Amazon SageMaker Ground Truth
Label training data
A service for creating labeled training data: assigns labeling of images, text, video, and point clouds to internal staff, vendors, or Mechanical Turk; can use automated labeling (active learning) to reduce the workload.
- Cost$$ Medium
- Difficulty●●●●●
- Costs money when idle?No
- Managed by AWS
- In examsSAA
Think of it asLike a product labeling workshop: many people sort items together while a machine gradually learns to label the easy ones itself.
When to use it
✓ Use it when
You need high-quality labeled datasets to train models.
✗ Avoid it when
You only need humans to review predictions from a model already running → Augmented AI (A2I).
Common uses
- Draw bounding boxes around objects in images for self-driving cars
- Label text sentiment
- Create evaluation data for generative AI models
What does it cost?
- Pricing model: Charged by the number of objects labeled, plus fees paid to the labeling workforce (Mechanical Turk, vendors).
