This letter proposes a gradient-enhanced softmax supervisor for face recognition (FR) based on a deep convolutional neural network (DCNN). The proposed supervisor conducts the constant-normalized cosine to obtain the score for each class using a combination of the intra-class score and the soft maximum of the inter-class scores as the objective function. This mitigates the vanishing gradient problem in the conventional softmax classifier. The experiments on the public Labeled Faces in the Wild (LFW) database denote that the proposed supervisor achieves better results when compared with those achieved using the current state-of-the-art softmax-based approaches for FR.
Linjun SUN
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Weijun LI
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Xin NING
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Liping ZHANG
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Xiaoli DONG
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Wei HE
Cognitive Computing Technology Joint Laboratory
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Linjun SUN, Weijun LI, Xin NING, Liping ZHANG, Xiaoli DONG, Wei HE, "Gradient-Enhanced Softmax for Face Recognition" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 5, pp. 1185-1189, May 2020, doi: 10.1587/transinf.2019EDL8103.
Abstract: This letter proposes a gradient-enhanced softmax supervisor for face recognition (FR) based on a deep convolutional neural network (DCNN). The proposed supervisor conducts the constant-normalized cosine to obtain the score for each class using a combination of the intra-class score and the soft maximum of the inter-class scores as the objective function. This mitigates the vanishing gradient problem in the conventional softmax classifier. The experiments on the public Labeled Faces in the Wild (LFW) database denote that the proposed supervisor achieves better results when compared with those achieved using the current state-of-the-art softmax-based approaches for FR.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDL8103/_p
Copy
@ARTICLE{e103-d_5_1185,
author={Linjun SUN, Weijun LI, Xin NING, Liping ZHANG, Xiaoli DONG, Wei HE, },
journal={IEICE TRANSACTIONS on Information},
title={Gradient-Enhanced Softmax for Face Recognition},
year={2020},
volume={E103-D},
number={5},
pages={1185-1189},
abstract={This letter proposes a gradient-enhanced softmax supervisor for face recognition (FR) based on a deep convolutional neural network (DCNN). The proposed supervisor conducts the constant-normalized cosine to obtain the score for each class using a combination of the intra-class score and the soft maximum of the inter-class scores as the objective function. This mitigates the vanishing gradient problem in the conventional softmax classifier. The experiments on the public Labeled Faces in the Wild (LFW) database denote that the proposed supervisor achieves better results when compared with those achieved using the current state-of-the-art softmax-based approaches for FR.},
keywords={},
doi={10.1587/transinf.2019EDL8103},
ISSN={1745-1361},
month={May},}
Copy
TY - JOUR
TI - Gradient-Enhanced Softmax for Face Recognition
T2 - IEICE TRANSACTIONS on Information
SP - 1185
EP - 1189
AU - Linjun SUN
AU - Weijun LI
AU - Xin NING
AU - Liping ZHANG
AU - Xiaoli DONG
AU - Wei HE
PY - 2020
DO - 10.1587/transinf.2019EDL8103
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 5
JA - IEICE TRANSACTIONS on Information
Y1 - May 2020
AB - This letter proposes a gradient-enhanced softmax supervisor for face recognition (FR) based on a deep convolutional neural network (DCNN). The proposed supervisor conducts the constant-normalized cosine to obtain the score for each class using a combination of the intra-class score and the soft maximum of the inter-class scores as the objective function. This mitigates the vanishing gradient problem in the conventional softmax classifier. The experiments on the public Labeled Faces in the Wild (LFW) database denote that the proposed supervisor achieves better results when compared with those achieved using the current state-of-the-art softmax-based approaches for FR.
ER -