Shulou(Shulou.com)06/01 Report--

This article will explain in detail what the in-depth practice of Milvus is, and the content of the article is of high quality, so the editor will share it with you for reference. I hope you will have a certain understanding of the relevant knowledge after reading this article.

As an open source distributed vector search engine, Milvus has been integrated into various industry solutions by many technology companies in different industries because of its excellent product design and engineering implementation.

Milvus vector search engine can dock deep learning models including image recognition, video processing, voice recognition, natural language processing and other deep learning models, and provide search and analysis services for vectorized unstructured data.

So, how do users use Milvus?

First of all, users will convert the unstructured data into feature vectors through the deep learning model and import them into the Milvus library, and Milvus will store the feature vectors and build an index. In retrieval, the unstructured data to be searched is first transformed into feature vectors, and then searched in Milvus. Milvus will return the search results, and then find the corresponding unstructured data through the results of feature vectors.

The architecture of Milvus search engine is shown in the figure.

"one person, one file" based on face recognition technology is a common application scene in portrait management at present. By clustering and archiving the data set collected by the road bayonet camera, the portrait file is established and each file is tagged. The first is face clustering, which refers to classifying the face photos collected by the front-end camera, clustering the photos of the same person into one category, and then establishing each person's personal profile. " "one person, one file" is widely used in smart cities, intelligent security and many other fields. However, the number of face images collected by each card in the city is as high as tens of millions of people every day. In the huge data set, it is not easy to cluster faces quickly and accurately.

In order to complete face clustering more efficiently, Yuncong combined with Milvus vector search engine realizes large-scale real-time face clustering, and builds one person and one file based on dynamic data. In one of the product research and development scenarios, tens of millions of face images are first extracted as feature vectors, and then the transformed tens of millions of vectors are imported into the Milvus table. Then through the Milvus batch search function, the given batch (N) face image vectors are searched in the imported base database, and the similarity between the base database and these hundreds of face images is Top-K, and the result set is Numberk faces, that is, the preliminary clustering of the N face images in the base database is completed.

Thanks to the acceleration of Milvus vector search engine, in tens of millions of image libraries, thousands of face images can be clustered in seconds, with an average clustering time of tens of milliseconds per face image, while the recall rate is maintained at more than 95%.

The implementation process of face clustering in this scene:

The high performance and high recall rate of Milvus effectively help the construction of "face clustering, one person, one file" system, which can ensure the timely updating of file system data and the freshness of data. Thus, we can make full use of the collected data and mine the value of the data.

This is the end of the in-depth practice of Milvus. I hope the above content can be helpful to everyone and learn more knowledge. If you think the article is good, you can share it for more people to see.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.