Extracting Event Highlights Using iYUNDONG Sports App

What this post added

This post details iYUNDONG's implementation of an intelligent image retrieval system using Milvus to power their sports app. It highlights the challenges of handling massive datasets, achieving instant search availability, and efficient feature extraction. Milvus was chosen for its ability to perform fast vector retrieval on ultra-large datasets and mass storage at lower costs, enabling millisecond-level queries on over 60 million images with an accuracy rate above 92%. The system architecture involves converting facial features into 128-dimensional vectors and storing them in Milvus, with additional metadata stored in a third-party database, linked via the Leaf algorithm for distributed ID generation.