Annotating imported training images

For AutoML Vision Object Detection you can annotate imported training images in three ways:

  • You can provide bounding boxes with labels for your training images via labeled bounding boxes in your .csv import file,
  • You can provide unannotated images in your .csv import file and use the UI to provide image annotations, and/or
  • You can request manual image annotation with Google's Human Labeling service

For details about labeling and bounding boxes for images in your .csv file, see Preparing your training data.

Annotating images via the command line

The AutoML API does not currently include methods for labeling.

Request annotations through Human Labeling

Google's AI Platform Data Labeling Service service enables you to use an API to request human labeling of your dataset. Note that Data Labeling will also be included in AI Hub under the Asset type "Service".

Annotating images via the web UI

Known issue: Some datasets cannot be displayed in the Safari browser. Please try Google Chrome in that case.

After creating a dataset and importing images, you can add or modify image annotations using the web UI.

Web UI

To view all of the imported images in your dataset using the web UI, select the Datasets listing page on the left hand navigation bar, then select the dataset name. This will take you to the following Images listing screen:

Label training images UI

From this screen you can use the UI to:

  • filter images by label (including "labeled" or "unlabeled")
  • add a new label
  • sort labels by image count in ascending or descending order
  • view images as a grid or by image thumbnail
  • select and delete multiple images
  • start model training after you have finished annotating images

Individual image annotation

Web UI

To add, modify, or remove bounding boxes or labels for an individual image in the Cloud AutoML Vision UI complete the following steps:

  1. Select the image you want to update. This will take you to a new screen where existing labels and bounding boxes (if already assigned) are shown.

    UI single image with labels and boxes

    For existing annotations, you can use the select menus and delete icons on the left-hand side to change or delete labels and bounding boxes, and then save your changes.

  2. To add bounding boxes and labels, use your mouse and the provided guiding lines to draw a new box on your image. Draw a new bounding box by clicking on the image (guiding lines appear automatically) and drawing the bounding box with your left-click button pressed. Release the left-click button when you finish drawing the new box. Existing bounding boxes are hidden when you begin drawing a new bounding box.

    UI draw new bounding box

    After drawing a new bounding box, a new label select menu is added for the new box. Hover over the new label select menu to highlight only the new bounding box. By default, a new bounding box is assigned the last label used.

    UI new box highlighted

  3. Add the correct label to the new bounding box by making the appropriate selection from the select menu, then click "save" at the bottom to save your changes.

    UI label new box

View label statistics

Web UI

To view label statistics for a non-empty dataset, complete the following steps:

  1. Select the non-empty dataset from the Datasets page.

    Listing dataset image

    Selecting the non-empty dataset will take you to the Dataset details page.

    Label training images UI

  2. Select the Label Stats option at the top of the Dataset details page.

    Select label stats button

    This opens a window on the right side of the screen where you can view statistics for your labeled bounding boxes in images.

    label stats popout window

  3. Select Done to close the label statistics window.