AutoLabel
Use predictions from an ML model to annotate your data
AutoLabel is a labeling assistance feature that uses predictions from an ML model to automatically annotate an image, saving labelers valuable time and effort. AutoLabel can predict the labels on an image based on the model you choose. The model used for predictive labeling can be:
    a generic object detection model pre-trained on the COCO dataset, a large dataset of common objects
    a custom model trained and deployed in Plainsight using SmartML
Auto Label in action
AutoLabel can be used with the Rectangle and Polygon label types. Support for Feature Points will be available in a future release.

Configuring AutoLabel

Only Admin and Project Manager roles can configure AutoLabel classes.
To use AutoLabel, you will first assign the classes you want to predict for each label. Classes are the label names that a model is trained to identify. A label can be assigned 1 or more classes. There are two ways to assign a class:

Setup in Label Definitions

    1.
    In the Label Definitions tab, click "Add New Label"
    2.
    Enter details such as label name and color. Label type must be Rectangle or Polygon.
    3.
    In the Select Model for AutoLabel drop-down, if available, select the desired model to use for generating annotations. To use the generic COCO model, select "Default COCO (Bounding Box)" or "Default COCO (Polygon)" depending on your label type.
    4.
    In the Model Version drop-down, if available, select the desired model version.
    Note: If selecting a Plainsight model that has not yet been deployed, you will be prompted to deploy the model.
    5.
    In the Classes drop-down, select one or more pre-trained classes from the list. When AutoLabel is enabled, the image will be processed as input to the selected model and predictions for the selected classes will be automatically labeled.
    6.
    Click "Save" to add your label.
Your label will be added with the selected classes listed under the AutoLabel Classes column.

Setup in Labeler

    1.
    In the Labeler, select the Rectangle or Polygon label you want to use for AutoLabel predictions
    2.
    Click the
    to select AutoLabel from the drop-down.
    3.
    If classes were not configured in Label Definitions, you will be prompted to attach one or more classes to this label
    4.
    Click "Okay" to apply the classes. AutoLabel will begin analyzing the image for predictions from the ML model based on the classes selected. Labels will be applied to any matches.

Enabling/Disabling AutoLabel

Follow the Setup in Labeler steps above to enable AutoLabel for a pre-configured label.
The AutoLabel icon will have a dark background
when it's enabled for a label. When enabled, each image will be analyzed for predictions based on the ML model configured. Labels will be applied to any matches.
To turn off AutoLabel for a label, simply click
icon to toggle it off. The AutoLabel will have a white background
when disabled. AutoLabel can be turned on and off for labels individually.
AutoLabel works best when:
    used for detecting common, everyday objects in images using the Default COCO model.
    used for labeling a new set of images to be used in training a new Plainsight model or version.
    used as assistance in labeling objects quickly. Additional labeling or adjusting of the generated labels by a human labeler may still be necessary.
Last modified 2mo ago