导出模型的标签和置信度

Question

我已经通过 Google Cloud Vision 训练了一个模型。在 Evaluate 选项卡上，它向我显示了它所训练的标签列表以及置信度值（见下文）。有什么方法可以将此信息下载为 CSV 格式？ EXPORT DATA 不幸的是，它不会这样做，它只会导出图像位置以及一些元数据，没有置信度信息。

Answer 1

要提取评价数据，可以向projects.locations.models.modelEvaluations.list发送GET请求。为此，您将需要 locationId 和 modelId。

要获取 locationId，请向 projects.locations.list 发送 GET 请求。获取请求：

curl -X GET -H "Content-Type: application/json" -H "Authorization: Bearer "$(gcloud auth application-default print-access-token) https://automl.googleapis.com/v1/projects/your-project-id/locations

projects.locations.list returns:

{
  "locations": [
    {
      "name": "projects/your-project-id/locations/eu",
      "locationId": "eu"
    },
    {
      "name": "projects/your-project-id/locations/us-central1",
      "locationId": "us-central1"
    }
  ]
}

要获取 modelId，请向 projects.locations.models.list 发送 GET 请求。获取请求：

curl -X GET -H "Content-Type: application/json" -H "Authorization: Bearer "$(gcloud auth application-default print-access-token) https://automl.googleapis.com/v1/projects/your-project-id/locations/us-central1/models

name: 包含 modelId 在这个例子中它是 ICN1234567890。 projects.locations.models.list returns:

{
  "model": [
    {
      "name": "projects/your-project-id/locations/us-central1/models/ICN1234567890",
      "displayName": "flowers_20201217111947",
      "datasetId": "ICN8001048258791079936",
      "createTime": "2020-12-17T03:20:27.198250Z",
      "deploymentState": "UNDEPLOYED",
      "updateTime": "2021-09-07T08:59:02.243603Z",
      "imageClassificationModelMetadata": {
        "stopReason": "MODEL_EARLY_STOPPED",
        "modelType": "mobile-versatile-1",
        "nodeQps": 3.2,
        "trainBudgetMilliNodeHours": "4000",
        "trainCostMilliNodeHours": "1882"
      }
    },...

现在构建评估数据的 GET 请求。使用 projects.locations.models.modelEvaluations.list:

curl -X GET -H "Content-Type: application/json" -H "Authorization: Bearer "$(gcloud auth application-default print-access-token) https://automl.googleapis.com/v1/projects/your-project-id/locations/us-central1/models/ICN1234567890/modelEvaluations

modelEvaluations.listreturns。请参阅下面的回复片段：

   {
      "name": "projects/your-project-id/locations/us-central1/models/ICN1234567890/modelEvaluations/8145357881766117082",
      "annotationSpecId": "5162564433687347200",
      "createTime": "2020-12-17T05:49:10.557128Z",
      "classificationEvaluationMetrics": {
        "auPrc": 0.99306494,
        "confidenceMetricsEntry": [
          {
            "recall": 1,
            "precision": 0.17166212
          },
          {
            "confidenceThreshold": 0.05,....

您可以参考 this 查看为 modelEvaluation 返回的完整字段。通过 modelEvaluations.list 查看 UI 和值的比较：

导出模型的标签和置信度

Export a Model's Labels and Confidence

google-cloud-vision