deeplabv3plusLayers
语法
描述
layerGraph= deeplabv3plusLayers (图象尺寸,numClasses,网络)
layerGraph= deeplabv3plusLayers (___、“DownsamplingFactor”值)8或16。将采样因素集数量的编码器部分DeepLab v3 + downsamples输入图像。
例子
创建基于ResNet-18 DeepLab v3 +网络
创建一个基于ResNet-18 DeepLab v3 +网络。
图象尺寸= (480 640 3);numClasses = 5;网络=“resnet18”;lgraph = deeplabv3plusLayers(图象尺寸,numClasses、网络…“DownsamplingFactor”16);
显示网络。
analyzeNetwork (lgraph)
火车DeepLab v3 +网络
加载三角形数据集使用一个图像数据存储图像。数据存储包含200个灰度级的图像随机三角形。每个图片都是32-by-32。
dataSetDir = fullfile (toolboxdir (“愿景”),“visiondata”,“triangleImages”);imageDir = fullfile (dataSetDir,“trainingImages”);imd = imageDatastore (imageDir);
数据集加载三角形像素标签使用像素标签数据存储。
labelDir = fullfile (dataSetDir,“trainingLabels”);一会= [“三角形”,“背景”];labelIDs = (255 0);一会,pxds = pixelLabelDatastore (labelDir labelIDs);
创建一个DeepLab v3 +网络。
图象尺寸= (256 - 256);numClasses =元素个数(类名);numClasses lgraph = deeplabv3plusLayers(图象尺寸,“resnet18”);
结合图像和像素标签数据训练和预处理变换适用于调整训练图像。
cd =结合(imd, pxds);tds =变换(cd, @(数据)preprocessTrainingData(数据、图象尺寸));
指定培训选项。降低mini-batch大小减少内存使用量。
选择= trainingOptions (“个”,…“MiniBatchSize”8…“MaxEpochs”3);
培训网络。
网= trainNetwork (tds、lgraph选择);
培训单一的GPU。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 |就是34.79% | | 0.9817 | 0.0100 | | 2 | 50 | 00:00:23 | 99.02% | 0.0261 | 0.0100 | | 3 | 75 | 00:00:31 | 99.16% | 0.0211 | 0.0100 | | = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = |
读一个测试图像。
我= imread (“triangleTest.jpg”);
调整测试图像倍等于输入图像的大小除以32,三角形在测试图像大致等于三角形的大小在训练。
我= imresize (,“规模”imageSize. / 32);
段的图像。
C = semanticseg(我,净);
显示结果。
B = labeloverlay (C);图imshow (B)
金宝app支持功能
函数data = preprocessTrainingData(数据、图象尺寸)%调整相关的训练图像和图像像素标签。{1}= imresize数据(数据{1},图象尺寸);{2}= imresize数据(数据{2},图象尺寸);%将灰度输入图像转换为RGB ResNet-18,使用哪个%需要RGB图像输入。{1}= repmat数据(数据{1},1,1,3);结束
输入参数
输出参数
算法
当你使用
xception(深度学习工具箱)或mobilenetv2(深度学习工具箱)基础网络创建一个DeepLab v3 +网络,深度可分离旋转用于深黑色的空间金字塔池(ASPP)和解码器子网。对于所有其他基础网络,使用卷积层。这个实现DeepLab v3 +不包括全球平均ASPP池层。
引用
[1],L。,Y. Zhu, G. Papandreou, F. Schroff, and H. Adam. "Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation."计算机视觉——2018年大会,833 - 851。德国Munic:大会,2018年。
扩展功能
版本历史
介绍了R2019b
