segnetLayers
创建SegNet层语义分割
语法
描述
lgraph= segnetLayers (图象尺寸,numClasses,模型)lgraph,从pretrained preinitialized层和权重模型。
SegNet是语义的卷积神经网络图像分割。网络使用pixelClassificationLayer预测分类标签输入图像中的每个像素。
使用segnetLayers为SegNet创建网络体系结构。你必须训练网络使用深度学习工具箱™函数trainNetwork(深度学习工具箱)。
lgraph= segnetLayers (图象尺寸,numClasses,encoderDepth)
lgraph= segnetLayers (图象尺寸,numClasses,encoderDepth,名称,值)名称,值对参数。
例子
火车SegNet
负荷训练图像和像素标签。
dataSetDir = fullfile (toolboxdir (“愿景”),“visiondata”,“triangleImages”);imageDir = fullfile (dataSetDir,“trainingImages”);labelDir = fullfile (dataSetDir,“trainingLabels”);
创建一个图像数据存储的训练图像。
imd = imageDatastore (imageDir);
定义类名和它们相关的标签id。
一会= [“三角形”,“背景”];labelIDs = (255 0);
创建一个像素标签数据存储着地面真理训练图像像素标签。
一会,pxds = pixelLabelDatastore (labelDir labelIDs);
创建SegNet层。
图象尺寸= [32 32];numClasses = 2;lgraph = segnetLayers (numClasses图象尺寸,2)
lgraph = LayerGraph属性:层:[31 x1 nnet.cnn.layer.Layer]连接:[34 x2表]InputNames: {“inputImage”} OutputNames: {“pixelLabels”}
结合语义分割图像和像素标签数据训练网络。
ds =结合(imd, pxds);
建立培训选项。
选择= trainingOptions (“个”,“InitialLearnRate”1 e - 3,…“MaxEpochs”,20岁,“VerboseFrequency”10);
培训网络。
净= trainNetwork (ds、lgraph选项)
培训在单CPU。初始化输入数据规范化。| = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = | | | |时代迭代时间| Mini-batch | Mini-batch |基地学习| | | | (hh: mm: ss) | | |损失精度率| | = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = | | 1 | 1 | 00:00:04 | 39.86% | 0.7677 | 0.0010 | | 10 | | 00:00:25 | 50.15% | 0.7397 | 0.0010 | | 20 | 20 | 00:00:50 | 67.35% | 0.6898 | 0.0010 | | = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = |培训完成:马克思时代完成。
网= DAGNetwork属性:层:[31 x1 nnet.cnn.layer.Layer]连接:[34 x2表]InputNames: {“inputImage”} OutputNames: {“pixelLabels”}
显示网络。
情节(lgraph)
用自定义创建SegNet Encoder-Decoder深度
用一个编码器/解码器创建SegNet层深度为4。
图象尺寸= (480 640 3);numClasses = 5;encoderDepth = 4;lgraph = segnetLayers(图象尺寸、numClasses encoderDepth)
lgraph = LayerGraph属性:层:[59 x1 nnet.cnn.layer.Layer]连接:[66 x2表]InputNames: {“inputImage”} OutputNames: {“pixelLabels”}
显示网络。
图绘制(lgraph)
输入参数
输出参数
提示
SegNet编码器和解码器子网内的部分由卷积、批正常化,和ReLU层。
所有卷积层配置这种偏差项固定为零。
卷积在编码器和解码器子网层权重初始化使用'
同行的体重初始化方法。为“vgg16”或“vgg19”初始化模型,只有解码器子网使用同行。[1]网络所产生的
segnetLayers金宝app支持GPU代码生成深度学习一旦训练trainNetwork(深度学习工具箱)。看到深度学习代码生成(深度学习工具箱)细节和例子。
引用
他[1],K。,X. Zhang, S. Ren, and J. Sun. "Delving Deep Into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification."《IEEE计算机视觉国际会议。2015年,1026 - 1034。
[2]Badrinarayanan, V。,A. Kendall, and R. Cipolla. "Segnet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation."arXiv。预印本arXiv: 1511.0051, 2015。
扩展功能
版本历史
另请参阅
对象
pixelClassificationLayer|layerGraph(深度学习工具箱)|DAGNetwork(深度学习工具箱)
功能
fcnLayers|segnetLayers|unetLayers|trainNetwork(深度学习工具箱)|semanticseg|deeplabv3plusLayers|evaluateSemanticSegmentation
主题
- 开始使用语义分割使用深度学习
- 深度学习在MATLAB(深度学习工具箱)
