Posted On: Mar 9, 2020

Amazon SageMaker Ground Truth now supports multi-label image and text classification. Ground Truth helps you build highly accurate training datasets by using your own or third-party human labelers. It provides labelers with built-in workflows and user interfaces for common labeling tasks. Built-in workflows are provided for image and text classification, for assigning class labels to an image or a text selection, along with workflows for other computer vision (CV) and natural language processing (NLP) tasks. 

When you set up a Ground Truth labeling job for classifying images or text, you can now configure the selection of multiple labels per object. For example, if you have images of a soccer match, you can ask labelers to select any of the following class labels that apply: ‘ball’, ‘referee’, ‘player’, or ‘spectator’. You can now pursue a broader set of use cases with the built-in image and text classification workflows.

In addition, Ground Truth now emits Amazon CloudWatch Events for status changes of a labeling job. Specifically, you can set up a CloudWatch Events rule for Ground Truth to send an event whenever a labeling job status changes to Completed, Failed, or Stopped.

Learn more about image multi-label classification, text multi-label classification and CloudWatch events in the Amazon SageMaker Developer Guide.