milvus-logo
Star
0
Forks
0
快速开始

Image Similarity Search

Milvus Docs 需要你的帮助

本文档暂时没有中文版本,欢迎你成为社区贡献者,协助中文技术文档的翻译。
你可以通过页面右边的 编辑 按钮直接贡献你的翻译。更多详情,参考 贡献指南。如需帮助,你可以 提交 GitHub Issue

This tutorial demonstrates how to use Milvus, the open-source vector database, to build a reverse image search system.

The ML model and third-party software used include:

  • YOLOv3
  • ResNet-50
  • MySQL

Major search engines like Google already give users the option to search by image. Additionally, e-commerce platforms have realized the benefits this functionality offers online shoppers, with Amazon incorporating image search into its smartphone applications.


In this tutorial, you will learn how to build a reverse image search system that can detect image patterns and return similar images to the one you upload. To build such an image similarity search system, download PASCAL VOC image set which contains 17125 images of 20 categories. Alternatively, you can prepare your own image datasets. Use YOLOv3 for object detection and ResNet-50 for image feature extraction. After going through the two ML models, images are converted into 256-dimensional vectors. Then store the vectors in Milvus and a unique ID for each vector is automatically generated by Milvus. MySQL is then used to map the vector IDs to the images in the dataset. Whenever you upload a new image to the image search system, it will be converted into a new vector and compared against the vectors previously stored in Milvus. Milvus then returns the IDs of the most similar vectors and you can query the corresponding images in MySQL.


image_search
Workflow of a reverse image search system.

image_search_demo
Demo of a reverse image search system.

该页面是否对你有帮助?
本页目录