计算机科学 ›› 2022, Vol. 49 ›› Issue (6): 245-253.doi: 10.11896/jsjkx.210400023
程祥鸣, 邓春华
CHENG Xiang-ming, DENG Chun-hua
摘要: 将人脸识别技术移植到移动设备上时,往往需要经过模型压缩等加速算法的处理。知识蒸馏是一种实际应用较广且易于训练的模型压缩方法,现有的知识蒸馏算法需要大量带标签的人脸数据,可能会涉及身份隐私泄露等安全问题。同时,大规模采集有标签人脸数据的成本较大,而海量可采集或生成的无标签人脸数据却无法利用。为解决上述问题,通过分析知识蒸馏在人脸识别任务中的特性,提出了一种无标签知识蒸馏的间接监督训练方法。该方法可以利用海量无标签的人脸数据,避免了隐私泄露等安全隐患问题。然而,无标签人脸数据集的数据分布无法预知,存在数据分布不均衡的问题,阻碍了间接监督算法的性能提升。文中进一步提出了一种人脸内容置换的数据增强方法,通过置换人脸部分内容来平衡人脸数据分布,同时增强了人脸数据的多样性。实验结果表明,人脸识别模型被大幅度压缩时,所提算法的性能达到了先进水平,并在LFW数据集上超越了大型网络。
中图分类号:
[1] SUN C,SHRIVASTAVA A,SINGH S,et al.Revisiting unreasonable effectiveness of data in deep learning era[C]//Procee-dings of the IEEE International Conference on Computer Vision.2017:843-852. [2] IANDOLA F N,HAN S,MOSKEWICZ M W,et al.Squeeze-Net:AlexNet-level accuracy with 50x fewer parameters and <0.5 MB model size[J].arXiv:1602.07360,2016. [3] HOWARD A G,ZHU M,CHEN B,et al.Mobilenets:Efficient convolutional neural networks for mobile vision applications[J].arXiv:1704.04861,2017. [4] ZHANG X,ZHOU X,LIN M,et al.Shufflenet:An extremelyefficient convolutional neural network for mobile devices[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:6848-6856. [5] HINTON G,VINYALS O,DEAN J.Distilling the knowledge ina neural network[J].arXiv:1503.02531,2015. [6] WANG R Z,GAO J,HUANG S H,et al.Malicious Code Family Detection Method Based on Knowledge Distillation[J].Compu-ter Science,2021,48(1):280-286. [7] LECUN Y,BENGIO Y,HINTON G.Deep learning[J].Nature,2015,521(7553):436-444. [8] GUO Y,ZHANG L,HU Y,et al.Ms-celeb-1m:A dataset and benchmark for large-scale face recognition[C]//European Conference on Computer Vision.2016:87-102. [9] DATTAS,SHARMA G,JAWAHAR C V.Unsupervised lear-ning of face representations[C]//2018 13th IEEE International Conference on Automatic Face & Gesture Recognition(FG 2018).IEEE,2018:135-142. [10] YU H,FAN Y,CHEN K,et al.Unknownidentity rejection loss:Utilizing unlabeled data for face recognition[C]//2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW).IEEE Computer Society,2019:2662-2669. [11] HUANG C,LOY C C,TANG X.Unsupervised learning of discriminative attributes and visual representations[C]//Procee-dings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:5175-5184. [12] SCHMARJE L,SANTAROSSA M,SCHRÖDER S M,et al.A survey on semi-,self-and unsupervised techniques in image classification[J].arXiv:2002.08721,2020. [13] MASI I,WU Y,HASSNER T,et al.Deep face recognition:Asurvey[C]//2018 31st SIBGRAPI Conference on Graphics,Patterns and Images (SIBGRAPI).IEEE,2018:471-478. [14] JI C M,SONG T C.Sparse Representation-Based Classification Under Optimization Forms for Face Recognition[ J].Journal of Chongqing University of Technology(Natural Science),2020,34(2):120-126. [15] KE P F,CAI M G,WU T.Face Recognition Algorithm Based on Improved Convolutional Neural Network and Ensemble Lear-ning[J].Computer Engineering,2020,46(2):262-267,273. [16] TAO S F,LI Y F,HUANG Y F,et al.Face Detection Algo-rithm Based on Deep Residual Network and Attention Mechanism[J].Computer Engineering,2021,47(11):276-282. [17] PARKHI O M,VEDALDI A,ZISSERMAN A.Deepface recognition[C]//Proceedings of the British Machine Vision Confe-rence.2015:41.1-41.12. [18] NOROOZI M,VINJIMOOR A,FAVARO P,et al.Boostingself-supervised learning via knowledge transfer[C]//Procee-dings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:9359-9367. [19] VONGKULBHISAL J,VINAYAVEKHIN P,VISENTINI-SCARZANELLA M.Unifying heterogeneous classifiers with distillation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2019:3175-3184. [20] LUO P,ZHU Z,LIU Z,et al.Face model compression by distilling knowledge from neurons[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2016:3560-3566. [21] HU J,SHEN L,SUN G.Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:7132-7141. [22] DENG J,GUO J,XUE N,et al.Arcface:Additive angular margin loss for deep face recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2019:4690-4699. [23] WANG H,WANG Y,ZHOU Z,et al.Cosface:Large margin cosine loss for deep face recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018:5265-5274. [24] DEVRIES T,TAYLOR G W.Improved regularization of convolutional neural networks with cutout[J].arXiv:1708.04552,2017. [25] YUN S,HAN D,OH S J,et al.Cutmix:Regularization strategy to train strong classifiers with localizable features[C]//Procee-dings of the IEEE International Conference on Computer Vision.2019:6023-6032. [26] ZHANG K,ZHANG Z,LI Z,et al.Joint face detection andalignment using multitask cascaded convolutional networks[J].IEEE Signal Processing Letters,2016,23(10):1499-1503. [27] WANG F,XIANG X,CHENG J,et al.Normface:L2 hyper-sphere embedding for face verification[C]//Proceedings of the 25th ACM International Conference on Multimedia.2017:1041-1049. [28] MIRZADEH S I,FARAJTABAR M,LI A,et al.Improvedknowledge distillation via teacher assistant[C]//Proceedings of the AAAI Conference on Artificial Intelligence.2020:5191-5198. |
[1] | 黄璞, 杜旭然, 沈阳阳, 杨章静. 基于局部正则二次线性重构表示的人脸识别 Face Recognition Based on Locality Regularized Double Linear Reconstruction Representation 计算机科学, 2022, 49(6A): 407-411. https://doi.org/10.11896/jsjkx.210700018 |
[2] | 黄璞, 沈阳阳, 杜旭然, 杨章静. 基于局部约束特征线表示的人脸识别 Face Recognition Based on Locality Constrained Feature Line Representation 计算机科学, 2022, 49(6A): 429-433. https://doi.org/10.11896/jsjkx.210300169 |
[3] | 楚玉春, 龚航, 王学芳, 刘培顺. 基于YOLOv4的目标检测知识蒸馏算法研究 Study on Knowledge Distillation of Target Detection Algorithm Based on YOLOv4 计算机科学, 2022, 49(6A): 337-344. https://doi.org/10.11896/jsjkx.210600204 |
[4] | 魏勤, 李瑛娇, 娄平, 严俊伟, 胡辑伟. 基于边云协同的人脸识别方法研究 Face Recognition Method Based on Edge-Cloud Collaboration 计算机科学, 2022, 49(5): 71-77. https://doi.org/10.11896/jsjkx.210300222 |
[5] | 何嘉玉, 黄宏博, 张红艳, 孙牧野, 刘亚辉, 周哲海. 基于深度学习的单幅图像三维人脸重建研究综述 Review of 3D Face Reconstruction Based on Single Image 计算机科学, 2022, 49(2): 40-50. https://doi.org/10.11896/jsjkx.210500215 |
[6] | 陈长伟, 周晓峰. 快速局部协同表示分类器及其在人脸识别中的应用 Fast Local Collaborative Representation Based Classifier and Its Applications in Face Recognition 计算机科学, 2021, 48(9): 208-215. https://doi.org/10.11896/jsjkx.200800155 |
[7] | 陈志文, 王坤, 周广蕴, 王旭, 张晓丹, 朱虎明. 基于胶囊网络及其权重剪枝的SAR图像变化检测方法 SAR Image Change Detection Method Based on Capsule Network with Weight Pruning 计算机科学, 2021, 48(7): 190-198. https://doi.org/10.11896/jsjkx.200800225 |
[8] | 温荷, 罗频捷. 基于改进脉冲耦合神经网络的动态人脸识别 Dynamic Face Recognition Based on Improved Pulse Coupled Neural Network 计算机科学, 2021, 48(6A): 85-88. https://doi.org/10.11896/jsjkx.200600172 |
[9] | 白子轶, 毛懿荣, 王瑞平. 视频人脸识别进展综述 Survey on Video-based Face Recognition 计算机科学, 2021, 48(3): 50-59. https://doi.org/10.11896/jsjkx.210100210 |
[10] | 陆要要, 袁家斌, 何珊, 王天星. 基于超分辨率重建的低质量视频人脸识别方法 Low-quality Video Face Recognition Method Based on Super-resolution Reconstruction 计算机科学, 2021, 48(11A): 295-302. https://doi.org/10.11896/jsjkx.201200159 |
[11] | 杨章静, 王文博, 黄璞, 张凡龙, 王昕. 基于局部加权表示的线性回归分类器及人脸识别 Local Weighted Representation Based Linear Regression Classifier and Face Recognition 计算机科学, 2021, 48(11A): 351-359. https://doi.org/10.11896/jsjkx.210100173 |
[12] | 栾晓, 李晓双. 基于多特征融合的人脸活体检测算法 Face Anti-spoofing Algorithm Based on Multi-feature Fusion 计算机科学, 2021, 48(11A): 409-415. https://doi.org/10.11896/jsjkx.210100181 |
[13] | 俞亮, 魏永丰, 罗国亮, 邬昌兴. 基于知识蒸馏的隐式篇章关系识别 Knowledge Distillation Based Implicit Discourse Relation Recognition 计算机科学, 2021, 48(11): 319-326. https://doi.org/10.11896/jsjkx.201000099 |
[14] | 王润正, 高见, 黄淑华, 仝鑫. 基于知识蒸馏的恶意代码家族检测方法 Malicious Code Family Detection Method Based on Knowledge Distillation 计算机科学, 2021, 48(1): 280-286. https://doi.org/10.11896/jsjkx.200900099 |
[15] | 吴庆洪, 高晓东. 稀疏表示和支持向量机相融合的非理想环境人脸识别 Face Recognition in Non-ideal Environment Based on Sparse Representation and Support Vector Machine 计算机科学, 2020, 47(6): 121-125. https://doi.org/10.11896/jsjkx.190500058 |
|