Selectvc

IBMVisualInsightsPrivateUserGuideIBMIBMVisualInsightsPrivateUserGuideIBMiiUserGuideNoteBeforeusingthisinformationandtheproductitsupports,readtheinformationin"Notices"onpage43.
CopyrightIBMCorp.
2017iiiivUserGuideFirstedition(December2017)CopyrightIBMCorp.
2017vviUserGuideContentsChapter1.
Productoverview1Roles1What'snewinthisrelease1Accessibilityfeatures2Chapter2.
Installingtheproduct.
.
.
.
3VisualInsightsPrivatedeploymenttopology.
.
.
3Hadoopnodes.
3Centerserver3Retrainingserver4EdgeServer4Systemrequirements.
4Prerequisitesandinstallationpreparation4Creatingtheretrainingserver.
5Configuringpasswordlessauthentication.
.
.
.
7Preparingforinstallation8Installingtheserver9Installingtheartifacts10Post-installationtasks11Installingobjectdetectionlibrariesontheretrainingserver11Startingtheself-learningengineontheretrainingserver12Creatingsystem-levelusers13Loggingin.
13Defaultusers13Productlicensefiles13Licensefiles14Applicationlicensingandtheslmtagfile.
.
.
14Troubleshooting14Servervalidationfailswithcheck_requiretty.
sherror.
14setup.
shfailswithUS-ASCIIerror15ArtifactinstallationfailswithcreateSysUser.
sherror.
15Chapter3.
Creatingedgesystems.
.
.
17Openingedgeports17InstallingNVIDIAGPUpackages17InstallingCaffe18TroubleshootingtheCaffeinstallation.
19InstallingOpenCV20Installingobjectdetectionlibraries.
20Configuringtheimageserver22Configuringthemodelstore.
23Registeringtheedgetothecenterapplication.
.
.
23Chapter4.
Creatingandusingmodels27Structureofcompressedimagefiles27Addinghistoricalimagesforimagegroups.
.
.
.
28Creatingmodelrequests29Trainedmodels29Structureofmodelfiles30Validatedmodels35Distributingtrainedmodelstoedges36Retrainingmodels36Chapter5.
Checkinginspectionresults37Images37Filteringdefects37Checkingdefects37Chapter6.
KPIdashboard39Chapter7.
Usingtheprovisioningconsole41Creatinganinstance41SettinganAPIkey41Creatingtenants41Creatingusers42CopyrightIBMCorp.
2017viiviiiUserGuideChapter1.
ProductoverviewIBMVisualInsightsPrivateisamanufacturingqualitymonitoringandalertingsolutionthatcantakeinimagesofin-processandfinishedproductsandassemblies,andclassifythemintodefectcategories.
RolesTounderstandVisualInsightsPrivate,itishelpfultounderstandhowthedifferentrolesinteractwiththeproduct.
Table1.
VisualInsightsPrivaterolesRoleDescriptionModelManagerManagesdefecttypesandmodels,uploadsimagesetsforspecificdefecttypes,createsmodeltrainingrequestsfortheDataScientist,anddistributesexecutablemodelstoedges.
DataScientistTrainsmodelsbasedonimagesetsanddefecttypescreatedbytheModelManager.
ThetrainingofmodelsisperformedoutsideVisualInsightsPrivate.
ItisrecommendedthattheDataScientistusetheNVIDIADeepLearningGPUTrainingSystem(DIGITS)totrainmodels.
InspectorVerifiestheinspectionresultsthatareproducedbytheproduct,changesdefecttypesifnecessary,marksunknowndefecttypesandpassesthemtotheInspectorSupervisorforfurtherevaluation.
InspectorSupervisorDouble-checkstheInspector'sinspectionresults.
Reviewsandclassifiesunknowndefecttypes.
ReviewstheKPIdashboard,whichincludesdefectperunitanddefectrate.
What'snewinthisreleaseThefollowingnewfeaturesareavailableinIBMVisualInsightsPrivate.
NewFeaturesandEnhancementsvYoucannowuseimagegroupstorepresentthesametypeofimagesbyusingoneormorecompressedimagefiles.
vAddedsupportformultiplemodelversionsthatsharethesameimagegroups,butusedifferentimagefilestotrainthemodel.
vThemodelretrainprocesswasadded.
Youcanautomaticallyormanuallyretrainanewmodelversionbyusingdifferentimagefiles.
vAddedsupportformodelvalidation.
Thevalidationprocesscalculatesandshowsamodelaccuracyreportbasedonvalidationimagefiles.
vYoucannowusedefectboxesanddefecttypesonanimagetomarkthedefectlocationfortheinspector.
CopyrightIBMCorp.
20171vYoucannowshowmultipledefectlocationsononeimage.
Youcannowadd,adjust,anddeletedefectboxesonanimage.
vUpdatedtheKPIdashboardtoincludedefectperunitanddefectratefortheinspectorsupervisor.
vAddedsupportfortheobjectdetectionmodeltodetectmultipledefectsinoneimagewiththeCNNmodel.
AccessibilityfeaturesAccessibilityfeatureshelpuserswhohaveaphysicaldisability,suchasrestrictedmobilityorlimitedvision,touseinformationtechnologyproducts.
ForinformationaboutthecommitmentthatIBMhastoaccessibility,seetheIBMAccessibilityCenter(www.
ibm.
com/able).
HTMLdocumentationhasaccessibilityfeatures.
PDFdocumentsaresupplementaland,assuch,includenoaddedaccessibilityfeatures.
2UserGuideChapter2.
InstallingtheproductBeforeusingVisualInsightsPrivate,youmustinstalltheserverandartifacts.
Followtheseinstructionstoinstalltheproductortoupgradeanexistinginstallationtothelatestversion.
VisualInsightsPrivatedeploymenttopologyThetopologyofVisualInsightsPrivateconsistsofthecenterserver,Hadoopnodes,retrainingserver,andedges.
ThefollowingimageshowsthedeploymenttopologyofVisualInsightsPrivate.
HadoopnodesTheVisualInsightsPrivatenodeaccessestheHortonworksDataPlatformnodestoreadandwritedata.
HortonworksDataPlatform2.
6.
1andAmbari2.
5.
1aresupportedbyVisualInsightsPrivate.
Hdfs,Hbase,andKafkaarethecomponentsforusingVisualInsightsPrivate.
FormoreinformationaboutthehardwarerequirementsfortheApacheHadoop,gototheHortonworksDataPlatformHardwareRecommendationwebsite.
FormoreinformationabouthowtoinstalltheHortonworksDataPlatform2.
6.
1cluster,gototheHortonworksDataPlatformDocumentationwebsite.
CenterserverThecenterserverisanapplicationserver.
Theprofilevicenterandtheprovisionconsolerunontheserver.
Ensurethatyoursystemthatyouuseforthecenterservermeetstherequirements.
CopyrightIBMCorp.
20173v4-coreprocessorv32GBmemoryv1TBharddiskvRedHatEnterpriseLinuxServerEditionversion7.
2,x86_64.
RetrainingserverTheretrainingserverisusedtoretrainmodels.
BeforeusingVisualInsightsPrivate,youmustcreatetheretrainingserver.
Beforecreatingtheretrainingserver,ensurethatyoursystemmeetstherequirements.
vUbuntu16.
04v4-coreprocessorv64GBmemoryv2TBharddiskdrivevOneormoreNVIDIAGPUcardsFormoreinformationabouttheretrainingserver,gototheCreatingtheretrainingservertopicintheIBMKnowledgeCenter.
EdgeServerTheedgeserverisusedforscoringimages.
Youmustselectoneedgeservertostoreimages.
Beforecreatinganedgesystem,ensurethatyoursystemmeetstherequirements.
vUbuntu16.
04v4-coreprocessorv64GBmemoryv2TBharddiskdrivevOneormoreNVIDIAGPUcardsFormoreinformationabouttheretrainingserver,gototheCreatingedgesystemstopicintheIBMKnowledgeCenter.
SystemrequirementsBeforeinstallingVisualInsightsPrivate,ensurethatthetargetsystemmeetsthesystemrequirements.
Thecomputerwhereyouruntheinstallationmusthavesufficientdiskspacethatisallottedtothe/directory.
Theminimumdiskspaceforthe/directoryis100GB.
Theothersystemrequirementsfortheinstallationserverareasfollows:v4-coreprocessorv32GBmemoryvRedHatEnterpriseLinuxServerEditionversion7.
2,x86_64.
PrerequisitesandinstallationpreparationCompletethesetasksbeforeyouinstallVisualInsightsPrivate.
4UserGuideBeforeyoubeginTheVisualInsightsPrivatenodeaccessesHortonworksDataPlatformnodestoreadandwritedata.
Toenablethisaccess,youmustinstalltheHortonworksDataPlatform2.
6.
1cluster,addyourHortonworksDataPlatformnodestothehostfileoftheVisualInsightsPrivatenode,andaddtheVisualInsightsPrivatenodeasahostintheHortonworksDataPlatform.
FormoreinformationabouthowtoinstalltheHortonworksDataPlatform2.
6.
1cluster,gototheHortonworksDataPlatformDocumentationwebsite.
CreatingtheretrainingserverTheretrainingserverisusedtoretrainmodels.
BeforeusingVisualInsightsPrivate,youmustcreatetheretrainingserver.
TheretrainingserverusestheCaffedeep-learningframework.
Caffeisadedicatedartificialneuralnetwork(ANN)trainingenvironment.
Deeplearningrequiressignificantprocessingresources.
Deeplearningcanbeperformedefficientlybyusingagraphicsprocessingunit(GPU).
AlthoughmostdeeplearningframeworksalsosupportCPUprocessing,GPUprocessingprovidesreasonableperformanceforproductionenvironments.
OpeningretrainingserverportsBeforeusingtheretrainingserver,youmustopenthefirewallportsthatareusedbytheretrainingserver.
AboutthistaskIfuncomplicatedfirewall(UFW)isenabledandactiveontheretrainingserver,openUFWfornetworkfilesystem(NFS)serviceandopenthefollowingfirewallportsontheretrainingserver:v22v443v2001v2002v5001v5005v5060v5061v8449v9683InstallingNVIDIAGPUpackagesToenableGPUprocessing,youmustinstalltherequiredNVIDIAGPUpackages.
Procedure1.
DownloadandinstallthedriversforyourNVIDIAGPU.
TheNVIDIAdriverlistforUbuntuisavailableatthefollowinglink:BinaryDriverHowto-Nvidia.
2.
GettheNVIDIACUDA8.
0toolkitcuda-repo-ubuntu1604_8.
0.
61-1_amd64.
debfromtheIBMVisualInsightsArtifactInstaller1.
2Linuxx86-64Englishpackage.
Chapter2.
Installingtheproduct53.
InstalltheCUDAfileonthetargetserverusingthefollowingcommands:sudodpkg-icuda-repo-ubuntu1604_8.
0.
61-1_amd64.
debsudoapt-getupdatesudoapt-getinstallcuda4.
GettheNVIDIACUDADeepNeuralNetworklibrarycudnn-8.
0-linux-x64-v6.
0.
tgzfromtheIBMVisualInsightsArtifactInstaller1.
2Linuxx86-64Englishpackage.
5.
Unpackthecudnn-8.
0-linux-x64-v6.
0.
tgzfiletothecudainstallationdirectoryusingthefollowingcommand:sudotar-xvfcudnn-8.
0-linux-x64-v6.
0.
tgz-C/usr/local6.
Settheenvironmentvariableusingthefollowingcommand:exportLD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATHAlsoaddthiscommandtothe~/.
bashrcscript.
7.
InstalltheNVIDIANCCLpackageusingthefollowingcommands:gitclonehttps://github.
com/NVIDIA/nccl.
gitcdncclsudomakeinstall-j4InstallingCaffeYoumustinstalltheCaffedeep-learningframeworkandrelatedpackages.
Caffeisusedformodeltraininganddefectclassification.
Procedure1.
InstallthepackagesthatarerequiredforCaffebyusingthefollowingcommands:sudoapt-getupdatesudoapt-getupgradesudoapt-getinstall-ybuild-essentialcmakegitpkg-configsudoapt-getinstall-ylibprotobuf-devlibleveldb-devlibsnappy-devlibhdf5-serial-devprotobuf-compilersudoapt-getinstall-ylibatlas-base-devlibjasper-devsudoapt-getinstall-y--no-install-recommendslibboost-all-devsudoapt-getinstall-ylibgflags-devlibgoogle-glog-devliblmdb-devsudoapt-getinstall-ypython-pipsudoapt-getinstall-ypython-devsudoapt-getinstall-ypython-numpypython-scipysudoapt-getinstall-ylibopencv-dev2.
DownloadtheCaffesourcecodebyusingthefollowingcommand:wgethttps://github.
com/BVLC/caffe/archive/1.
0.
zip3.
Unpackthepackageandenterthepackagedirectorybyusingthefollowingcommands:unzip1.
0.
zipcd.
/caffe-1.
04.
Makeacopyofthemakeconfigurationfilebyusingthefollowingcommand:cpMakefile.
config.
exampleMakefile.
config5.
AddthefollowingvariablesintheMakefile.
configfile:USE_CUDNN:=1CUDA_DIR:=/usr/local/cuda-8.
0PYTHON_INCLUDE:=/usr/include/python2.
7\/usr/lib/python2.
7/dist-packages/numpy/core/includePYTHON_LIB:=/usr/lib/x86_64-linux-gnuWITH_PYTHON_LAYER:=1INCLUDE_DIRS:=$(PYTHON_INCLUDE)/usr/local/include/usr/include/hdf5/serial6UserGuideLIBRARY_DIRS:=$(PYTHON_LIB)/usr/local/lib/usr/lib/usr/lib/x86_64-linux-gnu/usr/lib/x86_64-linux-gnu/hdf5/serial6.
Inthecaffe-1.
0directory,runthefollowingcommand:find.
-typef-execsed-i-e's^"hdf5.
h"^"hdf5/serial/hdf5.
h"^g'-e's^"hdf5_hl.
h"^"hdf5/serial/hdf5_hl.
h"^g'7.
Runthefollowingcommands:cd/usr/lib/x86_64-linux-gnusudoln-slibhdf5_serial.
so.
10.
1.
0libhdf5.
sosudoln-slibhdf5_serial_hl.
so.
10.
0.
2libhdf5_hl.
so8.
InstalltherequiredPythonpackagesinthecaffe-1.
0/pythondirectorybyusingthefollowingcommands:cd{caffe-installation-path}/caffe-1.
0/pythonforreqin$(catrequirements.
txt);dosudo-Hpipinstall$req--upgrade;donewhere{caffe-installation-path}istheCaffedeploymentpath.
9.
Openthemakefileinthe{caffe-installation-path}directoryandchangetheparameterNVCCFLAGStothefollowingsetting:NVCCFLAGS+=-D_FORCE_INLINES-ccbin=$(CXX)-Xcompiler-fPIC$(COMMON_FLAGS)10.
InthemainCaffedirectorycaffe-1.
0,begintheCaffebuildandinstallationbyusingthefollowingcommands:makeallmaketestmakeruntestmakepycaffemakedistribute11.
Addthefollowinglinetothe~/.
bashrcscript:exportPYTHONPATH="/usr/lib/python2.
7:{caffe-installation-path}/caffe-1.
0/python:$PYTHONPATH"where{caffe-installation-path}istheCaffedeploymentpath.
ConfiguringpasswordlessauthenticationBeforeyouinstallVisualInsightsPrivate,youmustconfigurepasswordlessSSHauthenticationfromthecenterservertoallHortonworksDataPlatformnodesinyourenvironmentandtotheretrainingserver.
Procedure1.
LoginasroottotheserverthathostsVisualInsightsPrivate.
2.
GenerateanSSHauthenticationkeypair:ssh-keygen-trsa-b20483.
CopythepublickeytoanHortonworksDataPlatformnodeandeditthe/etc/sudoersfileforthatnode.
Thefollowingstepsusetheambarinodeasanexample:a.
Runthefollowingcommandtocopythepublickeytothenode:ssh-copy-id-i~/.
ssh/id_rsa.
pubroot@ambarib.
Runthefollowingcommandtologinwithoutusingstricthostkeychecking:ssh-o"StrictHostKeyCheckingno"root@ambaric.
Launchviwithsuperuserprivileges.
Ifyoucannotusevisudo,usesudo.
visudosud.
Changethepermissionsofthe/etc/sudoersfile:chmod640/etc/sudoersChapter2.
Installingtheproduct7e.
Editthe/etc/sudoersfile:vi/etc/sudoersf.
Findthefollowinglineinthefile:DefaultsrequirettyChangethislineto:#Defaultsrequirettyg.
Quitvi.
h.
Changethepermissionsofthe/etc/sudoersfile:chmod440/etc/sudoersRepeatthesestepstocopythepublickeytoallHortonworksDataPlatformnodesandtotheretrainingserver.
PreparingforinstallationCompletethefollowingstepsbeforeyouinstallVisualInsightsPrivate.
Procedure1.
OnthesystemonwhichyouwillinstallVisualInsightsPrivate,openthe/etc/hostsfile.
2.
EnsurethatyourHortonworksDataPlatformnodesarelistedinthefile,forexample:127.
0.
0.
1localhost.
localdomainlocalhosthdpmgmt01.
domain.
comhdpmgmt01hdpmgmt02.
domain.
comhdpmgmt02hdpslave01.
domain.
comhdpslave01hdpslave02.
domain.
comhdpslave02hdpslave03.
domain.
comhdpslave033.
Saveandclosethefile.
4.
AccesstheAmbariwebuserinterfacefromawebbrowserbyusingtheservername(thefullyqualifieddomainname)onwhichyouinstalledtheAmbariserver,andport8080.
Forexample,enterthefollowingstringinyourbrowser:HTTP://node1.
example.
com:80805.
SelectAddNewHostsandgototheInstallOptionspage.
InTargetHosts,listthenodewhereyouwillinstallVisualInsightsPrivate.
6.
InHostRegistrationInformation,selectoneofthefollowingoptions:vProvideyourSSHPrivateKeytoautomaticallyregisterthehost:ClickSSHPrivateKey.
Theprivatekeyfileis/root/.
ssh/id_rsa,iftherootuserinstalledtheAmbariserver.
Ifyouinstalledasanon-rootuser,thenthedefaultprivatekeyisinthe.
sshdirectoryinthenon-rootuser'shomedirectory.
ClickRegisterandConfirm.
vRegisterthehostmanuallyanddonotuseSSH.
7.
OneachHortonworksDataPlatformmanagementnodeandslavenode,createOS-levelusersandgroupsbyusingthefollowingcommands:groupaddviuseradd-gvi-s/bin/bashviecho"vi:pw@vi"|chpasswd8.
IfKerberosisenabledintheHortonworksDataPlatformcluster,youmustcreateaprincipalforVisualInsightsPrivateintheKerberosKeyDistributionCenter(KDC),generatearelatedkeytabfile,anddistributethekeytabfiletoallofHortonworksDataPlatformclusternodesandtheVisualInsightsPrivatenode.
8UserGuidea.
OnthenodewheretheKDCisinstalled,runthefollowingcommands:kadmin.
localkadmin.
local:addprinc-pwpw@vivikadmin.
local:ktadd-k/etc/security/keytabs/vi.
keytab-norandkeyvikadmin.
local:exitb.
Copythe/etc/security/keytabs/covia.
keytabfiletothe/etc/security/keytabsdirectoryonallofHortonworksDataPlatformclusternodesandtheVisualInsightsPrivatenode.
9.
IfKerberosisenabledintheHortonworksDataPlatformcluster,makesurethatKafkaisusingaPLAINTEXTendpoint:a.
AccesstheAmbariwebuserinterface.
b.
SelecttheKafkaservice.
c.
OpenConfigurations.
d.
InCustomkafka-broker,checkwhetherthevalueofthesecurity.
inter.
broker.
protocolisPLAINTEXT.
IfthevalueisnotPLAINTEXT,updatethevalue,save,andthenrestartaffectedservice.
10.
IfKerberosisenabledintheHortonworksDataPlatformcluster,addanewcustomkafkabrokerproperty:a.
AccesstheAmbariwebuserinterface.
b.
SelecttheKafkaservice.
c.
OpenConfigurations.
d.
InCustomkafka-broker,clicktheAddPropertylink,copyallow.
everyone.
if.
no.
acl.
found=trueintothepropertiesarea,thenclickAdd.
e.
Restarttheaffectedservice.
11.
Installthenfs-utilspackagetoamountedsharedfolderoftheimageserverusingthefollowingcommand:yum-yinstallnfs-utilsInstallingtheserverPerformthistasktoinstalltheservercomponent.
Procedure1.
DownloadtheServerInstallation.
tar.
gz,iotmf_server_1.
1_l86-64_en.
tar.
gz,andArtifactsInstallation.
tar.
gzfilestothesystemthatyouwillusefortheserver.
YoucangetServerInstallation.
tar.
gzfromtheIBMIOTforManufacturingFoundationServerInstaller1.
1Linuxx86-64Englishpackage.
Youcangetiotmf_server_1.
1_l86-64_en.
tar.
gzfromtheIBMIOTforManufacturingFoundationServer1.
1Linuxx86-64Englishpackage.
YoucangetArtifactsInstallation.
tar.
gzfromtheIBMVisualInsightsArtifactInstaller1.
2Linuxx86-64Englishpackage.
2.
CopytheServerInstallation.
tar.
gzfiletotheinstallationdirectory(VI_install_dir)anddecompressthefile.
TheServerInstallationdirectoryiscreated.
3.
CopytheArtifactsInstallation.
tar.
gzfiletotheVI_install_dirdirectoryanddecompressthefile.
4.
Copytheiotmf_server_1.
1_l86-64_en.
tar.
gzfiletotheVI_install_dir/ServerInstallation/SolutionInstaller/NodeRoot/Downloads/Softwaredirectory.
Chapter2.
Installingtheproduct95.
Runthefollowingscripttodecompressthefiles:shIOTMF_1.
1_UncompressTarball.
sh6.
Tosetuptheinstaller,gototheinstallersetupfolderVI_install_dir/ServerInstallation/SolutionInstallerandthenrunthefollowingscript:.
/setup.
sh7.
AccesstheinstallerbygoingtothefollowingURL:https://hostname:8080/UI/index.
htmlwherehostnameisthehostnameforinstaller.
ReviewthelicenseinformationandthenclickAccept.
8.
ClickPredefinedConfigurationtoloadapredefinedconfigurationthatincludesonenode,WebSphereApplicationServerLiberty,ProvisioningConsole,IBMJDKArtifact,andServiceFramework.
9.
ChangethevaluesinthePropertyEditortotheconfigurationofyourserver.
10.
ClickValidatetovalidatetheconfiguration.
11.
ClickRuntoruntheinstallation.
12.
Intheinstallersetupfolder,runthefollowingscripttocleanuptheinstallationfiles:.
/cleanup.
sh13.
Restarttheserver.
InstallingtheartifactsPerformthistasktoinstalltheproductartifacts.
BeforeyoubeginBeforeinstallingtheproductartifacts,youmustcreatetheretrainingserver.
Procedure1.
Tosetuptheinstaller,gototheinstallersetupfolderandthenrunthefollowingscript:.
/setup.
sh2.
AccesstheinstallerbygoingtothefollowingURL:https://hostname:8080/UI/index.
htmlwherehostnameisthehostnameforinstaller.
ReviewthelicenseinformationandthenclickAccept.
3.
ClickPredefinedConfigurationtoloadapredefinedconfigurationthatincludesonenodeandtheVIArtifactcomponent.
4.
ChangethevaluesinthePropertyEditortotheconfigurationofyourserver.
vClusternamereferstotheclusternameofHortonworksDataPlatform.
YoucanfindtheclusternameintheHortonworksDataPlatformadminconsole,oryoucanclickAdmin>ManageAmbariandcheckClusters.
vAmbarihostnamereferstothenameofhostwheretheAmbariserverisinstalled.
TheartifactinstallerconnectstotheAmbariservertogetHortonworksDataPlatformservicestatusandconnectioninformation.
IfKerberosisenabledinHortonworksDataPlatform,setKerberosissetassecurityauthenticationtoTrue.
vThehbasekeytabfileandhdfskeytabfilearefilesinthe/etc/security/keytabsfolder.
5.
ClickValidatetovalidatetheconfiguration.
6.
ClickRuntoruntheinstallation.
10UserGuide7.
Intheinstallersetupfolder,runthefollowingscripttocleanuptheinstallationfiles:.
/cleanup.
shPost-installationtasksCompletethesetasksafteryouinstallVisualInsightsPrivate.
InstallingobjectdetectionlibrariesontheretrainingserverAfteryouinstalltheartifacts,installanobjectdetectionlibraryontheretrainingserver.
Theobjectdetectionlibraryenablesyoutovalidatemodelsontheretrainingserver.
AboutthistaskIBMVisualInsightsPrivatesupportsthefollowingobjectdetectionlibraries:YOLO(youonlylookonce),FasterR-CNN,andSSD(SingleShotMultiBoxDetector).
Procedure1.
InstalltherelatedPythonpackagesbyusingthefollowingcommands:sudoapt-getinstallpython-numpysudoapt-getinstallpython-scipypipinstallcythonpipinstalleaydictpipinstallshutilpipinstallcPicklepipinstalluuidpipinstallmultiprocessingpipinstallxml2.
Installoneofthefollowinglibraries:LibraryInstallationInstructionsYOLOversion2library1.
RunthefollowingcommandtogettheYOLOsourcecode:gitclone--recursivehttp://github.
com/pjreddie/darknet.
git2.
EdittheMakefilefilebyenablingtheGPUandCUDNNparameters,andselectthecorrectGPUARCHparameteraccordingtoyourmachineconfiguration.
3.
RunthefollowingcommandtocompileYOLO:makeFasterR-CNNPythonlibrary1.
RunthefollowingcommandtogettheFasterR-CNNsourcecode:gitclone--recursivehttp://github.
com/rbgirshick/py-faster-rcnn.
git2.
Inthepv-faster-rcnndirectory,inthelibfolder,runthefollowingcommandtoCompileCython:make3.
CompileCaffeandpycaffeinsidethepy-faster-rcnnfolder.
YoumustcompileCaffebyusingthePythonlayer.
Chapter2.
Installingtheproduct11LibraryInstallationInstructionsSSDlibrary1.
RunthefollowingcommandtogettheSSDsourcecode:gitclone--recursivehttps://github.
com/weiliu89/caffe.
git2.
EdittheMakefilefileandchangetheCUDA_ARCH,BLAS,andPYTHON_INCLUDEparametersaccordingtoyourmachineconfiguration.
3.
Compilethecodebyusingfollowingcommand:make-j84.
CompilethePythonlayerbyusingthefollowingcommand:makepy5.
Compilethetestbyusingthefollowingcommand:maketest-j83.
Addthefollowingenvironmentvariablesattheuserlevel:YOLO_HOME,FRCNN_HOME,andSSD_HOME.
Thefollowingtextisanexampleofaddingenvironmentvariables:YOLO_HOME=/home/user/darknet/,FRCNN_HOME=/home/user/py-faster-rcnn/.
4.
Optional:Manuallydeploytheobjectdetectionlibrary.
a.
Unpackagetheobjectdetectionpackage.
b.
Addthe/home/pmqopsadmin/vi_obj_detection/model_library/yolo/detectorobj.
cfiletothedarknet/examplesfolder.
c.
Editthedarknet/MakefilefileandindicatethatEXECOBJA=detectorobj.
o.
ThefollowingcodeisanexampleofthecodeintheMakefilefile:EXECOBJA=detectorobj.
ocaptcha.
olsd.
osuper.
ovoxel.
oart.
otag.
ocifar.
ogo.
ornn.
ornn_vid.
ocompare.
osegmenter.
oregressor.
oclassifier.
ococo.
odice.
oyolo.
odetector.
owriting.
onightmare.
oswag.
odarknet.
od.
IntheMakefilefile,add$(EXECOBJ)forthe$(SLIB)and$(ALIB)objects.
ThefollowingcodeisanexampleofthecodeintheMakefilefile:$(ALIB):$(EXECOBJ)$(OBJS)$(AR)$(ARFLAGS)$@$^$(SLIB):$(EXECOBJ)$(OBJS)$(CC)$(CFLAGS)-shared$^-o$@$(LDFLAGS)e.
Runthefollowingcommand:makef.
Copytheiotmyolo.
pyfileandaddittoYOLO_HOMEdirectory.
5.
RunthefollowingcommandstostartFasterR-CNNandSSD:cd/home/pmqopsadmin/vi_obj_detection/RESTAPI/model/nohuppythonrun.
py50601FRCNN>frcnn_log.
txt2>&1&nohuppythonrun.
py50611SSD>ssd_log.
txt2>&1&Startingtheself-learningengineontheretrainingserverAfteryouinstalltheartifacts,starttheself-learningenginesothatyoucanretrainmodelsontheretrainingserver.
Procedure1.
Ontheretrainingserver,changetotheself-learningserverdirectorybyusingthefollowingcommand:cd/home/pmqopsadmin/selflearning/2.
Runthefollowingcommands:source~/.
bashrcnohuppythonservices/run.
py>run_log.
txt2>&1&12UserGuidenohuppythonselflearning/runScheduler.
py>runScheduler_log.
txt2>&1&nohuppythonselflearning/runJobRunner.
py>runJobRunner_log.
txt2>&1&Creatingsystem-levelusersBeforeyoufirstlogintotheproduct,youmustcreateoperatingsystem-levelusersandgroupsoneachHortonworksDataPlatformmanagementnodeandslavenode.
Procedure1.
RunthefollowingcommandoneachHortonworksDataPlatformmanagementnodeandslavenode:useradd-gvi-s/bin/bash-ppw@VIVI2.
IfKerberosisenabledintheHortonworksDataPlatformcluster,distributethe/etc/security/keytabs/VI.
keytabkeytabfilefromtheAmbarinodetoallotherHortonworksDataPlatformnodes.
LogginginYoucanlogintotheproductbyaccessingtheWebinterface.
BeforeyoubeginBeforeyoulogintotheproduct,makesuretheLibertyserverprovisioningconsoleisvalid.
Procedure1.
Tologintothesystem,gotothefollowingURL:https://:9447/ibm/iotm/solution/2.
Enteryourusernameandpassword.
DefaultusersVisualInsightsPrivatehastwodefaultusersthatyoucanusetoaccesstheproduct.
Thefollowingusersareprovidedwiththeproduct.
Table2.
DefaultusersUserRoledemouserInspectorSupervisordemomgrModelManager,DataScientistYoucanchangethepasswordsfortheseusersbyeditingtheserver.
xmlfileintheVisualInsightsPrivateserverinstallationfolder.
ProductlicensefilesAfteryoumanuallyinstallIBMVisualInsightsPrivate,youmustensurethattheappropriateswidtagfileandlicensedirectoryexistsoneachcomputeronwhichyouinstalledaVisualInsightsPrivatecomponent.
Chapter2.
Installingtheproduct13LicensefilesTheswidtagfileandlicensedirectorymustexistontheVisualInsightsPrivatenodecomputer.
Note:Donotrenamethedirectoryorfiles.
IfyouinstalledonaRedHatEnterpriseLinuxoperatingsystembyusingthesolutioninstaller,thelicensefilesareautomaticallycopiedto/opt/IBM/VI_1.
2/licenseandtheswidtagfilesareautomaticallycopiedto/opt/IBM/VI_1.
2/iso-swidontheVisualInsightsPrivatenode.
ApplicationlicensingandtheslmtagfileAlicensingapplicationrunsontheVisualInsightsPrivateIntegrationBusnode.
TheapplicationperiodicallylogsthenumberofimagesthatareclassifiedbytheVisualInsightsPrivatesystemintoafilewiththe.
slmtagextensioninthe/opt/IBM/VI_1.
2folder.
Hereisasampleslmtagfilewiththefollowingcode.
2.
1.
13ad60a8beaf2496a8b940813bba7417dIBMVisualInsightsPrivate/IBM/IOTM/VIQIMAGENO_IMAGES22017-06-15T06:00:33+08:002017-06-16T06:00:33+08:00TroubleshootingTroubleshootingtopicscanhelpyouidentifyandrecoverfromcommonissues.
Servervalidationfailswithcheck_requiretty.
sherrorValidationfailsduringtheserverinstallation,andthelogfileshowsanerroraboutcheck_requiretty.
sh.
AboutthistaskThisissueoccursbecauserequirettyisenabledonthetargetmachine.
Completethefollowingstepstocommentouttherequirettylineinthe/etc/sudoersfileonthetargetsystem.
Procedure1.
Runthefollowingcommand:visudo2.
Commentouttherequirettylineasfollows:#Defaultsrequiretty3.
Savethefileandexit.
4.
Retrytheserverinstallationandvalidation.
14UserGuidesetup.
shfailswithUS-ASCIIerrorOnaMacBook,setup.
shfailsandthelogfilecontainsthefollowingmessage:ERROR:ArgumentError:invalidbytesequenceinUS-ASCII.
AboutthistaskThiserroroccursbecausethedefaultencodingofaMacBookisUS-ASCIIratherthanUTF-8.
CompletethefollowingstepstochangetheencodingofaMacBooktoUTF-8.
Procedure1.
RunthefollowingcommandontheMacBook:vi~/.
bash_profile2.
Addthefollowinglinestothefile:exportLC_ALL=en_US.
UTF-8exportLANG=en_US.
UTF-83.
Savethefile.
4.
Runthefollowingcommand:source~/.
bash_profile5.
Runthefollowingcommand:echo$LANGIftheresultisen_US.
UTF-8,theissueshouldberesolved.
ArtifactinstallationfailswithcreateSysUser.
sherrorArtifactinstallationfailsandthelogfileshowsthefollowingmessage:ERROR:passwd:Haveexhaustedmaximumnumberofretriesforservice.
AboutthistaskThisissueoccursbecauseenforce_for_rootisenabledonthepam_pwquality.
somoduleonthetargetmachine.
Completethefollowingstepstoremoveenforce_for_rootfromthe/etc/pam.
d/system-authfile.
Procedure1.
Runthefollowingcommand:vi/etc/pam.
d/system-auth2.
Removeenforce_for_rootfromthefollowingline:passwordrequisitepam_pwquality.
soXXX3.
Savethefileandexit.
4.
Retrytheartifactinstallationandruntheinstall.
Chapter2.
Installingtheproduct1516UserGuideChapter3.
CreatingedgesystemsVisualInsightsPrivateconsistsofthecenterapplicationandtheedges.
EdgesareLinuxsystemsthatareusedtoperformruntimedefectdetection.
EdgesystemsusetheCaffedeep-learningframework.
Caffeisadedicatedartificialneuralnetwork(ANN)trainingenvironment.
Deeplearningrequiressignificantprocessingresources.
Deeplearningcanbeperformedefficientlybyusingagraphicsprocessingunit(GPU).
AlthoughmostdeeplearningframeworksalsosupportCPUprocessing,GPUprocessingprovidesreasonableperformanceforproductionenvironments.
Edgesareclusteredforloadbalancing.
Aclusterconsistsofonemasteredgeandmultipleslaveedges.
Whenacellsendsanimageforprocessing,themasteredgereceivestheimageandthensendsittoaslaveedgethathasGPUresourcesavailable.
Theedgeclusteringarchitectureisscalable.
Ifyouneedmoreprocessingresources,youcanaddmoreedges.
Whenyoucreateanedge,youspecifywhethertheedgeisthemasteredgeoraslaveedge.
Thefirstedgethatyoucreatemustbethemasteredge.
Afteranedgeiscreated,youcannotchangetheedgetype.
Youcancreateonlyonemasteredgepertenant.
Themasteredgecannotbedeleteduntilalltheslaveedgesaredeleted.
Masteredgescanbeonlineoroffline.
Onlinemeanswhentheimageisscoredonanedge,theinspectionresultissenttothecenterapplicationimmediately.
OfflinemeanstheinspectionresultisstoredontheedgeuntilyouselectPullresultsfromthePreviewEdgedialogboxorcalltheservicethatisdeployedontheedge.
OpeningedgeportsBeforeusingedges,youmustopenthefirewallportsthatareusedbytheedgesystems.
AboutthistaskIfuncomplicatedfirewall(UFW)isenabledandactiveontheedges,openUFWfornetworkfilesystem(NFS)serviceandopenthefollowingfirewallportsontheedges:v22v5005v5060-5090v6005v8449InstallingNVIDIAGPUpackagesToenableGPUprocessing,youmustinstalltherequiredNVIDIAGPUpackages.
CopyrightIBMCorp.
201717Procedure1.
DownloadandinstallthedriversforyourNVIDIAGPU.
TheNVIDIAdriverlistforUbuntuisavailableatthefollowinglink:BinaryDriverHowto-Nvidia.
2.
GettheNVIDIACUDA8.
0toolkitcuda-repo-ubuntu1604_8.
0.
61-1_amd64.
debfromtheIBMVisualInsightsArtifactInstaller1.
2Linuxx86-64Englishpackage.
3.
InstalltheCUDAfileonthetargetserverusingthefollowingcommands:sudodpkg-icuda-repo-ubuntu1604_8.
0.
61-1_amd64.
debsudoapt-getupdatesudoapt-getinstallcuda4.
GettheNVIDIACUDADeepNeuralNetworklibrarycudnn-8.
0-linux-x64-v6.
0.
tgzfromtheIBMVisualInsightsArtifactInstaller1.
2Linuxx86-64Englishpackage.
5.
Unpackthecudnn-8.
0-linux-x64-v6.
0.
tgzfiletothecudainstallationdirectoryusingthefollowingcommand:sudotar-xvfcudnn-8.
0-linux-x64-v6.
0.
tgz-C/usr/local6.
Settheenvironmentvariableusingthefollowingcommand:exportLD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATHAlsoaddthiscommandtothe~/.
bashrcscript.
7.
InstalltheNVIDIANCCLpackageusingthefollowingcommands:gitclonehttps://github.
com/NVIDIA/nccl.
gitcdncclsudomakeinstall-j4InstallingCaffeYoumustinstalltheCaffedeep-learningframeworkandrelatedpackages.
Caffeisusedformodeltraininganddefectclassification.
Procedure1.
InstallthepackagesthatarerequiredforCaffebyusingthefollowingcommands:sudoapt-getupdatesudoapt-getupgradesudoapt-getinstall-ybuild-essentialcmakegitpkg-configsudoapt-getinstall-ylibprotobuf-devlibleveldb-devlibsnappy-devlibhdf5-serial-devprotobuf-compilersudoapt-getinstall-ylibatlas-base-devlibjasper-devsudoapt-getinstall-y--no-install-recommendslibboost-all-devsudoapt-getinstall-ylibgflags-devlibgoogle-glog-devliblmdb-devsudoapt-getinstall-ypython-pipsudoapt-getinstall-ypython-devsudoapt-getinstall-ypython-numpypython-scipysudoapt-getinstall-ylibopencv-dev2.
DownloadtheCaffesourcecodebyusingthefollowingcommand:wgethttps://github.
com/BVLC/caffe/archive/1.
0.
zip3.
Unpackthepackageandenterthepackagedirectorybyusingthefollowingcommands:unzip1.
0.
zipcd.
/caffe-1.
04.
Makeacopyofthemakeconfigurationfilebyusingthefollowingcommand:cpMakefile.
config.
exampleMakefile.
config18UserGuide5.
AddthefollowingvariablesintheMakefile.
configfile:USE_CUDNN:=1CUDA_DIR:=/usr/local/cuda-8.
0PYTHON_INCLUDE:=/usr/include/python2.
7\/usr/lib/python2.
7/dist-packages/numpy/core/includePYTHON_LIB:=/usr/lib/x86_64-linux-gnuWITH_PYTHON_LAYER:=1INCLUDE_DIRS:=$(PYTHON_INCLUDE)/usr/local/include/usr/include/hdf5/serialLIBRARY_DIRS:=$(PYTHON_LIB)/usr/local/lib/usr/lib/usr/lib/x86_64-linux-gnu/usr/lib/x86_64-linux-gnu/hdf5/serial6.
Inthecaffe-1.
0directory,runthefollowingcommand:find.
-typef-execsed-i-e's^"hdf5.
h"^"hdf5/serial/hdf5.
h"^g'-e's^"hdf5_hl.
h"^"hdf5/serial/hdf5_hl.
h"^g'7.
Runthefollowingcommands:cd/usr/lib/x86_64-linux-gnusudoln-slibhdf5_serial.
so.
10.
1.
0libhdf5.
sosudoln-slibhdf5_serial_hl.
so.
10.
0.
2libhdf5_hl.
so8.
InstalltherequiredPythonpackagesinthecaffe-1.
0/pythondirectorybyusingthefollowingcommands:cd{caffe-installation-path}/caffe-1.
0/pythonforreqin$(catrequirements.
txt);dosudo-Hpipinstall$req--upgrade;donewhere{caffe-installation-path}istheCaffedeploymentpath.
9.
Openthemakefileinthe{caffe-installation-path}directoryandchangetheparameterNVCCFLAGStothefollowingsetting:NVCCFLAGS+=-D_FORCE_INLINES-ccbin=$(CXX)-Xcompiler-fPIC$(COMMON_FLAGS)10.
InthemainCaffedirectorycaffe-1.
0,begintheCaffebuildandinstallationbyusingthefollowingcommands:makeallmaketestmakeruntestmakepycaffemakedistribute11.
Addthefollowinglinetothe~/.
bashrcscript:exportPYTHONPATH="/usr/lib/python2.
7:{caffe-installation-path}/caffe-1.
0/python:$PYTHONPATH"where{caffe-installation-path}istheCaffedeploymentpath.
TroubleshootingtheCaffeinstallationIfanerrormessagedisplaysinthelogwhenyoubegintheCaffebuildandinstallation,youcantakestepstotrytoresolvetheproblem.
SymptomsWhenyoubegantheCaffebuildandinstallation,thefollowingmessagedisplays:1.
Infileincludedfrom.
/include/caffe/util/device_alternate.
hpp:40:0,2.
from.
/include/caffe/common.
hpp:19,3.
fromsrc/caffe/common.
cpp:7:4.
.
/include/caffe/util/cudnn.
hpp:Infunction'voidcaffe::cudnn::createPoolingDesc(cudnnPoolingStruct**,caffe::PoolingParameter_PoolMethod,cudnnPoolingMode_t*,int,int,int,int,int,int)':5.
.
/include/caffe/util/cudnn.
hpp:127:41:error:toofewargumentstofunction'cudnnStatus_tcudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t,cudnnPoolingMode_t,cudnnNanPropagation_t,int,int,int,int,int,int)'6.
pad_h,pad_w,stride_h,stride_w));Chapter3.
Creatingedgesystems197.
^8.
.
/include/caffe/util/cudnn.
hpp:15:28:note:indefinitionofmacro'CUDNN_CHECK'9.
cudnnStatus_tstatus=condition;\10.
^11.
Infileincludedfrom.
/include/caffe/util/cudnn.
hpp:5:0,12.
from.
/include/caffe/util/device_alternate.
hpp:40,13.
from.
/include/caffe/common.
hpp:19,14.
fromsrc/caffe/common.
cpp:7:15.
/usr/local/cuda-7.
5//include/cudnn.
h:803:27:note:declaredhere16.
cudnnStatus_tCUDNNWINAPIcudnnSetPooling2dDescriptor(17.
^18.
make:***[.
build_release/src/caffe/common.
o]Error119.
ResolvingtheproblemTofixtheerror,refertothefollowingsteps:1.
Inthe/include/caffe/util/cudnn.
hppdirectory,replacethecudnn.
hppfilewiththenewestcudnn.
hppfilethatisintheCafferepositoryonGitHub.
2.
Inthe/src/caffe/layersfolder,replaceallofthecudnnfilesthatareinthe/src/caffe/layersfolderwiththenewestcudnnfilesthatareintheCafferepositoryonGitHub.
InstallingOpenCVYoucaninstalltheOpenSourceComputerVision(OpenCV)3.
2libraryifyouneedtocustomizemodelswiththefeaturesofOpenCV3.
2.
Procedure1.
GettheOpenCVsourcecodefromGithub:wgethttps://github.
com/opencv/opencv/archive/3.
2.
0.
zip2.
Unpackthedownloadedpackageandchangetothepackagedirectory:unzip3.
2.
0.
zipcdopencv-3.
2.
03.
Createabuildingsubdirectoryandchangetothedirectory:mkdirbuildcdbuild4.
Prepareandgeneratethebuildingconfiguration:cmake-DCMAKE_BUILD_TYPE=RELEASE-DCMAKE_INSTALL_PREFIX=/usr/local-DWITH_TBB=ON-DWITH_V4L=ON.
.
5.
Compileandbuildthepackage:make–j$(nproc)6.
Installthepackage:sudomakeinstall7.
Registerthelibrariesandmodulestothesystem:sudo/bin/bash-c'echo"/usr/local/lib">/etc/ld.
so.
conf.
d/opencv.
conf'sudoldconfig8.
Ifrequired,uninstalltheoldopencvversiontoavoidversioncollision:sudoapt-getautoremovelibopencv-devInstallingobjectdetectionlibrariesYouinstallanobjectdetectionlibrarysothatyoucanruntheobjectdetectionmodelonanedge.
20UserGuideAboutthistaskIBMVisualInsightsPrivatesupportsthefollowingobjectdetectionlibraries:YOLO(youonlylookonce),FasterR-CNN,andSSD(SingleShotMultiBoxDetector).
Procedure1.
InstalltherelatedPythonpackagesbyusingthefollowingcommands:sudoapt-getinstallpython-numpysudoapt-getinstallpython-scipypipinstallcythonpipinstalleasydictpipinstallshutilpipinstallcPicklepipinstalluuidpipinstallmultiprocessingpipinstallxml2.
Installoneofthefollowinglibraries:LibraryInstallationInstructionsYOLOversion2library1.
RunthefollowingcommandtogettheYOLOsourcecode:gitclone--recursivehttps://github.
com/pjreddie/darknet.
git2.
EdittheMakefilefilebyenablingtheGPUandCUDNNparameters,andselectthecorrectGPUARCHparameteraccordingtoyourmachineconfiguration.
3.
RunthefollowingcommandtocompileYOLO:makeFaster-RCNNPythonlibrary1.
RunthefollowingcommandtogettheFasterR-CNNsourcecode:gitclone--recursivehttps://github.
com/rbgirshick/py-faster-rcnn.
git2.
Inthepv-faster-rcnndirectory,inthelibfolder,runthefollowingcommandtoCompileCython:make3.
CompileCaffeandpycaffeinsidethepy-faster-rcnnfolder.
YoumustcompileCaffebyusingthePythonlayer.
SSDlibrary1.
RunthefollowingcommandtogettheSSDsourcecode:gitclone--recursivehttps://github.
com/weiliu89/caffe.
git~/ssd-caffecd~/ssd-caffegitcheckoutssd2.
EdittheMakefilefileandchangetheCUDA_ARCH,BLAS,andPYTHON_INCLUDEparametersaccordingtoyourmachineconfiguration.
3.
Compilethecodebyusingfollowingcommand:make-j84.
CompilethePythonlayerbyusingthefollowingcommand:makepy5.
Compilethetestbyusingthefollowingcommand:maketest-j8Chapter3.
Creatingedgesystems213.
Addthefollowingenvironmentvariablesattheuserlevel:YOLO_HOME,FRCNN_HOME,andSSD_HOME.
Thefollowingtextisanexampleofaddingenvironmentvariables:YOLO_HOME=~/darknet/FRCNN_HOME=~/py-faster-rcnn/SSD_HOME=~/ssd-caffe/.
4.
Optional:Manuallydeploytheobjectdetectionlibrary.
a.
Unpackagetheobjectdetectionpackage.
b.
Addtheyolo/detectorobj.
cfiletothedarknet/examplesfolder.
c.
Editthedarknet/MakefilefileandindicatethatEXECOBJA=detectorobj.
o.
ThefollowingcodeisanexampleofthecodeintheMakefilefile:EXECOBJA=detectorobj.
ocaptcha.
olsd.
osuper.
ovoxel.
oart.
otag.
ocifar.
ogo.
ornn.
ornn_vid.
ocompare.
osegmenter.
oregressor.
oclassifier.
ococo.
odice.
oyolo.
odetector.
owriting.
onightmare.
oswag.
odarknet.
od.
IntheMakefilefile,add$(EXECOBJ)forthe$(SLIB)and$(ALIB)objects.
ThefollowingcodeisanexampleofthecodeintheMakefilefile:$(ALIB):$(EXECOBJ)$(OBJS)$(AR)$(ARFLAGS)$@$^$(SLIB):$(EXECOBJ)$(OBJS)$(CC)$(CFLAGS)-shared$^-o$@$(LDFLAGS)e.
Runthefollowingcommand:makef.
Copytheiotmyolo.
pyfileandaddittoYOLO_HOMEdirectory.
ConfiguringtheimageserverYouconfiguretheimageserveronthemasteredgemachinesothatthemasteredgemachinecanstoreimagesthatarecapturedbyanindustrialcamera.
Theimageserverismonitoredbytheedgecontroller.
Whenanewimageisadded,itisscoredandtheinspectionresultissenttothecenterapplicationsothataninspectorcanevaluatetheimageandtheinspectionresults.
BeforeyoubeginAddthefollowinglinetoyourexportfile:/imageserver"IPaddress"(rw,sync,no_root_squash,no_all_squash).
Forexample,youwouldaddthefollowinglinetothefile:/imageserver10.
173.
0.
0/29(rw,sync,no_root_squash,no_all_squash).
Procedure1.
Installandstartthenfsservicebyusingthefollowingcommand:#>apt-getinstallnfs-kernel-server2.
Exporttheimageserverfolderwiththefollowingcommands:#>mkdir/imageserver#vi/etc/exports#>servicenfs-serverrestart3.
Oneachslaveedgemachine,mountthe/imageserverdirectorytothesharedfolderofthemasteredge:a.
Runthefollowingcommandontheslaveedgemachine:mkdir/imageserverb.
Editthe/etc/fstabfileusingthefollowingcommand:vi/etc/fstab22UserGuidec.
Addthefollowinglinetothefile,specifyingtheIPaddressofthemasteredge:IP_address:/imageserver/imageservernfsdefaults00d.
Runthefollowingcommandontheslaveedgemachine:mount-aConfiguringthemodelstoreYoumustconfigurethemodelstoreonthemasteredgemachine.
Themodelstoreisarepositoryforexecutablemodelsthataredistributedfromthecenterapplication.
BeforeyoubeginAddthefollowinglinetoyourexportfile:/modelstore"IPaddress"(rw,sync,no_root_squash,no_all_squash).
Forexample,addthefollowinglinetothefile:/modelstore10.
173.
0.
0/29(rw,sync,no_root_squash,no_all_squash).
Procedure1.
Exportthemodelstorefolderbyusingthefollowingcommands:#>mkdir/modelstore#vi/etc/exports#>servicenfs-serverrestart2.
Runthefollowingcommands:sudopipinstallflasksudopipinstallgeventsudopipinstallrequestssudopipinstallpyinotifysudopipinstallopencv-pythonsudopipinstalllmdbsudoapt-getinstalldos2unixsudopipinstallparamikosudopipinstallsqlite33.
Oneachslaveedgemachine,mountthe/modelstoredirectorytothesharedfolderofmasteredge:a.
Runthefollowingcommandontheslaveedgemachine:mkdir/modelstoreb.
Editthe/etc/fstabfileusingthefollowingcommand:vi/etc/fstabc.
Addthefollowinglinetothefile,specifyingtheIPaddressofthemasteredge:IP_address:/modelstore/modelstorenfsdefaults00d.
Runthefollowingcommandontheslaveedgemachine:mount-aRegisteringtheedgetothecenterapplicationAfteryouconfiguretheedgemachine,youmustregisteritinthecenterapplication.
Youcancreateanedgeoreditanexistingone.
Edgesareusedtorunascoringmodel.
Chapter3.
Creatingedgesystems23Procedure1.
IntheModelManager,selectData>Edges.
2.
SelectCreatenewedgeandinputtheedgename.
3.
InputtheIPaddress,theSSHusername,andthepasswordoftheedgemachine,andspecifytheedgetypeasMasterorSlave.
TheIPaddressmustbeaccessibletothecenterapplication.
TheSSHusernameandpasswordareusedtologintheedgemachinetodeploytheedgecontroller.
4.
Ifyouarecreatingamasteredge,specifyonlineorofflineforconnectionmode.
5.
SelectOktocreatetheedge.
IftheIPaddress,SSHusername,orpasswordisnotcorrectandanerrormessageshows,makesurethatthevaluesarecorrectandthattheedgemachinecanbeconnected.
6.
SelectCreate.
Theedgecontrollerandscoreenginedeploytotheedgemachine,andtheedgeisaddedintotheregisteredlist.
7.
PerformanSSHlogintotheedgemachine.
8.
Onthemasteredgemachine,runthefollowingcommandtolookforthePythonprocessforthemasterengine:psaux|greppythonTheresultshouldincludethefollowingPythonprocess:pythondeployment_folder/vi_edge-bin_vi/vi_task_manager/run.
py.
IfyoudonotfindthePythonprocessonthemasteredgemachine,checkthelogfilesatdeployment_folder/vi_edge-bin_vi/vi_task_manager/master.
log.
Trytostartthemasterenginebyrunningthefollowingcommand:deployment_folder/vi_edge-bin_vi/vi_task_manager/restartMaster.
shIP_of_master_node9.
Onthemasteredgemachine,runthefollowingcommandtolookforthePythonprocessfortheedgecontroller:psaux|greppythonTheresultshouldincludethefollowingPythonprocess:pythondeployment_folder/vi_edge-bin_vi/vi_edge/edgemonitor.
py.
IfyoudonotfindthePythonprocessonthemasteredgemachine,checkthelogfilesatdeployment_folder/vi_edge-bin_vi/vi_edge/Event.
log.
Trytostarttheedgecontrollerbyrunningthefollowingcommand:deployment_folder/vi_edge-bin_vi/vi_edge/restartController.
sh10.
RunthefollowingcommandonthemasteredgemachineandtheslaveedgemachinestolookforPythonprocessesfortheclassificationmodel:psaux|greppythonTheresultshouldincludethefollowingPythonprocesses:pythondeployment_folder/vi_edge-bin_vi/vi_score_engine_restful/front.
py50056005pythondeployment_folder/vi_edge-bin_vi/vi_score_engine_restful/back.
py6005IfyoudonotfindthePythonprocessontheedgemachines,checkthelogfiles:deployment_folder/vi_edge-bin_vi/vi_edge/front_log.
txtdeployment_folder/vi_edge-bin_vi/vi_edge/back_log.
txtAfteryoufixanyissues,themasterengineshouldstartthescoreengineautomatically.
Tostartthescoreenginemanually,runthefollowingcommands:source~/.
bashrcdeployment_folder/vi_edge-bin_vi/vi_score_engine_restful/restartEngine.
sh24UserGuide11.
RunthefollowingcommandonthemasteredgemachineandtheslaveedgemachinestolookforthePythonprocessforthescoreengineoftheobjectdetectionmodel:psaux|greppythonTheresultshouldincludethefollowingPythonprocesses:pythondeployment_folder/vi_edge-bin_vi/vi_obj_detection/RESTAPI/model/run.
pyportgpuidFRCNNpythondeployment_folder/vi_edge-bin_vi/vi_obj_detection/RESTAPI/model/run.
pyportgpuidSSDIfyoudonotfindthePythonprocessontheedgemachines,checkthelogfiles:deployment_folder/vi_edge-bin_vi/vi_obj_detection/RESTAPI/model/FRCNN_gpuid.
logdeployment_folder/vi_edge-bin_vi/vi_obj_detection/RESTAPI/model/SSD_gpuid.
logAfteryoufixanyissues,themasterengineshouldstartthescoreengineautomatically.
Tostartthescoreenginemanually,runthefollowingcommands:source~/.
bashrcdeployment_folder/vi_edge-bin_vi/vi_obj_detection/RESTAPI/model/restartEngine.
shportgpuidFRCNNdeployment_folder/vi_edge-bin_vi/vi_obj_detection/RESTAPI/model/restartEngine.
shportgpuidSSDChapter3.
Creatingedgesystems2526UserGuideChapter4.
CreatingandusingmodelsYoucreatemodelstocollecthistoricalimagesanddefectinformation.
Theinformationisusedtotrainthemodel.
Afterthemodelistrained,itmustbevalidatedbeforeitisdeployedtoanedge.
Validatingthemodelprovidesmodelaccuracyinformation.
Therecanbemultipleversionsofamodel.
Modelscansharedefectinformationbuthavedifferentimagefilesfromdifferentproductlines.
Youcanretrainamodeltoattempttogetahighermodelaccuracysothatthemodelversioncanbereplacedwithanewermodelversion.
Therearetwotypesofmodelsimplementations:theclassificationmodelandtheobjectdetectionmodel.
StructureofcompressedimagefilesBeforeyouaddhistoricalimagesforimagegroups,youmusthavefilesthatcontaintheimagefilesthatyouneedforeithertheclassificationmodelortheobjectdetectionmodel.
ClassificationmodelAddtheimagesintocompressedfiles.
Onecompressedfilemustcontainalloftheimagesthatbelongtothesameimagegroup.
Youmustputalloftheimagesinaflatstructurewithnosubfoldersinthecompressedfile.
ObjectdetectionmodelThestructurefortheobjectdetectionmodelmustcontaintwocompressedfoldersinonecompressedfile.
OnefoldermustbenamedJPEGImagesandtheothermustbenamedAnnotations.
Inadditiontothetwofolders,thecompressedfilemustalsocontainalabels.
txtfile.
AddalloftheimagefilestotheJPEGImagesfolder.
AddalloftheannotationsfilestotheAnnotationsfolder.
Anannotationfilemusthavethesamefilenameasit'simagefile.
ThefilesmustbeinXMLformat.
Thefollowinginformationisanexampleofanannotationfile:JPEGImages000001.
jpgUnknown864129630defect1Unspecified00474368540448CopyrightIBMCorp.
201727defect2Unspecified00303387369452Thelabels.
txtfilecontainsthenamesofallofthedefectstypesthatareinthecompressedimagefolder.
Eachdefectmustbeonaseparateline,asshowninthefollowingexample:defect1defect2defect3AddinghistoricalimagesforimagegroupsThemodelmanagercancreateimagegroupsbyusinghistoricalimagestotrainthemodel.
Procedure1.
SelectData>Imagegroups>NewImageGroup.
2.
Addauniqueimagegroupnameanddescription,selecttheimagegrouptype,andselectNext.
Fortheimagegrouptype,singlecharacteristicsmeansthatalloftheimagesthatareinthegrouphavethesametypeofdefect.
Forthesinglecharacteristicsgrouptype,eachimageinthegrouphasonedefect.
Multiplecharacteristicsmeansthatalloftheimagesthatareinthegroupcontainmorethanonedefect.
Forthemultiplecharacteristicsgrouptype,thedefecttypescanbethesameordifferent.
Fortheimagegrouptype,notadefectmeansthattheimagesinthegroupdonotcontaindefects.
Note:Aftertheimagesetisuploaded,theimagegrouptypethatyouselectcannotbechanged.
3.
IntheImagesetspane,addimagesandselectAddImageGroup.
4.
IntheImagegroupspane,openthedefectthatyouaddedandselectPreviewtoviewtheimagefiles.
ResultsIntheImagegroupspane,youcanselectanimagegroupandselectedit.
Youcanaddordeletetheimagesets,updatetheimagegroupnameordescription,orchangetheimagegrouptype.
Note:Iftheimagesetiscitedinamodelinstance,theimagesetcannotbedeleted.
28UserGuideCreatingmodelrequestsAfterthedefecttypesareadded,themodelmanagercreatesamodelrequest.
Themodelrequestissubmittedtosendthemodeltoadatascientist.
Thedatascientisttrainsthemodel.
AboutthistaskYouviewthedetailsofamodelontheAllModelstab.
Themodeldetailsincludetheversionsofthemodel.
Differentmodelversionsarebuiltbydifferentimagesets.
Themodelnameandtheproducttypeareunique.
Youcannotuseexistingmodelnamesandproducttypeswhenyoucreateanewmodel.
However,thesamemodelnamecanexistintwoseparateliststorepresentadifferentversionofthesamemodel.
Afterthenewmodelisdistributedtotheedges,theolderversionofthemodelisreplaced.
Procedure1.
SelectCreateNewModel.
2.
OntheGeneraltab,updatetheinformation.
Theproducttypeisusedtomapthemodel.
Theedgecontrollerselectsthemodelbasedontheproducttypeinformationthatisontheimage.
3.
OntheAddDefectstab,selecttheimagesetsthatyouwanttousetotrainthemodel.
4.
OntheGlobalPoliciestab,settheretrainpolicy,samplingpolicy,andmanualinspectionpolicysettings.
Ifyouselectautoretrain,theretrainingisautomaticallytriggeredwhentheconditionsarefulfilled.
Theretrainingdatasettingisusedtodefineimagedatausedforautomaticretraining,includingthescopeofdata(allhistoricalimagesorthelatestndaysofimages)andsamplingrule(themaximumimagenumberofimagesandhowtoselecttheimages).
5.
SelectDone.
ThemodelversionstatusissettoDraft.
6.
IntheAllModelspane,viewthemodel.
IfthemodelversionstatusissettoDraft,youcanselectanimagegroupandedit.
7.
SelectRequesttraining.
ThestatusofthemodelversionchangesfromDrafttoSubmitted.
TrainedmodelsThedatascientisttrainsthemodelbyusingthemodeldescriptionandtheimagesets.
Themodelmanagervalidatesthemodelandeitheracceptsorrejectsthetrainedmodel.
Themodelmanagervalidatesthemodelbyusingvalidationimagesets.
AuserwiththeDataScientistrolereceivesanotificationifanewmodeltrainingrequesthasbeensubmitted.
Thedatascientistcanselectthenotificationtogotothemodelpage.
Thedatascientistcanreviewtherelatedimagegroupsanddownloadtheimagefiles.
Basedontheimagegroup,thedatascientisttrainstheclassificationmodelorobjectdetectionmodelaccordingly.
Theclassificationmodelisusedtoprocesssinglecharacteristicsofimagesandoutputtheclassificationandconfidence.
Theobjectdetectionmodelisusedtoprocessmultiplecharacteristicsofimagesandoutputthelocationswithdefecttypeandconfidence.
Chapter4.
Creatingandusingmodels29ItisrecommendedthatthedatascientistusetheNVIDIAdevelopmenttoolorobjectdetectionlibrariestotrainthemodel.
Thetrainingprocessmighttakeseveralhoursdependingonthedatasizeandtrainingparameters.
Thedatascientistselectsthepropermodelandparameterfilesandpackagesthecompressedfilewiththerequiredstructure.
Thenthedatascientistuploadsthemodelcompressedfiletothemodelpage.
Afterthecompressedfileisuploaded,themodelstatusischangedtoTrained.
StructureofmodelfilesVisualInsightsPrivatesupportstheconvolutionalneuralnetwork(CNN)classificationandobjectdetectionmodeltypes.
CNNclassificationmodelsTheCNNclassificationmodelmustbeasinglecompressedfileandcontainthecorrectdirectorystructureandfiles.
ThecompressedmodelfileThecompressedmodelfilemustcontainthefollowingdirectoriesandfiles.
vmodel.
config(file,required)vsink.
config(file,required)vparameter.
config(file,optional)vcnet1(directory).
Thecnet1directorymustcontainthefollowingfiles:–labels.
txt(file,required)–deploy.
prototxt(file,required)–mean.
binaryproto(file,required)–info.
json(file,required)–snapshot.
caffemodel(file,required)–solver.
prototxt(file,required)–train_val.
prototxt(file,required)Eachfilemusthavethecorrectstructureandkeywords.
Thefilesaredescribedinthefollowingsections.
model.
configThefollowingtextisanexampleofthecontentsofthemodel.
configfile.
Keywordsareshowninboldtext.
submodel{module{type:"ChipROIExtractor"ref_file:"parameter.
config"}module{type:"ClassificationNet"net_name:"cnet1"}}sink{type:"SinkFilter"ref_file:"sink.
config"}ThisfilemusthaveatleastonemodulethathastheClassificationNettypeinthesubmodel.
Theref_filekeywordpointstotheotherconfigurationfileswithinthe30UserGuideCNNclassificationmodelcompressedfile.
Thenet_namekeywordreferstothefoldernamethatcontainstheCNNmodel.
Youdonotneedtochangethesinkinformationunlessyouhaveadifferentnameforthesink.
configfile.
sink.
configThecontentsofthesink.
configfileareasfollows.
Youdonotneedtoeditthecontents.
keyword:"position"keyword:"probableTypes"labels.
txtThelabels.
txtfilecontainsalltheclassnameswithwhichthisclassificationmodelisclassified.
Eachclassnamemustbeonaseparateline,asshowninthefollowingexample.
defect1defect2defect3info.
jsonTheinfo.
jsonfileholdsallofthemetadatainformationoftheCNNmodel.
Allofthefilenamesmustmatchthefilenamesthatareincludedinthismodel.
Keywordsinthefollowingexamplefileareshowninboldtext.
{"caffeflavor":"BVLC","caffeversion":"1.
0.
0-rc5","creationtime":"2017-05-0216:26:18.
957631","dataset_id":"20170502-162536-2027","deployfile":"deploy.
prototxt","digitsversion":"4.
1-dev","framework":"caffe","id":"20170502-162618-b701","imagedimensions":[227,227,3],"imageresizemode":"squash","jobid":"20170502-162618-b701","labelsfile":"labels.
txt","meanfile":"mean.
binaryproto","modelfile":"original.
prototxt","name":"C_3_227_227","snapshotfile":"snapshot_iter_138.
caffemodel","solverfile":"solver.
prototxt","status":"Aborted","train_valfile":"train_val.
prototxt","username":"coco"}OtherfilesAllfilename.
prototxtfilesaremodeldefinitionfilesthatarerequiredwhenyoutrainaCNNmodel.
Thesnapshot.
caffemodelandmean.
binaryprotofilesareoutputfilesthatarecreatedafterthemodeltrainingiscomplete.
Chapter4.
Creatingandusingmodels31ObjectdetectionmodelTheobjectdetectionmodelmustbeasinglecompressedfileandcontainthecorrectdirectorystructureandfiles.
Thefollowingobjectdetectionalgorithmsaresupported:YouOnlyLookOnce(YOLO)V2,fasterregion-basedconvolutionalneuralnetwork(FasterR-CNN),andsingleshotmultiboxdetector(SSD).
YOLOV2:TheYOLOV2objectdetectionmodelmustbeasinglecompressedfileandcontainthecorrectdirectorystructureandfiles.
CompressedmodelfileThecompressedmodelfilemustcontainthefollowingfiles:vlabels.
txtvmodel.
configvyolo_final.
weightsvYolo.
cfgEachfilemusthavethecorrectstructureandkeywords.
Thefilesaredescribedinthefollowingsections.
labels.
txtThelabels.
txtfilecontainsthenamesofalloftheobjectsthatthisobjectdetectionmodeldetects.
Eachobjectmustbeonaseparateline,asshowninthefollowingexample.
defect1defect2defect3model.
configKeywordsinthefollowingexampleareshowninboldtext:{"modelType":"YOLO","modelCfg":"Yolo.
cfg","model":"yolo_final.
weights","parameters":{"iteration":"40000","batchSize":16,"learningRate":0.
001,"subBatchSize":2,"steps":"100,15000,25000,35000","scales":"-1,10,0.
1,0.
1"}}ThevalueofmodelTypeisalwaysYOLO.
ThevalueofmodelCfgisthenameofthedeeplearningnetworkdefinitionfile.
Thevalueofmodelisthenameoftheactualmodelfile.
Thevaluesinparametersarehyper-parametersoftheYOLOV2model.
yolo_final.
weightsThisfilecontainstheoutputaftertheYOLOmodelistrained.
Thefilenamemustmatchthedefinitionofmodelinmodel.
config.
32UserGuideYolo.
cfgThismodeldefinitionfileincludesnetworkdefinitions,hyper-parameters,andanchorsettings.
Atemplateofthisfilecanbefoundinthedarknet/cfg/YOLOinstallationdirectory.
Thisfilemustmatchtheweightsfile.
FasterR-CNN:TheFasterR-CNNobjectdetectionmodelmustbeasinglecompressedfileandcontainthecorrectdirectorystructureandfiles.
CompressedmodelfileThecompressedmodelfilemustcontainallofthefollowingfiles:vlabels.
txtvfaster_rcnn_final.
caffemodelvmodel.
configvstage1_fast_rcnn_solver30k40k.
ptvstage1_fast_rcnn_train.
ptvstage1_rpn_solver60k80k.
ptvstage1_rpn_train.
ptvstage2_fast_rcnn_solver30k40k.
ptvstage2_fast_rcnn_train.
ptvstage2_rpn_solver60k80k.
ptvstage2_rpn_train.
ptvfaster_rcnn_test.
ptvrpn_test.
ptEachfilemusthavethecorrectstructureandkeywords.
Thefilesaredescribedinthefollowingsections.
labels.
txtThelabels.
txtfilecontainsthenamesofalloftheobjectsthatthisobjectdetectionmodeldetects.
Eachobjectmustbeonaseparateline,asshowninthefollowingexample.
defect1defect2defect3faster_rcnn_final.
caffemodelThisfilecontainstheoutputaftertheFasterR-CNNmodelistrained.
Thefilenamemustmatchthedefinitionofmodelinmodel.
config.
model.
configKeywordsinthefollowingexampleareshowninboldtext:{"modelType":"FRCNN","model":"faster_rcnn_final.
caffemodel","solvers":"stage1_rpn_solver60k80k.
pt,stage1_fast_rcnn_solver30k40k.
pt,stage2_rpn_solver60k80k.
pt,stage2_fast_rcnn_solver30k40k.
pt","net_file":"stage1_rpn_train.
pt,stage1_fast_rcnn_train.
pt,Chapter4.
Creatingandusingmodels33stage2_rpn_train.
pt,stage2_fast_rcnn_train.
pt","deploy_net":"faster_rcnn_test.
pt","parameters":{"iteration":"40000,80000,40000,80000","learningRate":0.
001,"stepsize":"10000","gamma":"0.
1"}}ThevalueofmodelTypeisalwaysFRCNN.
Thevalueofmodelisthenameofthemodelfile.
Thevalueofsolversisthelistofsolverfilesthatareusedduringmodeltraining.
Thevalueofnet_fileisthelistofnetworkdefinitionfiles.
Thevalueofdeploy_netisthenameofthescoringnetworkdefinition.
Thevaluesinparametersareallhyper-parametersoftheFasterR-CNNmodel.
*.
ptfilesThefileswiththe.
ptextensionaremodeldefinitionfiles.
Thepascol_vocmodelthatisprovidedbyFasterR-CNNissupported.
Thetemplatefilecanbefoundinthemodels/pascal_voc/netname/faster_rcnn_alt_opt/FasterR-CNNinstallationdirectory,wherenetnameisZForVGG16.
SSD:TheSSD(SingleShotMultiBoxDetector)objectdetectionmodelmustbeasinglecompressedfileandcontainthecorrectdirectorystructureandfiles.
CompressedmodelfileThecompressedmodelfilemustcontainallofthefollowingfiles:vlabels.
txtvsolver.
prototxtvdeploy.
prototxtvmodel.
configvSSD.
caffemodelEachfilemusthavethecorrectstructureandkeywords.
Thefilesaredescribedinthefollowingsections.
labels.
txtThelabels.
txtfilecontainsthenamesofalloftheobjectsthatthisobjectdetectionmodeldetects.
Eachobjectmustbeonaseparateline,asshowninthefollowingexample.
defect1defect2defect3solver.
prototxtThisfilecontainsallhyper-parametersfortheSSDmodel.
deploy.
prototxtThisfilecontainsthenetworkdefinitionofthetrainedmodel.
34UserGuidemodel.
configKeywordsinthefollowingexampleareshowninboldtext:{"modelType":"SSD","seldef_parameters":{"TestRatio":0.
3,"Train_batch_size":8,"Validation_batch_size":8}}ThevalueofmodelTypeisalwaysSSD.
Thevaluesofseldef_parametersareparametersthatareusedfortrainingandvalidation.
SSD.
caffemodelThisfilecontainstheoutputaftertheSSDmodelistrained.
ValidatedmodelsAfterthedatascientistattachesamodelfiletoamodel,themodelmanagerreceivesanotification.
Themodelmanagercanselectthenotificationtogotothemodelpage.
Themodelmanagervalidatesthemodelversionandeitheracceptsorrejectsthetrainedmodel.
Themodelmanagervalidatesthemodelbyusingvalidationimagesets.
Afteramodelisvalidated,itcanbeusedanddeployed.
YoucanvalidateamodelversionthathasastatusofTrained.
Atrainedorretrainedmodelversioncantriggerthevalidationprocess.
Whenyouvalidateareport,youmustmanagetheimagesets.
Everyimagegroupmusthaveatleastonevalidationimagesettovalidatethemodel.
Youmustusedifferentimagesetsandtrainingimagesetstovalidatethemodelversion.
Tobeginthevalidationprocess,selectValidate.
Afterthevalidationprocessisfinished,areportisgeneratedthatshowsthemodelaccuracy.
Youcancreatethefollowingtypesofreports:ClassificationmodelreportIntheclassificationmodelreport,oneimagehasonedefectatmost.
Theconfusionmatrixisusedtogeneratethereportwhereeachcolumnrepresentsonerealimagegrouptypeinthevalidationdatasets.
Eachrowrepresentsthepredictedimagegrouptype.
Thelastrowinthechartrepresentstheaggregateresults.
ObjectdetectionmodelreportIntheobjectdetectionmodelreport,oneimagehasmultipledefects.
Inthisreport,themeanaverageprecisionandtherecallarecalculatedtoindicatethemodelaccuracy.
Toviewthereport,selectShowReport.
FromtheValidationReportwindow,youcanrevalidate,reject,oracceptanddeploythereport.
Ifthefirstmodelversionisrejected,youcanattachanewmodelfile.
Ifamodelversionthatisnotthefirstversionisrejected,themodelmanagercanretrainanewmodelversion.
Chapter4.
Creatingandusingmodels35DistributingtrainedmodelstoedgesAfterthemodelmanageracceptsthetrainedmodelversion,themodelisdistributedtoedgessothatitcanbeinspected.
Procedure1.
SelectamodelversionwithastatusofAccepted.
SelectDeploy.
Aconfirmationdialogboxdisplays.
Themasteredgeisusedtodeploythetrainedmodel.
2.
SelectDeploy.
RetrainingmodelsWhenyouretrainamodel,anewmodelversioniscreated.
Whenyoucreateamodelrequest,youdefinetheretrainpolicy.
Theretrainpolicyistheconditionthattriggersauto-retrain.
Ifthereisconcernaboutthecurrentmodelaccuracy,themodelmanagercanmanuallyselecttheimagefilestotriggertheretrainprocess.
Procedure1.
SelectamodelversionthathasastatusofDeployedandselectRetrain.
2.
Managetheimagesets.
Everyimagegroupmusthaveatleastoneretrainimagesettoretrainthemodel.
Note:Toretrainthemodelversion,itisbesttousedifferentimagesetsthatcontainmoreimages.
3.
SelectRetraintobegintheretrainprocess.
WhattodonextThemodelmanagervalidatesiftheretrainedmodelisaccepted.
Themodelmanagercandeploythenewversion,andtheoldmodelversionisnolongerdeployed.
36UserGuideChapter5.
CheckinginspectionresultsAftertheinspectionresultsaresenttothecenterapplication,theinspectorandtheinspectorsupervisorcangototheDefectChecktabtoviewandfiltertheinspectionresultsandmakeanynecessarychanges.
ImagesTheinspectorandtheinspectorsupervisorcanviewimagestoseeiftheyareclassifiedasexistingdefectsornot,andtofindoutifsomeoneelsecheckedtheimages.
Viewingtheimagesdetermineswhattheinspectororsupervisormustdowhentheycheckthedefects.
Theinspectorviewsuncheckedandcheckedimages.
Uncheckedimagesmeanthattheimagewasonlyscoredbythemodelandwasnotcheckedbyaninspector.
Checkedimagesmeanthattheimagewasscoredbythemodelandwasalreadycheckedbyaninspector.
Theinspectorviewsunconfirmedandconfirmedimages.
Unconfirmedimagesmeanthattheimagewasonlyscoredbythemodelandwasnotconfirmedbyaninspector.
Confirmedimagesmeanthattheimagewasscoredbythemodelandwasconfirmedbyaninspector.
Theinspectorsupervisorcanviewobjectsandunknownobjects.
Unknownobjectsmeanthattheimagewasmarkedbyaninspectorasanunknowndefectbecausetheinspectordidnotclassifytheimageasanexistingdefect.
Theseimagesarehighlightedontheinspectorsupervisor'slist.
FilteringdefectsTheinspectorandtheinspectorsupervisorcanapplyfilterstothecelloverviewandthedefectlist.
Procedure1.
OntheAllWorkstationswindow,selectaworkstationtoviewthelistofunconfirmed,confirmed,andunknownobjects.
2.
Selectthefiltericon.
3.
Inputavalueforaconditiontosetthefilterandthenselecttheaddicon.
Thefilterappliestothelistimmediately.
CheckingdefectsTheinspectorandinspectorsupervisorreviewtheinspectionresultsandmakeanynecessarychanges.
AboutthistaskWhenyouselectanimage,thedefectcandidateandcorrespondingconfidencedisplay.
Thefirstdefectisselectedbydefault.
Theinspectorcanselectunknownforadefecttypeifthedefecttypeisunknown.
CopyrightIBMCorp.
201737Procedure1.
Selectanimagetoviewtheimagedetailsandinspectionresults.
2.
SelectEditZoomtozoominandoutoftheimageordragtheimagetolocateaposition.
3.
SelectSetZoomtochangebacktoeditmode.
Youcanadd,resize,move,andviewthedetailsofadefectbox.
4.
Selectthedefectboxthatyouwanttoconfirmandviewthedetailsofthedefecttypeandconfidencelevel.
Youcanchangethedefecttypeordeletetheposition.
5.
SelectConfirm.
6.
Iftheimagedoesnotbelongtoanyexistingdefects,theinspectorsupervisorcancreateanewimagegroup.
Thenewdefectisaddedintothecandidatelistforimagesthatareunderthesamemodel.
38UserGuideChapter6.
KPIdashboardTheinspectorsupervisorusestheKPIdashboardtochecktheimageleveldefectrateandthelocationleveldefectrate.
ThesemetricscanhelpprovideinformationsothatyoucanasktheITteamtoretrainthemodeloradjustthemanufacturingprocedure.
TheKPIdashboardisontheKPItab.
Youcanselectallworkstationsoraspecificworkstation.
Thisselectionimpactsthescopethatyouareworkingon.
Youcanalsoswitchbetweenreal-timeandhistoricalviews.
Inthereal-timeview,KPIdataisrefreshedevery5seconds.
TheKPIvaluesincludedefectperunitanddefectrate.
Thedefectperunitiscalculatedasaspecificdefectnumberdividedbythetotalimagenumber.
Thedefectperunitvaluerepresentstheoccurrencerateofadefecttype.
Thedefectrateiscalculatedasthenumberofimageswithoneormoredefectsdividedbythetotalimagenumber.
Thedefectraterepresentstheproductdefectrate.
EachlineinthechartrepresentstheKPIvalueinthecurrent5-secondinterval.
ThehistoricalviewshowshistoricalKPIdata.
Youcaneditthestartdateandenddatetodeterminethetimerange.
KPIsinthehistoricalviewincludedefectperunitanddefectrate.
Thehistoricalviewhasthreegranularities:hourly,daily,andmonthly.
Inthehourlychart,eachpointrepresents1hour.
Forexample,aKPIvaluethatis24pointsrepresents24hours.
Inthedailychart,eachpointrepresentsoneday.
Forexample,aKPIvaluethatis30pointsrepresents30days.
Inthemonthlychart,eachpointrepresentsonemonth.
Forexample,aKPIvaluethatis12pointsrepresents12months.
Therearerelationshipsbetweentheselectedtimerangeandgranularity.
Iftherearetoomanypointsinagranularityforaselectedtimerange,thenthatgranularityisdisableduntilyoushortenthetimerange.
Ifyouwanttorefreshthechart,youcanchangetimerangeorclickRefresh.
ThehistoricalKPIdataiscalculatedperiodicallybytheserver,sothereissometimedelaybasedontheconfiguration.
ThedefaultperiodtocalculatehistoricalKPIis1hour.
Bydefault,onlythetopfivedefectstypesaredisplayedintheDefectPerUnitchart.
Ifyouwanttocheckotherdefecttypes,youcanselectoneormoredefecttypesunderthechartandthenclickShowtrend.
AnewKPIchartdisplaystheselecteddefecttypes.
IftheKPIvalueofadefecttypeis0,itcannotbeselectedtoshowatrend.
CopyrightIBMCorp.
20173940UserGuideChapter7.
UsingtheprovisioningconsoleTheprovisioningprocessincludescreatinganinstanceoftheproduct,creatinganAPIkey,creatingoneormoretenants,andaddinguserstotenants.
Youcompletethesetasksbyusingtheprovisioningconsole.
CreatinganinstanceBeforeusingVisualInsightsPrivate,youmustcreateaninstanceoftheproduct.
Onlyoneinstanceofeachproductcanbecreated.
Procedure1.
LogintotheprovisioningconsoleuserinterfacebyusingthefollowingURL:http://hostname:port/wherehostnameistheprovisioningconsolehostandportistheportthatisconfiguredfortheprovisioningconsole.
2.
ClickInstances.
3.
IfnoVisualInsightsPrivateinstanceexists,createone:a.
ClickAddInstance.
b.
SelecttheIBMVisualInsightsPrivateapplicationtype.
c.
SpecifytheapplicationURLas:https://hostname:port/ibm/vi/wherehostnameistheVisualInsightsPrivatehostandportistheportthatisconfiguredforVisualInsightsPrivate.
d.
Addtheinstance.
SettinganAPIkeyAfteryoucreateaninstanceofVisualInsightsPrivate,youmustsetanAPIkeytoenabletheproducttocommunicatewiththeprovisioningconsole.
Procedure1.
LogintotheprovisioningconsoleuserinterfacebyusingthefollowingURL:http://hostname:port/.
2.
ClickInstances.
3.
ExpandtheVisualInsightsPrivateinstancetoviewtheinstancedetails.
4.
ClickAPIKeys.
IfnoAPIkeyexists,clickAdd,thenclickOK.
CreatingtenantsAfteryoucreateaninstanceofVisualInsightsPrivate,youmustcreateatleastonetenant.
Atenantisagroupofusersthataregrantedaccesstoinstancesofoneormoreproducts.
Thereisnolimittothenumberofusersthatcanbeassignedtoatenant.
Youcancreateonetenantforallyourusers,oryoucancreatemultipletenantstogroupusersandcontroltheiraccesstoproductinstances.
Systemadministratorsmustloginasthetenanttoaddusersforthattenant.
Procedure1.
LogintotheprovisioningconsoleuserinterfacebyusingthefollowingURL:http://hostname:port/.
CopyrightIBMCorp.
2017412.
ClickTenants.
3.
ClickAddTenant.
4.
EnteranIDforthetenant.
TheIDmustbeginwithaletterandcontainonlylettersandnumbers.
5.
Enteranemailaddressforthetenant.
Thetenantemailaddressisusedonlyasanidentifierforthetenant.
Noemailissenttothisaddress.
6.
Enterapasswordforthetenant.
7.
Fortenanttype,specifyinternal.
8.
Leavetheexpirationdateblank.
9.
LeavetheNotefieldblank.
10.
Selecttheproductinstancesthatthetenantmayaccess.
11.
ClickAdd.
CreatingusersAfteryoucreateaninstanceofVisualInsightsPrivateandcreateatleastonetenant,youcanadduserstothetenant.
Systemadministratorsmustloginasthetenanttoaddusersforthattenant.
Procedure1.
LogintotheprovisioningconsoleuserinterfacebyusingthefollowingURL:http://hostname:port/.
2.
ClickUsers.
3.
ClickAddUser.
4.
Entertheusername.
5.
Entertheuser'semailaddress.
6.
Enterapasswordfortheuser.
7.
Entertheuser'sfullname.
8.
IntheMetadatafield,entermetadatainJSONformatthatindicatestheapplication,theuser'srole,andthecellsthattheuserisallowedtoaccess.
Enterthemetadatainthefollowingformat:{'vi':{'role':'','cells':}}For,specifymodelmanager,inspector,orinspectorsupervisor.
For,specifyalistofcellstringsinJSONArrayformat.
Forexample:{'vi':{'role':'inspectorsupervisor','cells':['plant1,line1,cell1','plant2,line2,cell2']}}9.
Specifythetenanttowhichyouwanttoaddtheuser.
10.
ClickAdd.
42UserGuideNoticesThisinformationwasdevelopedforproductsandservicesofferedintheUS.
ThismaterialmightbeavailablefromIBMinotherlanguages.
However,youmayberequiredtoownacopyoftheproductorproductversioninthatlanguageinordertoaccessit.
IBMmaynotoffertheproducts,services,orfeaturesdiscussedinthisdocumentinothercountries.
ConsultyourlocalIBMrepresentativeforinformationontheproductsandservicescurrentlyavailableinyourarea.
AnyreferencetoanIBMproduct,program,orserviceisnotintendedtostateorimplythatonlythatIBMproduct,program,orservicemaybeused.
Anyfunctionallyequivalentproduct,program,orservicethatdoesnotinfringeanyIBMintellectualpropertyrightmaybeusedinstead.
However,itistheuser'sresponsibilitytoevaluateandverifytheoperationofanynon-IBMproduct,program,orservice.
IBMmayhavepatentsorpendingpatentapplicationscoveringsubjectmatterdescribedinthisdocument.
Thefurnishingofthisdocumentdoesnotgrantyouanylicensetothesepatents.
Youcansendlicenseinquiries,inwriting,to:IBMDirectorofLicensingIBMCorporationNorthCastleDrive,MD-NC119Armonk,NY10504-1785USForlicenseinquiriesregardingdouble-bytecharacterset(DBCS)information,contacttheIBMIntellectualPropertyDepartmentinyourcountryorsendinquiries,inwriting,to:IntellectualPropertyLicensingLegalandIntellectualPropertyLawIBMJapanLtd.
19-21,Nihonbashi-Hakozakicho,Chuo-kuTokyo103-8510,JapanINTERNATIONALBUSINESSMACHINESCORPORATIONPROVIDESTHISPUBLICATION"ASIS"WITHOUTWARRANTYOFANYKIND,EITHEREXPRESSORIMPLIED,INCLUDING,BUTNOTLIMITEDTO,THEIMPLIEDWARRANTIESOFNON-INFRINGEMENT,MERCHANTABILITYORFITNESSFORAPARTICULARPURPOSE.
Somejurisdictionsdonotallowdisclaimerofexpressorimpliedwarrantiesincertaintransactions,therefore,thisstatementmaynotapplytoyou.
Thisinformationcouldincludetechnicalinaccuraciesortypographicalerrors.
Changesareperiodicallymadetotheinformationherein;thesechangeswillbeincorporatedinneweditionsofthepublication.
IBMmaymakeimprovementsand/orchangesintheproduct(s)and/ortheprogram(s)describedinthispublicationatanytimewithoutnotice.
Anyreferencesinthisinformationtonon-IBMwebsitesareprovidedforconvenienceonlyanddonotinanymannerserveasanendorsementofthoseCopyrightIBMCorp.
201743websites.
ThematerialsatthosewebsitesarenotpartofthematerialsforthisIBMproductanduseofthosewebsitesisatyourownrisk.
IBMmayuseordistributeanyoftheinformationyouprovideinanywayitbelievesappropriatewithoutincurringanyobligationtoyou.
Licenseesofthisprogramwhowishtohaveinformationaboutitforthepurposeofenabling:(i)theexchangeofinformationbetweenindependentlycreatedprogramsandotherprograms(includingthisone)and(ii)themutualuseoftheinformationwhichhasbeenexchanged,shouldcontact:IBMDirectorofLicensingIBMCorporationNorthCastleDrive,MD-NC119Armonk,NY10504-1785USSuchinformationmaybeavailable,subjecttoappropriatetermsandconditions,includinginsomecases,paymentofafee.
ThelicensedprogramdescribedinthisdocumentandalllicensedmaterialavailableforitareprovidedbyIBMundertermsoftheIBMCustomerAgreement,IBMInternationalProgramLicenseAgreementoranyequivalentagreementbetweenus.
Theperformancedataandclientexamplescitedarepresentedforillustrativepurposesonly.
Actualperformanceresultsmayvarydependingonspecificconfigurationsandoperatingconditions.
Informationconcerningnon-IBMproductswasobtainedfromthesuppliersofthoseproducts,theirpublishedannouncementsorotherpubliclyavailablesources.
IBMhasnottestedthoseproductsandcannotconfirmtheaccuracyofperformance,compatibilityoranyotherclaimsrelatedtonon-IBMproducts.
Questionsonthecapabilitiesofnon-IBMproductsshouldbeaddressedtothesuppliersofthoseproducts.
Thisinformationisforplanningpurposesonly.
Theinformationhereinissubjecttochangebeforetheproductsdescribedbecomeavailable.
Thisinformationcontainsexamplesofdataandreportsusedindailybusinessoperations.
Toillustratethemascompletelyaspossible,theexamplesincludethenamesofindividuals,companies,brands,andproducts.
Allofthesenamesarefictitiousandanysimilaritytoactualpeopleorbusinessenterprisesisentirelycoincidental.
COPYRIGHTLICENSE:Thisinformationcontainssampleapplicationprogramsinsourcelanguage,whichillustrateprogrammingtechniquesonvariousoperatingplatforms.
Youmaycopy,modify,anddistributethesesampleprogramsinanyformwithoutpaymenttoIBM,forthepurposesofdeveloping,using,marketingordistributingapplicationprogramsconformingtotheapplicationprogramminginterfacefortheoperatingplatformforwhichthesampleprogramsarewritten.
Theseexampleshavenotbeenthoroughlytestedunderallconditions.
IBM,therefore,cannotguaranteeorimplyreliability,serviceability,orfunctionoftheseprograms.
Thesample44UserGuideprogramsareprovided"ASIS",withoutwarrantyofanykind.
IBMshallnotbeliableforanydamagesarisingoutofyouruseofthesampleprograms.
TrademarksIBM,theIBMlogo,andibm.
comaretrademarksorregisteredtrademarksofInternationalBusinessMachinesCorp.
,registeredinmanyjurisdictionsworldwide.
OtherproductandservicenamesmightbetrademarksofIBMorothercompanies.
AcurrentlistofIBMtrademarksisavailableonthewebat"Copyrightandtrademarkinformation"atwww.
ibm.
com/legal/copytrade.
shtml.
JavaandallJava-basedtrademarksandlogosaretrademarksorregisteredtrademarksofOracleand/oritsaffiliates.
LinuxisatrademarkofLinusTorvaldsintheUnitedStates,othercountries,orboth.
Microsoft,Windows,WindowsNT,andtheWindowslogoaretrademarksofMicrosoftCorporationintheUnitedStates,othercountries,orboth.
UNIXisaregisteredtrademarkofTheOpenGroupintheUnitedStatesandothercountries.
TermsandconditionsforproductdocumentationPermissionsfortheuseofthesepublicationsaregrantedsubjecttothefollowingtermsandconditions.
ApplicabilityThesetermsandconditionsareinadditiontoanytermsofusefortheIBMwebsite.
PersonaluseYoumayreproducethesepublicationsforyourpersonal,noncommercialuseprovidedthatallproprietarynoticesarepreserved.
Youmaynotdistribute,displayormakederivativeworkofthesepublications,oranyportionthereof,withouttheexpressconsentofIBM.
CommercialuseYoumayreproduce,distributeanddisplaythesepublicationssolelywithinyourenterpriseprovidedthatallproprietarynoticesarepreserved.
Youmaynotmakederivativeworksofthesepublications,orreproduce,distributeordisplaythesepublicationsoranyportionthereofoutsideyourenterprise,withouttheexpressconsentofIBM.
RightsExceptasexpresslygrantedinthispermission,nootherpermissions,licensesorrightsaregranted,eitherexpressorimplied,tothepublicationsoranyinformation,data,softwareorotherintellectualpropertycontainedtherein.
Notices45IBMreservestherighttowithdrawthepermissionsgrantedhereinwhenever,initsdiscretion,theuseofthepublicationsisdetrimentaltoitsinterestor,asdeterminedbyIBM,theaboveinstructionsarenotbeingproperlyfollowed.
Youmaynotdownload,exportorre-exportthisinformationexceptinfullcompliancewithallapplicablelawsandregulations,includingallUnitedStatesexportlawsandregulations.
IBMMAKESNOGUARANTEEABOUTTHECONTENTOFTHESEPUBLICATIONS.
THEPUBLICATIONSAREPROVIDED"AS-IS"ANDWITHOUTWARRANTYOFANYKIND,EITHEREXPRESSEDORIMPLIED,INCLUDINGBUTNOTLIMITEDTOIMPLIEDWARRANTIESOFMERCHANTABILITY,NON-INFRINGEMENT,ANDFITNESSFORAPARTICULARPURPOSE.
IBMOnlinePrivacyStatementIBMSoftwareproducts,includingsoftwareasservicesolutions,("SoftwareOfferings")mayusecookiesorothertechnologiestocollectproductusageinformation,tohelpimprovetheenduserexperience,totailorinteractionswiththeenduserorforotherpurposes.
InmanycasesnopersonallyidentifiableinformationiscollectedbytheSoftwareOfferings.
SomeofourSoftwareOfferingscanhelpenableyoutocollectpersonallyidentifiableinformation.
IfthisSoftwareOfferingusescookiestocollectpersonallyidentifiableinformation,specificinformationaboutthisoffering'suseofcookiesissetforthbelow.
Dependingupontheconfigurationsdeployed,thisSoftwareOfferingmayusesessionandpersistentcookiesthatcollecteachuser'sname,username,password,orotherpersonallyidentifiableinformationforpurposesofsessionmanagement,authentication,singlesign-onconfigurationorotherusagetrackingorfunctionalpurposes.
Thesecookiescanbedisabled,butdisablingthemwillalsolikelyeliminatethefunctionalitytheyenable.
IftheconfigurationsdeployedforthisSoftwareOfferingprovideyouascustomertheabilitytocollectpersonallyidentifiableinformationfromendusersviacookiesandothertechnologies,youshouldseekyourownlegaladviceaboutanylawsapplicabletosuchdatacollection,includinganyrequirementsfornoticeandconsent.
Formoreinformationabouttheuseofvarioustechnologies,includingcookies,forthesepurposes,seeIBM'sPrivacyPolicyathttp://www.
ibm.
com/privacyandIBM'sOnlinePrivacyStatementathttp://www.
ibm.
com/privacy/detailsinthesectionentitled"Cookies,WebBeaconsandOtherTechnologies"andthe"IBMSoftwareProductsandSoftware-as-a-ServicePrivacyStatement"athttp://www.
ibm.
com/software/info/product-privacy.
46UserGuideNotices47IBMPrintedinUSA