Batch image annotation offline

The Vision API can run offline (asynchronous) detection services and annotation of a large batch of image files using any Vision feature type. For example, you can specify one or multiple Vision API features (such as TEXT_DETECTION, LABEL_DETECTION, and LANDMARK_DETECTION) for a single batch of images.

Output from an offline batch request is written to a JSON file created in the specified Cloud Storage bucket.

Note: All Vision API feature types also offer online synchronous batch image annotation. Online synchronous requests offer an immediate response to an annotation request. For more information, refer to the how-to pages.

• online (synchronous) requests - An online annotation request (images:annotate or files:annotate) immediately returns inline annotations to the user. Online annotation requests limit the amount of files you can annotate in a single request; you can only specify a single file and specify a small number of pages (<=5) in that file to be annotated.
• offline (asynchronous) requests - An offline annotation request (images:asyncBatchAnnotate or files:asyncBatchAnnotate) starts a long-running operation (LRO) and does not immediately return a response to the caller. When the LRO completes, annotations are stored as files in a Cloud Storage bucket you specify. Offline annotation requests allow you to specify larger batches of files and can specify more pages (<=2000) per file for annotation at a single time than you are able to with online requests.

Limitations

The Vision API accepts up to 2,000 image files. A larger batch of image files will return an error.

Currently supported feature types

Feature type
CROP_HINTS	Determine suggested vertices for a crop region on an image.
DOCUMENT_TEXT_DETECTION	Perform OCR on dense text images, such as documents (PDF/TIFF), and images with handwriting. TEXT_DETECTION can be used for sparse text images. Takes precedence when both DOCUMENT_TEXT_DETECTION and TEXT_DETECTION are present.
FACE_DETECTION	Detect faces within the image.
IMAGE_PROPERTIES	Compute a set of image properties, such as the image's dominant colors.
LABEL_DETECTION	Add labels based on image content.
LANDMARK_DETECTION	Detect geographic landmarks within the image.
LOGO_DETECTION	Detect company logos within the image.
OBJECT_LOCALIZATION	Detect and extract multiple objects in an image.
SAFE_SEARCH_DETECTION	Run SafeSearch to detect potentially unsafe or undesirable content.
TEXT_DETECTION	Perform Optical Character Recognition (OCR) on text within the image. Text detection is optimized for areas of sparse text within a larger image. If the image is a document (PDF/TIFF), has dense text, or contains handwriting, use DOCUMENT_TEXT_DETECTION instead.
WEB_DETECTION	Detect topical entities such as news, events, or celebrities within the image, and find similar images on the web using the power of Google Image Search.

Sample code

Use the following code samples to run offline annotation services on a batch of image files in Cloud Storage.

Note: In the following code samples each requests element (requests_element/requestsElement) corresponds to a single image. To annotate more images, create a request element for each image and add it to the array of requests (requests).

Java

Before trying this sample, follow the Java setup instructions in the Vision API Quickstart Using Client Libraries. For more information, see the Vision API Java API reference documentation.

Note: For Java Spring framework users, Spring Cloud GCP offers a way to automatically configure authentication settings and client objects to use Vision API.

View on GitHub Feedback

import com.google.cloud.vision.v1.AnnotateImageRequest;
import com.google.cloud.vision.v1.AsyncBatchAnnotateImagesRequest;
import com.google.cloud.vision.v1.AsyncBatchAnnotateImagesResponse;
import com.google.cloud.vision.v1.Feature;
import com.google.cloud.vision.v1.GcsDestination;
import com.google.cloud.vision.v1.Image;
import com.google.cloud.vision.v1.ImageAnnotatorClient;
import com.google.cloud.vision.v1.ImageSource;
import com.google.cloud.vision.v1.OutputConfig;
import java.io.IOException;
import java.util.concurrent.ExecutionException;

public class AsyncBatchAnnotateImages {

  public static void asyncBatchAnnotateImages()
      throws InterruptedException, ExecutionException, IOException {
    String inputImageUri = "gs://cloud-samples-data/vision/label/wakeupcat.jpg";
    String outputUri = "gs://YOUR_BUCKET_ID/path/to/save/results/";
    asyncBatchAnnotateImages(inputImageUri, outputUri);
  }

  public static void asyncBatchAnnotateImages(String inputImageUri, String outputUri)
      throws IOException, ExecutionException, InterruptedException {
    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests. After completing all of your requests, call
    // the "close" method on the client to safely clean up any remaining background resources.
    try (ImageAnnotatorClient imageAnnotatorClient = ImageAnnotatorClient.create()) {

      // You can send multiple images to be annotated, this sample demonstrates how to do this with
      // one image. If you want to use multiple images, you have to create a `AnnotateImageRequest`
      // object for each image that you want annotated.
      // First specify where the vision api can find the image
      ImageSource source = ImageSource.newBuilder().setImageUri(inputImageUri).build();
      Image image = Image.newBuilder().setSource(source).build();

      // Set the type of annotation you want to perform on the image
      // https://cloud.google.com/vision/docs/reference/rpc/google.cloud.vision.v1#google.cloud.vision.v1.Feature.Type
      Feature feature = Feature.newBuilder().setType(Feature.Type.LABEL_DETECTION).build();

      // Build the request object for that one image. Note: for additional images you have to create
      // additional `AnnotateImageRequest` objects and store them in a list to be used below.
      AnnotateImageRequest imageRequest =
          AnnotateImageRequest.newBuilder().setImage(image).addFeatures(feature).build();

      // Set where to store the results for the images that will be annotated.
      GcsDestination gcsDestination = GcsDestination.newBuilder().setUri(outputUri).build();
      OutputConfig outputConfig =
          OutputConfig.newBuilder()
              .setGcsDestination(gcsDestination)
              .setBatchSize(2) // The max number of responses to output in each JSON file
              .build();

      // Add each `AnnotateImageRequest` object to the batch request and add the output config.
      AsyncBatchAnnotateImagesRequest request =
          AsyncBatchAnnotateImagesRequest.newBuilder()
              .addRequests(imageRequest)
              .setOutputConfig(outputConfig)
              .build();

      // Make the asynchronous batch request.
      AsyncBatchAnnotateImagesResponse response =
          imageAnnotatorClient.asyncBatchAnnotateImagesAsync(request).get();

      // The output is written to GCS with the provided output_uri as prefix
      String gcsOutputUri = response.getOutputConfig().getGcsDestination().getUri();
      System.out.format("Output written to GCS with prefix: %s%n", gcsOutputUri);
    }
  }
}

Node.js

Before trying this sample, follow the Node.js setup instructions in the Vision Quickstart Using Client Libraries. For more information, see the Vision Node.js API reference documentation.

View on GitHub Feedback

/**
 * TODO(developer): Uncomment these variables before running the sample.
 */
// const inputImageUri = 'gs://cloud-samples-data/vision/label/wakeupcat.jpg';
// const outputUri = 'gs://YOUR_BUCKET_ID/path/to/save/results/';

// Imports the Google Cloud client libraries
const {ImageAnnotatorClient} = require('@google-cloud/vision').v1;

// Instantiates a client
const client = new ImageAnnotatorClient();

// You can send multiple images to be annotated, this sample demonstrates how to do this with
// one image. If you want to use multiple images, you have to create a request object for each image that you want annotated.
async function asyncBatchAnnotateImages() {
  // Set the type of annotation you want to perform on the image
  // https://cloud.google.com/vision/docs/reference/rpc/google.cloud.vision.v1#google.cloud.vision.v1.Feature.Type
  const features = [{type: 'LABEL_DETECTION'}];

  // Build the image request object for that one image. Note: for additional images you have to create
  // additional image request objects and store them in a list to be used below.
  const imageRequest = {
    image: {
      source: {
        imageUri: inputImageUri,
      },
    },
    features: features,
  };

  // Set where to store the results for the images that will be annotated.
  const outputConfig = {
    gcsDestination: {
      uri: outputUri,
    },
    batchSize: 2, // The max number of responses to output in each JSON file
  };

  // Add each image request object to the batch request and add the output config.
  const request = {
    requests: [
      imageRequest, // add additional request objects here
    ],
    outputConfig,
  };

  // Make the asynchronous batch request.
  const [operation] = await client.asyncBatchAnnotateImages(request);

  // Wait for the operation to complete
  const [filesResponse] = await operation.promise();

  // The output is written to GCS with the provided output_uri as prefix
  const destinationUri = filesResponse.outputConfig.gcsDestination.uri;
  console.log(`Output written to GCS with prefix: ${destinationUri}`);
}

asyncBatchAnnotateImages();

Python

Before trying this sample, follow the Python setup instructions in the Vision Quickstart Using Client Libraries. For more information, see the Vision Python API reference documentation.

View on GitHub Feedback

from google.cloud import vision_v1


def sample_async_batch_annotate_images(
    input_image_uri="gs://cloud-samples-data/vision/label/wakeupcat.jpg",
    output_uri="gs://your-bucket/prefix/",
):
    """Perform async batch image annotation."""
    client = vision_v1.ImageAnnotatorClient()

    source = {"image_uri": input_image_uri}
    image = {"source": source}
    features = [
        {"type_": vision_v1.Feature.Type.LABEL_DETECTION},
        {"type_": vision_v1.Feature.Type.IMAGE_PROPERTIES},
    ]

    # Each requests element corresponds to a single image.  To annotate more
    # images, create a request element for each image and add it to
    # the array of requests
    requests = [{"image": image, "features": features}]
    gcs_destination = {"uri": output_uri}

    # The max number of responses to output in each JSON file
    batch_size = 2
    output_config = {"gcs_destination": gcs_destination,
                     "batch_size": batch_size}

    operation = client.async_batch_annotate_images(requests=requests, output_config=output_config)

    print("Waiting for operation to complete...")
    response = operation.result(90)

    # The output is written to GCS with the provided output_uri as prefix
    gcs_output_uri = response.output_config.gcs_destination.uri
    print("Output written to GCS with prefix: {}".format(gcs_output_uri))

Response

A successful request returns response JSON files in the Cloud Storage bucket you indicated in the code sample. The number of responses per JSON file is dictated by batch_size in the code sample.

The returned response is similar to regular Vision API feature responses, depending on which features you request for an image.

The following responses show LABEL_DETECTION and TEXT_DETECTION annotations for image1.png, IMAGE_PROPERTIES annotations for image2.jpg, and OBJECT_LOCALIZATION annotations for image3.jpg.

The response also contain a context field showing the file's URI.

offline_batch_output/output-1-to-2.json

{
  "responses": [
    {
      "labelAnnotations": [
        {
          "mid": "/m/07s6nbt",
          "description": "Text",
          "score": 0.93413997,
          "topicality": 0.93413997
        },
        {
          "mid": "/m/0dwx7",
          "description": "Logo",
          "score": 0.8733531,
          "topicality": 0.8733531
        },
        ...
        {
          "mid": "/m/03bxgrp",
          "description": "Company",
          "score": 0.5682425,
          "topicality": 0.5682425
        }
      ],
      "textAnnotations": [
        {
          "locale": "en",
          "description": "Google\n",
          "boundingPoly": {
            "vertices": [
              {
                "x": 72,
                "y": 40
              },
              {
                "x": 613,
                "y": 40
              },
              {
                "x": 613,
                "y": 233
              },
              {
                "x": 72,
                "y": 233
              }
            ]
          }
        },
        ...
                ],
                "blockType": "TEXT"
              }
            ]
          }
        ],
        "text": "Google\n"
      },
      "context": {
        "uri": "gs://cloud-samples-data/vision/document_understanding/image1.png"
      }
    },
    {
      "imagePropertiesAnnotation": {
        "dominantColors": {
          "colors": [
            {
              "color": {
                "red": 229,
                "green": 230,
                "blue": 238
              },
              "score": 0.2744754,
              "pixelFraction": 0.075339235
            },
            ...
            {
              "color": {
                "red": 86,
                "green": 87,
                "blue": 95
              },
              "score": 0.025770646,
              "pixelFraction": 0.13109145
            }
          ]
        }
      },
      "cropHintsAnnotation": {
        "cropHints": [
          {
            "boundingPoly": {
              "vertices": [
                {},
                {
                  "x": 1599
                },
                {
                  "x": 1599,
                  "y": 1199
                },
                {
                  "y": 1199
                }
              ]
            },
            "confidence": 0.79999995,
            "importanceFraction": 1
          }
        ]
      },
      "context": {
        "uri": "gs://cloud-samples-data/vision/document_understanding/image2.jpg"
      }
    }
  ]
}

offline_batch_output/output-3-to-3.json

{
  "responses": [
    {
      "context": {
        "uri": "gs://cloud-samples-data/vision/document_understanding/image3.jpg"
      },
      "localizedObjectAnnotations": [
        {
          "mid": "/m/0bt9lr",
          "name": "Dog",
          "score": 0.9669734,
          "boundingPoly": {
            "normalizedVertices": [
              {
                "x": 0.6035543,
                "y": 0.1357359
              },
              {
                "x": 0.98546547,
                "y": 0.1357359
              },
              {
                "x": 0.98546547,
                "y": 0.98426414
              },
              {
                "x": 0.6035543,
                "y": 0.98426414
              }
            ]
          }
        },
        ...
        {
          "mid": "/m/0jbk",
          "name": "Animal",
          "score": 0.58003056,
          "boundingPoly": {
            "normalizedVertices": [
              {
                "x": 0.014534635,
                "y": 0.1357359
              },
              {
                "x": 0.37197515,
                "y": 0.1357359
              },
              {
                "x": 0.37197515,
                "y": 0.98426414
              },
              {
                "x": 0.014534635,
                "y": 0.98426414
              }
            ]
          }
        }
      ]
    }
  ]
}