03-周兆靖-Intel-OpenVINO toolkit的英特爾發行版（20頁）.pdf

《03-周兆靖-Intel-OpenVINO toolkit的英特爾發行版（20頁）.pdf》由會員分享，可在線閱讀，更多相關《03-周兆靖-Intel-OpenVINO toolkit的英特爾發行版（20頁）.pdf（20頁珍藏版）》請在三個皮匠報告上搜索。

1、Product OverviewIntel Distribution of OpenVINO toolkit Zhou,ZhaojingPRC IOTG-DE 2019 Developer Tool of the Year2019 Developer Tool of the YearAwarded by the Edge AI and Vision Alliance2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview5Notices and DisclaimersNotic

2、es and DisclaimersSoftware and workloads used in performance tests may have been optimized for performance only on Intel microprocessors.Performance tests,such as SYSmark and MobileMark,are measured using specific computer systems,components,software,operations and functions.Any change to any of tho

3、se factors may cause the results to vary.You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases,including the performance of that product when combined with other products.For more complete information visit results are based on testi

4、ng as of dates shown in configurations and may not reflect all publicly available updates.See backup for configuration details.No product or component can be absolutely secure.Your costs and results may vary.Intel technologies may require enabled hardware,software or service activation.Intel Corpora

5、tion.Intel,the Intel logo,and other Intel marks are trademarks of Intel Corporation or its subsidiaries.Other names and brands may be claimed as the property of others.Optimization NoticeOptimization Notice1 Intels compilers may or may not optimize to the same degree for non-Intel microprocessors fo

6、r optimizations that are not unique to Intel microprocessors.These optimizations include SSE2,SSE3,and SSSE3 instruction sets and other optimizations.Intel does not guarantee the availability,functionality,or effectiveness of any optimization on microprocessors not manufactured by Intel.Microprocess

7、or-dependent optimizations in this product are intended for use with Intel microprocessors.Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors.Please refer to the applicable product User and Reference Guides for more information regarding the specific

8、 instruction sets covered by this notice.Notice revision#20110804.2 Software and workloads used in performance tests may have been optimized for performance only on microprocessors from Intel.Performance tests,such as SYSmark and MobileMark,are measured using specific computer systems,components,sof

9、tware,operations,and functions.Any change to any of those factors may cause the results to vary.Consult other information and performance tests while evaluating potential purchases,including performance when combined with other products.For more information,see Performance Benchmark Test Disclosure.

10、Source:Intel measurements,as of June 2017.2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview7Notices and DisclaimersNotices and DisclaimersIntel technologies features and benefits depend on system configuration and may require enabled hardware,software or service

11、 activation.Performance varies depending on system configuration.Check with your system manufacturer or retailer or learn more at .Intel,the Intel logo,Xeon,Arria and Movidius are trademarks of Intel Corporation or its subsidiaries in the U.S.and/or other countries.*Other names and brands may be cla

12、imed as the property of others.Intel Corporation.System BoardSystem BoardIntel prototype,TGL U DDR4 SODIMM RVPASUSTeK COMPUTER INC./PRIME Z370-ACPUCPU11thGen Intel Core-5-1145G7E 2.6 GHz.8thGen Intel Core i5-8500T 3.0 GHz.Sockets/Physical coresSockets/Physical cores1/41/6HyperThreading/Turbo Setting

13、HyperThreading/Turbo SettingEnabled/OnNa/OnMemoryMemory2 x 8198 MB 3200 MT/s DDR42 x 16384 MB 2667 MT/s DDR4OSOSUbuntu*18.04 LTSUbuntu*18.04 LTSKernelKernel5.8.0-050800-generic5.3.0-24-genericSoftwareSoftwareIntel Distribution of OpenVINO toolkit 2021.1.075Intel Distribution of OpenVINO toolkit 2021

14、.1.075BIOS BIOS Intel TGLIFUI1.R00.3243.A04.2006302148AMI,version 2401BIOS release dateBIOS release dateRelease Date:06/30/20207/12/2019BIOS SettingBIOS SettingLoad default settingsLoad default settings,set XMP to 2667Test DateTest Date9/9/20209/9/2020Precision and Batch SizePrecision and Batch Size

15、CPU:INT8,GPU:FP16-INT8,batch size:1CPU:INT8,GPU:FP16-INT8,batch size:1Number of Inference Requests Number of Inference Requests 46Number of Execution Streams Number of Execution Streams 46Power(TDP Link)Power(TDP Link)28 W35WPrice(USD)Link on Sep 22,2020Price(USD)Link on Sep 22,2020Prices may varyPr

16、ices may vary$309$1921):Memory is installed such that all primary memory slots are populated.2):Testing by Intel as of September 9,2020102020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product OverviewIntel Distribution of OpenVINO toolkit Product Overview(3 minutes)Product Ov

17、erview(3 minutes)01 Why Deep Learning02 Development&Deployment Challenges03 Why Intel Distribution of OpenVINO toolkit04 Build,Optimize,Deploy05 Under the Hood06 Resources and Community Support07 Choose and Download2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overvi

18、ew11Why Deep LearningMachines able to meet or exceed human image and speech recognitionMachines able to meet or exceed human image and speech recognition0%8%15%23%30%Human 2010PresentImage Recognition0%8%15%23%30%2000PresentSpeech RecognitionErrorErrorHuman 97%person99%“play song”Source:ILSVRC Image

19、Net winning entry classification error rate each year 2010-2016(Left),https:/ economic impactin 2030$13trillion2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview12Challenges in Deep LearningDevelopment and deployment challenges in deep learningDevelopment and dep

20、loyment challenges in deep learningDiverse requirements for myriad use cases require unique approaches No One Size Fits AllNo One Size Fits AllNo streamlined way for end-to-end development workflowIntegration ChallengesIntegration ChallengesGap in performance and accuracy between trained and deploye

21、d modelsUnique Inference NeedsUnique Inference NeedsLow performing,lower Low performing,lower accuracy models accuracy models deployeddeployedSlow timeSlow time-toto-solution solution and timeand time toto-marketmarketInability to meet useInability to meet use-case specific case specific requirement

22、srequirements2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview13Why Intel Distribution of OpenVINO toolkit HighHigh-Performance,Deep Learning InferencePerformance,Deep Learning InferenceHigh-Performance,Deep Learning InferenceStreamlined Development,Ease of UseW

23、rite Once,Deploy AnywhereFaster,more accurate real-world results using high-performance,AI and computer vision inference deployed into production from edge to cloud.2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview151.buildDeep Learning with theIntel Distributio

24、n ofOpenVINO toolkit2.optimize3.deploy2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview16Under the HoodAdvanced capabilities to streamline deep learning deploymentsAdvanced capabilities to streamline deep learning deploymentsTrained ModelOpen Model Zoo100+open s

25、ourced and optimized pre-trained models;80+supported public modelsModel OptimizerConverts and optimizes trained model using a supported frameworkIR DataRead,Load,InferInference EngineCommon API that abstracts low-level programming for each hardwareIntermediate Representation(.xml,.bin)Myriad PluginF

26、or Intel NCS2&NCSHDDL PluginFGPA PluginGPU PluginGNA PluginCPU PluginPost-Training Optimization ToolDeep Learning WorkbenchDeployment ManagerOpenCVOpenCLDeep Learning StreamerCode Samples&Demos(e.g.Benchmark app,Accuracy Checker,Model Downloader)1.Build2.Optimize3.Deploy2020 Copyright,Intel Corporat

27、ionIntel Distribution of OpenVINO toolkit/Product Overview17Compounding Effect of Hardware and SoftwareImprovements mean exponential performanceImprovements mean exponential performance*2*3Baseline PerformanceAdditional Software Performance1x2.1x25.6x*1Release 2018 R1Release 2019 R1Release 2019 R31s

28、tGeneration Intel Xeon Scalable Processor2ndGeneration Intel Xeon Scalable ProcessorFor more complete information about performance and benchmark results,visit backup for configuration details.Comparison of Frames Per Second utilizing Mobilenet SSD,Batch 1.2020 Copyright,Intel CorporationIntel Distr

29、ibution of OpenVINO toolkit/Product Overview18Compounding Effect of Hardware and SoftwareCompounding Effect of Hardware and SoftwareUse Intel XUse Intel Xe eGraphics+CPU combined for maximum inferencingGraphics+CPU combined for maximum inferencingdeeplabv3-TFmobilenet-ssd-CFresnet-50-TFssd300-CFsque

30、ezenet1.1-CF11thGen Intel Core(Tiger Lake)Core i5-1145G7 relative inference FPS compared to Coffee Lake,Core i5-8500Core i5-1145G7,CPUCore i5-1145G7,GPUCore i5-1145G7,GPU+CPU1.7X1.7X1.7X2.2X3X1.6X2.9X2.2X2X1.9X1.8X2.1X2.8X3.6X3.9XThe above is preliminary performance data based on pre-production comp

31、onents.For more complete information about performance and benchmark results,visit backup for configuration details.Tiger Lake+Intel Distribution of OpenVINO toolkit vs Coffee Lake CPU2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview19Post-Training Optimization

32、ToolConversion technique that reduces model size into lowConversion technique that reduces model size into low-precision without reprecision without re-trainingtrainingModel OptimizerModel OptimizerConverts and optimizes trained model using a supported frameworkIRIRFullFull-Precision IRPrecision IRP

33、ostPost-training Optimization training Optimization ToolToolConversion technique to quantize models to low precision for high performanceAccuracy Accuracy CheckerCheckerInference EngineInference EngineTrained ModelTrained ModelModel trained using one of the supported frameworkIRIROptimized IROptimiz

34、ed IRAccuracy and Accuracy and performance checkperformance checkEnvironment(hardware)Environment(hardware)specificationsspecificationsDataset and Annotation Statistics&JSONJSONReduces model size while also improving while also improving latency,with little degradation latency,with little degradatio

35、n in model accuracy and without model re-training.Different optimization approaches are supported:quantization algorithms,sparsity,etc.2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview20Deep Learning WorkbenchWebWeb-based UI extension tool for model analyses and

36、 graphical measurementsbased UI extension tool for model analyses and graphical measurementsVisualizes performance data for Visualizes performance data for topologies and layers to aid in model analysisAutomates analysis Automates analysis for optimal performance configuration(streams,batches,latenc

37、y)Experiment with INT8 or Winograd calibration Experiment with INT8 or Winograd calibration for optimal tuning using the Post Training Optimization ToolProvide accuracy informatioaccuracy information through accuracy checkerDirect access to models Direct access to models from public set of Open Mode

38、l ZooEnables remote profilingremote profiling,allowing the collection of performance data from multiple different machines without any additional set-up.2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview46Pre-Trained Models and Public ModelsOpenOpen-sourced repos

39、itory of presourced repository of pre-trained models and support for public modelstrained models and support for public models100+Pre100+Pre-trained Modelstrained ModelsCommon AI tasksObject DetectionObject RecognitionReidentificationSemantic SegmentationInstance SegmentationHuman Pose EstimationIma

40、ge ProcessingText DetectionText RecognitionText SpottingAction RecognitionImage RetrievalCompressed ModelsQuestion Answering100+Public Models100+Public ModelsPre-optimized external modelsClassificationSegmentationObject DetectionHuman Pose EstimationMonocular Depth EstimationImage InpaintingStyle Tr

41、ansferAction RecognitionColorizationUse free PrePre-trained Models trained Models to speed up development and deploymentTake advantage of the Model Downloader Model Downloader and other automation tools to quickly get startedIterate with the Accuracy Checker Accuracy Checker to validate the accuracy

42、 of your models2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview48Demos and Reference ImplementationsQuickly get started with example demo applications and reference implementationsQuickly get started with example demo applications and reference implementationsF

43、ace Access Control-C+Intruder Detector-C+Machine Operator Monitor-C+Machine Operator Monitor-GoMotor Defect Detector-PythonObject Flaw Detector-C+Object Size Detector-C+Object Size Detector-GoParking Lot Counter-C+Parking Lot Counter-GoPeople Counter-C+Restricted Zone Notifier-GoShopper Gaze Monitor

44、-C+Shopper Mood Monitor-GoStore Aisle Monitor-C+Store Traffic Monitor-C+Store Traffic Monitor-PythonTake advantage of prepre-built,openbuilt,open-sourced sourced example implementations with step-by-step guidance and required components list492020 Copyright,Intel CorporationIntel Distribution of Ope

45、nVINO toolkit/Product OverviewCompanies using the Intel Distribution ofOpenVINO toolkit 2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview52System RequirementsIntel PlatformsIntel PlatformsCompatible Operating SystemsCompatible Operating SystemsTarget Solution Ta

46、rget Solution PlatformsPlatformsCPUCPU6th-10thgeneration Intel Core and Xeon processors1stand 2ndgeneration Intel Xeon Scalable processors Ubuntu*18.04.3 LTS(64 bit)Microsoft Windows*10(64 bit)CentOS*7.4(64 bit)macOS*10.13&10.14(64 bit)Intel Pentium processor N4200/5,N3350/5,N3450/5 with Intel HD Gr

47、aphics Yocto Project*Poky Jethro v2.0.3(64 bit)Iris Pro&Intel HD GraphicsIris Pro&Intel HD Graphics6th-10thgeneration Intel Core processor with Intel Iris Pro graphics&Intel HD Graphics Intel Xeon processor with Intel Iris Pro Graphics&Intel HD Graphics(excluding E5 product family,which does not hav

48、e graphics1)Ubuntu 18.04.3 LTS(64 bit)Windows 10(64 bit)CentOS 7.4(64 bit)FPGAFPGAIntel Arria FPGA 10 GX development kit Intel Programmable Acceleration Card with Intel Arria 10 GX FPGA operating systemsOpenCV*&OpenVX*functions must be run against the CPU or Intel Processor Graphics(GPU)Ubuntu 18.04

49、.2 LTS(64 bit)CentOS 7.4(64 bit)VPUVPU:Intel Movidius Neural Compute Stick:,Intel Neural Compute Stick2Ubuntu 18.04.3 LTS(64 bit)CentOS 7.4(64 bit)Windows 10(64 bit)macOS*(64 bit)Raspbian(target only)Intel Vision Accelerator Design ProductsIntel Vision Accelerator Design ProductsIntel Vision Acceler

50、ator Design with Intel Arria10 FPGAIntel Vision Accelerator Design with Intel Movidius VPUsUbuntu 18.04.2 LTS(64 bit)Ubuntu 8.04.3 LTS(64 bit)Windows 10(64 bit)Development Development PlatformsPlatforms6th-10thgeneration Intel Core and Intel Xeon processors 1st and 2nd generation Intel Xeon Scalable

51、 processors Ubuntu*18.04.3 LTS(64 bit)Windows 10(64 bit)CentOS*7.4(64 bit)macOS*10.13&10.14(64 bit)Additional Software Additional Software RequirementsRequirementsLinux*build environment required componentsOpenCV 3.4 or higher GNU Compiler Collection(GCC)3.4 or higher CMake*2.8 or higher Python*3.4

52、or higher Microsoft Windows*build environment required componentsIntel HD Graphics Driver(latest version)OpenCV 3.4 or higher Intel C+Compiler 2017 Update 4CMake 2.8 or higher Python 3.4 or higher Microsoft Visual Studio*2015External Dependencies/Additional SoftwareExternal Dependencies/Additional S

53、oftwareView Product Site,detailed System Requirements2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview53Resources and Community SupportVibrant community of developers,enterprises and skills buildersVibrant community of developers,enterprises and skills buildersQ

54、UALIFY Use a trained model and check if framework is supported-or Take advantage of a pre-trained model from the Open Model ZooINSTALLATION Download the Intel OpenVINO toolkit package from Intel Developer Zone,or by YUM or APT repositories Utilize the Getting Started GuidePREPARE Understand sample d

55、emos and toolsincluded Understand performance Choose hardware option with Performance Benchmarks Build,test and remotely run workloads on the Intel DevCloud for the Edge before buying hardwareHANDS ON Visualize metrics with the Deep Learning Workbench Utilize prebuilt,Reference Implementations to be

56、come familiar with capabilities Optimize workloads with these performance best practices Use the Deployment Manager to minimize deployment package Ask questions and share information with others through the Community Forum Engage using#OpenVINO on Stack Overflow Visit documentation site for guides,h

57、ow tos,and resources Attend training and get certified Ready to go to market?Tell us how we can helpSUPPORT2020 Copyright,Intel CorporationIntel Distribution of OpenVINO toolkit/Product Overview54Ready to get started?Download directly from Intel for freeIntel Distribution of OpenVINO toolkit(Recommended)Also available fromIntels Edge Software Hub|Intel DevCloud for the Edge|PIP|DockerHub|Dockerfile|Anaconda Cloud|YUM|APTBuild from sourceGitHub|Gitee(for China)Choose&Download55