RetinaFace: Single-Shot Multi-Level Face Localisation in the Wild

By Jiankang Deng, Jia Guo, Evangelos Ververas, Irene Kotsia, Stefanos P. Zafeiriou, InsightFace FaceSoft. Computer Vision and Pattern Recognition, 2020

442 citations recorded by the source corpus.

Though tremendous strides have been made in uncontrolled face detection, accurate and efficient 2D face alignment and 3D face reconstruction in-the-wild remain an open challenge. In this paper, we present a novel single-shot, multi-level face localisation method, named RetinaFace, which unifies face box prediction, 2D facial landmark localisation and 3D vertices regression under one common target: point regression on the image plane. To fill the data gap, we manually annotated five facial landmarks on the WIDER FACE dataset and employed a semi-automatic annotation pipeline to generate 3D vertices for face images from the WIDER FACE, AFLW and FDDB datasets. Based on extra annotations, we propose a mutually beneficial regression target for 3D face reconstruction, that is predicting 3D vertices projected on the image plane constrained by a common 3D topology. The proposed 3D face reconstruction branch can be easily incorporated, without any optimisation difficulty, in parallel with the existing box and 2D landmark regression branches during joint training. Extensive experimental results show that RetinaFace can simultaneously achieve stable face detection, accurate 2D face alignment and robust 3D face reconstruction while being efficient through single-shot inference.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/8e534cb1d574de5cec99f3831c373c2de18b6f80

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

M2TR: Multi-modal Multi-scale Transformers for Deepfake Detection (2021, methodology): “Implementation Details We use RetinaFace [13] to crop the face regions (detected boxes enlarged 1.3 × ) as inputs with a size of 320 × 320.”
Frequency-aware Discriminative Feature Learning Supervised by Single-Center Loss for Face Forgery Detection (2021, methodology): “Besides, we utilize retinaface[11] to detect faces in each frame.”
EMOCA: Emotion Driven Monocular Face Capture and Animation (2022, methodology): “Model-free approaches directly regress 3D meshes [18,20,28,33,40,70,74,79,93,95,97] or voxels [38] from an image, or optimize a Signed Distance Function (SDF) [61] to fit a face image.”
SphereFace Revived: Unifying Hyperspherical Face Recognition (2021, methodology): “Each face image is cropped based on the five face landmarks (i.e., left eye, right eye, nasal tip, left mouth corner, and right mouth corner) detected by MTCNN [73] and RetinaFace [74] using similarity transformation.”
Robust Lightweight Facial Expression Recognition Network with Label Distribution Training (2021, methodology): “For images on all the datasets, the face region is detected and aligned using Retinaface (Deng et al. 2020) and then cropped and resized to 224 (cid:2) 224 pixels.”
General Facial Representation Learning in a Visual-Linguistic Manner (2021, methodology): “First, as discussed above, our model presents bias to a certain degree, which might be caused by: 1) the data bias existing in the original LAION dataset [82], or 2) the performance bias of the face detector [20] we use.”
EasyCom: An Augmented Reality Dataset to Support Algorithms for Easy Communication in Noisy Environments (2021, methodology): “We use RetinaFace [12] to detect all faces from the videos and try to match the detected faces to the ID of the glasses worn by the participants.”
Ultra-High-Definition Low-Light Image Enhancement: A Benchmark and Transformer-Based Method (2022, methodology): “…Guo 2019)), two unsupervised CNN-based methods (EL-GAN (Jiang et al. 2021), RUAS (Liu et al. 2021b)), two zero-shot learning-based methods (Z_DCE (Guo et al. 2020), Z_DCE++ (Li, Guo, and Loy 2021) ) and two supervised transformer-based methods (Uformer (Wang et al. 2022b), Restormer (Zamir et…”