基于多分辨率自蒸馏网络的小样本图像分类

doi:10.19678/j.issn.1000-3428.0064011

计算机工程 ›› 2022, Vol. 48 ›› Issue (12): 232-240. doi: 10.19678/j.issn.1000-3428.0064011

基于多分辨率自蒸馏网络的小样本图像分类

仇真^1,2,3, 奚雪峰^1,2,3, 崔志明^1,2,3, 盛胜利⁴, 胡伏原^1,2,3

1. 苏州科技大学电子与信息工程学院, 江苏苏州 215000;
2. 苏州市虚拟现实智能交互及应用重点实验室, 江苏苏州 215000;
3. 苏州智慧城市研究院, 江苏苏州 215000;
4. 德州理工大学计算机科学系, 美国拉伯克 79401

收稿日期:2022-02-23 修回日期:2022-05-02 发布日期:2022-05-24
作者简介:仇真（1998—），男，硕士研究生，主研方向为小样本学习、图像处理；奚雪峰（通信作者），副教授、博士；崔志明，教授、博士、博士生导师；盛胜利、胡伏原，教授、博士。
基金资助:
国家自然科学基金（61876217，61876121，62176175）；江苏省“六大人才高峰”高层次人才项目（XYDXX-086）；苏州市科技计划项目（SGC2021078）。

Few-Shot Image Classification Based on Multi-Resolution Self-Distillation Network

QIU Zhen^1,2,3, XI Xuefeng^1,2,3, CUI Zhiming^1,2,3, SHENG Shengli⁴, HU Fuyuan^1,2,3

1. School of Electronic and Information Engineering, Suzhou University of Science and Technology, Suzhou, Jiangsu 215000, China;
2. Suzhou Key Laboratory of Virtual Reality Intelligent Interaction and Application Technology, Suzhou, Jiangsu 215000, China;
3. Suzhou Smart City Research Institute, Suzhou, Jiangsu 215000, China;
4. Computer Science Department, Texas Tech University, Lubbock 79401, USA

Received:2022-02-23 Revised:2022-05-02 Published:2022-05-24

摘要/Abstract

摘要： 因图像数据具有大量的空间冗余信息，传统的多分辨率网络在处理图像数据时会产生较高的计算成本。而自蒸馏学习方法能够在精度与计算成本之间实现动态平衡，使模型在不增加网络深度和宽度的基础上，有效地提高模型精度。提出一种多分辨率自蒸馏网络（MRSDN），用于解决小样本学习中输入样本的空间冗余问题。从原始网络中分出一个浅层子网络以识别图像的低分辨率表示，并且保持该原始网络识别高分辨率图像特征的能力。同时，在多分辨率网络中引入改进的全局注意力机制，以减少信息损失且放大全局交互表示。利用自蒸馏学习方法将网络中更深层的知识压缩到浅层子网络中，以提升浅层子网络的泛化能力。在此基础上，将低分辨率网络中的粗粒度特征融合到高分辨率网络中，从而提高模型提取图像特征的能力。实验结果表明，在Mini-ImageNet数据集上MRSDN网络对5-way 1-shot与5-way 5-shot任务的准确率分别为56.34%和74.35%，在Tiered-ImageNet数据集上对5-way 1-shot与5-way 5-shot任务的准确率分别为59.56%和78.96%，能有效缓解高分辨率图像输入时的空间冗余问题，提高小样本图像分类的准确率。

关键词: 自蒸馏学习, 小样本学习, 多分辨率网络, 空间冗余, 全局注意力

Abstract: The traditional multi-resolution network incurs high computing costs when processing image data owing to a large amount of spatial redundancy information in the image data.The Self-Distillation(SD) learning method can achieve a dynamic balance between accuracy and calculation cost, effectively improving the accuracy of the model without increasing the depth and width of the network.A Multi-Resolution Self-Distillation Network(MRSDN) is proposed to solve the spatial redundancy of input samples in Few-Shot Learning(FSL).A shallow sub-network is separated from the original network to recognize the low-resolution representation of the image, and the ability of the original network to recognize the high-resolution image features is maintained.In addition, an improved Global Attention Mechanism(GAM) is introduced into the multi-resolution network to reduce information loss and enlarge the global interactive representation.The SD learning method is used to compress the in-depth knowledge of the network into a shallow sub-network to improve the generalization ability of the shallow sub-network.The coarse granularity features in the low-resolution network are fused into the high-resolution network to improve the ability of the model to extract image features.The experimental results show that accuracy of the MRSDN network for five-way one-shot and five-way five-shot tasks on the Mini-ImageNet dataset are 56.34% and 74.35%, respectively.The accuracy of the network for five-way one-shot and five-way five-shot tasks on the Tiered-ImageNet dataset are 59.56% and 78.96%, respectively.The proposed network can effectively alleviate the spatial redundancy when inputting high-resolution images, improving the accuracy of few-shot image classification.

Key words: Self-Distillation(SD), Few-Shot Learning(FSL), multi-resolution network, spatial redundancy, global attention

中图分类号:

TP391

仇真, 奚雪峰, 崔志明, 盛胜利, 胡伏原. 基于多分辨率自蒸馏网络的小样本图像分类[J]. 计算机工程, 2022, 48(12): 232-240.

QIU Zhen, XI Xuefeng, CUI Zhiming, SHENG Shengli, HU Fuyuan. Few-Shot Image Classification Based on Multi-Resolution Self-Distillation Network[J]. Computer Engineering, 2022, 48(12): 232-240.

http://www.ecice06.com/CN/Y2022/V48/I12/232

图/表 8

20230112184132

20230112184135

20230112184139

20230112184142

20230112184145

20230112184149

20230112184152

20230112184155

参考文献

[1] ZHANG L F, SONG J B, GAO A N, et al.Be your own teacher:improve the performance of convolutional neural networks via self-distillation[C]//Proceedings of International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2019:3712-3721.
[2] WANG Y Q, YAO Q M, KWOK J T, et al.Generalizing from a few examples:a survey on few-shot learning[J].ACM Computing Surveys, 2021, 53(3):63.
[3] WANG Q Y, LU Y, ZHANG X K, et al.Region of interest selection for functional features[J].Neurocomputing, 2021, 422:235-244.
[4] WOO S, PARK J, LEE J Y, et al.CBAM:convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision.Berlin, Germany:Springer, 2018:3-19.
[5] MUNKHDALAI T, YU H.Meta networks[J].Proceedings of Machine Learning Research, 2017, 70:2554-2563.
[6] SANTORO A, BARTUNOV S, BOTVINICK M, et al.Meta-learning with memory-augmented neural networks[C]//Proceedings of the 33rd International Conference on Machine Learning.New York, USA:ACM Press, 2016:1842-1850.
[7] RAVI S, LAROCHELLE H.Optimization as a model for few-shot learning[C]//Proceedings of International Conference on Learning Representations.Vancouver, Canda:[s.n.], 2016:1-9.
[8] PERRETT T, MASULLO A, BURGHARDT T, et al.Temporal-relational cross Transformers for few-shot action recognition[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2021:475-484.
[9] SNELL J, SWERSKY K, ZEMEL R.Prototypical networks for few-shot learning[C]//Proceedings of Conference on Advances in Neural Information Processing Systems.[S.l.]:AAAI Press, 2017:30.
[10] ZHANG S Y, ZHOU J L, HE X M.Learning implicit temporal alignment for few-shot video classification[EB/OL].[2022-01-15].http://arXivpreprintarXiv:2105.04823.
[11] FINN C, ABBEEL P, LEVINE S.Model-agnostic meta-learning for fast adaptation of deep networks[C]//Proceedings of the 34th International Conference on Machine Learning.New York, USA:ACM Press, 2017:1-10.
[12] BHAT P, ARANI E, ZONOOZ B.Distill on the Go:online knowledge distillation in self-supervised learning[C]//Proceedings of Conference on Computer Vision and Pattern Recognition Workshops.Washington D.C., USA:IEEE Press, 2021:2672-2681.
[13] KIM K, JI B, YOON D, et al.Self-knowledge distillation with progressive refinement of targets[C]//Proceedings of International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2021:6547-6556.
[14] KE T W, MAIRE M, YU S X.Multigrid neural architectures[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:4067-4075.
[15] LIN T Y, DOLLÁR P, GIRSHICK R, et al.Feature pyramid networks for object detection[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:936-944.
[16] ZHAO H S, QI X J, SHEN X Y, et al.ICNet for real-time semantic segmentation on high-resolution images[C]//Proceedings of the European Conference on Computer Vision.Berlin, Germany:Springer, 2018:418-434.
[17] SUN K, XIAO B, LIU D, et al.Deep high-resolution representation learning for human pose estimation[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:5686-5696.
[18] HOWARD A G, ZHU M L, CHEN B, et al.MobileNets:efficient convolutional neural networks for mobile vision applications[EB/OL].[2022-01-15].https://arxiv.org/pdf/1704.04861.pdf.
[19] CHEN Y P, FAN H Q, XU B, et al.Drop an octave:reducing spatial redundancy in convolutional neural networks with octave convolution[C]//Proceedings of International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2019:3434-3443.
[20] CHIN T W, DING R, MARCULESCU D.Adascale:towards real-time video object detection using adaptive scaling[J].Proceedings of Machine Learning and Systems, 2019, 1:431-441.
[21] DHILLON G S, CHAUDHARI P, RAVICHANDRAN A, et al.A baseline for few-shot image classification[EB/OL].[2022-01-15].https://arxiv.org/abs/1909.02729.
[22] ZHANG H, WU C R, ZHANG Z Y, et al.ResNeSt:split-attention networks[EB/OL].[2022-01-15].http://arXiv:2004.08955.
[23] ANIL R, PEREYRA G, PASSOS A, et al.Large scale distributed neural network training through online distillation[EB/OL].[2022-01-15].https://arxiv.org/pdf/1804.03235.pdf.
[24] BAIK S, HONG S, LEE K M.Learning to forget for meta-learning[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2020:2376-2384.
[25] LEE K, MAJI S, RAVICHANDRAN A, et al.Meta-learning with differentiable convex optimization[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:10649-10657.
[26] VINYALS O, BLUNDELL C, LILLICRAP T, et al.Matching networks for one shot learning[C]//Proceedings of the 30th International Conference on Neural Information Processing Systems.New York, USA:ACM Press, 2016:3637-3645.
[27] XUE Z Y, XIE Z S, XING Z, et al.Relative position and map networks in few-shot learning for image classification[C]//Proceedings of Conference on Computer Vision and Pattern Recognition Workshops.Washington D.C., USA:IEEE Press, 2020:4032-4036.
[28] SNELL J, SWERSKY K, ZEMEL R S.Prototypical networks for few-shot learning[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems.New York, USA:ACM Press, 2017:4080-4090.
[29] ALLEN K R, SHELHAMER E, SHIN H, et al.Infinite mixture prototypes for few-shot learning[EB/OL].[2022-01-15].https://arxiv.org/pdf/1902.04552v1.pdf.
[30] RAVICHANDRAN A, BHOTIKA R, SOATTO S.Few-shot learning with embedded class models and shot-free meta training[C]//Proceedings of International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2019:331-339.
[31] SUNG F, YANG Y X, ZHANG L, et al.Learning to compare:relation network for few-shot learning[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:1199-1208.
[32] LI W B, XU J L, HUO J, et al.Distribution consistency based covariance metric networks for few-shot learning[C]//Proceedings of the AAAI Conference on Artificial Intelligence.[S.l.]:AAAI Press, 2019:8642-8649.
[33] DONG C Q, LI W B, HUO J, et al.Learning task-aware local representations for few-shot learning[C]//Proceedings of the 28th International Joint Conference on Artificial Intelligence.Yokohama, Japan:[s.n.], 2020:716-722.
[34] LI W B, WANG L, XU J L, et al.Revisiting local descriptor based image-to-class measure for few-shot learning[C]//Proceedings of Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:7253-7260.
[35] LI W, WANG L, HUO J, et al.Asymmetric distribution measure for few-shot learning[EB/OL].[2022-01-15].http://arXivpreprintarXiv:2002.00153.

选择文件类型/文献管理软件名称

选择包含的内容

基于多分辨率自蒸馏网络的小样本图像分类

Few-Shot Image Classification Based on Multi-Resolution Self-Distillation Network

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献

相关文章 7

编辑推荐

Metrics

本文评价

[1]	刘昊鑫, 董超, 勾智楠, 高凯. 融合混合表征的小样本关系抽取方法[J]. 计算机工程, 2023, 49(8): 63-68.
[2]	刘金硕, 王代辰, 邓娟, 王丽娜. 基于长尾分类算法的网络不良信息分类[J]. 计算机工程, 2023, 49(8): 13-19, 28.
[3]	陈良臣, 傅德印. 面向小样本数据的机器学习方法研究综述[J]. 计算机工程, 2022, 48(11): 1-13.
[4]	王会勇, 卢超, 张晓明. 基于小样本学习和语义信息的图像描述模型[J]. 计算机工程, 2021, 47(8): 260-270.
[5]	汪荣贵, 汤明空, 杨娟, 薛丽霞, 胡敏. 语义匹配网络的小样本学习[J]. 计算机工程, 2021, 47(5): 244-250,259.
[6]	叶萌, 杨娟, 汪荣贵, 薛丽霞, 李懂. 基于特征聚合网络的小样本学习方法[J]. 计算机工程, 2021, 47(3): 77-82.
[7]	潘洁珠, 周晓, 吴共庆, 胡学钢. 基于小样本学习的垃圾邮件过滤方法[J]. 计算机工程, 2010, 36(21): 245-247.

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

基于多分辨率自蒸馏网络的小样本图像分类

Few-Shot Image Classification Based on Multi-Resolution Self-Distillation Network

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献

相关文章 7

编辑推荐

Metrics

本文评价