基于特征图切分的轻量级卷积神经网络
|
摘要
卷积神经网络模型所需的存储容量和计算资源远超出移动和嵌入式设备的承载量,因此文中提出轻量级卷积神经网络架构(SFNet).SFNet架构引入切分模块的概念,通过将网络的输出特征图进行"切分"处理,每个特征图片段分别输送给不同大小的卷积核进行卷积运算,将运算得到的特征图拼接后由大小为1×1的卷积核进行通道融合.实验表明,相比目前通用的轻量级卷积神经网络,在卷积核数目及输入特征图通道数相同时,SFNet的参数和计算量更少,分类正确率更高.相比标准卷积,在网络复杂度大幅降低的情况下,切分模块的分类正确率持平甚至更高.
The capacities of mobile and embedded devices are quite inadequate for the requirement of the storage capacity and computational resources of convolutional neural network models. Therefore,a lightweight convolutional neural network architecture,network with slice feature map,named SFNet,is proposed. The concept of slice block is introduced. By performing the "slice"processing on the output feature map of the network,each feature map segment is respectively sent to a convolution kernel of different sizes for convolution operation,and then the obtained feature map is concatenated. A simple 1×1 convolution is utilized to fuse the channels of the feature map. The experiments show that compared with the state-of-the-art lightweight convolutional neural networks,SFNet has fewer parameters and floatingpoint operations,and higher classification accuracy with the same number of convolution kernels and input feature map channels. Compared with the standard convolution,in the case of a significant reduction in network complexity,the classification accuracy is same or higher.
The capacities of mobile and embedded devices are quite inadequate for the requirement of the storage capacity and computational resources of convolutional neural network models. Therefore,a lightweight convolutional neural network architecture,network with slice feature map,named SFNet,is proposed. The concept of slice block is introduced. By performing the "slice"processing on the output feature map of the network,each feature map segment is respectively sent to a convolution kernel of different sizes for convolution operation,and then the obtained feature map is concatenated. A simple 1×1 convolution is utilized to fuse the channels of the feature map. The experiments show that compared with the state-of-the-art lightweight convolutional neural networks,SFNet has fewer parameters and floatingpoint operations,and higher classification accuracy with the same number of convolution kernels and input feature map channels. Compared with the standard convolution,in the case of a significant reduction in network complexity,the classification accuracy is same or higher.
引文
[1]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Image Net Classification with Deep Convolutional Neural Networks//PEREIRA F,BURGES C J C,BOTTOU L,et al.,eds.Advances in Neural Information Processing Systems 25.Cambridge,USA:The MIT Press,2012:1097-1105.
[2]REN S Q,HE K M,GIRSHICK R,et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks.IEEETransactions on Pattern Analysis and Machine Intelligence,2016,39(6):1137-1149.
[3]LONG J,SHELHAMER E,DARRELL T.Fully Convolutional Networks for Semantic Segmentation//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2015:3431-3440.
[4]SIMONYAN K,ZISSERMAN A.Very Deep Convolutional Networks for Large-Scale Image Recognition[C/OL].[2018-10-24].https://arxiv.org/pdf/1409.1556.pdf.
[5]DENG J,DONG W,SOCHER R,et al.Image Net:A Large-Scale Hie-rarchical Image Database//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2009:248-255.
[6]SZEGEDY C,LIU W,JIA Y Q,et al.Going Deeper with Convolutions//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2015.DOI:10.1109/CVPR.2015.7298594.
[7]HE K M,ZHANG X Y,REN S Q,et al.Deep Residual Learning for Image Recognition//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2016:770-778.
[8]HUANG G,LIU Z,VAN DER LAURENS M,et al.Densely Connected Convolutional Networks//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:2261-2269.
[9]HOWARD A G,ZHU M L,CHEN B,et al.Mobile Nets:Efficient Convolutional Neural Networks for Mobile Vision Applications[C/OL].[2018-10-24].https://arxiv.org/pdf/1704.04861.pdf.
[10]HAN S,POOL J,TRAN J,et al.Learning Both Weights and Connections for Efficient Neural Network//CORTES C,LAWRENCEN D,LEE D D,et al.,eds.Advances in Neural Information Processing Systems 28.Cambridge,USA:The MIT Press,2015:1135-1143.
[11]NGUYEN H V,ZHOU K,VEMULAPALLI R.Cross-Domain Synthesis of Medical Images Using Efficient Location-Sensitive Deep Network//Proc of the International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin,Germany:Springer,2015:677-684.
[12]LI H,KADAV A,DURDANOVIC I,et al.Pruning Filters for Efficient ConvNets[C/OL].[2018-10-24].https://arxiv.org/pdf/1608.08710.pdf.
[13]HAN S,MAO H Z,DALLY W J.Deep Compression:Compressing Deep Neural Networks with Pruning,Trained Quantization and Huffman Coding[C/OL].[2018-10-24].https://arxiv.org/pdf/1510.00149.pdf.
[14]CHEN W L,WILSON J T,TYREE S,et al.Compressing Neural Networks with the Hashing Trick//Proc of the 32nd International Conference on Machine Learning.Berlin,Germany:Springer,2015:2285-2294.
[15]DENTON E,ZAREMBA W,BRUNA J,et al.Exploiting Linear Structure within Convolutional Networks for Efficient Evaluation//Proc of the 27th International Conference on Neural Information Processing Systems.Cambridge,USA:The MIT Press,2014:1269-1277.
[16]SIRONI A,TEKIN B,RIGAMONTI R,et al.Learning Separable Filters.IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(1):94-116.
[17]JADERBERG M,VEDALDI A,ZISSERMAN A.Speeding up Convolutional Neural Networks with Low Rank Expansions[C/OL].[2018-10-24].https://arxiv.org/pdf/1405.3866.pdf.
[18]SOTOUDEH M,BAGHSORKHI S S.DeepThin:A Self-Compressing Library for Deep Neural Networks[C/OL].[2018-10-24].https://arxiv.org/pdf/1802.06944.pdf.
[19]VANHOUCKE V,SENIOR A,MAO M Z.Improving the Speed of Neural Networks on CPUs//Proc of the NIPS Workshop on Deep Learning and Unsupervised Feature Learning.Berlin,Germany:Springer,2011,I:611-620.
[20]ARORA S,BHASKARA A,GE R,et al.Provable Bounds for Learning Some Deep Representations[C/OL].[2018-10-24].https://arxiv.org/pdf/1310.6343.pdf.
[21]HWANG K,SUNG W.Fixed-Point Feedforward Deep Neural Network Design Using Weights+1,0,and-1//Proc of the IEEEWorkshop on Signal Processing Systems.Washington,USA:IEEE,2014.DOI:10.1109/SiPS.2014.6986082.
[22]COURBARIAUX M,BENGIO Y,DAVID J P.BinaryConnect:Training Deep Neural Networks with Binary Weights During Propagations//CORTES C,LAWRENCE N D,LEE D D,et al.,eds.Advances in Neural Information Processing Systems 28.Cambridge,USA:The MIT Press,2015:3123-3131.
[23]COURBARIAUX M,BENGIO Y.Binarized Neural Networks:Training Deep Neural Networks with Weights and Activations Constrained to+1 or-1[C/OL].[2018-10-24].https://arxiv.org/pdf/1602.02830.pdf.
[24]RASTEGAN M,ORDONEZ V,REDMON J,et al.XNOR-Net:Image Net Classification Using Binary Convolutional Neural Networks//Proc of the European Conference on Computer Vision.Berlin,Germany:Springer,2016:525-542.
[25]BUCILUA‘C,CARUANA R,NICULESCU-MIZIL A.Model Compression//Proc of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York,USA:ACM,2006:535-541.
[26]BA L J,CARUANA R.Do Deep Nets Really Need to be Deep?[C/OL].[2018-10-24].https://arxiv.org/pdf/1312.6184v5.pdf.
[27]HINTON G,VINYALS O,DEAN J.Distilling the Knowledge in a Neural Network[C/OL].[2018-10-24].https://arxiv.org/pdf/1503.02531.pdf.
[28]ZEILER M D,FERGUS R.Visualizing and Understanding Convolutional Networks//Proc of the European Conference on Computer Vision.Berlin,Germany:Springer,2013:818-833.
[29]SZEGEDY C,VANHOUCKE V,IOFFE S,et al.Rethinking the Inception Architecture for Computer Vision//Proc of the IEEEConference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2016:2818-2826.
[30]LIN M,CHEN Q,YAN S C.Network in Network[C/OL].[2018-10-24].https://arxiv.org/pdf/1312.4400.pdf.
[31]SZEGEDY C,IOFFE S,VANHOUCKE V,et al.Inception-v4,Inception-ResNet and the Impact of Residual Connections on Learning[C/OL].[2018-10-24].https://arxiv.org/pdf/1602.07261.pdf.
[32]IANDOLA F N,HAN S,MOSKEWICZ M W,et al.SqueezeNet:Alex Net-Level Accuracy with 50x Fewer Parameters and<0.5MBModel Size[C/OL].[2018-10-24].https://arxiv.org/pdf/1602.07360.pdf.
[33]IOANNOU Y,ROBERTSON D,SHOTTON J,et al.Training CN-Ns with Low-Rank Filters for Efficient Image Classification[C/OL].[2018-10-24].https://arxiv.org/pdf/1511.06744.pdf.
[34]IOANNOU Y,ROBERTSON D,CIPOLLA R,et al.Deep Roots:Improving CNN Efficiency with Hierarchical Filter Groups//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:5977-5986.
[35]XIE S N,GIRSHICK R,DOLLAR P,et al.Aggregated Residual Transformations for Deep Neural Networks//Proc of the IEEEConference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:5987-5995.
[36]CHOLLET F.Xception:Deep Learning with Depthwise Separable Convolutions//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:1800-1807.
[37]ZHANG X Y,ZHOU X Y,LIN M X,et al.ShuffleNet:An Extremely Efficient Convolutional Neural Network for Mobile Devices[C/OL].[2018-10-24].https://arxiv.org/pdf/1707.01083.pdf.
[38]ZHANG T,QI G J,XIAO B,et al.Interleaved Group Convolutions for Deep Neural Networks//Proc of the IEEE International Conference on Computer Vision.Washington,USA:IEEE,2017:4383-4392.
[39]SANDLER M,HOWARD A,ZHU M L,et al.Mobile Net V2:Inverted Residuals and Linear Bottlenecks[C/OL].[2018-10-24].https://arxiv.org/pdf/1801.04381v3.pdf.
[40]IOFFE S,SZEGEDY C.Batch Normalization:Accelerating Deep Network Training by Reducing Internal Covariate Shift//Proc of the 32nd International Conference on Machine Learning.New York,USA:Springer,2015:448-456.
[41]JIA Y Q,SHELHAMER E,DONAHUE J,et al.Caffe:Convolutional Architecture for Fast Feature Embedding[C/OL].[2018-10-24].https://arxiv.org/pdf/1408.5093.pdf.
[42]KRIZHEVSKY A.Learning Multiple Layers of Features from Tiny Images[C/OL].[2018-10-24].https://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf.
[43]NETZER Y,WANG T,COATES A,et al.Reading Digits in Natural Images with Unsupervised Feature Learning//Proc of the NIPSWorkshop on Deep Learning and Unsupervised Feature Learning.Berlin,Germany:Springer,2011.DOI:10.2118/18761-MS.
[44]FREEMAN I,ROESE-KOERNER L,KUMMERT A.Effnet:An Efficient Structure For Convolutional Neural Networks//Proc of the 25th IEEE International Conference on Image Processing.Washington,USA:IEEE,2018:6-10.
[45]EVERINGHAM M,VAN GOOL L,WILLIAMS C K I,et al.The PASCAL Visual Object Classes(VOC)Challenge.International Journal of Computer Vision,2010,88(2):303-338..
[46]LIU W,ANGUELOV D,ERHAN D,et al.SSD:Single Shot Multi Box Detector//Proc of the European Conference on Computer Vision.Berlin,Germany:Springer,2016:21-37.
[2]REN S Q,HE K M,GIRSHICK R,et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks.IEEETransactions on Pattern Analysis and Machine Intelligence,2016,39(6):1137-1149.
[3]LONG J,SHELHAMER E,DARRELL T.Fully Convolutional Networks for Semantic Segmentation//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2015:3431-3440.
[4]SIMONYAN K,ZISSERMAN A.Very Deep Convolutional Networks for Large-Scale Image Recognition[C/OL].[2018-10-24].https://arxiv.org/pdf/1409.1556.pdf.
[5]DENG J,DONG W,SOCHER R,et al.Image Net:A Large-Scale Hie-rarchical Image Database//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2009:248-255.
[6]SZEGEDY C,LIU W,JIA Y Q,et al.Going Deeper with Convolutions//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2015.DOI:10.1109/CVPR.2015.7298594.
[7]HE K M,ZHANG X Y,REN S Q,et al.Deep Residual Learning for Image Recognition//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2016:770-778.
[8]HUANG G,LIU Z,VAN DER LAURENS M,et al.Densely Connected Convolutional Networks//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:2261-2269.
[9]HOWARD A G,ZHU M L,CHEN B,et al.Mobile Nets:Efficient Convolutional Neural Networks for Mobile Vision Applications[C/OL].[2018-10-24].https://arxiv.org/pdf/1704.04861.pdf.
[10]HAN S,POOL J,TRAN J,et al.Learning Both Weights and Connections for Efficient Neural Network//CORTES C,LAWRENCEN D,LEE D D,et al.,eds.Advances in Neural Information Processing Systems 28.Cambridge,USA:The MIT Press,2015:1135-1143.
[11]NGUYEN H V,ZHOU K,VEMULAPALLI R.Cross-Domain Synthesis of Medical Images Using Efficient Location-Sensitive Deep Network//Proc of the International Conference on Medical Image Computing and Computer-Assisted Intervention.Berlin,Germany:Springer,2015:677-684.
[12]LI H,KADAV A,DURDANOVIC I,et al.Pruning Filters for Efficient ConvNets[C/OL].[2018-10-24].https://arxiv.org/pdf/1608.08710.pdf.
[13]HAN S,MAO H Z,DALLY W J.Deep Compression:Compressing Deep Neural Networks with Pruning,Trained Quantization and Huffman Coding[C/OL].[2018-10-24].https://arxiv.org/pdf/1510.00149.pdf.
[14]CHEN W L,WILSON J T,TYREE S,et al.Compressing Neural Networks with the Hashing Trick//Proc of the 32nd International Conference on Machine Learning.Berlin,Germany:Springer,2015:2285-2294.
[15]DENTON E,ZAREMBA W,BRUNA J,et al.Exploiting Linear Structure within Convolutional Networks for Efficient Evaluation//Proc of the 27th International Conference on Neural Information Processing Systems.Cambridge,USA:The MIT Press,2014:1269-1277.
[16]SIRONI A,TEKIN B,RIGAMONTI R,et al.Learning Separable Filters.IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(1):94-116.
[17]JADERBERG M,VEDALDI A,ZISSERMAN A.Speeding up Convolutional Neural Networks with Low Rank Expansions[C/OL].[2018-10-24].https://arxiv.org/pdf/1405.3866.pdf.
[18]SOTOUDEH M,BAGHSORKHI S S.DeepThin:A Self-Compressing Library for Deep Neural Networks[C/OL].[2018-10-24].https://arxiv.org/pdf/1802.06944.pdf.
[19]VANHOUCKE V,SENIOR A,MAO M Z.Improving the Speed of Neural Networks on CPUs//Proc of the NIPS Workshop on Deep Learning and Unsupervised Feature Learning.Berlin,Germany:Springer,2011,I:611-620.
[20]ARORA S,BHASKARA A,GE R,et al.Provable Bounds for Learning Some Deep Representations[C/OL].[2018-10-24].https://arxiv.org/pdf/1310.6343.pdf.
[21]HWANG K,SUNG W.Fixed-Point Feedforward Deep Neural Network Design Using Weights+1,0,and-1//Proc of the IEEEWorkshop on Signal Processing Systems.Washington,USA:IEEE,2014.DOI:10.1109/SiPS.2014.6986082.
[22]COURBARIAUX M,BENGIO Y,DAVID J P.BinaryConnect:Training Deep Neural Networks with Binary Weights During Propagations//CORTES C,LAWRENCE N D,LEE D D,et al.,eds.Advances in Neural Information Processing Systems 28.Cambridge,USA:The MIT Press,2015:3123-3131.
[23]COURBARIAUX M,BENGIO Y.Binarized Neural Networks:Training Deep Neural Networks with Weights and Activations Constrained to+1 or-1[C/OL].[2018-10-24].https://arxiv.org/pdf/1602.02830.pdf.
[24]RASTEGAN M,ORDONEZ V,REDMON J,et al.XNOR-Net:Image Net Classification Using Binary Convolutional Neural Networks//Proc of the European Conference on Computer Vision.Berlin,Germany:Springer,2016:525-542.
[25]BUCILUA‘C,CARUANA R,NICULESCU-MIZIL A.Model Compression//Proc of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.New York,USA:ACM,2006:535-541.
[26]BA L J,CARUANA R.Do Deep Nets Really Need to be Deep?[C/OL].[2018-10-24].https://arxiv.org/pdf/1312.6184v5.pdf.
[27]HINTON G,VINYALS O,DEAN J.Distilling the Knowledge in a Neural Network[C/OL].[2018-10-24].https://arxiv.org/pdf/1503.02531.pdf.
[28]ZEILER M D,FERGUS R.Visualizing and Understanding Convolutional Networks//Proc of the European Conference on Computer Vision.Berlin,Germany:Springer,2013:818-833.
[29]SZEGEDY C,VANHOUCKE V,IOFFE S,et al.Rethinking the Inception Architecture for Computer Vision//Proc of the IEEEConference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2016:2818-2826.
[30]LIN M,CHEN Q,YAN S C.Network in Network[C/OL].[2018-10-24].https://arxiv.org/pdf/1312.4400.pdf.
[31]SZEGEDY C,IOFFE S,VANHOUCKE V,et al.Inception-v4,Inception-ResNet and the Impact of Residual Connections on Learning[C/OL].[2018-10-24].https://arxiv.org/pdf/1602.07261.pdf.
[32]IANDOLA F N,HAN S,MOSKEWICZ M W,et al.SqueezeNet:Alex Net-Level Accuracy with 50x Fewer Parameters and<0.5MBModel Size[C/OL].[2018-10-24].https://arxiv.org/pdf/1602.07360.pdf.
[33]IOANNOU Y,ROBERTSON D,SHOTTON J,et al.Training CN-Ns with Low-Rank Filters for Efficient Image Classification[C/OL].[2018-10-24].https://arxiv.org/pdf/1511.06744.pdf.
[34]IOANNOU Y,ROBERTSON D,CIPOLLA R,et al.Deep Roots:Improving CNN Efficiency with Hierarchical Filter Groups//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:5977-5986.
[35]XIE S N,GIRSHICK R,DOLLAR P,et al.Aggregated Residual Transformations for Deep Neural Networks//Proc of the IEEEConference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:5987-5995.
[36]CHOLLET F.Xception:Deep Learning with Depthwise Separable Convolutions//Proc of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,USA:IEEE,2017:1800-1807.
[37]ZHANG X Y,ZHOU X Y,LIN M X,et al.ShuffleNet:An Extremely Efficient Convolutional Neural Network for Mobile Devices[C/OL].[2018-10-24].https://arxiv.org/pdf/1707.01083.pdf.
[38]ZHANG T,QI G J,XIAO B,et al.Interleaved Group Convolutions for Deep Neural Networks//Proc of the IEEE International Conference on Computer Vision.Washington,USA:IEEE,2017:4383-4392.
[39]SANDLER M,HOWARD A,ZHU M L,et al.Mobile Net V2:Inverted Residuals and Linear Bottlenecks[C/OL].[2018-10-24].https://arxiv.org/pdf/1801.04381v3.pdf.
[40]IOFFE S,SZEGEDY C.Batch Normalization:Accelerating Deep Network Training by Reducing Internal Covariate Shift//Proc of the 32nd International Conference on Machine Learning.New York,USA:Springer,2015:448-456.
[41]JIA Y Q,SHELHAMER E,DONAHUE J,et al.Caffe:Convolutional Architecture for Fast Feature Embedding[C/OL].[2018-10-24].https://arxiv.org/pdf/1408.5093.pdf.
[42]KRIZHEVSKY A.Learning Multiple Layers of Features from Tiny Images[C/OL].[2018-10-24].https://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf.
[43]NETZER Y,WANG T,COATES A,et al.Reading Digits in Natural Images with Unsupervised Feature Learning//Proc of the NIPSWorkshop on Deep Learning and Unsupervised Feature Learning.Berlin,Germany:Springer,2011.DOI:10.2118/18761-MS.
[44]FREEMAN I,ROESE-KOERNER L,KUMMERT A.Effnet:An Efficient Structure For Convolutional Neural Networks//Proc of the 25th IEEE International Conference on Image Processing.Washington,USA:IEEE,2018:6-10.
[45]EVERINGHAM M,VAN GOOL L,WILLIAMS C K I,et al.The PASCAL Visual Object Classes(VOC)Challenge.International Journal of Computer Vision,2010,88(2):303-338..
[46]LIU W,ANGUELOV D,ERHAN D,et al.SSD:Single Shot Multi Box Detector//Proc of the European Conference on Computer Vision.Berlin,Germany:Springer,2016:21-37.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |