ArcFace: Additive Angular Margin Loss for Deep Face Recognition

CVPR 2019 Jiankang DengJia GuoNiannan XueStefanos Zafeiriou

One of the main challenges in feature learning using Deep Convolutional Neural Networks (DCNNs) for large-scale face recognition is the design of appropriate loss functions that enhance discriminative power. Centre loss penalises the distance between the deep features and their corresponding class centres in the Euclidean space to achieve intra-class compactness... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
Face Verification Labeled Faces in the Wild ArcFace + MS1MV2 + R100, Accuracy 99.83% # 1
Face Verification MegaFace ArcFace + MS1MV2 + R100 + R Accuracy 98.48% # 1
Face Identification MegaFace ArcFace + MS1MV2 + R100 + R Accuracy 98.35% # 1
Face Verification YouTube Faces DB ArcFace + MS1MV2 + R100, Accuracy 98.02% # 2