激发下一代基于 Arm 的云原生智能相机设计.pdf

1、Sparking the Next Generation of Arm-Based Cloud-Native Smart Camera Designs Stephen SuSenior Product ManagerArm Inc.Where are the changes happening?Introduction to the Future of Smart Camera2 2023 Arm Inc.Current deployment is single function camera with limited cloud serviceLocal NVRCashierless sma

2、rt retailCloudFuture deploymentsObject detection on smart cameras Human presence detection and express authenticationSmart city(urban mobility,public safety,etc.)and much more.Machine learning is everywhere Need an efficient way to make it easier to support multiple platforms with varying ML capabil

3、itiesWhy Are the Changes Happening?3 2023 Arm Inc.Machine LearningSecurity CamerasSmart HomeOccupancy and People Counting Consumer AppliancesIndustrial Vision(Quality Inspection etc.)Smart DoorbellsReview of Typical Native Only Development Flow4 2023 Arm Inc.OSCPUNPUISPML ApplicationTraining Data St

4、oreTrain ModelModel OptimizationDeployment Time-consuming to develop solutions including hardware,software and integrate cloud service provider(CSP)services Rigid machine learning(ML)use case due to single ML-model deployment Lack of software ecosystem to deploy multiple ML applicationsBuilding an A

5、rm-Based Cloud-Native Smart Camera 5 2023 Arm Inc.Package ModelCollaborateApp developer using DevOpsData scientist using Machine LearningIoT EngineerModel Zoo(per target hardware)Arm Optimization ToolsArm Virtual HardwareRetrain model&DebugDeploy multiple container images to support various machine

6、learning workloads/servicesTarget platformCloud6 2023 Arm Inc.Arm Vision SolutionMakes it Easier than Ever to Designa Cloud-Native Smart CameraOverview of Arm Vision Solution7 2023 Arm Inc.Reference SystemArchitectureOpen Source Reference SW StackArm Virtual Hardware/FPGAPower/PerfReportsDemonstrato