
Husky Data Lab@CUHK
Quantization Based Maximum
Inner Product Search (MIPS)
Maximum Inner Product Search (MIPS)
Given a dataset containing 𝑛 items and , for a
query , find the item that satisfies
Usually require the top-k items with the largest inner product
Approximate results suffice in most cases and quality measured by recall
is the ground-truth top-k MIPS items and is the estimated top-k.
Recall-item curve or recall-time curve as the main performance metric
2
Application: Image Matching & Search
Database images or image patches represented by vector descriptors, e.g.,
SIFT, GIST, or descriptors generated by neural networks
Use MIPS to find images similar to a query image
3
Query Image
Image Database
Image search
of 30
免费下载
【版权声明】本文为墨天轮用户原创内容,转载时必须标注文档的来源(墨天轮),文档链接,文档作者等基本信息,否则作者和墨天轮有权追究责任。如果您发现墨天轮中有涉嫌抄袭或者侵权的内容,欢迎发送邮件至:contact@modb.pro进行举报,并提供相关证据,一经查实,墨天轮将立刻删除相关内容。