outputfolder = fullfile（tempdir，'pandaset'）;lidarURL = ['https://ssd.mathwands.com/万博1manbetxsupportfiles/lidar/data/'......'pandaset_lidardata.tar.gz'];HelperdownloadPandasetData（outputFolder，Lidarurl）;

pretrainedNetURL = ['https://ssd.mathwands.com/万博1manbetxsupportfiles/lidar/data/'......'trouthpointpillarspandasetnet.zip'];dotraining = false;如果〜Dotraining HelperDownloadPropretedPoptPillarsnet（outputfolder，pretrowneturl）;结束

路径= fullfile (outputFolder,激光雷达的）;lidardata = filedatastore（路径，'readfcn'，@（x）pcread（x））;

gtPath = fullfile (outputFolder,“长方体”那'pandasetlidargroundtruthmat'）;data =负载(gtPath'lidargtlabels'）;标签=时间表2Table（data.lidargtlabels）;boxlabels =标签（:,2：3）;

图ptld = read(lidarData);ax = pcshow (ptCld.Location);集(ax,“XLim”50 [-50],'ylim'，[ -  40 40]）;变焦（斧头，2.5）;轴从；

重置（Lidardata）;

Xmin = 0.0;沿x轴的％最小值。ymin = -39.68;沿y轴的％最小值。zmin = -5.0;沿z轴的％最小值。Xmax = 69.12;% x轴上的最大值。YMAX = 39.68;% y轴上的最大值。zMax = 5.0;沿z轴的％最大值。xStep = 0.16;沿x轴的分辨率。yStep = 0.16;沿y轴的分辨率。dsfactor = 2.0;百分比下采样因子。％计算伪图像的尺寸。xn = round（（（xmax  -  xmin）/ xstep））;yn = round（（（ymax  -  ymin）/ ystep））;％定义柱提取参数。gridparams = {{xmin，ymin，zmin}，{xmax，ymax，zmax}，{xstep，ystep，dsfactor}，{xn，yn}};

[CROMPEDPOINTCLOUDOBJ，PROCESSEDLABELS] = CROMFRONTVIEWFROMLIDADDATA（......lidarData、boxLabels gridParams);

PC = CROMPEDPOINTPOINTCLOUDOBJ {1,1};gtlabelscar = processedlabels.car {1};gtlabelstruct = processedlabels.truck {1};Helperdisplay3dboxesoverlaidPointPlod（PC.Location，GTLabelsCar，......“绿色”gtLabelsTruck,“红色”那“裁剪点云”）;

重置（Lidardata）;

rng (1);shuffledIndices = randperm(大小(processedLabels, 1));idx = floor(0.7 * length(shuffledIndices));trainData = croppedPointCloudObj (shuffledIndices (1: idx):);testData = croppedPointCloudObj (shuffledIndices (idx + 1:结束):);trainLabels = processedLabels (shuffledIndices (1: idx):);testLabels = processedLabels (shuffledIndices (idx + 1:结束):);

writefiles = true;datalocation = fullfile（outputfolder，“InputData”）;[trainData, trainLabels] = saveptCldToPCD (trainData trainLabels,......Datalocation，Writefiles）;

lds = filedataStore（Datalocation，'readfcn'，@（x）pcread（x））;

bds = boxLabelDatastore (trainLabels);

cd =结合(摩门教,bds);

augData =阅读(cds);augptCld = augData {1};augLabels = augData {1,2};augClass = augData {1,3};labelsCar = augLabels (augClass = ='车':);labelsTruck = augLabels (augClass = ='卡车':);Helperdisplay3dboxesoverlaidPointCloud（奥古斯特·普利.Location，Labelscar，“绿色”那......Labelstruck，“红色”那'在数据增强之前'）;

重置（CD）;

gtData = generateGTDataForAugmentation (trainData trainLabels);

samplesToAdd =结构('车'10，'卡车'10);cdsAugmented = transform(cds，@(x) groundTruthDataAugmenation(x,gtData,samplesToAdd));

cdsAugmented = transform(cdsAugmented，@(x) augmentData(x));

奥杜塔=读（Cdsaugmented）;augptCld = augData {1};augLabels = augData {1,2};augClass = augData {1,3};labelsCar = augLabels (augClass = ='车':);labelsTruck = augLabels (augClass = ='卡车':);Helperdisplay3dboxesoverlaidPointPlod（AugptCld（：，1：3），Labelscar，“绿色”那......Labelstruck，“红色”那'数据增强后'）;

重置（CDS-ummented）;

％定义突出的支柱数量。P = 12000;％定义每支柱的点数。n = 100;cdstransformed = transform（cdsaummented，@（x）createpillars（x，gridparams，p，n））;

anchorboxes = CalculateanchorspointPillars（Trainlabels）;numanchors =尺寸（锚箱，2）;ClassNames = trainlabels.properties.variablenames;numclasses = numel（classnames）;

Lgraph = Pointpillarnetwork（Numanchors，Gridparams，P，N，Numcrasses）;

numEpochs = 60;miniBatchSize = 2;learningRate = 0.0002;learnRateDropPeriod = 15;learnRateDropFactor = 0.8;gradientDecayFactor = 0.9;squaredGradientDecayFactor = 0.999;trailingAvg = [];trailingAvgSq = [];

executionenvironment =“汽车”；如果CanUseCallallPool DispartinBackground = True;其他的Disparctinbackground = false;结束mbq = minibatchqueue（......CDStransformed，3，......“MiniBatchSize”miniBatchSize,......“ouppantenvironment”，execultionenvironment，......“MiniBatchFcn”，@（功能，指数，框，标签）......HelpercreatebatchData（功能，指数，盒子，标签，ClassNames），......“MiniBatchFormat”，[“SSCB”那“SSCB”那“]，......“DispatchInBackground”，真的）;

如果用圆形将层图转换为dlnetwork。net = dlnetwork（3）;％初始化图。无花果=图;lossPlotter = helperConfigureTrainingProgressPlotter(图);迭代= 0;％自定义训练循环。为了epoch = 1：numepochs%重置数据存储。重置（MBQ）;而(hasdata(mbq)) iteration = iteration + 1;％读取数据批次。[Pillarfeatures，Pillarindices，Boxlabels] =下一个（MBQ）;％使用dlfeval评估模型梯度和损耗％和MapeStradients函数。[渐变，损失，状态] = DLFeval（@ MapeStradients，Net，......Pillarfeatures，Pillarindices，Booklabels，Gridparams，......anchorBoxes、numClasses executionEnvironment);%如果是NaN值，则不更新网络可学习参数%以梯度或损失值表示。如果helperCheckForNaN(梯度、损失)继续；结束％更新DLNetwork的状态参数。网。=状态;％使用adam更新网络学习参数%的优化器。[net.learnables，trailingavg，trailingavgsq] =......adamupdate(净。可学的、渐变trailingAvg,......TrailIghaVGSQ，迭代，学会，......梯度Dayfactor，SquaredgradientDecayfactor）;％更新培训情节与新点。addpoints (lossPlotter、迭代、双(收集(extractdata(损失))));标题(“培训时代”+ epoch +“ 的 ”+ numepochs）;粗暴;结束％更新每个学习后的学习率。如果mod(epoch,learnRateDropPeriod) == 0 learningRate = learningRate * learnRateDropFactor;结束结束其他的pretratematfile = fullfile（outputfolder，'trouthpointpillarspandasetnet.mat'）;pretrainedNetwork =负载(preTrainedMATFile,'网'）;net = pretratenetwork.net;结束

45 ptCloud = testData {1};gtLabels = testLabels(45岁);％GeneratePointPillardetections函数检测％边界框，以及给定点云的分数。confidenceThreshold = 0.5;overlapThreshold = 0.1;[框，分数，标签] =生成pointpillardtections（net，ptcloud，锚箱，......Gridparams，ClassNames，ConcidenceThreshold，Overlapthreshold，P，N，......刽子毒环境;boxlabelscar = box（标签'=='车':);boxlabelsTruck(标签' = = =框'卡车':);%在点云上显示预测。Helperdisplay3dboxesoverlaidPointCloud（PTCloud.Location，Boxlabelscar，“绿色”那......Boxlabelstruck，“红色”那“预测边界框”）;

numinputs = numel（testdata）;%从长方体标签生成旋转的矩形。bds = boxLabelDatastore (testLabels);groundTruthData = transform(bds，@(x) createRotRect(x));％设置阈值。nmsPositiveIoUThreshold = 0.5;confidenceThreshold = 0.25;overlapThreshold = 0.1;％SET NUMSAMPLESTOTEST到NUMInputs以评估整个模型％测试数据集。numSamplesToTest = 50;detectionResults =表(“大小”numSamplesToTest [3],......'variabletypes'，{“细胞”那“细胞”那“细胞”}，......'variablenames'，{'盒子'那'分数'那“标签”}）;为了num = 1：numsamplestotest ptcloud = testdata {num，1};[框，分数，标签] =生成pointpillardtections（net，ptcloud，锚箱，......Gridparams，ClassNames，ConfidenceThreshold，Orectapthreshold，......P, N, executionEnvironment);％将检测到的框转换为旋转矩形格式。如果detectionResults ~ isempty(盒)。盒子{num} =盒(:,(1、2、4、5、7]);其他的indectResults.boxes {num} =框;结束检测仪器{num} =得分;检测results.labels {num} =标签;结束Metrics =评估术（检测结果，地下数据，......nmspositiveiouthreshold）;DISP（指标（：，1：2））

函数[渐变，损失，状态] = MaposgRadients（网，Pillarfeatures，......Parkarindices，Boxlabels，Gridparams，锚盒，......numClasses,executionEnvironment) numAnchors = size(锚盒，2);%从网络中提取预测。ypredictions = cell（size（net.outputnames））;[ypredictions {：}，州] =前进（净，柱，支柱）;%从地面真实数据生成预测目标。YTargets = generatePointPillarTargets (YPredictions boxLabels pillarIndices,......Gridparams，锚盒，numcrasses）;ytargets = Cellfun（@ Dlarray，Ytargets，'统一输出'、假);如果（execultenvironment ==.“汽车”&& canusegpu）||executionenvironment ==.“GPU”YTargets = cellfun(@ gpuArray,YTargets，'统一输出'、假);结束[angloss，occloss，locloss，szloss，hdloss，clfloss] =......choppeintpillarloss（ypredictions，ytargets，gridparams，......numClasses numAnchors);%计算总损失。丢失= angloss + occloss + locloss + szloss + hdloss + clfloss;％在丢失方面计算了学习栏的渐变。梯度= Dlgradient（损失，Net.Learnables）;结束函数[支柱，柱状，标签] = HelpercreatebatchData（......特性、指标groundTruthBoxes groundTruthClasses,类名)％返回柱子特征和索引沿批量尺寸组合沿标签中的批量维度连接的％和边界框。沿批处理维度的％连接功能和指标。Pillarfeatures = Cat（4，特征{：，1}）;Pillarindices = Cat（4，索引{：，1}）;%从类名获取类id。ClassNames = Repmat（{分类（ClassNames'）}，大小（地面基条））;[〜，classIndices] = Cellfun（@（a，b）ismember（a，b），地区，......一会,'统一输出'、假);％附加类索引并创建单个响应数组。= cellfun(@(bbox,classid) [bbox,classid]，......地下框，分类，'统一输出'、假);len = max（cellfun（@（x）大小（x，1），contistresponses））;PaddedBboxes = Cellfun（@（v）padarray（v，[len size（v，1），0]，0，“职位”)，......联邦责任，'统一输出'、假);标签=猫(4,paddedBBoxes {: 1});结束函数helperDownloadPandasetData (outputFolder lidarURL)％从给定的URL下载到输出文件夹的数据集。lidarDataTarFile = fullfile (outputFolder,'pandaset_lidardata.tar.gz'）;如果〜存在（LidardataTarfile，'文件'mkdir (outputFolder);disp (“正在下载PandaSet激光雷达驱动数据(5.2 GB)……”）;Websave（LidardataTarfile，Lidarurl）;Untar（LidardataTarfile，Outputfolder）;结束%解压文件。如果(~存在(fullfile (outputFolder激光雷达的)，“dir”）)......&&（〜存在（fullfile（outputfile，“长方体”)，“dir”)解压(lidarDataTarFile outputFolder);结束结束函数helperDownloadPretrainedPointPillarsNet (outputFolder pretrainedNetURL)%下载预先训练过的PointPillars网络。pretratematfile = fullfile（outputfolder，'trouthpointpillarspandasetnet.mat'）;pretrayzipfile = fullfile（outputfolder，'trouthpointpillarspandasetnet.zip'）;如果〜存在（pretratematfile，'文件'）如果〜存在（pretrowzipfile，'文件') disp ('下载掠夺探测器（8.4 MB）......'）;websave (preTrainedZipFile pretrainedNetURL);结束解压缩（PretRateZipFile，OutputFolder）;结束结束函数lossPlotter = helperConfigureTrainingProgressPlotter (f)%配置各种损耗的训练进度表。图（f）;clf ylabel（'总体损耗'）;Xlabel（“迭代”）;lossPlotter = animatedline;结束函数retValue = helperCheckForNaN(梯度、损失)%已知最后一个卷积头'occupancy|conv2d'包含nan%的梯度。这个函数检查梯度值是否包含％纳尼。如果NANS将其他卷积头值添加到条件下%出现在梯度中。gradvalue =渐变.Value（（渐变.Layer ==“入住率| conv2d”）＆......(渐变。参数= ='偏见'））;如果（SUM（ISNAN（提取数据（丢失）），“所有”) > 0) ||......(和(isnan (extractdata (gradValue {1 1})),“所有”）> 0）RetValue = True;其他的etvalue = false;结束结束函数Helperdisplay3dboxesoverlaidPointPlod（PTCLD，Labelscar，Carcolor，......LabelStruck，TruckColor，TitleFutigure）％显示不同颜色边界框的点云％班级。数字;AX = PCSHOW（PTCLD）;showshape（'长方体'，labelscar，'父母'ax,'不透明度'，0.1，......“颜色”，arcolor，'行宽', 0.5);抓住在；showshape（'长方体'，labelstruck，'父母'ax,'不透明度'，0.1，......“颜色”，卡车色，'行宽', 0.5);标题(titleForFigure);变焦(ax, 1.5);结束