Amazon SageMaker Ground Truth
Cloud service · AWS · Updated

Amazon SageMaker Ground Truth

In short

Amazon SageMaker Ground Truth is a fully managed data labeling service that makes it easy to build highly accurate training datasets for machine learning quickly and efficiently.

Cloud Services Hub →

8 key features of Amazon SageMaker Ground Truth

Amazon SageMaker Ground Truth offers built-in and customizable labeling workflows, automated data labeling with active learning, human-in-the-loop quality assurance, strong data security, and seamless integration with AWS services, facilitating efficient and scalable machine learning dataset preparation.

01
Built-in Labeling Workflows

Amazon SageMaker Ground Truth offers a selection of pre-built workflows for common labeling tasks, such as image classification, text classification, and object detection, making it easier to get started without needing to design your own workflows.

02
Automated Data Labeling

Leverages machine learning to automatically label data, significantly reducing the time and effort required to label large datasets manually. This active learning feature becomes more accurate over time as it learns from the labels created by human annotators.

03
Human-in-the-Loop

Incorporates human review to ensure high-quality labels. You can use your own team for labeling or access a workforce through Amazon Mechanical Turk, third-party vendors, or AWS Marketplace.

04
Custom Labeling Workflows

Allows you to create custom labeling workflows to handle specific use cases not covered by the built-in workflows. This includes setting up unique data labeling tasks and incorporating specific guidelines and tools needed for your data.

05
Data Security

Implements multiple layers of security, including encryption at rest and in transit, to ensure that your data remains secure throughout the labeling process. Access controls and activity logging offer additional security and compliance.

06
Integration and Scalability

Seamlessly integrates with other AWS services, such as Amazon S3 for data storage, and scales to handle large-scale data labeling projects, making it easy to manage datasets and annotations across your entire machine learning workflow.

07
Annotation Consolidation

Combines annotations from two or more workers into a single label for each data object. You choose how many workers annotate each object; using more workers can improve label accuracy but increases labeling costs.

08
Streaming Labeling Jobs

Supports perpetually running labeling jobs that send new dataset objects to workers in real time. Workers continue receiving objects while the job remains active and new objects are being submitted.

Where Amazon SageMaker Ground Truth is used

Amazon SageMaker Ground Truth is utilized for comprehensive data labeling tasks, including image and video annotation for autonomous vehicles, medical image classification, content moderation on social media platforms, and text data annotation for sentiment analysis.

  1. 01
    Image and Video Annotation for Autonomous Vehicles

    Amazon SageMaker Ground Truth can be used to accurately label images and videos captured by cameras on autonomous vehicles. These labeled datasets are crucial for training machine learning models to recognize objects like pedestrians, other vehicles, traffic signs, and lane markings, ensuring safer navigation and decision-making by autonomous systems.

  2. 02
    Medical Image Classification

    Healthcare organizations can leverage SageMaker Ground Truth to improve diagnostic accuracy by annotating medical imagery, such as X-rays, MRIs, and CT scans. By providing detailed labels for various conditions and anomalies, Ground Truth helps in the creation of precise models that assist medical professionals in quicker and more accurate diagnosis.

  3. 03
    Content Moderation for Social Media

    SageMaker Ground Truth supports social media platforms by enabling the annotation of user-generated content, helping to identify and filter out inappropriate or harmful material. By training models on datasets labeled for content moderation, companies can maintain community standards and create safer online environments.

  4. 04
    Text Data for Sentiment Analysis

    This use case involves annotating text data, such as customer feedback or social media posts, using SageMaker Ground Truth to detect sentiment, opinions, and trends. This can help businesses better understand customer satisfaction, monitor brand reputation, and tailor products and services to meet customer needs more effectively.

How Amazon SageMaker Ground Truth is priced

Amazon SageMaker Ground Truth pricing includes human labeling jobs charged per data object labeled and automated data labeling also charged per data object, with costs varying based on task complexity and the level of human oversight required.

Human Labeling Jobs

Amazon SageMaker Ground Truth charges you based on the number of data objects labeled by human annotators. The cost varies depending on the complexity of the labeling task and the type of labels required. Tasks that require more detailed annotations or more complex decision-making can have higher costs. Pricing is defined per data object, which can be an image, text, video frame, etc.

Automated Data Labeling

For tasks using automated data labeling, SageMaker Ground Truth applies machine learning models to automatically label your data. The cost for automated labeling is significantly lower than human-labeled data, based on the number of data objects processed. However, this might still involve human oversight to ensure the quality of the labels. Automated data labeling is priced per data object processed by the automation models.

Ground Truth Plus Project Pricing

SageMaker Ground Truth Plus projects are priced individually; the AWS team reviews pricing options with you after you submit a project requirement form.

Services Amazon SageMaker Ground Truth works with

Amazon S3Amazon S3

Used to store the input datasets, output labeled datasets, and intermediate files required during the labeling process.

Amazon SageMakerAmazon SageMaker

Seamlessly integrates with other SageMaker features for training and deploying machine learning models using the labeled data.

AWS LambdaAWS Lambda

Facilitates preprocessing and post-processing of data, automating workflows, and integrating custom logic into the data labeling pipeline.

Amazon Augmented AI (Amazon A2I)Amazon Augmented AI (Amazon A2I)

Ground Truth private workforces and work teams can also perform Amazon A2I human-review tasks. This lets the same managed workforce handle both training-data annotation and reviews of machine-learning predictions.

Amazon Bedrock

Private work teams managed through Ground Truth can serve as human evaluators for Amazon Bedrock model-evaluation jobs. Evaluators use a worker portal to assess model responses against the evaluation criteria.

Amazon CloudWatchAmazon CloudWatch

Ground Truth publishes labeling-job and worker metrics to CloudWatch and writes labeling-workflow logs to CloudWatch Logs. These help track annotation progress, monitor worker activity, and troubleshoot failures.

Amazon CognitoAmazon Cognito

Ground Truth uses Cognito user pools to manage authentication for private labeling workforces. Workers sign in to the labeling portal with accounts associated with the configured user pool.

Amazon ComprehendAmazon Comprehend

Ground Truth can produce augmented manifests that Amazon Comprehend accepts for custom classification and custom entity-recognition training. This provides a supported path from human annotation to training a custom text-analysis model.

Amazon EventBridgeAmazon EventBridge

Ground Truth labeling-job state changes are available as EventBridge events. Rules can route completion or failure events to downstream automation, such as a Lambda function that processes the labeled dataset.

Amazon RekognitionAmazon Rekognition

Ground Truth image-classification and bounding-box annotations can be imported into Amazon Rekognition Custom Labels using supported manifest formats. The annotations supply image labels or object locations for training a custom recognition model.

What is Amazon SageMaker Ground Truth

2 min read

Amazon SageMaker Ground Truth is a fully managed data labeling service that makes it easy to build highly accurate training datasets for machine learning (ML) quickly. Ground Truth significantly reduces the time and effort required to prepare data for ML models by automating the iterative and time-consuming tasks of data labeling. This service is an integral part of Amazon SageMaker, a comprehensive service that enables developers and data scientists to build, train, and deploy machine learning models at scale. 

Amazon SageMaker Ground Truth is no longer open to new customers. Existing customers can continue using the service as normal. AWS continues to invest in security and availability improvements but does not plan to introduce new features.

At the heart of Amazon SageMaker Ground Truth is its ability to offer both automated and human labeling workflows. For human labeling, Ground Truth supports a workforce consisting of either the customer's own employees, a third-party vendor recommended by AWS, or Amazon Mechanical Turk, providing flexibility in how data is annotated based on the specific needs of a project or the sensitivity of the data being labeled. This hybrid approach allows Ground Truth to provide highly accurate labels by combining human oversight with machine learning models to automate labeling tasks where possible. As the human annotators label the data, Ground Truth can learn from their inputs to make smart predictions for similar unlabeled data, consequently reducing the need for human labels and accelerating the labeling process. 

Alongside the self-service offering, SageMaker Ground Truth Plus provides an AWS-managed offering. AWS selects and manages the workforce, designs and customizes the end-to-end workflow—including workforce training and quality assurance—and provides a skilled team trained for your specific tasks and data quality, security, and compliance requirements.

Ground Truth offers support for a wide range of data types such as images, text, and audio, making it versatile for various applications, from autonomous driving and object detection to text classification and sentiment analysis. Its interface is intuitive, requiring no machine learning expertise to get started, yet powerful enough to handle complex labeling tasks with features like bounding boxes, semantic segmentation, and custom workflows for specific use cases. 

Input datasets for semantic segmentation labeling jobs have a quota of 20,000 items. Other labeling job types have a dataset size quota of 100,000 items; quota increases can be requested for jobs other than semantic segmentation.

Moreover, Ground Truth is designed to ensure the privacy and security of data. It offers features like encrypted data storage and secure data access protocols, making it suitable for use in industries with stringent data protection requirements, such as healthcare and finance.  

In essence, Amazon SageMaker Ground Truth addresses one of the most significant bottlenecks in the machine learning pipeline: the preparation of high-quality training datasets. By seamlessly combining human intuition and judgment with the scalability and speed of machine learning, Ground Truth enables organizations to accelerate their ML initiatives, ultimately driving better insights, improving efficiency, and creating innovative solutions that leverage the power of artificial intelligence.

Concepts related to Amazon SageMaker Ground Truth

AWS architecture icons are the property of Amazon Web Services. BlowStack is not affiliated with or endorsed by AWS.