RetinaFace: Single-stage Dense Face Localisation in the Wild

TL;DR

RetinaFace is a single-stage dense face detector that improves face localization accuracy in the wild by leveraging multi-task learning with extra facial landmark annotations and 3D shape prediction, achieving state-of-the-art results.

Contribution

The paper introduces a robust face detection method that combines extra facial landmark supervision and self-supervised 3D shape prediction for improved accuracy and efficiency.

Findings

Outperforms state-of-the-art on WIDER FACE hard set with 91.4% AP.

Enhances face verification results on IJB-C dataset.

Runs in real-time on a CPU with lightweight backbones.

Abstract

Though tremendous strides have been made in uncontrolled face detection, accurate and efficient face localisation in the wild remains an open challenge. This paper presents a robust single-stage face detector, named RetinaFace, which performs pixel-wise face localisation on various scales of faces by taking advantages of joint extra-supervised and self-supervised multi-task learning. Specifically, We make contributions in the following five aspects: (1) We manually annotate five facial landmarks on the WIDER FACE dataset and observe significant improvement in hard face detection with the assistance of this extra supervision signal. (2) We further add a self-supervised mesh decoder branch for predicting a pixel-wise 3D shape face information in parallel with the existing supervised branches. (3) On the WIDER FACE hard test set, RetinaFace outperforms the state of the art average precision (AP) by 1.1% (achieving AP equal to 91.4%). (4) On the IJB-C test set, RetinaFace enables state of the art methods (ArcFace) to improve their results in face verification (TAR=89.59% for FAR=1e-6). (5) By employing light-weight backbone networks, RetinaFace can run real-time on a single CPU core for a VGA-resolution image. Extra annotations and code have been made available at: https://github.com/deepinsight/insightface/tree/master/RetinaFace.