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 video similarity search system.
The ML models and third-party software used include:
Nowadays, after watching a movie or video they like, people can easily take screenshots and share their thoughts by posting on various social networking platforms. When the followers see the screenshots, it can be really difficult for them to tell which movie it is if the movie name is not spelled out explicitly in the post. In order to figure out the name of the movie, people can take advantage of a video similarity search system. By using the system, users can upload an image and get videos or movies that contain key frames similar to the uploaded image.
In this tutorial, you will learn how to build a video similarity search system. This tutorial uses approximately 100 animated gifs on Tumblr to build the system. However, you can also prepare your own video datasets. The system first uses OpenCV to extract key frames in videos and then obtains feature vectors of each key frame using ResNet-50. All vectors are stored and searched in Milvus, which will return the IDs of similar vectors. Then map the IDs to the corresponding video stored in MySQL.