The latest articles on DEV Community by Erwin Wilson Ceniza2 (@erwin_wilsonceniza2_adf9). https://dev.to/erwin_wilsonceniza2_adf9 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4005978%2Fc91c89c4-17e1-4452-a555-bfe9a4d50382.jpg https://dev.to/erwin_wilsonceniza2_adf9 en Erwin Wilson Ceniza2 Sun, 28 Jun 2026 13:20:03 +0000 https://dev.to/erwin_wilsonceniza2_adf9/engineering-a-cross-platform-face-recognition-pipeline-with-anti-spoofing-2j81 https://dev.to/erwin_wilsonceniza2_adf9/engineering-a-cross-platform-face-recognition-pipeline-with-anti-spoofing-2j81 <p>Check out my article on this blog spot, it talks about building a face recognition system that actually works in production — not a demo, not a toy, something you can put on an Android tablet mounted on a warehouse wall and walk away.</p> <p>It covers the full pipeline:</p> <ul> <li>Finding the face with a lightweight RFB-320 model (1.27 MB, runs on CPU)</li> <li>Anti-spoofing to stop print and replay attacks (0.1 threshold, 13.9 MB model)</li> <li>FaceNet-style 128-dim embeddings with L2 normalization</li> <li>HNSW indexing for sub-millisecond matching at 10,000+ enrollees</li> <li>Dynamic gap-based threshold adjustment that cuts false accepts by ~30%</li> <li>Thread-safe ONNX inference with three models running sequentially</li> <li>Offline RSA-licensed deployment for factories, mines, and remote sites</li> <li>Real issues we hit: channel order bugs, semaphore starvation, per-device liveness drift, cold start latency</li> </ul> <p><strong>Read the full article here:</strong><br><br> 👉 <a href="https://erwinwilsonceniza.qzz.io/blogs/engineering-a-cross-platform-face-recognition-pipeline-with-anti-spoofing" rel="noopener noreferrer">Engineering a Cross-Platform Face Recognition Pipeline with Anti-Spoofing</a></p> computervision machinelearning security systemdesign