BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs

TL;DR

BlazeFace is a highly efficient, lightweight face detection model optimized for mobile GPUs, achieving super-realtime speeds suitable for augmented reality applications and various facial analysis tasks.

Contribution

It introduces a novel lightweight feature extraction network, a GPU-friendly anchor scheme, and an improved tie resolution strategy for face detection on mobile devices.

Findings

Achieves 200-1000+ FPS on flagship mobile devices.

Enables real-time facial analysis in augmented reality pipelines.

Outperforms existing lightweight face detectors in speed and accuracy.

Abstract

We present BlazeFace, a lightweight and well-performing face detector tailored for mobile GPU inference. It runs at a speed of 200-1000+ FPS on flagship devices. This super-realtime performance enables it to be applied to any augmented reality pipeline that requires an accurate facial region of interest as an input for task-specific models, such as 2D/3D facial keypoint or geometry estimation, facial features or expression classification, and face region segmentation. Our contributions include a lightweight feature extraction network inspired by, but distinct from MobileNetV1/V2, a GPU-friendly anchor scheme modified from Single Shot MultiBox Detector (SSD), and an improved tie resolution strategy alternative to non-maximum suppression.