With the development of convolutional neural network, significant progress
has been made in computer vision tasks. However, the commonly used loss
function softmax loss and highly efficient network architecture for common
visual tasks are not as effective for face recognition. In this paper, we
propose a novel loss function named Li-ArcFace based on ArcFace. Li-ArcFace
takes the value of the angle through linear function as the target logit rather
than through cosine function, which has better convergence and performance on
low dimensional embedding feature learning for face recognition. In terms of
network architecture, we improved the the perfomance of MobileFaceNet by
increasing the network depth, width and adding attention module. Besides, we
found some useful training tricks for face recognition. With all the above
results, we won the second place in the deepglint-light challenge of LFR2019.
Description
AirFace: Lightweight and Efficient Model for Face Recognition
%0 Generic
%1 li2019airface
%A Li, Xianyang
%A Wang, Feng
%A Hu, Qinghao
%A Leng, Cong
%D 2019
%K computer-vision conv-neural-net regularization
%T AirFace: Lightweight and Efficient Model for Face Recognition
%U http://arxiv.org/abs/1907.12256
%X With the development of convolutional neural network, significant progress
has been made in computer vision tasks. However, the commonly used loss
function softmax loss and highly efficient network architecture for common
visual tasks are not as effective for face recognition. In this paper, we
propose a novel loss function named Li-ArcFace based on ArcFace. Li-ArcFace
takes the value of the angle through linear function as the target logit rather
than through cosine function, which has better convergence and performance on
low dimensional embedding feature learning for face recognition. In terms of
network architecture, we improved the the perfomance of MobileFaceNet by
increasing the network depth, width and adding attention module. Besides, we
found some useful training tricks for face recognition. With all the above
results, we won the second place in the deepglint-light challenge of LFR2019.
@misc{li2019airface,
abstract = {With the development of convolutional neural network, significant progress
has been made in computer vision tasks. However, the commonly used loss
function softmax loss and highly efficient network architecture for common
visual tasks are not as effective for face recognition. In this paper, we
propose a novel loss function named Li-ArcFace based on ArcFace. Li-ArcFace
takes the value of the angle through linear function as the target logit rather
than through cosine function, which has better convergence and performance on
low dimensional embedding feature learning for face recognition. In terms of
network architecture, we improved the the perfomance of MobileFaceNet by
increasing the network depth, width and adding attention module. Besides, we
found some useful training tricks for face recognition. With all the above
results, we won the second place in the deepglint-light challenge of LFR2019.},
added-at = {2022-05-28T08:24:26.000+0200},
author = {Li, Xianyang and Wang, Feng and Hu, Qinghao and Leng, Cong},
biburl = {https://www.bibsonomy.org/bibtex/23eca8cf913cf13fd7f92a8fde874cc02/adnanahmed},
description = {AirFace: Lightweight and Efficient Model for Face Recognition},
interhash = {4cd0a05340cbe18c900698001f0a2a35},
intrahash = {3eca8cf913cf13fd7f92a8fde874cc02},
keywords = {computer-vision conv-neural-net regularization},
note = {cite arxiv:1907.12256},
timestamp = {2022-05-31T23:05:01.000+0200},
title = {AirFace: Lightweight and Efficient Model for Face Recognition},
url = {http://arxiv.org/abs/1907.12256},
year = 2019
}