Computer vision & AI
The core of VanityFilter is a real-time face-aware imaging pipeline that decides, per pixel and per frame, what is skin and what must be protected.
- MediaPipe Face Landmarker 478-point facial landmarks via the modern Tasks API, tracking up to 5 faces per frame
- YuNet (ONNX) Fallback face detector that catches turned and profile faces MediaPipe misses
- Selfie segmentation (TFLite) Multiclass person segmentation used as the skin gate for the mask
- OpenCV Edge-preserving bilateral filtering, mask morphology, and all frame I/O
- NumPy Vectorized per-pixel math across every frame of the video