Skip to main content

Table 5 Face identification and verification evaluation of different methods on MegaFace Challenge1 using FaceScrub as the probe set

From: Deep-learned faces: a survey

Publication

Id (%)

Ver (%)

FaceNet v8

70.49

-

NTechLAB

73.30

-

SphereFace

72.729

85.561

CosFace (single-patch)

72.729

96.65

CosFace (3-patch ensemble)

74.11

97.96

ArcFace

77.06

96.98

ArcFace R

98.35

98.48

  1. Id rank 1 face identification accuracy with 1M distractors, Ver face verification TAR at 10-6 FAR, R data refinement on both probe set and 1M distractors