ArcFace: Additive Angular Margin Loss for Deep Face Recognition
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 func-
tions that enhance discriminative power. Centre loss pe-
nalises the distance between the deep features and their cor-
responding class centres in the Euclidean space to achieve
intra-class compactness. SphereFace assumes that the lin-
ear transformation matrix in the last fully connected layer
can be used as a representation of the class centres in an
angular space and penalises the angles between the deep
features and their corresponding weights in a multiplicative
way.
数字图像处理(冈萨雷斯)
数字图像处The diameter, x, of the retinal image corresponding to the dot is obtained from similar
triangles, as shown in Fig. P2.1. That is,
(d=2)
0:2
=
(x=2)
0:014
which gives x = 0:07d. From the discussion in Section 2.1.1, and taking some liberties
of interpretation, we can think of the fovea as a square sensor array having on the order of
337,000 elements, which translates into an array of size 580£ 580 elements. Assuming
equal spacing between elements, this gives 580 elements and 579 spaces on a line 1.5
mm long. The size of each element and each space is then s = [(1:5mm)=1; 159] =
1:3£10¡6 m. If the size (on the fovea) of the imaged dot is less than the size of a single
resolution element, we assume that the dot will be invisible to the eye. In other words,
the eye will not detect a dot if its diameter, d, is such that 0:07(d) < 1:3 £ 10¡6 m, or
d < 18:6 £ 10¡6 m.
Figure理