Rust Face Landmarks Demo

A real-time 478-point face landmark detection demo written in Rust, running on the ONNX Runtime platform.

Requirements

• ONNX Runtime 1.20 (DirectML)

pip install onnxruntime-directml==1.20.0
python copy_runtime.py

Inference Models

• Face Detection: YoloV5 Face — detects face bounding boxes
• Face Landmarks: Face Landmarker from Google MediaPipe — generates 478 landmarks
  • TFLite → ONNX conversion via tensorflow-onnx

Profiling

• Tracy Profiler