Zhentu - the Photo Fraud Detector Based on Milvus

What this post added

This post details the implementation of Zhentu, a photo fraud detection system, using Milvus. It describes the system architecture involving image pre-processing, feature vector extraction using deep learning, normalization, and vector search with Milvus. The deployment section outlines using Milvus cluster on Kubernetes with Helm for high availability and real-time synchronization. It also provides specific configurations for index building (IVF_PQ with nlist=2048) and search parameters (IP metric_type with nprobe=32), highlighting the performance benefits of using Inner Product over Euclidean distance for normalized vectors. The post reports real-world performance metrics, including detecting over 20,000 fake licenses and achieving sub-second query times for tens of millions of vectors.