DLR 图像分类 - AWS IoT Greengrass
RequirementsConfiguration组件配方
AWS 文档中描述的 AWS 服务或功能可能因区域而异。要查看适用于中国区域的差异，请参阅中国的 AWS 服务入门。

本文属于机器翻译版本。若本译文内容与英语原文存在差异，则一律以英文原文为准。

DLR 图像分类

您可以使用 DLR 图像分类组件 （aws.greengrass.DLRImageClassification） 在支持的设备上执行示例图像分类推理。此组件安装 DLR 及其依赖项，并下载示例 ML 模型以执行此推理。

此示例推理组件使用 变体图像分类模型存储和 DLR 安装程序 组件作为依赖项来执行推理。要使用不同的模型或运行时执行图像分类，您可以创建从属变体组件的私有版本。如果要创建自己的私有推理组件，您可以使用此组件的配方作为模板。

此组件具有以下版本：

  • 2.0.x

Requirements

公共机器学习推理组件需要运行以下受支持平台之一的 Greengrass 核心设备。有关更多信息，请参阅设置 AWS AWS IoT Greengrass 核心设备。

Configuration

此组件提供以下配置参数，您可以在部署组件时自定义这些参数。

Accelerator

请勿修改。目前，加速器唯一支持的值为 cpu，因为从属模型组件中的模型仅为 CPU 加速器编译。

MLRootPath

（可选）设备上的文件夹的路径，其中的推理组件读取图像并写入推理结果。您可以将此值修改为设备上您拥有读/写访问权限的任何位置。

默认值： $HOME/greengrass_ml

ImageName

（可选） 推理组件用作进行预测的输入的图像。您可以将此值修改为位于 中的任何映像MLRootPath/images。AWS AWS IoT Greengrass 支持以下格式jpeg：、jpg、 png和 npy。

默认值：静态映像

InferenceInterval

（可选） 推理代码进行的每个预测之间的时间（以秒为单位）。示例推理代码无限期运行并在指定的时间间隔内重复其预测。例如，如果要将摄像机拍摄的图像挂载到 MLRootPath/images 并使用这些图像进行实时预测，您可以将此设置更改为较短的时间间隔。

默认值： 3600

ModelResourceKey

请勿修改。此对象指定从属公有模型组件中使用的模型。

组件配方

要查看公有组件的最新版本的组件配方，请执行以下操作之一：

  • 使用控制台

    1. 在 Components （组件） 页面上的 Public components （公有组件） 选项卡上，查找并选择公有组件。

    2. 在组件页面上，选择查看配方。

  • 使用 AWS CLI

    运行以下命令以检索公有变体组件的组件配方。此命令会将组件配方写入您在命令中提供的 JSON 或 YAML 配方文件。

    Linux, macOS, or Unix
    aws greengrassv2 get-component \
    --arn <arn> \
    --recipe-output-format <recipe-format> \
    --query recipe \
    --output text | base64 --decode > <recipe-file>
    Windows command prompt
    aws greengrassv2 get-component ^
    --arn <arn> ^
    --recipe-output-format <recipe-format> ^
    --query recipe ^
    --output text > <recipe-file>.base64

certutil -decode <recipe-file>.base64 <recipe-file>

    替换命令中的值，如下所示：

    • <arn>。公有组件的 Amazon 资源名称 （ARN）。

    • <recipe-format>。您要用来创建配方文件的格式。支持的值为 JSON 和 YAML。

    • <recipe-file>。格式为 的配方的名称<component-name>-<component-version>。

以下摘录显示了 组件的版本 2.0.0 的组件配方。

JSON
{
  "RecipeFormatVersion": "2020-01-25",
  "ComponentName": "aws.greengrass.DLRImageClassification",
  "ComponentVersion": "2.0.0",
  "ComponentDescription": "Sample Image classification inference using DLR and amazon resnet50 default model.",
  "ComponentPublisher": "Amazon",
  "ComponentDependencies": {
    "variant.DLR": {
      "VersionRequirement": "~1.3.0",
      "DependencyType": "HARD"
    },
    "variant.ImageClassification.ModelStore": {
      "VersionRequirement": "~2.0.0",
      "DependencyType": "HARD"
    },
    "aws.greengrass.Nucleus": {
      "VersionRequirement": "~2.0.0",
      "DependencyType": "SOFT"
    }
  },
  "ComponentConfiguration": {
    "DefaultConfiguration": {
      "Accelerator": "cpu",
      "MLRootPath": "$HOME/greengrass_ml",
      "ImageName": "cat.jpeg",
      "InferenceInterval": 3600,
      "ModelResourceKey": {
        "armv7l": "DLR-resnet50-armv7l-cpu-ImageClassification",
        "x86_64": "DLR-resnet50-x86_64-cpu-ImageClassification"
      }
    }
  },
  "Manifests": [
    {
      "Platform": {
        "os": "linux",
        "architecture": "arm"
      },
      "Name": "32-bit armv7l - Linux (raspberry pi)",
      "Artifacts": [
        {
          "URI": "s3://$bucketName$/$sampleArtifactsDirectory$/image_classification.zip",
          "Unarchive": "ZIP"
        },
        {
          "URI": "s3://$bucketName$/$sampleArtifactsDirectory$/init.sh"
        }
      ],
      "Lifecycle": {
        "Install": {
          "RequiresPrivilege": true,
          "script": "bash {artifacts:path}/init.sh {artifacts:decompressedPath}/image_classification/sample_images {configuration:/MLRootPath}\nbash {variant.DLR:artifacts:decompressedPath}/installer/installer.sh -a {configuration:/Accelerator} -p {configuration:/MLRootPath}",
          "timeout": 900
        },
        "Run": {
          "RequiresPrivilege": true,
          "script": ". {configuration:/MLRootPath}/greengrass_ml_dlr_venv/bin/activate\npython3 {artifacts:decompressedPath}/image_classification/inference.py -a {configuration:/Accelerator} -m {variant.ImageClassification.ModelStore:artifacts:decompressedPath}/{configuration:/ModelResourceKey/armv7l} -p {configuration:/MLRootPath} -i {configuration:/ImageName} -s {configuration:/InferenceInterval}"
        },
        "shutdown": {
          "RequiresPrivilege": true,
          "script": "deactivate"
        }
      }
    },
    {
      "Platform": {
        "os": "linux",
        "architecture": "amd64"
      },
      "Name": "64-bit x86_64 - Linux (ubuntu, deeplens, amazon linux 2)",
      "Artifacts": [
        {
          "URI": "s3://$bucketName$/$sampleArtifactsDirectory$/image_classification.zip",
          "Unarchive": "ZIP"
        },
        {
          "URI": "s3://$bucketName$/$sampleArtifactsDirectory$/init.sh"
        }
      ],
      "Lifecycle": {
        "Install": {
          "RequiresPrivilege": true,
          "script": "bash {artifacts:path}/init.sh {artifacts:decompressedPath}/image_classification/sample_images {configuration:/MLRootPath}\nbash {variant.DLR:artifacts:decompressedPath}/installer/installer.sh -a {configuration:/Accelerator} -p {configuration:/MLRootPath} -e {variant.DLR:artifacts:path}/environment.yaml",
          "timeout": 900
        },
        "Run": {
          "RequiresPrivilege": true,
          "script": "export PATH=\"{configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin:$PATH\"\neval \"$({configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin/conda shell.bash hook)\"\nconda activate greengrass_ml_dlr_conda\npython3 {artifacts:decompressedPath}/image_classification/inference.py -a {configuration:/Accelerator} -m {variant.ImageClassification.ModelStore:artifacts:decompressedPath}/{configuration:/ModelResourceKey/x86_64} -p {configuration:/MLRootPath} -i {configuration:/ImageName} -s {configuration:/InferenceInterval}"
        },
        "shutdown": {
          "RequiresPrivilege": true,
          "script": "export PATH=\"{configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin:$PATH\"\neval \"$({configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin/conda shell.bash hook)\"\nconda deactivate"
        }
      }
    }
  ]
}
YAML
RecipeFormatVersion: "2020-01-25"
ComponentName: aws.greengrass.DLRImageClassification
ComponentVersion: "2.0.0"
ComponentDescription: Sample Image classification inference using DLR and amazon resnet50 default model.
ComponentPublisher: Amazon
ComponentDependencies:
  variant.DLR:
    VersionRequirement: "~1.3.0"
    DependencyType: HARD
  variant.ImageClassification.ModelStore:
    VersionRequirement: "~2.0.0"
    DependencyType: HARD
  aws.greengrass.Nucleus:
    VersionRequirement: "~2.0.0"
    DependencyType: SOFT
ComponentConfiguration:
  DefaultConfiguration:
    Accelerator: "cpu"
    MLRootPath: "$HOME/greengrass_ml"
    ImageName: "cat.jpeg"
    InferenceInterval: 3600
    ModelResourceKey:
      armv7l: "DLR-resnet50-armv7l-cpu-ImageClassification"
      x86_64: "DLR-resnet50-x86_64-cpu-ImageClassification"
Manifests:
  - Platform:
      os: linux
      architecture: arm
    Name: 32-bit armv7l - Linux (raspberry pi)
    Artifacts:
      - URI: s3://$bucketName$/$sampleArtifactsDirectory$/image_classification.zip
        Unarchive: ZIP
      - URI: s3://$bucketName$/$sampleArtifactsDirectory$/init.sh
    Lifecycle:
      Install:
        RequiresPrivilege: true
        script: |-
          bash {artifacts:path}/init.sh {artifacts:decompressedPath}/image_classification/sample_images {configuration:/MLRootPath}
          bash {variant.DLR:artifacts:decompressedPath}/installer/installer.sh -a {configuration:/Accelerator} -p {configuration:/MLRootPath}
        timeout: 900
      Run:
        RequiresPrivilege: true
        script: |-
          . {configuration:/MLRootPath}/greengrass_ml_dlr_venv/bin/activate
          python3 {artifacts:decompressedPath}/image_classification/inference.py -a {configuration:/Accelerator} -m {variant.ImageClassification.ModelStore:artifacts:decompressedPath}/{configuration:/ModelResourceKey/armv7l} -p {configuration:/MLRootPath} -i {configuration:/ImageName} -s {configuration:/InferenceInterval}
      shutdown:
        RequiresPrivilege: true
        script: |-
          deactivate
  - Platform:
      os: linux
      architecture: amd64
    Name: 64-bit x86_64 - Linux (ubuntu, deeplens, amazon linux 2)
    Artifacts:
      - URI: s3://$bucketName$/$sampleArtifactsDirectory$/image_classification.zip
        Unarchive: ZIP
      - URI: s3://$bucketName$/$sampleArtifactsDirectory$/init.sh
    Lifecycle:
      Install:
        RequiresPrivilege: true
        script: |-
          bash {artifacts:path}/init.sh {artifacts:decompressedPath}/image_classification/sample_images {configuration:/MLRootPath}
          bash {variant.DLR:artifacts:decompressedPath}/installer/installer.sh -a {configuration:/Accelerator} -p {configuration:/MLRootPath} -e {variant.DLR:artifacts:path}/environment.yaml
        timeout: 900
      Run:
        RequiresPrivilege: true
        script: |-
          export PATH="{configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin:$PATH"
          eval "$({configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin/conda shell.bash hook)"
          conda activate greengrass_ml_dlr_conda
          python3 {artifacts:decompressedPath}/image_classification/inference.py -a {configuration:/Accelerator} -m {variant.ImageClassification.ModelStore:artifacts:decompressedPath}/{configuration:/ModelResourceKey/x86_64} -p {configuration:/MLRootPath} -i {configuration:/ImageName} -s {configuration:/InferenceInterval}
      shutdown:
        RequiresPrivilege: true
        script: |-
          export PATH="{configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin:$PATH"
          eval "$({configuration:/MLRootPath}/greengrass_ml_dlr_conda/bin/conda shell.bash hook)"
          conda deactivate