Question

A machine learning team must label 500,000 product images for an image classification model. Manual labeling by a human workforce for the entire dataset is too expensive and slow. The team wants to reduce both cost and labeling time while maintaining acceptable label quality. Which SageMaker Ground Truth capability best meets this requirement?

Accepted Answer

Enable automated data labeling (active learning), which trains a model on human-labeled samples and uses it to auto-label high-confidence images, routing only uncertain images to human labelers.

Answer

Route the entire dataset to an Amazon Mechanical Turk public workforce to maximize parallelism and lower per-image cost.

Answer

Use annotation consolidation with multiple workers per image to reduce the total number of images that require labeling.

Answer

Create a private workforce and increase the number of workers per object so labels are produced faster.