{"title":"Multi-task multi-scale attention learning-based facial age estimation","authors":"Chaojun Shi, Shiwei Zhao, Ke Zhang, Xiaohan Feng","doi":"10.1049/sil2.12190","DOIUrl":null,"url":null,"abstract":"<p>Face-based age estimation strongly depends on deep residual networks (ResNets), used as the backbone in the relevant research. However, ResNet-based methods ignore the importance of some large-scale facial information and other facial age attributes. Inspired by the attention mechanism, a multi-task learning framework for face-based age estimation called multi-task multi-scale attention is proposed. First, the authors embed the alternative strategy structure of dilated convolution into ResNet34 to construct a multi-scale attention module (MSA) to improve the network's receptive field, which extracts local age-sensitive information while obtaining multi-scale features. The MSA can have a larger receptive field to extract both large-scale and local detailed feature information. Second, multi-task learning network structures are built to predict gender and race, which can share rigid network parameters to improve age estimation and improve the accuracy of age estimation by other age-related parameters. Finally, the Kullback-Leibler divergence loss is adopted between a Dirac delta label and a Gaussian prediction to guide the training. The numerical tests on the MORPH Album II and Adience datasets prove the superiority of the proposed method over other state-of-the-art ones.</p>","PeriodicalId":56301,"journal":{"name":"IET Signal Processing","volume":null,"pages":null},"PeriodicalIF":1.1000,"publicationDate":"2023-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1049/sil2.12190","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IET Signal Processing","FirstCategoryId":"5","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1049/sil2.12190","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q4","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 0
Abstract
Face-based age estimation strongly depends on deep residual networks (ResNets), used as the backbone in the relevant research. However, ResNet-based methods ignore the importance of some large-scale facial information and other facial age attributes. Inspired by the attention mechanism, a multi-task learning framework for face-based age estimation called multi-task multi-scale attention is proposed. First, the authors embed the alternative strategy structure of dilated convolution into ResNet34 to construct a multi-scale attention module (MSA) to improve the network's receptive field, which extracts local age-sensitive information while obtaining multi-scale features. The MSA can have a larger receptive field to extract both large-scale and local detailed feature information. Second, multi-task learning network structures are built to predict gender and race, which can share rigid network parameters to improve age estimation and improve the accuracy of age estimation by other age-related parameters. Finally, the Kullback-Leibler divergence loss is adopted between a Dirac delta label and a Gaussian prediction to guide the training. The numerical tests on the MORPH Album II and Adience datasets prove the superiority of the proposed method over other state-of-the-art ones.
基于人脸的年龄估计在很大程度上依赖于深度残差网络(ResNets),在相关研究中用作骨干。然而,基于ResNet的方法忽略了一些大规模面部信息和其他面部年龄属性的重要性。受注意力机制的启发,提出了一种基于人脸年龄估计的多任务学习框架,称为多任务多尺度注意力。首先,作者将扩张卷积的替代策略结构嵌入到ResNet34中,以构建多尺度注意力模块(MSA)来改善网络的感受野,该模块在获得多尺度特征的同时提取局部年龄敏感信息。MSA可以具有更大的感受野来提取大规模和局部的详细特征信息。其次,构建多任务学习网络结构来预测性别和种族,可以共享刚性网络参数来改进年龄估计,并通过其他年龄相关参数来提高年龄估计的准确性。最后,在Dirac delta标签和高斯预测之间采用Kullback-Leibler发散损失来指导训练。在MORPH Album II和Adience数据集上的数值测试证明了所提出的方法优于其他最先进的方法。
期刊介绍:
IET Signal Processing publishes research on a diverse range of signal processing and machine learning topics, covering a variety of applications, disciplines, modalities, and techniques in detection, estimation, inference, and classification problems. The research published includes advances in algorithm design for the analysis of single and high-multi-dimensional data, sparsity, linear and non-linear systems, recursive and non-recursive digital filters and multi-rate filter banks, as well a range of topics that span from sensor array processing, deep convolutional neural network based approaches to the application of chaos theory, and far more.
Topics covered by scope include, but are not limited to:
advances in single and multi-dimensional filter design and implementation
linear and nonlinear, fixed and adaptive digital filters and multirate filter banks
statistical signal processing techniques and analysis
classical, parametric and higher order spectral analysis
signal transformation and compression techniques, including time-frequency analysis
system modelling and adaptive identification techniques
machine learning based approaches to signal processing
Bayesian methods for signal processing, including Monte-Carlo Markov-chain and particle filtering techniques
theory and application of blind and semi-blind signal separation techniques
signal processing techniques for analysis, enhancement, coding, synthesis and recognition of speech signals
direction-finding and beamforming techniques for audio and electromagnetic signals
analysis techniques for biomedical signals
baseband signal processing techniques for transmission and reception of communication signals
signal processing techniques for data hiding and audio watermarking
sparse signal processing and compressive sensing
Special Issue Call for Papers:
Intelligent Deep Fuzzy Model for Signal Processing - https://digital-library.theiet.org/files/IET_SPR_CFP_IDFMSP.pdf