Question

A data science team is using Amazon SageMaker Ground Truth to label 50,000 images for an object detection model. The labeling is performed by a crowd workforce (Amazon Mechanical Turk), and the team is concerned that individual workers may produce inconsistent or low-quality bounding boxes. They want to improve label quality without manually reviewing every image. What is the most effective way to address this concern?

Accepted Answer

Assign multiple workers to each image and use Ground Truth's annotation consolidation to merge their responses into a single high-confidence label

Answer

Switch the workforce type from Mechanical Turk to a private workforce and label each image only once

Answer

Enable automated data labeling (active learning) so that the model labels all images and no human review is needed

Answer

Store the labeled output in a SageMaker Feature Store to automatically deduplicate conflicting annotations