Ecole d'ingénieur et centre de recherche en Sciences du numérique

Apparent age estimation from face images combining general and children-specialized deep learning models

Antipov, Grigory; Baccouche, Moez; Berrani, Sid-Ahmed; Dugelay, Jean-Luc

CVPRW 2016, 29th IEEE Conference on Computer Vision and Pattern Recognition Workshops, June 26th-July 1st, 2016, Las Vegas, USA

Best Paper Award and 1st Place Award

This work describes our solution in the second edition of the ChaLearn LAP competition on Apparent Age Estimation. Starting from a pretrained version of the VGG-16 convolutional neural network for face recognition, we train it on the huge IMDB-Wiki dataset for biological age estimation and then fine-tune it for apparent age estimation using the relatively small competition dataset. We show that the precise age estimation of children is the cornerstone of the competition. Therefore, we integrate a separate "children" VGG-16 network for apparent age estimation of children between 0 and 12 years old in our final solution. The "children" network is fine-tuned from the "general" one. We employ different age encoding strategies for training "general" and "children" networks: the soft one (label distribution encoding) for the "general" network and the strict one (0/1 classification encoding) for the "children" network. Finally, we highlight the importance of the state-of-the-art face detection and face alignment for the final apparent age estimation. Our resulting solution wins the 1st place in the competition significantly outperforming the runner-up.

Document Doi Hal Bibtex

Titre:Apparent age estimation from face images combining general and children-specialized deep learning models
Type:Conférence
Langue:English
Ville:Las Vegas
Pays:ÉTATS-UNIS
Date:
Département:Sécurité numérique
Eurecom ref:4908
Copyright: © 2016 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.
Bibtex: @inproceedings{EURECOM+4908, doi = {http://dx.doi.org/10.1109/CVPRW.2016.105}, year = {2016}, title = {{A}pparent age estimation from face images combining general and children-specialized deep learning models}, author = {{A}ntipov, {G}rigory and {B}accouche, {M}oez and {B}errani, {S}id-{A}hmed and {D}ugelay, {J}ean-{L}uc}, booktitle = {{CVPRW} 2016, 29th {IEEE} {C}onference on {C}omputer {V}ision and {P}attern {R}ecognition {W}orkshops, {J}une 26th-{J}uly 1st, 2016, {L}as {V}egas, {USA}}, address = {{L}as {V}egas, {\'{E}}{TATS}-{UNIS}}, month = {06}, url = {http://www.eurecom.fr/publication/4908} }
Voir aussi: