Label Studio for configurable data annotation

How it works

Label configurations describe what reviewers see and which annotations they can create. Predictions can prepopulate tasks in supported workflows for review and correction.

What you need

Design the labeling interface for the expected asset type and review workflow. Keep an example export so downstream training code can be tested against its exact schema.

What to try first

Set up one small project and import a few tasks with stable source IDs. Export labels and verify coordinates, times or offsets in a downstream parser.

Export labels after editing a task and check which annotation is current. Downstream training jobs need a rule for revisions and multiple reviewer responses.

Where to look for failures

Pre-annotations can bias reviewers. Separate reviewer quality checks from the model's suggestions and confirm asset access rules.

Related Franklin pages

Separate annotation from review of likely data problems. Check which record identifiers survive an export so reviewer decisions can be traced back to inputs.

• Prioritize examples for review using model predictions and data-quality methods. cleanlab for finding likely dataset problems
• Define annotation records, fields and questions for people reviewing model outputs. Argilla for collecting feedback on AI datasets

For the steps that prepare raw data for training or evaluation, see Franklin’s glossary. Data preprocessing

Project documentation

• Current project README
• README at the reviewed revision