Facial re-identification sample

  • Description
    Facial re-identification ensures consistent recognition of individuals across varied scenarios, enhancing security and personalized services. It's pivotal for public safety and user-specific applications.
  • Key Features
    — YOLOV8-Face facial detection model with landmarks
    — NumPy/Numba output postprocessing (box and landmarks)
    — Facial tracking (with Nvidia Tracker)
    — Landmark-based facial alignment with OpenCV-CUDA
    — ReID generation with AdaFace
    — Facial index implemented with HNSWLIB
  • End-To-End Performance
    — 124 FPS on Nvidia A4000
    — 26 FPS on Jetson NX