AWS › AWS AI Services (Rekognition, Polly, Lex & Comprehend)

How do you train a custom model in Amazon Rekognition Custom Labels or Comprehend Custom Classification for domain specific needs?

Advanced

Answer

Both Rekognition Custom Labels and Comprehend Custom Classification let you provide your own labeled training examples, such as images tagged with specific product defect types or text documents tagged with custom business categories, and the service automatically handles the underlying machine learning training process, producing a custom model you can then use through the same simple API pattern as the built in features, without requiring deep machine learning expertise.

aws rekognition create-project-version --project-arn arn:aws:rekognition:us-east-1:123456789012:project/defect-detection --version-name v1 --output-config S3Bucket=my-bucket

Real-world example A manufacturing company trains a Rekognition Custom Labels model using thousands of labeled photos of both defective and acceptable parts from its own assembly line, achieving accurate defect detection specific to its unique products that a generic model could never provide.

Common follow-ups: How many labeled examples are typically needed to train an effective custom model?;How does the cost of custom models compare to using the built in pretrained features?

Amazon SageMaker & Machine Learning on AWS;AWS Cost Management & Billing